2023-02-09 11:18:06 -08:00
#!/bin/bash
red = '\033[0;31m'
green = '\033[0;32m'
2024-12-20 09:33:27 -08:00
blue = '\033[0;34m'
2023-02-09 11:18:06 -08:00
yellow = '\033[0;33m'
plain = '\033[0m'
2026-01-02 18:57:19 -08:00
xui_folder = " ${ XUI_MAIN_FOLDER : =/usr/local/x-ui } "
xui_service = " ${ XUI_SERVICE : =/etc/systemd/system } "
2023-02-09 11:18:06 -08:00
# check root
2026-05-19 04:09:35 -07:00
[ [ $EUID -ne 0 ] ] && echo -e " ${ red } Fatal error: ${ plain } Please run this script with root privilege \n " && exit 1
2023-02-09 11:18:06 -08:00
2023-03-11 07:05:35 -08:00
# Check OS and set release variable
if [ [ -f /etc/os-release ] ] ; then
source /etc/os-release
release = $ID
2026-05-04 04:20:24 -07:00
elif [ [ -f /usr/lib/os-release ] ] ; then
2023-03-11 07:05:35 -08:00
source /usr/lib/os-release
release = $ID
2023-02-09 11:18:06 -08:00
else
2026-05-19 04:09:35 -07:00
echo "Failed to check the system OS, please contact the author!" >& 2
exit 1
2023-02-09 11:18:06 -08:00
fi
2023-03-11 07:05:35 -08:00
echo " The OS release is: $release "
2023-02-09 11:18:06 -08:00
2024-04-01 01:42:28 -07:00
arch( ) {
2023-04-21 08:30:14 -07:00
case " $( uname -m) " in
2025-12-03 12:40:49 -08:00
x86_64 | x64 | amd64) echo 'amd64' ; ;
i*86 | x86) echo '386' ; ;
armv8* | armv8 | arm64 | aarch64) echo 'arm64' ; ;
armv7* | armv7 | arm) echo 'armv7' ; ;
armv6* | armv6) echo 'armv6' ; ;
armv5* | armv5) echo 'armv5' ; ;
s390x) echo 's390x' ; ;
2026-07-28 14:11:27 -07:00
*) echo -e " ${ green } Unsupported CPU architecture! ${ plain } " && rm -f " $( realpath " $0 " ) " && exit 1 ; ;
2023-04-21 08:30:14 -07:00
esac
}
2024-01-20 01:32:35 -08:00
2025-03-21 08:30:31 -07:00
echo " Arch: $( arch) "
2023-02-09 11:18:06 -08:00
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
# Non-interactive mode: triggered explicitly via XUI_NONINTERACTIVE=1, or
# implicitly when stdin is not a TTY (e.g. `curl ... | bash`, cloud-init).
# In this mode every prompt below is replaced by an env var or a sane default.
if [ [ " ${ XUI_NONINTERACTIVE :- 0 } " = = "1" ] ] || [ [ ! -t 0 ] ] ; then
NONINTERACTIVE = 1
else
NONINTERACTIVE = 0
fi
export NONINTERACTIVE
2025-12-27 15:03:33 -08:00
# Simple helpers
is_ipv4( ) {
[ [ " $1 " = ~ ^( [ 0-9] { 1,3} \. ) { 3} [ 0-9] { 1,3} $ ] ] && return 0 || return 1
}
is_ipv6( ) {
[ [ " $1 " = ~ : ] ] && return 0 || return 1
}
is_ip( ) {
is_ipv4 " $1 " || is_ipv6 " $1 "
}
is_domain( ) {
2026-01-11 17:53:43 -08:00
[ [ " $1 " = ~ ^( [ A-Za-z0-9] ( -*[ A-Za-z0-9] ) *\. ) +( xn--[ a-z0-9] { 2,} | [ A-Za-z] { 2,} ) $ ] ] && return 0 || return 1
2026-01-11 06:28:43 -08:00
}
2026-06-11 16:22:30 -07:00
# acme.sh's standalone server binds IPv4 by default; --listen-v6 makes it
# v6-only, which breaks HTTP-01 validation when the domain's A record points
# at this host's IPv4 (#4994). Only force IPv6 when the host has no global
# IPv4 address at all.
acme_listen_flag( ) {
if ip -4 addr show scope global 2> /dev/null | grep -q "inet " ; then
echo ""
else
echo "--listen-v6"
fi
}
2026-01-11 06:28:43 -08:00
# Port helpers
is_port_in_use( ) {
local port = " $1 "
2026-05-04 04:20:24 -07:00
if command -v ss > /dev/null 2>& 1; then
ss -ltn 2> /dev/null | awk -v p = " : ${ port } $" '$4 ~ p {exit 0} END {exit 1}'
2026-01-11 06:28:43 -08:00
return
fi
2026-05-04 04:20:24 -07:00
if command -v netstat > /dev/null 2>& 1; then
netstat -lnt 2> /dev/null | awk -v p = " : ${ port } " '$4 ~ p {exit 0} END {exit 1}'
2026-01-11 06:28:43 -08:00
return
fi
2026-05-04 04:20:24 -07:00
if command -v lsof > /dev/null 2>& 1; then
lsof -nP -iTCP:${ port } -sTCP:LISTEN > /dev/null 2>& 1 && return 0
2026-01-11 06:28:43 -08:00
fi
return 1
2025-12-27 15:03:33 -08:00
}
2026-05-19 04:09:35 -07:00
install_base( ) {
case " ${ release } " in
ubuntu | debian | armbian)
apt-get update && apt-get install -y -q cron curl tar tzdata socat ca-certificates openssl
; ;
fedora | amzn | virtuozzo | rhel | almalinux | rocky | ol)
2026-07-02 15:01:54 -07:00
dnf makecache -y && dnf install -y -q cronie curl tar tzdata socat ca-certificates openssl
2026-05-19 04:09:35 -07:00
; ;
centos)
if [ [ " ${ VERSION_ID } " = ~ ^7 ] ] ; then
2026-07-02 15:01:54 -07:00
yum makecache -y && yum install -y cronie curl tar tzdata socat ca-certificates openssl
2026-05-19 04:09:35 -07:00
else
2026-07-02 15:01:54 -07:00
dnf makecache -y && dnf install -y -q cronie curl tar tzdata socat ca-certificates openssl
2026-05-19 04:09:35 -07:00
fi
; ;
arch | manjaro | parch)
2026-07-05 11:03:11 -07:00
pacman -Sy --noconfirm cronie curl tar tzdata socat ca-certificates openssl
2026-05-19 04:09:35 -07:00
; ;
opensuse-tumbleweed | opensuse-leap)
zypper refresh && zypper -q install -y cron curl tar timezone socat ca-certificates openssl
; ;
alpine)
apk update && apk add dcron curl tar tzdata socat ca-certificates openssl
; ;
*)
apt-get update && apt-get install -y -q cron curl tar tzdata socat ca-certificates openssl
; ;
esac
}
2024-06-24 05:43:39 -07:00
gen_random_string( ) {
local length = " $1 "
2026-05-04 04:20:24 -07:00
openssl rand -base64 $(( length * 2 )) \
2026-04-01 04:59:48 -07:00
| tr -dc 'a-zA-Z0-9' \
| head -c " $length "
2024-06-24 05:43:39 -07:00
}
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
# prompt_or_default VARNAME "prompt text" "default" [ENV_NAME]
# Interactive: read into VARNAME. Non-interactive: VARNAME = ${ENV_NAME:-default}.
# ENV_NAME defaults to VARNAME when omitted. Keeps every interactive prompt
# string byte-for-byte identical to the original `read -rp`.
prompt_or_default( ) {
local __var = " $1 " __prompt = " $2 " __default = " $3 " __env = " ${ 4 :- $1 } "
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
printf -v " $__var " '%s' " ${ !__env :- $__default } "
else
# shellcheck disable=SC2229
read -rp " $__prompt " " $__var "
fi
}
# write_install_result <user> <pass> <port> <webpath> <scheme> <host> <token> <dbtype>
# Persists a parseable, root-only credentials file consumed by cloud-init/MOTD.
# Values are written with printf '%q' so a pinned password/username containing
# spaces, quotes, $(...) or backticks is shell-escaped and the file stays safely
# source-able (consumers do '. install-result.env'). For the alphanumeric random
# values gen_random_string emits, %q is a no-op. This is a DIFFERENT file from the
# Postgres env file (/etc/default/x-ui).
write_install_result( ) {
local u = " $1 " p = " $2 " port = " $3 " wbp = " $4 " scheme = " $5 " host = " $6 " token = " $7 " dbtype = " $8 "
local result_file = "/etc/x-ui/install-result.env"
local url_host = " ${ host :- SERVER_IP_UNKNOWN } "
install -d -m 755 /etc/x-ui 2> /dev/null
local prev_umask
prev_umask = $( umask )
umask 077
if ! {
printf 'XUI_USERNAME=%q\n' " $u "
printf 'XUI_PASSWORD=%q\n' " $p "
printf 'XUI_PANEL_PORT=%q\n' " $port "
printf 'XUI_WEB_BASE_PATH=%q\n' " $wbp "
printf 'XUI_ACCESS_URL=%q\n' " ${ scheme } :// ${ url_host } : ${ port } / ${ wbp } "
printf 'XUI_API_TOKEN=%q\n' " $token "
printf 'XUI_DB_TYPE=%q\n' " $dbtype "
} > " $result_file " ; then
umask " $prev_umask "
echo -e " ${ yellow } Warning: failed to write ${ result_file } . ${ plain } " >& 2
return 1
fi
umask " $prev_umask "
chmod 600 " $result_file " 2> /dev/null
chown root:root " $result_file " 2> /dev/null || true
echo -e " ${ green } Install result written to ${ result_file } (mode 600). ${ plain } "
}
2026-07-05 11:06:08 -07:00
# RHEL-family initdb writes pg_hba.conf host rules with ident auth, which
# compares the OS username against the Postgres role and always rejects the
# randomly generated panel role over TCP (#5806). Prepend password-auth rules
# scoped to the panel database; first match wins, and md5 also accepts
# scram-sha-256-stored verifiers, so this works on every supported distro.
pg_ensure_hba_password_auth( ) {
local pg_db = " $1 "
local hba_file
hba_file = $( sudo -u postgres psql -tAc 'SHOW hba_file' 2> /dev/null | tr -d '[:space:]' )
[ [ -n " ${ hba_file } " && -f " ${ hba_file } " ] ] || return 0
grep -Eq " ^host[[:space:]]+ ${ pg_db } [[:space:]] " " ${ hba_file } " && return 0
local tmp
tmp = $( mktemp) || return 1
{
echo "# Added by 3x-ui: allow password logins for the panel database."
echo " host ${ pg_db } all 127.0.0.1/32 md5 "
echo " host ${ pg_db } all ::1/128 md5 "
cat " ${ hba_file } "
} > " ${ tmp } " || {
rm -f " ${ tmp } "
return 1
}
cat " ${ tmp } " > " ${ hba_file } " || {
rm -f " ${ tmp } "
return 1
}
rm -f " ${ tmp } "
sudo -u postgres psql -tAc 'SELECT pg_reload_conf()' > /dev/null 2>& 1 || true
}
2026-05-23 10:52:37 -07:00
install_postgres_local( ) {
2026-05-28 08:20:16 -07:00
local pg_user pg_pass
2026-05-23 10:52:37 -07:00
pg_pass = $( gen_random_string 24)
2026-05-28 08:20:16 -07:00
local pg_db = "xui"
local pg_host = "127.0.0.1"
local pg_port = "5432"
2026-05-23 10:52:37 -07:00
case " ${ release } " in
ubuntu | debian | armbian)
apt-get update >& 2 && apt-get install -y -q postgresql >& 2 || return 1
; ;
fedora | amzn | virtuozzo | rhel | almalinux | rocky | ol)
dnf install -y -q postgresql-server postgresql-contrib >& 2 || return 1
[ [ -d /var/lib/pgsql/data && -f /var/lib/pgsql/data/PG_VERSION ] ] || postgresql-setup --initdb >& 2 || return 1
; ;
centos)
if [ [ " ${ VERSION_ID } " = ~ ^7 ] ] ; then
yum install -y postgresql-server postgresql-contrib >& 2 || return 1
else
dnf install -y -q postgresql-server postgresql-contrib >& 2 || return 1
fi
[ [ -d /var/lib/pgsql/data && -f /var/lib/pgsql/data/PG_VERSION ] ] || postgresql-setup --initdb >& 2 || return 1
; ;
arch | manjaro | parch)
2026-07-05 11:03:11 -07:00
pacman -Sy --noconfirm postgresql >& 2 || return 1
2026-05-23 10:52:37 -07:00
if [ [ ! -f /var/lib/postgres/data/PG_VERSION ] ] ; then
sudo -u postgres initdb -D /var/lib/postgres/data >& 2 || return 1
fi
; ;
opensuse-tumbleweed | opensuse-leap)
zypper -q install -y postgresql-server postgresql-contrib >& 2 || return 1
if [ [ ! -f /var/lib/pgsql/data/PG_VERSION ] ] ; then
install -d -o postgres -g postgres -m 700 /var/lib/pgsql/data >& 2 || return 1
su - postgres -c "initdb -D /var/lib/pgsql/data" >& 2 || return 1
fi
; ;
alpine)
apk add --no-cache postgresql postgresql-contrib >& 2 || return 1
if [ [ ! -f /var/lib/postgresql/data/PG_VERSION ] ] ; then
/etc/init.d/postgresql setup >& 2 || return 1
fi
rc-update add postgresql default >& 2 2> /dev/null || true
rc-service postgresql start >& 2 || return 1
; ;
*)
echo -e " ${ red } Unsupported distro for automatic PostgreSQL install: ${ release } ${ plain } " >& 2
return 1
; ;
esac
if [ [ " ${ release } " != "alpine" ] ] ; then
systemctl enable --now postgresql >& 2 || return 1
fi
# Wait briefly for the server to accept connections.
local i
for i in 1 2 3 4 5; do
sudo -u postgres psql -tAc 'SELECT 1' > /dev/null 2>& 1 && break
sleep 1
done
2026-05-28 08:20:16 -07:00
local existing_owner = ""
existing_owner = $( sudo -u postgres psql -tAc \
" SELECT pg_catalog.pg_get_userbyid(datdba) FROM pg_database WHERE datname=' ${ pg_db } ' " 2> /dev/null \
| tr -d '[:space:]' )
if [ [ -n " ${ existing_owner } " && " ${ existing_owner } " != "postgres" ] ] ; then
pg_user = " ${ existing_owner } "
else
pg_user = $( gen_random_string 8)
fi
# Idempotent role/db creation. Identifiers are double-quoted because a
# random username may start with a digit, which Postgres rejects unquoted.
2026-05-23 10:52:37 -07:00
sudo -u postgres psql -tAc " SELECT 1 FROM pg_roles WHERE rolname=' ${ pg_user } ' " 2> /dev/null \
| grep -q 1 \
2026-05-28 08:20:16 -07:00
|| sudo -u postgres psql -c " CREATE USER \" ${ pg_user } \" WITH PASSWORD ' ${ pg_pass } '; " >& 2 || return 1
2026-05-23 10:52:37 -07:00
sudo -u postgres psql -tAc " SELECT 1 FROM pg_database WHERE datname=' ${ pg_db } ' " 2> /dev/null \
| grep -q 1 \
2026-05-28 08:20:16 -07:00
|| sudo -u postgres psql -c " CREATE DATABASE \" ${ pg_db } \" OWNER \" ${ pg_user } \"; " >& 2 || return 1
2026-05-23 10:52:37 -07:00
2026-05-28 08:20:16 -07:00
sudo -u postgres psql -c " ALTER USER \" ${ pg_user } \" WITH PASSWORD ' ${ pg_pass } '; " >& 2 || return 1
2026-05-23 10:52:37 -07:00
2026-07-05 11:06:08 -07:00
pg_ensure_hba_password_auth " ${ pg_db } " \
|| echo -e " ${ yellow } Warning: could not update pg_hba.conf; PostgreSQL may reject the panel's TCP login (ident auth). ${ plain } " >& 2
2026-05-23 10:52:37 -07:00
local pg_pass_enc
pg_pass_enc = $( printf '%s' " ${ pg_pass } " | sed -e 's/%/%25/g' -e 's/:/%3A/g' -e 's/@/%40/g' -e 's|/|%2F|g' -e 's/?/%3F/g' -e 's/#/%23/g' )
2026-05-28 08:20:16 -07:00
if [ [ -n " ${ PG_CRED_FILE :- } " ] ] ; then
local prev_umask
prev_umask = $( umask )
umask 077
if ! cat > " ${ PG_CRED_FILE } " << EOF; then
PG_USER = ${ pg_user }
PG_PASS = ${ pg_pass }
PG_HOST = ${ pg_host }
PG_PORT = ${ pg_port }
PG_DB = ${ pg_db }
EOF
umask " ${ prev_umask } "
echo -e " ${ red } Failed to write PostgreSQL credentials to ${ PG_CRED_FILE } ${ plain } " >& 2
return 1
fi
umask " ${ prev_umask } "
fi
echo " postgres:// ${ pg_user } : ${ pg_pass_enc } @ ${ pg_host } : ${ pg_port } / ${ pg_db } ?sslmode=disable "
2026-05-23 10:52:37 -07:00
return 0
}
2026-05-31 08:53:34 -07:00
ensure_pg_client( ) {
if command -v pg_dump > /dev/null 2>& 1 && command -v pg_restore > /dev/null 2>& 1; then
return 0
fi
echo -e " ${ yellow } Installing PostgreSQL client tools (pg_dump/pg_restore) for in-panel backup... ${ plain } " >& 2
case " ${ release } " in
ubuntu | debian | armbian)
apt-get update >& 2 && apt-get install -y -q postgresql-client >& 2 || return 1
; ;
fedora | amzn | virtuozzo | rhel | almalinux | rocky | ol)
dnf install -y -q postgresql >& 2 || return 1
; ;
centos)
if [ [ " ${ VERSION_ID } " = ~ ^7 ] ] ; then
yum install -y postgresql >& 2 || return 1
else
dnf install -y -q postgresql >& 2 || return 1
fi
; ;
arch | manjaro | parch)
pacman -Sy --noconfirm postgresql >& 2 || return 1
; ;
opensuse-tumbleweed | opensuse-leap)
zypper -q install -y postgresql >& 2 || return 1
; ;
alpine)
apk add --no-cache postgresql-client >& 2 || return 1
; ;
*)
return 1
; ;
esac
command -v pg_dump > /dev/null 2>& 1 && command -v pg_restore > /dev/null 2>& 1
}
2025-12-27 15:03:33 -08:00
install_acme( ) {
echo -e " ${ green } Installing acme.sh for SSL certificate management... ${ plain } "
cd ~ || return 1
2026-05-04 04:20:24 -07:00
curl -s https://get.acme.sh | sh > /dev/null 2>& 1
2025-12-27 15:03:33 -08:00
if [ $? -ne 0 ] ; then
echo -e " ${ red } Failed to install acme.sh ${ plain } "
return 1
else
echo -e " ${ green } acme.sh installed successfully ${ plain } "
fi
return 0
}
setup_ssl_certificate( ) {
local domain = " $1 "
local server_ip = " $2 "
local existing_port = " $3 "
local existing_webBasePath = " $4 "
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
echo -e " ${ green } Setting up SSL certificate... ${ plain } "
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
# Check if acme.sh is installed
2026-05-04 04:20:24 -07:00
if ! command -v ~/.acme.sh/acme.sh & > /dev/null; then
2025-12-27 15:03:33 -08:00
install_acme
if [ $? -ne 0 ] ; then
echo -e " ${ yellow } Failed to install acme.sh, skipping SSL setup ${ plain } "
return 1
fi
fi
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
# Create certificate directory
local certPath = " /root/cert/ ${ domain } "
mkdir -p " $certPath "
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
# Issue certificate
echo -e " ${ green } Issuing SSL certificate for ${ domain } ... ${ plain } "
echo -e " ${ yellow } Note: Port 80 must be open and accessible from the internet ${ plain } "
2026-05-04 04:20:24 -07:00
~/.acme.sh/acme.sh --set-default-ca --server letsencrypt --force > /dev/null 2>& 1
2026-06-11 16:22:30 -07:00
~/.acme.sh/acme.sh --issue -d ${ domain } $( acme_listen_flag) --standalone --httpport 80 --force
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
if [ $? -ne 0 ] ; then
echo -e " ${ yellow } Failed to issue certificate for ${ domain } ${ plain } "
echo -e " ${ yellow } Please ensure port 80 is open and try again later with: x-ui ${ plain } "
2026-06-04 08:15:33 -07:00
rm -rf ~/.acme.sh/${ domain } ~/.acme.sh/${ domain } _ecc 2> /dev/null
2026-05-04 04:20:24 -07:00
rm -rf " $certPath " 2> /dev/null
2025-12-27 15:03:33 -08:00
return 1
fi
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
# Install certificate
2026-07-02 04:20:31 -07:00
~/.acme.sh/acme.sh --installcert --force -d ${ domain } \
2025-12-27 15:03:33 -08:00
--key-file /root/cert/${ domain } /privkey.pem \
--fullchain-file /root/cert/${ domain } /fullchain.pem \
2026-05-04 04:20:24 -07:00
--reloadcmd "systemctl restart x-ui" > /dev/null 2>& 1
2025-12-27 15:03:33 -08:00
if [ $? -ne 0 ] ; then
echo -e " ${ yellow } Failed to install certificate ${ plain } "
return 1
fi
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
# Enable auto-renew
2026-05-04 04:20:24 -07:00
~/.acme.sh/acme.sh --upgrade --auto-upgrade > /dev/null 2>& 1
2026-05-19 04:09:35 -07:00
# Secure permissions: private key readable only by owner
2026-05-04 04:20:24 -07:00
chmod 600 $certPath /privkey.pem 2> /dev/null
chmod 644 $certPath /fullchain.pem 2> /dev/null
2025-12-27 15:03:33 -08:00
# Set certificate for panel
local webCertFile = " /root/cert/ ${ domain } /fullchain.pem "
local webKeyFile = " /root/cert/ ${ domain } /privkey.pem "
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
if [ [ -f " $webCertFile " && -f " $webKeyFile " ] ] ; then
2026-05-04 04:20:24 -07:00
${ xui_folder } /x-ui cert -webCert " $webCertFile " -webCertKey " $webKeyFile " > /dev/null 2>& 1
2025-12-27 15:03:33 -08:00
echo -e " ${ green } SSL certificate installed and configured successfully! ${ plain } "
return 0
else
echo -e " ${ yellow } Certificate files not found ${ plain } "
return 1
fi
}
2026-01-04 20:28:02 -08:00
# Issue Let's Encrypt IP certificate with shortlived profile (~6 days validity)
# Requires acme.sh and port 80 open for HTTP-01 challenge
setup_ip_certificate( ) {
local ipv4 = " $1 "
2026-05-04 04:20:24 -07:00
local ipv6 = " $2 " # optional
2025-12-27 15:03:33 -08:00
2026-01-04 20:28:02 -08:00
echo -e " ${ green } Setting up Let's Encrypt IP certificate (shortlived profile)... ${ plain } "
echo -e " ${ yellow } Note: IP certificates are valid for ~6 days and will auto-renew. ${ plain } "
2026-01-11 06:28:43 -08:00
echo -e " ${ yellow } Default listener is port 80. If you choose another port, ensure external port 80 forwards to it. ${ plain } "
2025-12-27 15:03:33 -08:00
2026-01-04 20:28:02 -08:00
# Check for acme.sh
2026-05-04 04:20:24 -07:00
if ! command -v ~/.acme.sh/acme.sh & > /dev/null; then
2026-01-04 20:28:02 -08:00
install_acme
if [ $? -ne 0 ] ; then
echo -e " ${ red } Failed to install acme.sh ${ plain } "
return 1
fi
2025-12-27 15:03:33 -08:00
fi
2026-01-04 20:28:02 -08:00
# Validate IP address
if [ [ -z " $ipv4 " ] ] ; then
echo -e " ${ red } IPv4 address is required ${ plain } "
2025-12-27 15:03:33 -08:00
return 1
fi
2026-01-04 20:28:02 -08:00
if ! is_ipv4 " $ipv4 " ; then
echo -e " ${ red } Invalid IPv4 address: $ipv4 ${ plain } "
return 1
fi
# Create certificate directory
local certDir = "/root/cert/ip"
mkdir -p " $certDir "
# Build domain arguments
local domain_args = " -d ${ ipv4 } "
if [ [ -n " $ipv6 " ] ] && is_ipv6 " $ipv6 " ; then
domain_args = " ${ domain_args } -d ${ ipv6 } "
echo -e " ${ green } Including IPv6 address: ${ ipv6 } ${ plain } "
fi
2026-05-19 04:09:35 -07:00
# Set reload command for auto-renewal (add || true so it doesn't fail during first install)
2026-01-04 20:28:02 -08:00
local reloadCmd = "systemctl restart x-ui 2>/dev/null || rc-service x-ui restart 2>/dev/null || true"
2026-01-11 06:28:43 -08:00
# Choose port for HTTP-01 listener (default 80, prompt override)
local WebPort = ""
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
prompt_or_default WebPort "Port to use for ACME HTTP-01 listener (default 80): " "80" XUI_ACME_HTTP_PORT
2026-01-11 06:28:43 -08:00
WebPort = " ${ WebPort :- 80 } "
if ! [ [ " ${ WebPort } " = ~ ^[ 0-9] +$ ] ] || ( ( WebPort < 1 || WebPort > 65535) ) ; then
echo -e " ${ red } Invalid port provided. Falling back to 80. ${ plain } "
WebPort = 80
fi
echo -e " ${ green } Using port ${ WebPort } for standalone validation. ${ plain } "
if [ [ " ${ WebPort } " -ne 80 ] ] ; then
echo -e " ${ yellow } Reminder: Let's Encrypt still connects on port 80; forward external port 80 to ${ WebPort } . ${ plain } "
fi
# Ensure chosen port is available
while true; do
if is_port_in_use " ${ WebPort } " ; then
2026-05-19 04:09:35 -07:00
echo -e " ${ yellow } Port ${ WebPort } is in use. ${ plain } "
2026-01-11 06:28:43 -08:00
local alt_port = ""
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
echo -e " ${ red } Port ${ WebPort } is busy; cannot proceed in non-interactive mode. ${ plain } "
return 1
fi
2026-01-11 06:28:43 -08:00
read -rp "Enter another port for acme.sh standalone listener (leave empty to abort): " alt_port
alt_port = " ${ alt_port // / } "
if [ [ -z " ${ alt_port } " ] ] ; then
echo -e " ${ red } Port ${ WebPort } is busy; cannot proceed. ${ plain } "
return 1
fi
if ! [ [ " ${ alt_port } " = ~ ^[ 0-9] +$ ] ] || ( ( alt_port < 1 || alt_port > 65535) ) ; then
echo -e " ${ red } Invalid port provided. ${ plain } "
return 1
fi
WebPort = " ${ alt_port } "
continue
else
echo -e " ${ green } Port ${ WebPort } is free and ready for standalone validation. ${ plain } "
break
fi
done
2026-01-04 20:28:02 -08:00
# Issue certificate with shortlived profile
echo -e " ${ green } Issuing IP certificate for ${ ipv4 } ... ${ plain } "
2026-05-04 04:20:24 -07:00
~/.acme.sh/acme.sh --set-default-ca --server letsencrypt --force > /dev/null 2>& 1
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
[ [ -n " ${ XUI_ACME_EMAIL :- } " ] ] && ~/.acme.sh/acme.sh --register-account -m " ${ XUI_ACME_EMAIL } " > /dev/null 2>& 1
2026-05-04 04:20:24 -07:00
2026-01-04 20:28:02 -08:00
~/.acme.sh/acme.sh --issue \
${ domain_args } \
--standalone \
--server letsencrypt \
--certificate-profile shortlived \
--days 6 \
2026-01-11 06:28:43 -08:00
--httpport ${ WebPort } \
2026-01-04 20:28:02 -08:00
--force
if [ $? -ne 0 ] ; then
echo -e " ${ red } Failed to issue IP certificate ${ plain } "
2026-01-11 06:28:43 -08:00
echo -e " ${ yellow } Please ensure port ${ WebPort } is reachable (or forwarded from external port 80) ${ plain } "
2026-01-04 20:28:02 -08:00
# Cleanup acme.sh data for both IPv4 and IPv6 if specified
2026-06-04 08:15:33 -07:00
rm -rf ~/.acme.sh/${ ipv4 } ~/.acme.sh/${ ipv4 } _ecc 2> /dev/null
[ [ -n " $ipv6 " ] ] && rm -rf ~/.acme.sh/${ ipv6 } ~/.acme.sh/${ ipv6 } _ecc 2> /dev/null
2026-05-04 04:20:24 -07:00
rm -rf ${ certDir } 2> /dev/null
2026-01-04 20:28:02 -08:00
return 1
fi
echo -e " ${ green } Certificate issued successfully, installing... ${ plain } "
# Install certificate
# Note: acme.sh may report "Reload error" and exit non-zero if reloadcmd fails,
# but the cert files are still installed. We check for files instead of exit code.
2026-07-02 04:20:31 -07:00
~/.acme.sh/acme.sh --installcert --force -d ${ ipv4 } \
2026-01-04 20:28:02 -08:00
--key-file " ${ certDir } /privkey.pem " \
--fullchain-file " ${ certDir } /fullchain.pem " \
--reloadcmd " ${ reloadCmd } " 2>& 1 || true
# Verify certificate files exist (don't rely on exit code - reloadcmd failure causes non-zero)
if [ [ ! -f " ${ certDir } /fullchain.pem " || ! -f " ${ certDir } /privkey.pem " ] ] ; then
echo -e " ${ red } Certificate files not found after installation ${ plain } "
# Cleanup acme.sh data for both IPv4 and IPv6 if specified
2026-06-04 08:15:33 -07:00
rm -rf ~/.acme.sh/${ ipv4 } ~/.acme.sh/${ ipv4 } _ecc 2> /dev/null
[ [ -n " $ipv6 " ] ] && rm -rf ~/.acme.sh/${ ipv6 } ~/.acme.sh/${ ipv6 } _ecc 2> /dev/null
2026-05-04 04:20:24 -07:00
rm -rf ${ certDir } 2> /dev/null
2026-01-04 20:28:02 -08:00
return 1
fi
2026-05-04 04:20:24 -07:00
2026-01-04 20:28:02 -08:00
echo -e " ${ green } Certificate files installed successfully ${ plain } "
# Enable auto-upgrade for acme.sh (ensures cron job runs)
2026-05-04 04:20:24 -07:00
~/.acme.sh/acme.sh --upgrade --auto-upgrade > /dev/null 2>& 1
2026-01-04 20:28:02 -08:00
2026-05-19 04:09:35 -07:00
# Secure permissions: private key readable only by owner
2026-05-04 04:20:24 -07:00
chmod 600 ${ certDir } /privkey.pem 2> /dev/null
chmod 644 ${ certDir } /fullchain.pem 2> /dev/null
2026-01-04 20:28:02 -08:00
# Configure panel to use the certificate
echo -e " ${ green } Setting certificate paths for the panel... ${ plain } "
${ xui_folder } /x-ui cert -webCert " ${ certDir } /fullchain.pem " -webCertKey " ${ certDir } /privkey.pem "
2026-05-19 04:09:35 -07:00
2026-01-04 20:28:02 -08:00
if [ $? -ne 0 ] ; then
2026-05-19 04:09:35 -07:00
echo -e " ${ yellow } Warning: Could not set certificate paths automatically ${ plain } "
echo -e " ${ yellow } Certificate files are at: ${ plain } "
echo -e " Cert: ${ certDir } /fullchain.pem "
echo -e " Key: ${ certDir } /privkey.pem "
2026-01-04 20:28:02 -08:00
else
2026-05-19 04:09:35 -07:00
echo -e " ${ green } Certificate paths configured successfully ${ plain } "
2026-01-04 20:28:02 -08:00
fi
echo -e " ${ green } IP certificate installed and configured successfully! ${ plain } "
echo -e " ${ green } Certificate valid for ~6 days, auto-renews via acme.sh cron job. ${ plain } "
2026-05-19 04:09:35 -07:00
echo -e " ${ yellow } acme.sh will automatically renew and reload x-ui before expiry. ${ plain } "
2025-12-27 15:03:33 -08:00
return 0
}
# Comprehensive manual SSL certificate issuance via acme.sh
ssl_cert_issue( ) {
2026-01-02 18:57:19 -08:00
local existing_webBasePath = $( ${ xui_folder } /x-ui setting -show true | grep 'webBasePath:' | awk -F': ' '{print $2}' | tr -d '[:space:]' | sed 's#^/##' )
local existing_port = $( ${ xui_folder } /x-ui setting -show true | grep 'port:' | awk -F': ' '{print $2}' | tr -d '[:space:]' )
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
# check for acme.sh first
2026-05-04 04:20:24 -07:00
if ! command -v ~/.acme.sh/acme.sh & > /dev/null; then
2025-12-27 15:03:33 -08:00
echo "acme.sh could not be found. Installing now..."
cd ~ || return 1
curl -s https://get.acme.sh | sh
if [ $? -ne 0 ] ; then
echo -e " ${ red } Failed to install acme.sh ${ plain } "
return 1
else
echo -e " ${ green } acme.sh installed successfully ${ plain } "
fi
fi
# get the domain here, and we need to verify it
local domain = ""
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
domain = " ${ XUI_DOMAIN // / } "
if [ [ -z " $domain " ] ] || ! is_domain " $domain " ; then
echo -e " ${ red } XUI_SSL_MODE=domain requires a valid XUI_DOMAIN (got: ' ${ XUI_DOMAIN :- } '). ${ plain } "
return 1
2025-12-27 15:03:33 -08:00
fi
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
else
while true; do
read -rp "Please enter your domain name: " domain
domain = " ${ domain // / } " # Trim whitespace
2026-05-04 04:20:24 -07:00
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ -z " $domain " ] ] ; then
echo -e " ${ red } Domain name cannot be empty. Please try again. ${ plain } "
continue
fi
2026-05-04 04:20:24 -07:00
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if ! is_domain " $domain " ; then
echo -e " ${ red } Invalid domain format: ${ domain } . Please enter a valid domain name. ${ plain } "
continue
fi
break
done
fi
2025-12-27 15:03:33 -08:00
echo -e " ${ green } Your domain is: ${ domain } , checking it... ${ plain } "
2026-04-17 03:19:45 -07:00
SSL_ISSUED_DOMAIN = " ${ domain } "
2025-12-27 15:03:33 -08:00
2026-06-04 08:15:33 -07:00
# detect existing certificate and reuse it only if its files are actually
# present and non-empty. acme.sh stores ECC certs under ${domain}_ecc and RSA
# certs under ${domain}; a failed issuance can leave a domain entry in --list
# with no usable cert files, which must not be reused (it produces a 0-byte
# fullchain.pem). Broken partial state is cleaned up so issuance can proceed.
2026-04-17 03:19:45 -07:00
local cert_exists = 0
2026-05-04 04:20:24 -07:00
if ~/.acme.sh/acme.sh --list 2> /dev/null | awk '{print $1}' | grep -Fxq " ${ domain } " ; then
2026-06-04 08:15:33 -07:00
local acmeCertDir = ""
if [ [ -s ~/.acme.sh/${ domain } _ecc/fullchain.cer && -s ~/.acme.sh/${ domain } _ecc/${ domain } .key ] ] ; then
acmeCertDir = ~/.acme.sh/${ domain } _ecc
elif [ [ -s ~/.acme.sh/${ domain } /fullchain.cer && -s ~/.acme.sh/${ domain } /${ domain } .key ] ] ; then
acmeCertDir = ~/.acme.sh/${ domain }
fi
if [ [ -n " ${ acmeCertDir } " ] ] ; then
cert_exists = 1
local certInfo = $( ~/.acme.sh/acme.sh --list 2> /dev/null | grep -F " ${ domain } " )
echo -e " ${ yellow } Existing certificate found for ${ domain } , will reuse it. ${ plain } "
[ [ -n " ${ certInfo } " ] ] && echo " $certInfo "
else
echo -e " ${ yellow } Found incomplete acme.sh state for ${ domain } (no valid certificate files); cleaning it up and re-issuing. ${ plain } "
rm -rf ~/.acme.sh/${ domain } ~/.acme.sh/${ domain } _ecc
fi
fi
if [ [ ${ cert_exists } -eq 0 ] ] ; then
2025-12-27 15:03:33 -08:00
echo -e " ${ green } Your domain is ready for issuing certificates now... ${ plain } "
fi
# create a directory for the certificate
certPath = " /root/cert/ ${ domain } "
if [ ! -d " $certPath " ] ; then
mkdir -p " $certPath "
else
rm -rf " $certPath "
mkdir -p " $certPath "
fi
# get the port number for the standalone server
local WebPort = 80
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
prompt_or_default WebPort "Please choose which port to use (default is 80): " "80" XUI_ACME_HTTP_PORT
2026-07-06 03:44:37 -07:00
if [ [ -z ${ WebPort } ] ] ; then
WebPort = 80
elif [ [ ! ${ WebPort } = ~ ^[ 1-9] [ 0-9] *$ || ${ WebPort } -gt 65535 ] ] ; then
2025-12-27 15:03:33 -08:00
echo -e " ${ yellow } Your input ${ WebPort } is invalid, will use default port 80. ${ plain } "
WebPort = 80
fi
echo -e " ${ green } Will use port: ${ WebPort } to issue certificates. Please make sure this port is open. ${ plain } "
# Stop panel temporarily
echo -e " ${ yellow } Stopping panel temporarily... ${ plain } "
2026-05-04 04:20:24 -07:00
systemctl stop x-ui 2> /dev/null || rc-service x-ui stop 2> /dev/null
2025-12-27 15:03:33 -08:00
2026-04-17 03:19:45 -07:00
if [ [ ${ cert_exists } -eq 0 ] ] ; then
# issue the certificate
~/.acme.sh/acme.sh --set-default-ca --server letsencrypt --force
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
[ [ -n " ${ XUI_ACME_EMAIL :- } " ] ] && ~/.acme.sh/acme.sh --register-account -m " ${ XUI_ACME_EMAIL } " > /dev/null 2>& 1
2026-06-11 16:22:30 -07:00
~/.acme.sh/acme.sh --issue -d ${ domain } $( acme_listen_flag) --standalone --httpport ${ WebPort } --force
2026-04-17 03:19:45 -07:00
if [ $? -ne 0 ] ; then
echo -e " ${ red } Issuing certificate failed, please check logs. ${ plain } "
2026-06-04 08:15:33 -07:00
rm -rf ~/.acme.sh/${ domain } ~/.acme.sh/${ domain } _ecc
2026-05-04 04:20:24 -07:00
systemctl start x-ui 2> /dev/null || rc-service x-ui start 2> /dev/null
2026-04-17 03:19:45 -07:00
return 1
else
echo -e " ${ green } Issuing certificate succeeded, installing certificates... ${ plain } "
fi
2025-12-27 15:03:33 -08:00
else
2026-04-17 03:19:45 -07:00
echo -e " ${ green } Using existing certificate, installing certificates... ${ plain } "
2025-12-27 15:03:33 -08:00
fi
# Setup reload command
reloadCmd = "systemctl restart x-ui || rc-service x-ui restart"
echo -e " ${ green } Default --reloadcmd for ACME is: ${ yellow } systemctl restart x-ui || rc-service x-ui restart ${ plain } "
echo -e " ${ green } This command will run on every certificate issue and renew. ${ plain } "
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
setReloadcmd = "n"
else
read -rp "Would you like to modify --reloadcmd for ACME? (y/n): " setReloadcmd
fi
2025-12-27 15:03:33 -08:00
if [ [ " $setReloadcmd " = = "y" || " $setReloadcmd " = = "Y" ] ] ; then
echo -e " \n ${ green } \t1. ${ plain } Preset: systemctl reload nginx ; systemctl restart x-ui "
echo -e " ${ green } \t2. ${ plain } Input your own command "
echo -e " ${ green } \t0. ${ plain } Keep default reloadcmd "
read -rp "Choose an option: " choice
case " $choice " in
2026-05-04 04:20:24 -07:00
1)
echo -e " ${ green } Reloadcmd is: systemctl reload nginx ; systemctl restart x-ui ${ plain } "
reloadCmd = "systemctl reload nginx ; systemctl restart x-ui"
; ;
2)
echo -e " ${ yellow } It's recommended to put x-ui restart at the end ${ plain } "
read -rp "Please enter your custom reloadcmd: " reloadCmd
echo -e " ${ green } Reloadcmd is: ${ reloadCmd } ${ plain } "
; ;
*)
echo -e " ${ green } Keeping default reloadcmd ${ plain } "
; ;
2025-12-27 15:03:33 -08:00
esac
fi
# install the certificate
2026-04-17 03:19:45 -07:00
local installOutput = ""
2026-07-02 04:20:31 -07:00
installOutput = $( ~/.acme.sh/acme.sh --installcert --force -d ${ domain } \
2025-12-27 15:03:33 -08:00
--key-file /root/cert/${ domain } /privkey.pem \
2026-04-17 03:19:45 -07:00
--fullchain-file /root/cert/${ domain } /fullchain.pem --reloadcmd " ${ reloadCmd } " 2>& 1)
local installRc = $?
