* fix(settings): keep the stored port when a port field is cleared
Clearing the panel-port, subscription-port or LDAP-port InputNumber
fired onChange(null), which the handlers coerced to 0; on blur Ant
Design clamped the empty field to min=1 and the next save silently
persisted port 1. For subPort that breaks the generated subscription
links; for webPort it moves the panel itself to port 1 and locks the
admin out until the port is fixed via the x-ui CLI.
Ignore null changes so clearing a port field snaps back to the last
valid value instead of committing a bogus port.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(settings): pin cleared port fields to the stored value
Clearing the subscription-port field must not reach updateSetting at
all, while typed ports still pass through unchanged. Pins the fix so a
handler refactor cannot silently reintroduce the clamped port 1.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
The Nodes page cache is overwritten wholesale by the heartbeat websocket
push, but the job broadcast a raw []*model.Node while the REST list returns
[]*service.NodeView. model.Node tags the api token json:"-" and carries no
hasApiToken field, so every push stripped the flag the edit form reads to
decide whether a token is already stored.
One 5s tick after the page loaded, editing any non-mTLS node then failed
with "Name, address, port and API token are required" — and stayed failed,
because setQueryData refreshes dataUpdatedAt, so the query never goes stale
and never refetches the intact REST payload.
Broadcast the NodeView read contract instead.
The frontend's golden fixtures are the panel's model of an xray config, but
nothing ever asked xray-core whether it would accept them: the snapshots only
prove the Zod schemas agree with themselves. Building every fixture through the
same config builders the panel hands its config to — conf.InboundDetourConfig
for the full-config and AddInbound paths, conf.RouterConfig for
ApplyRoutingConfig, conf.DNSConfig for the dns section — found seven the core
refuses, three of them reachable from the panel's own UI. A refusal is not
scoped to one inbound: the config fails to load and every inbound stays down.
Hysteria: xray-core builds version 2 only, in both the protocol settings and
the transport settings, but the inbound settings schema accepted any version
from 1 up and its comment claimed upstream still supported v1. Both fixtures
carried version 1. The schema now pins 2, GenXrayInboundConfig heals stored
rows on the way out the way it already heals shadowsocks ciphers and wireguard
peers, and the share link drops the dead hysteria:// scheme — the subscription
server already emitted hysteria2:// for the same inbound.
XHTTP uplinkDataPlacement: both transport forms offered "query", which the core
has never accepted for that field (auto and body always, cookie and header in
packet-up mode). Replaced with auto, which was missing, and the default label
now names auto rather than body.
FinalMask items: switching an item to the rand-driven array kind wrote
packet:[] next to the rand. xray-core counts an empty array as a packet and
every item kind is exclusive, so noise answers "len(item.Packet) > 0 &&
item.Rand.To > 0" and header-custom "exactly one item kind must be set". The
editor now clears the packet, and GetXrayConfig strips the residue from rows
already saved with it.
The remaining four were stale fixtures: an xmc mask still on the usernames
shape v26.7.28 replaced with profiles, a fragment mask with no length, and
header-custom and noise items passing an array to the string packet kind — all
shapes the panel's own editors cannot produce.
golden_fixtures_xray_test.go keeps this from drifting again: every fixture in
every category is built through xray-core on each run, with a self-signed pair
standing in for the deployment certificate paths, so the next core bump reports
which fixture it broke.
Exercising the whole XrayAPI surface against a real xray-core 26.7.28 (the
version go.mod pins) turned up a way for ordinary panel activity to kill the
core process, plus two smaller mismatches with what the core actually does.
buildUserAccount picked the shadowsocks account type by falling through to a
2022 account whenever the cipher was not one of six hardcoded names. xray's
legacy and 2022 inbounds cast the account they are handed without checking
(proxy/shadowsocks/validator.go, proxy/shadowsocks_2022/inbound_multi.go), so
the wrong type is not an error — it panics the core and drops every connection
on the server. The fallback was reachable without any misconfiguration:
autoRenewClients hands AddUser the client object straight out of the inbound's
settings, where the cipher lives under "method", never "cipher", so every
auto-renewed client on a legacy-cipher shadowsocks inbound took xray down. The
xray-valid aead_* aliases hit it too. The cipher is now read from either key,
matched with the same table (and case-insensitivity) the core's own conf
package uses, and an unrecognized one is an error instead of a guess.
The legacy shadowsocks validator is also the only one that accepts a second
user under an email it already holds, and RemoveUser then drops just one of
them — a disabled or expired client kept connecting. AddUser now drops the
email first on that account type so a single removal fully revokes the client.
GetTraffic skipped every stat the first time it saw it. xray creates a
counter on a user's first use, so that dropped a new client's traffic for a
whole polling interval, as did the counter reset after a core restart. Only
the first poll of a process is a baseline now; later, unseen and rewound
counters both count from zero.
Also fixes three unchecked settings["method"].(string) assertions that panic
the panel on a shadowsocks inbound whose settings carry no method, and bounds
TestRoute's port so an out-of-range value cannot wrap into the uint32 the
core is asked about.
Tests: api_users_e2e_test.go drives add/remove for every protocol against a
real core and asserts it survives each one (skipped unless XRAY_E2E_BINARY is
set); the account-type, traffic-delta and renew paths get unit coverage.
Bump xtls/xray-core to 5ca6f4b7d4dc (v26.7.28) and move the three binary
pins (DockerInit.sh, the Linux and Windows URLs in release.yml) in lockstep
so the in-process conf.Build() validation and the child binary agree.
