Deluan Quintão 59810c3d59
feat(jellyfin): non-expiring, audience-scoped tokens revocable by password change (#6013)
* feat(auth): add per-user token_epoch column and bump method

* feat(auth): add aud and ep claims, omitted when zero

* feat(auth): add CreateAPIToken for non-expiring, audience-scoped tokens

* feat(auth): add CheckClaims for epoch and audience validation

* feat(jellyfin): issue non-expiring, jellyfin-scoped access tokens

* fix(subsonic): reject API-scoped and revoked tokens on the jwt path

* fix(server): reject API-scoped and revoked tokens on the native API

* fix(server): pin the token-subject guard and stop leaking test config

Adds a regression spec for the DevAutoLogin/ExtAuth guard in
tokenAllowed, switches its comparison to case-insensitive to match
the user lookup's own COLLATE NOCASE semantics, and restores Subsonic
JWT test config after each spec instead of leaking SessionTimeout.

* feat(request): add a token epoch holder for handler-to-middleware signalling

* refactor(server): write the refreshed JWT header after the handler runs

* feat(auth): revoke all tokens for a user when their password changes

* fix(server): restore Unwrap on the JWT refresh writer so SSE write deadlines apply

* test(auth): pin that non-session tokens reject API access tokens

* test(jellyfin): pin token scoping and epoch revocation end to end

Exercises auth.CreateAPIToken and CheckClaims against the real Jellyfin
router and SQLite DB: the minted token has no exp and is aud-scoped to
jellyfin, and bumping token_epoch through the real UserRepository revokes
an already-issued token on the next protected request.

* test(nativeapi): pin the token-epoch handoff through a real password-change request

Drive a self password change through the real Authenticator/JWTRefresher
chain and a real SQLite-backed userRepository, so the epoch handoff between
Put and the refreshed-token writer is verified end to end, not as two
separately-tested halves. Also fix tokenAllowed to read the enriched ctx it
was given instead of r.Context(), so its warning log carries the username.

* refactor(server): drop tokenAllowed's now-unused request parameter

Finding-2 already moved every use to ctx; r was dead weight. Also note
in the new nativeapi test why it must stay the package's only real-DB
spec: db.Db() is a process-wide singleton its cleanup closes for good.

* refactor(auth): remove duplication in claim decoding and token minting

* refactor(auth): group aud with the standard JWT claims

* refactor(auth): read aud with the standard-claim accessor pattern

* fix(log): redact every api_key spelling the Jellyfin API accepts

* fix(auth): bind session tokens to the user id, not just the username

* fix(auth): return the token epoch from the same atomic increment

* fix(auth): bump the token epoch in the same statement as the password write

* chore(auth): trim comments to the why-only budget
2026-08-22 20:36:24 -04:00
..

Jellyfin API

This package implements a subset of the Jellyfin REST API on top of Navidrome's existing library, users, playlists and scrobbling infrastructure. It lets Jellyfin-compatible clients (e.g. Finamp, jftui) browse and stream a Navidrome library without requiring a real Jellyfin server.

It is not a full Jellyfin server implementation: only the endpoints needed to browse a music library, stream audio, manage favorites/ratings for songs, albums, artists, and playlists, report playback, and manage playlists are implemented. Video, live TV, plugins, and Jellyfin's admin/dashboard APIs are out of scope.

Enabling

The Jellyfin API is disabled by default. Enable it via navidrome.toml:

[Jellyfin]
Enabled = true
# Optional: override the server name reported to clients (defaults to "Navidrome <version>")
ServerName = "My Music Server"
# Optional: usernames to show in the client login user-picker (default: none). See "Public user list".
ExposedPublicUsers = "alice, bob"
# Optional: max collection responses streaming at once (default: half the DB connection pool,
# min 2). Each streaming response holds a DB connection for its whole duration; excess requests
# queue rather than fail.
MaxConcurrentStreams = 4

or via environment variables:

ND_JELLYFIN_ENABLED=true
ND_JELLYFIN_SERVERNAME="My Music Server"
ND_JELLYFIN_EXPOSEDPUBLICUSERS="alice,bob"

Once enabled, the API is mounted at:

http://<host>:<port>/jellyfin

All the paths below are relative to that base URL (e.g. System/Info/Public means http://localhost:4533/jellyfin/System/Info/Public). Routes are matched case-insensitively, since real Jellyfin clients (and jellyfin-apiclient-python) send mixed-case paths.

