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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
- 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>
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>
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>
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.
The bot's add-client flow already serialised client_Flow into the VLESS
JSON template but never exposed a way to set it from Telegram, so every
client ended up with an empty flow regardless of the inbound's transport.
Added an inline "Flow" row to the VLESS protocol keyboard with three
choices — None, xtls-rprx-vision, and xtls-rprx-vision-udp443 — and a
matching i18n key in all 13 locale files. The row is only shown when
the inbound can actually use Vision flow (mirrors the frontend's
canEnableTlsFlow check: VLESS over TCP with TLS or Reality); on other
transports it's hidden and any stale client_Flow value is reset, so the
generated JSON stays consistent with the inbound's stream settings.
SetInboundEnable called rt.DelInbound for every runtime, but Remote.DelInbound
hits panel/api/inbounds/del/:id on the node — a real row delete, not just a
"stop serving" hint like Local.DelInbound. Flipping the enable switch on a
remote inbound therefore wiped the row on the node entirely.
Route remote inbounds through UpdateInbound instead so the row stays and only
the enable flag is patched. Local path keeps the Del+Add flow since that's
how Xray's gRPC API expects to be driven.
Fixes#4402
The fast-probe mode hard-coded net.DialTimeout("tcp", ...), so testing a
WARP/WireGuard or Hysteria outbound always failed with an i/o timeout —
those transports only listen on UDP, never on TCP.
Probe is now transport-aware: extractOutboundEndpoints tags each endpoint
with the network the proxy actually listens on (UDP for wireguard,
hysteria, and any outbound whose streamSettings.network is hysteria, kcp,
or quic; TCP otherwise). probeUDPEndpoint dials UDP, writes a single
sentinel byte so the kernel can surface ICMP errors, and treats a read
timeout as success (WireGuard ignores invalid packets, so silence is the
expected reply from a reachable server). The result's mode field now
reflects what was probed, so the UI badge shows UDP for these outbounds
instead of mislabelling them as TCP.
* fix: prevent online clients from randomly disappearing from panel UI
Online status was determined solely by whether a client transferred
bytes in the current 5-second polling window. The online list was
completely replaced each cycle, so idle-but-connected clients with no
traffic delta in that window were dropped from the UI.
Now online status is computed from lastOnline DB timestamps with a
5-second grace period via RefreshOnlineClientsFromMap(), so clients
remain visible across idle polling windows.
Closes#4384
* fix: extend online client grace period to survive idle poll cycles
The 5s grace period equalled the traffic-poll interval, so a client
whose Xray stats reported a zero delta for one cycle was still dropped
on the very next tick. Bump to 20s (~4 polls) so idle-but-connected
sessions stay visible across momentary counter gaps without lingering
long after a real disconnect.
Refs #4384
---------
Co-authored-by: MHSanaei <ho3ein.sanaei@gmail.com>
- Frontend: Only include streamSettings in toJson() for vmess, vless, trojan, shadowsocks, and hysteria protocols
- Frontend: Hide Stream tab in Advanced section for unsupported protocols
- Frontend: Clear streamSettings in Advanced tab when switching to unsupported protocols
- Frontend: Add CodeMirror JSON editor to config view in index page with mobile responsive design
- Backend: Add normalizeStreamSettings() to clear streamSettings for tunnel, mixed, http, tun, and wireguard protocols
- Backend: Apply normalization in AddInbound() and UpdateInbound()
- Backend: Add omitempty JSON tag to StreamSettings field to exclude null values from Xray config
Two bugs combined to leave per-client traffic / remained / all-time
columns stuck at stale numbers while only the inbound-level row and
the online badge refreshed:
1. Backend (xray + node sync traffic jobs) only included the per-client
array in the client_stats broadcast when activeEmails / touched
was non-empty. Cycles with no client deltas — or any node sync that
failed to fetch a snapshot — shipped only the inbound summary, so
the frontend had nothing to merge for clients. Replaced both code
paths with a single GetAllClientTraffics() snapshot per cycle; the
broadcast now always carries the full client list.
2. Frontend mutated dbInbound.clientStats[i] in place. DBInbound is a
plain class instance (not wrapped in reactive()), so Vue could not
see the field-level changes and ClientRowTable's statsMap computed
stayed cached forever. Added a statsVersion tick bumped on every
merge and read inside statsMap so the computed re-evaluates and the
template pulls fresh up/down/allTime/expiryTime each push.