XMC finalmask (#6487) is the breaking change. The mask's `usernames` string
list is gone, replaced by a required `profiles` array whose entries each need
a 3-16 character [A-Za-z0-9_] username, a parseable UUID and both Mojang
texture fields; the "default to Dream when empty" fallback was removed, so an
xmc mask saved by an older panel now fails to build and takes the whole
config down with it rather than degrading one inbound.
The textures are a signed blob only Mojang's session server can issue, so a
legacy username cannot be upgraded automatically. The panel now:
- rejects an incomplete xmc mask at save time (AddInbound/UpdateInbound),
pointing at the specific field that is missing;
- drops only the offending mask when generating the core config, for rows
that never went through the form (upgrade, node sync, restored backup,
direct DB edit), warning which inbound lost its obfuscation instead of
leaving every inbound offline;
- carries legacy usernames into profile stubs in the finalmask form so the
operator keeps their player names and sees exactly what still needs
filling in, and edits profiles through a list editor.
No destructive DB migration: unlike the removed shadowsocks ciphers there is
no valid replacement to rewrite to, and dropping the mask from stored rows
would discard the operator's hostname and password for config they can still
repair. The generation-time strip already prevents the startup failure.
Also track the core's xmux maxConnections fallback, lowered from 6 to 3 for
anti-TSPU, in the fresh-XMUX seed so a new panel config matches what the core
would pick on its own.
TUN gained a `desc` key and random utunN naming, but the Go validator no
longer accepts TUN inbounds and the panel only renders legacy saved rows, so
nothing there needs adapting. The remaining commits are REALITY log-warning
wording, gRPC/XHTTP localAddr accuracy and a routing tweak, none of which
change the JSON config surface.
Tests cross-check the panel's profile predicate against conf.XMCProfile.Build()
so a future core release that tightens or relaxes the rules fails loudly
rather than silently emitting configs the core refuses to start on.
Domain-based Routing rules could never match RouteThroughXray traffic: an
AmneziaWG peer resolves DNS itself, through the tunnel, before ever sending
a packet, so the decapsulated traffic TPROXY hands to the bridge is already
a bare destination IP with no domain name attached at the network layer.
Every other inbound recovers this via sniffing (confirmed working for the
stock wireguard inbound, which does have it configured); the bridge never
got a sniffing block at all, so only tag/IP/network-based rules could ever
match it -- any domain rule above it in the list was silently unreachable.
The dokodemo-door TPROXY bridge every AmneziaWG peer's traffic is routed
through has no per-user identity, so Xray's own access log never carries an
"email:" token for these lines -- the Access Logs modal showed a blank
Email column for every in-*-udp row, even though every other protocol's
rows show the client normally.
The peer's decapsulated tunnel IP does survive as the log's "from" address,
and that IP deterministically maps to exactly one configured peer. Builds a
"<inbound tag>|<ip>" -> email index from the same AmneziaWG inbounds already
parsed elsewhere (amneziawg.InstanceFromInbound), and fills in Email from it
whenever the raw log line didn't have one.
The Inbounds list only special-cased isWireguard/isHysteria for the "UDP"
network badge, so an AmneziaWG row showed just the bare protocol tag with
no transport badge next to it. Added the missing isAmneziawg flag (mirrors
isWireguard exactly) and wired it into the same branch.
Client-row protocol-color maps in ClientsPage/HostList had no amneziawg
entry, silently falling back to grey -- ClientInfoModal already had
amneziawg: 'yellow' from earlier work, these two just never got it.
TPROXY never rewrites a packet's own destination address, only the routing
decision. A default-deny firewall whose INPUT chain sanity-checks "is this
destination actually local" (UFW's ufw-not-local, via addrtype --dst-type
LOCAL, is a concrete example) silently drops the redirected packet before
Xray's socket ever sees it -- RouteThroughXray looked fully configured
(TPROXY rule present and counting, Xray listening with IP_TRANSPARENT set)
yet every peer's traffic vanished with no trace on either side.
Adds an idempotent, never-torn-down "iptables -I INPUT 1 -m mark --mark
<fwmark> -j ACCEPT" alongside the existing shared policy route, so this
works regardless of which firewall manager owns the rest of the INPUT chain.
This reverts commit c004c18d90.
Showing {{EMAIL}}/{{USERNAME}} on the first subscription-body link only is
intentional, not an oversight in 876d55f2. Restoring the behaviour and the
tests that pin it.
Making the identity tokens configurable is the sanctioned route for the
operators asking for them on every link (#5935), rather than flipping the
default for everyone.
CLAUDE.md rules out // line comments in committed Go. The rationale they
carried is in the commit messages for each fix; doc comments that already
existed are kept, updated where the code they describe changed.
Also replaces reflect.Ptr with reflect.Pointer and rewrites the YAML keyword
alternation as a lookup table, both flagged by golangci-lint.
finalClients was a nil slice, so an inbound that has a clients key but whose
clients are all filtered out — disabled by an admin, or cut by the traffic
job for quota or expiry — was handed to xray-core as "clients": null.
The panel already treats a stored null client list as invalid data and
coerces it to [] at startup, and null is what reporters see in bin/config.json
when they go looking for a connectivity problem, which sends the diagnosis
after a serialization bug that is not there. Build the slice empty so the
same state serializes as [].