Authentication

Jellyfin clients authenticate with POST /Users/AuthenticateByName using the user's Navidrome username/password, and get back an AccessToken (a Navidrome JWT). That token is then sent on every subsequent request as the X-Emby-Token header (or embedded in the X-Emby-Authorization/Authorization header's Token="..." field, or as an api_key/ApiKey query param — all forms are accepted, matching what different clients do).

POST /Users/AuthenticateByName is rate-limited per IP with the same limiter as the native /auth/login (AuthRequestLimit/AuthWindowLength), since it's an unauthenticated brute-force surface.

Access tokens do not expire, matching real Jellyfin. They are revoked by a password change, which bumps the user's token epoch.

Public user list (login picker)

GET /Users/Public lets a client render a login user-picker (tap a user, then just type the password) instead of a blank username field. It's unauthenticated, so by default it exposes no users. Set Jellyfin.ExposedPublicUsers to a comma-separated list of usernames to advertise:

[Jellyfin]
ExposedPublicUsers = "alice, bob"

Only the named users are listed (never the full user table), resolved live per request; a configured name that doesn't exist is skipped and logged at Warn. Each entry is a minimal DTO (Name, Id) with no Policy/Configuration, so admin status isn't leaked to unauthenticated callers, and no avatar (PrimaryImageTag omitted — Navidrome has no per-user profile images).

Players and sessions

Every authenticated request registers (or refreshes) the calling device as a Navidrome player, mirroring Subsonic's getPlayer — so a Jellyfin client shows up in the players list (and scrobbling has a player) as soon as it makes any authenticated call, not only when it reports playback. The player id is the device id from X-Emby-Authorization (DeviceId="..."); the player name is Client [Device]. Those field values are URL-decoded, since some clients percent-encode them (Jellify sends Device="Pixel%208%20Pro", Finamp sends it raw). A request that carries no client/device info (e.g. the GET socket handshake, which authenticates via ?api_key= only) is skipped, so it doesn't create a nameless player.

Multi-library behavior

Jellyfin has no native concept of multiple music libraries the way Navidrome does, so each Navidrome library the current user can access is exposed as its own top-level Jellyfin "CollectionFolder" view (GET /UserViews), instead of merging every library into a single view. Browsing (/Items), artists, and the "Latest" list are all scoped to the libraries the authenticated user has access to; a library (or item within it) the user cannot access returns 404, never 403, so ids can't be used as an existence oracle.

Browsing filters

GET /Items accepts the filter params clients use to build screens: ParentId (a library view id for scoping, an artist id when browsing into an artist's albums, or an album id when browsing into an album's tracks); AlbumArtistIds/ArtistIds/contributingArtistIds (an artist's albums or tracks — Finamp's artist screen sends these alongside ParentId=<libraryId>); AlbumIds (an album's tracks — Feishin fetches them this way instead of ParentId); GenreIds (a genre's albums or tracks — Finamp's genre screen sends it the same way; /Artists/AlbumArtists and MusicArtist queries accept it too, matching artists credited on an album of that genre); SearchTerm; Filters (IsFavorite, IsFavoriteOrLikes, IsPlayed, IsUnplayed) and the standalone isFavorite/isPlayed booleans it can also be expressed as — Filters wins when both are sent, as in Jellyfin; Likes, Dislikes, IsFolder, IsNotFolder and IsResumable have no Navidrome equivalent and are ignored; SortBy/SortOrder (every recognized key is applied in order, so secondary keys break ties; unrecognized keys are skipped, and Random always sorts alone); StartIndex/Limit; and Ids (batch fetch by id). Recursive=false with a library ParentId returns direct children only (no tracks — no track is a library's direct child).

