Skip to content

Fixable CVEs in ghcr.io/immich-app/immich-server (2 critical, 80 high) #979

Open

Description

@github-actions

The weekly CVE scan found fixable HIGH or CRITICAL vulnerabilities in ghcr.io/immich-app/immich-server. This body is rewritten by every scan, and the issue closes itself when a scan finds the pinned digest clean.

The fix is normally the Dependabot bump for this image. If upstream has not released one, the fixed versions below say what to wait for.

ghcr.io/immich-app/immich-server:v3.4.1

Stacks: sensitive. Digest: sha256:7c489add298fc2f92cda4c777de6462aa206877e04087b815f62c7664af0b53b.

Severity ID Package Installed Fixed in
CRITICAL CVE-2026-42216 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
CRITICAL CVE-2026-42217 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-109518 apt 3.0.3 3.0.4
HIGH CVE-2026-102276 brace-expansion 5.0.9 5.0.10, 3.0.7, 2.1.5, 1.1.19
HIGH CVE-2026-102278 brace-expansion 5.0.9 5.0.11, 3.0.8, 2.1.6, 1.1.20
HIGH CVE-2026-18446 fast-uri 3.1.4 2.4.4, 3.1.5, 4.1.2
HIGH CVE-2026-75899 fast-uri 3.1.4 2.4.5, 3.1.6, 4.1.3
HIGH CVE-2026-75931 fast-uri 3.1.4 2.4.5, 3.1.6, 4.1.3
HIGH CVE-2026-75975 fast-uri 3.1.4 2.4.5, 3.1.6, 4.1.3
HIGH CVE-2026-76172 fast-uri 3.1.4 2.4.5, 3.1.6, 4.1.3
HIGH CVE-2026-84292 fast-uri 3.1.4 2.4.6, 3.1.7, 4.1.4
HIGH CVE-2026-63209 github.com/klauspost/compress v1.18.6 1.18.7
HIGH CVE-2026-56854 golang.org/x/crypto v0.52.0 0.55.0
HIGH CVE-2026-46600 golang.org/x/net v0.55.0 0.56.0
HIGH CVE-2026-78661 golang.org/x/net v0.55.0 0.62.0
HIGH CVE-2026-78669 golang.org/x/net v0.55.0 0.60.0
HIGH CVE-2026-56851 golang.org/x/text v0.37.0 0.41.0
HIGH CVE-2026-56852 golang.org/x/text v0.37.0 0.39.0
HIGH CVE-2026-84304 google.golang.org/grpc v1.81.1 1.83.1
HIGH CVE-2026-84445 google.golang.org/grpc v1.81.1 1.82.2, 1.83.2, 1.84.0-dev.0.20260825144003-d5a41119e0e3, 1.85.0-dev.0.20260825072537-93e31b48545e
HIGH GHSA-hrxh-6v49-42gf google.golang.org/grpc v1.81.1 1.82.1
HIGH CVE-2026-109518 libapt-pkg7.0 3.0.3 3.0.4
HIGH CVE-2026-27622 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-34379 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-34543 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-34544 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-34588 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-40244 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-40250 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-41142 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-59184 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-59186 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-59189 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-59981 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-59982 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-68515 libopenexr-3-1-30 3.1.13-2 3.2.12-1~deb13u1
HIGH CVE-2026-113872 multer 2.4.0 2.5.0
HIGH CVE-2026-116304 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116306 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116307 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116309 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116314 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116315 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116316 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116317 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116319 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116320 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116323 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116325 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116326 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116328 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116330 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116331 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116332 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116335 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116336 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120139 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120141 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120142 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120143 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120145 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120147 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120148 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120149 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120150 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120151 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-120153 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-14661 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-18447 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-6460 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-6467 postgresql-client-17 17.11-1.pgdg13+2 17.12-0+deb13u1
HIGH CVE-2026-116576 systeminformation 5.33.15 5.33.16
HIGH CVE-2026-109383 undici 6.28.0 6.29.1, 7.30.1, 8.12.2
HIGH CVE-2026-109383 undici 8.11.2 6.29.1, 7.30.1, 8.12.2
HIGH CVE-2026-109829 undici 8.11.2 7.30.1, 8.12.2
HIGH CVE-2026-109869 undici 6.28.0 6.29.1, 7.30.1, 8.12.2
HIGH CVE-2026-109869 undici 8.11.2 6.29.1, 7.30.1, 8.12.2
HIGH CVE-2026-19534 undici 6.28.0 6.28.1, 7.29.1, 8.10.2
HIGH CVE-2026-87018 undici 6.28.0 6.29.1, 7.30.1, 8.12.2
HIGH CVE-2026-87018 undici 8.11.2 6.29.1, 7.30.1, 8.12.2
HIGH CVE-2026-76844 zlib1g 1:1.3.dfsg+really1.3.1-1+b1 1:1.3.dfsg+really1.3.1-1+deb13u1
HIGH CVE-2026-85091 zlib1g 1:1.3.dfsg+really1.3.1-1+b1 1:1.3.dfsg+really1.3.1-1+deb13u1