Removed the now-dead emailSet helper from node_traffic_sync_job and
the activeEmails filter from xray_traffic_job.
* tunnel: rename settings to Xray's current schema (address →
rewriteAddress, port → rewritePort, network → allowedNetwork) in
the model, form modal, info modal, and the bundled API inbound
template; expose portMap so per-port forwarding can be configured
from the panel.
* tun: add the full TUN protocol form and read-only info blocks
(name, mtu, gateway, dns, userLevel, autoSystemRoutingTable,
autoOutboundsInterface) — previously the protocol was selectable
but the form rendered blank.
* hysteria: surface the stream-level version, obfs password, and
udpIdleTimeout fields that the model already supported.
Refs https://xtls.github.io/config/inbounds/tunnel.html
Refs https://xtls.github.io/config/inbounds/tun.html
Refs https://xtls.github.io/config/transports/hysteria.html
The license update was always failing because the Cloudflare response has
no `success` field — the check rejected every successful PUT. On real
errors (e.g. "Too many connected devices."), the toast leaked the raw URL
+ JSON body. Now the WARP API's error envelope is parsed into a clean
message and shown inline next to the Update button.
InboundFormModal: switching out of the Advanced tab now parses the three
JSON textareas and rebuilds the structured Inbound via Inbound.fromJson,
so the Basic tab reflects what was pasted. Invalid JSON keeps the user
on Advanced with a specific parse error.
XrayPage: Save now parses xraySetting upfront and snaps the user back to
the Advanced tab on invalid JSON instead of letting the backend reject a
generic blob.
Replace the single regenerable API token with a named-token list:
- New ApiToken model + service with constant-time auth matching
- Seeder migrates the legacy `apiToken` setting into a "default" row
- Security tab gets create/enable/delete UI; api-docs page links to it
- Dedicated "API Tokens" section in the in-panel docs
URL anchors now reflect the active tab/section on Settings, Xray, and
API Docs pages, so deep links like `/panel/settings#security` work.
Translations for the 8 new SecurityTab strings added across all locales.
Add UserService.BumpLoginEpoch and call it from updateSetting when
TwoFactorEnable flips false → true. Existing cookies (issued under
the looser no-2FA policy) get a 401 on their next request and are
forced through the login flow. Disabling 2FA is a relaxation and
does not bump the epoch — sessions stay valid.
Also fix the dev-mode 401 redirect: targeting `${basePath}login.html`
breaks when basePath isn't "/" (Vite has no file at e.g.
"/test/login.html"; the SPA fallback loops the 401). Navigate to
basePath instead — Vite's bypassMigratedRoute and Go's index
handler both serve login.html for that path.
Strip stale doc-comment from netsafe and IndexController.logout
in line with the project's no-inline-comments convention.
AddInboundClient and UpdateInboundClient previously accepted an
empty Email field for every protocol except shadowsocks (where
email doubles as the client ID). Empty emails break downstream
features that key off email — IP-limit logging, traffic stats,
client-online tracking, subscription remarks.
Reject empty/whitespace-only emails at the service layer so the
API surface (POST /panel/api/inbounds/addClient and
/updateClient/:id) returns a clear error instead of persisting
an unidentifiable client.
Also drop the stale `len(Email) > 0` guard in UpdateInboundClient
that became dead code once empty emails are rejected.
The Node.Probe and Remote.do paths built outbound URLs by string-
formatting admin-controlled fields (Scheme/Address/Port/BasePath)
straight into requests, then dialed the result with the default
transport. CodeQL flagged this as go/request-forgery — an admin
(or anyone who compromises the admin account) could point a node
at internal infrastructure (cloud metadata, RFC1918 ranges, etc.)
and the panel would dutifully fetch it.
Add util/netsafe with a shared TOCTOU-safe DialContext that
resolves the host, rejects private/internal IPs unless the
per-request context whitelists them (per-node AllowPrivateAddress
flag, plumbed through context.Value), and dials the resolved IP
directly so the IP that passed the check is the IP we connect to.
This closes the DNS-rebinding window where a hostname could
resolve to a public IP at check time and a private one at dial.
Also tighten address validation (NormalizeHost rejects anything
that isn't a bare hostname or IP literal — no embedded paths,
userinfo, schemes) and switch URL construction from fmt.Sprintf to
url.URL{} + net.JoinHostPort so admin-supplied values can't smuggle
URL components.
custom_geo.go's isBlockedIP now delegates to netsafe so there's
one source of truth.