echo " ${ installOutput } "
2025-12-27 15:03:33 -08:00
2026-04-17 03:19:45 -07:00
local installWroteFiles = 0
if echo " ${ installOutput } " | grep -q "Installing key to:" && echo " ${ installOutput } " | grep -q "Installing full chain to:" ; then
installWroteFiles = 1
fi
2026-05-04 04:20:24 -07:00
if [ [ -f " /root/cert/ ${ domain } /privkey.pem " && -f " /root/cert/ ${ domain } /fullchain.pem " && ( ${ installRc } -eq 0 || ${ installWroteFiles } -eq 1) ] ] ; then
2026-04-17 03:19:45 -07:00
echo -e " ${ green } Installing certificate succeeded, enabling auto renew... ${ plain } "
else
2025-12-27 15:03:33 -08:00
echo -e " ${ red } Installing certificate failed, exiting. ${ plain } "
2026-04-17 03:19:45 -07:00
if [ [ ${ cert_exists } -eq 0 ] ] ; then
2026-06-04 08:15:33 -07:00
rm -rf ~/.acme.sh/${ domain } ~/.acme.sh/${ domain } _ecc
2026-04-17 03:19:45 -07:00
fi
2026-05-04 04:20:24 -07:00
systemctl start x-ui 2> /dev/null || rc-service x-ui start 2> /dev/null
2025-12-27 15:03:33 -08:00
return 1
fi
# enable auto-renew
~/.acme.sh/acme.sh --upgrade --auto-upgrade
if [ $? -ne 0 ] ; then
echo -e " ${ yellow } Auto renew setup had issues, certificate details: ${ plain } "
ls -lah /root/cert/${ domain } /
2026-05-19 04:09:35 -07:00
# Secure permissions: private key readable only by owner
chmod 600 $certPath /privkey.pem 2> /dev/null
chmod 644 $certPath /fullchain.pem 2> /dev/null
2025-12-27 15:03:33 -08:00
else
echo -e " ${ green } Auto renew succeeded, certificate details: ${ plain } "
ls -lah /root/cert/${ domain } /
2026-05-19 04:09:35 -07:00
# Secure permissions: private key readable only by owner
chmod 600 $certPath /privkey.pem 2> /dev/null
chmod 644 $certPath /fullchain.pem 2> /dev/null
2025-12-27 15:03:33 -08:00
fi
2026-05-19 04:09:35 -07:00
# start panel
2026-05-04 04:20:24 -07:00
systemctl start x-ui 2> /dev/null || rc-service x-ui start 2> /dev/null
2025-12-27 15:03:33 -08:00
# Prompt user to set panel paths after successful certificate installation
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
setPanel = "y"
else
read -rp "Would you like to set this certificate for the panel? (y/n): " setPanel
fi
2025-12-27 15:03:33 -08:00
if [ [ " $setPanel " = = "y" || " $setPanel " = = "Y" ] ] ; then
local webCertFile = " /root/cert/ ${ domain } /fullchain.pem "
local webKeyFile = " /root/cert/ ${ domain } /privkey.pem "
if [ [ -f " $webCertFile " && -f " $webKeyFile " ] ] ; then
2026-01-02 18:57:19 -08:00
${ xui_folder } /x-ui cert -webCert " $webCertFile " -webCertKey " $webKeyFile "
2025-12-27 15:03:33 -08:00
echo -e " ${ green } Certificate paths set for the panel ${ plain } "
echo -e " ${ green } Certificate File: $webCertFile ${ plain } "
echo -e " ${ green } Private Key File: $webKeyFile ${ plain } "
echo ""
echo -e " ${ green } Access URL: https:// ${ domain } : ${ existing_port } / ${ existing_webBasePath } ${ plain } "
echo -e " ${ yellow } Panel will restart to apply SSL certificate... ${ plain } "
2026-05-04 04:20:24 -07:00
systemctl restart x-ui 2> /dev/null || rc-service x-ui restart 2> /dev/null
2025-12-27 15:03:33 -08:00
else
echo -e " ${ red } Error: Certificate or private key file not found for domain: $domain . ${ plain } "
fi
else
echo -e " ${ yellow } Skipping panel path setting. ${ plain } "
fi
2026-05-04 04:20:24 -07:00
2025-12-27 15:03:33 -08:00
return 0
}
2026-05-19 04:09:35 -07:00
# Reusable interactive SSL setup (domain or IP)
# Sets global `SSL_HOST` to the chosen domain/IP for Access URL usage
2025-12-27 15:03:33 -08:00
prompt_and_setup_ssl( ) {
local panel_port = " $1 "
2026-05-19 04:09:35 -07:00
local web_base_path = " $2 "
2025-12-27 15:03:33 -08:00
local server_ip = " $3 "
local ssl_choice = ""
2026-05-19 04:09:35 -07:00
SSL_SCHEME = "https"
2025-12-27 15:03:33 -08:00
echo -e " ${ yellow } Choose SSL certificate setup method: ${ plain } "
2026-01-04 20:28:02 -08:00
echo -e " ${ green } 1. ${ plain } Let's Encrypt for Domain (90-day validity, auto-renews) "
echo -e " ${ green } 2. ${ plain } Let's Encrypt for IP Address (6-day validity, auto-renews) "
2026-01-21 07:47:36 -08:00
echo -e " ${ green } 3. ${ plain } Custom SSL Certificate (Path to existing files) "
2026-05-19 04:09:35 -07:00
echo -e " ${ green } 4. ${ plain } Skip SSL (advanced — behind reverse proxy / SSH tunnel only) "
2026-01-21 07:47:36 -08:00
echo -e " ${ blue } Note: ${ plain } Options 1 & 2 require port 80 open. Option 3 requires manual paths. "
2026-05-19 04:09:35 -07:00
echo -e " ${ blue } Note: ${ plain } Option 4 serves the panel over plain HTTP — only safe behind nginx/Caddy or an SSH tunnel. "
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
case " ${ XUI_SSL_MODE :- none } " in
domain) ssl_choice = "1" ; ;
ip) ssl_choice = "2" ; ;
none | "" ) ssl_choice = "4" ; ;
*)
echo -e " ${ yellow } Unknown XUI_SSL_MODE=' ${ XUI_SSL_MODE } ', defaulting to none (HTTP). ${ plain } "
ssl_choice = "4"
; ;
esac
else
read -rp "Choose an option (default 2 for IP): " ssl_choice
ssl_choice = " ${ ssl_choice // / } " # Trim whitespace
2026-05-04 04:20:24 -07:00
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
# Default to 2 (IP cert) if input is empty or invalid (not 1, 3 or 4)
if [ [ " $ssl_choice " != "1" && " $ssl_choice " != "3" && " $ssl_choice " != "4" ] ] ; then
ssl_choice = "2"
fi
2025-12-27 15:03:33 -08:00
fi
case " $ssl_choice " in
2026-05-04 04:20:24 -07:00
1)
# User chose Let's Encrypt domain option
echo -e " ${ green } Using Let's Encrypt for domain certificate... ${ plain } "
if ssl_cert_issue; then
local cert_domain = " ${ SSL_ISSUED_DOMAIN } "
if [ [ -z " ${ cert_domain } " ] ] ; then
cert_domain = $( ~/.acme.sh/acme.sh --list 2> /dev/null | tail -1 | awk '{print $1}' )
fi
2026-04-17 03:19:45 -07:00
2026-05-04 04:20:24 -07:00
if [ [ -n " ${ cert_domain } " ] ] ; then
SSL_HOST = " ${ cert_domain } "
echo -e " ${ green } ✓ SSL certificate configured successfully with domain: ${ cert_domain } ${ plain } "
else
echo -e " ${ yellow } SSL setup may have completed, but domain extraction failed ${ plain } "
SSL_HOST = " ${ server_ip } "
fi
2026-04-17 03:19:45 -07:00
else
2026-05-04 04:20:24 -07:00
echo -e " ${ red } SSL certificate setup failed for domain mode. ${ plain } "
2026-04-17 03:19:45 -07:00
SSL_HOST = " ${ server_ip } "
fi
2026-05-04 04:20:24 -07:00
; ;
2)
# User chose Let's Encrypt IP certificate option
echo -e " ${ green } Using Let's Encrypt for IP certificate (shortlived profile)... ${ plain } "
2026-01-21 07:47:36 -08:00
2026-07-11 13:35:31 -07:00
# Confirm the auto-detected IP before issuing for it: with asymmetric
# routing / multi-WAN the echo services can return a transit address.
if [ [ " $NONINTERACTIVE " != "1" ] ] ; then
local ip_confirm = ""
read -rp " Is ${ server_ip } the correct incoming public IPv4 address for this server? [Default y]: " ip_confirm
if [ [ -n " $ip_confirm " && " $ip_confirm " != "y" && " $ip_confirm " != "Y" ] ] ; then
server_ip = ""
while [ [ -z " $server_ip " ] ] ; do
read -rp "Please enter your server's public IPv4 address: " server_ip
server_ip = " ${ server_ip // / } "
if [ [ ! " $server_ip " = ~ ^[ 0-9] +\. [ 0-9] +\. [ 0-9] +\. [ 0-9] +$ ] ] ; then
echo -e " ${ red } Invalid IPv4 address. Please try again. ${ plain } "
server_ip = ""
fi
done
fi
fi
2026-05-04 04:20:24 -07:00
# Ask for optional IPv6
local ipv6_addr = ""
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
prompt_or_default ipv6_addr "Do you have an IPv6 address to include? (leave empty to skip): " "" XUI_SSL_IPV6
2026-05-04 04:20:24 -07:00
ipv6_addr = " ${ ipv6_addr // / } " # Trim whitespace
2026-01-21 07:47:36 -08:00
2026-05-04 04:20:24 -07:00
# Stop panel if running (port 80 needed)
if [ [ $release = = "alpine" ] ] ; then
rc-service x-ui stop > /dev/null 2>& 1
2026-01-21 07:47:36 -08:00
else
2026-05-04 04:20:24 -07:00
systemctl stop x-ui > /dev/null 2>& 1
2026-01-21 07:47:36 -08:00
fi
2026-05-04 04:20:24 -07:00
setup_ip_certificate " ${ server_ip } " " ${ ipv6_addr } "
if [ $? -eq 0 ] ; then
SSL_HOST = " ${ server_ip } "
echo -e " ${ green } ✓ Let's Encrypt IP certificate configured successfully ${ plain } "
2026-01-21 07:47:36 -08:00
else
2026-05-04 04:20:24 -07:00
echo -e " ${ red } ✗ IP certificate setup failed. Please check port 80 is open. ${ plain } "
SSL_HOST = " ${ server_ip } "
2026-01-21 07:47:36 -08:00
fi
2026-05-04 04:20:24 -07:00
; ;
3)
# User chose Custom Paths (User Provided) option
echo -e " ${ green } Using custom existing certificate... ${ plain } "
local custom_cert = ""
local custom_key = ""
local custom_domain = ""
2026-01-21 07:47:36 -08:00
2026-05-04 04:20:24 -07:00
# 3.1 Request Domain to compose Panel URL later
read -rp "Please enter domain name certificate issued for: " custom_domain
custom_domain = " ${ custom_domain // / } " # Remove spaces
# 3.2 Loop for Certificate Path
while true; do
read -rp "Input certificate path (keywords: .crt / fullchain): " custom_cert
# Strip quotes if present
custom_cert = $( echo " $custom_cert " | tr -d '"' | tr -d "'" )
if [ [ -f " $custom_cert " && -r " $custom_cert " && -s " $custom_cert " ] ] ; then
break
elif [ [ ! -f " $custom_cert " ] ] ; then
echo -e " ${ red } Error: File does not exist! Try again. ${ plain } "
elif [ [ ! -r " $custom_cert " ] ] ; then
echo -e " ${ red } Error: File exists but is not readable (check permissions)! ${ plain } "
else
echo -e " ${ red } Error: File is empty! ${ plain } "
fi
done
# 3.3 Loop for Private Key Path
while true; do
read -rp "Input private key path (keywords: .key / privatekey): " custom_key
# Strip quotes if present
custom_key = $( echo " $custom_key " | tr -d '"' | tr -d "'" )
if [ [ -f " $custom_key " && -r " $custom_key " && -s " $custom_key " ] ] ; then
break
elif [ [ ! -f " $custom_key " ] ] ; then
echo -e " ${ red } Error: File does not exist! Try again. ${ plain } "
elif [ [ ! -r " $custom_key " ] ] ; then
echo -e " ${ red } Error: File exists but is not readable (check permissions)! ${ plain } "
else
echo -e " ${ red } Error: File is empty! ${ plain } "
fi
done
# 3.4 Apply Settings via x-ui binary
${ xui_folder } /x-ui cert -webCert " $custom_cert " -webCertKey " $custom_key " > /dev/null 2>& 1
# Set SSL_HOST for composing Panel URL
if [ [ -n " $custom_domain " ] ] ; then
SSL_HOST = " $custom_domain "
else
SSL_HOST = " ${ server_ip } "
fi
echo -e " ${ green } ✓ Custom certificate paths applied. ${ plain } "
echo -e " ${ yellow } Note: You are responsible for renewing these files externally. ${ plain } "
systemctl restart x-ui > /dev/null 2>& 1 || rc-service x-ui restart > /dev/null 2>& 1
; ;
2026-05-19 04:09:35 -07:00
4)
echo ""
echo -e " ${ red } ⚠ Panel will be installed WITHOUT SSL/TLS. ${ plain } "
echo -e " ${ yellow } Login credentials and cookies will travel as plain HTTP. ${ plain } "
echo -e " ${ yellow } Only safe when: ${ plain } "
echo -e " ${ yellow } • A reverse proxy (nginx, Caddy, Traefik) terminates TLS for you, or ${ plain } "
echo -e " ${ yellow } • You access the panel exclusively via SSH tunnel ${ plain } "
echo ""
SSL_SCHEME = "http"
SSL_HOST = " ${ server_ip } "
local bind_local = ""
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
# Cloud images must stay reachable on their public interface.
bind_local = "n"
else
read -rp "Bind the panel to 127.0.0.1 only? (recommended — forces SSH tunnel / reverse-proxy access) [y/N]: " bind_local
fi
2026-05-19 04:09:35 -07:00
if [ [ " $bind_local " = = "y" || " $bind_local " = = "Y" ] ] ; then
${ xui_folder } /x-ui setting -listenIP "127.0.0.1" > /dev/null 2>& 1
SSL_HOST = "127.0.0.1"
echo -e " ${ green } ✓ Panel bound to 127.0.0.1 only. It is now unreachable from the public internet. ${ plain } "
echo ""
echo -e " ${ green } SSH Port Forwarding — open the panel from your local machine via: ${ plain } "
echo -e " Standard SSH command:"
echo -e " ${ yellow } ssh -L 2222:127.0.0.1: ${ panel_port } root@ ${ server_ip } ${ plain } "
echo -e " If using an SSH key:"
echo -e " ${ yellow } ssh -i <sshkeypath> -L 2222:127.0.0.1: ${ panel_port } root@ ${ server_ip } ${ plain } "
echo -e " Then open in your browser:"
echo -e " ${ yellow } http://localhost:2222/ ${ web_base_path } ${ plain } "
echo ""
echo -e " ${ yellow } Alternative: point a reverse proxy (nginx/Caddy) at 127.0.0.1: ${ panel_port } and let it terminate TLS. ${ plain } "
else
echo -e " ${ yellow } Panel will listen on all interfaces over plain HTTP. Make sure something else is terminating TLS in front of it. ${ plain } "
fi
systemctl restart x-ui > /dev/null 2>& 1 || rc-service x-ui restart > /dev/null 2>& 1
echo -e " ${ green } ✓ SSL setup skipped. ${ plain } "
; ;
2026-05-04 04:20:24 -07:00
*)
echo -e " ${ red } Invalid option. Skipping SSL setup. ${ plain } "
2026-01-21 07:47:36 -08:00
SSL_HOST = " ${ server_ip } "
2026-05-04 04:20:24 -07:00
; ;
2025-12-27 15:03:33 -08:00
esac
}
2026-05-19 04:09:35 -07:00
config_after_install( ) {
local existing_hasDefaultCredential = $( ${ xui_folder } /x-ui setting -show true | grep -Eo 'hasDefaultCredential: .+' | awk '{print $2}' )
Feat/multi inbound clients (#4469)
* feat(clients): add shadow tables for first-class client promotion
Introduces three new GORM-backed tables (clients, client_inbounds,
inbound_fallback_children) and a populate-only seeder that backfills
them from each inbound's existing settings.clients JSON. Duplicate
emails across inbounds auto-merge under one client row, with each
field conflict logged. Existing services are unchanged and continue
reading from settings.clients — this commit is groundwork only.
* feat(clients): make clients+client_inbounds the runtime source of truth
Adds ClientService.SyncInbound that reconciles the new tables from
each inbound's clients list whenever existing service paths mutate
settings.clients. Wires it into AddInbound, UpdateInbound,
AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, DelDepletedClients, autoRenewClients, and
the timestamp-backfill path in adjustTraffics, plus DetachInbound
on DelInbound.
GetXrayConfig now builds settings.clients from the new tables before
writing config.json, and getInboundsBySubId joins through them
instead of JSON_EACH on settings JSON. Live Xray config and
subscription endpoints are now driven by the relational view;
settings.clients JSON stays in step as a side effect of every write.
* feat(clients): add top-level Clients tab and CRUD API
Adds /panel/api/clients endpoints (list, get, add, update, del,
attach, detach) backed by ClientService methods that orchestrate
the per-inbound Add/Update/Del flows so a single client row is
created once and attached to many inbounds in one operation.
The frontend gains a dedicated Clients page (frontend/clients.html
+ src/pages/clients/) with an AntD table, multi-inbound attach
modal, and full CRUD. Axios interceptor learns to honour
Content-Type: application/json so the JSON endpoints work
alongside the legacy form-encoded ones.
The legacy per-inbound client modal stays untouched in this PR —
both flows now write to the same source of truth.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): add Port-with-Fallback inbound type
Adds a new "portfallback" protocol that emits as a VLESS-TLS inbound
under the hood but is paired with a sidecar table of child inbounds.
Panel auto-builds settings.fallbacks at Xray-config-gen time from the
sidecar — each child's listen+port becomes the fallback dest, with
SNI/ALPN/path/xver match criteria pulled from the row. No more typing
loopback ports by hand or keeping settings.fallbacks in sync.
Backend: new FallbackService (Get/SetChildren, BuildFallbacksJSON);
two new routes (GET/POST /panel/api/inbounds/:id/fallbackChildren);
xray.GetXrayConfig injects fallbacks for PortFallback inbounds; the
inbound model emits protocol="vless" so Xray accepts the config.
Frontend: PORTFALLBACK joins the protocol dropdown; selecting it
shows the standard VLESS controls plus a Fallback Children table
(inbound picker + per-row SNI/ALPN/path/xver). Children are loaded
on edit and replaced atomically on save.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reset Traffic, QR Code, Info actions + Online/Remaining columns
The Clients page table gains:
- Online column — green/grey tag driven by /panel/api/inbounds/onlines,
polled every 10s.
- Remaining column — bytes-remaining tag, coloured green/orange/red
against quota, purple infinity when unlimited.
- Action icons per row: QR, Info, Reset traffic, Edit, Delete.
ClientInfoModal shows the full client detail (uuid/password/auth,
traffic ↑/↓ + remaining + all-time, expiry absolute + relative,
attached inbounds chip list, online + last-online).
ClientQrModal fetches links for the client's subId via
/panel/api/inbounds/getSubLinks/:subId and renders each one through
the existing QrPanel component.
Reset Traffic confirms then calls the existing per-inbound endpoint
on the client's first attached inbound (the traffic row is keyed on
email globally, so any attached inbound resets the shared counter).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): expose Attached inbounds in edit mode
The multi-select was gated on add-only, so editing a client had no way
to change which inbounds it belonged to. The picker now shows in both
modes, and on submit the modal diffs the picked set against the
original attachedIds — additions go through the /attach endpoint,
removals through /detach, both after the field update lands so the
new attachments get the latest values.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): unbreak template parsing + stale i18n keys
- InboundFormModal: split the multi-line help string in the
PortFallback section onto one line — Vue's template parser was
bailing on Unterminated string constant because a single-quoted
literal spanned two lines inside a {{ }} interpolation.
- ClientInfoModal: t('disable') was missing at the root level, so
vue-i18n returned the key path literally. Use t('disabled') which
exists.
- Linter cleanup elsewhere: pages.client.* references renamed to
pages.clients.* to match the merged i18n block; whitespace
normalisation in a few unrelated Vue templates.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* 1
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(traffic): drop all-time traffic tracking
Removes the AllTime field from Inbound and ClientTraffic and migrates
existing DBs by dropping the all_time columns on startup. The counter
duplicated up+down without adding signal, and the per-event accumulator
ran on every traffic write.
Frontend: drop the All-time column from the inbound list and the
client-row table, the All-time row from the client info modal, and the
All-Time Total Usage tile from the inbounds summary card. The
allTimeTraffic/allTimeTrafficUsage i18n keys are removed across every
locale.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): mobile cards, multi-select, bulk add
Adds the same row-card layout the inbounds page uses on mobile: the
table is suppressed under the mobile breakpoint and each client renders
as a compact card with a status dot, email, Info button, Enable switch,
and overflow menu. All the per-client detail (traffic, remaining,
expiry, attached inbounds, flow, created/updated, URL, subscription)
opens through the existing info modal.
Multi-select with bulk delete wires AntD row-selection on desktop and
a per-card checkbox on mobile; a Delete (N) button appears in the
toolbar when anything is selected.
Bulk add reuses the five email-generation modes from the inbound bulk
modal but takes a multi-inbound picker so one bulk run can attach to
several inbounds at once. Submits client-by-client through the
existing /panel/api/clients/add endpoint.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): remove legacy per-inbound client UI
Now that clients live as first-class rows attached to one or many
inbounds, the per-inbound client UI on the inbounds page is dead
weight — every client action either has a global equivalent on the
Clients page or makes no sense in a many-to-many world.
Deletes ClientFormModal, ClientBulkModal, CopyClientsModal, and
ClientRowTable from inbounds/. Strips the matching emits, refs,
handlers, and dropdown menu items from InboundList and InboundsPage,
and removes the dead mobile expand-chevron state and the desktop
expanded-row plumbing that drove the inline client table.
The InboundFormModal Clients tab still works in add-mode (one inline
client at inbound creation) — that flow goes through ClientService.
SyncInbound on save and remains useful.
Fixes a stray "</a-dropdown>" left over by an earlier toolbar edit
in ClientsPage that broke the template parser.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Delete depleted action
Mirrors the legacy delDepletedClients action that lived under the
inbounds page, but as a first-class /panel/api/clients/delDepleted
endpoint backed by ClientService. The new path goes through
ClientService.Delete for each depleted email, so the new clients +
client_inbounds + xray_client_traffic tables stay consistent.
Adds a danger-styled toolbar button on the Clients page (next to
Reset all client traffic) with a confirm dialog and a toast
reporting the deleted count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): move every client-shaped endpoint off /inbounds onto /clients
After the multi-inbound client migration, client state belongs to the
client API surface, not the inbound one. Twelve routes that were
crammed under /panel/api/inbounds/* now live where they belong, under
/panel/api/clients/*.
Moved (route, handler, doc):
POST /clientIps/:email
POST /clearClientIps/:email
POST /onlines
POST /lastOnline
POST /updateClientTraffic/:email
POST /resetAllClientTraffics/:id
POST /delDepletedClients/:id
POST /:id/resetClientTraffic/:email
GET /getClientTraffics/:email
GET /getClientTrafficsById/:id
GET /getSubLinks/:subId
GET /getClientLinks/:id/:email
Their /clients/* counterparts are:
POST /clients/clientIps/:email
POST /clients/clearClientIps/:email
POST /clients/onlines
POST /clients/lastOnline
POST /clients/updateTraffic/:email
POST /clients/resetTraffic/:email (email-only, fans out)
GET /clients/traffic/:email
GET /clients/traffic/byId/:id
GET /clients/subLinks/:subId
GET /clients/links/:id/:email
per-inbound resetAllClientTraffics and delDepletedClients are dropped
entirely — the Clients page already exposes global Reset All Traffic
and Delete depleted actions, and per-inbound resets are meaningless
once a client can be attached to many inbounds.
ClientService.ResetTrafficByEmail is the new email-only reset path:
it looks up every inbound the client is attached to and pushes the
counter reset + Xray re-add through inboundService.ResetClientTraffic
for each one, so depleted users come back online instantly.
Frontend callers (ClientsPage, useClients, ClientQrModal,
ClientInfoModal, InboundInfoModal, InboundsPage, useInbounds) all
switched to the new paths. The Inbounds page drops its per-inbound
"Reset client traffic" and "Delete depleted clients" dropdown items —
users do those at the client level now. api-docs is rebuilt to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): switch tgbot + ldap callers to ClientService
Adds two thin helpers to ClientService (CreateOne, DetachByEmail) and
rewrites tgbot.SubmitAddClient and ldap_sync_job to call ClientService
directly. Removes the JSON-blob payloads (BuildJSONForProtocol output for
add, clientsToJSON/clientToJSON helpers) that callers previously fed to
InboundService.AddInboundClient/DelInboundClient.
ldap_sync_job.batchSetEnable now loops InboundService.SetClientEnableByEmail
per email instead of trying to coerce AddInboundClient into doing the
update — the old path would have failed duplicate-email validation for
existing clients anyway.
The legacy InboundService.AddInboundClient/UpdateInboundClient/
DelInboundClient methods stay in place; they are now only used internally
by ClientService Create/Update/Delete/Attach. Inlining + deleting them
follows in a separate commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): move all client mutation methods to ClientService
Moves the client mutation surface out of InboundService and into
ClientService. These methods all operate on a single client (identity
fields, traffic limits, expiry, ip limit, enable state, telegram tg id)
and didn't belong on the inbound aggregate.
Moved (12 methods): AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, checkEmailsExistForClients, SetClientTelegramUserID,
checkIsEnabledByEmail, ToggleClientEnableByEmail, SetClientEnableByEmail,
ResetClientIpLimitByEmail, ResetClientExpiryTimeByEmail,
ResetClientTrafficLimitByEmail.
Each method now takes an explicit *InboundService for the helpers that
legitimately stay on InboundService (GetInbound, GetClients, runtimeFor,
AddClientStat / UpdateClientStat / DelClientStat, DelClientIPs /
UpdateClientIPs, emailUsedByOtherInbounds, getAllEmailSubIDs,
GetClientInboundByEmail / GetClientInboundByTrafficID,
GetClientTrafficByEmail).
Stays on InboundService: ResetClientTrafficByEmail and
ResetClientTraffic(id, email) — these mutate xray_client_traffic rows,
not client identity, so they're inbound-side bookkeeping.
Callers updated: tgbot (6 calls), ldap_sync_job (1 call),
InboundService internal (writeBackClientSubID, CopyInboundClients,
AddInbound's email-uniqueness check), ClientService Create/Update/
Delete/Attach/Detach.
Also removes a dead resetAllClientTraffics controller handler whose
route was already gone after the previous /clients API migration.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): finish migrating to ClientService + tidy IP routes
Two related cleanups in the new /clients surface:
1. Move ResetAllClientTraffics (bulk-reset of xray_client_traffic +
last_traffic_reset_time, with node-runtime propagation) from
InboundService to ClientService. PeriodicTrafficResetJob now holds
a clientService and calls
j.clientService.ResetAllClientTraffics(&j.inboundService, id).
The last client-mutation method on InboundService is gone.
2. Shorten redundantly-named routes/handlers under /panel/api/clients:
- /clientIps/:email -> /ips/:email (handler getIps)
- /clearClientIps/:email -> /clearIps/:email (handler clearIps)
The "client" prefix was redundant inside the clients namespace.
Frontend (InboundInfoModal) and api-docs updated to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds,clients): clean up inbound modal + enrich client modal
Inbound modal rework (InboundFormModal.vue + inbound.js):
- Drop the embedded Client subform in the Protocol tab. Multi-inbound
clients are managed exclusively from the Clients page now; a fresh
inbound is created with zero clients (settings constructors default
to []) and the user attaches clients afterwards.
- Hide the Protocol tab entirely when it has nothing to render
(VMESS, Trojan without fallbacks, Hysteria). Auto-switches active
tab to Basic when the tab disappears while focused.
- Move the Security section (Security selector + TLS block with certs
and ECH + Reality block) out of the Stream tab into its own
Security tab, sharing the canEnableStream gate.
Client modal additions (ClientFormModal.vue + ClientBulkAddModal.vue):
- Flow select (xtls-rprx-vision / -udp443) appears only when the
panel actually has a Vision-capable inbound (VLESS or PortFallback
on TCP with TLS or Reality). Hidden otherwise, and cleared when
it disappears.
- IP Limit input is disabled when the panel-level ipLimitEnable
setting is off, fetched into useClients alongside subSettings and
threaded through ClientsPage to both modals.
- Edit modal now shows an "IP Log" section listing IPs that have
connected with the client's credentials, with refresh and clear
buttons (calls the renamed /panel/api/clients/ips and /clearIps
endpoints).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): drop manual Fallbacks UI from inbound modal
The PortFallback protocol type now covers the common
VLESS-master-plus-children case with auto-wired dests, so the manual
Fallbacks editor (showFallbacks block in the Protocol tab) is mostly
redundant. Removed:
- the v-if="showFallbacks" template block (SNI/ALPN/Path/dest/PROXY rows)
- the showFallbacks computed
- the addFallback / delFallback helpers
- the .fallbacks-header / .fallbacks-title styles
- the showFallbacks gate from hasProtocolTabContent (so Trojan-over-TCP
no longer shows an empty Protocol tab)
Power users who need a non-inbound fallback dest (nginx, static site)
can still author settings.fallbacks via the Advanced JSON tab.
* feat(clients,inbounds): move search/filter to Clients page + small fixes
Search/filter relocation:
- Remove the search/filter toolbar (search switch + filter radio +
protocol/node selects + the visibleInbounds projection +
inboundsFilterState localStorage + filter CSS + the SearchOutlined/
FilterOutlined/ObjectUtil/Inbound imports it required) from
InboundList. The filters were all client-oriented buckets bolted
onto the inbound row.
- Add a search/filter toolbar to ClientsPage with the same shape:
switch between deep-text search and bucket filter (active /
deactive / depleted / expiring / online) + protocol filter that
matches clients attached to at least one inbound with the chosen
protocol. State persists in clientsFilterState localStorage.
filteredClients drives both the desktop table and the mobile card
list, and select-all / allSelected / someSelected only span the
visible subset.
- useClients now also fetches expireDiff and trafficDiff from
/panel/setting/defaultSettings (used to detect the expiring
bucket); ClientsPage threads them into the client-bucket helper.
Loose fixes folded in:
- Add Client: email field is auto-filled with a random handle on
open, matching uuid/subId/password/auth.
- Inbound clone: parse and reuse the source settings JSON (with
clients reset to []) instead of building a fresh defaulted
Settings, so VLESS Encryption/Decryption and other non-client
fields survive the clone.
- en-US.json: add the ipLog string used by the edit-client modal.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reverse tag field for VLESS-attached clients
Mirrors the Flow field's pattern: a Reverse tag input appears in the
Add/Edit Client modal whenever at least one selected inbound is VLESS
or PortFallback. The value rides over the wire as
client.reverse = { tag: '...' } so it lands directly in model.Client's
*ClientReverse field; an empty value omits the reverse key entirely.
On edit the field is hydrated from props.client.reverse?.tag, and the
showReverseTag watcher clears the field if the user drops the last
VLESS-like inbound from the selection.
* fix(xray): emit only protocol-relevant fields per client entry
The Xray config synthesizer was writing every identifier field (id,
password, flow, auth, security/method, reverse) on every client entry
regardless of the inbound's protocol. Xray ignores unknown fields, so
the config worked, but it diverged from the spec and leaked secrets
across protocols when one client was attached to multiple inbounds —
a VLESS inbound's generated config carried the same client's Trojan
password and Hysteria auth alongside its uuid.
Switch on inbound.Protocol when building each entry:
- VLESS / PortFallback: id, flow, reverse
- VMess: id, security
- Trojan: password, flow
- Shadowsocks: password, method
- Hysteria / Hysteria2: auth
email is emitted for every protocol.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): restore auto-disable kick under new schema
disableInvalidClients still resolved (inbound_tag, email) pairs via
JSON_EACH(inbounds.settings.clients), which is empty after migrating
to the clients + client_inbounds tables. Result: xrayApi.RemoveUser
never ran for depleted clients, clients.enable stayed true so the UI
showed them as active, and only xray_client_traffic.enable got flipped
- making "Restart Xray After Auto Disable" only half-work.
Resolve the targets via a JOIN through the new schema, flip clients.enable
so the Clients page reflects the state, and drop the legacy JSON
write-back plus the subId cascade workaround (email is unique now).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): live WebSocket updates + Ended status surfacing
ClientsPage now subscribes to traffic / client_stats / invalidate
WebSocket events instead of polling /onlines every 10s. Per-row
traffic counters refresh in place, online state stays current, and
list-level mutations elsewhere trigger a refresh.
The client roll-up summary moves from InboundsPage to ClientsPage
where it belongs, restructured into six labeled stat tiles
(Total / Online / Ended / Expiring / Disabled / Active) with email
popovers on the ones with issues.
Auto-disabled clients (traffic exhausted or expiry passed) now
classify as 'depleted' even though clients.enable=false, so they
show up under the Ended filter and render a red Ended tag instead
of looking indistinguishable from an operator-disabled row.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(nodes): per-node client roll-up and panel version
Added transient inboundCount / clientCount / onlineCount /
depletedCount fields to model.Node, populated by NodeService.GetAll
via aggregated queries (one join across inbounds + client_inbounds,
one over client_traffics intersected with the in-memory online
emails). The Nodes list renders these as colored chips on a new
"Clients" column so an operator can see at a glance how many users
each node carries and how many are currently online or depleted.
Also exposes the remote panel's version. The central panel adds
panelVersion to its /api/server/status payload (sourced from
config.GetVersion). Probe reads that field and persists it on the
node row, mirroring how xrayVersion already flows. NodesPage gets
a new column next to Xray Version, in both desktop and mobile
views, with English and Persian strings.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): stop node sync from resurrecting deleted clients
Several related issues around node-managed clients:
- Remote runtime: drop the per-inbound resetAllClientTraffics path
and point traffic/onlines/lastOnline fetches at the new
/panel/api/clients/* routes.
- Delete from master: always push the updated inbound to the node
even when the client was already disabled or depleted, so the
node actually loses the user instead of silently keeping it.
- setRemoteTraffic: mirror remote clients into the central tables
only on first discovery of a node inbound. Matched inbounds let
the master own the join table, so a stale snap can no longer
re-create a ClientRecord (and join row) for a client that was
just deleted on the master.
- ClientService.Delete: route through submitTrafficWrite so deletes
serialize with node traffic merges, and switch the final
ClientRecord delete to an explicit Where("id = ?") clause.
- setRemoteTraffic UNIQUE-constraint fix: use clause.OnConflict on
inserts and email-keyed UPDATEs for client_traffics, so mirroring
a snap doesn't trip the unique email index.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): switch client API endpoints from id to email
All client-scoped routes now use the unique email as the path key
(get, update, del, attach, detach, links). Email is the stable,
protocol-independent identifier — UUIDs don't exist for trojan or
shadowsocks, and internal numeric ids leaked panel implementation
detail into the public API.
Removed the redundant /traffic/byId/:id endpoint (covered by
/traffic/:email) and collapsed /links/:id/:email into /links/:email,
which now returns links across every attached inbound for the client.
Frontend selection, bulk delete, and toggle state are now keyed by
email as well, dropping the id→email lookup workaround.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(server): move cached state and helpers into ServerService
ServerController had grown to hold its own status cache, version-list
TTL cache, history-bucket whitelist, and the loop that drove all three
— concerns that belong in the service layer. Pull them out:
- lastStatus + the @2s refresh become ServerService.RefreshStatus and
ServerService.LastStatus; the controller's cron now just orchestrates
the cross-service side effects (xrayMetrics sample, websocket broadcast).
- The 15-minute Xray-versions cache (with stale-on-error fallback) moves
into ServerService.GetXrayVersionsCached, collapsing the controller
handler to a single call.
- The freedom/blackhole outbound-tag walk used by /xraylogs becomes
ServerService.GetDefaultLogOutboundTags.
- The allowed-history-bucket whitelist moves to package-level
service.IsAllowedHistoryBucket, so both NodeController and
ServerController validate against the same list.
Net result: web/controller/server.go drops from 458 to 365 lines and
contains only HTTP wiring + presentation-y side effects.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): emit JSON-text columns as nested objects
Inbound, ClientRecord, and InboundClientIps store settings /
streamSettings / sniffing / reverse / ips as JSON-text in the DB. The
API was passing that text through verbatim, so every consumer had to
JSON.parse a string inside a string. Add MarshalJSON / UnmarshalJSON so
the wire format is a real nested object, while still accepting the
legacy escaped-string shape on write. Frontend dbinbound.js gets a
matching coerceInboundJsonField helper for the same dual-shape read
path, and inbound.js toJson stops emitting empty/placeholder fields
(externalProxy [], sniffing destOverride when disabled, etc.) so the
new normalised JSON stays terse. api-docs and the inbound-clone path
are updated to the new shape. Controller route lists are regrouped so
all GETs sit above POSTs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): include inboundIds and traffic in /clients/list
ClientRecord got its own MarshalJSON in the previous commit, and
ClientWithAttachments embeds it to add inboundIds and traffic. Go
promotes the embedded MarshalJSON to the outer struct, so the encoder
was calling ClientRecord.MarshalJSON for the whole value and silently
dropping the extras. The frontend reads row.inboundIds / row.traffic
from /clients/list, so attached inbounds didn't render and newly added
clients looked like they hadn't saved. Add an explicit MarshalJSON on
ClientWithAttachments that splices the extras in.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): gate IP Log on ipLimitEnable + clean access-log dropdown
Legacy panel hid the IP Log section when access logging was off; the
Vue 3 migration left it gated on isEdit only, so the section showed
even when xray's access log was 'none' and nothing was being recorded.
Restore the ipLimitEnable gate on the edit modal's IP Log form-item.
While here, clean up the Xray Settings access-log dropdown: previously
two 'none' entries appeared (an empty value labelled with t('none') and
the literal 'none' from the options array). Drop the empty option for
access log (the literal 'none' covers it) and relabel the empty option
for error log / mask address to t('empty') so they're distinguishable.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(nodes): route per-client ops through node clients API + orphan sweep
Adds Runtime methods AddClient, UpdateUser, and DeleteUser so master
mutates clients on a node via /panel/api/clients/{add,update,del} rather
than pushing the whole inbound. The previous rt.UpdateInbound path made
the node DelInbound+AddInbound on every single-client change, briefly
cycling every other user on the same inbound.
DelInbound no longer filters by enable=true, so a disabled node inbound
actually gets removed from the node instead of being resurrected by the
next snap.
setRemoteTrafficLocked now sweeps any ClientRecord with zero
ClientInbound rows after SyncInbound rebuilds the attachments, which is
how a node-side delete propagates back to master instead of leaving a
detached ghost. ClientService.Delete tombstones the email first so a
snap arriving mid-delete can't re-create the record.
WebSocket broadcasts an "invalidate(clients)" message on every client
mutation so the Clients page refreshes without manual reload.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(balancers): allow fallback on all strategies + feed burstObservatory from random/roundRobin
Drops the random/roundRobin gate on the Fallback field in
BalancerFormModal so every strategy can pick a fallback outbound.
syncObservatories now feeds burstObservatory from leastLoad +
random + roundRobin balancers (was leastLoad only), matching how
leastPing feeds observatory.
Fix the JsonEditor "Unexpected end of JSON input" that appeared
when switching a balancer between leastPing and another strategy:
the obsView watcher was gated on showObsEditor (a boolean OR of
the two flags) and missed the case where one observatory
swapped for the other in the same tick. Watch the individual
flags instead so obsView flips to the surviving editor and the
getter stops pointing at a deleted key.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): use sortedInbounds for mobile empty-state check
InboundList referenced an undefined visibleInbounds in the mobile
card list's empty-state guard, throwing "Cannot read properties of
undefined (reading 'length')" and breaking the entire mobile render.
* feat(clients): sortable table columns
Adds the same sortState / sortableCol / sortFns pattern InboundList
uses, wrapping filteredClients in sortedClients so sort composes with
the existing search/filter pipeline. Sortable: enable, email,
inboundIds (attachment count), traffic, remaining, expiryTime;
actions and online stay unsorted.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(shadowsocks): generate valid ss2022 keys and per-client method for legacy ciphers
The Add Client flow on shadowsocks inbounds was producing xray configs
that failed to start:
- 2022-blake3-* ciphers need a base64-encoded key of an exact byte
length per cipher. fillProtocolDefaults was assigning a uuid-style
string, which xray rejects as "bad key". Now the password is
generated (or replaced if invalid) via random.Base64Bytes(n) sized
to the chosen cipher.
- Legacy ciphers (aes-256-gcm, chacha20-*, xchacha20-*) require a
per-client method field in multi-user mode; model.Client has no
Method, so settings.clients was stored without one and xray failed
with "unsupported cipher method:". applyShadowsocksClientMethod
now injects the top-level method into each client on add/update,
and healShadowsocksClientMethods backfills it at xray-config-build
time so existing inbounds heal on the next start.
- xray/api.go ssCipherType switch was missing aes-256-gcm, which
fell through to ss2022 path.
- SSMethods dropdown now offers aes-256-gcm.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): preserve ClientRecord on inbound delete + filter Attached inbounds to multi-client protocols
Replace the global orphan sweep in setRemoteTrafficLocked with a
per-inbound diff cleanup: only delete a ClientRecord whose email
disappeared from a snap-tracked inbound (i.e. a node-side delete).
Inbounds that vanished entirely from the snap (e.g. admin deleted
the inbound on master) aren't iterated, so a client whose last
attachment came from that inbound is now left alone instead of
being deleted alongside the inbound.
ClientFormModal and ClientBulkAddModal now filter the Attached
inbounds dropdown to protocols that actually support multiple
clients: shadowsocks, vless, vmess, trojan, hysteria, hysteria2,
and portfallback (which routes through VLESS settings).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): make empty-state text readable on dark/ultra themes
The "No clients yet" empty state had a hardcoded black color
(rgba(0,0,0,0.45)) that vanished against the dark backgrounds.
Drop the inline color, let it inherit from the AntD theme, and
fade with opacity like the mobile card empty state already does.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): client-first tgbot add flow, tgId field, lightweight inbound options
- tgbot: drop legacy per-protocol Add Client UI in favour of a client-first
multi-inbound flow. New BuildClientDraftMessage / getInboundsAttachPicker
let an admin pick one or more inbounds and submit a single client; per-
protocol secrets are now generated server-side via fillProtocolDefaults.
Drops awaiting_id/awaiting_password_tr/awaiting_password_sh state cases
and add_client_ch_default_id/pass_tr/pass_sh/flow callbacks. Adds a
setTGUser button + awaiting_tg_id state so the bot can set Client.TgID
during Add.
- clients UI: add Telegram user ID input to ClientFormModal (0 = none).
Hide IP Limit field entirely when ipLimitEnable is off — disabled fields
still take layout space, this collapses Auth(Hysteria) to full width.
- inbounds API: new GET /panel/api/inbounds/options that returns just
{id, remark, protocol, port, tlsFlowCapable}. Used by the clients page
pickers so the dropdown payload stays small on panels with thousands of
clients (drops settings JSON, clientStats, streamSettings). Server-side
TlsFlowCapable mirrors Inbound.canEnableTlsFlow so the modal no longer
needs to parse streamSettings client-side.