Written by .github/workflows/cve-scan.yml in this run.

Activity

  1. added
    securityA credential, an exposure, or a control that is weaker than recorded
    sensitivestacks/sensitive on trinity (VLAN 99)
    on Oct 7, 2026
  2. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (2 critical, 50 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 48 high)[/+] on Oct 10, 2026
  3. github-actions commented on Oct 10, 2026

    @github-actions
    Author

    New since the last scan: CVE-2026-106445, CVE-2026-106446.

  4. Gerrrt commented on Oct 10, 2026

    @Gerrrt
    Owner

    Triage, 2026-10-10: the bump we could take has merged but isn't deployed. What's left waits on Immich.

    • This body is stale. It describes v3.2.4, scanned before #1129 merged today and moved the pin to v3.3.1@sha256:db996e35…. trinity hasn't been redeployed yet: sensitive-immich-server and -machine-learning are still running v3.2.4 (docker ps, 2026-10-10).
    • v3.3.1 fixes 14 HIGH. I rescanned it with trivy 0.75.0 (--severity HIGH,CRITICAL --ignore-unfixed) and got 4 critical and 34 high, against 4 and 48 for v3.2.4. That matches chore(sensitive): Immich v3.2.4 -> v3.3.1, ahead of Sunday's Dependabot (#979) #1129's numbers. Debian's gzip, sqlite and openssl, plus npm's compression, multer and sharp, are fixed.
    • The criticals carry over. handlebars 4.7.9 (×2, fixed in 4.7.10), proxy-addr 2.0.7 (fixed in 2.0.8) and tar 7.5.16 (fixed in 7.5.19). Immich main's pnpm-lock.yaml still pins all three at those versions, so the next release won't fix them unless someone bumps them first.
    • The remaining HIGHs are all transitive npm dependencies in upstream's build (brace-expansion ×11, fast-uri ×6, tar, undici, js-yaml, nodemailer, ws, socket.io-parser, engine.io, grpc-js, ip-address, source-map-js), plus one libpcre2. Only an Immich release can fix them.
    • Exposure: the server is reached only through Caddy on the sensitive tier, as immich.matrix.elysium. I didn't trace which of these packages sit on a request path, so I'm not claiming any are unreachable.

    Next steps:

    1. Deploy chore(sensitive): Immich v3.2.4 -> v3.3.1, ahead of Sunday's Dependabot (#979) #1129 on trinity (make up on the sensitive stack, then make check-container-health). Sunday's scan then reads v3.3.1 and this body drops to 4 / 34.
    2. Wait for an Immich release that bumps handlebars, proxy-addr and tar. Dependabot will offer it.

    Revisit this when: an Immich release's lockfile moves those three packages, or one of these findings turns out to be reachable through Caddy.