The reported inbound also needs the clients table to be in sync, which is a
separate question still open on the issue.
informTrafficToExternalAPI posted through the package-level fasthttp.Do,
which carries no read or write deadline. Run() is scheduled @every 5s under
cron.SkipIfStillRunning, so a receiver that accepts the connection and then
neither answers nor closes did not just delay one notification — it held the
job, and every following tick was skipped for the duration.
What stops with it is more than counters: AddTraffic runs autoRenewClients
and disableInvalidClients in the same call, so quota and expiry enforcement
stall too, and an over-quota client keeps transiting for the whole hang. The
online-client refresh and the websocket broadcasts sit later in the same tick.
Give the endpoint its own client with read/write deadlines and a DoTimeout
budget under the poll cadence, close the connection rather than pooling it
for a call this infrequent, and skip the POST outright when there is nothing
to report. Retries stay off: the payload carries per-tick deltas, so a resend
after a failed response leg would double-count on the receiver.
Verified against a listener that accepts and stalls: fasthttp.Do was still
blocked after 8s, the new client returns at its 3s budget.
A REALITY short-id like 2351e1 is valid hex, but as a bare YAML scalar the
resolution rules read it as the float 23510. mihomo hex-decodes the resulting
five-digit string, fails with "invalid REALITY short ID", and the whole
provider loads zero nodes — one proxy takes the entire subscription down.
The encoder quotes the forms it recognises (plain integers, hex, booleans)
but not the exponent-float form, and its own parser reads that token back as
a string, so nothing in a round-trip through it reveals the problem. Check
the values against the resolution rules instead, and force quotes on any
plain scalar that would resolve to a non-string.
Applied to every string in the document rather than to short-id alone: the
panel's own short-id generator emits random hex, and passwords, obfs-
passwords and pre-shared keys reach the output the same way. Unambiguous
values are untouched, so the document is otherwise byte-identical.
The existing Clash tests assert on the config map, never on the serialized
text, which is why this survived; the new tests assert on the output.
Real incident: an AmneziaWG inbound with RouteThroughXray enabled lost
all internet on that connection after a migration. Root-caused on the
live box -- iptables TPROXY counters were incrementing (packets
correctly redirected to 127.0.0.1:63110), but nothing was actually
listening there (ss showed nothing on that port) until a full
`systemctl restart x-ui`, after which the bridge came up immediately.
Xray-core's gRPC AddInbound reports success for a new sockopt.tproxy
inbound (internal/amneziawg's own Xray egress bridge is the only kind
this fork ever generates) but doesn't reliably bind a working listener
for it outside of process startup -- the bridge silently never comes
up, and RouteThroughXray traffic goes nowhere until the next full
restart happens to occur for an unrelated reason.
diffInbounds already has this exact defensive pattern for REALITY
inbounds ("a gRPC remove+add does not reliably rebuild the REALITY
authenticator"), just never extended to TPROXY, and only in the
already-existing-then-changed branch -- the "brand new inbound" branch
had no such guard at all, which is exactly the path a freshly-enabled
RouteThroughXray bridge takes. Added inboundUsesTproxy and wired it
into both branches.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
876d55f2 put EMAIL/USERNAME in the same first-link-only bucket as the usage
tokens, so a client attached to several inbounds got its email on whichever
inbound sorted first and bare inbound names on all the rest. With the shipped
default template ({{INBOUND}}-{{EMAIL}}|...) that makes every profile after
the first indistinguishable between clients — the point of the token.
The two are not alike: the usage block repeats identical numbers on every
link, while the identity is what tells one imported profile from another.
Restore identity on all body links and leave usage first-link-only.
Reported again in #6029 and #5659, which asked for the same revert.
client_global_traffics rows are keyed by (master_guid, email) and are only
ever overwritten by a push from that same master. A master that stops
pushing — decommissioned, reinstalled under a fresh GUID, or detached from
the node — therefore leaves its last snapshot behind permanently.
depletedClientsCond's cross-panel EXISTS branch matched any such row, so a
node kept comparing a client's quota against counters frozen weeks earlier.
Once they exceeded the quota the node disabled the client on every traffic
poll, and the node -> master enable merge latched that off on the master too,
where nothing sets it back. The reported symptom is exactly this: a client at
11 GB of a 24 GB quota, enabled on two nodes, disabled on the third, which
still held a 27-day-old row from a previous master reporting 30 GB.
Bound both the enforcement predicate and the display overlay to rows a master
refreshed within globalTrafficFreshWindow. Masters push every 30s, so a live
master is never affected; a master that is merely unreachable for a while
keeps enforcing for a full day before its numbers are set aside.
The one-way enable merge that makes such a disable permanent on the master is
deliberate (12d84c2a, #4917) and is left alone.
AmneziaWG is this fork's whole reason for existing, so requiring an
explicit flag/prompt answer to actually get it on every fresh install,
migration, or update was the wrong default -- confirmed today by a real
migration where it silently stayed uninstalled (interactive default was
"no", and non-interactive/unattended installs skipped it outright with
nothing to answer the prompt).
should_install_amneziawg now defaults to yes in both the interactive
prompt and the non-interactive/unattended path; XUI_INSTALL_AMNEZIAWG
still works as an explicit override in either direction (true/false),
so anyone who genuinely doesn't want the DKMS module + host-wide IP
forwarding can still opt out.