* refactor(session): store user ID in session instead of full struct
Replaces storing the full User object in the session cookie with just
the user ID. GetLoginUser now re-fetches the user from the database on
every request so credential/permission changes take effect immediately
without requiring a re-login. Includes a backward-compatible migration
path for existing sessions that still carry the old struct payload.
* feat(auth): block panel with default admin/admin credentials and guide credential change
checkLogin middleware now detects default admin/admin credentials and
redirects every panel route to /panel/settings until they are changed.
The settings page auto-opens the Authentication tab, shows a
non-dismissible error banner, and lists 'Default credentials' first in
the security checklist. Login response includes mustChangeCredentials
so the login page can redirect directly. Logout is now POST-only.
Password must be at least 10 characters and cannot be admin/admin.
* feat(settings): redact secrets in AllSettingView and add TrustedProxyCIDRs
Introduces AllSettingView which strips tgBotToken, twoFactorToken,
ldapPassword, apiToken and warp/nord secrets before sending them to
the browser, replacing them with boolean hasFoo presence flags. A new
/panel/setting/secret endpoint allows updating individual secrets by
key. Secrets that arrive blank on a save are preserved from the DB
rather than overwritten. Adds TrustedProxyCIDRs as a configurable
setting (defaults to localhost CIDRs). URL fields are validated before
save.
* fix(security): SSRF prevention, trusted-proxy header gating, CSP nonce, HTTP timeouts
Adds SanitizeHTTPURL / SanitizePublicHTTPURL to reject private-range
and loopback targets before any outbound HTTP request (node probe,
xray download, outbound test, external traffic inform, tgbot API
server, panel updater). Forwarded headers (X-Real-IP, X-Forwarded-For,
X-Forwarded-Host) are now only trusted when the direct connection
arrives from a CIDR in TrustedProxyCIDRs. CSP policy is tightened with
a per-request nonce. HTTP server gains read/write/idle timeouts. Panel
updater downloads the script to a temp file instead of piping curl into
shell. Xray archive download adds a size cap and response-code check.
backuptotgbot is changed from GET to POST.
* feat(nodes): add allow-private-address toggle per node
Adds AllowPrivateAddress to the Node model (DB default false). When
enabled it bypasses the SSRF private-range check for that node's probe
URL, allowing nodes hosted on RFC-1918 or loopback addresses (e.g.
a private VPN or LAN setup).
* chore: frontend UX improvements, CI pipeline, and dev tooling
- AppSidebar: logout via POST /logout instead of navigating to GET
- InboundList: persist filter state (search, protocol, node) to
localStorage across page reloads; add protocol and node filter dropdowns
- IndexPage: add health status strip (Xray, CPU, Memory, Update) with
quick-action buttons
- dependabot: weekly go mod and npm update schedule
- ci.yml: add GitHub Actions workflow for build and vet
- .nvmrc: pin Node 22 for local development
- frontend: bump package.json and package-lock.json
- SubPage, DnsPresetsModal, api-docs: minor fixes
* fix(ci): stub web/dist before go list to satisfy go:embed at compile time
* chore(ui): remove health-strip bar from dashboard top
* Revert "feat(auth): block panel with default admin/admin credentials and guide credential change"
This reverts commit 56ce6073ce.
* fix(auth): make logout POST+CSRF and propagate session loss to other tabs
- Switch /logout from GET to POST with CSRFMiddleware so it matches the
SPA's existing HttpUtil.post('/logout') call (previously 404'd silently)
and blocks GET-based logout via image tags or link prefetchers. Handler
now returns JSON; the SPA already navigates client-side.
- Return 401 (instead of 404) from /panel/api/* when the caller is a
browser XHR (X-Requested-With: XMLHttpRequest) so the axios interceptor
redirects to the login page on logout-in-another-tab, cookie expiry,
and server restart. Anonymous callers still get 404 to keep endpoints
hidden from casual scanners.
- One-shot the 401 redirect in axios-init.js and hang the rejected
promise so queued polls don't stack reloads or surface error toasts
while the browser is navigating away.
- Add the CSP nonce to the runtime-injected <script> in dist.go so the
panel loads under the existing script-src 'nonce-...' policy.
- Update api-docs endpoints.js: GET /logout doc entry was missing.
* fix(settings): POST /logout after credential change
* fix(auth): invalidate other sessions when credentials change
When the admin changes username/password from one machine, sessions
on every other machine kept working until they manually logged out
because session storage is a signed client-side cookie — there is
no server-side session list to revoke.