  5. Gerrrt commented on Oct 10, 2026

    @Gerrrt
    Owner

    Deployed, 2026-10-10: trinity now runs Immich v3.3.1 (#1129). This is step 1 of the triage above.

    • The deploy: make up STACK=sensitive from main at a79e3e2. All 20 healthchecks pass, and the 13 single-file config mounts match the repo.
    • The containers: sensitive-immich-server and sensitive-immich-machine-learning are both on v3.3.1. The server log shows four migrations succeeded, then Immich Server is listening … [v3.3.1].
    • Through Caddy: https://immich.matrix.elysium/api/server/version returns 3.3.1, and the UI returns 200.

    The next weekly scan should rewrite this body to v3.3.1's 4 critical and 34 high. The issue stays open for step 2: an Immich release that bumps handlebars, proxy-addr and tar.

  6. Gerrrt commented on Oct 10, 2026

    @Gerrrt
    Owner

    Deployed and verified on trinity, 2026-10-10

    #1129 (squash 1c628f3) moved immich-server and immich-machine-learning to v3.3.1. Both were running on trinity by 14:36 UTC.

    What was checked:

    • Migrations. The server log shows the four v3.3 migrations succeeded, including PersonSharing. It then printed Immich Server is listening … [v3.3.1].
    • Container health. make check-container-health STACK=sensitive passed: 20 of 20 healthy. The ML container sits behind the ml profile, so that check skips it. Docker reports it healthy, and it answers pong to the server on immich-machine-learning:3003.
    • ML inference. The server sent /predict a text query using ViT-B-32__openai. It got back a 512-dimension CLIP embedding.
    • Vector search. That embedding was used for a nearest-neighbour query on smart_search. It ran through clip_index (VectorChord 0.4.3, unchanged) and returned ranked matches from the 601 embedded assets.
    • End to end. A smart search in the web UI, as the user, returned relevant results in a few seconds.

    This issue stays open. v3.3.1 cuts the server's fixable findings from 4 critical / 48 high to 4 / 34. What remains are transitive npm dependencies inside upstream's build. The weekly scan closes this once a pinned digest scans clean.

    🤖 Posted by Claude Code

  7. Gerrrt commented on Oct 10, 2026

    @Gerrrt
    Owner

    Deployed: immich-server v3.3.1 is running on trinity (2026-10-10, ~14:36Z), via #1129 (squash 1c628f3).

    Why this issue stays open. Scanned with the pinned trivy, --ignore-unfixed, HIGH/CRITICAL, on 2026-10-10:

    Critical High
    v3.2.4 (was deployed) 4 48
    v3.3.1 (deployed now) 4 34
    • Fixed: Debian's gzip, sqlite and openssl, and pcre2 in part; npm's compression, multer and sharp, and nodemailer in part.
    • Still present: proxy-addr, tar, brace-expansion, fast-uri, undici, ws, socket.io-parser, @grpc/grpc-js, js-yaml, ip-address and handlebars. These are transitive npm dependencies inside upstream's build. Only an Immich release can fix them.

    Next step: take the next Dependabot bump. The weekly scan closes this issue itself once the deployed digest scans clean.

  8. Gerrrt commented on Oct 10, 2026

    @Gerrrt
    Owner

    The four criticals, traced through Caddy, 2026-10-10

    This closes the gap the triage above left open: "I didn't trace which of these packages sit on a request path." I traced each one against the deployed immich-server:v3.3.1@sha256:db996e35…, using trivy 0.75.0's PkgPath and each advisory's precondition, then reading the image's own server/dist and node_modules/.pnpm. None of the four is reachable as deployed.