- clientInfoMsg now shows attached inbound remarks, and getInboundUsages
reports the attached client count per inbound.
- api-docs: document the new /options endpoint and add tgId / flow to the
clients add/update bodies.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): keep Node column visible for node-attached inbounds
The Node column was bound to hasActiveNode, so disabling every node hid
the column even when inbounds were still attached to those nodes — the
admin lost the visual cue that those inbounds belonged to a node and
would come back when it was re-enabled. Combine hasActiveNode with a
new hasNodeAttachedInbound check (any dbInbound with nodeId != null) so
the column survives node-disable.
* fix(api-docs): accept functional-component icons in EndpointSection
AntD-Vue icons (SafetyCertificateOutlined, etc.) are functional
components, so the icon prop's type: Object validator was rejecting
them with a "Expected Object, got Function" warning at runtime.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* test: cover crypto, random, netsafe, sub helpers, xray equals, websocket hub, node service
Adds ~110 unit tests across previously untested packages. Focus on
pure-logic and concurrency surfaces where regressions would silently
affect users:
- util/crypto, util/random: password hashing round-trip, ss2022 key
generation, alphabet/length invariants.
- util/netsafe: IsBlockedIP edge cases, NormalizeHost validation,
SSRF guard with AllowPrivate context bypass.
- util/common, util/json_util: traffic formatter, Combine nil-skip,
RawMessage empty-as-null and copy-on-unmarshal.
- sub: splitLinkLines, searchKey/searchHost, kcp share fields,
finalmask normalization, buildVmessLink round-trip.
- xray: Config.Equals and InboundConfig.Equals field-by-field,
getRequiredUserString/getOptionalUserString type checks.
- web/websocket: hub registration, throttling, slow-client eviction,
nil-receiver safety, concurrent register/unregister.
- web/service: NodeService.normalize validation, normalizeBasePath,
HeartbeatPatch.ToUI mapping.
- web/job: atomicBool concurrent set/takeAndReset semantics.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n(clients): replace English fallbacks with proper translation keys
Pulls every hard-coded English label/title in the Clients page and its
four modals through the i18n layer so localized panels stop leaking
English. New keys live under pages.clients (auth, hysteriaAuth, uuid,
flow, flowNone, reverseTag, reverseTagPlaceholder, telegramId,
telegramIdPlaceholder, created, updated, ipLimit) plus refresh at the
root and toasts.bulkDeletedMixed / bulkCreatedMixed for partial-failure
toasts. Also switches the add-client modal's primary button from "Add"
to "Create" for consistency with other create flows.
The bulk-add Random/Random+Prefix/... email-method options stay
hard-coded by request - they're identifier-shaped strings.
* i18n: backfill 99 missing keys across all 12 non-English locales
Brings every translation file up to parity with en-US.json so the
Clients page, the fallback-children inbound section, the new refresh
verb, the Nodes panel-version label and a handful of older holes stop
falling through to the English fallback. New strings span:
- pages.clients.* (labels, confirmations, toasts, emailMethods)
- pages.inbounds.portFallback.* (Reality fallback inbound section)
- pages.nodes.panelVersion, menu.clients, refresh
Technical identifiers (Auth, UUID, Flow, Reverse tag) are intentionally
left untranslated since they correspond to xray-core field names.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n: drop stale pages.client block duplicated in every non-English locale
Every non-English locale carried a pages.client (singular) section with
30 entries that duplicated pages.clients (plural). The plural namespace
is what the Vue code actually consumes; the singular one was dead
weight from an older rename that never got cleaned up in the
non-English files. Removing it brings every locale to exactly 984
keys, matching en-US.json.
* chore: apply modernize analyzer fixes across codebase
Mechanical replacements suggested by golang.org/x/tools/.../modernize:
strings.Cut/CutPrefix/SplitSeq, slices.Contains, maps.Copy, min(),
range-over-int, new(expr), strings.Builder for hot += loops,
reflect.TypeFor[T](), sync.WaitGroup.Go(), drop legacy //+build lines.
* feat(database): add PostgreSQL as an optional backend alongside SQLite
Lets operators with large client counts or multi-node setups pick PostgreSQL
at install time without breaking the existing SQLite default. Backend is
selected at runtime via XUI_DB_TYPE/XUI_DB_DSN, a small dialect layer keeps
the five JSON_EXTRACT/JSON_EACH queries portable, and a new `x-ui migrate-db`
subcommand copies SQLite data into PostgreSQL in FK-aware order.
* fix(inbounds): gate node selector to multi-node-capable protocols
Hide the Deploy-To selector and clear nodeId when switching to a
protocol that can't run on a remote node. Also:
- subs: return 404 (not 400) when subId matches no inbounds, so VPN
clients distinguish "deleted/unknown" from a server error
- hysteria link gen: use the inbound's resolved address so node-managed
inbounds advertise the node host instead of the central panel
- shadowsocks: default network to 'tcp' (udp was causing issues for some
clients on first-create)
- vite dev proxy: rewrite migrated-route bypass against the live base
path instead of a hardcoded single-segment regex
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): bulk add/delete correctness + perf, working pagination, delayed-start in form
Bulk add/delete were serial on the frontend (one toast per call, N round-trips)
and the backend race exposed by parallelizing them lost client attachments and
hit UNIQUE constraint failed on client_inbounds. The single add/edit modal also
had no Start-After-First-Use option, and the table never showed the delayed
duration.
Backend (web/service/client.go):
- Per-inbound mutex on Add/Update/Del InboundClient so concurrent writers on
the same inbound don't lose the read-modify-write of settings JSON.
- SyncInbound skips create+join when the email is tombstoned so a concurrent
maintenance pass (adjustTraffics, autoRenewClients, markClientsDisabledIn-
Settings) that did a stale RMW can't resurrect a just-deleted client with a
fresh id.
- compactOrphans sweeps settings.clients entries whose ClientRecord no longer
exists, applied in Add/DelInboundClient + DelInboundClientByEmail so each
user-initiated mutation self-heals the inbound's settings.
- DelInboundClient uses Pluck instead of First for the stats lookup so a
missing row doesn't abort the delete with a noisy ErrRecordNotFound log.
Frontend:
- HttpUtil.{get,post} accept a silent option that suppresses the auto-toast.
- ClientBulkAddModal fires creates in parallel + silent + one summary toast.
- useClients.removeMany runs deletes in parallel + silent and refreshes once;
ClientsPage bulk delete uses it and shows one aggregate toast.
- useClients.applyInvalidate debounces 200 ms so the burst of N WebSocket
invalidate events from the backend collapses into a single refresh.
- ClientsPage pagination is reactive (paginationState ref + tablePagination
computed); onTableChange persists page-size and page changes.
- ClientFormModal gains a Start-After-First-Use switch + Duration days input
alongside the existing Expiry Date picker; on edit-mode open a negative
expiryTime is decoded back to delayed mode + days; on submit the payload
sends -86400000 * days or the absolute timestamp.
- ClientsPage table shows the delayed-start duration (blue tag Nd, tooltip
Start After First Use: Nd) instead of infinity.
- Telegram ID field in the form is hidden when /panel/setting/defaultSettings
reports tgBotEnable=false; Comment then fills the row.
- Form row 3 collapses UUID (span 12) + Total GB (span 8) + Limit IP (span 4)
when ipLimitEnable is on, else UUID + Total GB at 12/12.
- useInbounds.rollupClients counts only clients with a matching clientStats
row, so orphans in settings.clients no longer inflate the inbound's count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(windows): clean shutdown, working panel restart, harden kernel32 load
Three Windows-specific issues addressed:
1. Orphaned xray-windows-amd64 after VS Code debugger stop. Delve's
"Stop" sends TerminateProcess to the Go binary, which is uncatchable
— our signal handlers never run, so xrayService.StopXray() is skipped
and xray is left dangling. Spawn xray as a child of a Job Object with
JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE so the OS kills xray when our
handle to the job is closed (which happens even on TerminateProcess).
Also trap os.Interrupt in main so Ctrl+C in the terminal runs the
graceful path.
2. /panel/setting/restartPanel logged "failed to send SIGHUP signal: not
supported by windows" because Windows can't deliver arbitrary signals.
Add a restart hook in web/global; main registers it to push SIGHUP
into its own signal channel, and RestartPanel calls the hook before
falling back to the (Unix-only) signal path. Same restart-loop code
runs in both cases.
3. util/sys/sys_windows.go now uses windows.NewLazySystemDLL so the
kernel32.dll resolve is pinned to %SystemRoot%\System32 (prevents
DLL hijacking by a planted DLL next to the binary). Local filetime
type replaced with windows.Filetime, and the unreliable
syscall.GetLastError() fallback replaced with a type assertion on the
errno captured at call time.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(sys): correct CPU/connection accounting on linux + darwin
util/sys/sys_linux.go:
- GetTCPCount/GetUDPCount were counting the column header row in
/proc/net/{tcp,udp}[6] as a connection, inflating the reported total
by 1 per non-empty file (so the panel status line always showed 2
more connections than actually existed). Replace getLinesNum +
safeGetLinesNum with a single bufio.Scanner-based countConnections
that skips the header.
- CPUPercentRaw now opens HostProc("stat") instead of a hardcoded
/proc/stat so HOST_PROC overrides apply, matching the connection
counters in the same file.
- Simplify CPU field unpacking: pad nums to 8 once instead of guarding
every assignment with a len check.
util/sys/sys_darwin.go:
- Fix swapped idle/intr indices on kern.cp_time. BSD CPUSTATES order
is user, nice, sys, intr, idle (CP_INTR=3, CP_IDLE=4) — gopsutil's
cpu_darwin_nocgo.go reads the same layout. The previous code used
out[3] as idle and out[4] as intr, so busy = total - dIdle was
actually subtracting interrupt time, making the panel report CPU
usage close to 100% on macOS regardless of actual load.
- Collapse the per-field delta math into a single loop.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(xray): rotate crash reports into log folder, prevent overwrites
writeCrashReport had two flaws: it wrote to the bin folder (alongside the
xray binary) which conflates artifacts, and the second-precision timestamp
meant a tight restart-loop crash burst overwrote prior reports. Write to
the log folder with nanosecond precision and keep the last 10 reports.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* revert(inbounds): drop unreleased portfallback protocol
The Port-with-Fallback inbound (commit 62fd9f9d) was confusing as a
standalone protocol — fallbacks belong on a regular VLESS/Trojan TCP-TLS
inbound, the way Xray models them natively. Rip out the entire feature
cleanly (no migration needed since it was never released): protocol
constant, fallback children DB table, FallbackService, 2 API endpoints,
all UI rows, related translations and api-docs. A native fallback flow
attached to VLESS/Trojan TCP-TLS/Reality will land in a follow-up commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): native fallbacks on VLESS/Trojan TCP-TLS, with working child links
A VLESS or Trojan inbound on TCP with TLS or Reality can now act as a
fallback master: pick existing inbounds as children and the panel auto-
fills the SNI / ALPN / path / xver routing fields from each child's
transport, auto-builds settings.fallbacks at config-gen time, and
rewrites the child's client-share link so it advertises the master's
reachable endpoint and TLS state instead of the child's loopback listen.
Layout matches the Xray All-in-One Nginx example: master at :443 with
clients + TLS, each child on 127.0.0.1 with its own transport+clients.
Order matters (Xray walks fallbacks top-to-bottom) — reorder via the
per-row up/down arrows. Path / SNI / ALPN are exposed under a per-row
Edit toggle for the rare cases where the auto-derivation needs
overriding; otherwise just pick a child and you're done.
Backend: new InboundFallback table + FallbackService (GetByMaster /
SetByMaster / GetParentForChild / BuildFallbacksJSON); two routes
(GET / POST /panel/api/inbounds/:id/fallbacks); xray.GetXrayConfig
injects settings.fallbacks for any VLESS/Trojan TCP-TLS/Reality
inbound; GetInbounds annotates each child with FallbackParent so the
frontend can rewrite links without an extra round-trip.
Link projection covers every emission path — clients-page QR/links,
per-inbound Get URL, raw subscription, sub-JSON, sub-Clash, and the
inbounds-page link/info/QR — via a shared projectThroughFallbackMaster
on the backend and a shared projectChildThroughMaster on the frontend
that both handle the panel-tracked relationship and the legacy
unix-socket (@vless-ws) convention.
Strings translated into all 12 non-English locales.
* docs: rewrite CONTRIBUTING with full local-dev setup
The prior three-line CONTRIBUTING left newcomers guessing at every
non-trivial step: which Go / Node versions, where xray comes from, why
the panel goes blank when XUI_DEBUG=true is flipped on, how the Vue
multi-page setup is wired, what to do on Windows when go build trips
on the CGo SQLite driver.
Now covers prerequisites, MinGW-w64 install on Windows (niXman builds
or MSYS2), one-shot first-time setup, two frontend dev workflows with
the XUI_DEBUG asset-cache gotcha called out, the architecture and
conventions of the Vue side, a project-layout map, useful env vars,
and the PR checklist.
---------
2026-05-19 03:16:42 -07:00
local existing_webBasePath = $( ${ xui_folder } /x-ui setting -show true | grep -Eo 'webBasePath: .+' | awk '{print $2}' | sed 's#^/##' )
2026-05-19 04:09:35 -07:00
local existing_port = $( ${ xui_folder } /x-ui setting -show true | grep -Eo 'port: .+' | awk '{print $2}' )
# Properly detect empty cert by checking if cert: line exists and has content after it
local existing_cert = $( ${ xui_folder } /x-ui setting -getCert true | grep 'cert:' | awk -F': ' '{print $2}' | tr -d '[:space:]' )
2025-08-21 05:24:25 -07:00
local URL_lists = (
"https://api4.ipify.org"
2025-12-03 12:40:49 -08:00
"https://ipv4.icanhazip.com"
"https://v4.api.ipinfo.io/ip"
"https://ipv4.myexternalip.com/raw"
"https://4.ident.me"
"https://check-host.net/ip"
2025-08-21 05:24:25 -07:00
)
local server_ip = ""
for ip_address in " ${ URL_lists [@] } " ; do
2026-05-04 04:20:24 -07:00
local response = $( curl -s -w "\n%{http_code}" --max-time 3 " ${ ip_address } " 2> /dev/null)
2026-02-11 12:32:23 -08:00
local http_code = $( echo " $response " | tail -n1)
2026-05-07 15:51:28 -07:00
local ip_result = $( echo " $response " | head -n-1 | tr -d '[:space:]"' )
if [ [ " ${ http_code } " = = "200" && " ${ ip_result } " = ~ ^[ 0-9] +\. [ 0-9] +\. [ 0-9] +\. [ 0-9] +$ ] ] ; then
2026-02-11 12:32:23 -08:00
server_ip = " ${ ip_result } "
2025-07-27 08:24:11 -07:00
break
fi
done
2026-05-04 04:20:24 -07:00
2026-05-07 15:51:28 -07:00
if [ [ -z " $server_ip " ] ] ; then
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
# Panel binds 0.0.0.0 regardless; the IP is only used to compose the
# displayed access URL. Fall back to XUI_SERVER_IP or leave blank.
server_ip = " ${ XUI_SERVER_IP :- } "
else
echo -e " ${ yellow } Could not auto-detect server IP from any provider. ${ plain } "
while [ [ -z " $server_ip " ] ] ; do
read -rp "Please enter your server's public IPv4 address: " server_ip
server_ip = " ${ server_ip // / } "
if [ [ ! " $server_ip " = ~ ^[ 0-9] +\. [ 0-9] +\. [ 0-9] +\. [ 0-9] +$ ] ] ; then
echo -e " ${ red } Invalid IPv4 address. Please try again. ${ plain } "
server_ip = ""
fi
done
fi
2026-05-07 15:51:28 -07:00
fi
2024-10-25 02:05:22 -07:00
if [ [ ${# existing_webBasePath } -lt 4 ] ] ; then
2026-05-19 04:09:35 -07:00
if [ [ " $existing_hasDefaultCredential " = = "true" ] ] ; then
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
local config_webBasePath = " ${ XUI_WEB_BASE_PATH :- $( gen_random_string 18) } "
local config_username = " ${ XUI_USERNAME :- $( gen_random_string 10) } "
local config_password = " ${ XUI_PASSWORD :- $( gen_random_string 10) } "
local config_port = ""
2026-05-04 04:20:24 -07:00
2026-05-23 10:52:37 -07:00
local db_label = "SQLite (/etc/x-ui/x-ui.db)"
echo ""
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } Database Selection ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
2026-05-28 08:20:16 -07:00
echo -e " 1) SQLite (default — recommended for < 500 clients)"
2026-05-23 10:52:37 -07:00
echo -e " 2) PostgreSQL (recommended for high client counts / many nodes)"
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
if [ [ " ${ XUI_DB_TYPE :- sqlite } " = = "postgres" ] ] ; then
db_choice = "2"
else
db_choice = "1"
fi
else
read -rp "Choose [1]: " db_choice
db_choice = " ${ db_choice :- 1 } "
fi
2026-05-23 10:52:37 -07:00
if [ [ " $db_choice " = = "2" ] ] ; then
local xui_env_file
case " ${ release } " in
ubuntu | debian | armbian)
xui_env_file = "/etc/default/x-ui"
; ;
arch | manjaro | parch | alpine)
xui_env_file = "/etc/conf.d/x-ui"
; ;
*)
xui_env_file = "/etc/sysconfig/x-ui"
; ;
esac
local xui_dsn = ""
local pg_mode = ""
2026-05-28 08:20:16 -07:00
local pg_local_installed = 0
2026-05-23 10:52:37 -07:00
while [ [ -z " $xui_dsn " ] ] ; do
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
if [ [ -n " ${ XUI_DB_DSN :- } " ] ] ; then
xui_dsn = " ${ XUI_DB_DSN } "
db_label = "PostgreSQL (external)"
break
fi
echo -e " ${ yellow } Installing PostgreSQL locally (non-interactive)... ${ plain } "
local pg_cred_file
pg_cred_file = $( mktemp 2> /dev/null) || pg_cred_file = $( mktemp -t x-ui-pg-creds.XXXXXXXX)
if [ [ -n " ${ pg_cred_file } " ] ] && xui_dsn = $( PG_CRED_FILE = " ${ pg_cred_file } " install_postgres_local) ; then
pg_local_installed = 1
if [ [ -r " ${ pg_cred_file } " ] ] ; then
# shellcheck disable=SC1090
source " ${ pg_cred_file } "
fi
rm -f " ${ pg_cred_file } "
db_label = " PostgreSQL ( ${ PG_USER } @ ${ PG_HOST } : ${ PG_PORT } / ${ PG_DB } ) "
break
fi
rm -f " ${ pg_cred_file } "
echo -e " ${ red } PostgreSQL installation failed in non-interactive mode; aborting. ${ plain } "
echo -e " ${ yellow } Set XUI_DB_DSN to use an existing server, or XUI_DB_TYPE=sqlite. ${ plain } "
exit 1
fi
2026-05-23 10:52:37 -07:00
echo ""
echo -e " 1) Install PostgreSQL locally and create a dedicated user/db (recommended)"
echo -e " 2) Use an existing PostgreSQL server (enter DSN)"
read -rp "Choose [1]: " pg_mode
pg_mode = " ${ pg_mode :- 1 } "
if [ [ " $pg_mode " = = "2" ] ] ; then
while [ [ -z " $xui_dsn " ] ] ; do
read -rp "Enter PostgreSQL DSN (postgres://user:pass@host:port/dbname?sslmode=disable): " xui_dsn
xui_dsn = " ${ xui_dsn // / } "
done
db_label = "PostgreSQL (external)"
else
echo -e " ${ yellow } Installing PostgreSQL — this may take a moment... ${ plain } "
2026-05-28 08:20:16 -07:00
local pg_cred_file
pg_cred_file = $( mktemp 2> /dev/null) || pg_cred_file = $( mktemp -t x-ui-pg-creds.XXXXXXXX)
if [ [ -z " ${ pg_cred_file } " ] ] ; then
echo -e " ${ red } Failed to create temporary credentials file. ${ plain } "
xui_dsn = ""
continue
fi
if xui_dsn = $( PG_CRED_FILE = " ${ pg_cred_file } " install_postgres_local) ; then
pg_local_installed = 1
if [ [ -r " ${ pg_cred_file } " ] ] ; then
# shellcheck disable=SC1090
source " ${ pg_cred_file } "
fi
rm -f " ${ pg_cred_file } "
db_label = " PostgreSQL ( ${ PG_USER } @ ${ PG_HOST } : ${ PG_PORT } / ${ PG_DB } ) "
2026-05-23 10:52:37 -07:00
else
2026-05-28 08:20:16 -07:00
rm -f " ${ pg_cred_file } "
2026-05-23 10:52:37 -07:00
echo ""
echo -e " ${ red } PostgreSQL installation failed. ${ plain } "
echo -e " 1) Retry local install"
echo -e " 2) Enter an external DSN instead"
echo -e " 3) Abort install"
echo -e " 4) Fall back to SQLite"
read -rp "Choose [1]: " pg_fail
pg_fail = " ${ pg_fail :- 1 } "
case " $pg_fail " in
2) pg_mode = "2" ; ;
2026-05-28 08:20:16 -07:00
3)
echo -e " ${ red } Install aborted. ${ plain } "
exit 1
; ;
4)
db_choice = "1"
xui_dsn = ""
break
; ;
2026-05-23 10:52:37 -07:00
*) xui_dsn = "" ; ;
esac
fi
fi
done
if [ [ -n " $xui_dsn " ] ] ; then
install -d -m 755 " $( dirname " $xui_env_file " ) "
umask 077
cat > " $xui_env_file " << EOF
XUI_DB_TYPE = postgres
XUI_DB_DSN = ${ xui_dsn }
EOF
chmod 600 " $xui_env_file "
umask 022
export XUI_DB_TYPE = postgres
export XUI_DB_DSN = " ${ xui_dsn } "
2026-05-31 08:53:34 -07:00
ensure_pg_client || echo -e " ${ yellow } ⚠ Could not install pg_dump/pg_restore. In-panel database backup/restore will be unavailable until you install the postgresql-client package. ${ plain } "
2026-05-23 10:52:37 -07:00
fi
fi
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
if [ [ " $NONINTERACTIVE " = = "1" ] ] ; then
if [ [ -n " ${ XUI_PANEL_PORT :- } " ] ] ; then
config_port = " ${ XUI_PANEL_PORT } "
echo -e " ${ yellow } Your Panel Port is: ${ config_port } ${ plain } "
else
config_port = $( shuf -i 1024-62000 -n 1)
echo -e " ${ yellow } Generated random port: ${ config_port } ${ plain } "
fi
Feat/multi inbound clients (#4469)
* feat(clients): add shadow tables for first-class client promotion
Introduces three new GORM-backed tables (clients, client_inbounds,
inbound_fallback_children) and a populate-only seeder that backfills
them from each inbound's existing settings.clients JSON. Duplicate
emails across inbounds auto-merge under one client row, with each
field conflict logged. Existing services are unchanged and continue
reading from settings.clients — this commit is groundwork only.
* feat(clients): make clients+client_inbounds the runtime source of truth
Adds ClientService.SyncInbound that reconciles the new tables from
each inbound's clients list whenever existing service paths mutate
settings.clients. Wires it into AddInbound, UpdateInbound,
AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, DelDepletedClients, autoRenewClients, and
the timestamp-backfill path in adjustTraffics, plus DetachInbound
on DelInbound.
GetXrayConfig now builds settings.clients from the new tables before
writing config.json, and getInboundsBySubId joins through them
instead of JSON_EACH on settings JSON. Live Xray config and
subscription endpoints are now driven by the relational view;
settings.clients JSON stays in step as a side effect of every write.
* feat(clients): add top-level Clients tab and CRUD API
Adds /panel/api/clients endpoints (list, get, add, update, del,
attach, detach) backed by ClientService methods that orchestrate
the per-inbound Add/Update/Del flows so a single client row is
created once and attached to many inbounds in one operation.
The frontend gains a dedicated Clients page (frontend/clients.html
+ src/pages/clients/) with an AntD table, multi-inbound attach
modal, and full CRUD. Axios interceptor learns to honour
Content-Type: application/json so the JSON endpoints work
alongside the legacy form-encoded ones.
The legacy per-inbound client modal stays untouched in this PR —
both flows now write to the same source of truth.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): add Port-with-Fallback inbound type
Adds a new "portfallback" protocol that emits as a VLESS-TLS inbound
under the hood but is paired with a sidecar table of child inbounds.
Panel auto-builds settings.fallbacks at Xray-config-gen time from the
sidecar — each child's listen+port becomes the fallback dest, with
SNI/ALPN/path/xver match criteria pulled from the row. No more typing
loopback ports by hand or keeping settings.fallbacks in sync.
Backend: new FallbackService (Get/SetChildren, BuildFallbacksJSON);
two new routes (GET/POST /panel/api/inbounds/:id/fallbackChildren);
xray.GetXrayConfig injects fallbacks for PortFallback inbounds; the
inbound model emits protocol="vless" so Xray accepts the config.
Frontend: PORTFALLBACK joins the protocol dropdown; selecting it
shows the standard VLESS controls plus a Fallback Children table
(inbound picker + per-row SNI/ALPN/path/xver). Children are loaded
on edit and replaced atomically on save.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reset Traffic, QR Code, Info actions + Online/Remaining columns
The Clients page table gains:
- Online column — green/grey tag driven by /panel/api/inbounds/onlines,
polled every 10s.
- Remaining column — bytes-remaining tag, coloured green/orange/red
against quota, purple infinity when unlimited.
- Action icons per row: QR, Info, Reset traffic, Edit, Delete.
ClientInfoModal shows the full client detail (uuid/password/auth,
traffic ↑/↓ + remaining + all-time, expiry absolute + relative,
attached inbounds chip list, online + last-online).
ClientQrModal fetches links for the client's subId via
/panel/api/inbounds/getSubLinks/:subId and renders each one through
the existing QrPanel component.
Reset Traffic confirms then calls the existing per-inbound endpoint
on the client's first attached inbound (the traffic row is keyed on
email globally, so any attached inbound resets the shared counter).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): expose Attached inbounds in edit mode
The multi-select was gated on add-only, so editing a client had no way
to change which inbounds it belonged to. The picker now shows in both
modes, and on submit the modal diffs the picked set against the
original attachedIds — additions go through the /attach endpoint,
removals through /detach, both after the field update lands so the
new attachments get the latest values.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): unbreak template parsing + stale i18n keys
- InboundFormModal: split the multi-line help string in the
PortFallback section onto one line — Vue's template parser was
bailing on Unterminated string constant because a single-quoted
literal spanned two lines inside a {{ }} interpolation.
- ClientInfoModal: t('disable') was missing at the root level, so
vue-i18n returned the key path literally. Use t('disabled') which
exists.
- Linter cleanup elsewhere: pages.client.* references renamed to
pages.clients.* to match the merged i18n block; whitespace
normalisation in a few unrelated Vue templates.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* 1
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(traffic): drop all-time traffic tracking
Removes the AllTime field from Inbound and ClientTraffic and migrates
existing DBs by dropping the all_time columns on startup. The counter
duplicated up+down without adding signal, and the per-event accumulator
ran on every traffic write.
Frontend: drop the All-time column from the inbound list and the
client-row table, the All-time row from the client info modal, and the
All-Time Total Usage tile from the inbounds summary card. The
allTimeTraffic/allTimeTrafficUsage i18n keys are removed across every
locale.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): mobile cards, multi-select, bulk add
Adds the same row-card layout the inbounds page uses on mobile: the
table is suppressed under the mobile breakpoint and each client renders
as a compact card with a status dot, email, Info button, Enable switch,
and overflow menu. All the per-client detail (traffic, remaining,
expiry, attached inbounds, flow, created/updated, URL, subscription)
opens through the existing info modal.
Multi-select with bulk delete wires AntD row-selection on desktop and
a per-card checkbox on mobile; a Delete (N) button appears in the
toolbar when anything is selected.
Bulk add reuses the five email-generation modes from the inbound bulk
modal but takes a multi-inbound picker so one bulk run can attach to
several inbounds at once. Submits client-by-client through the
existing /panel/api/clients/add endpoint.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): remove legacy per-inbound client UI
Now that clients live as first-class rows attached to one or many
inbounds, the per-inbound client UI on the inbounds page is dead
weight — every client action either has a global equivalent on the
Clients page or makes no sense in a many-to-many world.
Deletes ClientFormModal, ClientBulkModal, CopyClientsModal, and
ClientRowTable from inbounds/. Strips the matching emits, refs,
handlers, and dropdown menu items from InboundList and InboundsPage,
and removes the dead mobile expand-chevron state and the desktop
expanded-row plumbing that drove the inline client table.
The InboundFormModal Clients tab still works in add-mode (one inline
client at inbound creation) — that flow goes through ClientService.
SyncInbound on save and remains useful.
Fixes a stray "</a-dropdown>" left over by an earlier toolbar edit
in ClientsPage that broke the template parser.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Delete depleted action
Mirrors the legacy delDepletedClients action that lived under the
inbounds page, but as a first-class /panel/api/clients/delDepleted
endpoint backed by ClientService. The new path goes through
ClientService.Delete for each depleted email, so the new clients +
client_inbounds + xray_client_traffic tables stay consistent.
Adds a danger-styled toolbar button on the Clients page (next to
Reset all client traffic) with a confirm dialog and a toast
reporting the deleted count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): move every client-shaped endpoint off /inbounds onto /clients
After the multi-inbound client migration, client state belongs to the
client API surface, not the inbound one. Twelve routes that were
crammed under /panel/api/inbounds/* now live where they belong, under
/panel/api/clients/*.
Moved (route, handler, doc):
POST /clientIps/:email
POST /clearClientIps/:email
POST /onlines
POST /lastOnline
POST /updateClientTraffic/:email
POST /resetAllClientTraffics/:id
POST /delDepletedClients/:id
POST /:id/resetClientTraffic/:email
GET /getClientTraffics/:email
GET /getClientTrafficsById/:id
GET /getSubLinks/:subId
GET /getClientLinks/:id/:email
Their /clients/* counterparts are:
POST /clients/clientIps/:email
POST /clients/clearClientIps/:email
POST /clients/onlines
POST /clients/lastOnline
POST /clients/updateTraffic/:email
POST /clients/resetTraffic/:email (email-only, fans out)
GET /clients/traffic/:email
GET /clients/traffic/byId/:id
GET /clients/subLinks/:subId
GET /clients/links/:id/:email
per-inbound resetAllClientTraffics and delDepletedClients are dropped
entirely — the Clients page already exposes global Reset All Traffic
and Delete depleted actions, and per-inbound resets are meaningless
once a client can be attached to many inbounds.
ClientService.ResetTrafficByEmail is the new email-only reset path:
it looks up every inbound the client is attached to and pushes the
counter reset + Xray re-add through inboundService.ResetClientTraffic
for each one, so depleted users come back online instantly.
Frontend callers (ClientsPage, useClients, ClientQrModal,
ClientInfoModal, InboundInfoModal, InboundsPage, useInbounds) all
switched to the new paths. The Inbounds page drops its per-inbound
"Reset client traffic" and "Delete depleted clients" dropdown items —
users do those at the client level now. api-docs is rebuilt to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): switch tgbot + ldap callers to ClientService
Adds two thin helpers to ClientService (CreateOne, DetachByEmail) and
rewrites tgbot.SubmitAddClient and ldap_sync_job to call ClientService
directly. Removes the JSON-blob payloads (BuildJSONForProtocol output for
add, clientsToJSON/clientToJSON helpers) that callers previously fed to
InboundService.AddInboundClient/DelInboundClient.
ldap_sync_job.batchSetEnable now loops InboundService.SetClientEnableByEmail
per email instead of trying to coerce AddInboundClient into doing the
update — the old path would have failed duplicate-email validation for
existing clients anyway.
The legacy InboundService.AddInboundClient/UpdateInboundClient/
DelInboundClient methods stay in place; they are now only used internally
by ClientService Create/Update/Delete/Attach. Inlining + deleting them
follows in a separate commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): move all client mutation methods to ClientService
Moves the client mutation surface out of InboundService and into
ClientService. These methods all operate on a single client (identity
fields, traffic limits, expiry, ip limit, enable state, telegram tg id)
and didn't belong on the inbound aggregate.
Moved (12 methods): AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, checkEmailsExistForClients, SetClientTelegramUserID,
checkIsEnabledByEmail, ToggleClientEnableByEmail, SetClientEnableByEmail,
ResetClientIpLimitByEmail, ResetClientExpiryTimeByEmail,
ResetClientTrafficLimitByEmail.
Each method now takes an explicit *InboundService for the helpers that
legitimately stay on InboundService (GetInbound, GetClients, runtimeFor,
AddClientStat / UpdateClientStat / DelClientStat, DelClientIPs /
UpdateClientIPs, emailUsedByOtherInbounds, getAllEmailSubIDs,
GetClientInboundByEmail / GetClientInboundByTrafficID,
GetClientTrafficByEmail).
Stays on InboundService: ResetClientTrafficByEmail and
ResetClientTraffic(id, email) — these mutate xray_client_traffic rows,
not client identity, so they're inbound-side bookkeeping.
Callers updated: tgbot (6 calls), ldap_sync_job (1 call),
InboundService internal (writeBackClientSubID, CopyInboundClients,
AddInbound's email-uniqueness check), ClientService Create/Update/
Delete/Attach/Detach.
Also removes a dead resetAllClientTraffics controller handler whose
route was already gone after the previous /clients API migration.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): finish migrating to ClientService + tidy IP routes
Two related cleanups in the new /clients surface:
1. Move ResetAllClientTraffics (bulk-reset of xray_client_traffic +
last_traffic_reset_time, with node-runtime propagation) from
InboundService to ClientService. PeriodicTrafficResetJob now holds
a clientService and calls
j.clientService.ResetAllClientTraffics(&j.inboundService, id).
The last client-mutation method on InboundService is gone.
2. Shorten redundantly-named routes/handlers under /panel/api/clients:
- /clientIps/:email -> /ips/:email (handler getIps)
- /clearClientIps/:email -> /clearIps/:email (handler clearIps)
The "client" prefix was redundant inside the clients namespace.
Frontend (InboundInfoModal) and api-docs updated to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds,clients): clean up inbound modal + enrich client modal
Inbound modal rework (InboundFormModal.vue + inbound.js):
- Drop the embedded Client subform in the Protocol tab. Multi-inbound
clients are managed exclusively from the Clients page now; a fresh
inbound is created with zero clients (settings constructors default
to []) and the user attaches clients afterwards.
- Hide the Protocol tab entirely when it has nothing to render
(VMESS, Trojan without fallbacks, Hysteria). Auto-switches active
tab to Basic when the tab disappears while focused.
- Move the Security section (Security selector + TLS block with certs
and ECH + Reality block) out of the Stream tab into its own
Security tab, sharing the canEnableStream gate.
Client modal additions (ClientFormModal.vue + ClientBulkAddModal.vue):
- Flow select (xtls-rprx-vision / -udp443) appears only when the
panel actually has a Vision-capable inbound (VLESS or PortFallback
on TCP with TLS or Reality). Hidden otherwise, and cleared when
it disappears.
- IP Limit input is disabled when the panel-level ipLimitEnable
setting is off, fetched into useClients alongside subSettings and
threaded through ClientsPage to both modals.
- Edit modal now shows an "IP Log" section listing IPs that have
connected with the client's credentials, with refresh and clear
buttons (calls the renamed /panel/api/clients/ips and /clearIps
endpoints).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): drop manual Fallbacks UI from inbound modal
The PortFallback protocol type now covers the common
VLESS-master-plus-children case with auto-wired dests, so the manual
Fallbacks editor (showFallbacks block in the Protocol tab) is mostly
redundant. Removed:
- the v-if="showFallbacks" template block (SNI/ALPN/Path/dest/PROXY rows)
- the showFallbacks computed
- the addFallback / delFallback helpers
- the .fallbacks-header / .fallbacks-title styles
- the showFallbacks gate from hasProtocolTabContent (so Trojan-over-TCP
no longer shows an empty Protocol tab)
Power users who need a non-inbound fallback dest (nginx, static site)
can still author settings.fallbacks via the Advanced JSON tab.
* feat(clients,inbounds): move search/filter to Clients page + small fixes
Search/filter relocation:
- Remove the search/filter toolbar (search switch + filter radio +
protocol/node selects + the visibleInbounds projection +
inboundsFilterState localStorage + filter CSS + the SearchOutlined/
FilterOutlined/ObjectUtil/Inbound imports it required) from
InboundList. The filters were all client-oriented buckets bolted
onto the inbound row.
- Add a search/filter toolbar to ClientsPage with the same shape:
switch between deep-text search and bucket filter (active /
deactive / depleted / expiring / online) + protocol filter that
matches clients attached to at least one inbound with the chosen
protocol. State persists in clientsFilterState localStorage.
filteredClients drives both the desktop table and the mobile card
list, and select-all / allSelected / someSelected only span the
visible subset.
- useClients now also fetches expireDiff and trafficDiff from
/panel/setting/defaultSettings (used to detect the expiring
bucket); ClientsPage threads them into the client-bucket helper.
Loose fixes folded in:
- Add Client: email field is auto-filled with a random handle on
open, matching uuid/subId/password/auth.
- Inbound clone: parse and reuse the source settings JSON (with
clients reset to []) instead of building a fresh defaulted
Settings, so VLESS Encryption/Decryption and other non-client
fields survive the clone.
- en-US.json: add the ipLog string used by the edit-client modal.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reverse tag field for VLESS-attached clients
Mirrors the Flow field's pattern: a Reverse tag input appears in the
Add/Edit Client modal whenever at least one selected inbound is VLESS
or PortFallback. The value rides over the wire as
client.reverse = { tag: '...' } so it lands directly in model.Client's
*ClientReverse field; an empty value omits the reverse key entirely.
On edit the field is hydrated from props.client.reverse?.tag, and the
showReverseTag watcher clears the field if the user drops the last
VLESS-like inbound from the selection.
* fix(xray): emit only protocol-relevant fields per client entry
The Xray config synthesizer was writing every identifier field (id,
password, flow, auth, security/method, reverse) on every client entry
regardless of the inbound's protocol. Xray ignores unknown fields, so
the config worked, but it diverged from the spec and leaked secrets
across protocols when one client was attached to multiple inbounds —
a VLESS inbound's generated config carried the same client's Trojan
password and Hysteria auth alongside its uuid.
Switch on inbound.Protocol when building each entry:
- VLESS / PortFallback: id, flow, reverse
- VMess: id, security
- Trojan: password, flow
- Shadowsocks: password, method
- Hysteria / Hysteria2: auth
email is emitted for every protocol.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): restore auto-disable kick under new schema
disableInvalidClients still resolved (inbound_tag, email) pairs via
JSON_EACH(inbounds.settings.clients), which is empty after migrating
to the clients + client_inbounds tables. Result: xrayApi.RemoveUser
never ran for depleted clients, clients.enable stayed true so the UI
showed them as active, and only xray_client_traffic.enable got flipped
- making "Restart Xray After Auto Disable" only half-work.
Resolve the targets via a JOIN through the new schema, flip clients.enable
so the Clients page reflects the state, and drop the legacy JSON
write-back plus the subId cascade workaround (email is unique now).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): live WebSocket updates + Ended status surfacing
ClientsPage now subscribes to traffic / client_stats / invalidate
WebSocket events instead of polling /onlines every 10s. Per-row
traffic counters refresh in place, online state stays current, and
list-level mutations elsewhere trigger a refresh.
The client roll-up summary moves from InboundsPage to ClientsPage
where it belongs, restructured into six labeled stat tiles
(Total / Online / Ended / Expiring / Disabled / Active) with email
popovers on the ones with issues.
Auto-disabled clients (traffic exhausted or expiry passed) now
classify as 'depleted' even though clients.enable=false, so they
show up under the Ended filter and render a red Ended tag instead
of looking indistinguishable from an operator-disabled row.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(nodes): per-node client roll-up and panel version
Added transient inboundCount / clientCount / onlineCount /
depletedCount fields to model.Node, populated by NodeService.GetAll
via aggregated queries (one join across inbounds + client_inbounds,
one over client_traffics intersected with the in-memory online
emails). The Nodes list renders these as colored chips on a new
"Clients" column so an operator can see at a glance how many users
each node carries and how many are currently online or depleted.
Also exposes the remote panel's version. The central panel adds
panelVersion to its /api/server/status payload (sourced from
config.GetVersion). Probe reads that field and persists it on the
node row, mirroring how xrayVersion already flows. NodesPage gets
a new column next to Xray Version, in both desktop and mobile
views, with English and Persian strings.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): stop node sync from resurrecting deleted clients
Several related issues around node-managed clients:
- Remote runtime: drop the per-inbound resetAllClientTraffics path
and point traffic/onlines/lastOnline fetches at the new
/panel/api/clients/* routes.
- Delete from master: always push the updated inbound to the node
even when the client was already disabled or depleted, so the
node actually loses the user instead of silently keeping it.
- setRemoteTraffic: mirror remote clients into the central tables
only on first discovery of a node inbound. Matched inbounds let
the master own the join table, so a stale snap can no longer
re-create a ClientRecord (and join row) for a client that was
just deleted on the master.
- ClientService.Delete: route through submitTrafficWrite so deletes
serialize with node traffic merges, and switch the final
ClientRecord delete to an explicit Where("id = ?") clause.
- setRemoteTraffic UNIQUE-constraint fix: use clause.OnConflict on
inserts and email-keyed UPDATEs for client_traffics, so mirroring
a snap doesn't trip the unique email index.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): switch client API endpoints from id to email
All client-scoped routes now use the unique email as the path key
(get, update, del, attach, detach, links). Email is the stable,
protocol-independent identifier — UUIDs don't exist for trojan or
shadowsocks, and internal numeric ids leaked panel implementation
detail into the public API.
Removed the redundant /traffic/byId/:id endpoint (covered by
/traffic/:email) and collapsed /links/:id/:email into /links/:email,
which now returns links across every attached inbound for the client.
Frontend selection, bulk delete, and toggle state are now keyed by
email as well, dropping the id→email lookup workaround.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(server): move cached state and helpers into ServerService
ServerController had grown to hold its own status cache, version-list
TTL cache, history-bucket whitelist, and the loop that drove all three
— concerns that belong in the service layer. Pull them out:
- lastStatus + the @2s refresh become ServerService.RefreshStatus and
ServerService.LastStatus; the controller's cron now just orchestrates
the cross-service side effects (xrayMetrics sample, websocket broadcast).