Implemented endpoints

Area Endpoints
Handshake / system GET System/Info/Public, GET System/Info (authenticated), GET/POST System/Ping, GET System/Endpoint (authenticated), GET QuickConnect/Enabled
Auth POST Users/AuthenticateByName, GET Users/Public
Users GET UserViews, GET Users/{userId}/Views, GET Users/Me, GET Users/{userId}
Browsing GET Items, GET Users/{userId}/Items, GET Items/{itemId}, GET Users/{userId}/Items/{itemId}, GET Users/{userId}/Items/Latest, DELETE Items/{itemId} (playlists only)
Artists / genres GET Artists, GET Artists/AlbumArtists, GET Genres, GET MusicGenres
Similar / mixes GET Artists/{itemId}/Similar, GET Items/{itemId}/Similar, GET Items/{itemId}/InstantMix
Images GET Items/{itemId}/Images/{type}[/{index}] (public), POST/DELETE Items/{itemId}/Images/{type} (playlist cover, authenticated)
Favorites / ratings for songs, albums, artists, and playlists POST/DELETE UserFavoriteItems/{itemId}, POST/DELETE Users/{userId}/FavoriteItems/{itemId}, POST/DELETE Users/{userId}/Items/{itemId}/Rating, GET UserItems/{itemId}/UserData, GET Users/{userId}/Items/{itemId}/UserData
Streaming GET Audio/{itemId}/stream[.{container}], GET Audio/{itemId}/universal, GET Audio/{itemId}/main.m3u8, GET Items/{itemId}/File, GET Items/{itemId}/Download, GET/POST Items/{itemId}/PlaybackInfo
Lyrics GET Audio/{itemId}/Lyrics
Playback reporting POST Sessions/Playing, POST Sessions/Playing/Progress, POST Sessions/Playing/Stopped, POST Sessions/Capabilities[/Full]
Playlists POST Playlists, GET Playlists/{playlistId}, POST Playlists/{playlistId} (rename / visibility / replace tracks), GET Playlists/{playlistId}/Items, POST/DELETE Playlists/{playlistId}/Items, GET Playlists/{playlistId}/Users[/{userId}]
Real-time GET socket (WebSocket; keeps clients like Finamp from 404-loop-reconnecting)
AudioMuse-AI (see below) GET AudioMuseAI/info, GET AudioMuseAI/health, GET AudioMuseAI/similar_tracks, GET AudioMuseAI/find_path

Any other path returns a 404 with a {} JSON body, and is logged server-side at Debug level as Jellyfin API: unhandled route (method + path). If a client you're testing needs an endpoint that isn't in the table above, check the server logs for these lines to see exactly what it's requesting.

Playlist management

Playlists are the main writable surface of this API:

  • Container expansion. When creating (POST Playlists), adding to (POST Playlists/{id}/Items) or replacing (POST Playlists/{id}) a playlist, the Ids may contain containers — album, artist or playlist ids — not just song ids. Each is expanded into its tracks (in order) before the write, matching how Jellyfin clients populate these lists. A bare song id passes through.
  • Id list encoding. POST/DELETE Playlists/{id}/Items accept the id list both ways clients spell it: repeated params (ids=X&ids=Y, how Jellify's @jellyfin/sdk serializes arrays) and a single comma-separated value (ids=X,Y, Finamp). Reading only the first value would add just one track of an expanded album.
  • Update (POST Playlists/{id}): with Ids present, the track list is replaced (Finamp uses this for reordering) — an explicit empty Ids ([]) clears the playlist, while an omitted Ids leaves the tracks untouched and only updates Name/IsPublic. IsPublic maps to Navidrome's Public flag, surfaced to clients as OpenAccess on GET Playlists/{id}.
  • Cover art: POST Items/{id}/Images/Primary uploads a playlist cover (raw or base64 body, JPEG/PNG/WebP/GIF detected by magic number, extension from Content-Type); DELETE removes it. Only playlists are writable through this API — album/artist covers come from tag/sidecar scanning, so a non-playlist id returns 501. Uploads honor the same gates as the native endpoint: they're bounded by MaxImageUploadSize and require EnableArtworkUpload for non-admins.
  • PlaylistItemId: GET Playlists/{id}/Items tags each entry with PlaylistItemId (the playlist-track row id, distinct from the song id) so a client can echo it back via DELETE Playlists/{id}/Items?EntryIds=... to remove one occurrence of a song that appears more than once in the same playlist.