    CVE Package What the bug needs What Immich v3.3.1 does Reachable?
    CVE-2026-106445 handlebars 4.7.9 An attacker-controlled template rendered with allowProtoMethodsByDefault enabled Handlebars is imported in exactly one file, services/storage-template.service.js. It compiles with { knownHelpers: undefined, strict: true } and renders template(substitutions) with no runtime options, so the prototype-method access the bug needs is never turned on. The template is the admin-only storage template: PUT /api/system-config requires admin: true plus SystemConfigUpdate. No
    CVE-2026-106446 handlebars 4.7.9 Passing a pre-parsed AST object to compile() instead of a string The only compile() input is storageTemplate.template, which the config schema validates as z.string() (dtos/config.dto.js:283). An object is rejected before it reaches handlebars, and only an admin can set it. No
    CVE-2026-90711 proxy-addr 2.0.7 (via express 5.2.1) A trust entry written as IPv4-mapped IPv6 with an IPv4-sized prefix, e.g. ::ffff:10.0.0.0/8 app.set('trust proxy', ['loopback', ...IMMICH_TRUSTED_PROXIES]). We set IMMICH_TRUSTED_PROXIES: 172.28.99.2, Caddy's fixed address, as a plain IPv4 (stacks/sensitive/compose.yaml:880). Neither entry is the misparsed form. No, unless that env var is ever rewritten in ::ffff: notation
    CVE-2026-59873 tar 7.5.16 Extracting an attacker-supplied gzip In the image's pnpm store, the only package depending on tar@7 is node-gyp@13.0.0, the native-addon build tool. Nothing in server/dist imports tar. No, it's install-time tooling

    What would change this:

    • Immich starts using handlebars anywhere else, such as the email templates. Today those are z.string() config fields that handlebars doesn't render.
    • IMMICH_TRUSTED_PROXIES changes to an IPv6-mapped CIDR.
    • A server code path starts using tar at runtime, for example backup restore.

    Each would show up in the image's dist/ or in our compose diff. The remaining 34 HIGHs aren't covered by this trace.

    To reproduce:

    IMG=ghcr.io/immich-app/immich-server:v3.3.1@sha256:db996e352359771c6a3db121a0ed8761516b22f727fc092a0f46dcfef82c1bc1
    docker run --rm --entrypoint sh "$IMG" -c 'cd /usr/src/app/server; grep -rln handlebars dist; grep -n "handlebar.compile" dist/services/storage-template.service.js; grep -n "trust proxy" dist/app.common.js; ls -d node_modules/.pnpm/*/node_modules/tar | grep -v /tar@'

    🤖 Traced by Claude Code

  9. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 48 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 34 high)[/+] on Oct 10, 2026
  10. github-actions commented on Oct 10, 2026

    @github-actions
    Author

    New since the last scan: CVE-2026-90776.

  11. Gerrrt commented on Oct 10, 2026

    @Gerrrt
    Owner

    The HIGHs, traced against the live server, 2026-10-10: three are reachable before login

    This follows the criticals trace above, against the running sensitive-immich-server on trinity (v3.3.1@sha256:db996e35…). The 34 rows are 26 distinct (CVE, package) pairs. For each one I mapped runtime dependents in the image's pnpm store, then checked server/dist and the live config. Three are reachable before authentication, from any device on Hicks: denial of service only, with no data access. The rest aren't reachable.

    Reachable: the WebSocket stack (socket.io → engine.io → ws)

    Immich's gateway is @WebSocketGateway({ path: '/api/socket.io', transports: ['websocket'], cors: true }). Caddy passes immich.matrix.elysium straight to immich-server:2283, WebSocket upgrades included. Caddy publishes 0.0.0.0:443 on trinity, and docs/network.md passes vlan50 → 10.0.99.40:443, by segment rather than host, so any device on Hicks can open /api/socket.io. Immich's auth runs in handleConnection, after the transport and parser have already run. On failure it calls client.disconnect(), which in socket.io defaults to close = false: it leaves the namespace but doesn't close the engine.io connection.