- The 15-minute Xray-versions cache (with stale-on-error fallback) moves
into ServerService.GetXrayVersionsCached, collapsing the controller
handler to a single call.
- The freedom/blackhole outbound-tag walk used by /xraylogs becomes
ServerService.GetDefaultLogOutboundTags.
- The allowed-history-bucket whitelist moves to package-level
service.IsAllowedHistoryBucket, so both NodeController and
ServerController validate against the same list.
Net result: web/controller/server.go drops from 458 to 365 lines and
contains only HTTP wiring + presentation-y side effects.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): emit JSON-text columns as nested objects
Inbound, ClientRecord, and InboundClientIps store settings /
streamSettings / sniffing / reverse / ips as JSON-text in the DB. The
API was passing that text through verbatim, so every consumer had to
JSON.parse a string inside a string. Add MarshalJSON / UnmarshalJSON so
the wire format is a real nested object, while still accepting the
legacy escaped-string shape on write. Frontend dbinbound.js gets a
matching coerceInboundJsonField helper for the same dual-shape read
path, and inbound.js toJson stops emitting empty/placeholder fields
(externalProxy [], sniffing destOverride when disabled, etc.) so the
new normalised JSON stays terse. api-docs and the inbound-clone path
are updated to the new shape. Controller route lists are regrouped so
all GETs sit above POSTs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): include inboundIds and traffic in /clients/list
ClientRecord got its own MarshalJSON in the previous commit, and
ClientWithAttachments embeds it to add inboundIds and traffic. Go
promotes the embedded MarshalJSON to the outer struct, so the encoder
was calling ClientRecord.MarshalJSON for the whole value and silently
dropping the extras. The frontend reads row.inboundIds / row.traffic
from /clients/list, so attached inbounds didn't render and newly added
clients looked like they hadn't saved. Add an explicit MarshalJSON on
ClientWithAttachments that splices the extras in.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): gate IP Log on ipLimitEnable + clean access-log dropdown
Legacy panel hid the IP Log section when access logging was off; the
Vue 3 migration left it gated on isEdit only, so the section showed
even when xray's access log was 'none' and nothing was being recorded.
Restore the ipLimitEnable gate on the edit modal's IP Log form-item.
While here, clean up the Xray Settings access-log dropdown: previously
two 'none' entries appeared (an empty value labelled with t('none') and
the literal 'none' from the options array). Drop the empty option for
access log (the literal 'none' covers it) and relabel the empty option
for error log / mask address to t('empty') so they're distinguishable.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(nodes): route per-client ops through node clients API + orphan sweep
Adds Runtime methods AddClient, UpdateUser, and DeleteUser so master
mutates clients on a node via /panel/api/clients/{add,update,del} rather
than pushing the whole inbound. The previous rt.UpdateInbound path made
the node DelInbound+AddInbound on every single-client change, briefly
cycling every other user on the same inbound.
DelInbound no longer filters by enable=true, so a disabled node inbound
actually gets removed from the node instead of being resurrected by the
next snap.
setRemoteTrafficLocked now sweeps any ClientRecord with zero
ClientInbound rows after SyncInbound rebuilds the attachments, which is
how a node-side delete propagates back to master instead of leaving a
detached ghost. ClientService.Delete tombstones the email first so a
snap arriving mid-delete can't re-create the record.
WebSocket broadcasts an "invalidate(clients)" message on every client
mutation so the Clients page refreshes without manual reload.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(balancers): allow fallback on all strategies + feed burstObservatory from random/roundRobin
Drops the random/roundRobin gate on the Fallback field in
BalancerFormModal so every strategy can pick a fallback outbound.
syncObservatories now feeds burstObservatory from leastLoad +
random + roundRobin balancers (was leastLoad only), matching how
leastPing feeds observatory.
Fix the JsonEditor "Unexpected end of JSON input" that appeared
when switching a balancer between leastPing and another strategy:
the obsView watcher was gated on showObsEditor (a boolean OR of
the two flags) and missed the case where one observatory
swapped for the other in the same tick. Watch the individual
flags instead so obsView flips to the surviving editor and the
getter stops pointing at a deleted key.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): use sortedInbounds for mobile empty-state check
InboundList referenced an undefined visibleInbounds in the mobile
card list's empty-state guard, throwing "Cannot read properties of
undefined (reading 'length')" and breaking the entire mobile render.
* feat(clients): sortable table columns
Adds the same sortState / sortableCol / sortFns pattern InboundList
uses, wrapping filteredClients in sortedClients so sort composes with
the existing search/filter pipeline. Sortable: enable, email,
inboundIds (attachment count), traffic, remaining, expiryTime;
actions and online stay unsorted.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(shadowsocks): generate valid ss2022 keys and per-client method for legacy ciphers
The Add Client flow on shadowsocks inbounds was producing xray configs
that failed to start:
- 2022-blake3-* ciphers need a base64-encoded key of an exact byte
length per cipher. fillProtocolDefaults was assigning a uuid-style
string, which xray rejects as "bad key". Now the password is
generated (or replaced if invalid) via random.Base64Bytes(n) sized
to the chosen cipher.
- Legacy ciphers (aes-256-gcm, chacha20-*, xchacha20-*) require a
per-client method field in multi-user mode; model.Client has no
Method, so settings.clients was stored without one and xray failed
with "unsupported cipher method:". applyShadowsocksClientMethod
now injects the top-level method into each client on add/update,
and healShadowsocksClientMethods backfills it at xray-config-build
time so existing inbounds heal on the next start.
- xray/api.go ssCipherType switch was missing aes-256-gcm, which
fell through to ss2022 path.
- SSMethods dropdown now offers aes-256-gcm.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): preserve ClientRecord on inbound delete + filter Attached inbounds to multi-client protocols
Replace the global orphan sweep in setRemoteTrafficLocked with a
per-inbound diff cleanup: only delete a ClientRecord whose email
disappeared from a snap-tracked inbound (i.e. a node-side delete).
Inbounds that vanished entirely from the snap (e.g. admin deleted
the inbound on master) aren't iterated, so a client whose last
attachment came from that inbound is now left alone instead of
being deleted alongside the inbound.
ClientFormModal and ClientBulkAddModal now filter the Attached
inbounds dropdown to protocols that actually support multiple
clients: shadowsocks, vless, vmess, trojan, hysteria, hysteria2,
and portfallback (which routes through VLESS settings).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): make empty-state text readable on dark/ultra themes
The "No clients yet" empty state had a hardcoded black color
(rgba(0,0,0,0.45)) that vanished against the dark backgrounds.
Drop the inline color, let it inherit from the AntD theme, and
fade with opacity like the mobile card empty state already does.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): client-first tgbot add flow, tgId field, lightweight inbound options
- tgbot: drop legacy per-protocol Add Client UI in favour of a client-first
multi-inbound flow. New BuildClientDraftMessage / getInboundsAttachPicker
let an admin pick one or more inbounds and submit a single client; per-
protocol secrets are now generated server-side via fillProtocolDefaults.
Drops awaiting_id/awaiting_password_tr/awaiting_password_sh state cases
and add_client_ch_default_id/pass_tr/pass_sh/flow callbacks. Adds a
setTGUser button + awaiting_tg_id state so the bot can set Client.TgID
during Add.
- clients UI: add Telegram user ID input to ClientFormModal (0 = none).
Hide IP Limit field entirely when ipLimitEnable is off — disabled fields
still take layout space, this collapses Auth(Hysteria) to full width.
- inbounds API: new GET /panel/api/inbounds/options that returns just
{id, remark, protocol, port, tlsFlowCapable}. Used by the clients page
pickers so the dropdown payload stays small on panels with thousands of
clients (drops settings JSON, clientStats, streamSettings). Server-side
TlsFlowCapable mirrors Inbound.canEnableTlsFlow so the modal no longer
needs to parse streamSettings client-side.
- clientInfoMsg now shows attached inbound remarks, and getInboundUsages
reports the attached client count per inbound.
- api-docs: document the new /options endpoint and add tgId / flow to the
clients add/update bodies.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): keep Node column visible for node-attached inbounds
The Node column was bound to hasActiveNode, so disabling every node hid
the column even when inbounds were still attached to those nodes — the
admin lost the visual cue that those inbounds belonged to a node and
would come back when it was re-enabled. Combine hasActiveNode with a
new hasNodeAttachedInbound check (any dbInbound with nodeId != null) so
the column survives node-disable.
* fix(api-docs): accept functional-component icons in EndpointSection
AntD-Vue icons (SafetyCertificateOutlined, etc.) are functional
components, so the icon prop's type: Object validator was rejecting
them with a "Expected Object, got Function" warning at runtime.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* test: cover crypto, random, netsafe, sub helpers, xray equals, websocket hub, node service
Adds ~110 unit tests across previously untested packages. Focus on
pure-logic and concurrency surfaces where regressions would silently
affect users:
- util/crypto, util/random: password hashing round-trip, ss2022 key
generation, alphabet/length invariants.
- util/netsafe: IsBlockedIP edge cases, NormalizeHost validation,
SSRF guard with AllowPrivate context bypass.
- util/common, util/json_util: traffic formatter, Combine nil-skip,
RawMessage empty-as-null and copy-on-unmarshal.
- sub: splitLinkLines, searchKey/searchHost, kcp share fields,
finalmask normalization, buildVmessLink round-trip.
- xray: Config.Equals and InboundConfig.Equals field-by-field,
getRequiredUserString/getOptionalUserString type checks.
- web/websocket: hub registration, throttling, slow-client eviction,
nil-receiver safety, concurrent register/unregister.
- web/service: NodeService.normalize validation, normalizeBasePath,
HeartbeatPatch.ToUI mapping.
- web/job: atomicBool concurrent set/takeAndReset semantics.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n(clients): replace English fallbacks with proper translation keys
Pulls every hard-coded English label/title in the Clients page and its
four modals through the i18n layer so localized panels stop leaking
English. New keys live under pages.clients (auth, hysteriaAuth, uuid,
flow, flowNone, reverseTag, reverseTagPlaceholder, telegramId,
telegramIdPlaceholder, created, updated, ipLimit) plus refresh at the
root and toasts.bulkDeletedMixed / bulkCreatedMixed for partial-failure
toasts. Also switches the add-client modal's primary button from "Add"
to "Create" for consistency with other create flows.
The bulk-add Random/Random+Prefix/... email-method options stay
hard-coded by request - they're identifier-shaped strings.
* i18n: backfill 99 missing keys across all 12 non-English locales
Brings every translation file up to parity with en-US.json so the
Clients page, the fallback-children inbound section, the new refresh
verb, the Nodes panel-version label and a handful of older holes stop
falling through to the English fallback. New strings span:
- pages.clients.* (labels, confirmations, toasts, emailMethods)
- pages.inbounds.portFallback.* (Reality fallback inbound section)
- pages.nodes.panelVersion, menu.clients, refresh
Technical identifiers (Auth, UUID, Flow, Reverse tag) are intentionally
left untranslated since they correspond to xray-core field names.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n: drop stale pages.client block duplicated in every non-English locale
Every non-English locale carried a pages.client (singular) section with
30 entries that duplicated pages.clients (plural). The plural namespace
is what the Vue code actually consumes; the singular one was dead
weight from an older rename that never got cleaned up in the
non-English files. Removing it brings every locale to exactly 984
keys, matching en-US.json.
* chore: apply modernize analyzer fixes across codebase
Mechanical replacements suggested by golang.org/x/tools/.../modernize:
strings.Cut/CutPrefix/SplitSeq, slices.Contains, maps.Copy, min(),
range-over-int, new(expr), strings.Builder for hot += loops,
reflect.TypeFor[T](), sync.WaitGroup.Go(), drop legacy //+build lines.
* feat(database): add PostgreSQL as an optional backend alongside SQLite
Lets operators with large client counts or multi-node setups pick PostgreSQL
at install time without breaking the existing SQLite default. Backend is
selected at runtime via XUI_DB_TYPE/XUI_DB_DSN, a small dialect layer keeps
the five JSON_EXTRACT/JSON_EACH queries portable, and a new `x-ui migrate-db`
subcommand copies SQLite data into PostgreSQL in FK-aware order.
* fix(inbounds): gate node selector to multi-node-capable protocols
Hide the Deploy-To selector and clear nodeId when switching to a
protocol that can't run on a remote node. Also:
- subs: return 404 (not 400) when subId matches no inbounds, so VPN
clients distinguish "deleted/unknown" from a server error
- hysteria link gen: use the inbound's resolved address so node-managed
inbounds advertise the node host instead of the central panel
- shadowsocks: default network to 'tcp' (udp was causing issues for some
clients on first-create)
- vite dev proxy: rewrite migrated-route bypass against the live base
path instead of a hardcoded single-segment regex
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): bulk add/delete correctness + perf, working pagination, delayed-start in form
Bulk add/delete were serial on the frontend (one toast per call, N round-trips)
and the backend race exposed by parallelizing them lost client attachments and
hit UNIQUE constraint failed on client_inbounds. The single add/edit modal also
had no Start-After-First-Use option, and the table never showed the delayed
duration.
Backend (web/service/client.go):
- Per-inbound mutex on Add/Update/Del InboundClient so concurrent writers on
the same inbound don't lose the read-modify-write of settings JSON.
- SyncInbound skips create+join when the email is tombstoned so a concurrent
maintenance pass (adjustTraffics, autoRenewClients, markClientsDisabledIn-
Settings) that did a stale RMW can't resurrect a just-deleted client with a
fresh id.
- compactOrphans sweeps settings.clients entries whose ClientRecord no longer
exists, applied in Add/DelInboundClient + DelInboundClientByEmail so each
user-initiated mutation self-heals the inbound's settings.
- DelInboundClient uses Pluck instead of First for the stats lookup so a
missing row doesn't abort the delete with a noisy ErrRecordNotFound log.
Frontend:
- HttpUtil.{get,post} accept a silent option that suppresses the auto-toast.
- ClientBulkAddModal fires creates in parallel + silent + one summary toast.
- useClients.removeMany runs deletes in parallel + silent and refreshes once;
ClientsPage bulk delete uses it and shows one aggregate toast.
- useClients.applyInvalidate debounces 200 ms so the burst of N WebSocket
invalidate events from the backend collapses into a single refresh.
- ClientsPage pagination is reactive (paginationState ref + tablePagination
computed); onTableChange persists page-size and page changes.
- ClientFormModal gains a Start-After-First-Use switch + Duration days input
alongside the existing Expiry Date picker; on edit-mode open a negative
expiryTime is decoded back to delayed mode + days; on submit the payload
sends -86400000 * days or the absolute timestamp.
- ClientsPage table shows the delayed-start duration (blue tag Nd, tooltip
Start After First Use: Nd) instead of infinity.
- Telegram ID field in the form is hidden when /panel/setting/defaultSettings
reports tgBotEnable=false; Comment then fills the row.
- Form row 3 collapses UUID (span 12) + Total GB (span 8) + Limit IP (span 4)
when ipLimitEnable is on, else UUID + Total GB at 12/12.
- useInbounds.rollupClients counts only clients with a matching clientStats
row, so orphans in settings.clients no longer inflate the inbound's count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(windows): clean shutdown, working panel restart, harden kernel32 load
Three Windows-specific issues addressed:
1. Orphaned xray-windows-amd64 after VS Code debugger stop. Delve's
"Stop" sends TerminateProcess to the Go binary, which is uncatchable
— our signal handlers never run, so xrayService.StopXray() is skipped
and xray is left dangling. Spawn xray as a child of a Job Object with
JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE so the OS kills xray when our
handle to the job is closed (which happens even on TerminateProcess).
Also trap os.Interrupt in main so Ctrl+C in the terminal runs the
graceful path.
2. /panel/setting/restartPanel logged "failed to send SIGHUP signal: not
supported by windows" because Windows can't deliver arbitrary signals.
Add a restart hook in web/global; main registers it to push SIGHUP
into its own signal channel, and RestartPanel calls the hook before
falling back to the (Unix-only) signal path. Same restart-loop code
runs in both cases.
3. util/sys/sys_windows.go now uses windows.NewLazySystemDLL so the
kernel32.dll resolve is pinned to %SystemRoot%\System32 (prevents
DLL hijacking by a planted DLL next to the binary). Local filetime
type replaced with windows.Filetime, and the unreliable
syscall.GetLastError() fallback replaced with a type assertion on the
errno captured at call time.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(sys): correct CPU/connection accounting on linux + darwin
util/sys/sys_linux.go:
- GetTCPCount/GetUDPCount were counting the column header row in
/proc/net/{tcp,udp}[6] as a connection, inflating the reported total
by 1 per non-empty file (so the panel status line always showed 2
more connections than actually existed). Replace getLinesNum +
safeGetLinesNum with a single bufio.Scanner-based countConnections
that skips the header.
- CPUPercentRaw now opens HostProc("stat") instead of a hardcoded
/proc/stat so HOST_PROC overrides apply, matching the connection
counters in the same file.
- Simplify CPU field unpacking: pad nums to 8 once instead of guarding
every assignment with a len check.
util/sys/sys_darwin.go:
- Fix swapped idle/intr indices on kern.cp_time. BSD CPUSTATES order
is user, nice, sys, intr, idle (CP_INTR=3, CP_IDLE=4) — gopsutil's
cpu_darwin_nocgo.go reads the same layout. The previous code used
out[3] as idle and out[4] as intr, so busy = total - dIdle was
actually subtracting interrupt time, making the panel report CPU
usage close to 100% on macOS regardless of actual load.
- Collapse the per-field delta math into a single loop.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(xray): rotate crash reports into log folder, prevent overwrites
writeCrashReport had two flaws: it wrote to the bin folder (alongside the
xray binary) which conflates artifacts, and the second-precision timestamp
meant a tight restart-loop crash burst overwrote prior reports. Write to
the log folder with nanosecond precision and keep the last 10 reports.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* revert(inbounds): drop unreleased portfallback protocol
The Port-with-Fallback inbound (commit 62fd9f9d) was confusing as a
standalone protocol — fallbacks belong on a regular VLESS/Trojan TCP-TLS
inbound, the way Xray models them natively. Rip out the entire feature
cleanly (no migration needed since it was never released): protocol
constant, fallback children DB table, FallbackService, 2 API endpoints,
all UI rows, related translations and api-docs. A native fallback flow
attached to VLESS/Trojan TCP-TLS/Reality will land in a follow-up commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): native fallbacks on VLESS/Trojan TCP-TLS, with working child links
A VLESS or Trojan inbound on TCP with TLS or Reality can now act as a
fallback master: pick existing inbounds as children and the panel auto-
fills the SNI / ALPN / path / xver routing fields from each child's
transport, auto-builds settings.fallbacks at config-gen time, and
rewrites the child's client-share link so it advertises the master's
reachable endpoint and TLS state instead of the child's loopback listen.
Layout matches the Xray All-in-One Nginx example: master at :443 with
clients + TLS, each child on 127.0.0.1 with its own transport+clients.
Order matters (Xray walks fallbacks top-to-bottom) — reorder via the
per-row up/down arrows. Path / SNI / ALPN are exposed under a per-row
Edit toggle for the rare cases where the auto-derivation needs
overriding; otherwise just pick a child and you're done.
Backend: new InboundFallback table + FallbackService (GetByMaster /
SetByMaster / GetParentForChild / BuildFallbacksJSON); two routes
(GET / POST /panel/api/inbounds/:id/fallbacks); xray.GetXrayConfig
injects settings.fallbacks for any VLESS/Trojan TCP-TLS/Reality
inbound; GetInbounds annotates each child with FallbackParent so the
frontend can rewrite links without an extra round-trip.
Link projection covers every emission path — clients-page QR/links,
per-inbound Get URL, raw subscription, sub-JSON, sub-Clash, and the
inbounds-page link/info/QR — via a shared projectThroughFallbackMaster
on the backend and a shared projectChildThroughMaster on the frontend
that both handle the panel-tracked relationship and the legacy
unix-socket (@vless-ws) convention.
Strings translated into all 12 non-English locales.
* docs: rewrite CONTRIBUTING with full local-dev setup
The prior three-line CONTRIBUTING left newcomers guessing at every
non-trivial step: which Go / Node versions, where xray comes from, why
the panel goes blank when XUI_DEBUG=true is flipped on, how the Vue
multi-page setup is wired, what to do on Windows when go build trips
on the CGo SQLite driver.
Now covers prerequisites, MinGW-w64 install on Windows (niXman builds
or MSYS2), one-shot first-time setup, two frontend dev workflows with
the XUI_DEBUG asset-cache gotcha called out, the architecture and
conventions of the Vue side, a project-layout map, useful env vars,
and the PR checklist.
---------
2026-05-19 03:16:42 -07:00
else
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
read -rp "Would you like to customize the Panel Port settings? (If not, a random port will be applied) [y/n]: " config_confirm
if [ [ " ${ config_confirm } " = = "y" || " ${ config_confirm } " = = "Y" ] ] ; then
read -rp "Please set up the panel port: " config_port
echo -e " ${ yellow } Your Panel Port is: ${ config_port } ${ plain } "
else
config_port = $( shuf -i 1024-62000 -n 1)
echo -e " ${ yellow } Generated random port: ${ config_port } ${ plain } "
fi
2026-05-19 04:09:35 -07:00
fi
${ xui_folder } /x-ui setting -username " ${ config_username } " -password " ${ config_password } " -port " ${ config_port } " -webBasePath " ${ config_webBasePath } "
echo ""
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } SSL Certificate Setup (RECOMMENDED) ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ yellow } SSL is strongly recommended. Skip only if a reverse proxy ${ plain } "
echo -e " ${ yellow } or SSH tunnel handles TLS for you. ${ plain } "
echo -e " ${ yellow } Let's Encrypt now supports both domains and IP addresses! ${ plain } "
echo ""
prompt_and_setup_ssl " ${ config_port } " " ${ config_webBasePath } " " ${ server_ip } "
# Retrieve the API token for display
local config_apiToken = $( ${ xui_folder } /x-ui setting -getApiToken true | grep -Eo 'apiToken: .+' | awk '{print $2}' )
# Display final credentials and access information
echo ""
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } Panel Installation Complete! ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } Username: ${ config_username } ${ plain } "
echo -e " ${ green } Password: ${ config_password } ${ plain } "
echo -e " ${ green } Port: ${ config_port } ${ plain } "
echo -e " ${ green } WebBasePath: ${ config_webBasePath } ${ plain } "
2026-05-23 10:52:37 -07:00
echo -e " ${ green } Database: ${ db_label } ${ plain } "
2026-05-19 04:09:35 -07:00
echo -e " ${ green } Access URL: ${ SSL_SCHEME } :// ${ SSL_HOST } : ${ config_port } / ${ config_webBasePath } ${ plain } "
echo -e " ${ green } API Token: ${ config_apiToken } ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ yellow } ⚠ IMPORTANT: Save these credentials securely! ${ plain } "
if [ [ " $SSL_SCHEME " = = "https" ] ] ; then
echo -e " ${ yellow } ⚠ SSL Certificate: Enabled and configured ${ plain } "
else
echo -e " ${ yellow } ⚠ SSL Certificate: Skipped — panel is HTTP-only. Use a reverse proxy or SSH tunnel. ${ plain } "
Feat/multi inbound clients (#4469)
* feat(clients): add shadow tables for first-class client promotion
Introduces three new GORM-backed tables (clients, client_inbounds,
inbound_fallback_children) and a populate-only seeder that backfills
them from each inbound's existing settings.clients JSON. Duplicate
emails across inbounds auto-merge under one client row, with each
field conflict logged. Existing services are unchanged and continue
reading from settings.clients — this commit is groundwork only.
* feat(clients): make clients+client_inbounds the runtime source of truth
Adds ClientService.SyncInbound that reconciles the new tables from
each inbound's clients list whenever existing service paths mutate
settings.clients. Wires it into AddInbound, UpdateInbound,
AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, DelDepletedClients, autoRenewClients, and
the timestamp-backfill path in adjustTraffics, plus DetachInbound
on DelInbound.
GetXrayConfig now builds settings.clients from the new tables before
writing config.json, and getInboundsBySubId joins through them
instead of JSON_EACH on settings JSON. Live Xray config and
subscription endpoints are now driven by the relational view;
settings.clients JSON stays in step as a side effect of every write.
* feat(clients): add top-level Clients tab and CRUD API
Adds /panel/api/clients endpoints (list, get, add, update, del,
attach, detach) backed by ClientService methods that orchestrate
the per-inbound Add/Update/Del flows so a single client row is
created once and attached to many inbounds in one operation.
The frontend gains a dedicated Clients page (frontend/clients.html
+ src/pages/clients/) with an AntD table, multi-inbound attach
modal, and full CRUD. Axios interceptor learns to honour
Content-Type: application/json so the JSON endpoints work
alongside the legacy form-encoded ones.
The legacy per-inbound client modal stays untouched in this PR —
both flows now write to the same source of truth.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): add Port-with-Fallback inbound type
Adds a new "portfallback" protocol that emits as a VLESS-TLS inbound
under the hood but is paired with a sidecar table of child inbounds.
Panel auto-builds settings.fallbacks at Xray-config-gen time from the
sidecar — each child's listen+port becomes the fallback dest, with
SNI/ALPN/path/xver match criteria pulled from the row. No more typing
loopback ports by hand or keeping settings.fallbacks in sync.
Backend: new FallbackService (Get/SetChildren, BuildFallbacksJSON);
two new routes (GET/POST /panel/api/inbounds/:id/fallbackChildren);
xray.GetXrayConfig injects fallbacks for PortFallback inbounds; the
inbound model emits protocol="vless" so Xray accepts the config.
Frontend: PORTFALLBACK joins the protocol dropdown; selecting it
shows the standard VLESS controls plus a Fallback Children table
(inbound picker + per-row SNI/ALPN/path/xver). Children are loaded
on edit and replaced atomically on save.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reset Traffic, QR Code, Info actions + Online/Remaining columns
The Clients page table gains:
- Online column — green/grey tag driven by /panel/api/inbounds/onlines,
polled every 10s.
- Remaining column — bytes-remaining tag, coloured green/orange/red
against quota, purple infinity when unlimited.
- Action icons per row: QR, Info, Reset traffic, Edit, Delete.
ClientInfoModal shows the full client detail (uuid/password/auth,
traffic ↑/↓ + remaining + all-time, expiry absolute + relative,
attached inbounds chip list, online + last-online).
ClientQrModal fetches links for the client's subId via
/panel/api/inbounds/getSubLinks/:subId and renders each one through
the existing QrPanel component.
Reset Traffic confirms then calls the existing per-inbound endpoint
on the client's first attached inbound (the traffic row is keyed on
email globally, so any attached inbound resets the shared counter).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): expose Attached inbounds in edit mode
The multi-select was gated on add-only, so editing a client had no way
to change which inbounds it belonged to. The picker now shows in both
modes, and on submit the modal diffs the picked set against the
original attachedIds — additions go through the /attach endpoint,
removals through /detach, both after the field update lands so the
new attachments get the latest values.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): unbreak template parsing + stale i18n keys
- InboundFormModal: split the multi-line help string in the
PortFallback section onto one line — Vue's template parser was
bailing on Unterminated string constant because a single-quoted
literal spanned two lines inside a {{ }} interpolation.
- ClientInfoModal: t('disable') was missing at the root level, so
vue-i18n returned the key path literally. Use t('disabled') which
exists.
- Linter cleanup elsewhere: pages.client.* references renamed to
pages.clients.* to match the merged i18n block; whitespace
normalisation in a few unrelated Vue templates.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* 1
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(traffic): drop all-time traffic tracking
Removes the AllTime field from Inbound and ClientTraffic and migrates
existing DBs by dropping the all_time columns on startup. The counter
duplicated up+down without adding signal, and the per-event accumulator
ran on every traffic write.
Frontend: drop the All-time column from the inbound list and the
client-row table, the All-time row from the client info modal, and the
All-Time Total Usage tile from the inbounds summary card. The
allTimeTraffic/allTimeTrafficUsage i18n keys are removed across every
locale.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): mobile cards, multi-select, bulk add
Adds the same row-card layout the inbounds page uses on mobile: the
table is suppressed under the mobile breakpoint and each client renders
as a compact card with a status dot, email, Info button, Enable switch,
and overflow menu. All the per-client detail (traffic, remaining,
expiry, attached inbounds, flow, created/updated, URL, subscription)
opens through the existing info modal.
Multi-select with bulk delete wires AntD row-selection on desktop and
a per-card checkbox on mobile; a Delete (N) button appears in the
toolbar when anything is selected.
Bulk add reuses the five email-generation modes from the inbound bulk
modal but takes a multi-inbound picker so one bulk run can attach to
several inbounds at once. Submits client-by-client through the
existing /panel/api/clients/add endpoint.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): remove legacy per-inbound client UI
Now that clients live as first-class rows attached to one or many
inbounds, the per-inbound client UI on the inbounds page is dead
weight — every client action either has a global equivalent on the
Clients page or makes no sense in a many-to-many world.
Deletes ClientFormModal, ClientBulkModal, CopyClientsModal, and
ClientRowTable from inbounds/. Strips the matching emits, refs,
handlers, and dropdown menu items from InboundList and InboundsPage,
and removes the dead mobile expand-chevron state and the desktop
expanded-row plumbing that drove the inline client table.
The InboundFormModal Clients tab still works in add-mode (one inline
client at inbound creation) — that flow goes through ClientService.
SyncInbound on save and remains useful.
Fixes a stray "</a-dropdown>" left over by an earlier toolbar edit
in ClientsPage that broke the template parser.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Delete depleted action
Mirrors the legacy delDepletedClients action that lived under the
inbounds page, but as a first-class /panel/api/clients/delDepleted
endpoint backed by ClientService. The new path goes through
ClientService.Delete for each depleted email, so the new clients +
client_inbounds + xray_client_traffic tables stay consistent.
Adds a danger-styled toolbar button on the Clients page (next to
Reset all client traffic) with a confirm dialog and a toast
reporting the deleted count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): move every client-shaped endpoint off /inbounds onto /clients
After the multi-inbound client migration, client state belongs to the
client API surface, not the inbound one. Twelve routes that were
crammed under /panel/api/inbounds/* now live where they belong, under
/panel/api/clients/*.
Moved (route, handler, doc):
POST /clientIps/:email
POST /clearClientIps/:email
POST /onlines
POST /lastOnline
POST /updateClientTraffic/:email
POST /resetAllClientTraffics/:id
POST /delDepletedClients/:id
POST /:id/resetClientTraffic/:email
GET /getClientTraffics/:email
GET /getClientTrafficsById/:id
GET /getSubLinks/:subId
GET /getClientLinks/:id/:email
Their /clients/* counterparts are:
POST /clients/clientIps/:email
POST /clients/clearClientIps/:email
POST /clients/onlines
POST /clients/lastOnline
POST /clients/updateTraffic/:email
POST /clients/resetTraffic/:email (email-only, fans out)
GET /clients/traffic/:email
GET /clients/traffic/byId/:id
GET /clients/subLinks/:subId
GET /clients/links/:id/:email
per-inbound resetAllClientTraffics and delDepletedClients are dropped
entirely — the Clients page already exposes global Reset All Traffic
and Delete depleted actions, and per-inbound resets are meaningless
once a client can be attached to many inbounds.
ClientService.ResetTrafficByEmail is the new email-only reset path:
it looks up every inbound the client is attached to and pushes the
counter reset + Xray re-add through inboundService.ResetClientTraffic
for each one, so depleted users come back online instantly.
Frontend callers (ClientsPage, useClients, ClientQrModal,
ClientInfoModal, InboundInfoModal, InboundsPage, useInbounds) all
switched to the new paths. The Inbounds page drops its per-inbound
"Reset client traffic" and "Delete depleted clients" dropdown items —
users do those at the client level now. api-docs is rebuilt to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): switch tgbot + ldap callers to ClientService
Adds two thin helpers to ClientService (CreateOne, DetachByEmail) and
rewrites tgbot.SubmitAddClient and ldap_sync_job to call ClientService
directly. Removes the JSON-blob payloads (BuildJSONForProtocol output for
add, clientsToJSON/clientToJSON helpers) that callers previously fed to
InboundService.AddInboundClient/DelInboundClient.
ldap_sync_job.batchSetEnable now loops InboundService.SetClientEnableByEmail
per email instead of trying to coerce AddInboundClient into doing the
update — the old path would have failed duplicate-email validation for
existing clients anyway.
The legacy InboundService.AddInboundClient/UpdateInboundClient/
DelInboundClient methods stay in place; they are now only used internally
by ClientService Create/Update/Delete/Attach. Inlining + deleting them
follows in a separate commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): move all client mutation methods to ClientService
Moves the client mutation surface out of InboundService and into
ClientService. These methods all operate on a single client (identity
fields, traffic limits, expiry, ip limit, enable state, telegram tg id)
and didn't belong on the inbound aggregate.
Moved (12 methods): AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, checkEmailsExistForClients, SetClientTelegramUserID,
checkIsEnabledByEmail, ToggleClientEnableByEmail, SetClientEnableByEmail,
ResetClientIpLimitByEmail, ResetClientExpiryTimeByEmail,
ResetClientTrafficLimitByEmail.
Each method now takes an explicit *InboundService for the helpers that
legitimately stay on InboundService (GetInbound, GetClients, runtimeFor,
AddClientStat / UpdateClientStat / DelClientStat, DelClientIPs /
UpdateClientIPs, emailUsedByOtherInbounds, getAllEmailSubIDs,
GetClientInboundByEmail / GetClientInboundByTrafficID,
GetClientTrafficByEmail).
Stays on InboundService: ResetClientTrafficByEmail and
ResetClientTraffic(id, email) — these mutate xray_client_traffic rows,
not client identity, so they're inbound-side bookkeeping.
Callers updated: tgbot (6 calls), ldap_sync_job (1 call),
InboundService internal (writeBackClientSubID, CopyInboundClients,
AddInbound's email-uniqueness check), ClientService Create/Update/
Delete/Attach/Detach.
Also removes a dead resetAllClientTraffics controller handler whose
route was already gone after the previous /clients API migration.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): finish migrating to ClientService + tidy IP routes
Two related cleanups in the new /clients surface:
1. Move ResetAllClientTraffics (bulk-reset of xray_client_traffic +
last_traffic_reset_time, with node-runtime propagation) from
InboundService to ClientService. PeriodicTrafficResetJob now holds
a clientService and calls
j.clientService.ResetAllClientTraffics(&j.inboundService, id).
The last client-mutation method on InboundService is gone.
2. Shorten redundantly-named routes/handlers under /panel/api/clients:
- /clientIps/:email -> /ips/:email (handler getIps)
- /clearClientIps/:email -> /clearIps/:email (handler clearIps)
The "client" prefix was redundant inside the clients namespace.
Frontend (InboundInfoModal) and api-docs updated to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds,clients): clean up inbound modal + enrich client modal
Inbound modal rework (InboundFormModal.vue + inbound.js):
- Drop the embedded Client subform in the Protocol tab. Multi-inbound
clients are managed exclusively from the Clients page now; a fresh
inbound is created with zero clients (settings constructors default
to []) and the user attaches clients afterwards.
- Hide the Protocol tab entirely when it has nothing to render
(VMESS, Trojan without fallbacks, Hysteria). Auto-switches active
tab to Basic when the tab disappears while focused.
- Move the Security section (Security selector + TLS block with certs
and ECH + Reality block) out of the Stream tab into its own
Security tab, sharing the canEnableStream gate.
Client modal additions (ClientFormModal.vue + ClientBulkAddModal.vue):
- Flow select (xtls-rprx-vision / -udp443) appears only when the
panel actually has a Vision-capable inbound (VLESS or PortFallback
on TCP with TLS or Reality). Hidden otherwise, and cleared when
it disappears.
- IP Limit input is disabled when the panel-level ipLimitEnable
setting is off, fetched into useClients alongside subSettings and
threaded through ClientsPage to both modals.
- Edit modal now shows an "IP Log" section listing IPs that have
connected with the client's credentials, with refresh and clear
buttons (calls the renamed /panel/api/clients/ips and /clearIps
endpoints).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): drop manual Fallbacks UI from inbound modal
The PortFallback protocol type now covers the common
VLESS-master-plus-children case with auto-wired dests, so the manual
Fallbacks editor (showFallbacks block in the Protocol tab) is mostly
redundant. Removed:
- the v-if="showFallbacks" template block (SNI/ALPN/Path/dest/PROXY rows)
- the showFallbacks computed
- the addFallback / delFallback helpers
- the .fallbacks-header / .fallbacks-title styles
- the showFallbacks gate from hasProtocolTabContent (so Trojan-over-TCP
no longer shows an empty Protocol tab)
Power users who need a non-inbound fallback dest (nginx, static site)
can still author settings.fallbacks via the Advanced JSON tab.
* feat(clients,inbounds): move search/filter to Clients page + small fixes
Search/filter relocation:
- Remove the search/filter toolbar (search switch + filter radio +
protocol/node selects + the visibleInbounds projection +
inboundsFilterState localStorage + filter CSS + the SearchOutlined/
FilterOutlined/ObjectUtil/Inbound imports it required) from
InboundList. The filters were all client-oriented buckets bolted
onto the inbound row.
- Add a search/filter toolbar to ClientsPage with the same shape:
switch between deep-text search and bucket filter (active /
deactive / depleted / expiring / online) + protocol filter that
matches clients attached to at least one inbound with the chosen
protocol. State persists in clientsFilterState localStorage.
filteredClients drives both the desktop table and the mobile card
list, and select-all / allSelected / someSelected only span the
visible subset.
- useClients now also fetches expireDiff and trafficDiff from
/panel/setting/defaultSettings (used to detect the expiring
bucket); ClientsPage threads them into the client-bucket helper.
Loose fixes folded in:
- Add Client: email field is auto-filled with a random handle on
open, matching uuid/subId/password/auth.
- Inbound clone: parse and reuse the source settings JSON (with
clients reset to []) instead of building a fresh defaulted
Settings, so VLESS Encryption/Decryption and other non-client
fields survive the clone.
- en-US.json: add the ipLog string used by the edit-client modal.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reverse tag field for VLESS-attached clients
Mirrors the Flow field's pattern: a Reverse tag input appears in the
Add/Edit Client modal whenever at least one selected inbound is VLESS
or PortFallback. The value rides over the wire as
client.reverse = { tag: '...' } so it lands directly in model.Client's
*ClientReverse field; an empty value omits the reverse key entirely.
On edit the field is hydrated from props.client.reverse?.tag, and the
showReverseTag watcher clears the field if the user drops the last
VLESS-like inbound from the selection.
* fix(xray): emit only protocol-relevant fields per client entry
The Xray config synthesizer was writing every identifier field (id,
password, flow, auth, security/method, reverse) on every client entry
regardless of the inbound's protocol. Xray ignores unknown fields, so
the config worked, but it diverged from the spec and leaked secrets
across protocols when one client was attached to multiple inbounds —
a VLESS inbound's generated config carried the same client's Trojan
password and Hysteria auth alongside its uuid.
Switch on inbound.Protocol when building each entry:
- VLESS / PortFallback: id, flow, reverse
- VMess: id, security
- Trojan: password, flow
- Shadowsocks: password, method
- Hysteria / Hysteria2: auth
email is emitted for every protocol.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): restore auto-disable kick under new schema
disableInvalidClients still resolved (inbound_tag, email) pairs via
JSON_EACH(inbounds.settings.clients), which is empty after migrating
to the clients + client_inbounds tables. Result: xrayApi.RemoveUser
never ran for depleted clients, clients.enable stayed true so the UI
showed them as active, and only xray_client_traffic.enable got flipped
- making "Restart Xray After Auto Disable" only half-work.
Resolve the targets via a JOIN through the new schema, flip clients.enable
so the Clients page reflects the state, and drop the legacy JSON
write-back plus the subId cascade workaround (email is unique now).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): live WebSocket updates + Ended status surfacing
ClientsPage now subscribes to traffic / client_stats / invalidate
WebSocket events instead of polling /onlines every 10s. Per-row
traffic counters refresh in place, online state stays current, and
list-level mutations elsewhere trigger a refresh.
The client roll-up summary moves from InboundsPage to ClientsPage
where it belongs, restructured into six labeled stat tiles
(Total / Online / Ended / Expiring / Disabled / Active) with email
popovers on the ones with issues.
Auto-disabled clients (traffic exhausted or expiry passed) now
classify as 'depleted' even though clients.enable=false, so they
show up under the Ended filter and render a red Ended tag instead
of looking indistinguishable from an operator-disabled row.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(nodes): per-node client roll-up and panel version
Added transient inboundCount / clientCount / onlineCount /
depletedCount fields to model.Node, populated by NodeService.GetAll
via aggregated queries (one join across inbounds + client_inbounds,
one over client_traffics intersected with the in-memory online
emails). The Nodes list renders these as colored chips on a new
"Clients" column so an operator can see at a glance how many users
each node carries and how many are currently online or depleted.
Also exposes the remote panel's version. The central panel adds
panelVersion to its /api/server/status payload (sourced from
config.GetVersion). Probe reads that field and persists it on the
node row, mirroring how xrayVersion already flows. NodesPage gets
a new column next to Xray Version, in both desktop and mobile
views, with English and Persian strings.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): stop node sync from resurrecting deleted clients
Several related issues around node-managed clients:
- Remote runtime: drop the per-inbound resetAllClientTraffics path
and point traffic/onlines/lastOnline fetches at the new
/panel/api/clients/* routes.
- Delete from master: always push the updated inbound to the node
even when the client was already disabled or depleted, so the
node actually loses the user instead of silently keeping it.
- setRemoteTraffic: mirror remote clients into the central tables
only on first discovery of a node inbound. Matched inbounds let
the master own the join table, so a stale snap can no longer
re-create a ClientRecord (and join row) for a client that was
just deleted on the master.
- ClientService.Delete: route through submitTrafficWrite so deletes
serialize with node traffic merges, and switch the final
ClientRecord delete to an explicit Where("id = ?") clause.
- setRemoteTraffic UNIQUE-constraint fix: use clause.OnConflict on
inserts and email-keyed UPDATEs for client_traffics, so mirroring
a snap doesn't trip the unique email index.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): switch client API endpoints from id to email
All client-scoped routes now use the unique email as the path key
(get, update, del, attach, detach, links). Email is the stable,
protocol-independent identifier — UUIDs don't exist for trojan or
shadowsocks, and internal numeric ids leaked panel implementation
detail into the public API.
Removed the redundant /traffic/byId/:id endpoint (covered by
/traffic/:email) and collapsed /links/:id/:email into /links/:email,
which now returns links across every attached inbound for the client.