Ownership is enforced by core/playlists: a non-owner editing/deleting a playlist gets 403 if it is visible to them (public) or 404 if it is not (private) — the API never reveals that someone else's private playlist exists.

Images

The GET Items/{itemId}/Images/{type} route is intentionally public (artwork isn't sensitive, matching Jellyfin's lenient image handling), so it carries no authenticated user. Artwork is therefore resolved under an elevated admin context — the same approach core/artwork's cache warmer uses — so user-scoped items like private playlists still resolve their cover instead of falling back to the placeholder. Album, artist, media-file and playlist ids are all resolved to their Navidrome ArtworkID.

Item ids are GUIDs

Jellyfin item ids are GUIDs, serialized as 32 lowercase hex chars with no dashes (Guid.ToString("N")). Navidrome ids are canonical 22-char base62 encodings of a 128-bit value, so dto.EncodeID/dto.DecodeID map between the two via model/id — losslessly except for the ~2⁻⁹⁶ chance an id's 128-bit value falls in the reserved space below (leading 12 bytes all zero).

Three emitted ids aren't 128-bit values: integer library ids, the synthetic playlists folder, and PlaylistItemId (a playlist entry positionplaylist_tracks.id is an integer column). They use a reserved GUID space — 12 zero bytes, a non-zero kind tag, a 24-bit payload — so library 1 is 00000000000000000000000001000001. The tag is never zero, because Jellyfin serializes the all-zero GUID as null.

DecodeID accepts dashed and uppercase GUIDs (Jellyfin's Guid.Parse does) and returns ok=false for anything malformed — including "" — which handlers surface as a 404.

The wire format must stay GUID-shaped for this reason: Finamp's saved-queue persistence bit-packs each item id into exactly 16 bytes (packIds() in lib/models/finamp_models.dart), so a 32-hex GUID round-trips exactly, whereas a longer id would be silently truncated.

Streaming and transcoding

The stream endpoints reuse the same transcode-decision pipeline as the Subsonic /stream endpoint:

  • GET Audio/{id}/stream[.{container}] / universal — the target format comes from the .{container} path suffix, the container param, or (when neither is present) audioCodec. audioBitRate/maxStreamingBitrate are bits/sec, per Jellyfin convention. static=true forces direct play (raw), never a transcode.
  • GET Items/{id}/File / Download — always the original file bytes, matching real Jellyfin. Finamp plays through File when its transcoding setting is off, so an undecodable format (e.g. DSF) can't be rescued server-side on this path.
  • GET Audio/{id}/main.m3u8 — the endpoint Finamp plays through when its transcoding setting is on. Implemented as a single-segment HLS VOD playlist whose one segment is the progressive transcode endpoint above, so the whole pipeline (decision, cache, forced transcoding) is reused. Segment codec honors audioCodec but is limited to what HLS packed-audio can carry (aac, mp3); anything else falls back to aac. Seeking re-reads from the start, like Subsonic transcoded streams.
  • Server-forced transcoding. A format/bitrate configured on the registered player (Settings → Players) is applied to stream, universal and main.m3u8 — same override semantics as Subsonic. File/Download stay raw. For HLS clients, force aac or mp3; other formats are advertised and served but packed-audio players won't decode them.

AudioMuse-AI compatible endpoints

Compatibility shim for Jellyfin front-ends that integrate AudioMuse-AI — e.g. Symfonium can use these endpoints for sonic mixes when connected as a Jellyfin client. Backed natively by Navidrome's core/sonic engine (the SonicSimilarity plugin capability) — no external AudioMuse-AI backend or proxy is involved. The endpoints are gated on a SonicSimilarity plugin being loaded, like the Subsonic sonicSimilarity OpenSubsonic extension.