    CVE Package What it does Why it's reachable here
    CVE-2026-48779 ws 8.20.1 (under engine.io) A peer sending floods of tiny fragments forces memory far beyond maxPayload, until the process is OOM-killed ws parses frames before anything else runs. No session or auth is involved.
    CVE-2026-69185 socket.io-parser 4.2.6 A packet announcing a large number of binary attachments makes the server buffer them, until OOM socket.io's per-connection decoder runs on every incoming packet, before namespace auth. A failed auth leaves the connection open.
    CVE-2026-102599 engine.io 6.6.8 An "upgrade" with a mismatched or omitted EIO and a crafted heartbeat can throw an uncaught exception, which crashes the process Even with transports: ['websocket']: a second WebSocket request carrying an existing sid goes onWebSocket → createTransport → client._maybeUpgrade, gated only by upgrading/upgraded, which a directly opened WebSocket session has neither of. verify() skips the transport check for upgrades. The sid comes from the attacker's own unauthenticated handshake. (This is from reading engine.io 6.6.8's server.js/socket.js in the image. I didn't exercise it, and won't against the live service.)

    Impact: Immich's API process (immich-api) dies or is OOM-killed. The container runs restart: unless-stopped under a mem_limit of 3 GiB. An OOM kill of the container restarts it. I haven't tested whether a crash of just the immich-api child is respawned by the parent immich process or takes the container down. In-flight uploads and jobs are interrupted, and a repeated attack keeps it down. Nothing is read or written.

    Upstream: Immich main's pnpm-lock.yaml (last changed 2026-10-07) still pins engine.io@6.6.8, socket.io-parser@4.2.6 and ws@8.20.1, so the next release won't fix these unless they're bumped first. The fixes are engine.io 6.6.10, socket.io-parser 4.2.7 and ws 8.21.0.

    Not reachable

    CVE(s) Package Why
    6 fast-uri 3.1.4 Only ajv ← conf ← react-email, the email-preview CLI. It isn't loaded by dist/.
    5 brace-expansion minimatch ← glob/readdir-glob ← archiver and react-email. Downloads use archiver('zip') with appended files, never archive.glob. The external-library crawler uses fast-glob (micromatch/braces, not brace-expansion), with exclusion patterns that only the admin sets. The /usr/local/lib copies belong to global npm.
    2 js-yaml 4.3.0 utils/config.js reads IMMICH_CONFIG_FILE, which isn't set. The other dependents are swagger and nest-commander (CLI).
    2 nodemailer 9.1.1 (addressparser) Only email.repository.js. SMTP isn't configured: system-config has no notifications.smtp.enabled, so it defaults to false (read-only query on immich-db).
    1 @grpc/grpc-js Only under OpenTelemetry's OTLP exporters, as a client. The bug is in server-side getAuthContext, and no IMMICH_TELEMETRY_*/OTEL_* is set.
    1 source-map-js postcss under tailwind for email templates. It needs an attacker-supplied indexed source map.
    2 tar, undici Under node-gyp (bcrypt/sharp install time), plus global npm's copies.
    1 ip-address Global npm's copy.
    1 libpcre2-8 (Debian) It is mapped by immich and immich-api, through glib (libglib/libgio/libgobject) under libvips. The CVE needs an attacker-controlled regex through PCRE2's JIT API, and glib's GRegex patterns come from library code, not image data.

    Options (for the decision, not taken)

    1. Accept until an Immich release ships the bumped deps. It's household-only and DoS-only, with auto-restart.
    2. Ask upstream to bump engine.io, socket.io-parser and ws. Immich's Renovate may already be about to.
    3. Narrow at Caddy, for example limiting concurrent connections or request rate on /api/socket.io. The web UI and the mobile app need it for live updates, so it can't be blocked outright.

    To reproduce:

    docker exec sensitive-immich-server sh -c 'cd /usr/src/app/server; grep -nE "transports|path:|disconnect\(" dist/repositories/websocket.repository.js; E=$(ls -d node_modules/.pnpm/engine.io@6.6.8*/node_modules/engine.io/build); sed -n 595,640p $E/server.js'
    grep -n -A3 'immich.matrix.elysium' stacks/sensitive/Caddyfile; grep -n 'trinity' docs/network.md