Frontend selection, bulk delete, and toggle state are now keyed by
email as well, dropping the id→email lookup workaround.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(server): move cached state and helpers into ServerService
ServerController had grown to hold its own status cache, version-list
TTL cache, history-bucket whitelist, and the loop that drove all three
— concerns that belong in the service layer. Pull them out:
- lastStatus + the @2s refresh become ServerService.RefreshStatus and
ServerService.LastStatus; the controller's cron now just orchestrates
the cross-service side effects (xrayMetrics sample, websocket broadcast).
- The 15-minute Xray-versions cache (with stale-on-error fallback) moves
into ServerService.GetXrayVersionsCached, collapsing the controller
handler to a single call.
- The freedom/blackhole outbound-tag walk used by /xraylogs becomes
ServerService.GetDefaultLogOutboundTags.
- The allowed-history-bucket whitelist moves to package-level
service.IsAllowedHistoryBucket, so both NodeController and
ServerController validate against the same list.
Net result: web/controller/server.go drops from 458 to 365 lines and
contains only HTTP wiring + presentation-y side effects.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): emit JSON-text columns as nested objects
Inbound, ClientRecord, and InboundClientIps store settings /
streamSettings / sniffing / reverse / ips as JSON-text in the DB. The
API was passing that text through verbatim, so every consumer had to
JSON.parse a string inside a string. Add MarshalJSON / UnmarshalJSON so
the wire format is a real nested object, while still accepting the
legacy escaped-string shape on write. Frontend dbinbound.js gets a
matching coerceInboundJsonField helper for the same dual-shape read
path, and inbound.js toJson stops emitting empty/placeholder fields
(externalProxy [], sniffing destOverride when disabled, etc.) so the
new normalised JSON stays terse. api-docs and the inbound-clone path
are updated to the new shape. Controller route lists are regrouped so
all GETs sit above POSTs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): include inboundIds and traffic in /clients/list
ClientRecord got its own MarshalJSON in the previous commit, and
ClientWithAttachments embeds it to add inboundIds and traffic. Go
promotes the embedded MarshalJSON to the outer struct, so the encoder
was calling ClientRecord.MarshalJSON for the whole value and silently
dropping the extras. The frontend reads row.inboundIds / row.traffic
from /clients/list, so attached inbounds didn't render and newly added
clients looked like they hadn't saved. Add an explicit MarshalJSON on
ClientWithAttachments that splices the extras in.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): gate IP Log on ipLimitEnable + clean access-log dropdown
Legacy panel hid the IP Log section when access logging was off; the
Vue 3 migration left it gated on isEdit only, so the section showed
even when xray's access log was 'none' and nothing was being recorded.
Restore the ipLimitEnable gate on the edit modal's IP Log form-item.
While here, clean up the Xray Settings access-log dropdown: previously
two 'none' entries appeared (an empty value labelled with t('none') and
the literal 'none' from the options array). Drop the empty option for
access log (the literal 'none' covers it) and relabel the empty option
for error log / mask address to t('empty') so they're distinguishable.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(nodes): route per-client ops through node clients API + orphan sweep
Adds Runtime methods AddClient, UpdateUser, and DeleteUser so master
mutates clients on a node via /panel/api/clients/{add,update,del} rather
than pushing the whole inbound. The previous rt.UpdateInbound path made
the node DelInbound+AddInbound on every single-client change, briefly
cycling every other user on the same inbound.
DelInbound no longer filters by enable=true, so a disabled node inbound
actually gets removed from the node instead of being resurrected by the
next snap.
setRemoteTrafficLocked now sweeps any ClientRecord with zero
ClientInbound rows after SyncInbound rebuilds the attachments, which is
how a node-side delete propagates back to master instead of leaving a
detached ghost. ClientService.Delete tombstones the email first so a
snap arriving mid-delete can't re-create the record.
WebSocket broadcasts an "invalidate(clients)" message on every client
mutation so the Clients page refreshes without manual reload.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(balancers): allow fallback on all strategies + feed burstObservatory from random/roundRobin
Drops the random/roundRobin gate on the Fallback field in
BalancerFormModal so every strategy can pick a fallback outbound.
syncObservatories now feeds burstObservatory from leastLoad +
random + roundRobin balancers (was leastLoad only), matching how
leastPing feeds observatory.
Fix the JsonEditor "Unexpected end of JSON input" that appeared
when switching a balancer between leastPing and another strategy:
the obsView watcher was gated on showObsEditor (a boolean OR of
the two flags) and missed the case where one observatory
swapped for the other in the same tick. Watch the individual
flags instead so obsView flips to the surviving editor and the
getter stops pointing at a deleted key.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): use sortedInbounds for mobile empty-state check
InboundList referenced an undefined visibleInbounds in the mobile
card list's empty-state guard, throwing "Cannot read properties of
undefined (reading 'length')" and breaking the entire mobile render.
* feat(clients): sortable table columns
Adds the same sortState / sortableCol / sortFns pattern InboundList
uses, wrapping filteredClients in sortedClients so sort composes with
the existing search/filter pipeline. Sortable: enable, email,
inboundIds (attachment count), traffic, remaining, expiryTime;
actions and online stay unsorted.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(shadowsocks): generate valid ss2022 keys and per-client method for legacy ciphers
The Add Client flow on shadowsocks inbounds was producing xray configs
that failed to start:
- 2022-blake3-* ciphers need a base64-encoded key of an exact byte
length per cipher. fillProtocolDefaults was assigning a uuid-style
string, which xray rejects as "bad key". Now the password is
generated (or replaced if invalid) via random.Base64Bytes(n) sized
to the chosen cipher.
- Legacy ciphers (aes-256-gcm, chacha20-*, xchacha20-*) require a
per-client method field in multi-user mode; model.Client has no
Method, so settings.clients was stored without one and xray failed
with "unsupported cipher method:". applyShadowsocksClientMethod
now injects the top-level method into each client on add/update,
and healShadowsocksClientMethods backfills it at xray-config-build
time so existing inbounds heal on the next start.
- xray/api.go ssCipherType switch was missing aes-256-gcm, which
fell through to ss2022 path.
- SSMethods dropdown now offers aes-256-gcm.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): preserve ClientRecord on inbound delete + filter Attached inbounds to multi-client protocols
Replace the global orphan sweep in setRemoteTrafficLocked with a
per-inbound diff cleanup: only delete a ClientRecord whose email
disappeared from a snap-tracked inbound (i.e. a node-side delete).
Inbounds that vanished entirely from the snap (e.g. admin deleted
the inbound on master) aren't iterated, so a client whose last
attachment came from that inbound is now left alone instead of
being deleted alongside the inbound.
ClientFormModal and ClientBulkAddModal now filter the Attached
inbounds dropdown to protocols that actually support multiple
clients: shadowsocks, vless, vmess, trojan, hysteria, hysteria2,
and portfallback (which routes through VLESS settings).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): make empty-state text readable on dark/ultra themes
The "No clients yet" empty state had a hardcoded black color
(rgba(0,0,0,0.45)) that vanished against the dark backgrounds.
Drop the inline color, let it inherit from the AntD theme, and
fade with opacity like the mobile card empty state already does.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): client-first tgbot add flow, tgId field, lightweight inbound options
- tgbot: drop legacy per-protocol Add Client UI in favour of a client-first
multi-inbound flow. New BuildClientDraftMessage / getInboundsAttachPicker
let an admin pick one or more inbounds and submit a single client; per-
protocol secrets are now generated server-side via fillProtocolDefaults.
Drops awaiting_id/awaiting_password_tr/awaiting_password_sh state cases
and add_client_ch_default_id/pass_tr/pass_sh/flow callbacks. Adds a
setTGUser button + awaiting_tg_id state so the bot can set Client.TgID
during Add.
- clients UI: add Telegram user ID input to ClientFormModal (0 = none).
Hide IP Limit field entirely when ipLimitEnable is off — disabled fields
still take layout space, this collapses Auth(Hysteria) to full width.
- inbounds API: new GET /panel/api/inbounds/options that returns just
{id, remark, protocol, port, tlsFlowCapable}. Used by the clients page
pickers so the dropdown payload stays small on panels with thousands of
clients (drops settings JSON, clientStats, streamSettings). Server-side
TlsFlowCapable mirrors Inbound.canEnableTlsFlow so the modal no longer
needs to parse streamSettings client-side.
- clientInfoMsg now shows attached inbound remarks, and getInboundUsages
reports the attached client count per inbound.
- api-docs: document the new /options endpoint and add tgId / flow to the
clients add/update bodies.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): keep Node column visible for node-attached inbounds
The Node column was bound to hasActiveNode, so disabling every node hid
the column even when inbounds were still attached to those nodes — the
admin lost the visual cue that those inbounds belonged to a node and
would come back when it was re-enabled. Combine hasActiveNode with a
new hasNodeAttachedInbound check (any dbInbound with nodeId != null) so
the column survives node-disable.
* fix(api-docs): accept functional-component icons in EndpointSection
AntD-Vue icons (SafetyCertificateOutlined, etc.) are functional
components, so the icon prop's type: Object validator was rejecting
them with a "Expected Object, got Function" warning at runtime.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* test: cover crypto, random, netsafe, sub helpers, xray equals, websocket hub, node service
Adds ~110 unit tests across previously untested packages. Focus on
pure-logic and concurrency surfaces where regressions would silently
affect users:
- util/crypto, util/random: password hashing round-trip, ss2022 key
generation, alphabet/length invariants.
- util/netsafe: IsBlockedIP edge cases, NormalizeHost validation,
SSRF guard with AllowPrivate context bypass.
- util/common, util/json_util: traffic formatter, Combine nil-skip,
RawMessage empty-as-null and copy-on-unmarshal.
- sub: splitLinkLines, searchKey/searchHost, kcp share fields,
finalmask normalization, buildVmessLink round-trip.
- xray: Config.Equals and InboundConfig.Equals field-by-field,
getRequiredUserString/getOptionalUserString type checks.
- web/websocket: hub registration, throttling, slow-client eviction,
nil-receiver safety, concurrent register/unregister.
- web/service: NodeService.normalize validation, normalizeBasePath,
HeartbeatPatch.ToUI mapping.
- web/job: atomicBool concurrent set/takeAndReset semantics.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n(clients): replace English fallbacks with proper translation keys
Pulls every hard-coded English label/title in the Clients page and its
four modals through the i18n layer so localized panels stop leaking
English. New keys live under pages.clients (auth, hysteriaAuth, uuid,
flow, flowNone, reverseTag, reverseTagPlaceholder, telegramId,
telegramIdPlaceholder, created, updated, ipLimit) plus refresh at the
root and toasts.bulkDeletedMixed / bulkCreatedMixed for partial-failure
toasts. Also switches the add-client modal's primary button from "Add"
to "Create" for consistency with other create flows.
The bulk-add Random/Random+Prefix/... email-method options stay
hard-coded by request - they're identifier-shaped strings.
* i18n: backfill 99 missing keys across all 12 non-English locales
Brings every translation file up to parity with en-US.json so the
Clients page, the fallback-children inbound section, the new refresh
verb, the Nodes panel-version label and a handful of older holes stop
falling through to the English fallback. New strings span:
- pages.clients.* (labels, confirmations, toasts, emailMethods)
- pages.inbounds.portFallback.* (Reality fallback inbound section)
- pages.nodes.panelVersion, menu.clients, refresh
Technical identifiers (Auth, UUID, Flow, Reverse tag) are intentionally
left untranslated since they correspond to xray-core field names.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n: drop stale pages.client block duplicated in every non-English locale
Every non-English locale carried a pages.client (singular) section with
30 entries that duplicated pages.clients (plural). The plural namespace
is what the Vue code actually consumes; the singular one was dead
weight from an older rename that never got cleaned up in the
non-English files. Removing it brings every locale to exactly 984
keys, matching en-US.json.
* chore: apply modernize analyzer fixes across codebase
Mechanical replacements suggested by golang.org/x/tools/.../modernize:
strings.Cut/CutPrefix/SplitSeq, slices.Contains, maps.Copy, min(),
range-over-int, new(expr), strings.Builder for hot += loops,
reflect.TypeFor[T](), sync.WaitGroup.Go(), drop legacy //+build lines.
* feat(database): add PostgreSQL as an optional backend alongside SQLite
Lets operators with large client counts or multi-node setups pick PostgreSQL
at install time without breaking the existing SQLite default. Backend is
selected at runtime via XUI_DB_TYPE/XUI_DB_DSN, a small dialect layer keeps
the five JSON_EXTRACT/JSON_EACH queries portable, and a new `x-ui migrate-db`
subcommand copies SQLite data into PostgreSQL in FK-aware order.
* fix(inbounds): gate node selector to multi-node-capable protocols
Hide the Deploy-To selector and clear nodeId when switching to a
protocol that can't run on a remote node. Also:
- subs: return 404 (not 400) when subId matches no inbounds, so VPN
clients distinguish "deleted/unknown" from a server error
- hysteria link gen: use the inbound's resolved address so node-managed
inbounds advertise the node host instead of the central panel
- shadowsocks: default network to 'tcp' (udp was causing issues for some
clients on first-create)
- vite dev proxy: rewrite migrated-route bypass against the live base
path instead of a hardcoded single-segment regex
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): bulk add/delete correctness + perf, working pagination, delayed-start in form
Bulk add/delete were serial on the frontend (one toast per call, N round-trips)
and the backend race exposed by parallelizing them lost client attachments and
hit UNIQUE constraint failed on client_inbounds. The single add/edit modal also
had no Start-After-First-Use option, and the table never showed the delayed
duration.
Backend (web/service/client.go):
- Per-inbound mutex on Add/Update/Del InboundClient so concurrent writers on
the same inbound don't lose the read-modify-write of settings JSON.
- SyncInbound skips create+join when the email is tombstoned so a concurrent
maintenance pass (adjustTraffics, autoRenewClients, markClientsDisabledIn-
Settings) that did a stale RMW can't resurrect a just-deleted client with a
fresh id.
- compactOrphans sweeps settings.clients entries whose ClientRecord no longer
exists, applied in Add/DelInboundClient + DelInboundClientByEmail so each
user-initiated mutation self-heals the inbound's settings.
- DelInboundClient uses Pluck instead of First for the stats lookup so a
missing row doesn't abort the delete with a noisy ErrRecordNotFound log.
Frontend:
- HttpUtil.{get,post} accept a silent option that suppresses the auto-toast.
- ClientBulkAddModal fires creates in parallel + silent + one summary toast.
- useClients.removeMany runs deletes in parallel + silent and refreshes once;
ClientsPage bulk delete uses it and shows one aggregate toast.
- useClients.applyInvalidate debounces 200 ms so the burst of N WebSocket
invalidate events from the backend collapses into a single refresh.
- ClientsPage pagination is reactive (paginationState ref + tablePagination
computed); onTableChange persists page-size and page changes.
- ClientFormModal gains a Start-After-First-Use switch + Duration days input
alongside the existing Expiry Date picker; on edit-mode open a negative
expiryTime is decoded back to delayed mode + days; on submit the payload
sends -86400000 * days or the absolute timestamp.
- ClientsPage table shows the delayed-start duration (blue tag Nd, tooltip
Start After First Use: Nd) instead of infinity.
- Telegram ID field in the form is hidden when /panel/setting/defaultSettings
reports tgBotEnable=false; Comment then fills the row.
- Form row 3 collapses UUID (span 12) + Total GB (span 8) + Limit IP (span 4)
when ipLimitEnable is on, else UUID + Total GB at 12/12.
- useInbounds.rollupClients counts only clients with a matching clientStats
row, so orphans in settings.clients no longer inflate the inbound's count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(windows): clean shutdown, working panel restart, harden kernel32 load
Three Windows-specific issues addressed:
1. Orphaned xray-windows-amd64 after VS Code debugger stop. Delve's
"Stop" sends TerminateProcess to the Go binary, which is uncatchable
— our signal handlers never run, so xrayService.StopXray() is skipped
and xray is left dangling. Spawn xray as a child of a Job Object with
JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE so the OS kills xray when our
handle to the job is closed (which happens even on TerminateProcess).
Also trap os.Interrupt in main so Ctrl+C in the terminal runs the
graceful path.
2. /panel/setting/restartPanel logged "failed to send SIGHUP signal: not
supported by windows" because Windows can't deliver arbitrary signals.
Add a restart hook in web/global; main registers it to push SIGHUP
into its own signal channel, and RestartPanel calls the hook before
falling back to the (Unix-only) signal path. Same restart-loop code
runs in both cases.
3. util/sys/sys_windows.go now uses windows.NewLazySystemDLL so the
kernel32.dll resolve is pinned to %SystemRoot%\System32 (prevents
DLL hijacking by a planted DLL next to the binary). Local filetime
type replaced with windows.Filetime, and the unreliable
syscall.GetLastError() fallback replaced with a type assertion on the
errno captured at call time.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(sys): correct CPU/connection accounting on linux + darwin
util/sys/sys_linux.go:
- GetTCPCount/GetUDPCount were counting the column header row in
/proc/net/{tcp,udp}[6] as a connection, inflating the reported total
by 1 per non-empty file (so the panel status line always showed 2
more connections than actually existed). Replace getLinesNum +
safeGetLinesNum with a single bufio.Scanner-based countConnections
that skips the header.
- CPUPercentRaw now opens HostProc("stat") instead of a hardcoded
/proc/stat so HOST_PROC overrides apply, matching the connection
counters in the same file.
- Simplify CPU field unpacking: pad nums to 8 once instead of guarding
every assignment with a len check.
util/sys/sys_darwin.go:
- Fix swapped idle/intr indices on kern.cp_time. BSD CPUSTATES order
is user, nice, sys, intr, idle (CP_INTR=3, CP_IDLE=4) — gopsutil's
cpu_darwin_nocgo.go reads the same layout. The previous code used
out[3] as idle and out[4] as intr, so busy = total - dIdle was
actually subtracting interrupt time, making the panel report CPU
usage close to 100% on macOS regardless of actual load.
- Collapse the per-field delta math into a single loop.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(xray): rotate crash reports into log folder, prevent overwrites
writeCrashReport had two flaws: it wrote to the bin folder (alongside the
xray binary) which conflates artifacts, and the second-precision timestamp
meant a tight restart-loop crash burst overwrote prior reports. Write to
the log folder with nanosecond precision and keep the last 10 reports.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* revert(inbounds): drop unreleased portfallback protocol
The Port-with-Fallback inbound (commit 62fd9f9d) was confusing as a
standalone protocol — fallbacks belong on a regular VLESS/Trojan TCP-TLS
inbound, the way Xray models them natively. Rip out the entire feature
cleanly (no migration needed since it was never released): protocol
constant, fallback children DB table, FallbackService, 2 API endpoints,
all UI rows, related translations and api-docs. A native fallback flow
attached to VLESS/Trojan TCP-TLS/Reality will land in a follow-up commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): native fallbacks on VLESS/Trojan TCP-TLS, with working child links
A VLESS or Trojan inbound on TCP with TLS or Reality can now act as a
fallback master: pick existing inbounds as children and the panel auto-
fills the SNI / ALPN / path / xver routing fields from each child's
transport, auto-builds settings.fallbacks at config-gen time, and
rewrites the child's client-share link so it advertises the master's
reachable endpoint and TLS state instead of the child's loopback listen.
Layout matches the Xray All-in-One Nginx example: master at :443 with
clients + TLS, each child on 127.0.0.1 with its own transport+clients.
Order matters (Xray walks fallbacks top-to-bottom) — reorder via the
per-row up/down arrows. Path / SNI / ALPN are exposed under a per-row
Edit toggle for the rare cases where the auto-derivation needs
overriding; otherwise just pick a child and you're done.
Backend: new InboundFallback table + FallbackService (GetByMaster /
SetByMaster / GetParentForChild / BuildFallbacksJSON); two routes
(GET / POST /panel/api/inbounds/:id/fallbacks); xray.GetXrayConfig
injects settings.fallbacks for any VLESS/Trojan TCP-TLS/Reality
inbound; GetInbounds annotates each child with FallbackParent so the
frontend can rewrite links without an extra round-trip.
Link projection covers every emission path — clients-page QR/links,
per-inbound Get URL, raw subscription, sub-JSON, sub-Clash, and the
inbounds-page link/info/QR — via a shared projectThroughFallbackMaster
on the backend and a shared projectChildThroughMaster on the frontend
that both handle the panel-tracked relationship and the legacy
unix-socket (@vless-ws) convention.
Strings translated into all 12 non-English locales.
* docs: rewrite CONTRIBUTING with full local-dev setup
The prior three-line CONTRIBUTING left newcomers guessing at every
non-trivial step: which Go / Node versions, where xray comes from, why
the panel goes blank when XUI_DEBUG=true is flipped on, how the Vue
multi-page setup is wired, what to do on Windows when go build trips
on the CGo SQLite driver.
Now covers prerequisites, MinGW-w64 install on Windows (niXman builds
or MSYS2), one-shot first-time setup, two frontend dev workflows with
the XUI_DEBUG asset-cache gotcha called out, the architecture and
conventions of the Vue side, a project-layout map, useful env vars,
and the PR checklist.
---------
2026-05-19 03:16:42 -07:00
fi
2026-05-28 08:20:16 -07:00
2026-05-31 08:53:34 -07:00
if [ [ " $db_choice " = = "2" ] ] ; then
echo ""
echo -e " ${ green } PostgreSQL backup & restore is built into the panel: ${ plain } "
echo -e " ${ blue } ${ SSL_SCHEME } :// ${ SSL_HOST } : ${ config_port } / ${ config_webBasePath } ${ plain } → Backup & Restore "
echo -e " ${ yellow } Back Up downloads a pg_dump .dump file; Restore reloads it via pg_restore. ${ plain } "
fi
2026-05-28 08:20:16 -07:00
if [ [ " $db_choice " = = "2" && " $pg_local_installed " = = "1" ] ] ; then
echo ""
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } PostgreSQL Credentials ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } DB Name: ${ PG_DB } ${ plain } "
echo -e " ${ green } Username: ${ PG_USER } ${ plain } "
echo -e " ${ green } Password: ${ PG_PASS } ${ plain } "
echo -e " ${ green } Host: ${ PG_HOST } ${ plain } "
echo -e " ${ green } Port: ${ PG_PORT } ${ plain } "
echo -e " ${ green } DSN: ${ xui_dsn } ${ plain } "
echo -e " ${ green } Env file: ${ xui_env_file } ${ plain } "
echo -e " ${ green } ------------------------------------------- ${ plain } "
echo -e " ${ green } Connect from this server: ${ plain } "
echo -e " ${ blue } sudo -u postgres psql -d ${ PG_DB } ${ plain } (as the postgres superuser) "
echo -e " ${ blue } PGPASSWORD=' ${ PG_PASS } ' psql -h ${ PG_HOST } -p ${ PG_PORT } -U ${ PG_USER } -d ${ PG_DB } ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ yellow } ⚠ The panel reads these credentials from ${ xui_env_file } . ${ plain } "
echo -e " ${ yellow } ⚠ Save the password — it is not stored anywhere else in plain text. ${ plain } "
unset PG_USER PG_PASS PG_HOST PG_PORT PG_DB
fi
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
# Persist a machine-parseable credentials file for cloud-init / MOTD.
: " ${ SSL_SCHEME : =https } "
: " ${ SSL_HOST : = ${ server_ip } } "
local db_type_out = "sqlite"
[ [ " $db_choice " = = "2" ] ] && db_type_out = "postgres"
write_install_result " ${ config_username } " " ${ config_password } " " ${ config_port } " \
" ${ config_webBasePath } " " ${ SSL_SCHEME } " " ${ SSL_HOST } " " ${ config_apiToken } " " ${ db_type_out } "
Feat/multi inbound clients (#4469)
* feat(clients): add shadow tables for first-class client promotion
Introduces three new GORM-backed tables (clients, client_inbounds,
inbound_fallback_children) and a populate-only seeder that backfills
them from each inbound's existing settings.clients JSON. Duplicate
emails across inbounds auto-merge under one client row, with each
field conflict logged. Existing services are unchanged and continue
reading from settings.clients — this commit is groundwork only.
* feat(clients): make clients+client_inbounds the runtime source of truth
Adds ClientService.SyncInbound that reconciles the new tables from
each inbound's clients list whenever existing service paths mutate
settings.clients. Wires it into AddInbound, UpdateInbound,
AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, DelDepletedClients, autoRenewClients, and
the timestamp-backfill path in adjustTraffics, plus DetachInbound
on DelInbound.
GetXrayConfig now builds settings.clients from the new tables before
writing config.json, and getInboundsBySubId joins through them
instead of JSON_EACH on settings JSON. Live Xray config and
subscription endpoints are now driven by the relational view;
settings.clients JSON stays in step as a side effect of every write.
* feat(clients): add top-level Clients tab and CRUD API
Adds /panel/api/clients endpoints (list, get, add, update, del,
attach, detach) backed by ClientService methods that orchestrate
the per-inbound Add/Update/Del flows so a single client row is
created once and attached to many inbounds in one operation.
The frontend gains a dedicated Clients page (frontend/clients.html
+ src/pages/clients/) with an AntD table, multi-inbound attach
modal, and full CRUD. Axios interceptor learns to honour
Content-Type: application/json so the JSON endpoints work
alongside the legacy form-encoded ones.
The legacy per-inbound client modal stays untouched in this PR —
both flows now write to the same source of truth.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): add Port-with-Fallback inbound type
Adds a new "portfallback" protocol that emits as a VLESS-TLS inbound
under the hood but is paired with a sidecar table of child inbounds.
Panel auto-builds settings.fallbacks at Xray-config-gen time from the
sidecar — each child's listen+port becomes the fallback dest, with
SNI/ALPN/path/xver match criteria pulled from the row. No more typing
loopback ports by hand or keeping settings.fallbacks in sync.
Backend: new FallbackService (Get/SetChildren, BuildFallbacksJSON);
two new routes (GET/POST /panel/api/inbounds/:id/fallbackChildren);
xray.GetXrayConfig injects fallbacks for PortFallback inbounds; the
inbound model emits protocol="vless" so Xray accepts the config.
Frontend: PORTFALLBACK joins the protocol dropdown; selecting it
shows the standard VLESS controls plus a Fallback Children table
(inbound picker + per-row SNI/ALPN/path/xver). Children are loaded
on edit and replaced atomically on save.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reset Traffic, QR Code, Info actions + Online/Remaining columns
The Clients page table gains:
- Online column — green/grey tag driven by /panel/api/inbounds/onlines,
polled every 10s.
- Remaining column — bytes-remaining tag, coloured green/orange/red
against quota, purple infinity when unlimited.
- Action icons per row: QR, Info, Reset traffic, Edit, Delete.
ClientInfoModal shows the full client detail (uuid/password/auth,
traffic ↑/↓ + remaining + all-time, expiry absolute + relative,
attached inbounds chip list, online + last-online).
ClientQrModal fetches links for the client's subId via
/panel/api/inbounds/getSubLinks/:subId and renders each one through
the existing QrPanel component.
Reset Traffic confirms then calls the existing per-inbound endpoint
on the client's first attached inbound (the traffic row is keyed on
email globally, so any attached inbound resets the shared counter).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): expose Attached inbounds in edit mode
The multi-select was gated on add-only, so editing a client had no way
to change which inbounds it belonged to. The picker now shows in both
modes, and on submit the modal diffs the picked set against the
original attachedIds — additions go through the /attach endpoint,
removals through /detach, both after the field update lands so the
new attachments get the latest values.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): unbreak template parsing + stale i18n keys
- InboundFormModal: split the multi-line help string in the
PortFallback section onto one line — Vue's template parser was
bailing on Unterminated string constant because a single-quoted
literal spanned two lines inside a {{ }} interpolation.
- ClientInfoModal: t('disable') was missing at the root level, so
vue-i18n returned the key path literally. Use t('disabled') which
exists.
- Linter cleanup elsewhere: pages.client.* references renamed to
pages.clients.* to match the merged i18n block; whitespace
normalisation in a few unrelated Vue templates.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* 1
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(traffic): drop all-time traffic tracking
Removes the AllTime field from Inbound and ClientTraffic and migrates
existing DBs by dropping the all_time columns on startup. The counter
duplicated up+down without adding signal, and the per-event accumulator
ran on every traffic write.
Frontend: drop the All-time column from the inbound list and the
client-row table, the All-time row from the client info modal, and the
All-Time Total Usage tile from the inbounds summary card. The
allTimeTraffic/allTimeTrafficUsage i18n keys are removed across every
locale.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): mobile cards, multi-select, bulk add
Adds the same row-card layout the inbounds page uses on mobile: the
table is suppressed under the mobile breakpoint and each client renders
as a compact card with a status dot, email, Info button, Enable switch,
and overflow menu. All the per-client detail (traffic, remaining,
expiry, attached inbounds, flow, created/updated, URL, subscription)
opens through the existing info modal.
Multi-select with bulk delete wires AntD row-selection on desktop and
a per-card checkbox on mobile; a Delete (N) button appears in the
toolbar when anything is selected.
Bulk add reuses the five email-generation modes from the inbound bulk
modal but takes a multi-inbound picker so one bulk run can attach to
several inbounds at once. Submits client-by-client through the
existing /panel/api/clients/add endpoint.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): remove legacy per-inbound client UI
Now that clients live as first-class rows attached to one or many
inbounds, the per-inbound client UI on the inbounds page is dead
weight — every client action either has a global equivalent on the
Clients page or makes no sense in a many-to-many world.
Deletes ClientFormModal, ClientBulkModal, CopyClientsModal, and
ClientRowTable from inbounds/. Strips the matching emits, refs,
handlers, and dropdown menu items from InboundList and InboundsPage,
and removes the dead mobile expand-chevron state and the desktop
expanded-row plumbing that drove the inline client table.
The InboundFormModal Clients tab still works in add-mode (one inline
client at inbound creation) — that flow goes through ClientService.
SyncInbound on save and remains useful.
Fixes a stray "</a-dropdown>" left over by an earlier toolbar edit
in ClientsPage that broke the template parser.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Delete depleted action
Mirrors the legacy delDepletedClients action that lived under the
inbounds page, but as a first-class /panel/api/clients/delDepleted
endpoint backed by ClientService. The new path goes through
ClientService.Delete for each depleted email, so the new clients +
client_inbounds + xray_client_traffic tables stay consistent.
Adds a danger-styled toolbar button on the Clients page (next to
Reset all client traffic) with a confirm dialog and a toast
reporting the deleted count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): move every client-shaped endpoint off /inbounds onto /clients
After the multi-inbound client migration, client state belongs to the
client API surface, not the inbound one. Twelve routes that were
crammed under /panel/api/inbounds/* now live where they belong, under
/panel/api/clients/*.
Moved (route, handler, doc):
POST /clientIps/:email
POST /clearClientIps/:email
POST /onlines
POST /lastOnline
POST /updateClientTraffic/:email
POST /resetAllClientTraffics/:id
POST /delDepletedClients/:id
POST /:id/resetClientTraffic/:email
GET /getClientTraffics/:email
GET /getClientTrafficsById/:id
GET /getSubLinks/:subId
GET /getClientLinks/:id/:email
Their /clients/* counterparts are:
POST /clients/clientIps/:email
POST /clients/clearClientIps/:email
POST /clients/onlines
POST /clients/lastOnline
POST /clients/updateTraffic/:email
POST /clients/resetTraffic/:email (email-only, fans out)
GET /clients/traffic/:email
GET /clients/traffic/byId/:id
GET /clients/subLinks/:subId
GET /clients/links/:id/:email
per-inbound resetAllClientTraffics and delDepletedClients are dropped
entirely — the Clients page already exposes global Reset All Traffic
and Delete depleted actions, and per-inbound resets are meaningless
once a client can be attached to many inbounds.
ClientService.ResetTrafficByEmail is the new email-only reset path:
it looks up every inbound the client is attached to and pushes the
counter reset + Xray re-add through inboundService.ResetClientTraffic
for each one, so depleted users come back online instantly.
Frontend callers (ClientsPage, useClients, ClientQrModal,
ClientInfoModal, InboundInfoModal, InboundsPage, useInbounds) all
switched to the new paths. The Inbounds page drops its per-inbound
"Reset client traffic" and "Delete depleted clients" dropdown items —
users do those at the client level now. api-docs is rebuilt to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): switch tgbot + ldap callers to ClientService
Adds two thin helpers to ClientService (CreateOne, DetachByEmail) and
rewrites tgbot.SubmitAddClient and ldap_sync_job to call ClientService
directly. Removes the JSON-blob payloads (BuildJSONForProtocol output for
add, clientsToJSON/clientToJSON helpers) that callers previously fed to
InboundService.AddInboundClient/DelInboundClient.
ldap_sync_job.batchSetEnable now loops InboundService.SetClientEnableByEmail
per email instead of trying to coerce AddInboundClient into doing the
update — the old path would have failed duplicate-email validation for
existing clients anyway.
The legacy InboundService.AddInboundClient/UpdateInboundClient/
DelInboundClient methods stay in place; they are now only used internally
by ClientService Create/Update/Delete/Attach. Inlining + deleting them
follows in a separate commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): move all client mutation methods to ClientService
Moves the client mutation surface out of InboundService and into
ClientService. These methods all operate on a single client (identity
fields, traffic limits, expiry, ip limit, enable state, telegram tg id)
and didn't belong on the inbound aggregate.
Moved (12 methods): AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, checkEmailsExistForClients, SetClientTelegramUserID,
checkIsEnabledByEmail, ToggleClientEnableByEmail, SetClientEnableByEmail,
ResetClientIpLimitByEmail, ResetClientExpiryTimeByEmail,
ResetClientTrafficLimitByEmail.
Each method now takes an explicit *InboundService for the helpers that
legitimately stay on InboundService (GetInbound, GetClients, runtimeFor,
AddClientStat / UpdateClientStat / DelClientStat, DelClientIPs /
UpdateClientIPs, emailUsedByOtherInbounds, getAllEmailSubIDs,
GetClientInboundByEmail / GetClientInboundByTrafficID,
GetClientTrafficByEmail).
Stays on InboundService: ResetClientTrafficByEmail and
ResetClientTraffic(id, email) — these mutate xray_client_traffic rows,
not client identity, so they're inbound-side bookkeeping.
Callers updated: tgbot (6 calls), ldap_sync_job (1 call),
InboundService internal (writeBackClientSubID, CopyInboundClients,
AddInbound's email-uniqueness check), ClientService Create/Update/
Delete/Attach/Detach.
Also removes a dead resetAllClientTraffics controller handler whose
route was already gone after the previous /clients API migration.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): finish migrating to ClientService + tidy IP routes
Two related cleanups in the new /clients surface:
1. Move ResetAllClientTraffics (bulk-reset of xray_client_traffic +
last_traffic_reset_time, with node-runtime propagation) from
InboundService to ClientService. PeriodicTrafficResetJob now holds
a clientService and calls
j.clientService.ResetAllClientTraffics(&j.inboundService, id).
The last client-mutation method on InboundService is gone.
2. Shorten redundantly-named routes/handlers under /panel/api/clients:
- /clientIps/:email -> /ips/:email (handler getIps)
- /clearClientIps/:email -> /clearIps/:email (handler clearIps)
The "client" prefix was redundant inside the clients namespace.
Frontend (InboundInfoModal) and api-docs updated to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds,clients): clean up inbound modal + enrich client modal
Inbound modal rework (InboundFormModal.vue + inbound.js):
- Drop the embedded Client subform in the Protocol tab. Multi-inbound
clients are managed exclusively from the Clients page now; a fresh
inbound is created with zero clients (settings constructors default
to []) and the user attaches clients afterwards.
- Hide the Protocol tab entirely when it has nothing to render
(VMESS, Trojan without fallbacks, Hysteria). Auto-switches active
tab to Basic when the tab disappears while focused.
- Move the Security section (Security selector + TLS block with certs
and ECH + Reality block) out of the Stream tab into its own
Security tab, sharing the canEnableStream gate.
Client modal additions (ClientFormModal.vue + ClientBulkAddModal.vue):
- Flow select (xtls-rprx-vision / -udp443) appears only when the
panel actually has a Vision-capable inbound (VLESS or PortFallback
on TCP with TLS or Reality). Hidden otherwise, and cleared when
it disappears.
- IP Limit input is disabled when the panel-level ipLimitEnable
setting is off, fetched into useClients alongside subSettings and
threaded through ClientsPage to both modals.
- Edit modal now shows an "IP Log" section listing IPs that have
connected with the client's credentials, with refresh and clear
buttons (calls the renamed /panel/api/clients/ips and /clearIps
endpoints).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): drop manual Fallbacks UI from inbound modal
The PortFallback protocol type now covers the common
VLESS-master-plus-children case with auto-wired dests, so the manual
Fallbacks editor (showFallbacks block in the Protocol tab) is mostly
redundant. Removed:
- the v-if="showFallbacks" template block (SNI/ALPN/Path/dest/PROXY rows)
- the showFallbacks computed
- the addFallback / delFallback helpers
- the .fallbacks-header / .fallbacks-title styles
- the showFallbacks gate from hasProtocolTabContent (so Trojan-over-TCP
no longer shows an empty Protocol tab)
Power users who need a non-inbound fallback dest (nginx, static site)
can still author settings.fallbacks via the Advanced JSON tab.
* feat(clients,inbounds): move search/filter to Clients page + small fixes
Search/filter relocation:
- Remove the search/filter toolbar (search switch + filter radio +
protocol/node selects + the visibleInbounds projection +
inboundsFilterState localStorage + filter CSS + the SearchOutlined/
FilterOutlined/ObjectUtil/Inbound imports it required) from
InboundList. The filters were all client-oriented buckets bolted
onto the inbound row.
- Add a search/filter toolbar to ClientsPage with the same shape:
switch between deep-text search and bucket filter (active /
deactive / depleted / expiring / online) + protocol filter that
matches clients attached to at least one inbound with the chosen
protocol. State persists in clientsFilterState localStorage.
filteredClients drives both the desktop table and the mobile card
list, and select-all / allSelected / someSelected only span the
visible subset.
- useClients now also fetches expireDiff and trafficDiff from
/panel/setting/defaultSettings (used to detect the expiring
bucket); ClientsPage threads them into the client-bucket helper.
Loose fixes folded in:
- Add Client: email field is auto-filled with a random handle on
open, matching uuid/subId/password/auth.
- Inbound clone: parse and reuse the source settings JSON (with
clients reset to []) instead of building a fresh defaulted
Settings, so VLESS Encryption/Decryption and other non-client
fields survive the clone.
- en-US.json: add the ipLog string used by the edit-client modal.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reverse tag field for VLESS-attached clients
Mirrors the Flow field's pattern: a Reverse tag input appears in the
Add/Edit Client modal whenever at least one selected inbound is VLESS
or PortFallback. The value rides over the wire as
client.reverse = { tag: '...' } so it lands directly in model.Client's
*ClientReverse field; an empty value omits the reverse key entirely.
On edit the field is hydrated from props.client.reverse?.tag, and the
showReverseTag watcher clears the field if the user drops the last
VLESS-like inbound from the selection.
* fix(xray): emit only protocol-relevant fields per client entry
The Xray config synthesizer was writing every identifier field (id,
password, flow, auth, security/method, reverse) on every client entry
regardless of the inbound's protocol. Xray ignores unknown fields, so
the config worked, but it diverged from the spec and leaked secrets
across protocols when one client was attached to multiple inbounds —
a VLESS inbound's generated config carried the same client's Trojan
password and Hysteria auth alongside its uuid.
Switch on inbound.Protocol when building each entry:
- VLESS / PortFallback: id, flow, reverse
- VMess: id, security
- Trojan: password, flow
- Shadowsocks: password, method
- Hysteria / Hysteria2: auth
email is emitted for every protocol.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): restore auto-disable kick under new schema
disableInvalidClients still resolved (inbound_tag, email) pairs via
JSON_EACH(inbounds.settings.clients), which is empty after migrating
to the clients + client_inbounds tables. Result: xrayApi.RemoveUser
never ran for depleted clients, clients.enable stayed true so the UI
showed them as active, and only xray_client_traffic.enable got flipped
- making "Restart Xray After Auto Disable" only half-work.
Resolve the targets via a JOIN through the new schema, flip clients.enable
so the Clients page reflects the state, and drop the legacy JSON
write-back plus the subId cascade workaround (email is unique now).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): live WebSocket updates + Ended status surfacing
ClientsPage now subscribes to traffic / client_stats / invalidate
WebSocket events instead of polling /onlines every 10s. Per-row
traffic counters refresh in place, online state stays current, and
list-level mutations elsewhere trigger a refresh.
The client roll-up summary moves from InboundsPage to ClientsPage
where it belongs, restructured into six labeled stat tiles
(Total / Online / Ended / Expiring / Disabled / Active) with email
popovers on the ones with issues.
Auto-disabled clients (traffic exhausted or expiry passed) now
classify as 'depleted' even though clients.enable=false, so they
show up under the Ended filter and render a red Ended tag instead
of looking indistinguishable from an operator-disabled row.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(nodes): per-node client roll-up and panel version
Added transient inboundCount / clientCount / onlineCount /
depletedCount fields to model.Node, populated by NodeService.GetAll
via aggregated queries (one join across inbounds + client_inbounds,
one over client_traffics intersected with the in-memory online
emails). The Nodes list renders these as colored chips on a new
"Clients" column so an operator can see at a glance how many users
each node carries and how many are currently online or depleted.
Also exposes the remote panel's version. The central panel adds
panelVersion to its /api/server/status payload (sourced from
config.GetVersion). Probe reads that field and persists it on the
node row, mirroring how xrayVersion already flows. NodesPage gets
a new column next to Xray Version, in both desktop and mobile
views, with English and Persian strings.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): stop node sync from resurrecting deleted clients
Several related issues around node-managed clients:
- Remote runtime: drop the per-inbound resetAllClientTraffics path
and point traffic/onlines/lastOnline fetches at the new
/panel/api/clients/* routes.
- Delete from master: always push the updated inbound to the node
even when the client was already disabled or depleted, so the
node actually loses the user instead of silently keeping it.
- setRemoteTraffic: mirror remote clients into the central tables
only on first discovery of a node inbound. Matched inbounds let
the master own the join table, so a stale snap can no longer
re-create a ClientRecord (and join row) for a client that was
just deleted on the master.
- ClientService.Delete: route through submitTrafficWrite so deletes
serialize with node traffic merges, and switch the final
ClientRecord delete to an explicit Where("id = ?") clause.