Also fixed the printed opt-out hint to show the actual working syntax
for setting the variable before a later re-run -- `VAR=val curl ... |
bash` only exports VAR into curl's environment, not bash's, since
they're separate processes joined by a pipe. Hit this exact gotcha
live: XUI_INSTALL_AMNEZIAWG=true prefixed onto the curl command was
silently ignored.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The sidebar's version badge and the subscription-templates docs link
still hardcoded MHSanaei/3x-ui. install.sh and every README already
point at Kuzz007/3x-ui; these were the last two functional (non-credit)
repo links left pointing upstream. DONATE_URL/DOCS_URL intentionally
left alone -- those point at the original author's own donation page
and docs site, which is deliberate per this fork's existing convention
of crediting/linking upstream for general (non-fork-specific) info.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
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>
GitHub Pages was never enabled in this fork's repo settings, and the
workflow's own comment ties it to docs.sanaei.dev, upstream's custom
domain, not something this fork has. Failed both times it ever ran
(including once before today, unrelated to any of this session's
other changes) with "Ensure GitHub Pages has been enabled". Docs CI
(docs-ci.yml, lint/build-check only, no deploy) is untouched.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Never used by this fork's own distribution (install.sh/x-ui.sh is the
only supported install path), AmneziaWG structurally can't run in the
Alpine-based image anyway, and Docker Hub publishing was failing on
every release for lack of configured credentials. Removed Dockerfile,
docker-compose.yml, DockerEntrypoint.sh, DockerInit.sh, .dockerignore,
and the docker.yml CI workflow; dropped the now-dead "Docker" README
subsection (all 7 languages), the "Docker image"/"Docker Compose"
options from the issue/PR templates, and corrected claude-bot.yml's
now-stale references to the deleted files, the never-actually-ours
ghcr.io/mhsanaei/3x-ui image, and (caught in passing) an already-stale
claim that Windows is a supported platform.
Left untouched: generic container-runtime adaptations that apply
regardless of image source (x-ui.sh's running-in-docker detection,
the virtual-interface-name filters, the DNS-over-container-network
note) and deploy/test/smoke-noninteractive.sh, which uses Docker only
as its own test sandbox, not as something this repo ships.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The push/pull_request paths filter didn't include .github/workflows/**,
so the previous commit (fixing the Audit step) never actually ran
through CI -- editing the workflow itself silently skipped the trigger.
A broken workflow-syntax change could merge untested the same way.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
npm audit --audit-level=high doesn't read code comments -- it was still
exiting 1 on the one remaining, already-investigated-as-unfixable advisory
(eslint-plugin-jsx-a11y's own pinned minimatch, GHSA-mh99-v99m-4gvg), so
every run on main has actually been red despite the prior commit's comment
calling it "known-accepted". That advisory lives entirely in
devDependencies (lint tooling, never shipped), so --omit=dev is the
correct signal for "doesn't apply to what we ship" -- verified locally,
0 vulnerabilities. Runtime dependencies are still audited at high severity.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The AmneziaWG share link (both the panel's per-client copy-link/QR and
the subscription endpoint) used an invented amneziawg://user@host:port
URI the real AmneziaVPN app can't parse -- it only recognizes its own
vpn:// scheme. Reverse-engineered the real app's import path (reading
amnezia-vpn/amnezia-client's own source) and confirmed it just needs
base64url(no padding) of a plain AmneziaWG .conf text -- no JSON schema
or qCompress framing to replicate, since qUncompress falls back to the
raw bytes for plain text and the parser reads a flat "Key = Value" bag
regardless of section. Both link generators now wrap the same .conf
text their own "download config" feature already produces correctly.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Downloading the raw subscription URL as a .txt file duplicates the
adjacent copy button for the vast majority of actual usage, and the
row already has a real, useful download further down (the per-client
config, e.g. AmneziaWG's own .conf). Left the JSON/Clash rows' download
buttons alone -- only the plain SUB row's was removed.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two real, forward-compatible fixes for npm audit's high-severity
advisories (not the downgrades npm audit fix --force offers):
- react-router (GHSA-qwww-vcr4-c8h2, RSC CSRF bypass): react-router-dom
is frozen at 7.18.1, pinning the vulnerable react-router@7.18.1 -- no
newer react-router-dom release exists pointing at the fixed line.
react-router itself has shipped the real fix at 8.3.0. Migrated the 9
files importing from react-router-dom (all using plain
createBrowserRouter/RouterProvider/useLocation/useNavigate/Outlet, no
RSC anywhere) to import from react-router directly instead.
- brace-expansion/minimatch (GHSA-mh99-v99m-4gvg): fixed for eslint's
own dependency chain (minimatch@10.2.5, used by eslint itself,
storybook, typescript-eslint, swagger-client) via a scoped
"minimatch@^10" override forcing brace-expansion to the now-published
5.0.8 patch -- within the range minimatch@10.2.5 already declares
wanting (^5.0.5), so this isn't a version-pin workaround, just
unblocking a patch release npm's resolver hadn't picked up.