    🤖 Traced by Claude Code

  12. Gerrrt commented on Oct 11, 2026

    @Gerrrt
    Owner

    Option 2 taken, 2026-10-11: noted on Immich's open Renovate PR immich-app/immich#32128 (comment) that its lockfile still resolves engine.io 6.6.8 / socket.io-parser 4.2.6 although socket.io 4.8.4's ranges allow the patched 6.6.10 / 4.2.7 (which also brings ws 8.21). It's a public dependency-hygiene note with no exploit detail. It isn't a security report, since Immich's SECURITY.md wants a tested reproduction for those. When the Immich release that carries those versions lands, scan it before bumping, then the weekly scan closes this. Options 1 (accept meanwhile) and 3 (Caddy limits on /api/socket.io) still stand.

  13. Gerrrt commented on Oct 11, 2026

    @Gerrrt
    Owner

    Option 3 (limits at Caddy) checked, 2026-10-11: not available, and it wouldn't help.

    • No limiter in our Caddy. Every stack pins caddy:2.11.7@sha256:f2a1290d…, and trinity's sensitive-caddy runs that digest. caddy list-modules shows 135 standard modules and no rate or connection limiter. The only match for "limit"/"conn" is http.reverse_proxy.selection_policies.least_conn, a load-balancing policy. A limiter would need the third-party caddy-ratelimit module in a custom build, and this repository builds no images.
    • It wouldn't stop these three anyway. Each needs only one WebSocket connection to /api/socket.io: the ws fragment flood and the socket.io-parser attachment count work within a single connection, and the engine.io upgrade needs one handshake plus one upgrade. A per-client rate or connection limit still admits that connection. Blocking /api/socket.io outright would break live updates in the web UI and the mobile app.

    That leaves option 1 (accept: household-only, DoS-only, auto-restart) while option 2 (the note on immich-app/immich#32128) waits for an Immich release that resolves engine.io ≥ 6.6.10, socket.io-parser ≥ 4.2.7 and ws ≥ 8.21.0. Scan that release before bumping. Nothing in the repo changes.

  14. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 34 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 46 high)[/+] on Oct 19, 2026
  15. github-actions commented on Oct 19, 2026

    @github-actions
    Author
  16. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 46 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 52 high)[/+] on Oct 26, 2026
  17. github-actions commented on Oct 26, 2026

    @github-actions
    Author
  18. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (4 critical, 52 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (5 critical, 55 high)[/+] on Nov 2, 2026
  19. github-actions commented on Nov 2, 2026

    @github-actions
    Author

    New since the last scan: CVE-2026-109383, CVE-2026-116576, CVE-2026-87018.

  20. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (5 critical, 55 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (5 critical, 40 high)[/+] on Nov 9, 2026
  21. github-actions commented on Nov 9, 2026

    @github-actions
    Author

    New since the last scan: CVE-2026-109518, CVE-2026-76844, CVE-2026-85091.

  22. Gerrrt commented on Nov 9, 2026

    @Gerrrt
    Owner

    v3.3.6 triage, 2026-11-09: one more DoS before login. v3.4.0 fixes them all.

    Where things stand. trinity runs immich-server:v3.3.6@sha256:e1981983… (#1321), and it's Immich's latest stable release. There's no bump to take yet. The three WebSocket DoS rows from the 2026-10-10 trace (ws, socket.io-parser, engine.io) are still present and still reachable in v3.3.6.

    What's new since the v3.3.1 trace

    I traced these against the running container, by reading code and process maps only. Nothing was sent to the service.

