mirror of
https://github.com/navidrome/navidrome.git
synced 2026-11-19 15:23:36 +00:00
* fix(playlists): bound the M3U playlist upload size Uploading an M3U to POST /api/playlist had no limit on body size or entry count. Every entry, duplicates included, was kept in memory, copied into the playlist, saved and written back in the response, so one large upload could use enough memory to stop the server. The upload body is now capped at 32 MiB and the import at 50,000 entries. Both return 413 and nothing is saved. Read errors from the body are no longer dropped by the line scanner, so a cut-off body does not save a partial playlist. Playlists imported from disk by the scanner keep no entry cap, because real on-disk playlists can be larger (90k+ tracks). * refactor(playlists): map upload limit errors in writePlaylistError The 413 mapping for oversized or too-long M3U uploads now lives in writePlaylistError with the other playlist error statuses, so the upload handler has a single error path. The bare 0 passed to parseM3U for on-disk playlists is now the named noEntryLimit. * fix(playlists): run M3U upload imports one at a time One upload at the 50,000 entry limit can peak at a few hundred MB, so several sent together could still use up a small server's memory. Upload imports now take a single package-wide slot before reading the body. Other uploads wait for it, or return when their request is cancelled. Measured on a live server: three concurrent 50,000-entry uploads peaked at 1,030 MB before this change and 553 MB after.