One advisory remains genuinely unfixable from our side:
eslint-plugin-jsx-a11y pins minimatch@^3.1.2 (old major, never
patched); forcing it to the 10.x line via override breaks npm's own
dependency-tree validation (a real incompatibility, not just an npm
quirk), so this needs an eslint-plugin-jsx-a11y release bumping its own
minimatch. Lint-time only, no untrusted input reaches it -- ci.yml's
Audit step comment updated to reflect the new, smaller remaining scope.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This reverts commit 638458bd. Reordering new rules to the top by
creation-time was a blunt heuristic: it fixed a rule silently shadowed
by an unrestricted catch-all above it, but breaks the opposite,
equally valid setup -- e.g. a pre-existing "youtube -> interfaceA" rule
(no inboundTag restriction, meant to apply everywhere) followed by a
new "amneziawg -> interfaceB" rule. Under the reverted logic the newer,
broader-matching amneziawg rule would jump above the youtube rule and
shadow it for amneziawg-sourced traffic -- the exact same class of bug
this was meant to fix, just aimed the other way.
Xray's rule order encodes real admin intent (which rule should win for
overlapping traffic) that creation time can't safely stand in for.
Back to plain append-at-the-end; admins reorder manually with the
existing moveUp/moveDown controls, same as upstream's own behavior.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A new routing rule created through the Routing tab was always appended
to the end of the list. Xray matches rules top-to-bottom, first hit
wins, so any pre-existing broader rule (e.g. a block rule with no
inboundTag restriction) silently shadows a newly created, more specific
rule forever -- it looks saved and enabled but never actually fires.
Root-caused this from a report that an AmneziaWG inbound routed through
a real outbound lost all connectivity while routing it "direct" worked
fine: the AmneziaWG bridge's own TPROXY/interface code turned out to be
entirely fine (tunnel stayed up, client stayed online) -- the new rule
was just sitting below existing RU-IP/bittorrent/domain block rules
with no inbound restriction, so those matched first. Confirmed directly:
manually dragging the rule to sit right after the pinned api rule fixed
it. This isn't AmneziaWG-specific -- any protocol's newly added rule can
be shadowed the same way.
New rules now insert right after the pinned api rule (or at the very
top if it's absent) instead of at the end, so a newly created rule takes
effect by default; drag it lower afterward if a lower priority is
actually wanted.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Speed column showed "--" for AmneziaWG (and MTProto, which has the
identical gap) even while cumulative traffic totals were correct.
XrayTrafficJob drives live speed by querying xray-core's own stats API
and broadcasting the delta over websocket -- but AmneziaWG/MTProto never
run inside xray-core's own runtime inbounds, so they're invisible to
that API. Their own jobs already compute the same per-poll delta shape
(that's what keeps cumulative totals correct) but never broadcast it.
Reusing the existing "traffics"/"clientTraffics" broadcast would have
two real bugs: the frontend's existing scope/replace logic would let
each side clobber the other's speed on its next unrelated tick, and the
websocket hub's per-message-type throttle is keyed only by message type,
not caller -- since both sidecar jobs run on identical "@every 10s"
grids registered milliseconds apart, one would silently lose almost
every broadcast if both protocols were ever configured together.
Fixed with a small unthrottled broadcast path (both sidecar jobs are
already self-rate-limited by their own cron cadence) and protocol-
namespaced wire keys, tracked in their own frontend state and merged
into the existing inboundSpeed/clientSpeed only at read time -- so every
existing consumer needs zero changes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The fork's README only ever documented AmneziaWG as "what's different
here" -- the new geosite/geoip routing-rule autocomplete had nowhere
appropriate to go. Added a dedicated, appendable section for smaller
fork-specific improvements beyond AmneziaWG, and described the new
autocomplete feature there in all 7 languages. Future fork-specific
features should get a short entry here too, in every README.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Writing a routing rule today means remembering exact geosite:/geoip:/
ext:file:code syntax by hand, with no way to discover what categories
actually exist in the .dat files sitting in the bin folder -- including
custom ones like geosite_roscom.dat added via the Geodata auto-update
feature. The Domain/IP fields in the rule editor now suggest categories
as you type (e.g. "you" -> "geosite:youtube"), built live from whatever
.dat files are actually on disk, while still accepting any free-typed
value exactly as before.
Backend: GET /panel/api/xray/getGeodataCategories scans the bin folder,
parses matched geosite*/geoip*.dat files via xray-core's own exported
protobuf types, and formats each category as the exact rule syntax
xray-core's parser accepts -- geosite:/geoip: for the default files,
ext:<file>:<code> for anything else (there's no shorthand for custom
files). Cached in memory keyed by each file's (name, size, modTime) so
a request-time scan is cheap until a file actually changes.
Frontend: the Domain/IP inputs become Select "tags" fields fed by a new
useGeodataCategories() query hook, with an explicit substring filter so
"you" matches "geosite:youtube" (not a prefix). The array<->CSV-string
adapter lives entirely at the FormField transform boundary, so the
underlying form schema and saved rule shape are unchanged.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The published v3.5.0-awg.1 release was built before 814369da, so it still
ships the version-comparison bug that commit fixed. Cutting a new tag from
the corrected tree so the panel's own self-check is actually accurate for
anyone who updates.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
internal/config/version was never bumped past 3.5.0 when v3.5.0-awg.1 was
tagged, so a freshly-updated panel kept reporting its own version as the
plain upstream base. On top of that, parseVersionParts (Go and its
TypeScript mirror) required exactly 3 dot-separated numeric parts, so it
rejected the -awg.N suffix entirely and fell back to a raw string
inequality that reports "update available" any time the strings merely
differ -- which they always do here, even when already on the latest tag.