Add a per-user LoginEpoch counter stamped into the session at login
and re-verified on every authenticated request. UpdateUser and
UpdateFirstUser bump the epoch (UpdateUser via gorm.Expr so a single
update statement is atomic), so any cookie issued before the change
no longer matches the user's current epoch and GetLoginUser returns
nil — the SPA's 401 interceptor then redirects to the login page.
Backward compatible: the column defaults to 0 and missing cookie
values are treated as 0, so sessions issued before this change
remain valid until the first credential update.
---------
Co-authored-by: Sanaei <ho3ein.sanaei@gmail.com>
Since v2.9.4, disableInvalidClients sets c['enable']=false in inbound settings JSON when a client hits its limit. autoRenewClients only updated client_traffics.enable - never flipped the JSON field back. The Xray config generator checks both, so client stayed excluded after renewal.
* feat(traffic_writer): enhance traffic writer with concurrency safety and state management
* Revert "feat(traffic_writer): enhance traffic writer with concurrency safety and state management"
This reverts commit e6760ae396.
* feat(vless): clarify VLESS encryption auth selection and enhance parsing logic
* feat(traffic_writer): enhance traffic writer with concurrency safety and state management
* Revert "feat(traffic_writer): enhance traffic writer with concurrency safety and state management"
This reverts commit e6760ae396.
* feat(traffic_writer): enhance traffic writer with concurrency safety and state management
* feat(web): implement panel-only start/stop methods for in-process restarts
Polls xray's /debug/vars on the 2s status tick, stores memstats and per-outbound observatory delay in the metric history ring buffer, and exposes them through a new XrayMetricsModal opened from the Charts card. Restructures the dashboard to consolidate uptime, usage, version, and Telegram link into stat-style or action-style cards consistent with the existing AntD aesthetic.
After a SIGHUP-driven panel restart (which is exactly what the frontend
triggers after a successful DB import via /panel/setting/restartPanel),
the previous implementation deadlocked:
1. server.Stop() called StopTrafficWriter — cancels the context and waits
for the consumer goroutine to exit. The goroutine dies.
2. server.Start() called StartTrafficWriter, but sync.Once had already
fired, so it was a no-op. twQueue still pointed to the old channel
with no consumer.
3. startTask() → RestartXray(true) → GetXrayConfig() →
InboundService.AddTraffic(nil, nil) → submitTrafficWrite. The send
to twQueue succeeded (buffer space) but <-req.done blocked forever
because no goroutine was draining the channel.
4. RestartXray held the global xray lock for the entire hang, so every
subsequent restart attempt from the panel UI also blocked on
lock.Lock(). User-visible symptom: xray stopped silently after DB
import and no panel action could revive it.
Replace sync.Once with a mutex-guarded Start that spawns a fresh
goroutine on each cycle, and a Stop that resets the package state so
the next Start works. runTrafficWriter now takes its channels as
parameters instead of reading package vars, so the old goroutine can't
interfere with a new one if their lifetimes briefly overlap.
- New GET /panel/api/inbounds/getSubLinks/:subId and /getClientLinks/:id/:email
return the same protocol URLs the panel UI's Copy button emits, honouring
X-Forwarded-Host / X-Forwarded-Proto. Documented in the API docs page.
- Refactor: sub package no longer imports web. The embedded dist FS is
injected via sub.SetDistFS, and the link generator is registered with the
service layer via service.RegisterSubLinkProvider, avoiding the circular
import the new endpoints would otherwise introduce.
- Security: stop emitting window.X_UI_CUR_VER on login.html and drop the
visible version chip from the login page, so the panel version is no
longer pre-auth info disclosure. Authenticated pages still receive it.
- Bump config/version.
Different nodes are different machines, so same port + transport across
NodeIDs shouldn't conflict. resolveInboundTag now keeps a caller-supplied
unique tag verbatim so central and node panels stay in agreement instead
of regenerating into a UNIQUE constraint failure on sync.
Frontend (NordModal.vue):
- Server selector gets show-search with the option label set to
`${cityName} ${name} ${hostname}` so admins can find a specific
server inside a 100+ entry country list by typing.
- Each option renders the load as a colored a-tag (green <30%,
orange 30-70%, red >70%) instead of plain text — quicker visual
scan when sorting through servers in the dropdown.
Backend (nord.go):
- GetCountries / GetServers now check resp.StatusCode and return
"NordVPN API error: <status>" on non-200, matching the pattern
GetCredentials already used. Previously a 4xx/5xx body was
returned as a "success" string and the frontend silently failed
to parse it, surfacing only as an empty "No servers found".