- setRemoteTraffic UNIQUE-constraint fix: use clause.OnConflict on
inserts and email-keyed UPDATEs for client_traffics, so mirroring
a snap doesn't trip the unique email index.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): switch client API endpoints from id to email
All client-scoped routes now use the unique email as the path key
(get, update, del, attach, detach, links). Email is the stable,
protocol-independent identifier — UUIDs don't exist for trojan or
shadowsocks, and internal numeric ids leaked panel implementation
detail into the public API.
Removed the redundant /traffic/byId/:id endpoint (covered by
/traffic/:email) and collapsed /links/:id/:email into /links/:email,
which now returns links across every attached inbound for the client.
Frontend selection, bulk delete, and toggle state are now keyed by
email as well, dropping the id→email lookup workaround.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(server): move cached state and helpers into ServerService
ServerController had grown to hold its own status cache, version-list
TTL cache, history-bucket whitelist, and the loop that drove all three
— concerns that belong in the service layer. Pull them out:
- lastStatus + the @2s refresh become ServerService.RefreshStatus and
ServerService.LastStatus; the controller's cron now just orchestrates
the cross-service side effects (xrayMetrics sample, websocket broadcast).
- The 15-minute Xray-versions cache (with stale-on-error fallback) moves
into ServerService.GetXrayVersionsCached, collapsing the controller
handler to a single call.
- The freedom/blackhole outbound-tag walk used by /xraylogs becomes
ServerService.GetDefaultLogOutboundTags.
- The allowed-history-bucket whitelist moves to package-level
service.IsAllowedHistoryBucket, so both NodeController and
ServerController validate against the same list.
Net result: web/controller/server.go drops from 458 to 365 lines and
contains only HTTP wiring + presentation-y side effects.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): emit JSON-text columns as nested objects
Inbound, ClientRecord, and InboundClientIps store settings /
streamSettings / sniffing / reverse / ips as JSON-text in the DB. The
API was passing that text through verbatim, so every consumer had to
JSON.parse a string inside a string. Add MarshalJSON / UnmarshalJSON so
the wire format is a real nested object, while still accepting the
legacy escaped-string shape on write. Frontend dbinbound.js gets a
matching coerceInboundJsonField helper for the same dual-shape read
path, and inbound.js toJson stops emitting empty/placeholder fields
(externalProxy [], sniffing destOverride when disabled, etc.) so the
new normalised JSON stays terse. api-docs and the inbound-clone path
are updated to the new shape. Controller route lists are regrouped so
all GETs sit above POSTs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): include inboundIds and traffic in /clients/list
ClientRecord got its own MarshalJSON in the previous commit, and
ClientWithAttachments embeds it to add inboundIds and traffic. Go
promotes the embedded MarshalJSON to the outer struct, so the encoder
was calling ClientRecord.MarshalJSON for the whole value and silently
dropping the extras. The frontend reads row.inboundIds / row.traffic
from /clients/list, so attached inbounds didn't render and newly added
clients looked like they hadn't saved. Add an explicit MarshalJSON on
ClientWithAttachments that splices the extras in.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): gate IP Log on ipLimitEnable + clean access-log dropdown
Legacy panel hid the IP Log section when access logging was off; the
Vue 3 migration left it gated on isEdit only, so the section showed
even when xray's access log was 'none' and nothing was being recorded.
Restore the ipLimitEnable gate on the edit modal's IP Log form-item.
While here, clean up the Xray Settings access-log dropdown: previously
two 'none' entries appeared (an empty value labelled with t('none') and
the literal 'none' from the options array). Drop the empty option for
access log (the literal 'none' covers it) and relabel the empty option
for error log / mask address to t('empty') so they're distinguishable.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(nodes): route per-client ops through node clients API + orphan sweep
Adds Runtime methods AddClient, UpdateUser, and DeleteUser so master
mutates clients on a node via /panel/api/clients/{add,update,del} rather
than pushing the whole inbound. The previous rt.UpdateInbound path made
the node DelInbound+AddInbound on every single-client change, briefly
cycling every other user on the same inbound.
DelInbound no longer filters by enable=true, so a disabled node inbound
actually gets removed from the node instead of being resurrected by the
next snap.
setRemoteTrafficLocked now sweeps any ClientRecord with zero
ClientInbound rows after SyncInbound rebuilds the attachments, which is
how a node-side delete propagates back to master instead of leaving a
detached ghost. ClientService.Delete tombstones the email first so a
snap arriving mid-delete can't re-create the record.
WebSocket broadcasts an "invalidate(clients)" message on every client
mutation so the Clients page refreshes without manual reload.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(balancers): allow fallback on all strategies + feed burstObservatory from random/roundRobin
Drops the random/roundRobin gate on the Fallback field in
BalancerFormModal so every strategy can pick a fallback outbound.
syncObservatories now feeds burstObservatory from leastLoad +
random + roundRobin balancers (was leastLoad only), matching how
leastPing feeds observatory.
Fix the JsonEditor "Unexpected end of JSON input" that appeared
when switching a balancer between leastPing and another strategy:
the obsView watcher was gated on showObsEditor (a boolean OR of
the two flags) and missed the case where one observatory
swapped for the other in the same tick. Watch the individual
flags instead so obsView flips to the surviving editor and the
getter stops pointing at a deleted key.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): use sortedInbounds for mobile empty-state check
InboundList referenced an undefined visibleInbounds in the mobile
card list's empty-state guard, throwing "Cannot read properties of
undefined (reading 'length')" and breaking the entire mobile render.
* feat(clients): sortable table columns
Adds the same sortState / sortableCol / sortFns pattern InboundList
uses, wrapping filteredClients in sortedClients so sort composes with
the existing search/filter pipeline. Sortable: enable, email,
inboundIds (attachment count), traffic, remaining, expiryTime;
actions and online stay unsorted.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(shadowsocks): generate valid ss2022 keys and per-client method for legacy ciphers
The Add Client flow on shadowsocks inbounds was producing xray configs
that failed to start:
- 2022-blake3-* ciphers need a base64-encoded key of an exact byte
length per cipher. fillProtocolDefaults was assigning a uuid-style
string, which xray rejects as "bad key". Now the password is
generated (or replaced if invalid) via random.Base64Bytes(n) sized
to the chosen cipher.
- Legacy ciphers (aes-256-gcm, chacha20-*, xchacha20-*) require a
per-client method field in multi-user mode; model.Client has no
Method, so settings.clients was stored without one and xray failed
with "unsupported cipher method:". applyShadowsocksClientMethod
now injects the top-level method into each client on add/update,
and healShadowsocksClientMethods backfills it at xray-config-build
time so existing inbounds heal on the next start.
- xray/api.go ssCipherType switch was missing aes-256-gcm, which
fell through to ss2022 path.
- SSMethods dropdown now offers aes-256-gcm.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): preserve ClientRecord on inbound delete + filter Attached inbounds to multi-client protocols
Replace the global orphan sweep in setRemoteTrafficLocked with a
per-inbound diff cleanup: only delete a ClientRecord whose email
disappeared from a snap-tracked inbound (i.e. a node-side delete).
Inbounds that vanished entirely from the snap (e.g. admin deleted
the inbound on master) aren't iterated, so a client whose last
attachment came from that inbound is now left alone instead of
being deleted alongside the inbound.
ClientFormModal and ClientBulkAddModal now filter the Attached
inbounds dropdown to protocols that actually support multiple
clients: shadowsocks, vless, vmess, trojan, hysteria, hysteria2,
and portfallback (which routes through VLESS settings).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): make empty-state text readable on dark/ultra themes
The "No clients yet" empty state had a hardcoded black color
(rgba(0,0,0,0.45)) that vanished against the dark backgrounds.
Drop the inline color, let it inherit from the AntD theme, and
fade with opacity like the mobile card empty state already does.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): client-first tgbot add flow, tgId field, lightweight inbound options
- tgbot: drop legacy per-protocol Add Client UI in favour of a client-first
multi-inbound flow. New BuildClientDraftMessage / getInboundsAttachPicker
let an admin pick one or more inbounds and submit a single client; per-
protocol secrets are now generated server-side via fillProtocolDefaults.
Drops awaiting_id/awaiting_password_tr/awaiting_password_sh state cases
and add_client_ch_default_id/pass_tr/pass_sh/flow callbacks. Adds a
setTGUser button + awaiting_tg_id state so the bot can set Client.TgID
during Add.
- clients UI: add Telegram user ID input to ClientFormModal (0 = none).
Hide IP Limit field entirely when ipLimitEnable is off — disabled fields
still take layout space, this collapses Auth(Hysteria) to full width.
- inbounds API: new GET /panel/api/inbounds/options that returns just
{id, remark, protocol, port, tlsFlowCapable}. Used by the clients page
pickers so the dropdown payload stays small on panels with thousands of
clients (drops settings JSON, clientStats, streamSettings). Server-side
TlsFlowCapable mirrors Inbound.canEnableTlsFlow so the modal no longer
needs to parse streamSettings client-side.
- clientInfoMsg now shows attached inbound remarks, and getInboundUsages
reports the attached client count per inbound.
- api-docs: document the new /options endpoint and add tgId / flow to the
clients add/update bodies.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): keep Node column visible for node-attached inbounds
The Node column was bound to hasActiveNode, so disabling every node hid
the column even when inbounds were still attached to those nodes — the
admin lost the visual cue that those inbounds belonged to a node and
would come back when it was re-enabled. Combine hasActiveNode with a
new hasNodeAttachedInbound check (any dbInbound with nodeId != null) so
the column survives node-disable.
* fix(api-docs): accept functional-component icons in EndpointSection
AntD-Vue icons (SafetyCertificateOutlined, etc.) are functional
components, so the icon prop's type: Object validator was rejecting
them with a "Expected Object, got Function" warning at runtime.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* test: cover crypto, random, netsafe, sub helpers, xray equals, websocket hub, node service
Adds ~110 unit tests across previously untested packages. Focus on
pure-logic and concurrency surfaces where regressions would silently
affect users:
- util/crypto, util/random: password hashing round-trip, ss2022 key
generation, alphabet/length invariants.
- util/netsafe: IsBlockedIP edge cases, NormalizeHost validation,
SSRF guard with AllowPrivate context bypass.
- util/common, util/json_util: traffic formatter, Combine nil-skip,
RawMessage empty-as-null and copy-on-unmarshal.
- sub: splitLinkLines, searchKey/searchHost, kcp share fields,
finalmask normalization, buildVmessLink round-trip.
- xray: Config.Equals and InboundConfig.Equals field-by-field,
getRequiredUserString/getOptionalUserString type checks.
- web/websocket: hub registration, throttling, slow-client eviction,
nil-receiver safety, concurrent register/unregister.
- web/service: NodeService.normalize validation, normalizeBasePath,
HeartbeatPatch.ToUI mapping.
- web/job: atomicBool concurrent set/takeAndReset semantics.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n(clients): replace English fallbacks with proper translation keys
Pulls every hard-coded English label/title in the Clients page and its
four modals through the i18n layer so localized panels stop leaking
English. New keys live under pages.clients (auth, hysteriaAuth, uuid,
flow, flowNone, reverseTag, reverseTagPlaceholder, telegramId,
telegramIdPlaceholder, created, updated, ipLimit) plus refresh at the
root and toasts.bulkDeletedMixed / bulkCreatedMixed for partial-failure
toasts. Also switches the add-client modal's primary button from "Add"
to "Create" for consistency with other create flows.
The bulk-add Random/Random+Prefix/... email-method options stay
hard-coded by request - they're identifier-shaped strings.
* i18n: backfill 99 missing keys across all 12 non-English locales
Brings every translation file up to parity with en-US.json so the
Clients page, the fallback-children inbound section, the new refresh
verb, the Nodes panel-version label and a handful of older holes stop
falling through to the English fallback. New strings span:
- pages.clients.* (labels, confirmations, toasts, emailMethods)
- pages.inbounds.portFallback.* (Reality fallback inbound section)
- pages.nodes.panelVersion, menu.clients, refresh
Technical identifiers (Auth, UUID, Flow, Reverse tag) are intentionally
left untranslated since they correspond to xray-core field names.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n: drop stale pages.client block duplicated in every non-English locale
Every non-English locale carried a pages.client (singular) section with
30 entries that duplicated pages.clients (plural). The plural namespace
is what the Vue code actually consumes; the singular one was dead
weight from an older rename that never got cleaned up in the
non-English files. Removing it brings every locale to exactly 984
keys, matching en-US.json.
* chore: apply modernize analyzer fixes across codebase
Mechanical replacements suggested by golang.org/x/tools/.../modernize:
strings.Cut/CutPrefix/SplitSeq, slices.Contains, maps.Copy, min(),
range-over-int, new(expr), strings.Builder for hot += loops,
reflect.TypeFor[T](), sync.WaitGroup.Go(), drop legacy //+build lines.
* feat(database): add PostgreSQL as an optional backend alongside SQLite
Lets operators with large client counts or multi-node setups pick PostgreSQL
at install time without breaking the existing SQLite default. Backend is
selected at runtime via XUI_DB_TYPE/XUI_DB_DSN, a small dialect layer keeps
the five JSON_EXTRACT/JSON_EACH queries portable, and a new `x-ui migrate-db`
subcommand copies SQLite data into PostgreSQL in FK-aware order.
* fix(inbounds): gate node selector to multi-node-capable protocols
Hide the Deploy-To selector and clear nodeId when switching to a
protocol that can't run on a remote node. Also:
- subs: return 404 (not 400) when subId matches no inbounds, so VPN
clients distinguish "deleted/unknown" from a server error
- hysteria link gen: use the inbound's resolved address so node-managed
inbounds advertise the node host instead of the central panel
- shadowsocks: default network to 'tcp' (udp was causing issues for some
clients on first-create)
- vite dev proxy: rewrite migrated-route bypass against the live base
path instead of a hardcoded single-segment regex
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): bulk add/delete correctness + perf, working pagination, delayed-start in form
Bulk add/delete were serial on the frontend (one toast per call, N round-trips)
and the backend race exposed by parallelizing them lost client attachments and
hit UNIQUE constraint failed on client_inbounds. The single add/edit modal also
had no Start-After-First-Use option, and the table never showed the delayed
duration.
Backend (web/service/client.go):
- Per-inbound mutex on Add/Update/Del InboundClient so concurrent writers on
the same inbound don't lose the read-modify-write of settings JSON.
- SyncInbound skips create+join when the email is tombstoned so a concurrent
maintenance pass (adjustTraffics, autoRenewClients, markClientsDisabledIn-
Settings) that did a stale RMW can't resurrect a just-deleted client with a
fresh id.
- compactOrphans sweeps settings.clients entries whose ClientRecord no longer
exists, applied in Add/DelInboundClient + DelInboundClientByEmail so each
user-initiated mutation self-heals the inbound's settings.
- DelInboundClient uses Pluck instead of First for the stats lookup so a
missing row doesn't abort the delete with a noisy ErrRecordNotFound log.
Frontend:
- HttpUtil.{get,post} accept a silent option that suppresses the auto-toast.
- ClientBulkAddModal fires creates in parallel + silent + one summary toast.
- useClients.removeMany runs deletes in parallel + silent and refreshes once;
ClientsPage bulk delete uses it and shows one aggregate toast.
- useClients.applyInvalidate debounces 200 ms so the burst of N WebSocket
invalidate events from the backend collapses into a single refresh.
- ClientsPage pagination is reactive (paginationState ref + tablePagination
computed); onTableChange persists page-size and page changes.
- ClientFormModal gains a Start-After-First-Use switch + Duration days input
alongside the existing Expiry Date picker; on edit-mode open a negative
expiryTime is decoded back to delayed mode + days; on submit the payload
sends -86400000 * days or the absolute timestamp.
- ClientsPage table shows the delayed-start duration (blue tag Nd, tooltip
Start After First Use: Nd) instead of infinity.
- Telegram ID field in the form is hidden when /panel/setting/defaultSettings
reports tgBotEnable=false; Comment then fills the row.
- Form row 3 collapses UUID (span 12) + Total GB (span 8) + Limit IP (span 4)
when ipLimitEnable is on, else UUID + Total GB at 12/12.
- useInbounds.rollupClients counts only clients with a matching clientStats
row, so orphans in settings.clients no longer inflate the inbound's count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(windows): clean shutdown, working panel restart, harden kernel32 load
Three Windows-specific issues addressed:
1. Orphaned xray-windows-amd64 after VS Code debugger stop. Delve's
"Stop" sends TerminateProcess to the Go binary, which is uncatchable
— our signal handlers never run, so xrayService.StopXray() is skipped
and xray is left dangling. Spawn xray as a child of a Job Object with
JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE so the OS kills xray when our
handle to the job is closed (which happens even on TerminateProcess).
Also trap os.Interrupt in main so Ctrl+C in the terminal runs the
graceful path.
2. /panel/setting/restartPanel logged "failed to send SIGHUP signal: not
supported by windows" because Windows can't deliver arbitrary signals.
Add a restart hook in web/global; main registers it to push SIGHUP
into its own signal channel, and RestartPanel calls the hook before
falling back to the (Unix-only) signal path. Same restart-loop code
runs in both cases.
3. util/sys/sys_windows.go now uses windows.NewLazySystemDLL so the
kernel32.dll resolve is pinned to %SystemRoot%\System32 (prevents
DLL hijacking by a planted DLL next to the binary). Local filetime
type replaced with windows.Filetime, and the unreliable
syscall.GetLastError() fallback replaced with a type assertion on the
errno captured at call time.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(sys): correct CPU/connection accounting on linux + darwin
util/sys/sys_linux.go:
- GetTCPCount/GetUDPCount were counting the column header row in
/proc/net/{tcp,udp}[6] as a connection, inflating the reported total
by 1 per non-empty file (so the panel status line always showed 2
more connections than actually existed). Replace getLinesNum +
safeGetLinesNum with a single bufio.Scanner-based countConnections
that skips the header.
- CPUPercentRaw now opens HostProc("stat") instead of a hardcoded
/proc/stat so HOST_PROC overrides apply, matching the connection
counters in the same file.
- Simplify CPU field unpacking: pad nums to 8 once instead of guarding
every assignment with a len check.
util/sys/sys_darwin.go:
- Fix swapped idle/intr indices on kern.cp_time. BSD CPUSTATES order
is user, nice, sys, intr, idle (CP_INTR=3, CP_IDLE=4) — gopsutil's
cpu_darwin_nocgo.go reads the same layout. The previous code used
out[3] as idle and out[4] as intr, so busy = total - dIdle was
actually subtracting interrupt time, making the panel report CPU
usage close to 100% on macOS regardless of actual load.
- Collapse the per-field delta math into a single loop.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(xray): rotate crash reports into log folder, prevent overwrites
writeCrashReport had two flaws: it wrote to the bin folder (alongside the
xray binary) which conflates artifacts, and the second-precision timestamp
meant a tight restart-loop crash burst overwrote prior reports. Write to
the log folder with nanosecond precision and keep the last 10 reports.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* revert(inbounds): drop unreleased portfallback protocol
The Port-with-Fallback inbound (commit 62fd9f9d) was confusing as a
standalone protocol — fallbacks belong on a regular VLESS/Trojan TCP-TLS
inbound, the way Xray models them natively. Rip out the entire feature
cleanly (no migration needed since it was never released): protocol
constant, fallback children DB table, FallbackService, 2 API endpoints,
all UI rows, related translations and api-docs. A native fallback flow
attached to VLESS/Trojan TCP-TLS/Reality will land in a follow-up commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): native fallbacks on VLESS/Trojan TCP-TLS, with working child links
A VLESS or Trojan inbound on TCP with TLS or Reality can now act as a
fallback master: pick existing inbounds as children and the panel auto-
fills the SNI / ALPN / path / xver routing fields from each child's
transport, auto-builds settings.fallbacks at config-gen time, and
rewrites the child's client-share link so it advertises the master's
reachable endpoint and TLS state instead of the child's loopback listen.
Layout matches the Xray All-in-One Nginx example: master at :443 with
clients + TLS, each child on 127.0.0.1 with its own transport+clients.
Order matters (Xray walks fallbacks top-to-bottom) — reorder via the
per-row up/down arrows. Path / SNI / ALPN are exposed under a per-row
Edit toggle for the rare cases where the auto-derivation needs
overriding; otherwise just pick a child and you're done.
Backend: new InboundFallback table + FallbackService (GetByMaster /
SetByMaster / GetParentForChild / BuildFallbacksJSON); two routes
(GET / POST /panel/api/inbounds/:id/fallbacks); xray.GetXrayConfig
injects settings.fallbacks for any VLESS/Trojan TCP-TLS/Reality
inbound; GetInbounds annotates each child with FallbackParent so the
frontend can rewrite links without an extra round-trip.
Link projection covers every emission path — clients-page QR/links,
per-inbound Get URL, raw subscription, sub-JSON, sub-Clash, and the
inbounds-page link/info/QR — via a shared projectThroughFallbackMaster
on the backend and a shared projectChildThroughMaster on the frontend
that both handle the panel-tracked relationship and the legacy
unix-socket (@vless-ws) convention.
Strings translated into all 12 non-English locales.
* docs: rewrite CONTRIBUTING with full local-dev setup
The prior three-line CONTRIBUTING left newcomers guessing at every
non-trivial step: which Go / Node versions, where xray comes from, why
the panel goes blank when XUI_DEBUG=true is flipped on, how the Vue
multi-page setup is wired, what to do on Windows when go build trips
on the CGo SQLite driver.
Now covers prerequisites, MinGW-w64 install on Windows (niXman builds
or MSYS2), one-shot first-time setup, two frontend dev workflows with
the XUI_DEBUG asset-cache gotcha called out, the architecture and
conventions of the Vue side, a project-layout map, useful env vars,
and the PR checklist.
---------
2026-05-19 03:16:42 -07:00
else
2026-05-19 04:09:35 -07:00
local config_webBasePath = $( gen_random_string 18)
echo -e " ${ yellow } WebBasePath is missing or too short. Generating a new one... ${ plain } "
${ xui_folder } /x-ui setting -webBasePath " ${ config_webBasePath } "
echo -e " ${ green } New WebBasePath: ${ config_webBasePath } ${ plain } "
# If the panel is already installed but no certificate is configured, prompt for SSL now
if [ [ -z " ${ existing_cert } " ] ] ; then
echo ""
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } SSL Certificate Setup (RECOMMENDED) ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ yellow } Let's Encrypt now supports both domains and IP addresses! ${ plain } "
echo ""
prompt_and_setup_ssl " ${ existing_port } " " ${ config_webBasePath } " " ${ server_ip } "
echo -e " ${ green } Access URL: ${ SSL_SCHEME } :// ${ SSL_HOST } : ${ existing_port } / ${ config_webBasePath } ${ plain } "
Feat/multi inbound clients (#4469)
* feat(clients): add shadow tables for first-class client promotion
Introduces three new GORM-backed tables (clients, client_inbounds,
inbound_fallback_children) and a populate-only seeder that backfills
them from each inbound's existing settings.clients JSON. Duplicate
emails across inbounds auto-merge under one client row, with each
field conflict logged. Existing services are unchanged and continue
reading from settings.clients — this commit is groundwork only.
* feat(clients): make clients+client_inbounds the runtime source of truth
Adds ClientService.SyncInbound that reconciles the new tables from
each inbound's clients list whenever existing service paths mutate
settings.clients. Wires it into AddInbound, UpdateInbound,
AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, DelDepletedClients, autoRenewClients, and
the timestamp-backfill path in adjustTraffics, plus DetachInbound
on DelInbound.
GetXrayConfig now builds settings.clients from the new tables before
writing config.json, and getInboundsBySubId joins through them
instead of JSON_EACH on settings JSON. Live Xray config and
subscription endpoints are now driven by the relational view;
settings.clients JSON stays in step as a side effect of every write.
* feat(clients): add top-level Clients tab and CRUD API
Adds /panel/api/clients endpoints (list, get, add, update, del,
attach, detach) backed by ClientService methods that orchestrate
the per-inbound Add/Update/Del flows so a single client row is
created once and attached to many inbounds in one operation.
The frontend gains a dedicated Clients page (frontend/clients.html
+ src/pages/clients/) with an AntD table, multi-inbound attach
modal, and full CRUD. Axios interceptor learns to honour
Content-Type: application/json so the JSON endpoints work
alongside the legacy form-encoded ones.
The legacy per-inbound client modal stays untouched in this PR —
both flows now write to the same source of truth.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): add Port-with-Fallback inbound type
Adds a new "portfallback" protocol that emits as a VLESS-TLS inbound
under the hood but is paired with a sidecar table of child inbounds.
Panel auto-builds settings.fallbacks at Xray-config-gen time from the
sidecar — each child's listen+port becomes the fallback dest, with
SNI/ALPN/path/xver match criteria pulled from the row. No more typing
loopback ports by hand or keeping settings.fallbacks in sync.
Backend: new FallbackService (Get/SetChildren, BuildFallbacksJSON);
two new routes (GET/POST /panel/api/inbounds/:id/fallbackChildren);
xray.GetXrayConfig injects fallbacks for PortFallback inbounds; the
inbound model emits protocol="vless" so Xray accepts the config.
Frontend: PORTFALLBACK joins the protocol dropdown; selecting it
shows the standard VLESS controls plus a Fallback Children table
(inbound picker + per-row SNI/ALPN/path/xver). Children are loaded
on edit and replaced atomically on save.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reset Traffic, QR Code, Info actions + Online/Remaining columns
The Clients page table gains:
- Online column — green/grey tag driven by /panel/api/inbounds/onlines,
polled every 10s.
- Remaining column — bytes-remaining tag, coloured green/orange/red
against quota, purple infinity when unlimited.
- Action icons per row: QR, Info, Reset traffic, Edit, Delete.
ClientInfoModal shows the full client detail (uuid/password/auth,
traffic ↑/↓ + remaining + all-time, expiry absolute + relative,
attached inbounds chip list, online + last-online).
ClientQrModal fetches links for the client's subId via
/panel/api/inbounds/getSubLinks/:subId and renders each one through
the existing QrPanel component.
Reset Traffic confirms then calls the existing per-inbound endpoint
on the client's first attached inbound (the traffic row is keyed on
email globally, so any attached inbound resets the shared counter).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): expose Attached inbounds in edit mode
The multi-select was gated on add-only, so editing a client had no way
to change which inbounds it belonged to. The picker now shows in both
modes, and on submit the modal diffs the picked set against the
original attachedIds — additions go through the /attach endpoint,
removals through /detach, both after the field update lands so the
new attachments get the latest values.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): unbreak template parsing + stale i18n keys
- InboundFormModal: split the multi-line help string in the
PortFallback section onto one line — Vue's template parser was
bailing on Unterminated string constant because a single-quoted
literal spanned two lines inside a {{ }} interpolation.
- ClientInfoModal: t('disable') was missing at the root level, so
vue-i18n returned the key path literally. Use t('disabled') which
exists.
- Linter cleanup elsewhere: pages.client.* references renamed to
pages.clients.* to match the merged i18n block; whitespace
normalisation in a few unrelated Vue templates.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* 1
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(traffic): drop all-time traffic tracking
Removes the AllTime field from Inbound and ClientTraffic and migrates
existing DBs by dropping the all_time columns on startup. The counter
duplicated up+down without adding signal, and the per-event accumulator
ran on every traffic write.
Frontend: drop the All-time column from the inbound list and the
client-row table, the All-time row from the client info modal, and the
All-Time Total Usage tile from the inbounds summary card. The
allTimeTraffic/allTimeTrafficUsage i18n keys are removed across every
locale.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): mobile cards, multi-select, bulk add
Adds the same row-card layout the inbounds page uses on mobile: the
table is suppressed under the mobile breakpoint and each client renders
as a compact card with a status dot, email, Info button, Enable switch,
and overflow menu. All the per-client detail (traffic, remaining,
expiry, attached inbounds, flow, created/updated, URL, subscription)
opens through the existing info modal.
Multi-select with bulk delete wires AntD row-selection on desktop and
a per-card checkbox on mobile; a Delete (N) button appears in the
toolbar when anything is selected.
Bulk add reuses the five email-generation modes from the inbound bulk
modal but takes a multi-inbound picker so one bulk run can attach to
several inbounds at once. Submits client-by-client through the
existing /panel/api/clients/add endpoint.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): remove legacy per-inbound client UI
Now that clients live as first-class rows attached to one or many
inbounds, the per-inbound client UI on the inbounds page is dead
weight — every client action either has a global equivalent on the
Clients page or makes no sense in a many-to-many world.
Deletes ClientFormModal, ClientBulkModal, CopyClientsModal, and
ClientRowTable from inbounds/. Strips the matching emits, refs,
handlers, and dropdown menu items from InboundList and InboundsPage,
and removes the dead mobile expand-chevron state and the desktop
expanded-row plumbing that drove the inline client table.
The InboundFormModal Clients tab still works in add-mode (one inline
client at inbound creation) — that flow goes through ClientService.
SyncInbound on save and remains useful.
Fixes a stray "</a-dropdown>" left over by an earlier toolbar edit
in ClientsPage that broke the template parser.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Delete depleted action
Mirrors the legacy delDepletedClients action that lived under the
inbounds page, but as a first-class /panel/api/clients/delDepleted
endpoint backed by ClientService. The new path goes through
ClientService.Delete for each depleted email, so the new clients +
client_inbounds + xray_client_traffic tables stay consistent.
Adds a danger-styled toolbar button on the Clients page (next to
Reset all client traffic) with a confirm dialog and a toast
reporting the deleted count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): move every client-shaped endpoint off /inbounds onto /clients
After the multi-inbound client migration, client state belongs to the
client API surface, not the inbound one. Twelve routes that were
crammed under /panel/api/inbounds/* now live where they belong, under
/panel/api/clients/*.
Moved (route, handler, doc):
POST /clientIps/:email
POST /clearClientIps/:email
POST /onlines
POST /lastOnline
POST /updateClientTraffic/:email
POST /resetAllClientTraffics/:id
POST /delDepletedClients/:id
POST /:id/resetClientTraffic/:email
GET /getClientTraffics/:email
GET /getClientTrafficsById/:id
GET /getSubLinks/:subId
GET /getClientLinks/:id/:email
Their /clients/* counterparts are:
POST /clients/clientIps/:email
POST /clients/clearClientIps/:email
POST /clients/onlines
POST /clients/lastOnline
POST /clients/updateTraffic/:email
POST /clients/resetTraffic/:email (email-only, fans out)
GET /clients/traffic/:email
GET /clients/traffic/byId/:id
GET /clients/subLinks/:subId
GET /clients/links/:id/:email
per-inbound resetAllClientTraffics and delDepletedClients are dropped
entirely — the Clients page already exposes global Reset All Traffic
and Delete depleted actions, and per-inbound resets are meaningless
once a client can be attached to many inbounds.
ClientService.ResetTrafficByEmail is the new email-only reset path:
it looks up every inbound the client is attached to and pushes the
counter reset + Xray re-add through inboundService.ResetClientTraffic
for each one, so depleted users come back online instantly.
Frontend callers (ClientsPage, useClients, ClientQrModal,
ClientInfoModal, InboundInfoModal, InboundsPage, useInbounds) all
switched to the new paths. The Inbounds page drops its per-inbound
"Reset client traffic" and "Delete depleted clients" dropdown items —
users do those at the client level now. api-docs is rebuilt to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): switch tgbot + ldap callers to ClientService
Adds two thin helpers to ClientService (CreateOne, DetachByEmail) and
rewrites tgbot.SubmitAddClient and ldap_sync_job to call ClientService
directly. Removes the JSON-blob payloads (BuildJSONForProtocol output for
add, clientsToJSON/clientToJSON helpers) that callers previously fed to
InboundService.AddInboundClient/DelInboundClient.
ldap_sync_job.batchSetEnable now loops InboundService.SetClientEnableByEmail
per email instead of trying to coerce AddInboundClient into doing the
update — the old path would have failed duplicate-email validation for
existing clients anyway.
The legacy InboundService.AddInboundClient/UpdateInboundClient/
DelInboundClient methods stay in place; they are now only used internally
by ClientService Create/Update/Delete/Attach. Inlining + deleting them
follows in a separate commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(service): move all client mutation methods to ClientService
Moves the client mutation surface out of InboundService and into
ClientService. These methods all operate on a single client (identity
fields, traffic limits, expiry, ip limit, enable state, telegram tg id)
and didn't belong on the inbound aggregate.
Moved (12 methods): AddInboundClient, UpdateInboundClient, DelInboundClient,
DelInboundClientByEmail, checkEmailsExistForClients, SetClientTelegramUserID,
checkIsEnabledByEmail, ToggleClientEnableByEmail, SetClientEnableByEmail,
ResetClientIpLimitByEmail, ResetClientExpiryTimeByEmail,
ResetClientTrafficLimitByEmail.
Each method now takes an explicit *InboundService for the helpers that
legitimately stay on InboundService (GetInbound, GetClients, runtimeFor,
AddClientStat / UpdateClientStat / DelClientStat, DelClientIPs /
UpdateClientIPs, emailUsedByOtherInbounds, getAllEmailSubIDs,
GetClientInboundByEmail / GetClientInboundByTrafficID,
GetClientTrafficByEmail).
Stays on InboundService: ResetClientTrafficByEmail and
ResetClientTraffic(id, email) — these mutate xray_client_traffic rows,
not client identity, so they're inbound-side bookkeeping.
Callers updated: tgbot (6 calls), ldap_sync_job (1 call),
InboundService internal (writeBackClientSubID, CopyInboundClients,
AddInbound's email-uniqueness check), ClientService Create/Update/
Delete/Attach/Detach.
Also removes a dead resetAllClientTraffics controller handler whose
route was already gone after the previous /clients API migration.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): finish migrating to ClientService + tidy IP routes
Two related cleanups in the new /clients surface:
1. Move ResetAllClientTraffics (bulk-reset of xray_client_traffic +
last_traffic_reset_time, with node-runtime propagation) from
InboundService to ClientService. PeriodicTrafficResetJob now holds
a clientService and calls
j.clientService.ResetAllClientTraffics(&j.inboundService, id).
The last client-mutation method on InboundService is gone.
2. Shorten redundantly-named routes/handlers under /panel/api/clients:
- /clientIps/:email -> /ips/:email (handler getIps)
- /clearClientIps/:email -> /clearIps/:email (handler clearIps)
The "client" prefix was redundant inside the clients namespace.
Frontend (InboundInfoModal) and api-docs updated to match.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds,clients): clean up inbound modal + enrich client modal
Inbound modal rework (InboundFormModal.vue + inbound.js):
- Drop the embedded Client subform in the Protocol tab. Multi-inbound
clients are managed exclusively from the Clients page now; a fresh
inbound is created with zero clients (settings constructors default
to []) and the user attaches clients afterwards.
- Hide the Protocol tab entirely when it has nothing to render
(VMESS, Trojan without fallbacks, Hysteria). Auto-switches active
tab to Basic when the tab disappears while focused.
- Move the Security section (Security selector + TLS block with certs
and ECH + Reality block) out of the Stream tab into its own
Security tab, sharing the canEnableStream gate.
Client modal additions (ClientFormModal.vue + ClientBulkAddModal.vue):
- Flow select (xtls-rprx-vision / -udp443) appears only when the
panel actually has a Vision-capable inbound (VLESS or PortFallback
on TCP with TLS or Reality). Hidden otherwise, and cleared when
it disappears.
- IP Limit input is disabled when the panel-level ipLimitEnable
setting is off, fetched into useClients alongside subSettings and
threaded through ClientsPage to both modals.
- Edit modal now shows an "IP Log" section listing IPs that have
connected with the client's credentials, with refresh and clear
buttons (calls the renamed /panel/api/clients/ips and /clearIps
endpoints).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(inbounds): drop manual Fallbacks UI from inbound modal
The PortFallback protocol type now covers the common
VLESS-master-plus-children case with auto-wired dests, so the manual
Fallbacks editor (showFallbacks block in the Protocol tab) is mostly
redundant. Removed:
- the v-if="showFallbacks" template block (SNI/ALPN/Path/dest/PROXY rows)
- the showFallbacks computed
- the addFallback / delFallback helpers
- the .fallbacks-header / .fallbacks-title styles
- the showFallbacks gate from hasProtocolTabContent (so Trojan-over-TCP
no longer shows an empty Protocol tab)
Power users who need a non-inbound fallback dest (nginx, static site)
can still author settings.fallbacks via the Advanced JSON tab.
* feat(clients,inbounds): move search/filter to Clients page + small fixes
Search/filter relocation:
- Remove the search/filter toolbar (search switch + filter radio +
protocol/node selects + the visibleInbounds projection +
inboundsFilterState localStorage + filter CSS + the SearchOutlined/
FilterOutlined/ObjectUtil/Inbound imports it required) from
InboundList. The filters were all client-oriented buckets bolted
onto the inbound row.
- Add a search/filter toolbar to ClientsPage with the same shape:
switch between deep-text search and bucket filter (active /
deactive / depleted / expiring / online) + protocol filter that
matches clients attached to at least one inbound with the chosen
protocol. State persists in clientsFilterState localStorage.
filteredClients drives both the desktop table and the mobile card
list, and select-all / allSelected / someSelected only span the
visible subset.
- useClients now also fetches expireDiff and trafficDiff from
/panel/setting/defaultSettings (used to detect the expiring
bucket); ClientsPage threads them into the client-bucket helper.
Loose fixes folded in:
- Add Client: email field is auto-filled with a random handle on
open, matching uuid/subId/password/auth.
- Inbound clone: parse and reuse the source settings JSON (with
clients reset to []) instead of building a fresh defaulted
Settings, so VLESS Encryption/Decryption and other non-client
fields survive the clone.
- en-US.json: add the ipLog string used by the edit-client modal.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): add Reverse tag field for VLESS-attached clients
Mirrors the Flow field's pattern: a Reverse tag input appears in the
Add/Edit Client modal whenever at least one selected inbound is VLESS
or PortFallback. The value rides over the wire as
client.reverse = { tag: '...' } so it lands directly in model.Client's
*ClientReverse field; an empty value omits the reverse key entirely.
On edit the field is hydrated from props.client.reverse?.tag, and the
showReverseTag watcher clears the field if the user drops the last
VLESS-like inbound from the selection.
* fix(xray): emit only protocol-relevant fields per client entry
The Xray config synthesizer was writing every identifier field (id,
password, flow, auth, security/method, reverse) on every client entry
regardless of the inbound's protocol. Xray ignores unknown fields, so
the config worked, but it diverged from the spec and leaked secrets
across protocols when one client was attached to multiple inbounds —
a VLESS inbound's generated config carried the same client's Trojan
password and Hysteria auth alongside its uuid.
Switch on inbound.Protocol when building each entry:
- VLESS / PortFallback: id, flow, reverse
- VMess: id, security
- Trojan: password, flow
- Shadowsocks: password, method
- Hysteria / Hysteria2: auth
email is emitted for every protocol.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): restore auto-disable kick under new schema
disableInvalidClients still resolved (inbound_tag, email) pairs via
JSON_EACH(inbounds.settings.clients), which is empty after migrating
to the clients + client_inbounds tables. Result: xrayApi.RemoveUser
never ran for depleted clients, clients.enable stayed true so the UI
showed them as active, and only xray_client_traffic.enable got flipped
- making "Restart Xray After Auto Disable" only half-work.
Resolve the targets via a JOIN through the new schema, flip clients.enable
so the Clients page reflects the state, and drop the legacy JSON
write-back plus the subId cascade workaround (email is unique now).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): live WebSocket updates + Ended status surfacing
ClientsPage now subscribes to traffic / client_stats / invalidate
WebSocket events instead of polling /onlines every 10s. Per-row
traffic counters refresh in place, online state stays current, and
list-level mutations elsewhere trigger a refresh.
The client roll-up summary moves from InboundsPage to ClientsPage
where it belongs, restructured into six labeled stat tiles
(Total / Online / Ended / Expiring / Disabled / Active) with email
popovers on the ones with issues.
Auto-disabled clients (traffic exhausted or expiry passed) now
classify as 'depleted' even though clients.enable=false, so they
show up under the Ended filter and render a red Ended tag instead
of looking indistinguishable from an operator-disabled row.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(nodes): per-node client roll-up and panel version
Added transient inboundCount / clientCount / onlineCount /
depletedCount fields to model.Node, populated by NodeService.GetAll
via aggregated queries (one join across inbounds + client_inbounds,
one over client_traffics intersected with the in-memory online
emails). The Nodes list renders these as colored chips on a new
"Clients" column so an operator can see at a glance how many users
each node carries and how many are currently online or depleted.
Also exposes the remote panel's version. The central panel adds
panelVersion to its /api/server/status payload (sourced from
config.GetVersion). Probe reads that field and persists it on the
node row, mirroring how xrayVersion already flows. NodesPage gets
a new column next to Xray Version, in both desktop and mobile
views, with English and Persian strings.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): stop node sync from resurrecting deleted clients
Several related issues around node-managed clients:
- Remote runtime: drop the per-inbound resetAllClientTraffics path
and point traffic/onlines/lastOnline fetches at the new
/panel/api/clients/* routes.
- Delete from master: always push the updated inbound to the node
even when the client was already disabled or depleted, so the
node actually loses the user instead of silently keeping it.
- setRemoteTraffic: mirror remote clients into the central tables
only on first discovery of a node inbound. Matched inbounds let
the master own the join table, so a stale snap can no longer
re-create a ClientRecord (and join row) for a client that was
just deleted on the master.
- ClientService.Delete: route through submitTrafficWrite so deletes
serialize with node traffic merges, and switch the final
ClientRecord delete to an explicit Where("id = ?") clause.
- setRemoteTraffic UNIQUE-constraint fix: use clause.OnConflict on
inserts and email-keyed UPDATEs for client_traffics, so mirroring
a snap doesn't trip the unique email index.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(clients): switch client API endpoints from id to email
All client-scoped routes now use the unique email as the path key
(get, update, del, attach, detach, links). Email is the stable,
protocol-independent identifier — UUIDs don't exist for trojan or
shadowsocks, and internal numeric ids leaked panel implementation
detail into the public API.
Removed the redundant /traffic/byId/:id endpoint (covered by
/traffic/:email) and collapsed /links/:id/:email into /links/:email,
which now returns links across every attached inbound for the client.
Frontend selection, bulk delete, and toggle state are now keyed by
email as well, dropping the id→email lookup workaround.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(server): move cached state and helpers into ServerService
ServerController had grown to hold its own status cache, version-list
TTL cache, history-bucket whitelist, and the loop that drove all three
— concerns that belong in the service layer. Pull them out:
- lastStatus + the @2s refresh become ServerService.RefreshStatus and
ServerService.LastStatus; the controller's cron now just orchestrates
the cross-service side effects (xrayMetrics sample, websocket broadcast).