Bump the embedded version to 3.5.0-awg.1 and extend both parsers to treat
"-awg.N" as an optional 4th, lower-priority component (defaulting to 0 for
a plain tag), so e.g. 3.5.0-awg.2 > 3.5.0-awg.1 > 3.5.0 and a matching tag
compares equal instead of always looking outdated.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Neither is exploitable in this app (react-router CVE is RSC-only, we use
createBrowserRouter; jsx-a11y's brace-expansion chain only runs against our
own lint globs), and npm's suggested fixes are both downgrades with no real
forward patch published yet -- left as-is rather than trading a working
version for one that doesn't fix anything reachable here.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
v3.5.0-awg.1 is now tagged and promoted to Latest, so the plain
no-argument install.sh command resolves to something real for the first
time. Restructured Quick Start in all 7 READMEs to lead with it, followed
by the explicit-version and dev-channel variants (matching upstream's own
three-command Quick Start layout), replacing the now-outdated "this fork
only ever publishes dev-latest" note.
Investigated multi-node interaction with AmneziaWG: the master's own
reconcile (DesiredAmneziaWGInstances) and Xray config generation
(injectAmneziawgEgress, the GenXrayInboundConfig protocol skip) all
correctly filter on NodeID IS NULL, so a node-assigned AmneziaWG (or
MTProto) inbound would never be managed by the master. But nothing
stopped one from being created that way: NODE_ELIGIBLE_PROTOCOLS
(frontend/src/pages/inbounds/form/InboundFormModal.tsx) only hides the
node picker client-side -- a direct API call could set nodeId on an
AmneziaWG inbound, which every node then reconciles as an ordinary
local inbound (nodes run the identical binary, full cron suite
included), leaving it running unmanaged and untracked by the master's
own AmneziaWG bookkeeping.
Added isNodeEligibleProtocol (inbound_protocol.go), mirroring the
frontend's allowlist, and enforced it in both AddInbound (the actually
exploitable path -- nodeId comes straight from the request) and
UpdateInbound (defense in depth; NodeID is already restored from the
stored row there before this check, so it mainly guards against a
protocol change on an existing node-hosted inbound).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
getInboundsBySubId's SQL protocol allowlist never had 'amneziawg' added,
so every AmneziaWG client was silently excluded from all three
subscription formats (plain/individual links, JSON, Clash) and from the
Telegram bot's QR/individual-link buttons, which fetch through the same
path. genAmneziaWGLink itself was already fully implemented and already
wired into GetLink's dispatch switch -- it just never got a chance to
run. Same bug shape as the earlier TRACKED_PROTOCOLS frontend gap: a
hardcoded protocol list one entry short.
Found while investigating whether the Telegram bot needed AmneziaWG-
specific client-management code -- it doesn't (the bot itself is fully
protocol-agnostic), but this is the actual root cause of "can't share
an AmneziaWG client's config via the bot."
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Investigated: the image is Alpine-based, and AmneziaWG's own packaging
(DKMS module + amneziawg-tools) doesn't target Alpine/musl at all --
unlike the Debian/Ubuntu/Fedora/Arch paths install.sh already handles,
there's no package to apk add even with full host network/capabilities.
The panel already degrades gracefully (IsAwgInstalled() logs one warning
instead of retrying forever), so no code change is needed -- just made
the reason explicit at the point where a user would reach for cap_add/
network_mode to try to work around it.
- manager.go: serverAddress assumed subnetIp always ends in ".0"; a
base like "10.8.1.5" was used verbatim as the server's own address,
eventually colliding with peer allocation (which starts at .2
upward). Now derives the first host of the actual subnetIp/subnetCidr
network via netip, matching serverAddressV6's own approach. A /32
base (no host bits at all) is still used as-is. (Finding 12, partial
-- the /16 pool-widening half of this finding only exists on the
upstream-pr/amneziawg branch's merged client_wireguard.go, not here;
handled separately on that branch.)
- manager.go: ensureLocked carried the previous per-peer traffic
counters (`last`) forward even through a full restart, but
awg-quick down+up resets the kernel's own counters to zero -- the
next CollectTraffic computed a large negative delta (clamped to 0),
silently discarding real traffic. Extracted the decision into
nextTrafficBaseline: only a reload (syncconf) preserves the
baseline. (Finding 13)
- portfwd.go: exported ForwardedPortsInclude; inbound_amneziawg.go's
new checkForwardedPortsConflict uses it to reject, at save time, a
client's forwardedPorts that would DNAT the panel's own port or
another enabled inbound's port to the tunnel client --
portForwardLines has no destination restriction, so this collision
was previously silent. Wired into both the single-client update path
and the add-client path (client_inbound_apply.go), plus
normalizeAmneziaWGSettings for the whole-inbound save path. (Finding 14)
- inbound.go: InboundOption.AwgServer sent the whole ServerSettings
struct including PrivateKey to GetInboundOptions callers -- a
shared, admin-wide dropdown-filling endpoint the frontend's own
AwgServerOptionSchema never reads that field from. Redacted it
before assigning. (Finding 11)
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Addresses Finding 10 from the automated PR review: an always-on TPROXY
bridge makes every AmneziaWG tunnel hard-depend on Xray being up (all
traffic, including DNS, drops whenever Xray restarts), and forces a full
awg-quick down+up bounce on any client add/remove/re-IP, permanently
losing the syncconf fast path.
Adds ServerSettings.RouteThroughXray (off by default):
- defaultPostUpDown only emits the TPROXY/policy-route rules when it's
on; a plain AmneziaWG tunnel now has zero Xray dependency out of the
box.