  • GET /AudioMuseAI/info — returns {"Version": <navidrome version>, "AvailableEndpoints": [...]} (200). AvailableEndpoints lists the endpoints below only when a provider is loaded; otherwise it is empty.
  • GET /AudioMuseAI/health — liveness probe: 200 with an empty body when a provider is loaded, else 404.
  • GET /AudioMuseAI/similar_tracks?item_id=<id>&n=10&eliminate_duplicates=true — 404 when no provider is loaded; otherwise a JSON array of {author, distance, item_id, title} (200; [] when there is no match or no item_id). eliminate_duplicates (default true) limits results to one track per artist.
  • GET /AudioMuseAI/find_path?start_song_id=<id>&end_song_id=<id>&max_steps=25 — 404 when no provider is loaded; otherwise {"path": [{author, item_id, title, tempo?}], "total_distance": <float>} (200), or 400 with start_song_id and end_song_id are required. when either id is missing.

item_id/start_song_id/end_song_id are the GUID-form ids Navidrome hands Jellyfin clients. tempo comes from the track's BPM when known; the richer AudioMuse per-track features (energy, key, mood_vector, scale, other_features) are not provided. In multi-library setups, find_path's path and total_distance only reflect hops through tracks in libraries the caller can access, since hops through inaccessible libraries are filtered out of the result.

curl walkthrough

This mirrors the sequence a real client (e.g. Finamp) follows: handshake, login, browse the library hierarchy, fetch playback info, stream, favorite, report playback, and manage a playlist.

BASE=http://localhost:4533/jellyfin

# 1. Handshake (no auth required)
curl -s "$BASE/System/Info/Public" | jq .

# 2. Login - capture the AccessToken
TOKEN=$(curl -s -X POST "$BASE/Users/AuthenticateByName" \
  -H 'Content-Type: application/json' \
  -d '{"Username":"admin","Pw":"password"}' | jq -r .AccessToken)

AUTH=(-H "X-Emby-Token: $TOKEN")

# 3. List the user's views (one per accessible library)
curl -s "${AUTH[@]}" "$BASE/UserViews" | jq .

# 4. Browse artists
curl -s "${AUTH[@]}" "$BASE/Items?IncludeItemTypes=MusicArtist" | jq .
ARTIST_ID=$(curl -s "${AUTH[@]}" "$BASE/Items?IncludeItemTypes=MusicArtist&Limit=1" | jq -r '.Items[0].Id')

# 5. Drill into that artist's albums (ParentId with no IncludeItemTypes defaults to MusicAlbum)
ALBUM_ID=$(curl -s "${AUTH[@]}" "$BASE/Items?ParentId=$ARTIST_ID" | jq -r '.Items[0].Id')

# 6. List the album's songs
USER_ID=$(curl -s "${AUTH[@]}" "$BASE/Users/Me" | jq -r .Id)
SONG_ID=$(curl -s "${AUTH[@]}" "$BASE/Users/$USER_ID/Items?ParentId=$ALBUM_ID&IncludeItemTypes=Audio" \
  | jq -r '.Items[0].Id')

# 7. Ask for playback info, then stream the song
curl -s -X POST "${AUTH[@]}" "$BASE/Items/$SONG_ID/PlaybackInfo" | jq .
curl -s "${AUTH[@]}" "$BASE/Audio/$SONG_ID/stream" -o /tmp/song.audio

# 8. Favorite the song
curl -s -X POST "${AUTH[@]}" "$BASE/Users/$USER_ID/FavoriteItems/$SONG_ID" | jq .

# 9. Report playback start/stop (also drives scrobbling)
curl -s -X POST "${AUTH[@]}" -H 'Content-Type: application/json' \
  -d "{\"ItemId\":\"$SONG_ID\",\"PositionTicks\":0}" "$BASE/Sessions/Playing"
curl -s -X POST "${AUTH[@]}" -H 'Content-Type: application/json' \
  -d "{\"ItemId\":\"$SONG_ID\",\"PositionTicks\":1200000000}" "$BASE/Sessions/Playing/Stopped"

# 10. Create a playlist from a whole album (the album id is expanded to its tracks)
PLAYLIST_ID=$(curl -s -X POST "${AUTH[@]}" -H 'Content-Type: application/json' \
  -d "{\"Name\":\"My Playlist\",\"Ids\":[\"$ALBUM_ID\"]}" "$BASE/Playlists" | jq -r .Id)

