Eyevinn / moqlivemock / 37214498113

56%

main: 56%

Pull #154

github

fix(drm): encrypt each MoQ group as one decode sequence

Each chunk was encrypted with mp4.EncryptFragment, which makes a new FragmentEncryptor every time: a new AES key schedule, sample protector and IV copy, and for AV1 a new frame header decoder. An AV1 inter frame was therefore parsed without the reference frames before it, which decide parts of its header, so the clear range of a protected sample could end before or after the start of the tile data.

GenMoQGroup now creates one FragmentEncryptor per group, which starts at a random-access point, passes it to GenCMAFChunk and GenLocmafChunk, and stores its next IV on the track afterwards. A nil encryptor encrypts a chunk on its own, as before, for the bitrate measurement and the tests. The IV chain is unchanged, so every encrypted AVC, HEVC, AAC, Opus and AC-3 object is byte-identical. Encrypted tracks are a further 13% faster with 9% fewer allocations (geomean).

For AV1, mp4ff now finds the tile data where ffmpeg's trace_headers does for all 250 frames of video_400kbps_av1 with the sequence header of its keyframes, against 139 frames when each chunk was encrypted on its own. The av1C of the AV1 test assets holds a sequence header that differs from the one in the keyframes, and mp4ff parses with the av1C one, so the ranges of these assets stay wrong until that is fixed separately.

50 of 61 new or added lines in 2 files covered. (81.97%)

3399 of 6076 relevant lines covered (55.94%)

1288.58 hits per line