- structuralFingerprint covers it (toggling it changes whether PostUp/
PostDown contain any TPROXY rules at all -- structural, not a
per-peer host-rule). hostRulesFingerprint's IPv4 tracking is now
itself conditional on RouteThroughXray (and IPv6 tracking on
IPv6Enabled), so an instance that never uses either keeps the
syncconf fast path for a plain peer re-IP.
- injectAmneziawgEgress only creates a bridge for inbounds that opted
in; checkAmneziawgEgressConflict (the Finding-7 fix) now parses each
candidate through InstanceFromInbound so a non-routed inbound's port
is correctly never treated as reserved.
- New inbound-level Switch in the AmneziaWG form; the actual outbound
decision is still made entirely through the panel's stock Routing
page, same as before -- only whether the bridge exists at all is now
a choice.
Translation keys added to all 13 locales in the same commit this time,
not backfilled later (see Finding 9's lesson).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Each is independently reproducible; fixed together since one review pass
found all of them.
- manager.go: the shared "ip rule add fwmark" policy route had no
existence check, so it duplicated in "ip rule show" on every interface
bounce (which hostRulesFingerprint forces on any client add/remove/
re-IP). Now checked via "ip rule list | grep -q ..." first. (Finding 2)
- params.go: ExternalInterface, IPv6ExternalInterface, and subnetIp/
subnetCidr are interpolated unescaped into a shell-executed PostUp/
PostDown line, but only obfuscation and the IPv6 subnet were validated
before save. Added ValidateInterfaceName (a strict charset+length
pattern) and ValidateSubnetIPv4 (netip.ParsePrefix), wired into
normalizeAmneziaWGSettings. (Finding 3)
- amneziawg_job.go: IsAwgInstalled() existed but nothing ever called it,
so a host without awg/awg-quick (the Docker image, RHEL, Arch, a failed
install.sh PPA step) logged a reconcile failure every 10s forever. Now
checked once an inbound actually needs it, warning once instead of
spamming. (Finding 4)
- client_inbound_apply.go: the WireGuard/AmneziaWG credential
carry-forward (added so a metadata-only client edit doesn't rotate
keys) never covered ForwardedPorts, so a partial edit -- an API call or
Telegram-bot toggle that omits the field -- silently wiped a client's
port-forwarding spec. Carried forward and written back the same way the
key fields already are. (Finding 5)
- manager.go: hostRulesFingerprint keyed each peer on its IPv4 address
only, and structuralFingerprint omitted IPv6Enabled/IPv6ExternalInterface
entirely, so an IPv6-only change could pick the syncconf reload path
(which never re-runs PostUp, leaving a stale NDP-proxy entry) or be a
complete no-op. Both fingerprints now cover the IPv6 fields. (Finding 6)
- port_conflict.go: the AmneziaWG egress bridge (injectAmneziawgEgress)
binds 127.0.0.1:63100+id with no collision check anywhere, since it
isn't a database row the ordinary port-conflict query can see -- same
blind spot the reserved Xray API port already has its own check for.
Added the equivalent check for the AmneziaWG bridge port. (Finding 7)
- install.sh: install_amneziawg ran unconditionally for every install/
update, building a DKMS kernel module and enabling host-wide IPv4/IPv6
forwarding whether or not the feature is ever used. Gated behind a new
should_install_amneziawg (XUI_INSTALL_AMNEZIAWG=true/false, or an
interactive y/N prompt defaulting to no). Also replaced the deprecated
apt-key adv with a dedicated keyring + signed-by= on the Debian branch,
and guarded its sources.list appends against duplication on a retried
install. (Finding 8)
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two gaps left an AmneziaWG interface stuck outside the manager's
control after a crash (kill -9/OOM/panic skips StopAll):
- ensureRestart's teardown was gated on the in-memory `exists` map,
which is always empty on a fresh process, so a survived interface
never got interfaceDown before interfaceUp tried `ip link add`
against a name the kernel already had — failing forever and never
populating m.ifaces, so traffic accounting silently stopped and the
inbound could never be removed. Gate on isInterfaceUp instead, which
checks real kernel state rather than this process's own bookkeeping.
- An inbound deleted from the database entirely while the panel was
down has no entry in `desired` ever again, so it never reaches the
per-id cleanup loop in Reconcile (which only walks m.ifaces). Add a
one-time sweepOrphansLocked scan of configDir, mirroring
mtproto.Manager.sweepOrphansLocked, that tears down and removes any
leftover interface/config not in the current desired set.
Found by the automated review on MHSanaei/3x-ui#6105 (Finding 1).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Fa/Ar/Zh/Es/Tr READMEs still described upstream verbatim (generic 3X-UI,
no AmneziaWG section, upstream install command, Windows listed as
supported, Contributing/donation/stargazers sections). Rewrite each to
match the already fork-ified README.md/README.ru_RU.md: fork framing,
the "What's different: AmneziaWG" section, dev-latest install command,
Windows removed from supported platforms, leaner developer-notes/credit
sections. Also restore the 7-language switcher line on all files now
that they're consistent again.
Only en-US/ru-RU ever got these 9 keys as each AmneziaWG feature landed
(the regenerate-obfuscation button, then Phase 2a's IPv6 fields, then
Phase 2b's per-client ForwardedPorts) — the other 11 locale files were
never backfilled, so i18next has been silently falling back to English
for all of them since Phase 1. Cosmetic-only (never broke anything),
but now closed for every shipped locale.