- The 15-minute Xray-versions cache (with stale-on-error fallback) moves
into ServerService.GetXrayVersionsCached, collapsing the controller
handler to a single call.
- The freedom/blackhole outbound-tag walk used by /xraylogs becomes
ServerService.GetDefaultLogOutboundTags.
- The allowed-history-bucket whitelist moves to package-level
service.IsAllowedHistoryBucket, so both NodeController and
ServerController validate against the same list.
Net result: web/controller/server.go drops from 458 to 365 lines and
contains only HTTP wiring + presentation-y side effects.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(api): emit JSON-text columns as nested objects
Inbound, ClientRecord, and InboundClientIps store settings /
streamSettings / sniffing / reverse / ips as JSON-text in the DB. The
API was passing that text through verbatim, so every consumer had to
JSON.parse a string inside a string. Add MarshalJSON / UnmarshalJSON so
the wire format is a real nested object, while still accepting the
legacy escaped-string shape on write. Frontend dbinbound.js gets a
matching coerceInboundJsonField helper for the same dual-shape read
path, and inbound.js toJson stops emitting empty/placeholder fields
(externalProxy [], sniffing destOverride when disabled, etc.) so the
new normalised JSON stays terse. api-docs and the inbound-clone path
are updated to the new shape. Controller route lists are regrouped so
all GETs sit above POSTs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): include inboundIds and traffic in /clients/list
ClientRecord got its own MarshalJSON in the previous commit, and
ClientWithAttachments embeds it to add inboundIds and traffic. Go
promotes the embedded MarshalJSON to the outer struct, so the encoder
was calling ClientRecord.MarshalJSON for the whole value and silently
dropping the extras. The frontend reads row.inboundIds / row.traffic
from /clients/list, so attached inbounds didn't render and newly added
clients looked like they hadn't saved. Add an explicit MarshalJSON on
ClientWithAttachments that splices the extras in.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): gate IP Log on ipLimitEnable + clean access-log dropdown
Legacy panel hid the IP Log section when access logging was off; the
Vue 3 migration left it gated on isEdit only, so the section showed
even when xray's access log was 'none' and nothing was being recorded.
Restore the ipLimitEnable gate on the edit modal's IP Log form-item.
While here, clean up the Xray Settings access-log dropdown: previously
two 'none' entries appeared (an empty value labelled with t('none') and
the literal 'none' from the options array). Drop the empty option for
access log (the literal 'none' covers it) and relabel the empty option
for error log / mask address to t('empty') so they're distinguishable.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(nodes): route per-client ops through node clients API + orphan sweep
Adds Runtime methods AddClient, UpdateUser, and DeleteUser so master
mutates clients on a node via /panel/api/clients/{add,update,del} rather
than pushing the whole inbound. The previous rt.UpdateInbound path made
the node DelInbound+AddInbound on every single-client change, briefly
cycling every other user on the same inbound.
DelInbound no longer filters by enable=true, so a disabled node inbound
actually gets removed from the node instead of being resurrected by the
next snap.
setRemoteTrafficLocked now sweeps any ClientRecord with zero
ClientInbound rows after SyncInbound rebuilds the attachments, which is
how a node-side delete propagates back to master instead of leaving a
detached ghost. ClientService.Delete tombstones the email first so a
snap arriving mid-delete can't re-create the record.
WebSocket broadcasts an "invalidate(clients)" message on every client
mutation so the Clients page refreshes without manual reload.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(balancers): allow fallback on all strategies + feed burstObservatory from random/roundRobin
Drops the random/roundRobin gate on the Fallback field in
BalancerFormModal so every strategy can pick a fallback outbound.
syncObservatories now feeds burstObservatory from leastLoad +
random + roundRobin balancers (was leastLoad only), matching how
leastPing feeds observatory.
Fix the JsonEditor "Unexpected end of JSON input" that appeared
when switching a balancer between leastPing and another strategy:
the obsView watcher was gated on showObsEditor (a boolean OR of
the two flags) and missed the case where one observatory
swapped for the other in the same tick. Watch the individual
flags instead so obsView flips to the surviving editor and the
getter stops pointing at a deleted key.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): use sortedInbounds for mobile empty-state check
InboundList referenced an undefined visibleInbounds in the mobile
card list's empty-state guard, throwing "Cannot read properties of
undefined (reading 'length')" and breaking the entire mobile render.
* feat(clients): sortable table columns
Adds the same sortState / sortableCol / sortFns pattern InboundList
uses, wrapping filteredClients in sortedClients so sort composes with
the existing search/filter pipeline. Sortable: enable, email,
inboundIds (attachment count), traffic, remaining, expiryTime;
actions and online stay unsorted.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(shadowsocks): generate valid ss2022 keys and per-client method for legacy ciphers
The Add Client flow on shadowsocks inbounds was producing xray configs
that failed to start:
- 2022-blake3-* ciphers need a base64-encoded key of an exact byte
length per cipher. fillProtocolDefaults was assigning a uuid-style
string, which xray rejects as "bad key". Now the password is
generated (or replaced if invalid) via random.Base64Bytes(n) sized
to the chosen cipher.
- Legacy ciphers (aes-256-gcm, chacha20-*, xchacha20-*) require a
per-client method field in multi-user mode; model.Client has no
Method, so settings.clients was stored without one and xray failed
with "unsupported cipher method:". applyShadowsocksClientMethod
now injects the top-level method into each client on add/update,
and healShadowsocksClientMethods backfills it at xray-config-build
time so existing inbounds heal on the next start.
- xray/api.go ssCipherType switch was missing aes-256-gcm, which
fell through to ss2022 path.
- SSMethods dropdown now offers aes-256-gcm.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): preserve ClientRecord on inbound delete + filter Attached inbounds to multi-client protocols
Replace the global orphan sweep in setRemoteTrafficLocked with a
per-inbound diff cleanup: only delete a ClientRecord whose email
disappeared from a snap-tracked inbound (i.e. a node-side delete).
Inbounds that vanished entirely from the snap (e.g. admin deleted
the inbound on master) aren't iterated, so a client whose last
attachment came from that inbound is now left alone instead of
being deleted alongside the inbound.
ClientFormModal and ClientBulkAddModal now filter the Attached
inbounds dropdown to protocols that actually support multiple
clients: shadowsocks, vless, vmess, trojan, hysteria, hysteria2,
and portfallback (which routes through VLESS settings).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): make empty-state text readable on dark/ultra themes
The "No clients yet" empty state had a hardcoded black color
(rgba(0,0,0,0.45)) that vanished against the dark backgrounds.
Drop the inline color, let it inherit from the AntD theme, and
fade with opacity like the mobile card empty state already does.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(clients): client-first tgbot add flow, tgId field, lightweight inbound options
- tgbot: drop legacy per-protocol Add Client UI in favour of a client-first
multi-inbound flow. New BuildClientDraftMessage / getInboundsAttachPicker
let an admin pick one or more inbounds and submit a single client; per-
protocol secrets are now generated server-side via fillProtocolDefaults.
Drops awaiting_id/awaiting_password_tr/awaiting_password_sh state cases
and add_client_ch_default_id/pass_tr/pass_sh/flow callbacks. Adds a
setTGUser button + awaiting_tg_id state so the bot can set Client.TgID
during Add.
- clients UI: add Telegram user ID input to ClientFormModal (0 = none).
Hide IP Limit field entirely when ipLimitEnable is off — disabled fields
still take layout space, this collapses Auth(Hysteria) to full width.
- inbounds API: new GET /panel/api/inbounds/options that returns just
{id, remark, protocol, port, tlsFlowCapable}. Used by the clients page
pickers so the dropdown payload stays small on panels with thousands of
clients (drops settings JSON, clientStats, streamSettings). Server-side
TlsFlowCapable mirrors Inbound.canEnableTlsFlow so the modal no longer
needs to parse streamSettings client-side.
- clientInfoMsg now shows attached inbound remarks, and getInboundUsages
reports the attached client count per inbound.
- api-docs: document the new /options endpoint and add tgId / flow to the
clients add/update bodies.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(inbounds): keep Node column visible for node-attached inbounds
The Node column was bound to hasActiveNode, so disabling every node hid
the column even when inbounds were still attached to those nodes — the
admin lost the visual cue that those inbounds belonged to a node and
would come back when it was re-enabled. Combine hasActiveNode with a
new hasNodeAttachedInbound check (any dbInbound with nodeId != null) so
the column survives node-disable.
* fix(api-docs): accept functional-component icons in EndpointSection
AntD-Vue icons (SafetyCertificateOutlined, etc.) are functional
components, so the icon prop's type: Object validator was rejecting
them with a "Expected Object, got Function" warning at runtime.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* test: cover crypto, random, netsafe, sub helpers, xray equals, websocket hub, node service
Adds ~110 unit tests across previously untested packages. Focus on
pure-logic and concurrency surfaces where regressions would silently
affect users:
- util/crypto, util/random: password hashing round-trip, ss2022 key
generation, alphabet/length invariants.
- util/netsafe: IsBlockedIP edge cases, NormalizeHost validation,
SSRF guard with AllowPrivate context bypass.
- util/common, util/json_util: traffic formatter, Combine nil-skip,
RawMessage empty-as-null and copy-on-unmarshal.
- sub: splitLinkLines, searchKey/searchHost, kcp share fields,
finalmask normalization, buildVmessLink round-trip.
- xray: Config.Equals and InboundConfig.Equals field-by-field,
getRequiredUserString/getOptionalUserString type checks.
- web/websocket: hub registration, throttling, slow-client eviction,
nil-receiver safety, concurrent register/unregister.
- web/service: NodeService.normalize validation, normalizeBasePath,
HeartbeatPatch.ToUI mapping.
- web/job: atomicBool concurrent set/takeAndReset semantics.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n(clients): replace English fallbacks with proper translation keys
Pulls every hard-coded English label/title in the Clients page and its
four modals through the i18n layer so localized panels stop leaking
English. New keys live under pages.clients (auth, hysteriaAuth, uuid,
flow, flowNone, reverseTag, reverseTagPlaceholder, telegramId,
telegramIdPlaceholder, created, updated, ipLimit) plus refresh at the
root and toasts.bulkDeletedMixed / bulkCreatedMixed for partial-failure
toasts. Also switches the add-client modal's primary button from "Add"
to "Create" for consistency with other create flows.
The bulk-add Random/Random+Prefix/... email-method options stay
hard-coded by request - they're identifier-shaped strings.
* i18n: backfill 99 missing keys across all 12 non-English locales
Brings every translation file up to parity with en-US.json so the
Clients page, the fallback-children inbound section, the new refresh
verb, the Nodes panel-version label and a handful of older holes stop
falling through to the English fallback. New strings span:
- pages.clients.* (labels, confirmations, toasts, emailMethods)
- pages.inbounds.portFallback.* (Reality fallback inbound section)
- pages.nodes.panelVersion, menu.clients, refresh
Technical identifiers (Auth, UUID, Flow, Reverse tag) are intentionally
left untranslated since they correspond to xray-core field names.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* i18n: drop stale pages.client block duplicated in every non-English locale
Every non-English locale carried a pages.client (singular) section with
30 entries that duplicated pages.clients (plural). The plural namespace
is what the Vue code actually consumes; the singular one was dead
weight from an older rename that never got cleaned up in the
non-English files. Removing it brings every locale to exactly 984
keys, matching en-US.json.
* chore: apply modernize analyzer fixes across codebase
Mechanical replacements suggested by golang.org/x/tools/.../modernize:
strings.Cut/CutPrefix/SplitSeq, slices.Contains, maps.Copy, min(),
range-over-int, new(expr), strings.Builder for hot += loops,
reflect.TypeFor[T](), sync.WaitGroup.Go(), drop legacy //+build lines.
* feat(database): add PostgreSQL as an optional backend alongside SQLite
Lets operators with large client counts or multi-node setups pick PostgreSQL
at install time without breaking the existing SQLite default. Backend is
selected at runtime via XUI_DB_TYPE/XUI_DB_DSN, a small dialect layer keeps
the five JSON_EXTRACT/JSON_EACH queries portable, and a new `x-ui migrate-db`
subcommand copies SQLite data into PostgreSQL in FK-aware order.
* fix(inbounds): gate node selector to multi-node-capable protocols
Hide the Deploy-To selector and clear nodeId when switching to a
protocol that can't run on a remote node. Also:
- subs: return 404 (not 400) when subId matches no inbounds, so VPN
clients distinguish "deleted/unknown" from a server error
- hysteria link gen: use the inbound's resolved address so node-managed
inbounds advertise the node host instead of the central panel
- shadowsocks: default network to 'tcp' (udp was causing issues for some
clients on first-create)
- vite dev proxy: rewrite migrated-route bypass against the live base
path instead of a hardcoded single-segment regex
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(clients): bulk add/delete correctness + perf, working pagination, delayed-start in form
Bulk add/delete were serial on the frontend (one toast per call, N round-trips)
and the backend race exposed by parallelizing them lost client attachments and
hit UNIQUE constraint failed on client_inbounds. The single add/edit modal also
had no Start-After-First-Use option, and the table never showed the delayed
duration.
Backend (web/service/client.go):
- Per-inbound mutex on Add/Update/Del InboundClient so concurrent writers on
the same inbound don't lose the read-modify-write of settings JSON.
- SyncInbound skips create+join when the email is tombstoned so a concurrent
maintenance pass (adjustTraffics, autoRenewClients, markClientsDisabledIn-
Settings) that did a stale RMW can't resurrect a just-deleted client with a
fresh id.
- compactOrphans sweeps settings.clients entries whose ClientRecord no longer
exists, applied in Add/DelInboundClient + DelInboundClientByEmail so each
user-initiated mutation self-heals the inbound's settings.
- DelInboundClient uses Pluck instead of First for the stats lookup so a
missing row doesn't abort the delete with a noisy ErrRecordNotFound log.
Frontend:
- HttpUtil.{get,post} accept a silent option that suppresses the auto-toast.
- ClientBulkAddModal fires creates in parallel + silent + one summary toast.
- useClients.removeMany runs deletes in parallel + silent and refreshes once;
ClientsPage bulk delete uses it and shows one aggregate toast.
- useClients.applyInvalidate debounces 200 ms so the burst of N WebSocket
invalidate events from the backend collapses into a single refresh.
- ClientsPage pagination is reactive (paginationState ref + tablePagination
computed); onTableChange persists page-size and page changes.
- ClientFormModal gains a Start-After-First-Use switch + Duration days input
alongside the existing Expiry Date picker; on edit-mode open a negative
expiryTime is decoded back to delayed mode + days; on submit the payload
sends -86400000 * days or the absolute timestamp.
- ClientsPage table shows the delayed-start duration (blue tag Nd, tooltip
Start After First Use: Nd) instead of infinity.
- Telegram ID field in the form is hidden when /panel/setting/defaultSettings
reports tgBotEnable=false; Comment then fills the row.
- Form row 3 collapses UUID (span 12) + Total GB (span 8) + Limit IP (span 4)
when ipLimitEnable is on, else UUID + Total GB at 12/12.
- useInbounds.rollupClients counts only clients with a matching clientStats
row, so orphans in settings.clients no longer inflate the inbound's count.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(windows): clean shutdown, working panel restart, harden kernel32 load
Three Windows-specific issues addressed:
1. Orphaned xray-windows-amd64 after VS Code debugger stop. Delve's
"Stop" sends TerminateProcess to the Go binary, which is uncatchable
— our signal handlers never run, so xrayService.StopXray() is skipped
and xray is left dangling. Spawn xray as a child of a Job Object with
JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE so the OS kills xray when our
handle to the job is closed (which happens even on TerminateProcess).
Also trap os.Interrupt in main so Ctrl+C in the terminal runs the
graceful path.
2. /panel/setting/restartPanel logged "failed to send SIGHUP signal: not
supported by windows" because Windows can't deliver arbitrary signals.
Add a restart hook in web/global; main registers it to push SIGHUP
into its own signal channel, and RestartPanel calls the hook before
falling back to the (Unix-only) signal path. Same restart-loop code
runs in both cases.
3. util/sys/sys_windows.go now uses windows.NewLazySystemDLL so the
kernel32.dll resolve is pinned to %SystemRoot%\System32 (prevents
DLL hijacking by a planted DLL next to the binary). Local filetime
type replaced with windows.Filetime, and the unreliable
syscall.GetLastError() fallback replaced with a type assertion on the
errno captured at call time.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(sys): correct CPU/connection accounting on linux + darwin
util/sys/sys_linux.go:
- GetTCPCount/GetUDPCount were counting the column header row in
/proc/net/{tcp,udp}[6] as a connection, inflating the reported total
by 1 per non-empty file (so the panel status line always showed 2
more connections than actually existed). Replace getLinesNum +
safeGetLinesNum with a single bufio.Scanner-based countConnections
that skips the header.
- CPUPercentRaw now opens HostProc("stat") instead of a hardcoded
/proc/stat so HOST_PROC overrides apply, matching the connection
counters in the same file.
- Simplify CPU field unpacking: pad nums to 8 once instead of guarding
every assignment with a len check.
util/sys/sys_darwin.go:
- Fix swapped idle/intr indices on kern.cp_time. BSD CPUSTATES order
is user, nice, sys, intr, idle (CP_INTR=3, CP_IDLE=4) — gopsutil's
cpu_darwin_nocgo.go reads the same layout. The previous code used
out[3] as idle and out[4] as intr, so busy = total - dIdle was
actually subtracting interrupt time, making the panel report CPU
usage close to 100% on macOS regardless of actual load.
- Collapse the per-field delta math into a single loop.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(xray): rotate crash reports into log folder, prevent overwrites
writeCrashReport had two flaws: it wrote to the bin folder (alongside the
xray binary) which conflates artifacts, and the second-precision timestamp
meant a tight restart-loop crash burst overwrote prior reports. Write to
the log folder with nanosecond precision and keep the last 10 reports.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* revert(inbounds): drop unreleased portfallback protocol
The Port-with-Fallback inbound (commit 62fd9f9d) was confusing as a
standalone protocol — fallbacks belong on a regular VLESS/Trojan TCP-TLS
inbound, the way Xray models them natively. Rip out the entire feature
cleanly (no migration needed since it was never released): protocol
constant, fallback children DB table, FallbackService, 2 API endpoints,
all UI rows, related translations and api-docs. A native fallback flow
attached to VLESS/Trojan TCP-TLS/Reality will land in a follow-up commit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(inbounds): native fallbacks on VLESS/Trojan TCP-TLS, with working child links
A VLESS or Trojan inbound on TCP with TLS or Reality can now act as a
fallback master: pick existing inbounds as children and the panel auto-
fills the SNI / ALPN / path / xver routing fields from each child's
transport, auto-builds settings.fallbacks at config-gen time, and
rewrites the child's client-share link so it advertises the master's
reachable endpoint and TLS state instead of the child's loopback listen.
Layout matches the Xray All-in-One Nginx example: master at :443 with
clients + TLS, each child on 127.0.0.1 with its own transport+clients.
Order matters (Xray walks fallbacks top-to-bottom) — reorder via the
per-row up/down arrows. Path / SNI / ALPN are exposed under a per-row
Edit toggle for the rare cases where the auto-derivation needs
overriding; otherwise just pick a child and you're done.
Backend: new InboundFallback table + FallbackService (GetByMaster /
SetByMaster / GetParentForChild / BuildFallbacksJSON); two routes
(GET / POST /panel/api/inbounds/:id/fallbacks); xray.GetXrayConfig
injects settings.fallbacks for any VLESS/Trojan TCP-TLS/Reality
inbound; GetInbounds annotates each child with FallbackParent so the
frontend can rewrite links without an extra round-trip.
Link projection covers every emission path — clients-page QR/links,
per-inbound Get URL, raw subscription, sub-JSON, sub-Clash, and the
inbounds-page link/info/QR — via a shared projectThroughFallbackMaster
on the backend and a shared projectChildThroughMaster on the frontend
that both handle the panel-tracked relationship and the legacy
unix-socket (@vless-ws) convention.
Strings translated into all 12 non-English locales.
* docs: rewrite CONTRIBUTING with full local-dev setup
The prior three-line CONTRIBUTING left newcomers guessing at every
non-trivial step: which Go / Node versions, where xray comes from, why
the panel goes blank when XUI_DEBUG=true is flipped on, how the Vue
multi-page setup is wired, what to do on Windows when go build trips
on the CGo SQLite driver.
Now covers prerequisites, MinGW-w64 install on Windows (niXman builds
or MSYS2), one-shot first-time setup, two frontend dev workflows with
the XUI_DEBUG asset-cache gotcha called out, the architecture and
conventions of the Vue side, a project-layout map, useful env vars,
and the PR checklist.
---------
2026-05-19 03:16:42 -07:00
else
2026-05-19 04:09:35 -07:00
# If a cert already exists, just show the access URL
echo -e " ${ green } Access URL: https:// ${ server_ip } : ${ existing_port } / ${ config_webBasePath } ${ plain } "
2025-09-25 05:34:12 -07:00
fi
2023-02-09 11:18:06 -08:00
fi
else
2026-05-19 04:09:35 -07:00
if [ [ " $existing_hasDefaultCredential " = = "true" ] ] ; then
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
local config_username = " ${ XUI_USERNAME :- $( gen_random_string 10) } "
local config_password = " ${ XUI_PASSWORD :- $( gen_random_string 10) } "
2026-05-19 04:09:35 -07:00
echo -e " ${ yellow } Default credentials detected. Security update required... ${ plain } "
${ xui_folder } /x-ui setting -username " ${ config_username } " -password " ${ config_password } "
echo -e "Generated new random login credentials:"
echo -e "###############################################"
echo -e " ${ green } Username: ${ config_username } ${ plain } "
echo -e " ${ green } Password: ${ config_password } ${ plain } "
echo -e "###############################################"
feat: release-driven golden-image & unattended-install deployment pipeline (#5323)
* feat(install): add non-interactive install path for cloud/golden-image use
Trigger non-interactive mode when XUI_NONINTERACTIVE=1 or stdin is not a
TTY (curl | bash, cloud-init). Every prompt is then replaced by an env var
or a sane default; interactive prompts stay byte-for-byte identical.
Honored env vars: XUI_USERNAME, XUI_PASSWORD, XUI_PANEL_PORT,
XUI_WEB_BASE_PATH (unset => random, as before), XUI_SSL_MODE=none|ip|domain
(default none), XUI_DOMAIN, XUI_ACME_EMAIL, XUI_DB_TYPE/XUI_DB_DSN, plus
additive XUI_ACME_HTTP_PORT, XUI_SSL_IPV6, XUI_SERVER_IP.
On success, write /etc/x-ui/install-result.env (mode 600) with the panel
creds + access URL + api token, in both interactive and non-interactive
modes, so cloud-init/MOTD can surface them. Postgres in non-interactive
mode requires XUI_DB_DSN or installs locally; never silently downgrades.
* feat(deploy): add first-boot per-instance credential generation
Golden images ship with no x-ui.db. x-ui-firstboot.sh runs once (guarded by
/etc/x-ui/.firstboot-done), before x-ui.service, and replaces the seeded
admin/admin with fresh random username/password on a random high port,
regenerates the session secret/panel GUID via 'x-ui setting -reset', mints an
API token, and writes the creds to /etc/x-ui/credentials.txt (600) + /etc/motd.
Idempotent: skips regeneration if a non-default admin already exists. The
oneshot unit is ordered After=network-online/cloud-init and Before=x-ui.service
so the panel never serves default credentials.
* chore(deploy): force LF for cloud-image deploy assets (.service/.hcl/.yaml)
* feat(deploy): add Packer config + provisioning scripts for golden image
One build, two sources: amazon-ebs (AWS AMI, Canonical Ubuntu 24.04 base via
source_ami_filter) and qemu (qcow2 + raw, NoCloud-seeded for build-time SSH).
Provisioner order is fixed: provision.sh -> harden.sh -> cleanup.sh.
- provision.sh: downloads the released x-ui tarball (no Go build), installs the
panel + firstboot unit, enables but does NOT start services, creates NO DB.
- harden.sh: key-only SSH, no root password login, locks default account
passwords, enables unattended-upgrades (scanner-compliant).
- cleanup.sh: wipes any DB/creds, SSH host keys, authorized_keys, machine-id,
cloud-init state, logs and history; fails the build if any secret survives.
packer fmt -check clean; packer validate passes for both sources.
* feat(deploy): add generic cloud-init user-data for unattended install
cloud-init.yaml installs the latest 3x-ui non-interactively (XUI_NONINTERACTIVE=1)
on any cloud-init platform, generating unique per-instance credentials and
surfacing them via /etc/x-ui/install-result.env, serial console and MOTD.
README documents per-provider usage (Hetzner/AWS/DO/Vultr/GCP/Azure/Oracle)
and all XUI_* knobs.
* ci: add image.yml to build cloud images on release
On release: published (or workflow_dispatch with a tag), waits for the
x-ui-linux-amd64.tar.gz asset (handles the release-matrix upload race), then:
- qemu-image (always): builds the qcow2 with Packer and attaches a compressed
.qcow2.xz + sha256 to the GitHub release. Uses KVM when /dev/kvm exists,
else TCG.
- ami-image (gated): builds the AWS AMI only when AWS creds exist (OIDC role
preferred, else access keys), so forks skip cleanly. Prints the AMI ID to the
job summary. No secrets or AMI IDs are committed.
* test(deploy): add container smoke tests for install + firstboot
smoke-noninteractive.sh: runs install.sh piped (no TTY) with
XUI_NONINTERACTIVE=1 in an Ubuntu container; asserts install-result.env (600)
holds random non-default creds, hasDefaultCredential is false, and the panel
serves HTTP.
smoke-firstboot.sh: installs the released binary with no DB, runs
x-ui-firstboot.sh; asserts per-instance creds + credentials.txt (600) + MOTD,
no admin/admin, and that a second run is a no-op (sentinel honored).
smoke.yml runs both as gated jobs on PRs/pushes touching install.sh or deploy/**.
Both pass locally against the v3.3.1 release binary.
* docs(deploy): add Packer/marketplace docs and link from README
- deploy/README.md: index of the cloud-deploy tooling and the two models
- deploy/packer/README.md: how to build locally, variables, first-boot behavior
- deploy/marketplace/aws/README.md: seller registration -> AMI scan ->
limited-visibility preview -> go-public checklist
- deploy/marketplace/hetzner/README.md: cloud-init-first guidance + snapshot
caveat (delete x-ui.db first) + hetznercloud/apps reference
- README.md: link the unattended-install / cloud-image docs from Quick Start
* feat(deploy): build golden images for arm64 as well as amd64
The install path was already multi-arch (install.sh auto-detects arch); this
extends the golden image + CI to arm64:
- packer: xui_arch (amd64|arm64, validated) now derives the base AMI filter and
the Ubuntu cloud image; the qemu source switches to qemu-system-aarch64 + virt
machine + AAVMF UEFI firmware for arm64. amd64 path unchanged.
- image.yml: arch matrix. AMIs for amd64 (t3.small) + arm64 (t4g.small/Graviton)
from one runner; qcow2 for amd64 on a standard runner and arm64 on a native
ubuntu-24.04-arm runner. Waits for both release tarballs.
- smoke.yml: run install + firstboot smoke tests on amd64 and arm64 runners;
smoke-firstboot.sh now resolves the arch tarball via dpkg.
- docs updated for both arches.
packer fmt/validate pass for amd64 and arm64; actionlint + shellcheck clean.
Verified locally: non-interactive install AND firstboot run on the real arm64
release binary under emulation (ELF aarch64, no admin/admin).
* chore(deploy): default AWS region to eu-central-1 (Frankfurt)
Replace the us-east-1 fallback in image.yml (4 sites) and the Packer 'region'
default + doc examples. Still overridable via the AWS_REGION repo variable / the
-var 'region=...' flag.
* feat(deploy): add Amazon Lightsail support (launch script + snapshot builder)
Lightsail can't launch from an EC2 AMI and its blueprint list isn't
self-publishable, so add the two self-service paths instead:
- launch-script.sh: paste into Lightsail 'Add launch script' (or --user-data) to
install 3x-ui non-interactively with unique per-instance credentials.
- snapshot-userdata.sh + build-snapshot.sh: AWS CLI pipeline that provisions a
build instance (panel installed, NO DB, firstboot enabled), runs the shared
cleanup.sh, then snapshots it. Instances launched from the snapshot mint their
own credentials on first boot. Optional --panel-port pins a known port for the
Lightsail firewall.
- README documents both paths, the firewall caveat, and the blueprint reality.
EC2 AMI / Marketplace path kept untouched alongside. All scripts shellcheck-clean.
* fix(deploy): address Copilot PR review findings
- install.sh + firstboot: write install-result.env / credentials.txt values with
printf %q so the files stay safe to source even if creds are pinned with shell
metacharacters (no-op for the alphanumeric random defaults).
- firstboot: fail closed if 'x-ui setting -show' can't be parsed to true/false —
exit without writing the sentinel so the next boot retries, instead of silently
skipping regeneration and risking admin/admin.
- firstboot + cloud-init + lightsail launch-script: keep secrets out of the
world-readable /etc/motd (show URL + username only; full creds via the mode-600
file / serial console).
- lightsail build-snapshot: handle download-default-key-pair returning either a
PEM or base64, and assert a valid PEM before using it for SSH.
- image.yml: pin hashicorp/setup-packer@v3 (was @main).
- deploy/README: document XUI_ACME_HTTP_PORT / XUI_SSL_IPV6 / XUI_SERVER_IP.
Both container smoke tests still pass; shellcheck + actionlint clean.
2026-06-14 09:08:35 -07:00
# Persist a machine-parseable credentials file for cloud-init / MOTD.
local config_apiToken
config_apiToken = $( ${ xui_folder } /x-ui setting -getApiToken true | grep -Eo 'apiToken: .+' | awk '{print $2}' )
: " ${ SSL_SCHEME : =https } "
: " ${ SSL_HOST : = ${ server_ip } } "
write_install_result " ${ config_username } " " ${ config_password } " " ${ existing_port } " \
" ${ existing_webBasePath } " " ${ SSL_SCHEME } " " ${ SSL_HOST } " " ${ config_apiToken } " " ${ XUI_DB_TYPE :- sqlite } "
2026-05-19 04:09:35 -07:00
else
echo -e " ${ green } Username, Password, and WebBasePath are properly set. ${ plain } "
fi
# Existing install: if no cert configured, prompt user for SSL setup
# Properly detect empty cert by checking if cert: line exists and has content after it
existing_cert = $( ${ xui_folder } /x-ui setting -getCert true | grep 'cert:' | awk -F': ' '{print $2}' | tr -d '[:space:]' )
if [ [ -z " $existing_cert " ] ] ; then
echo ""
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ green } SSL Certificate Setup (RECOMMENDED) ${ plain } "
echo -e " ${ green } ═══════════════════════════════════════════ ${ plain } "
echo -e " ${ yellow } Let's Encrypt now supports both domains and IP addresses! ${ plain } "
echo ""
prompt_and_setup_ssl " ${ existing_port } " " ${ existing_webBasePath } " " ${ server_ip } "
echo -e " ${ green } Access URL: ${ SSL_SCHEME } :// ${ SSL_HOST } : ${ existing_port } / ${ existing_webBasePath } ${ plain } "
else
echo -e " ${ green } SSL certificate already configured. No action needed. ${ plain } "
fi
2025-09-25 05:34:12 -07:00
fi
2026-05-04 04:20:24 -07:00
2026-05-19 04:09:35 -07:00
${ xui_folder } /x-ui migrate
}
2026-06-22 14:49:09 -07:00
# setup_fail2ban auto-installs and configures fail2ban for the IP Limit feature
# by invoking the freshly installed x-ui CLI. IP Limit is load-bearing on
# fail2ban (without it the panel disables the limitIp field and zeroes existing
# limits), so a fresh install should make it work out of the box, just like the
# Docker image already does. Non-fatal by design: a fail2ban failure must never
# abort the panel install.
setup_fail2ban( ) {
if [ [ -n " ${ XUI_ENABLE_FAIL2BAN +x } " && " ${ XUI_ENABLE_FAIL2BAN } " != "true" ] ] ; then
echo -e " ${ yellow } XUI_ENABLE_FAIL2BAN= ${ XUI_ENABLE_FAIL2BAN } , skipping Fail2ban auto-setup. ${ plain } "
return 0
fi
if [ [ ! -x /usr/bin/x-ui ] ] ; then
echo -e " ${ yellow } x-ui CLI not found; skipping Fail2ban auto-setup. ${ plain } "
return 0
fi
echo -e " ${ green } Setting up Fail2ban for the IP Limit feature... ${ plain } "
if /usr/bin/x-ui setup-fail2ban; then
echo -e " ${ green } Fail2ban setup complete. ${ plain } "
else
echo -e " ${ yellow } Fail2ban setup did not finish; IP Limit stays disabled until you run 'x-ui' and open the IP Limit menu. Continuing. ${ plain } "
fi
return 0
}
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
# Lands a systemd unit file at ${xui_service}/x-ui.service via a temp file +
# atomic mv, so a failed cp/curl or an interrupted mv never leaves a
# truncated unit file at the live path -- systemd would then fail to parse
# it on the next daemon-reload/start. Same pattern already used for
# /usr/bin/x-ui elsewhere in this script. source_is_url picks cp (from a
# file already extracted from the release tarball) vs curl (GitHub fallback).
_install_xui_service_unit( ) {
local source = " $1 "
local source_is_url = " $2 "
local dest = " ${ xui_service } /x-ui.service "
local temp_file = " ${ dest } .tmp. $$ "
rm -f " $temp_file "
if [ [ " $source_is_url " = = "true" ] ] ; then
curl -fLRo " $temp_file " " $source " > /dev/null 2>& 1
else
cp -f " $source " " $temp_file " > /dev/null 2>& 1
fi
if [ [ $? -ne 0 ] ] ; then
rm -f " $temp_file "
return 1
fi
if [ [ ! -s " $temp_file " ] ] ; then
rm -f " $temp_file "
return 1
fi
mv -f " $temp_file " " $dest "
if [ [ $? -ne 0 ] ] ; then
rm -f " $temp_file "
return 1
fi
return 0
}
2026-05-19 04:09:35 -07:00
install_x-ui( ) {
cd ${ xui_folder %/x-ui } /
# Download resources
if [ $# = = 0 ] ; then
2026-06-14 12:17:59 -07:00
tag_version = $( curl -Ls --retry 5 --retry-delay 3 --connect-timeout 15 --max-time 60 "https://api.github.com/repos/MHSanaei/3x-ui/releases/latest" | grep '"tag_name":' | sed -E 's/.*"([^"]+)".*/\1/' )
2026-05-19 04:09:35 -07:00
if [ [ ! -n " $tag_version " ] ] ; then
2026-06-22 12:52:38 -07:00
echo -e " ${ red } Failed to fetch x-ui version, it may be due to GitHub API restrictions, please try it later ${ plain } "
exit 1
2026-05-19 04:09:35 -07:00
fi
echo -e " Got x-ui latest version: ${ tag_version } , beginning the installation... "
2026-07-14 03:44:55 -07:00
curl -fLR --retry 5 --retry-delay 3 --connect-timeout 15 --speed-limit 1 --speed-time 300 -o ${ xui_folder } -linux-$( arch) .tar.gz https://github.com/MHSanaei/3x-ui/releases/download/${ tag_version } /x-ui-linux-$( arch) .tar.gz
2026-05-19 04:09:35 -07:00
if [ [ $? -ne 0 ] ] ; then
echo -e " ${ red } Downloading x-ui failed, please be sure that your server can access GitHub ${ plain } "
exit 1
fi
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if [ [ ! -s ${ xui_folder } -linux-$( arch) .tar.gz ] ] ; then
rm ${ xui_folder } -linux-$( arch) .tar.gz -f
echo -e " ${ red } Downloaded x-ui release archive is empty ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
else
tag_version = $1
2026-06-24 17:36:30 -07:00
# The rolling dev channel ships under a fixed, non-semver tag that is
# force-moved to the latest main commit on every push. Accept `dev` as a
# convenient alias and skip the numeric floor check for it.
if [ [ " $tag_version " = = "dev" || " $tag_version " = = "dev-latest" ] ] ; then
tag_version = "dev-latest"
echo -e " ${ yellow } Installing the rolling dev build (tag: dev-latest). This is a per-commit pre-release, not a stable version. ${ plain } "
else
tag_version_numeric = ${ tag_version #v }
min_version = "2.3.5"
2026-05-19 04:09:35 -07:00
2026-06-24 17:36:30 -07:00
if [ [ " $( printf '%s\n' " $min_version " " $tag_version_numeric " | sort -V | head -n1) " != " $min_version " ] ] ; then
echo -e " ${ red } Please use a newer version (at least v2.3.5). Exiting installation. ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
fi
url = " https://github.com/MHSanaei/3x-ui/releases/download/ ${ tag_version } /x-ui-linux- $( arch) .tar.gz "
2026-06-24 17:36:30 -07:00
echo -e " Beginning to install x-ui ${ tag_version } "
2026-07-14 03:44:55 -07:00
curl -fLR --retry 5 --retry-delay 3 --connect-timeout 15 --speed-limit 1 --speed-time 300 -o ${ xui_folder } -linux-$( arch) .tar.gz ${ url }
2026-05-19 04:09:35 -07:00
if [ [ $? -ne 0 ] ] ; then
2026-06-24 17:36:30 -07:00
echo -e " ${ red } Download x-ui ${ tag_version } failed, please check if the version exists ${ plain } "
2026-05-19 04:09:35 -07:00
exit 1
fi
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if [ [ ! -s ${ xui_folder } -linux-$( arch) .tar.gz ] ] ; then
rm ${ xui_folder } -linux-$( arch) .tar.gz -f
echo -e " ${ red } Downloaded x-ui release archive is empty ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
fi
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
local xui_script_temp = " /usr/bin/x-ui-temp. $$ "
rm -f " ${ xui_script_temp } "
curl -fLRo " ${ xui_script_temp } " https://raw.githubusercontent.com/MHSanaei/3x-ui/main/x-ui.sh
2026-05-19 04:09:35 -07:00
if [ [ $? -ne 0 ] ] ; then
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
rm -f " ${ xui_script_temp } "
2026-05-19 04:09:35 -07:00
echo -e " ${ red } Failed to download x-ui.sh ${ plain } "
exit 1
fi
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if [ [ ! -s " ${ xui_script_temp } " ] ] ; then
rm -f " ${ xui_script_temp } "
echo -e " ${ red } Downloaded x-ui.sh is empty ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
# Stop x-ui service and remove old resources
fix(install): preserve custom bin/ files (e.g. hand-added geoip) across updates (#6152)
* fix(install): preserve custom bin/ files (e.g. hand-added geoip) across updates
Every reinstall/update wipes /usr/local/x-ui/ wholesale and re-extracts
the release tarball, which only ships known assets (xray/mtg binaries,
the bundled geoip*/geosite*.dat sets). A user-reported real incident:
a hand-placed custom geoip file referenced from a routing rule via
"ext:<file>:<code>" got silently deleted on update, and Xray refused
to start at all afterward ("failed to open <file>: no such file or
directory"), taking down every inbound until the file was manually
restored from the user's own backup.
install_x-ui now backs up the old bin/ before the wipe and restores,
after extraction, only the files the fresh release doesn't provide --
bundled assets still get the newer per-release copy, nothing custom
silently disappears. Verified in isolation: standard files (geoip.dat,
the xray binary) end up as the fresh release's copy; a custom file
absent from the release survives untouched.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix: harden the bin/ snapshot-and-restore against the review round on #6152
- Replace the mktemp+cp snapshot with a same-filesystem mv of bin/ aside:
an unchecked mktemp failure previously made the very next line copy
bin/'s contents into "/" (empty custom_bin_backup + trailing slash),
and a silently-ignored cp failure (stderr redirected, exit code never
checked) could leave a truncated custom geo file that gets "restored"
as if it were intact. A rename is atomic and needs no extra disk space,
removing both failure modes at once; if it fails, back off cleanly and
say so instead of proceeding as if a backup exists.
- Add a trap so an interrupted update (Ctrl-C, signal) between the
backup and the restore doesn't leave the snapshot (which contains
bin/config.json and every mtproto client's FakeTLS secret) sitting
around indefinitely; the two exit-path cleanups this replaces are gone
since the trap now covers those exits too.
- Move the restore below the arm arch-rename/chmod block instead of
before it, so xray-linux-arm32/mtg-linux-arm already exist under their
final names and don't get needlessly restored-then-overwritten and
misreported as "custom".
- Exclude bin/config.json and bin/mtproto/*.toml from the restore: those
are the panel's own generated runtime state (internal/xray/process.go,
internal/mtproto/manager.go), not admin-placed files, and restoring a
stale one only resurrects dead state or recreates bin/mtproto/ with the
wrong (more permissive) directory mode.
- Match symlinks in the restore's find, not just plain files -- cp -a
already preserves them in the snapshot, but the restore loop was
silently dropping them, which is exactly the failure mode (a geo file
symlinked in from elsewhere) this PR set out to fix.
- Quote the two new xui_folder expansions.
- Extend the non-interactive smoke test to reinstall over an existing
install with a sentinel file in bin/, asserting it survives and that
the bundled geoip.dat is still the release's own copy -- the update
path this PR touches had no CI coverage at all before this.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-08-14 07:41:45 -07:00
local custom_bin_backup = ""
2026-05-19 04:09:35 -07:00
if [ [ -e ${ xui_folder } / ] ] ; then
if [ [ $release = = "alpine" ] ] ; then
rc-service x-ui stop
else
systemctl stop x-ui
fi
fix(mtproto): reap orphaned mtg, fix SysLog viewer, mtg log visibility, export remark (#5105) (#5107)
* fix(logs): render journalctl output in the SysLog viewer
The log viewer's parseLogLine only understood the app-log format
(2006/01/02 15:04:05 LEVEL - body). With SysLog ticked the backend
returns journalctl lines (Mon DD HH:MM:SS host ident[pid]: LEVEL - body),
so the parser mistook the journal time for the level and dropped the
body, leaving only timestamps. Detect and strip the journald prefix,
keep the journal timestamp as the stamp, then parse the real level and
body from the remainder.
* feat(mtproto): surface mtg output and add status reporting
mtg's stdout/stderr was captured by a writer that kept only the last
line and showed it nowhere, so the reason a proxy could not reach
Telegram was invisible. Stream mtg output line-by-line into the x-ui
log, tagged per inbound, so it appears in the panel log viewer and
journald.