# 11. Make it public, then remove one entry
curl -s -X POST "${AUTH[@]}" -H 'Content-Type: application/json' \
  -d '{"IsPublic":true}' "$BASE/Playlists/$PLAYLIST_ID"
ENTRY_ID=$(curl -s "${AUTH[@]}" "$BASE/Playlists/$PLAYLIST_ID/Items" | jq -r '.Items[0].PlaylistItemId')
curl -s -X DELETE "${AUTH[@]}" "$BASE/Playlists/$PLAYLIST_ID/Items?EntryIds=$ENTRY_ID"

# 12. Delete the playlist
curl -s -X DELETE "${AUTH[@]}" "$BASE/Items/$PLAYLIST_ID"

Testing

Handler-level unit tests live alongside each file (*_test.go). A full end-to-end suite in e2e/ exercises every endpoint through the real router against a real SQLite database and real repositories (only artwork/streaming/ffmpeg are stubbed), with per-Describe snapshot isolation — mirroring the Subsonic server/subsonic/e2e suite. Run it with:

make test PKG=./server/jellyfin/...

Known limitations

  • Genres are global. GET Genres/MusicGenres is not scoped to the current user's libraries (genre tags aren't per-library entities in Navidrome's model).
  • Artist item-access relies on list-time scoping. Unlike albums and songs (which each belong to exactly one library and are checked against user.HasLibraryAccess on every fetch), an artist can have content across multiple libraries via library_artist, so there's no single library id to gate a direct GET Items/{artistId} or favorite/rating call against. Access control for artists is enforced by scoping the Artists/Items?IncludeItemTypes=MusicArtist list to the user's libraries, plus the persistence layer's own defense-in-depth; a client that already has an artist id from elsewhere is not re-checked against library membership.
  • Blurhashes are synthetic, not computed from the artwork (follow-up). ImageBlurHashes is populated by dto/blurhash.go, which derives a well-formed 1-component (solid color) blurhash by hashing the item id — it never looks at the actual image. Real Jellyfin computes a multi-component blurhash from the cover's pixels (downscaled to 128×128) once at scan time and stores it per image, so its placeholder approximates the art. Ours satisfies the protocol (Finamp gets a valid value to use as a de-dup key and a placeholder, no missing-blurhash warning) but renders as a flat color while art loads. A proper implementation would compute the real blurhash in the core/artwork pipeline (where the image is already decoded), cache it keyed like the artwork, and have the mappers read it — keeping the synthetic value as a fallback for art that hasn't been rendered yet.
  • The WebSocket only keep-alives; it pushes no events (follow-up). GET socket sends a ForceKeepAlive and answers KeepAlive pings so real-time clients (Finamp) settle into a working session instead of 404-loop-reconnecting, but it never pushes anything. A follow-up would broadcast real session/playstate and library-change events over it (via server/events), mirroring Jellyfin's session messages.
  • Lyrics. GET Audio/{id}/Lyrics serves the main lyric track as a LyricDto (Start in 100ns ticks, word-level Cues when present), resolved through the full core/lyrics pipeline (embedded, .lrc sidecars, plugins per LyricsPriority) behind a 5-minute TTL cache that also caches misses — Jellify fetches for every played track, Feishin per song change, so lyric-less tracks are the hot path. No lyrics → 404 (never an empty 200), which all three clients degrade gracefully. Finamp gates its lyrics view on a Lyric MediaStream (not HasLyrics, which is just a list badge): browse lists advertise it from embedded lyrics only (the "[]" sentinel check — the column is never "" post-scan), while PlaybackInfo runs the full pipeline per track so sidecar/plugin lyrics also light up. Feishin additionally requires server version ≥ 10.9 — the reason jellyfinVersion is 10.9.11. Concurrent misses on the same track share one pipeline invocation (SimpleCache.GetWithLoader is singleflighted), and the load runs detached from the request context with a one-minute bound, so a cancelled request or hung plugin can't fail or pin the load for other waiters. Follow-up: tracks whose only lyrics are sidecar/plugin-sourced show no HasLyrics badge in lists (request-time sources can't be known at list time without per-row I/O).