A second audit of the hardened workflow found the "no shell at all"
claim in resolve-conflicts was still false, by two routes that live
outside this file.
The job runs the model in the workspace right after `gh pr checkout`,
so for a fork pull request the working directory is attacker-controlled.
claude-code-action writes `enableAllProjectMcpServers = true` into
~/.claude/settings.json before starting Claude Code
(base-action/src/setup-claude-code-settings.ts), and the CLI honours a
project `.mcp.json` unless `strictMcpConfig` is set, which the action
never sets. A contributor branch carrying an `.mcp.json` therefore got
its command spawned at session start, with --allowedTools gating tool
calls but not server startup. The same tree also supplied CLAUDE.md and
.claude/ as project instructions. The job now passes
`--strict-mcp-config` and `--setting-sources user`, so nothing in the
merged tree configures the session.
The second route was `Edit` with no path scope, the only unscoped file
grant left. Editing `.git/config` to set `core.fsmonitor` or a
`credential.helper` gets a command run by the next step's git calls,
which hold CLAUDE_BOT_PAT, and the stray-file guard could never see it
because `git diff --name-only` lists tracked paths only. The merge step
now emits one `Edit(//<workspace>/<file>)` rule per conflicted path and
the model gets exactly those plus /tmp, with `.git/**` denied outright
and Bash, WebFetch, WebSearch and Task denied by name. Hooks are
disabled for the run (`core.hooksPath=/dev/null`, `commit --no-verify`).
Conflict handling gets three real gaps closed: modify/delete, rename and
both-added conflicts (git status DD/AU/UD/DU/AA/UA) leave no markers, so
they used to sail through the marker check and get committed unresolved
- they are now detected up front and handed back untouched; the marker
scan covers `=======` and `|||||||`, not just the outer pair; and after
staging, `git diff --diff-filter=U` must come back empty or nothing is
committed. A `=======` markdown underline of exactly seven characters in
a conflicted file will now hand the merge back rather than commit it,
which is the safe direction.
Smaller things the audit was right about:
- the mutating gh rules are prefix rules, so `Bash(gh issue close:*)`
reached every issue in the repository. They now carry the triggering
number: `Bash(gh issue close ${{ github.event.issue.number }}:*)`.
- `Write(//tmp/**)` is granted alongside `Edit(//tmp/**)`: the docs say a
Write(path) rule is never matched by the file checks, so the Edit rule
is what authorises it, but the tool has to be listed to exist at all.
Without this the model could not create /tmp/comment.md.
- the mention prompt lost its thread context when it moved to agent mode
and referred to "<number>" literally; it now gets repo, number, title
and whether the thread is a pull request.
- `git log`/`git show` are gone from mention: `--output=<file>` makes
them a file-write primitive.
- `@claude resolve pr conflicts` on a plain issue matched no job at all.
- the commit step gated on `skip != 'true'`, so it also ran when the
merge step died before writing any output; it now needs `skip ==
'false'`.
- bot-authored pull requests (dependabot opens three ecosystems' worth)
no longer start a review run that the action refuses to serve.
- resolve-conflicts drops to `contents: read`, since the push is the
PAT's job, and fails with a comment when that PAT is missing.
Making the jobs read-only in the previous commit was not enough: two of
the mechanisms that grant write access were invisible in the workflow
file itself.
Every job now passes a `prompt:` input. Without one, claude-code-action
picks tag mode for a mention, and src/modes/tag/index.ts then appends
`--permission-mode acceptEdits`, its own allowedTools including
`Bash(git commit:*)` and a push wrapper, and calls setupBranch. So the
mention job could edit files and commit them no matter what its own
allowedTools said, and its system prompt claiming otherwise was simply
wrong. A `prompt:` selects agent mode, which adds nothing. It also
removes tag mode's hidden requirement that the comment contain the
trigger phrase, which would have made resolve-conflicts a no-op for a
comment that said only "resolve pr conflicts".
resolve-conflicts no longer hands git to the model. `Bash(git:*)` is a
prefix rule, so it permitted `git push origin HEAD:main`, `--force`,
`git remote set-url`, and shell execution through `git config alias.x
'!sh -c ...'` - the action ships scripts/git-push.sh precisely because
`git push:*` allows `--receive-pack='sh -c ...'`. The job now splits in
three: a step checks out the PR branch, merges the base and collects the
conflicted paths; the model gets Read/Glob/Grep/Edit and no shell at all;
a final step verifies and pushes. That step refuses to commit if a
conflict marker survives, if the model wrote /tmp/ABORT, or if anything
outside the conflicted set was touched, and it stages those paths
individually instead of `git add -A`. The PAT is now written to the push
URL only in that last step, after the model's session has ended, instead
of sitting in .git/config while untrusted branch content is read.
The bare `Write` grant in the three answering jobs becomes
`Edit(//tmp/**)`, since only prose kept it out of the checkout and out of
$GITHUB_ACTION_PATH, whose scripts run after the model step. Each prompt
now says to fall back to an inline --body if the write is refused, so a
denied write cannot silently cost a reply. mention gains the transcript
upload and the no-reply guard the other jobs already have, keyed to the
triggering comment's timestamp.
Restores the header note about the 21000-character expression cap, with
the current block sizes.
PathGao