- GetCredentials drops its own ad-hoc 10s http.Client and reuses
the shared nordHTTPClient (15s) — one client, one timeout.
- service.TestOutbound now dispatches on `mode`:
- "tcp": parallel net.DialTimeout to every server/peer endpoint
(vmess/vless/trojan/ss/socks/http/wireguard). No xray spin-up,
no semaphore — safe to run concurrently across outbounds.
- "http" (default): existing temp-xray + SOCKS path, now with an
httptrace.ClientTrace breakdown (DNS / Connect / TLS / TTFB)
alongside the total delay and status code.
- testSemaphore renamed to httpTestSemaphore — only HTTP probes
serialise, TCP runs free.
- TestOutboundResult carries the per-mode extras: timing fields for
HTTP, per-endpoint dial list for TCP, plus a `mode` echo.
- Controller reads `mode` from the form and passes it through.
- useXraySetting: testOutbound accepts mode (default "tcp"); new
testAllOutbounds(mode) runs a worker pool (concurrency 8 for TCP,
1 for HTTP) and skips blackhole / loopback / blocked outbounds —
also skips freedom / dns under TCP since they have no endpoint.
- OutboundsTab: TCP/HTTP radio toggle and a Test All button land in
the toolbar; the per-row ⚡ now uses the selected mode. Results
surface in a popover with the full timing breakdown plus the
endpoint list for TCP probes. Latency header replaces the duplicate
"check" column title.
Practical effect: testing ten outbounds in TCP mode drops from ~50–100s
(serial HTTP) to ~1–2s (parallel dial × 8). HTTP mode stays as the
authoritative probe and now shows where the latency actually lives.
- Traffic-writer single-consumer queue (web/service/traffic_writer.go)
serialises every DB write that touches up/down/all_time/last_online
(AddTraffic, SetRemoteTraffic, Reset*, UpdateClientTrafficByEmail) so
overlapping goroutines can no longer clobber each other's column-scoped
Updates with a stale tx.Save.
- DB pool: WAL + busy_timeout=10s + synchronous=NORMAL + _txlock=
immediate, MaxOpenConns=8 / MaxIdleConns=4. The immediate-tx PRAGMA
fixes residual "database is locked [0ms]" cases where deferred-tx
writer-upgrade conflicts bypass busy_timeout.
- SetRemoteTraffic full-mirrors node-authoritative state into central:
settings JSON, remark, listen, port, total, expiry, all_time, enable,
plus per-client total/expiry/reset/all_time. Inbounds and
client_traffics rows present on node but missing from central are
created; rows missing from snap are deleted (with cascading
client_traffics removal).
- NodeTrafficSyncJob detects structural changes from the mirror and
broadcasts invalidate(inbounds) so open central UIs re-fetch via REST
on node-side add/del/edit without manual refresh.
- XrayTrafficJob broadcasts invalidate(inbounds) when auto-disable flips
client_traffics.enable so the per-client toggle reflects depletion
without manual refresh.
- Frontend: inbounds page now subscribes to the BroadcastInbounds 'inbounds'
WS event (full-list pushes from add/del/update controllers were silently
dropped). Fixes invalidate payload field (dataType -> type). Restart-
panel modal switched from Promise-wrap to onOk-only so Cancel actually
cancels.
- Node files trimmed of stale prose-comments; cron cadence dropped
10s -> 5s to match the inbounds page UX.
- README badges and Go module path bumped v2 -> v3 to match module rename.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Several file operations used os.ModePerm (0777) which makes files
world-writable and world-readable, violating the principle of least
privilege:
- database/db.go: InitDB directory creation → 0755
- xray/process.go: Xray config write → 0644
- xray/process.go: Crash report write → 0644
- web/service/server.go: Binary extraction → 0755
Also removes unused "io/fs" imports from the affected files.
The Node model only carried `json:` tags, so when the panel's axios
posted form-encoded bodies to /panel/api/nodes/add and /test, Gin's
form binder produced a zero-valued Node — empty Name, empty Address,
Port=0 — surfacing as "node name is required" and a probe URL of
"https://:0/...". Add `form:` tags so add/test bind correctly.
Also skip inbounds with NodeID set when building the central xray
config; otherwise the central panel tried to listen on ports owned by
node-managed inbounds and xray-core failed to start with a bind
collision.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