Also fix mangled log lines: logger.Info uses fmt.Sprint, which drops
the space between adjacent string operands, producing output like
'inbound3on0.0.0.0:8443'. Switch the affected mtproto calls to the
formatted (*f) variants.
Add show_mtproto_status to x-ui.sh so 'x-ui status' reports each
mtproto inbound's mtg process state and bind address.
* fix(logs): parse all journalctl message shapes in SysLog viewer
Real journalctl output mixes four message shapes after the
'Mon DD HH:MM:SS host ident[pid]:' prefix: go-logging 'LEVEL - msg'
(x-ui/xray), Go std-log with an embedded date (net/http, runtime),
telego's '[timestamp] LEVEL msg', and systemd lines. The viewer only
understood the first, so std-log and telego lines — which never contain
' - ' — collapsed to a bare timestamp (e.g. the 8s telego 409 spam).
Extract the parser into a pure, testable module and teach it the other
shapes: strip the redundant Go std-log date, lift the level out of
telego brackets, and always keep the message body. Add a unit test
covering each shape with real captured lines.
* fix(mtproto): reap orphaned mtg sidecars so a stale one can't break new clients
On Linux x-ui does not kill its mtg children when it dies (no kill-on-exit,
unlike the Windows job object). After a crash, OOM, kill -9, or update, a
stale mtg keeps holding the inbound port with an OLD secret, so new clients
fail the FakeTLS handshake and get silently domain-fronted to the fakeTLS
domain instead of proxied to Telegram (a few MB of traffic, never connects).
Sweep orphans at startup: on the first reconcile, before x-ui starts any of
its own mtg, scan /proc and SIGKILL any process whose executable is our
mtg-<goos>-<goarch> binary. x-ui is the sole owner of mtg, so anything alive
then is an orphan. Runs once per process (swept guard), survives the
binary-deleted-during-update case via /proc/<pid>/cmdline, and is a no-op on
Windows (job object) and other platforms.
Also clear stray mtg in update.sh/install.sh after stopping x-ui, anchored to
the 'mtg-linux-<arch> run ' invocation so the pattern can't match unrelated
command lines (e.g. x-ui.sh's own 'grep mtg-linux').
* fix(logs): drop dead body initializer flagged by eslint no-useless-assignment
* fix(mtproto): drop remark fragment from tg://proxy export link
The mtproto export link appended the inbound remark as a URL fragment
(tg://proxy?server=...&port=...&secret=...#remark). Telegram Desktop
rejects a proxy deep link with a trailing fragment as 'This proxy link
is invalid', breaking one-click import, and a remark is meaningless for
proxy links across clients. Stop adding it in both the panel link
(genMtprotoLink) and the subscription service. Fixes #5105.
* fix(x-ui.sh): remove unused check_mtproto_status helper
show_mtproto_status does its own process check, so check_mtproto_status
was dead code. Drop it (per Copilot review on #5107).
2026-06-08 19:01:33 -07:00
# Kill any leftover mtg (MTProto) sidecars. x-ui runs them outside its own
# lifecycle, so on Linux a stale one can survive the stop and keep holding
# an inbound port with an outdated secret, silently breaking new clients.
# The freshly installed panel respawns a clean mtg per inbound on start.
pkill -f 'mtg-linux-[^ ]* run ' > /dev/null 2>& 1 || true
fix(install): preserve custom bin/ files (e.g. hand-added geoip) across updates (#6152)
* fix(install): preserve custom bin/ files (e.g. hand-added geoip) across updates
Every reinstall/update wipes /usr/local/x-ui/ wholesale and re-extracts
the release tarball, which only ships known assets (xray/mtg binaries,
the bundled geoip*/geosite*.dat sets). A user-reported real incident:
a hand-placed custom geoip file referenced from a routing rule via
"ext:<file>:<code>" got silently deleted on update, and Xray refused
to start at all afterward ("failed to open <file>: no such file or
directory"), taking down every inbound until the file was manually
restored from the user's own backup.
install_x-ui now backs up the old bin/ before the wipe and restores,
after extraction, only the files the fresh release doesn't provide --
bundled assets still get the newer per-release copy, nothing custom
silently disappears. Verified in isolation: standard files (geoip.dat,
the xray binary) end up as the fresh release's copy; a custom file
absent from the release survives untouched.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix: harden the bin/ snapshot-and-restore against the review round on #6152
- Replace the mktemp+cp snapshot with a same-filesystem mv of bin/ aside:
an unchecked mktemp failure previously made the very next line copy
bin/'s contents into "/" (empty custom_bin_backup + trailing slash),
and a silently-ignored cp failure (stderr redirected, exit code never
checked) could leave a truncated custom geo file that gets "restored"
as if it were intact. A rename is atomic and needs no extra disk space,
removing both failure modes at once; if it fails, back off cleanly and
say so instead of proceeding as if a backup exists.
- Add a trap so an interrupted update (Ctrl-C, signal) between the
backup and the restore doesn't leave the snapshot (which contains
bin/config.json and every mtproto client's FakeTLS secret) sitting
around indefinitely; the two exit-path cleanups this replaces are gone
since the trap now covers those exits too.
- Move the restore below the arm arch-rename/chmod block instead of
before it, so xray-linux-arm32/mtg-linux-arm already exist under their
final names and don't get needlessly restored-then-overwritten and
misreported as "custom".
- Exclude bin/config.json and bin/mtproto/*.toml from the restore: those
are the panel's own generated runtime state (internal/xray/process.go,
internal/mtproto/manager.go), not admin-placed files, and restoring a
stale one only resurrects dead state or recreates bin/mtproto/ with the
wrong (more permissive) directory mode.
- Match symlinks in the restore's find, not just plain files -- cp -a
already preserves them in the snapshot, but the restore loop was
silently dropping them, which is exactly the failure mode (a geo file
symlinked in from elsewhere) this PR set out to fix.
- Quote the two new xui_folder expansions.
- Extend the non-interactive smoke test to reinstall over an existing
install with a sentinel file in bin/, asserting it survives and that
the bundled geoip.dat is still the release's own copy -- the update
path this PR touches had no CI coverage at all before this.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-08-14 07:41:45 -07:00
# bin/ is about to be wiped wholesale by the tar extraction below. The
# release only ships known assets (xray/mtg binaries, the bundled
# geoip*/geosite*.dat sets) -- anything else in bin/ was placed there
# by the admin (e.g. a hand-added custom geoip/geosite file referenced
# from a routing rule via "ext:<file>:<code>") and would otherwise be
# silently deleted on every update, breaking Xray at next start with
# "failed to open <file>: no such file or directory" for any routing
# rule that references it. Moved aside rather than copied: a rename
# on the same filesystem is atomic (no truncated file if disk space
# runs out mid-copy, unlike `cp`) and keeps the snapshot under
# /usr/local rather than a separate, possibly small/tmpfs $TMPDIR.
if [ [ -d " ${ xui_folder } /bin " ] ] ; then
custom_bin_backup = " ${ xui_folder %/x-ui } /x-ui-bin-backup. $$ "
rm -rf " ${ custom_bin_backup } "
if ! mv " ${ xui_folder } /bin " " ${ custom_bin_backup } " ; then
custom_bin_backup = ""
echo -e " ${ yellow } Could not back up bin/ -- custom files there will not be preserved across this update ${ plain } "
fi
fi
# Sole cleanup path for the backup from here on -- covers both the
# two `exit 1`s below (extraction/binary-missing failures) and an
# interrupted update (Ctrl-C, signal) before the restore runs.
# Cleared once the restore below finishes normally.
trap '[[ -n "${custom_bin_backup}" ]] && rm -rf "${custom_bin_backup}"' EXIT INT TERM
2026-05-19 04:09:35 -07:00
rm ${ xui_folder } / -rf
fi
# Extract resources and set permissions
tar zxvf x-ui-linux-$( arch) .tar.gz
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if [ [ $? -ne 0 ] ] ; then
rm x-ui-linux-$( arch) .tar.gz -f
rm -f " ${ xui_script_temp } "
echo -e " ${ red } Failed to extract the x-ui release archive -- the previous installation has already been removed, so the panel will not start until this is fixed; try running the installer again ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
rm x-ui-linux-$( arch) .tar.gz -f
cd x-ui
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if [ [ $? -ne 0 || ! -s x-ui ] ] ; then
rm -f " ${ xui_script_temp } "
echo -e " ${ red } Extracted x-ui archive is missing the x-ui binary -- the previous installation has already been removed, so the panel will not start until this is fixed; try running the installer again ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
chmod +x x-ui
chmod +x x-ui.sh
2026-05-04 04:20:24 -07:00
2026-07-05 11:08:10 -07:00
# Check the system's architecture and rename the file accordingly.
# The panel binary maps GOARCH=arm to "arm32" (internal/xray/process.go),
# so the Xray binary must be named xray-linux-arm32; mtg keeps plain "arm".
2024-04-01 01:42:28 -07:00
if [ [ $( arch) = = "armv5" || $( arch) = = "armv6" || $( arch) = = "armv7" ] ] ; then
2026-07-05 11:08:10 -07:00
mv bin/xray-linux-$( arch) bin/xray-linux-arm32
chmod +x bin/xray-linux-arm32
feat(mtproto): add MTProto (FakeTLS) protocol via managed mtg sidecar (#5076)
* feat(mtproto): add MTProto (FakeTLS) protocol via managed mtg sidecar
Xray-core has no mtproto proxy, so mtproto inbounds run as standalone
mtg (9seconds/mtg) sidecar processes managed by the panel — one per
inbound — and are excluded from the generated Xray config entirely.
- model: MTProto protocol constant, validator, and FakeTLS secret
helpers (GenerateFakeTLSSecret/HealMtprotoSecret)
- mtproto package: per-inbound mtg process manager with reconcile,
graceful stop, and best-effort Prometheus traffic scraping
- runtime: delegate mtproto inbounds to the mtg manager instead of the
Xray gRPC API; skip mtproto when building the Xray config
- web: boot reconcile + StopAll wiring, periodic reconcile/traffic job,
port-conflict transport, secret healing on inbound add/update
- sub: tg:// proxy share-link generation
- frontend: protocol option, Zod schema, Protocol tab (FakeTLS domain +
regenerable secret), info-modal link, and i18n
- provisioning: fetch mtg v2.2.8 in install.sh, DockerInit.sh, and the
Linux + Windows release workflows
* fix
* fix
* fix: address Copilot review comments on mtproto PR
- web/web.go: create NewMtprotoJob once and reuse for cron + initial run
- mtproto/manager.go: StopAll cleans up per-inbound config files on shutdown
- mtproto/manager.go: CollectTraffic releases mutex before HTTP scrapes to
avoid blocking Ensure/Reconcile/Remove during network I/O
- database/model/model.go: panic on crypto/rand failure in mtprotoRandomMiddle
instead of silently producing a weak all-zero secret
- install.sh: fix chmod to handle renamed bin/mtg-linux-arm on armv5/v6/v7
2026-06-08 05:28:19 -07:00
if [ [ -f bin/mtg-linux-$( arch) ] ] ; then
mv bin/mtg-linux-$( arch) bin/mtg-linux-arm
chmod +x bin/mtg-linux-arm
fi
2024-01-15 16:01:15 -08:00
fi
2026-05-19 04:09:35 -07:00
chmod +x x-ui bin/xray-linux-$( arch)
feat(mtproto): add MTProto (FakeTLS) protocol via managed mtg sidecar (#5076)
* feat(mtproto): add MTProto (FakeTLS) protocol via managed mtg sidecar
Xray-core has no mtproto proxy, so mtproto inbounds run as standalone
mtg (9seconds/mtg) sidecar processes managed by the panel — one per
inbound — and are excluded from the generated Xray config entirely.
- model: MTProto protocol constant, validator, and FakeTLS secret
helpers (GenerateFakeTLSSecret/HealMtprotoSecret)
- mtproto package: per-inbound mtg process manager with reconcile,
graceful stop, and best-effort Prometheus traffic scraping
- runtime: delegate mtproto inbounds to the mtg manager instead of the
Xray gRPC API; skip mtproto when building the Xray config
- web: boot reconcile + StopAll wiring, periodic reconcile/traffic job,
port-conflict transport, secret healing on inbound add/update
- sub: tg:// proxy share-link generation
- frontend: protocol option, Zod schema, Protocol tab (FakeTLS domain +
regenerable secret), info-modal link, and i18n
- provisioning: fetch mtg v2.2.8 in install.sh, DockerInit.sh, and the
Linux + Windows release workflows
* fix
* fix
* fix: address Copilot review comments on mtproto PR
- web/web.go: create NewMtprotoJob once and reuse for cron + initial run
- mtproto/manager.go: StopAll cleans up per-inbound config files on shutdown
- mtproto/manager.go: CollectTraffic releases mutex before HTTP scrapes to
avoid blocking Ensure/Reconcile/Remove during network I/O
- database/model/model.go: panic on crypto/rand failure in mtprotoRandomMiddle
instead of silently producing a weak all-zero secret
- install.sh: fix chmod to handle renamed bin/mtg-linux-arm on armv5/v6/v7
2026-06-08 05:28:19 -07:00
if [ [ -f bin/mtg-linux-arm ] ] ; then
chmod +x bin/mtg-linux-arm
elif [ [ -f bin/mtg-linux-$( arch) ] ] ; then
chmod +x bin/mtg-linux-$( arch)
fi
2026-05-04 04:20:24 -07:00
fix(install): preserve custom bin/ files (e.g. hand-added geoip) across updates (#6152)
* fix(install): preserve custom bin/ files (e.g. hand-added geoip) across updates
Every reinstall/update wipes /usr/local/x-ui/ wholesale and re-extracts
the release tarball, which only ships known assets (xray/mtg binaries,
the bundled geoip*/geosite*.dat sets). A user-reported real incident:
a hand-placed custom geoip file referenced from a routing rule via
"ext:<file>:<code>" got silently deleted on update, and Xray refused
to start at all afterward ("failed to open <file>: no such file or
directory"), taking down every inbound until the file was manually
restored from the user's own backup.
install_x-ui now backs up the old bin/ before the wipe and restores,
after extraction, only the files the fresh release doesn't provide --
bundled assets still get the newer per-release copy, nothing custom
silently disappears. Verified in isolation: standard files (geoip.dat,
the xray binary) end up as the fresh release's copy; a custom file
absent from the release survives untouched.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix: harden the bin/ snapshot-and-restore against the review round on #6152
- Replace the mktemp+cp snapshot with a same-filesystem mv of bin/ aside:
an unchecked mktemp failure previously made the very next line copy
bin/'s contents into "/" (empty custom_bin_backup + trailing slash),
and a silently-ignored cp failure (stderr redirected, exit code never
checked) could leave a truncated custom geo file that gets "restored"
as if it were intact. A rename is atomic and needs no extra disk space,
removing both failure modes at once; if it fails, back off cleanly and
say so instead of proceeding as if a backup exists.
- Add a trap so an interrupted update (Ctrl-C, signal) between the
backup and the restore doesn't leave the snapshot (which contains
bin/config.json and every mtproto client's FakeTLS secret) sitting
around indefinitely; the two exit-path cleanups this replaces are gone
since the trap now covers those exits too.
- Move the restore below the arm arch-rename/chmod block instead of
before it, so xray-linux-arm32/mtg-linux-arm already exist under their
final names and don't get needlessly restored-then-overwritten and
misreported as "custom".
- Exclude bin/config.json and bin/mtproto/*.toml from the restore: those
are the panel's own generated runtime state (internal/xray/process.go,
internal/mtproto/manager.go), not admin-placed files, and restoring a
stale one only resurrects dead state or recreates bin/mtproto/ with the
wrong (more permissive) directory mode.
- Match symlinks in the restore's find, not just plain files -- cp -a
already preserves them in the snapshot, but the restore loop was
silently dropping them, which is exactly the failure mode (a geo file
symlinked in from elsewhere) this PR set out to fix.
- Quote the two new xui_folder expansions.
- Extend the non-interactive smoke test to reinstall over an existing
install with a sentinel file in bin/, asserting it survives and that
the bundled geoip.dat is still the release's own copy -- the update
path this PR touches had no CI coverage at all before this.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-08-14 07:41:45 -07:00
# Restore anything from the old bin/ that the fresh release doesn't ship
# (custom geoip/geosite files, or anything else an admin hand-placed
# there) -- never overwrites a same-named file the new release provides,
# so bundled assets (geoip.dat, geoip_RU.dat, ...) still get the fresh
# per-release copy. Runs after the arch-rename above so xray-linux-arm32/
# mtg-linux-arm already exist under their final names there and aren't
# mistaken for custom files needing a restore. Skips paths the panel
# itself regenerates at runtime (config.json, mtproto/*.toml -- see
# internal/xray/process.go, internal/mtproto/manager.go): those aren't
# admin-placed, and restoring a stale one only resurrects dead state (an
# orphaned mtg config for a since-deleted inbound) or the wrong
# directory permissions.
if [ [ -n " ${ custom_bin_backup } " ] ] ; then
local restored_custom_bin = ( )
while IFS = read -r -d '' f; do
local rel = " ${ f # " ${ custom_bin_backup } " / } "
case " ${ rel } " in
config.json | mtproto | mtproto/*) continue ; ;
esac
if [ [ ! -e " bin/ ${ rel } " ] ] ; then
mkdir -p " bin/ $( dirname " ${ rel } " ) "
cp -a " ${ f } " " bin/ ${ rel } "
restored_custom_bin += ( " ${ rel } " )
fi
done < <( find " ${ custom_bin_backup } " \( -type f -o -type l \) -print0)
rm -rf " ${ custom_bin_backup } "
custom_bin_backup = ""
if [ [ ${# restored_custom_bin [@] } -gt 0 ] ] ; then
echo -e " ${ green } Restored custom file(s) in bin/ not shipped by this release: ${ restored_custom_bin [*] } ${ plain } "
fi
fi
trap - EXIT INT TERM
2026-05-19 04:09:35 -07:00
# Update x-ui cli and se set permission
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
mv -f " ${ xui_script_temp } " /usr/bin/x-ui
if [ [ $? -ne 0 ] ] ; then
rm -f " ${ xui_script_temp } "
echo -e " ${ red } Failed to install x-ui.sh ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
chmod +x /usr/bin/x-ui
mkdir -p /var/log/x-ui
config_after_install
2026-01-02 18:13:00 -08:00
2026-05-19 04:09:35 -07:00
# Etckeeper compatibility
if [ -d "/etc/.git" ] ; then
if [ -f "/etc/.gitignore" ] ; then
if ! grep -q "x-ui/x-ui.db" "/etc/.gitignore" ; then
echo "" >> "/etc/.gitignore"
echo "x-ui/x-ui.db" >> "/etc/.gitignore"
echo -e " ${ green } Added x-ui.db to /etc/.gitignore for etckeeper ${ plain } "
fi
else
echo "x-ui/x-ui.db" > "/etc/.gitignore"
echo -e " ${ green } Created /etc/.gitignore and added x-ui.db for etckeeper ${ plain } "
2026-01-02 18:13:00 -08:00
fi
fi
2026-05-04 04:20:24 -07:00
2025-09-22 12:56:43 -07:00
if [ [ $release = = "alpine" ] ] ; then
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
xui_rc_temp = " /etc/init.d/x-ui.tmp. $$ "
rm -f " ${ xui_rc_temp } "
curl -fLRo " ${ xui_rc_temp } " https://raw.githubusercontent.com/MHSanaei/3x-ui/main/x-ui.rc
2025-09-25 05:34:12 -07:00
if [ [ $? -ne 0 ] ] ; then
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
rm -f " ${ xui_rc_temp } "
2026-05-19 04:09:35 -07:00
echo -e " ${ red } Failed to download x-ui.rc ${ plain } "
exit 1
fi
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if [ [ ! -s " ${ xui_rc_temp } " ] ] ; then
rm -f " ${ xui_rc_temp } "
echo -e " ${ red } Downloaded x-ui.rc is empty ${ plain } "
exit 1
fi
mv -f " ${ xui_rc_temp } " /etc/init.d/x-ui
if [ [ $? -ne 0 ] ] ; then
rm -f " ${ xui_rc_temp } "
echo -e " ${ red } Failed to install x-ui.rc ${ plain } "
exit 1
fi
2026-05-19 04:09:35 -07:00
chmod +x /etc/init.d/x-ui
rc-update add x-ui
rc-service x-ui start
else
# Install systemd service file
service_installed = false
if [ -f "x-ui.service" ] ; then
echo -e " ${ green } Found x-ui.service in extracted files, installing... ${ plain } "
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if _install_xui_service_unit "x-ui.service" "false" ; then
2026-05-19 04:09:35 -07:00
service_installed = true
2026-01-02 22:27:39 -08:00
fi
fi
2026-05-19 04:09:35 -07:00
if [ " $service_installed " = false ] ; then
2026-01-02 18:37:48 -08:00
case " ${ release } " in
ubuntu | debian | armbian)
2026-01-02 22:27:39 -08:00
if [ -f "x-ui.service.debian" ] ; then
2026-05-19 04:09:35 -07:00
echo -e " ${ green } Found x-ui.service.debian in extracted files, installing... ${ plain } "
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if _install_xui_service_unit "x-ui.service.debian" "false" ; then
2026-01-02 22:27:39 -08:00
service_installed = true
fi
fi
2026-05-04 04:20:24 -07:00
; ;
2026-01-18 06:41:07 -08:00
arch | manjaro | parch)
if [ -f "x-ui.service.arch" ] ; then
2026-05-19 04:09:35 -07:00
echo -e " ${ green } Found x-ui.service.arch in extracted files, installing... ${ plain } "
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if _install_xui_service_unit "x-ui.service.arch" "false" ; then
2026-01-18 06:41:07 -08:00
service_installed = true
fi
fi
2026-05-04 04:20:24 -07:00
; ;
2026-01-02 18:37:48 -08:00
*)
2026-01-02 22:27:39 -08:00
if [ -f "x-ui.service.rhel" ] ; then
2026-05-19 04:09:35 -07:00
echo -e " ${ green } Found x-ui.service.rhel in extracted files, installing... ${ plain } "
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if _install_xui_service_unit "x-ui.service.rhel" "false" ; then
2026-01-02 22:27:39 -08:00
service_installed = true
fi
fi
2026-05-04 04:20:24 -07:00
; ;
2026-01-02 18:37:48 -08:00
esac
2026-01-02 22:27:39 -08:00
fi
2026-05-19 04:09:35 -07:00
# If service file not found in tar.gz, download from GitHub
if [ " $service_installed " = false ] ; then
echo -e " ${ yellow } Service files not found in tar.gz, downloading from GitHub... ${ plain } "
case " ${ release } " in
ubuntu | debian | armbian)
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
service_unit_url = "https://raw.githubusercontent.com/MHSanaei/3x-ui/main/x-ui.service.debian"
2026-05-19 04:09:35 -07:00
; ;
arch | manjaro | parch)
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
service_unit_url = "https://raw.githubusercontent.com/MHSanaei/3x-ui/main/x-ui.service.arch"
2026-05-19 04:09:35 -07:00
; ;
*)
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
service_unit_url = "https://raw.githubusercontent.com/MHSanaei/3x-ui/main/x-ui.service.rhel"
2026-05-19 04:09:35 -07:00
; ;
esac
fix: make all self-managed file downloads/installs atomic, with real completion status (#5711)
* fix(script): download the live x-ui.sh script atomically before replacing it
update_menu(), update_shell(), and update.sh's update_x-ui() all overwrote
/usr/bin/x-ui in place via `curl -o`, truncating and rewriting the same
inode a currently-running x-ui process may still be reading from. A
network hiccup or slow write during that overwrite leaves a
half-old/half-new script on disk, which then fails with bogus syntax
errors on the next run. Download to /usr/bin/x-ui-temp and `mv -f` into
place instead, matching the atomic pattern install.sh already uses.
Also fixes update_menu() checking chmod's exit code instead of curl's,
which meant a failed download could still report "Update successful."
* fix(script): close remaining gaps in the atomic script-update path
Code review of the previous commit found the atomic mv fix was itself
incomplete:
- None of the mv -f calls checked their exit status, so a failed move
fell through to chmod and "success" messaging while /usr/bin/x-ui
stayed on the old file.
- update_shell()'s `[[ -s x-ui-temp ]]` guard couldn't tell "curl -z
got a 304, nothing to do" from "a stale temp file survived an
earlier crashed run" -- the latter could get moved into place with
no freshness check.
- update_menu(), update_shell(), and update_x-ui() all hardcoded the
same /usr/bin/x-ui-temp path, so two concurrent updates (e.g. a
cron auto-update racing an interactive menu update) could collide.
- update.sh's update_x-ui() was missing the non-empty-file guard
update_shell() already had.
x-ui.sh's update_menu() and update_shell() now share a
replace_xui_script() helper that uses a PID-suffixed temp path
(/usr/bin/x-ui-temp.$$), pre-cleans it before every attempt, and
checks the exit status of curl, the non-empty test, and mv before
treating the update as successful. update.sh's update_x-ui() gets the
same sequence inlined (it's fetched as a standalone script and can't
call x-ui.sh's function), closing the missing-guard gap and using its
own unique temp path.
* fix(script,panel): harden the remaining self-update download paths
install.sh had the same unguarded /usr/bin/x-ui-temp overwrite the two
already-fixed scripts had: no exit-status check on mv, and a fixed temp
name shared with x-ui.sh/update.sh's (now-unique) temp files. Give it
its own PID-suffixed temp path, an empty-file guard, and an mv
exit-status check, matching the pattern used there.
Audited the web dashboard's Go-native updater (panel.go) for the same
bug class: it already uses os.CreateTemp for a genuinely unique temp
file and cleans up via both a deferred Remove and a shell EXIT trap, so
it was never exposed to the fixed-path race. It was missing a check
for a zero-byte download (a 200 OK with an empty body would chmod +x
and exec an empty script) -- added that alongside the existing size
cap.
Not addressed here: once startUpdate()'s child process starts, the Go
service releases it and returns success immediately. If update.sh
fails partway through, the still-running old panel keeps answering
/status, so the frontend's poll can report success with no update
having happened. Fixing that needs update.sh to signal completion
status back and the frontend to check it -- a separate follow-up.
* feat(panel): report real completion status for the web self-update
Fixes the fire-and-forget gap flagged in the atomic-overwrite fix: once
startUpdate() launches update.sh detached, the Go service had no way to
learn whether it actually succeeded. If update.sh failed partway
(network drop, disk full, permission denied), the still-running old
panel kept answering /status, so the frontend's poll reported success
with nothing having changed.
update.sh now writes its outcome to a small JSON status file
(/etc/x-ui/update-status.json by default) via `trap ... EXIT`, which
covers every exit path in the script -- including the two bare `exit 1`
call sites that don't go through the existing _fail() helper. The Go
service generates a run ID before launching, passes it and the status
path to update.sh via XUI_UPDATE_RUN_ID/XUI_UPDATE_STATUS_FILE, and a
new GET /panel/api/server/getUpdateStatus endpoint reports it back. The
frontend now polls that instead of blindly trusting HTTP reachability,
and shows a distinct error or "couldn't confirm" message instead of
silently reloading into a false success.
Adversarial review of this surfaced three more issues, fixed here:
- No lock stopped two concurrent /updatePanel calls from launching two
update.sh runs that would race each other on the actual update work
(tar extraction, service unit swap). Added an in-memory guard with a
5-minute self-expiring window, so a run that never reaches a terminal
state doesn't lock out retries indefinitely.
- XUI_UPDATE_RUN_ID is read from the environment and was interpolated
unquoted into the status JSON; a malformed value would produce
invalid JSON. Now validated as digits-only before use.
- The run ID is a UnixNano timestamp (19 digits), sent as a raw JSON
number it would lose precision in JavaScript (past
Number.MAX_SAFE_INTEGER), letting two different runs round to the
same value on the wire and defeat the whole comparison. It's now a
decimal string end to end (Go, the status file, and the generated
frontend type).
install.sh's equivalent temp-file/mv path and the Go-native
downloadPanelUpdater() path were audited for the same bug classes
during this work; findings from that audit were addressed separately.
* fix(panel): release the update lock as soon as the run finishes
An exhaustive multi-angle review of the whole branch (12 finder angles,
3-vote adversarial verification, a fresh-eyes sweep) surfaced a real
bug in the concurrency guard added in the previous commit, plus several
smaller issues; this fixes what's actionable now.
The bug: acquireUpdateSlot only ever released on the 5-minute stale
timeout or if launching itself failed. If update.sh launched fine but
failed fast (bad GitHub API response, "x-ui not installed", any of its
early exit paths), the status file correctly reported "failed" within
seconds, but a retry was still rejected with "a panel update is
already in progress" for up to 5 more minutes -- the guard never
looked at the very status file this branch built to know a run was
done. It now tracks which run ID currently holds the slot and checks
that run's own status before falling back to the timeout, so a fast
failure clears the way for an immediate retry. Added a regression test
for this, plus one confirming a stale, unrelated runID can't be
mistaken for the current run finishing.
Also:
- Added a genuinely concurrent test for the guard: 200 goroutines
racing acquireUpdateSlot, asserting exactly one wins. The previous
tests only ever called it from one goroutine, so they gave no signal
if the mutex's check-then-set were silently broken -- verified this
by temporarily removing the lock and confirming the old tests still
passed while the new one caught it immediately under -race.
- Removed the redundant upfront "pending" status write: GetUpdateStatus
already defaults a missing/stale file to pending, and the frontend
matches by run ID regardless, so the write changed no observable
behavior. Deleted writeUpdateStatus entirely since that was its only
caller.
- Renamed replace_xui_script()'s unclear "conditional" parameter to
use_if_modified_since, matching what it actually controls.
- Added HTTP-level tests for the new getUpdateStatus endpoint,
including a regression test that the runId wire format is a JSON
string (decoding into a Go string field fails outright if it were
ever a bare number). updatePanel's actual launch path is not
covered: on a Linux test runner it would make a real network call
and could exec a real update.sh, so only its non-Linux guard path is
safely testable without mocking.
Not fixed here, tracked separately: the same unsafe-overwrite pattern
this branch eliminated for /usr/bin/x-ui is still present for the
systemd unit file install in update.sh and install.sh (lower severity
since systemd only reads it on daemon-reload, not continuously); and
startUpdate's systemd-run-vs-detached-fallback branching has no test
coverage since testing it safely needs dependency injection this fix
doesn't warrant bundling in.
* fix(script): make systemd unit file installation atomic
Same anti-pattern as the /usr/bin/x-ui overwrite fixed earlier: every
site that lands the systemd unit at ${xui_service}/x-ui.service --
copying it from the extracted release tarball, or falling back to a
GitHub download per distro family -- wrote straight onto the live
path via cp/curl, no temp file, no verification. A network drop
mid-download or an interrupted cp leaves the unit file truncated;
systemd then fails to parse it on the next daemon-reload/start,
leaving the panel unable to come up until an operator manually
re-copies a good unit file.
Lower severity than the /usr/bin/x-ui case (systemd only reads this
file on demand at daemon-reload time, not continuously the way bash
interprets a running script line by line), but it's the identical
gap, just left uncovered when that fix landed.
Added a small shared helper in both update.sh and install.sh --
_install_xui_service_unit() -- covering both source types (cp from
the tarball, curl from GitHub): write to a PID-suffixed temp file,
verify the copy/download succeeded and the result is non-empty, then
mv -f into place and check that exit status too, matching the pattern
already used for /usr/bin/x-ui. All 4 cp sites and the 3-way curl
fallback in each file now go through it; verified no other site
writes new content to the unit path (the remaining ${xui_service}
references are a pre-install existence check, an rm during old-version
cleanup, and the chown/chmod that already ran after the file is safely
in place -- none of those need atomicity).
Verified with bash -n on both files, plus a standalone scratch test
exercising cp-success, cp-with-missing-source, cp-with-empty-source,
and curl-failure paths: on every failure the previous, good unit file
content is left untouched and no temp file is leaked behind.
* fix(script): make Alpine's OpenRC init script install atomic; drop a stray comment
A final maximum-rigor review of the whole PR (12 finder angles including
a repo-wide sweep for any remaining instance of the bug class this PR
fixes) found two more real issues:
- Alpine's /etc/init.d/x-ui startup script is downloaded via a bare
`curl -fLRo` straight onto the live path in both update.sh and
install.sh -- the exact same unguarded-overwrite pattern already
fixed for /usr/bin/x-ui and the systemd unit file, just left
uncovered on the OpenRC side. A network drop mid-download truncates
the live init script; OpenRC then fails to source/execute it on the
next start, leaving the panel unable to come up. Fixed with the same
temp-file + non-empty check + mv -f (with its own exit-status check)
pattern used everywhere else in this PR. Verified with bash -n and a
standalone scratch-script test covering success, empty-download, and
destination-preserved-on-failure paths.
- internal/web/service/panel/panel_test.go had one line-level `//`
comment on a call site, which the root CLAUDE.md's hard rule ("No //
line comments in committed Go/TS... rename instead of annotating")
explicitly prohibits. The comment duplicated context already stated
in the test's own doc comment two lines above, so it's simply
removed rather than reworded.
Also flagged, deliberately not bundled here since it's a different
subsystem: x-ui.sh's update_geofiles() downloads Xray's live
geoip.dat/geosite.dat with the same unguarded curl -o pattern. Tracked
as its own follow-up.
* fix(script): make geo-data file downloads atomic
Same anti-pattern as /usr/bin/x-ui, the systemd unit file, and the
Alpine init script fixed in prior PRs: update_geofiles() downloaded
Xray's live geoip.dat/geosite.dat (and the IR/RU variants) with curl
writing straight onto the exact path Xray reads at runtime
(internal/xray/process.go's GetGeoipPath/GetGeositePath), no temp
file, no verification. The existing check only inspected the reported
HTTP status via -w '%{http_code}', not file integrity, so a network
drop mid-download could leave a truncated .dat file on disk that
passes the status check. Xray then fails to parse it on the next
restart/reload, breaking any routing rules that reference geoip:/
geosite:.
The -z conditional-GET usage needed care here: the original code
pointed both -z and -o at the same live path. Fixed by pointing -z at
the live file (to keep the "already current" freshness check) while
-o writes to a PID-suffixed temp file, matching the pattern already
proven in x-ui.sh's replace_xui_script(). Verified with a local HTTP
server that a 304 response leaves the temp file untouched/nonexistent
(so the existing "already up to date" branch still works unchanged),
and added a non-empty check plus a checked mv -f before treating a
download as installed.
Verified with bash -n and an end-to-end scratch test against a local
server covering: fresh download, 304-not-modified, empty response
body, and a 404 -- confirming a failure at any stage leaves the
previous good .dat file completely untouched and no temp file behind.
* fix(script): verify the release tarball extraction, not just the download
The final maximum-rigor review found the most significant remaining gap
in this whole effort: update.sh and install.sh check the tarball
download's exit status, but never check tar's exit status, and never
verify the extracted x-ui binary actually exists before continuing.
Worse, by the time extraction runs, the previous installation has
already been stopped and deleted -- there's no rollback. A truncated
download that still passes curl's own check, or a tar failure (disk
full, killed process), left the panel silently in a broken half-state:
chmod/config/service-install all continued to run against a missing or
empty binary, with no error surfaced anywhere. This is the same bug
class as everything else in this PR (unverified write to a path
something then depends on), just for the tarball itself rather than a
single file -- and it also covers the geo-data files this PR already
fixed once for the interactive/cron path, since they ship inside this
same tarball on every panel update.
Added: a non-empty check on the downloaded archive (both files, both
install.sh call sites) and a check that tar succeeded and produced a
non-empty x-ui binary before proceeding, failing loudly with a message
that explicitly says the previous install is already gone, since
silently continuing here is worse than anywhere else in this PR.
This doesn't make the multi-file extraction fully atomic (that would
mean extracting to a temp directory and atomically swapping the whole
install tree into place, a materially larger restructuring than
anything else in this PR) -- but it closes the "fails silently, user
discovers it days later when Xray can't start" gap, which was the
actual reported problem this whole effort traces back to.
Also fixed, all much smaller:
- replace_xui_script() in x-ui.sh implicitly returned chmod's exit
status instead of success, so a successful atomic install could be
reported as failed if chmod transiently failed after the mv already
landed the new script. Added an explicit `return 0`.
- update_geofiles() had no default case branch; an unrecognized
argument would silently reuse whatever dat_files/dat_source values a
previous call left in the un-scoped globals instead of failing.
Currently unreachable (all three call sites pass fixed literals) but
cheap, defensive, and worth having.
- internal/web/controller/server.go's updatePanel has one branch (an
unparseable "dev" form value) that's both untested and safe to test
on any platform, since it's rejected before any real exec/network
call. Added the missing test case.
Verified: bash -n on all three scripts; an empirical scratch test
covering an empty downloaded archive, a corrupt (non-gzip) archive,
and a successfully-extracting-but-empty archive, confirming each is
caught before the script proceeds; full go build/vet/test -race
across the whole module; frontend generation confirmed still in sync.
* fix(panel): base the update-slot staleness fallback on process liveness
Addresses the automated review on the upstream PR (MHSanaei/3x-ui#5711).
Blocking finding: acquireUpdateSlot's staleness fallback freed the
update slot purely on elapsed wall-clock time (5 minutes), with no
check on whether the update.sh process it launched was actually still
running. update.sh runs install_base() (apt-get/dnf/pacman update and
install) before update_x-ui even starts, plus several GitHub
downloads (release tarball, x-ui.sh, and possibly a service unit or
x-ui.rc) -- on a slow or throttled host, a small VPS being the typical
deployment target for this project, that alone can plausibly exceed 5
minutes with nothing wrong. A second /updatePanel call arriving in
that window (an admin retrying after the frontend's 90s poll times
out, or overlapping master-node bulk-update calls) would launch a
second update.sh, racing the exact rm/tar/mv/systemctl sequence this
whole PR exists to make safe.
Fixed by recording the launched process's PID (detached-fallback path
only; the systemd-run path's own process has already exited by the
time startUpdate returns, so it never learns update.sh's real PID) and
checking it via the standard POSIX kill(pid, 0) liveness probe before
treating a run as stale, following the existing panel_unix.go /
panel_other.go platform-split pattern already used for
setDetachedProcess. A confirmed-alive process now keeps the slot held
past updateStaleAfter (raised from 5 to 20 minutes as a safer baseline
for the systemd-run path, which still has no way to check liveness
directly). updateHardCeiling (2 hours) is an absolute backstop so a
genuinely wedged run can never lock out retries permanently even on
the PID-tracked path.
Added two regression tests exercising the new logic (gated to Linux,
since processAlive is a no-op stub elsewhere): a live PID keeps the
slot held past the stale window, and the hard ceiling overrides
liveness. Traced both by hand against the new acquireUpdateSlot logic;
could not execute-verify processAlive itself on this Windows dev
machine (no WSL distro installed, and installing one felt
disproportionate to validate kill(pid, 0), an extremely well-established
POSIX primitive), but cross-compiled clean for linux/amd64 and this
repo's CI runs the real test suite on Linux.
Also fixed, both suggestions from the same review:
- install.sh: two failure paths right after tarball extraction were
exiting without cleaning up the already-downloaded x-ui.sh temp file
(xui_script_temp), leaving it behind. Every other new failure branch
in this PR removes its temp file before exiting; these two now do
too.
- frontend/src/pages/api-docs/endpoints.ts: updatePanel's doc entry
did not reflect that a successful response now carries an obj with
runId. Added an inline response example matching the existing
pattern used for other ad hoc (non-schema-backed) responses like
getWebCertFiles.
Verified: go build/vet clean on both windows (native) and a linux/amd64
cross-compile; full go test ./... clean; go test -race on the panel
and controller packages; bash -n on all three shell scripts; npm run
gen confirms the openapi.json diff is exactly the new response example
with no stray changes to src/generated; TestAPIRoutesDocumented still
passes.
2026-07-02 09:19:33 -07:00
if ! _install_xui_service_unit " $service_unit_url " "true" ; then
2026-05-19 04:09:35 -07:00
echo -e " ${ red } Failed to install x-ui.service from GitHub ${ plain } "
exit 1
fi
service_installed = true
fi
if [ " $service_installed " = true ] ; then
echo -e " ${ green } Setting up systemd unit... ${ plain } "
chown root:root ${ xui_service } /x-ui.service > /dev/null 2>& 1
chmod 644 ${ xui_service } /x-ui.service > /dev/null 2>& 1
systemctl daemon-reload
systemctl enable x-ui
systemctl start x-ui
else
echo -e " ${ red } Failed to install x-ui.service file ${ plain } "
exit 1
fi
2025-09-22 12:56:43 -07:00
fi
2026-05-04 04:20:24 -07:00
2026-06-22 14:49:09 -07:00
# IP Limit relies on fail2ban; install + configure it now so the feature
# works out of the box (no-op when XUI_ENABLE_FAIL2BAN=false). Never fatal.
setup_fail2ban
2026-05-19 04:09:35 -07:00
echo -e " ${ green } x-ui ${ tag_version } ${ plain } installation finished, it is running now... "
2023-02-09 11:18:06 -08:00
echo -e ""
2024-12-20 09:33:27 -08:00
echo -e " ┌───────────────────────────────────────────────────────┐
│ ${ blue } x-ui control menu usages ( subcommands) :${ plain } │
│ │
│ ${ blue } x-ui${ plain } - Admin Management Script │
│ ${ blue } x-ui start${ plain } - Start │
│ ${ blue } x-ui stop${ plain } - Stop │
│ ${ blue } x-ui restart${ plain } - Restart │
│ ${ blue } x-ui status${ plain } - Current Status │
│ ${ blue } x-ui settings${ plain } - Current Settings │
│ ${ blue } x-ui enable${ plain } - Enable Autostart on OS Startup │
│ ${ blue } x-ui disable${ plain } - Disable Autostart on OS Startup │
│ ${ blue } x-ui log${ plain } - Check logs │
│ ${ blue } x-ui banlog${ plain } - Check Fail2ban ban logs │
│ ${ blue } x-ui update${ plain } - Update │
2025-11-07 10:26:43 -08:00
│ ${ blue } x-ui legacy${ plain } - Legacy version │
2024-12-20 09:33:27 -08:00
│ ${ blue } x-ui install${ plain } - Install │
│ ${ blue } x-ui uninstall${ plain } - Uninstall │
2025-12-03 15:05:21 -08:00
└───────────────────────────────────────────────────────┘"
2023-02-09 11:18:06 -08:00
}
echo -e " ${ green } Running... ${ plain } "
install_base
2026-05-19 04:09:35 -07:00
install_x-ui $1