    CVE Package Reachable? Why
    CVE-2026-111181 body-parser 2.3.0 Yes, before login (from reading the code; not tested) This is the same version and the same bug as Actual on #970. workers/api.js calls NestFactory.create(ApiModule, …) with Nest's default bodyParser, which mounts express's json/urlencoded parsers globally, before any guard. body-parser's contentstream() pipes a Content-Encoding: gzip body into zlib.createGunzip(). Caddy's immich.matrix.elysium block is a bare reverse_proxy.
    CVE-2026-113872 multer 2.4.0 After login multer runs in Nest interceptors (FileUploadInterceptor, and FileInterceptor on backup upload), and those run after the auth guard. Backup upload is admin-only. There are 0 shared links, so no anonymous upload path exists.
    CVE-2026-116576 systeminformation 5.31.17 No It's in the pnpm store only as fdir's dev dependency. Nothing in server/dist imports it.
    CVE-2026-109383, -109869, -19534, -87018 undici 6.28.0 No This copy is global npm's, under /usr/local/lib/node_modules/npm. Immich's own copy (6.26.0, already in the earlier trace) sits under node-gyp.
    CVE-2026-111497 libbrotli1 After login It's mapped by immich and immich-api through libvips (libjxl, freetype). Node's own gzip and brotli are bundled, so they don't use it. Attacker input reaches it only as uploaded media, which needs a login.
    CVE-2026-76844, -85091 zlib1g After login Same as libbrotli1: libvips's PNG/EXR/cairo decoders use it, and Node's bundled zlib doesn't.
    CVE-2026-109518 apt, libapt-pkg7.0 No Image build tooling.

    Impact, now known. When immich-api crashes, the parent immich process's onExit calls process.exit(exitCode). That stops the container, along with the microservices worker and any running jobs. restart: unless-stopped brings it back (restart count 0 so far). This answers the question the 2026-10-10 trace left open. Repeating the request keeps Immich down. Nothing is read or written.

    Upstream has fixed all of these in v3.4.0

    pnpm-lock.yaml at v3.4.0-rc.3 (2026-11-06), and on main, resolves:

    Package v3.3.6 v3.4.0-rc.3
    engine.io 6.6.8 6.6.11
    socket.io-parser 4.2.6 4.2.7
    ws (under engine.io) 8.20.1 8.22.1
    body-parser 2.3.0 2.4.0
    handlebars 4.7.9 4.7.10
    proxy-addr 2.0.7 2.0.8
    tar 7.5.16 7.5.22
    multer 2.4.0 2.5.0
    systeminformation 5.31.17 5.33.15

    ws 8.20.1 stays in the lockfile, but only under engine.io-client, which isn't on the server path. The Debian rows depend on v3.4.0's base image.

    Decision: wait for v3.4.0. No Caddy stopgap. A 415 matcher like #1343's was considered, but v3.4.0 stable looks days away.

    Next step: when the v3.4.0 Dependabot PR comes:

    1. Trivy the new digest before merging.
    2. Write Refs #979 in the PR body.
    3. Deploy with make up STACK=sensitive.
    4. Confirm in the running container that engine.io ≥ 6.6.10, socket.io-parser ≥ 4.2.7, ws ≥ 8.21.0 under engine.io, and body-parser ≥ 2.4.0.

    The weekly scan then closes this issue, or lists what's left.

    Revisit if v3.4.0 slips well past rc.3, or its stable lockfile differs from rc.3's. In either case, the #1343-style Caddy matcher becomes worth doing.

    To reproduce:

    docker exec sensitive-immich-server sh -c 'cd /usr/src/app/server; sed -n 1,20p dist/workers/api.js; grep -n "process.exit" dist/main.js; ls node_modules/.pnpm | grep -E "^(body-parser|engine.io|socket.io-parser|ws|undici|systeminformation)@"'
    gh api "repos/immich-app/immich/contents/pnpm-lock.yaml?ref=v3.4.0-rc.3" -H "Accept: application/vnd.github.raw" | grep -oE "^  (engine\.io|socket\.io-parser|body-parser|ws)@[0-9.]+" | sort -u

    🤖 Traced by Claude Code

  23. changed the title [-]Fixable CVEs in ghcr.io/immich-app/immich-server (5 critical, 40 high)[/-] [+]Fixable CVEs in ghcr.io/immich-app/immich-server (2 critical, 80 high)[/+] on Nov 16, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    securityA credential, an exposure, or a control that is weaker than recordedsensitivestacks/sensitive on trinity (VLAN 99)

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions