2 * Matroska file demuxer
3 * Copyright (c) 2003-2008 The FFmpeg Project
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * Matroska file demuxer
25 * @author Ronald Bultje <rbultje@ronald.bitfreak.net>
26 * @author with a little help from Moritz Bunkus <moritz@bunkus.org>
27 * @author totally reworked by Aurelien Jacobs <aurel@gnuage.org>
28 * @see specs available on the Matroska project page: http://www.matroska.org/
42 #include "libavutil/avstring.h"
43 #include "libavutil/base64.h"
44 #include "libavutil/dict.h"
45 #include "libavutil/intfloat.h"
46 #include "libavutil/intreadwrite.h"
47 #include "libavutil/lzo.h"
48 #include "libavutil/mathematics.h"
50 #include "libavcodec/bytestream.h"
51 #include "libavcodec/flac.h"
52 #include "libavcodec/mpeg4audio.h"
55 #include "avio_internal.h"
60 /* For ff_codec_get_id(). */
78 typedef const struct EbmlSyntax {
87 const struct EbmlSyntax *n;
107 uint64_t doctype_version;
113 } MatroskaTrackCompression;
118 } MatroskaTrackEncryption;
123 MatroskaTrackCompression compression;
124 MatroskaTrackEncryption encryption;
125 } MatroskaTrackEncoding;
129 uint64_t display_width;
130 uint64_t display_height;
131 uint64_t pixel_width;
132 uint64_t pixel_height;
134 uint64_t stereo_mode;
136 } MatroskaTrackVideo;
140 double out_samplerate;
144 /* real audio header (extracted from extradata) */
151 uint64_t buf_timecode;
153 } MatroskaTrackAudio;
158 } MatroskaTrackPlane;
161 EbmlList combine_planes;
162 } MatroskaTrackOperation;
173 uint64_t default_duration;
174 uint64_t flag_default;
175 uint64_t flag_forced;
176 uint64_t seek_preroll;
177 MatroskaTrackVideo video;
178 MatroskaTrackAudio audio;
179 MatroskaTrackOperation operation;
181 uint64_t codec_delay;
184 int64_t end_timecode;
186 uint64_t max_block_additional_id;
196 } MatroskaAttachment;
234 MatroskaTagTarget target;
254 AVFormatContext *ctx;
258 MatroskaLevel levels[EBML_MAX_DEPTH];
268 EbmlList attachments;
274 /* byte position of the segment inside the stream */
275 int64_t segment_start;
277 /* the packet queue */
284 /* What to skip before effectively reading a packet. */
285 int skip_to_keyframe;
286 uint64_t skip_to_timecode;
288 /* File has a CUES element, but we defer parsing until it is needed. */
289 int cues_parsing_deferred;
291 int current_cluster_num_blocks;
292 int64_t current_cluster_pos;
293 MatroskaCluster current_cluster;
295 /* File has SSA subtitles which prevent incremental cluster parsing. */
297 } MatroskaDemuxContext;
304 uint64_t additional_id;
306 int64_t discard_padding;
309 static EbmlSyntax ebml_header[] = {
310 { EBML_ID_EBMLREADVERSION, EBML_UINT, 0, offsetof(Ebml, version), { .u = EBML_VERSION } },
311 { EBML_ID_EBMLMAXSIZELENGTH, EBML_UINT, 0, offsetof(Ebml, max_size), { .u = 8 } },
312 { EBML_ID_EBMLMAXIDLENGTH, EBML_UINT, 0, offsetof(Ebml, id_length), { .u = 4 } },
313 { EBML_ID_DOCTYPE, EBML_STR, 0, offsetof(Ebml, doctype), { .s = "(none)" } },
314 { EBML_ID_DOCTYPEREADVERSION, EBML_UINT, 0, offsetof(Ebml, doctype_version), { .u = 1 } },
315 { EBML_ID_EBMLVERSION, EBML_NONE },
316 { EBML_ID_DOCTYPEVERSION, EBML_NONE },
320 static EbmlSyntax ebml_syntax[] = {
321 { EBML_ID_HEADER, EBML_NEST, 0, 0, { .n = ebml_header } },
325 static EbmlSyntax matroska_info[] = {
326 { MATROSKA_ID_TIMECODESCALE, EBML_UINT, 0, offsetof(MatroskaDemuxContext, time_scale), { .u = 1000000 } },
327 { MATROSKA_ID_DURATION, EBML_FLOAT, 0, offsetof(MatroskaDemuxContext, duration) },
328 { MATROSKA_ID_TITLE, EBML_UTF8, 0, offsetof(MatroskaDemuxContext, title) },
329 { MATROSKA_ID_WRITINGAPP, EBML_NONE },
330 { MATROSKA_ID_MUXINGAPP, EBML_UTF8, 0, offsetof(MatroskaDemuxContext, muxingapp) },
331 { MATROSKA_ID_DATEUTC, EBML_BIN, 0, offsetof(MatroskaDemuxContext, date_utc) },
332 { MATROSKA_ID_SEGMENTUID, EBML_NONE },
336 static EbmlSyntax matroska_track_video[] = {
337 { MATROSKA_ID_VIDEOFRAMERATE, EBML_FLOAT, 0, offsetof(MatroskaTrackVideo, frame_rate) },
338 { MATROSKA_ID_VIDEODISPLAYWIDTH, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_width), { .u=-1 } },
339 { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_height), { .u=-1 } },
340 { MATROSKA_ID_VIDEOPIXELWIDTH, EBML_UINT, 0, offsetof(MatroskaTrackVideo, pixel_width) },
341 { MATROSKA_ID_VIDEOPIXELHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, pixel_height) },
342 { MATROSKA_ID_VIDEOCOLORSPACE, EBML_BIN, 0, offsetof(MatroskaTrackVideo, color_space) },
343 { MATROSKA_ID_VIDEOALPHAMODE, EBML_UINT, 0, offsetof(MatroskaTrackVideo, alpha_mode) },
344 { MATROSKA_ID_VIDEOPIXELCROPB, EBML_NONE },
345 { MATROSKA_ID_VIDEOPIXELCROPT, EBML_NONE },
346 { MATROSKA_ID_VIDEOPIXELCROPL, EBML_NONE },
347 { MATROSKA_ID_VIDEOPIXELCROPR, EBML_NONE },
348 { MATROSKA_ID_VIDEODISPLAYUNIT, EBML_NONE },
349 { MATROSKA_ID_VIDEOFLAGINTERLACED, EBML_NONE },
350 { MATROSKA_ID_VIDEOSTEREOMODE, EBML_UINT, 0, offsetof(MatroskaTrackVideo, stereo_mode), { .u = MATROSKA_VIDEO_STEREOMODE_TYPE_NB } },
351 { MATROSKA_ID_VIDEOASPECTRATIO, EBML_NONE },
355 static EbmlSyntax matroska_track_audio[] = {
356 { MATROSKA_ID_AUDIOSAMPLINGFREQ, EBML_FLOAT, 0, offsetof(MatroskaTrackAudio, samplerate), { .f = 8000.0 } },
357 { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, EBML_FLOAT, 0, offsetof(MatroskaTrackAudio, out_samplerate) },
358 { MATROSKA_ID_AUDIOBITDEPTH, EBML_UINT, 0, offsetof(MatroskaTrackAudio, bitdepth) },
359 { MATROSKA_ID_AUDIOCHANNELS, EBML_UINT, 0, offsetof(MatroskaTrackAudio, channels), { .u = 1 } },
363 static EbmlSyntax matroska_track_encoding_compression[] = {
364 { MATROSKA_ID_ENCODINGCOMPALGO, EBML_UINT, 0, offsetof(MatroskaTrackCompression, algo), { .u = 0 } },
365 { MATROSKA_ID_ENCODINGCOMPSETTINGS, EBML_BIN, 0, offsetof(MatroskaTrackCompression, settings) },
369 static EbmlSyntax matroska_track_encoding_encryption[] = {
370 { MATROSKA_ID_ENCODINGENCALGO, EBML_UINT, 0, offsetof(MatroskaTrackEncryption,algo), {.u = 0} },
371 { MATROSKA_ID_ENCODINGENCKEYID, EBML_BIN, 0, offsetof(MatroskaTrackEncryption,key_id) },
372 { MATROSKA_ID_ENCODINGENCAESSETTINGS, EBML_NONE },
373 { MATROSKA_ID_ENCODINGSIGALGO, EBML_NONE },
374 { MATROSKA_ID_ENCODINGSIGHASHALGO, EBML_NONE },
375 { MATROSKA_ID_ENCODINGSIGKEYID, EBML_NONE },
376 { MATROSKA_ID_ENCODINGSIGNATURE, EBML_NONE },
379 static EbmlSyntax matroska_track_encoding[] = {
380 { MATROSKA_ID_ENCODINGSCOPE, EBML_UINT, 0, offsetof(MatroskaTrackEncoding, scope), { .u = 1 } },
381 { MATROSKA_ID_ENCODINGTYPE, EBML_UINT, 0, offsetof(MatroskaTrackEncoding, type), { .u = 0 } },
382 { MATROSKA_ID_ENCODINGCOMPRESSION, EBML_NEST, 0, offsetof(MatroskaTrackEncoding, compression), { .n = matroska_track_encoding_compression } },
383 { MATROSKA_ID_ENCODINGENCRYPTION, EBML_NEST, 0, offsetof(MatroskaTrackEncoding, encryption), { .n = matroska_track_encoding_encryption } },
384 { MATROSKA_ID_ENCODINGORDER, EBML_NONE },
388 static EbmlSyntax matroska_track_encodings[] = {
389 { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack, encodings), { .n = matroska_track_encoding } },
393 static EbmlSyntax matroska_track_plane[] = {
394 { MATROSKA_ID_TRACKPLANEUID, EBML_UINT, 0, offsetof(MatroskaTrackPlane,uid) },
395 { MATROSKA_ID_TRACKPLANETYPE, EBML_UINT, 0, offsetof(MatroskaTrackPlane,type) },
399 static EbmlSyntax matroska_track_combine_planes[] = {
400 { MATROSKA_ID_TRACKPLANE, EBML_NEST, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n = matroska_track_plane} },
404 static EbmlSyntax matroska_track_operation[] = {
405 { MATROSKA_ID_TRACKCOMBINEPLANES, EBML_NEST, 0, 0, {.n = matroska_track_combine_planes} },
409 static EbmlSyntax matroska_track[] = {
410 { MATROSKA_ID_TRACKNUMBER, EBML_UINT, 0, offsetof(MatroskaTrack, num) },
411 { MATROSKA_ID_TRACKNAME, EBML_UTF8, 0, offsetof(MatroskaTrack, name) },
412 { MATROSKA_ID_TRACKUID, EBML_UINT, 0, offsetof(MatroskaTrack, uid) },
413 { MATROSKA_ID_TRACKTYPE, EBML_UINT, 0, offsetof(MatroskaTrack, type) },
414 { MATROSKA_ID_CODECID, EBML_STR, 0, offsetof(MatroskaTrack, codec_id) },
415 { MATROSKA_ID_CODECPRIVATE, EBML_BIN, 0, offsetof(MatroskaTrack, codec_priv) },
416 { MATROSKA_ID_CODECDELAY, EBML_UINT, 0, offsetof(MatroskaTrack, codec_delay) },
417 { MATROSKA_ID_TRACKLANGUAGE, EBML_UTF8, 0, offsetof(MatroskaTrack, language), { .s = "eng" } },
418 { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack, default_duration) },
419 { MATROSKA_ID_TRACKTIMECODESCALE, EBML_FLOAT, 0, offsetof(MatroskaTrack, time_scale), { .f = 1.0 } },
420 { MATROSKA_ID_TRACKFLAGDEFAULT, EBML_UINT, 0, offsetof(MatroskaTrack, flag_default), { .u = 1 } },
421 { MATROSKA_ID_TRACKFLAGFORCED, EBML_UINT, 0, offsetof(MatroskaTrack, flag_forced), { .u = 0 } },
422 { MATROSKA_ID_TRACKVIDEO, EBML_NEST, 0, offsetof(MatroskaTrack, video), { .n = matroska_track_video } },
423 { MATROSKA_ID_TRACKAUDIO, EBML_NEST, 0, offsetof(MatroskaTrack, audio), { .n = matroska_track_audio } },
424 { MATROSKA_ID_TRACKOPERATION, EBML_NEST, 0, offsetof(MatroskaTrack, operation), { .n = matroska_track_operation } },
425 { MATROSKA_ID_TRACKCONTENTENCODINGS, EBML_NEST, 0, 0, { .n = matroska_track_encodings } },
426 { MATROSKA_ID_TRACKMAXBLKADDID, EBML_UINT, 0, offsetof(MatroskaTrack, max_block_additional_id) },
427 { MATROSKA_ID_SEEKPREROLL, EBML_UINT, 0, offsetof(MatroskaTrack, seek_preroll) },
428 { MATROSKA_ID_TRACKFLAGENABLED, EBML_NONE },
429 { MATROSKA_ID_TRACKFLAGLACING, EBML_NONE },
430 { MATROSKA_ID_CODECNAME, EBML_NONE },
431 { MATROSKA_ID_CODECDECODEALL, EBML_NONE },
432 { MATROSKA_ID_CODECINFOURL, EBML_NONE },
433 { MATROSKA_ID_CODECDOWNLOADURL, EBML_NONE },
434 { MATROSKA_ID_TRACKMINCACHE, EBML_NONE },
435 { MATROSKA_ID_TRACKMAXCACHE, EBML_NONE },
439 static EbmlSyntax matroska_tracks[] = {
440 { MATROSKA_ID_TRACKENTRY, EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext, tracks), { .n = matroska_track } },
444 static EbmlSyntax matroska_attachment[] = {
445 { MATROSKA_ID_FILEUID, EBML_UINT, 0, offsetof(MatroskaAttachment, uid) },
446 { MATROSKA_ID_FILENAME, EBML_UTF8, 0, offsetof(MatroskaAttachment, filename) },
447 { MATROSKA_ID_FILEMIMETYPE, EBML_STR, 0, offsetof(MatroskaAttachment, mime) },
448 { MATROSKA_ID_FILEDATA, EBML_BIN, 0, offsetof(MatroskaAttachment, bin) },
449 { MATROSKA_ID_FILEDESC, EBML_NONE },
453 static EbmlSyntax matroska_attachments[] = {
454 { MATROSKA_ID_ATTACHEDFILE, EBML_NEST, sizeof(MatroskaAttachment), offsetof(MatroskaDemuxContext, attachments), { .n = matroska_attachment } },
458 static EbmlSyntax matroska_chapter_display[] = {
459 { MATROSKA_ID_CHAPSTRING, EBML_UTF8, 0, offsetof(MatroskaChapter, title) },
460 { MATROSKA_ID_CHAPLANG, EBML_NONE },
464 static EbmlSyntax matroska_chapter_entry[] = {
465 { MATROSKA_ID_CHAPTERTIMESTART, EBML_UINT, 0, offsetof(MatroskaChapter, start), { .u = AV_NOPTS_VALUE } },
466 { MATROSKA_ID_CHAPTERTIMEEND, EBML_UINT, 0, offsetof(MatroskaChapter, end), { .u = AV_NOPTS_VALUE } },
467 { MATROSKA_ID_CHAPTERUID, EBML_UINT, 0, offsetof(MatroskaChapter, uid) },
468 { MATROSKA_ID_CHAPTERDISPLAY, EBML_NEST, 0, 0, { .n = matroska_chapter_display } },
469 { MATROSKA_ID_CHAPTERFLAGHIDDEN, EBML_NONE },
470 { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
471 { MATROSKA_ID_CHAPTERPHYSEQUIV, EBML_NONE },
472 { MATROSKA_ID_CHAPTERATOM, EBML_NONE },
476 static EbmlSyntax matroska_chapter[] = {
477 { MATROSKA_ID_CHAPTERATOM, EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext, chapters), { .n = matroska_chapter_entry } },
478 { MATROSKA_ID_EDITIONUID, EBML_NONE },
479 { MATROSKA_ID_EDITIONFLAGHIDDEN, EBML_NONE },
480 { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
481 { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
485 static EbmlSyntax matroska_chapters[] = {
486 { MATROSKA_ID_EDITIONENTRY, EBML_NEST, 0, 0, { .n = matroska_chapter } },
490 static EbmlSyntax matroska_index_pos[] = {
491 { MATROSKA_ID_CUETRACK, EBML_UINT, 0, offsetof(MatroskaIndexPos, track) },
492 { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos, pos) },
493 { MATROSKA_ID_CUERELATIVEPOSITION,EBML_NONE },
494 { MATROSKA_ID_CUEDURATION, EBML_NONE },
495 { MATROSKA_ID_CUEBLOCKNUMBER, EBML_NONE },
499 static EbmlSyntax matroska_index_entry[] = {
500 { MATROSKA_ID_CUETIME, EBML_UINT, 0, offsetof(MatroskaIndex, time) },
501 { MATROSKA_ID_CUETRACKPOSITION, EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex, pos), { .n = matroska_index_pos } },
505 static EbmlSyntax matroska_index[] = {
506 { MATROSKA_ID_POINTENTRY, EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext, index), { .n = matroska_index_entry } },
510 static EbmlSyntax matroska_simpletag[] = {
511 { MATROSKA_ID_TAGNAME, EBML_UTF8, 0, offsetof(MatroskaTag, name) },
512 { MATROSKA_ID_TAGSTRING, EBML_UTF8, 0, offsetof(MatroskaTag, string) },
513 { MATROSKA_ID_TAGLANG, EBML_STR, 0, offsetof(MatroskaTag, lang), { .s = "und" } },
514 { MATROSKA_ID_TAGDEFAULT, EBML_UINT, 0, offsetof(MatroskaTag, def) },
515 { MATROSKA_ID_TAGDEFAULT_BUG, EBML_UINT, 0, offsetof(MatroskaTag, def) },
516 { MATROSKA_ID_SIMPLETAG, EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag, sub), { .n = matroska_simpletag } },
520 static EbmlSyntax matroska_tagtargets[] = {
521 { MATROSKA_ID_TAGTARGETS_TYPE, EBML_STR, 0, offsetof(MatroskaTagTarget, type) },
522 { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget, typevalue), { .u = 50 } },
523 { MATROSKA_ID_TAGTARGETS_TRACKUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, trackuid) },
524 { MATROSKA_ID_TAGTARGETS_CHAPTERUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, chapteruid) },
525 { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, attachuid) },
529 static EbmlSyntax matroska_tag[] = {
530 { MATROSKA_ID_SIMPLETAG, EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags, tag), { .n = matroska_simpletag } },
531 { MATROSKA_ID_TAGTARGETS, EBML_NEST, 0, offsetof(MatroskaTags, target), { .n = matroska_tagtargets } },
535 static EbmlSyntax matroska_tags[] = {
536 { MATROSKA_ID_TAG, EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext, tags), { .n = matroska_tag } },
540 static EbmlSyntax matroska_seekhead_entry[] = {
541 { MATROSKA_ID_SEEKID, EBML_UINT, 0, offsetof(MatroskaSeekhead, id) },
542 { MATROSKA_ID_SEEKPOSITION, EBML_UINT, 0, offsetof(MatroskaSeekhead, pos), { .u = -1 } },
546 static EbmlSyntax matroska_seekhead[] = {
547 { MATROSKA_ID_SEEKENTRY, EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext, seekhead), { .n = matroska_seekhead_entry } },
551 static EbmlSyntax matroska_segment[] = {
552 { MATROSKA_ID_INFO, EBML_NEST, 0, 0, { .n = matroska_info } },
553 { MATROSKA_ID_TRACKS, EBML_NEST, 0, 0, { .n = matroska_tracks } },
554 { MATROSKA_ID_ATTACHMENTS, EBML_NEST, 0, 0, { .n = matroska_attachments } },
555 { MATROSKA_ID_CHAPTERS, EBML_NEST, 0, 0, { .n = matroska_chapters } },
556 { MATROSKA_ID_CUES, EBML_NEST, 0, 0, { .n = matroska_index } },
557 { MATROSKA_ID_TAGS, EBML_NEST, 0, 0, { .n = matroska_tags } },
558 { MATROSKA_ID_SEEKHEAD, EBML_NEST, 0, 0, { .n = matroska_seekhead } },
559 { MATROSKA_ID_CLUSTER, EBML_STOP },
563 static EbmlSyntax matroska_segments[] = {
564 { MATROSKA_ID_SEGMENT, EBML_NEST, 0, 0, { .n = matroska_segment } },
568 static EbmlSyntax matroska_blockmore[] = {
569 { MATROSKA_ID_BLOCKADDID, EBML_UINT, 0, offsetof(MatroskaBlock,additional_id) },
570 { MATROSKA_ID_BLOCKADDITIONAL, EBML_BIN, 0, offsetof(MatroskaBlock,additional) },
574 static EbmlSyntax matroska_blockadditions[] = {
575 { MATROSKA_ID_BLOCKMORE, EBML_NEST, 0, 0, {.n = matroska_blockmore} },
579 static EbmlSyntax matroska_blockgroup[] = {
580 { MATROSKA_ID_BLOCK, EBML_BIN, 0, offsetof(MatroskaBlock, bin) },
581 { MATROSKA_ID_BLOCKADDITIONS, EBML_NEST, 0, 0, { .n = matroska_blockadditions} },
582 { MATROSKA_ID_SIMPLEBLOCK, EBML_BIN, 0, offsetof(MatroskaBlock, bin) },
583 { MATROSKA_ID_BLOCKDURATION, EBML_UINT, 0, offsetof(MatroskaBlock, duration) },
584 { MATROSKA_ID_DISCARDPADDING, EBML_SINT, 0, offsetof(MatroskaBlock, discard_padding) },
585 { MATROSKA_ID_BLOCKREFERENCE, EBML_SINT, 0, offsetof(MatroskaBlock, reference) },
586 { MATROSKA_ID_CODECSTATE, EBML_NONE },
587 { 1, EBML_UINT, 0, offsetof(MatroskaBlock, non_simple), { .u = 1 } },
591 static EbmlSyntax matroska_cluster[] = {
592 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
593 { MATROSKA_ID_BLOCKGROUP, EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
594 { MATROSKA_ID_SIMPLEBLOCK, EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
595 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
596 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
600 static EbmlSyntax matroska_clusters[] = {
601 { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, { .n = matroska_cluster } },
602 { MATROSKA_ID_INFO, EBML_NONE },
603 { MATROSKA_ID_CUES, EBML_NONE },
604 { MATROSKA_ID_TAGS, EBML_NONE },
605 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
609 static EbmlSyntax matroska_cluster_incremental_parsing[] = {
610 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
611 { MATROSKA_ID_BLOCKGROUP, EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
612 { MATROSKA_ID_SIMPLEBLOCK, EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
613 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
614 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
615 { MATROSKA_ID_INFO, EBML_NONE },
616 { MATROSKA_ID_CUES, EBML_NONE },
617 { MATROSKA_ID_TAGS, EBML_NONE },
618 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
619 { MATROSKA_ID_CLUSTER, EBML_STOP },
623 static EbmlSyntax matroska_cluster_incremental[] = {
624 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
625 { MATROSKA_ID_BLOCKGROUP, EBML_STOP },
626 { MATROSKA_ID_SIMPLEBLOCK, EBML_STOP },
627 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
628 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
632 static EbmlSyntax matroska_clusters_incremental[] = {
633 { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, { .n = matroska_cluster_incremental } },
634 { MATROSKA_ID_INFO, EBML_NONE },
635 { MATROSKA_ID_CUES, EBML_NONE },
636 { MATROSKA_ID_TAGS, EBML_NONE },
637 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
641 static const char *const matroska_doctypes[] = { "matroska", "webm" };
643 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
645 AVIOContext *pb = matroska->ctx->pb;
647 matroska->current_id = 0;
648 matroska->num_levels = 0;
650 /* seek to next position to resync from */
651 if (avio_seek(pb, last_pos + 1, SEEK_SET) < 0)
656 // try to find a toplevel element
657 while (!avio_feof(pb)) {
658 if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
659 id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
660 id == MATROSKA_ID_SEEKHEAD || id == MATROSKA_ID_ATTACHMENTS ||
661 id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS) {
662 matroska->current_id = id;
665 id = (id << 8) | avio_r8(pb);
674 * Return: Whether we reached the end of a level in the hierarchy or not.
676 static int ebml_level_end(MatroskaDemuxContext *matroska)
678 AVIOContext *pb = matroska->ctx->pb;
679 int64_t pos = avio_tell(pb);
681 if (matroska->num_levels > 0) {
682 MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
683 if (pos - level->start >= level->length || matroska->current_id) {
684 matroska->num_levels--;
692 * Read: an "EBML number", which is defined as a variable-length
693 * array of bytes. The first byte indicates the length by giving a
694 * number of 0-bits followed by a one. The position of the first
695 * "one" bit inside the first byte indicates the length of this
697 * Returns: number of bytes read, < 0 on error
699 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
700 int max_size, uint64_t *number)
705 /* The first byte tells us the length in bytes - avio_r8() can normally
706 * return 0, but since that's not a valid first ebmlID byte, we can
707 * use it safely here to catch EOS. */
708 if (!(total = avio_r8(pb))) {
709 /* we might encounter EOS here */
710 if (!avio_feof(pb)) {
711 int64_t pos = avio_tell(pb);
712 av_log(matroska->ctx, AV_LOG_ERROR,
713 "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
715 return pb->error ? pb->error : AVERROR(EIO);
720 /* get the length of the EBML number */
721 read = 8 - ff_log2_tab[total];
722 if (read > max_size) {
723 int64_t pos = avio_tell(pb) - 1;
724 av_log(matroska->ctx, AV_LOG_ERROR,
725 "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
726 (uint8_t) total, pos, pos);
727 return AVERROR_INVALIDDATA;
730 /* read out length */
731 total ^= 1 << ff_log2_tab[total];
733 total = (total << 8) | avio_r8(pb);
741 * Read a EBML length value.
742 * This needs special handling for the "unknown length" case which has multiple
745 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
748 int res = ebml_read_num(matroska, pb, 8, number);
749 if (res > 0 && *number + 1 == 1ULL << (7 * res))
750 *number = 0xffffffffffffffULL;
755 * Read the next element as an unsigned int.
756 * 0 is success, < 0 is failure.
758 static int ebml_read_uint(AVIOContext *pb, int size, uint64_t *num)
763 return AVERROR_INVALIDDATA;
765 /* big-endian ordering; build up number */
768 *num = (*num << 8) | avio_r8(pb);
774 * Read the next element as a signed int.
775 * 0 is success, < 0 is failure.
777 static int ebml_read_sint(AVIOContext *pb, int size, int64_t *num)
782 return AVERROR_INVALIDDATA;
787 *num = sign_extend(avio_r8(pb), 8);
789 /* big-endian ordering; build up number */
791 *num = (*num << 8) | avio_r8(pb);
798 * Read the next element as a float.
799 * 0 is success, < 0 is failure.
801 static int ebml_read_float(AVIOContext *pb, int size, double *num)
806 *num = av_int2float(avio_rb32(pb));
808 *num = av_int2double(avio_rb64(pb));
810 return AVERROR_INVALIDDATA;
816 * Read the next element as an ASCII string.
817 * 0 is success, < 0 is failure.
819 static int ebml_read_ascii(AVIOContext *pb, int size, char **str)
823 /* EBML strings are usually not 0-terminated, so we allocate one
824 * byte more, read the string and NULL-terminate it ourselves. */
825 if (!(res = av_malloc(size + 1)))
826 return AVERROR(ENOMEM);
827 if (avio_read(pb, (uint8_t *) res, size) != size) {
839 * Read the next element as binary data.
840 * 0 is success, < 0 is failure.
842 static int ebml_read_binary(AVIOContext *pb, int length, EbmlBin *bin)
844 av_fast_padded_malloc(&bin->data, &bin->size, length);
846 return AVERROR(ENOMEM);
849 bin->pos = avio_tell(pb);
850 if (avio_read(pb, bin->data, length) != length) {
851 av_freep(&bin->data);
860 * Read the next element, but only the header. The contents
861 * are supposed to be sub-elements which can be read separately.
862 * 0 is success, < 0 is failure.
864 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
866 AVIOContext *pb = matroska->ctx->pb;
867 MatroskaLevel *level;
869 if (matroska->num_levels >= EBML_MAX_DEPTH) {
870 av_log(matroska->ctx, AV_LOG_ERROR,
871 "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
872 return AVERROR(ENOSYS);
875 level = &matroska->levels[matroska->num_levels++];
876 level->start = avio_tell(pb);
877 level->length = length;
883 * Read signed/unsigned "EBML" numbers.
884 * Return: number of bytes processed, < 0 on error
886 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
887 uint8_t *data, uint32_t size, uint64_t *num)
890 ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
891 return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
895 * Same as above, but signed.
897 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
898 uint8_t *data, uint32_t size, int64_t *num)
903 /* read as unsigned number first */
904 if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
907 /* make signed (weird way) */
908 *num = unum - ((1LL << (7 * res - 1)) - 1);
913 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
914 EbmlSyntax *syntax, void *data);
916 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
917 uint32_t id, void *data)
920 for (i = 0; syntax[i].id; i++)
921 if (id == syntax[i].id)
923 if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
924 matroska->num_levels > 0 &&
925 matroska->levels[matroska->num_levels - 1].length == 0xffffffffffffff)
926 return 0; // we reached the end of an unknown size cluster
927 if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
928 av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%"PRIX32"\n", id);
929 if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
930 return AVERROR_INVALIDDATA;
932 return ebml_parse_elem(matroska, &syntax[i], data);
935 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
938 if (!matroska->current_id) {
940 int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
943 matroska->current_id = id | 1 << 7 * res;
945 return ebml_parse_id(matroska, syntax, matroska->current_id, data);
948 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
953 for (i = 0; syntax[i].id; i++)
954 switch (syntax[i].type) {
956 *(uint64_t *) ((char *) data + syntax[i].data_offset) = syntax[i].def.u;
959 *(double *) ((char *) data + syntax[i].data_offset) = syntax[i].def.f;
963 // the default may be NULL
964 if (syntax[i].def.s) {
965 uint8_t **dst = (uint8_t **) ((uint8_t *) data + syntax[i].data_offset);
966 *dst = av_strdup(syntax[i].def.s);
968 return AVERROR(ENOMEM);
973 while (!res && !ebml_level_end(matroska))
974 res = ebml_parse(matroska, syntax, data);
979 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
980 EbmlSyntax *syntax, void *data)
982 static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
985 // max. 16 MB for strings
986 [EBML_STR] = 0x1000000,
987 [EBML_UTF8] = 0x1000000,
988 // max. 256 MB for binary data
989 [EBML_BIN] = 0x10000000,
990 // no limits for anything else
992 AVIOContext *pb = matroska->ctx->pb;
993 uint32_t id = syntax->id;
998 data = (char *) data + syntax->data_offset;
999 if (syntax->list_elem_size) {
1000 EbmlList *list = data;
1001 newelem = av_realloc_array(list->elem, list->nb_elem + 1, syntax->list_elem_size);
1003 return AVERROR(ENOMEM);
1004 list->elem = newelem;
1005 data = (char *) list->elem + list->nb_elem * syntax->list_elem_size;
1006 memset(data, 0, syntax->list_elem_size);
1010 if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
1011 matroska->current_id = 0;
1012 if ((res = ebml_read_length(matroska, pb, &length)) < 0)
1014 if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
1015 av_log(matroska->ctx, AV_LOG_ERROR,
1016 "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for syntax element %i\n",
1017 length, max_lengths[syntax->type], syntax->type);
1018 return AVERROR_INVALIDDATA;
1022 switch (syntax->type) {
1024 res = ebml_read_uint(pb, length, data);
1027 res = ebml_read_sint(pb, length, data);
1030 res = ebml_read_float(pb, length, data);
1034 res = ebml_read_ascii(pb, length, data);
1037 res = ebml_read_binary(pb, length, data);
1040 if ((res = ebml_read_master(matroska, length)) < 0)
1042 if (id == MATROSKA_ID_SEGMENT)
1043 matroska->segment_start = avio_tell(matroska->ctx->pb);
1044 return ebml_parse_nest(matroska, syntax->def.n, data);
1046 return ebml_parse_id(matroska, syntax->def.n, id, data);
1050 if (ffio_limit(pb, length) != length)
1051 return AVERROR(EIO);
1052 return avio_skip(pb, length) < 0 ? AVERROR(EIO) : 0;
1054 if (res == AVERROR_INVALIDDATA)
1055 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
1056 else if (res == AVERROR(EIO))
1057 av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
1061 static void ebml_free(EbmlSyntax *syntax, void *data)
1064 for (i = 0; syntax[i].id; i++) {
1065 void *data_off = (char *) data + syntax[i].data_offset;
1066 switch (syntax[i].type) {
1072 av_freep(&((EbmlBin *) data_off)->data);
1075 if (syntax[i].list_elem_size) {
1076 EbmlList *list = data_off;
1077 char *ptr = list->elem;
1078 for (j = 0; j < list->nb_elem;
1079 j++, ptr += syntax[i].list_elem_size)
1080 ebml_free(syntax[i].def.n, ptr);
1081 av_free(list->elem);
1083 ebml_free(syntax[i].def.n, data_off);
1093 static int matroska_probe(AVProbeData *p)
1096 int len_mask = 0x80, size = 1, n = 1, i;
1099 if (AV_RB32(p->buf) != EBML_ID_HEADER)
1102 /* length of header */
1104 while (size <= 8 && !(total & len_mask)) {
1110 total &= (len_mask - 1);
1112 total = (total << 8) | p->buf[4 + n++];
1114 /* Does the probe data contain the whole header? */
1115 if (p->buf_size < 4 + size + total)
1118 /* The header should contain a known document type. For now,
1119 * we don't parse the whole header but simply check for the
1120 * availability of that array of characters inside the header.
1121 * Not fully fool-proof, but good enough. */
1122 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1123 int probelen = strlen(matroska_doctypes[i]);
1124 if (total < probelen)
1126 for (n = 4 + size; n <= 4 + size + total - probelen; n++)
1127 if (!memcmp(p->buf + n, matroska_doctypes[i], probelen))
1128 return AVPROBE_SCORE_MAX;
1131 // probably valid EBML header but no recognized doctype
1132 return AVPROBE_SCORE_EXTENSION;
1135 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
1138 MatroskaTrack *tracks = matroska->tracks.elem;
1141 for (i = 0; i < matroska->tracks.nb_elem; i++)
1142 if (tracks[i].num == num)
1145 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
1149 static int matroska_decode_buffer(uint8_t **buf, int *buf_size,
1150 MatroskaTrack *track)
1152 MatroskaTrackEncoding *encodings = track->encodings.elem;
1153 uint8_t *data = *buf;
1154 int isize = *buf_size;
1155 uint8_t *pkt_data = NULL;
1156 uint8_t av_unused *newpktdata;
1157 int pkt_size = isize;
1161 if (pkt_size >= 10000000U)
1162 return AVERROR_INVALIDDATA;
1164 switch (encodings[0].compression.algo) {
1165 case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
1167 int header_size = encodings[0].compression.settings.size;
1168 uint8_t *header = encodings[0].compression.settings.data;
1170 if (header_size && !header) {
1171 av_log(NULL, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1178 pkt_size = isize + header_size;
1179 pkt_data = av_malloc(pkt_size);
1181 return AVERROR(ENOMEM);
1183 memcpy(pkt_data, header, header_size);
1184 memcpy(pkt_data + header_size, data, isize);
1188 case MATROSKA_TRACK_ENCODING_COMP_LZO:
1190 olen = pkt_size *= 3;
1191 newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING);
1193 result = AVERROR(ENOMEM);
1196 pkt_data = newpktdata;
1197 result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
1198 } while (result == AV_LZO_OUTPUT_FULL && pkt_size < 10000000);
1200 result = AVERROR_INVALIDDATA;
1207 case MATROSKA_TRACK_ENCODING_COMP_ZLIB:
1209 z_stream zstream = { 0 };
1210 if (inflateInit(&zstream) != Z_OK)
1212 zstream.next_in = data;
1213 zstream.avail_in = isize;
1216 newpktdata = av_realloc(pkt_data, pkt_size);
1218 inflateEnd(&zstream);
1221 pkt_data = newpktdata;
1222 zstream.avail_out = pkt_size - zstream.total_out;
1223 zstream.next_out = pkt_data + zstream.total_out;
1225 result = inflate(&zstream, Z_NO_FLUSH);
1227 result = Z_MEM_ERROR;
1228 } while (result == Z_OK && pkt_size < 10000000);
1229 pkt_size = zstream.total_out;
1230 inflateEnd(&zstream);
1231 if (result != Z_STREAM_END) {
1232 if (result == Z_MEM_ERROR)
1233 result = AVERROR(ENOMEM);
1235 result = AVERROR_INVALIDDATA;
1242 case MATROSKA_TRACK_ENCODING_COMP_BZLIB:
1244 bz_stream bzstream = { 0 };
1245 if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1247 bzstream.next_in = data;
1248 bzstream.avail_in = isize;
1251 newpktdata = av_realloc(pkt_data, pkt_size);
1253 BZ2_bzDecompressEnd(&bzstream);
1256 pkt_data = newpktdata;
1257 bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1258 bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1260 result = BZ2_bzDecompress(&bzstream);
1262 result = BZ_MEM_ERROR;
1263 } while (result == BZ_OK && pkt_size < 10000000);
1264 pkt_size = bzstream.total_out_lo32;
1265 BZ2_bzDecompressEnd(&bzstream);
1266 if (result != BZ_STREAM_END) {
1267 if (result == BZ_MEM_ERROR)
1268 result = AVERROR(ENOMEM);
1270 result = AVERROR_INVALIDDATA;
1277 return AVERROR_INVALIDDATA;
1281 *buf_size = pkt_size;
1290 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
1291 AVPacket *pkt, uint64_t display_duration)
1294 char *layer, *ptr = pkt->data, *end = ptr + pkt->size;
1296 for (; *ptr != ',' && ptr < end - 1; ptr++)
1301 for (; *ptr != ',' && ptr < end - 1; ptr++)
1304 int64_t end_pts = pkt->pts + display_duration;
1305 int sc = matroska->time_scale * pkt->pts / 10000000;
1306 int ec = matroska->time_scale * end_pts / 10000000;
1307 int sh, sm, ss, eh, em, es, len;
1321 len = 50 + end - ptr + FF_INPUT_BUFFER_PADDING_SIZE;
1322 if (!(line = av_buffer_alloc(len)))
1324 snprintf(line->data, len,
1325 "Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
1326 layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
1327 av_buffer_unref(&pkt->buf);
1329 pkt->data = line->data;
1330 pkt->size = strlen(line->data);
1334 static int matroska_merge_packets(AVPacket *out, AVPacket *in)
1336 int ret = av_grow_packet(out, in->size);
1340 memcpy(out->data + out->size - in->size, in->data, in->size);
1348 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1349 AVDictionary **metadata, char *prefix)
1351 MatroskaTag *tags = list->elem;
1355 for (i = 0; i < list->nb_elem; i++) {
1356 const char *lang = tags[i].lang &&
1357 strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1359 if (!tags[i].name) {
1360 av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1364 snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1366 av_strlcpy(key, tags[i].name, sizeof(key));
1367 if (tags[i].def || !lang) {
1368 av_dict_set(metadata, key, tags[i].string, 0);
1369 if (tags[i].sub.nb_elem)
1370 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1373 av_strlcat(key, "-", sizeof(key));
1374 av_strlcat(key, lang, sizeof(key));
1375 av_dict_set(metadata, key, tags[i].string, 0);
1376 if (tags[i].sub.nb_elem)
1377 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1380 ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1383 static void matroska_convert_tags(AVFormatContext *s)
1385 MatroskaDemuxContext *matroska = s->priv_data;
1386 MatroskaTags *tags = matroska->tags.elem;
1389 for (i = 0; i < matroska->tags.nb_elem; i++) {
1390 if (tags[i].target.attachuid) {
1391 MatroskaAttachment *attachment = matroska->attachments.elem;
1392 for (j = 0; j < matroska->attachments.nb_elem; j++)
1393 if (attachment[j].uid == tags[i].target.attachuid &&
1394 attachment[j].stream)
1395 matroska_convert_tag(s, &tags[i].tag,
1396 &attachment[j].stream->metadata, NULL);
1397 } else if (tags[i].target.chapteruid) {
1398 MatroskaChapter *chapter = matroska->chapters.elem;
1399 for (j = 0; j < matroska->chapters.nb_elem; j++)
1400 if (chapter[j].uid == tags[i].target.chapteruid &&
1402 matroska_convert_tag(s, &tags[i].tag,
1403 &chapter[j].chapter->metadata, NULL);
1404 } else if (tags[i].target.trackuid) {
1405 MatroskaTrack *track = matroska->tracks.elem;
1406 for (j = 0; j < matroska->tracks.nb_elem; j++)
1407 if (track[j].uid == tags[i].target.trackuid && track[j].stream)
1408 matroska_convert_tag(s, &tags[i].tag,
1409 &track[j].stream->metadata, NULL);
1411 matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1412 tags[i].target.type);
1417 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska,
1420 EbmlList *seekhead_list = &matroska->seekhead;
1421 uint32_t level_up = matroska->level_up;
1422 uint32_t saved_id = matroska->current_id;
1423 MatroskaSeekhead *seekhead = seekhead_list->elem;
1424 int64_t before_pos = avio_tell(matroska->ctx->pb);
1425 MatroskaLevel level;
1429 if (idx >= seekhead_list->nb_elem ||
1430 seekhead[idx].id == MATROSKA_ID_SEEKHEAD ||
1431 seekhead[idx].id == MATROSKA_ID_CLUSTER)
1435 offset = seekhead[idx].pos + matroska->segment_start;
1436 if (avio_seek(matroska->ctx->pb, offset, SEEK_SET) == offset) {
1437 /* We don't want to lose our seekhead level, so we add
1438 * a dummy. This is a crude hack. */
1439 if (matroska->num_levels == EBML_MAX_DEPTH) {
1440 av_log(matroska->ctx, AV_LOG_INFO,
1441 "Max EBML element depth (%d) reached, "
1442 "cannot parse further.\n", EBML_MAX_DEPTH);
1443 ret = AVERROR_INVALIDDATA;
1446 level.length = (uint64_t) -1;
1447 matroska->levels[matroska->num_levels] = level;
1448 matroska->num_levels++;
1449 matroska->current_id = 0;
1451 ret = ebml_parse(matroska, matroska_segment, matroska);
1453 /* remove dummy level */
1454 while (matroska->num_levels) {
1455 uint64_t length = matroska->levels[--matroska->num_levels].length;
1456 if (length == (uint64_t) -1)
1462 avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
1463 matroska->level_up = level_up;
1464 matroska->current_id = saved_id;
1469 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1471 EbmlList *seekhead_list = &matroska->seekhead;
1472 int64_t before_pos = avio_tell(matroska->ctx->pb);
1475 // we should not do any seeking in the streaming case
1476 if (!matroska->ctx->pb->seekable ||
1477 (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1480 for (i = 0; i < seekhead_list->nb_elem; i++) {
1481 MatroskaSeekhead *seekhead = seekhead_list->elem;
1482 if (seekhead[i].pos <= before_pos)
1485 // defer cues parsing until we actually need cue data.
1486 if (seekhead[i].id == MATROSKA_ID_CUES) {
1487 matroska->cues_parsing_deferred = 1;
1491 if (matroska_parse_seekhead_entry(matroska, i) < 0) {
1492 // mark index as broken
1493 matroska->cues_parsing_deferred = -1;
1499 static void matroska_add_index_entries(MatroskaDemuxContext *matroska)
1501 EbmlList *index_list;
1502 MatroskaIndex *index;
1503 int index_scale = 1;
1506 index_list = &matroska->index;
1507 index = index_list->elem;
1508 if (index_list->nb_elem &&
1509 index[0].time > 1E14 / matroska->time_scale) {
1510 av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1511 index_scale = matroska->time_scale;
1513 for (i = 0; i < index_list->nb_elem; i++) {
1514 EbmlList *pos_list = &index[i].pos;
1515 MatroskaIndexPos *pos = pos_list->elem;
1516 for (j = 0; j < pos_list->nb_elem; j++) {
1517 MatroskaTrack *track = matroska_find_track_by_num(matroska,
1519 if (track && track->stream)
1520 av_add_index_entry(track->stream,
1521 pos[j].pos + matroska->segment_start,
1522 index[i].time / index_scale, 0, 0,
1528 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1529 EbmlList *seekhead_list = &matroska->seekhead;
1530 MatroskaSeekhead *seekhead = seekhead_list->elem;
1533 for (i = 0; i < seekhead_list->nb_elem; i++)
1534 if (seekhead[i].id == MATROSKA_ID_CUES)
1536 av_assert1(i <= seekhead_list->nb_elem);
1538 if (matroska_parse_seekhead_entry(matroska, i) < 0)
1539 matroska->cues_parsing_deferred = -1;
1540 matroska_add_index_entries(matroska);
1543 static int matroska_aac_profile(char *codec_id)
1545 static const char *const aac_profiles[] = { "MAIN", "LC", "SSR" };
1548 for (profile = 0; profile < FF_ARRAY_ELEMS(aac_profiles); profile++)
1549 if (strstr(codec_id, aac_profiles[profile]))
1554 static int matroska_aac_sri(int samplerate)
1558 for (sri = 0; sri < FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1559 if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
1564 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
1567 /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
1568 time_t creation_time = date_utc / 1000000000 + 978307200;
1569 struct tm *ptm = gmtime(&creation_time);
1571 strftime(buffer, sizeof(buffer), "%Y-%m-%d %H:%M:%S", ptm);
1572 av_dict_set(metadata, "creation_time", buffer, 0);
1575 static int matroska_parse_flac(AVFormatContext *s,
1576 MatroskaTrack *track,
1579 AVStream *st = track->stream;
1580 uint8_t *p = track->codec_priv.data;
1581 int size = track->codec_priv.size;
1583 if (size < 8 + FLAC_STREAMINFO_SIZE || p[4] & 0x7f) {
1584 av_log(s, AV_LOG_WARNING, "Invalid FLAC private data\n");
1585 track->codec_priv.size = 0;
1589 track->codec_priv.size = 8 + FLAC_STREAMINFO_SIZE;
1591 p += track->codec_priv.size;
1592 size -= track->codec_priv.size;
1594 /* parse the remaining metadata blocks if present */
1596 int block_last, block_type, block_size;
1598 flac_parse_block_header(p, &block_last, &block_type, &block_size);
1602 if (block_size > size)
1605 /* check for the channel mask */
1606 if (block_type == FLAC_METADATA_TYPE_VORBIS_COMMENT) {
1607 AVDictionary *dict = NULL;
1608 AVDictionaryEntry *chmask;
1610 ff_vorbis_comment(s, &dict, p, block_size, 0);
1611 chmask = av_dict_get(dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", NULL, 0);
1613 uint64_t mask = strtol(chmask->value, NULL, 0);
1614 if (!mask || mask & ~0x3ffffULL) {
1615 av_log(s, AV_LOG_WARNING,
1616 "Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK\n");
1618 st->codec->channel_layout = mask;
1620 av_dict_free(&dict);
1630 static int matroska_parse_tracks(AVFormatContext *s)
1632 MatroskaDemuxContext *matroska = s->priv_data;
1633 MatroskaTrack *tracks = matroska->tracks.elem;
1638 for (i = 0; i < matroska->tracks.nb_elem; i++) {
1639 MatroskaTrack *track = &tracks[i];
1640 enum AVCodecID codec_id = AV_CODEC_ID_NONE;
1641 EbmlList *encodings_list = &track->encodings;
1642 MatroskaTrackEncoding *encodings = encodings_list->elem;
1643 uint8_t *extradata = NULL;
1644 int extradata_size = 0;
1645 int extradata_offset = 0;
1646 uint32_t fourcc = 0;
1648 char* key_id_base64 = NULL;
1651 /* Apply some sanity checks. */
1652 if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1653 track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1654 track->type != MATROSKA_TRACK_TYPE_SUBTITLE &&
1655 track->type != MATROSKA_TRACK_TYPE_METADATA) {
1656 av_log(matroska->ctx, AV_LOG_INFO,
1657 "Unknown or unsupported track type %"PRIu64"\n",
1661 if (!track->codec_id)
1664 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1665 if (!track->default_duration && track->video.frame_rate > 0)
1666 track->default_duration = 1000000000 / track->video.frame_rate;
1667 if (track->video.display_width == -1)
1668 track->video.display_width = track->video.pixel_width;
1669 if (track->video.display_height == -1)
1670 track->video.display_height = track->video.pixel_height;
1671 if (track->video.color_space.size == 4)
1672 fourcc = AV_RL32(track->video.color_space.data);
1673 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1674 if (!track->audio.out_samplerate)
1675 track->audio.out_samplerate = track->audio.samplerate;
1677 if (encodings_list->nb_elem > 1) {
1678 av_log(matroska->ctx, AV_LOG_ERROR,
1679 "Multiple combined encodings not supported");
1680 } else if (encodings_list->nb_elem == 1) {
1681 if (encodings[0].type) {
1682 if (encodings[0].encryption.key_id.size > 0) {
1683 /* Save the encryption key id to be stored later as a
1685 const int b64_size = AV_BASE64_SIZE(encodings[0].encryption.key_id.size);
1686 key_id_base64 = av_malloc(b64_size);
1687 if (key_id_base64 == NULL)
1688 return AVERROR(ENOMEM);
1690 av_base64_encode(key_id_base64, b64_size,
1691 encodings[0].encryption.key_id.data,
1692 encodings[0].encryption.key_id.size);
1694 encodings[0].scope = 0;
1695 av_log(matroska->ctx, AV_LOG_ERROR,
1696 "Unsupported encoding type");
1700 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1703 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1706 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO &&
1708 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP) {
1709 encodings[0].scope = 0;
1710 av_log(matroska->ctx, AV_LOG_ERROR,
1711 "Unsupported encoding type");
1712 } else if (track->codec_priv.size && encodings[0].scope & 2) {
1713 uint8_t *codec_priv = track->codec_priv.data;
1714 int ret = matroska_decode_buffer(&track->codec_priv.data,
1715 &track->codec_priv.size,
1718 track->codec_priv.data = NULL;
1719 track->codec_priv.size = 0;
1720 av_log(matroska->ctx, AV_LOG_ERROR,
1721 "Failed to decode codec private data\n");
1724 if (codec_priv != track->codec_priv.data)
1725 av_free(codec_priv);
1729 for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
1730 if (!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1731 strlen(ff_mkv_codec_tags[j].str))) {
1732 codec_id = ff_mkv_codec_tags[j].id;
1737 st = track->stream = avformat_new_stream(s, NULL);
1739 av_free(key_id_base64);
1740 return AVERROR(ENOMEM);
1743 if (key_id_base64) {
1744 /* export encryption key id as base64 metadata tag */
1745 av_dict_set(&st->metadata, "enc_key_id", key_id_base64, 0);
1746 av_freep(&key_id_base64);
1749 if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC") &&
1750 track->codec_priv.size >= 40 &&
1751 track->codec_priv.data) {
1752 track->ms_compat = 1;
1753 bit_depth = AV_RL16(track->codec_priv.data + 14);
1754 fourcc = AV_RL32(track->codec_priv.data + 16);
1755 codec_id = ff_codec_get_id(ff_codec_bmp_tags,
1758 codec_id = ff_codec_get_id(ff_codec_movvideo_tags,
1760 extradata_offset = 40;
1761 } else if (!strcmp(track->codec_id, "A_MS/ACM") &&
1762 track->codec_priv.size >= 14 &&
1763 track->codec_priv.data) {
1765 ffio_init_context(&b, track->codec_priv.data,
1766 track->codec_priv.size,
1767 0, NULL, NULL, NULL, NULL);
1768 ret = ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1771 codec_id = st->codec->codec_id;
1772 extradata_offset = FFMIN(track->codec_priv.size, 18);
1773 } else if (!strcmp(track->codec_id, "A_QUICKTIME")
1774 && (track->codec_priv.size >= 86)
1775 && (track->codec_priv.data)) {
1776 fourcc = AV_RL32(track->codec_priv.data + 4);
1777 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
1778 if (ff_codec_get_id(ff_codec_movaudio_tags, AV_RL32(track->codec_priv.data))) {
1779 fourcc = AV_RL32(track->codec_priv.data);
1780 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
1782 } else if (!strcmp(track->codec_id, "V_QUICKTIME") &&
1783 (track->codec_priv.size >= 21) &&
1784 (track->codec_priv.data)) {
1785 fourcc = AV_RL32(track->codec_priv.data + 4);
1786 codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
1787 if (ff_codec_get_id(ff_codec_movvideo_tags, AV_RL32(track->codec_priv.data))) {
1788 fourcc = AV_RL32(track->codec_priv.data);
1789 codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
1791 if (codec_id == AV_CODEC_ID_NONE && AV_RL32(track->codec_priv.data+4) == AV_RL32("SMI "))
1792 codec_id = AV_CODEC_ID_SVQ3;
1793 } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
1794 switch (track->audio.bitdepth) {
1796 codec_id = AV_CODEC_ID_PCM_U8;
1799 codec_id = AV_CODEC_ID_PCM_S24BE;
1802 codec_id = AV_CODEC_ID_PCM_S32BE;
1805 } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
1806 switch (track->audio.bitdepth) {
1808 codec_id = AV_CODEC_ID_PCM_U8;
1811 codec_id = AV_CODEC_ID_PCM_S24LE;
1814 codec_id = AV_CODEC_ID_PCM_S32LE;
1817 } else if (codec_id == AV_CODEC_ID_PCM_F32LE &&
1818 track->audio.bitdepth == 64) {
1819 codec_id = AV_CODEC_ID_PCM_F64LE;
1820 } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
1821 int profile = matroska_aac_profile(track->codec_id);
1822 int sri = matroska_aac_sri(track->audio.samplerate);
1823 extradata = av_mallocz(5 + FF_INPUT_BUFFER_PADDING_SIZE);
1825 return AVERROR(ENOMEM);
1826 extradata[0] = (profile << 3) | ((sri & 0x0E) >> 1);
1827 extradata[1] = ((sri & 0x01) << 7) | (track->audio.channels << 3);
1828 if (strstr(track->codec_id, "SBR")) {
1829 sri = matroska_aac_sri(track->audio.out_samplerate);
1830 extradata[2] = 0x56;
1831 extradata[3] = 0xE5;
1832 extradata[4] = 0x80 | (sri << 3);
1836 } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX - 12 - FF_INPUT_BUFFER_PADDING_SIZE) {
1837 /* Only ALAC's magic cookie is stored in Matroska's track headers.
1838 * Create the "atom size", "tag", and "tag version" fields the
1839 * decoder expects manually. */
1840 extradata_size = 12 + track->codec_priv.size;
1841 extradata = av_mallocz(extradata_size +
1842 FF_INPUT_BUFFER_PADDING_SIZE);
1844 return AVERROR(ENOMEM);
1845 AV_WB32(extradata, extradata_size);
1846 memcpy(&extradata[4], "alac", 4);
1847 AV_WB32(&extradata[8], 0);
1848 memcpy(&extradata[12], track->codec_priv.data,
1849 track->codec_priv.size);
1850 } else if (codec_id == AV_CODEC_ID_TTA) {
1851 extradata_size = 30;
1852 extradata = av_mallocz(extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
1854 return AVERROR(ENOMEM);
1855 ffio_init_context(&b, extradata, extradata_size, 1,
1856 NULL, NULL, NULL, NULL);
1857 avio_write(&b, "TTA1", 4);
1859 avio_wl16(&b, track->audio.channels);
1860 avio_wl16(&b, track->audio.bitdepth);
1861 if (track->audio.out_samplerate < 0 || track->audio.out_samplerate > INT_MAX)
1862 return AVERROR_INVALIDDATA;
1863 avio_wl32(&b, track->audio.out_samplerate);
1864 avio_wl32(&b, av_rescale((matroska->duration * matroska->time_scale),
1865 track->audio.out_samplerate,
1866 AV_TIME_BASE * 1000));
1867 } else if (codec_id == AV_CODEC_ID_RV10 ||
1868 codec_id == AV_CODEC_ID_RV20 ||
1869 codec_id == AV_CODEC_ID_RV30 ||
1870 codec_id == AV_CODEC_ID_RV40) {
1871 extradata_offset = 26;
1872 } else if (codec_id == AV_CODEC_ID_RA_144) {
1873 track->audio.out_samplerate = 8000;
1874 track->audio.channels = 1;
1875 } else if ((codec_id == AV_CODEC_ID_RA_288 ||
1876 codec_id == AV_CODEC_ID_COOK ||
1877 codec_id == AV_CODEC_ID_ATRAC3 ||
1878 codec_id == AV_CODEC_ID_SIPR)
1879 && track->codec_priv.data) {
1882 ffio_init_context(&b, track->codec_priv.data,
1883 track->codec_priv.size,
1884 0, NULL, NULL, NULL, NULL);
1886 flavor = avio_rb16(&b);
1887 track->audio.coded_framesize = avio_rb32(&b);
1889 track->audio.sub_packet_h = avio_rb16(&b);
1890 track->audio.frame_size = avio_rb16(&b);
1891 track->audio.sub_packet_size = avio_rb16(&b);
1893 track->audio.coded_framesize <= 0 ||
1894 track->audio.sub_packet_h <= 0 ||
1895 track->audio.frame_size <= 0 ||
1896 track->audio.sub_packet_size <= 0)
1897 return AVERROR_INVALIDDATA;
1898 track->audio.buf = av_malloc_array(track->audio.sub_packet_h,
1899 track->audio.frame_size);
1900 if (!track->audio.buf)
1901 return AVERROR(ENOMEM);
1902 if (codec_id == AV_CODEC_ID_RA_288) {
1903 st->codec->block_align = track->audio.coded_framesize;
1904 track->codec_priv.size = 0;
1906 if (codec_id == AV_CODEC_ID_SIPR && flavor < 4) {
1907 static const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1908 track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1909 st->codec->bit_rate = sipr_bit_rate[flavor];
1911 st->codec->block_align = track->audio.sub_packet_size;
1912 extradata_offset = 78;
1914 } else if (codec_id == AV_CODEC_ID_FLAC && track->codec_priv.size) {
1915 ret = matroska_parse_flac(s, track, &extradata_offset);
1918 } else if (codec_id == AV_CODEC_ID_PRORES && track->codec_priv.size == 4) {
1919 fourcc = AV_RL32(track->codec_priv.data);
1921 track->codec_priv.size -= extradata_offset;
1923 if (codec_id == AV_CODEC_ID_NONE)
1924 av_log(matroska->ctx, AV_LOG_INFO,
1925 "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
1927 if (track->time_scale < 0.01)
1928 track->time_scale = 1.0;
1929 avpriv_set_pts_info(st, 64, matroska->time_scale * track->time_scale,
1930 1000 * 1000 * 1000); /* 64 bit pts in ns */
1932 /* convert the delay from ns to the track timebase */
1933 track->codec_delay = av_rescale_q(track->codec_delay,
1934 (AVRational){ 1, 1000000000 },
1937 st->codec->codec_id = codec_id;
1939 if (strcmp(track->language, "und"))
1940 av_dict_set(&st->metadata, "language", track->language, 0);
1941 av_dict_set(&st->metadata, "title", track->name, 0);
1943 if (track->flag_default)
1944 st->disposition |= AV_DISPOSITION_DEFAULT;
1945 if (track->flag_forced)
1946 st->disposition |= AV_DISPOSITION_FORCED;
1948 if (!st->codec->extradata) {
1950 st->codec->extradata = extradata;
1951 st->codec->extradata_size = extradata_size;
1952 } else if (track->codec_priv.data && track->codec_priv.size > 0) {
1953 if (ff_alloc_extradata(st->codec, track->codec_priv.size))
1954 return AVERROR(ENOMEM);
1955 memcpy(st->codec->extradata,
1956 track->codec_priv.data + extradata_offset,
1957 track->codec_priv.size);
1961 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1962 MatroskaTrackPlane *planes = track->operation.combine_planes.elem;
1964 st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1965 st->codec->codec_tag = fourcc;
1967 st->codec->bits_per_coded_sample = bit_depth;
1968 st->codec->width = track->video.pixel_width;
1969 st->codec->height = track->video.pixel_height;
1970 av_reduce(&st->sample_aspect_ratio.num,
1971 &st->sample_aspect_ratio.den,
1972 st->codec->height * track->video.display_width,
1973 st->codec->width * track->video.display_height,
1975 if (st->codec->codec_id != AV_CODEC_ID_HEVC)
1976 st->need_parsing = AVSTREAM_PARSE_HEADERS;
1978 if (track->default_duration) {
1979 av_reduce(&st->avg_frame_rate.num, &st->avg_frame_rate.den,
1980 1000000000, track->default_duration, 30000);
1981 #if FF_API_R_FRAME_RATE
1982 if (st->avg_frame_rate.num < st->avg_frame_rate.den * 1000L)
1983 st->r_frame_rate = st->avg_frame_rate;
1987 /* export stereo mode flag as metadata tag */
1988 if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
1989 av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
1991 /* export alpha mode flag as metadata tag */
1992 if (track->video.alpha_mode)
1993 av_dict_set(&st->metadata, "alpha_mode", "1", 0);
1995 /* if we have virtual track, mark the real tracks */
1996 for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
1998 if (planes[j].type >= MATROSKA_VIDEO_STEREO_PLANE_COUNT)
2000 snprintf(buf, sizeof(buf), "%s_%d",
2001 ff_matroska_video_stereo_plane[planes[j].type], i);
2002 for (k=0; k < matroska->tracks.nb_elem; k++)
2003 if (planes[j].uid == tracks[k].uid) {
2004 av_dict_set(&s->streams[k]->metadata,
2005 "stereo_mode", buf, 0);
2009 // add stream level stereo3d side data if it is a supported format
2010 if (track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB &&
2011 track->video.stereo_mode != 10 && track->video.stereo_mode != 12) {
2012 int ret = ff_mkv_stereo3d_conv(st, track->video.stereo_mode);
2016 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2017 st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
2018 st->codec->sample_rate = track->audio.out_samplerate;
2019 st->codec->channels = track->audio.channels;
2020 if (!st->codec->bits_per_coded_sample)
2021 st->codec->bits_per_coded_sample = track->audio.bitdepth;
2022 if (st->codec->codec_id != AV_CODEC_ID_AAC)
2023 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2024 if (track->codec_delay > 0) {
2025 st->codec->delay = av_rescale_q(track->codec_delay,
2027 (AVRational){1, st->codec->sample_rate});
2029 if (track->seek_preroll > 0) {
2030 av_codec_set_seek_preroll(st->codec,
2031 av_rescale_q(track->seek_preroll,
2032 (AVRational){1, 1000000000},
2033 (AVRational){1, st->codec->sample_rate}));
2035 } else if (codec_id == AV_CODEC_ID_WEBVTT) {
2036 st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
2038 if (!strcmp(track->codec_id, "D_WEBVTT/CAPTIONS")) {
2039 st->disposition |= AV_DISPOSITION_CAPTIONS;
2040 } else if (!strcmp(track->codec_id, "D_WEBVTT/DESCRIPTIONS")) {
2041 st->disposition |= AV_DISPOSITION_DESCRIPTIONS;
2042 } else if (!strcmp(track->codec_id, "D_WEBVTT/METADATA")) {
2043 st->disposition |= AV_DISPOSITION_METADATA;
2045 } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2046 st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
2048 if (st->codec->codec_id == AV_CODEC_ID_SSA ||
2049 st->codec->codec_id == AV_CODEC_ID_ASS)
2051 if (st->codec->codec_id == AV_CODEC_ID_ASS)
2053 matroska->contains_ssa = 1;
2060 static int matroska_read_header(AVFormatContext *s)
2062 MatroskaDemuxContext *matroska = s->priv_data;
2063 EbmlList *attachments_list = &matroska->attachments;
2064 EbmlList *chapters_list = &matroska->chapters;
2065 MatroskaAttachment *attachments;
2066 MatroskaChapter *chapters;
2067 uint64_t max_start = 0;
2074 /* First read the EBML header. */
2075 if (ebml_parse(matroska, ebml_syntax, &ebml) ||
2076 ebml.version > EBML_VERSION ||
2077 ebml.max_size > sizeof(uint64_t) ||
2078 ebml.id_length > sizeof(uint32_t) ||
2079 ebml.doctype_version > 3 ||
2081 av_log(matroska->ctx, AV_LOG_ERROR,
2082 "EBML header using unsupported features\n"
2083 "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2084 ebml.version, ebml.doctype, ebml.doctype_version);
2085 ebml_free(ebml_syntax, &ebml);
2086 return AVERROR_PATCHWELCOME;
2087 } else if (ebml.doctype_version == 3) {
2088 av_log(matroska->ctx, AV_LOG_WARNING,
2089 "EBML header using unsupported features\n"
2090 "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2091 ebml.version, ebml.doctype, ebml.doctype_version);
2093 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
2094 if (!strcmp(ebml.doctype, matroska_doctypes[i]))
2096 if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
2097 av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
2098 if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
2099 ebml_free(ebml_syntax, &ebml);
2100 return AVERROR_INVALIDDATA;
2103 ebml_free(ebml_syntax, &ebml);
2105 /* The next thing is a segment. */
2106 pos = avio_tell(matroska->ctx->pb);
2107 res = ebml_parse(matroska, matroska_segments, matroska);
2108 // try resyncing until we find a EBML_STOP type element.
2110 res = matroska_resync(matroska, pos);
2113 pos = avio_tell(matroska->ctx->pb);
2114 res = ebml_parse(matroska, matroska_segment, matroska);
2116 matroska_execute_seekhead(matroska);
2118 if (!matroska->time_scale)
2119 matroska->time_scale = 1000000;
2120 if (matroska->duration)
2121 matroska->ctx->duration = matroska->duration * matroska->time_scale *
2122 1000 / AV_TIME_BASE;
2123 av_dict_set(&s->metadata, "title", matroska->title, 0);
2124 av_dict_set(&s->metadata, "encoder", matroska->muxingapp, 0);
2126 if (matroska->date_utc.size == 8)
2127 matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
2129 res = matroska_parse_tracks(s);
2133 attachments = attachments_list->elem;
2134 for (j = 0; j < attachments_list->nb_elem; j++) {
2135 if (!(attachments[j].filename && attachments[j].mime &&
2136 attachments[j].bin.data && attachments[j].bin.size > 0)) {
2137 av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
2139 AVStream *st = avformat_new_stream(s, NULL);
2142 av_dict_set(&st->metadata, "filename", attachments[j].filename, 0);
2143 av_dict_set(&st->metadata, "mimetype", attachments[j].mime, 0);
2144 st->codec->codec_id = AV_CODEC_ID_NONE;
2145 st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
2146 if (ff_alloc_extradata(st->codec, attachments[j].bin.size))
2148 memcpy(st->codec->extradata, attachments[j].bin.data,
2149 attachments[j].bin.size);
2151 for (i = 0; ff_mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
2152 if (!strncmp(ff_mkv_mime_tags[i].str, attachments[j].mime,
2153 strlen(ff_mkv_mime_tags[i].str))) {
2154 st->codec->codec_id = ff_mkv_mime_tags[i].id;
2158 attachments[j].stream = st;
2162 chapters = chapters_list->elem;
2163 for (i = 0; i < chapters_list->nb_elem; i++)
2164 if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid &&
2165 (max_start == 0 || chapters[i].start > max_start)) {
2166 chapters[i].chapter =
2167 avpriv_new_chapter(s, chapters[i].uid,
2168 (AVRational) { 1, 1000000000 },
2169 chapters[i].start, chapters[i].end,
2171 if (chapters[i].chapter) {
2172 av_dict_set(&chapters[i].chapter->metadata,
2173 "title", chapters[i].title, 0);
2175 max_start = chapters[i].start;
2178 matroska_add_index_entries(matroska);
2180 matroska_convert_tags(s);
2186 * Put one packet in an application-supplied AVPacket struct.
2187 * Returns 0 on success or -1 on failure.
2189 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
2192 if (matroska->num_packets > 0) {
2193 memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
2194 av_free(matroska->packets[0]);
2195 if (matroska->num_packets > 1) {
2197 memmove(&matroska->packets[0], &matroska->packets[1],
2198 (matroska->num_packets - 1) * sizeof(AVPacket *));
2199 newpackets = av_realloc(matroska->packets,
2200 (matroska->num_packets - 1) *
2201 sizeof(AVPacket *));
2203 matroska->packets = newpackets;
2205 av_freep(&matroska->packets);
2206 matroska->prev_pkt = NULL;
2208 matroska->num_packets--;
2216 * Free all packets in our internal queue.
2218 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
2220 matroska->prev_pkt = NULL;
2221 if (matroska->packets) {
2223 for (n = 0; n < matroska->num_packets; n++) {
2224 av_free_packet(matroska->packets[n]);
2225 av_free(matroska->packets[n]);
2227 av_freep(&matroska->packets);
2228 matroska->num_packets = 0;
2232 static int matroska_parse_laces(MatroskaDemuxContext *matroska, uint8_t **buf,
2233 int *buf_size, int type,
2234 uint32_t **lace_buf, int *laces)
2236 int res = 0, n, size = *buf_size;
2237 uint8_t *data = *buf;
2238 uint32_t *lace_size;
2242 *lace_buf = av_mallocz(sizeof(int));
2244 return AVERROR(ENOMEM);
2246 *lace_buf[0] = size;
2250 av_assert0(size > 0);
2254 lace_size = av_mallocz(*laces * sizeof(int));
2256 return AVERROR(ENOMEM);
2259 case 0x1: /* Xiph lacing */
2263 for (n = 0; res == 0 && n < *laces - 1; n++) {
2265 if (size <= total) {
2266 res = AVERROR_INVALIDDATA;
2271 lace_size[n] += temp;
2278 if (size <= total) {
2279 res = AVERROR_INVALIDDATA;
2283 lace_size[n] = size - total;
2287 case 0x2: /* fixed-size lacing */
2288 if (size % (*laces)) {
2289 res = AVERROR_INVALIDDATA;
2292 for (n = 0; n < *laces; n++)
2293 lace_size[n] = size / *laces;
2296 case 0x3: /* EBML lacing */
2300 n = matroska_ebmlnum_uint(matroska, data, size, &num);
2301 if (n < 0 || num > INT_MAX) {
2302 av_log(matroska->ctx, AV_LOG_INFO,
2303 "EBML block data error\n");
2304 res = n<0 ? n : AVERROR_INVALIDDATA;
2309 total = lace_size[0] = num;
2310 for (n = 1; res == 0 && n < *laces - 1; n++) {
2313 r = matroska_ebmlnum_sint(matroska, data, size, &snum);
2314 if (r < 0 || lace_size[n - 1] + snum > (uint64_t)INT_MAX) {
2315 av_log(matroska->ctx, AV_LOG_INFO,
2316 "EBML block data error\n");
2317 res = r<0 ? r : AVERROR_INVALIDDATA;
2322 lace_size[n] = lace_size[n - 1] + snum;
2323 total += lace_size[n];
2325 if (size <= total) {
2326 res = AVERROR_INVALIDDATA;
2329 lace_size[*laces - 1] = size - total;
2335 *lace_buf = lace_size;
2341 static int matroska_parse_rm_audio(MatroskaDemuxContext *matroska,
2342 MatroskaTrack *track, AVStream *st,
2343 uint8_t *data, int size, uint64_t timecode,
2346 int a = st->codec->block_align;
2347 int sps = track->audio.sub_packet_size;
2348 int cfs = track->audio.coded_framesize;
2349 int h = track->audio.sub_packet_h;
2350 int y = track->audio.sub_packet_cnt;
2351 int w = track->audio.frame_size;
2354 if (!track->audio.pkt_cnt) {
2355 if (track->audio.sub_packet_cnt == 0)
2356 track->audio.buf_timecode = timecode;
2357 if (st->codec->codec_id == AV_CODEC_ID_RA_288) {
2358 if (size < cfs * h / 2) {
2359 av_log(matroska->ctx, AV_LOG_ERROR,
2360 "Corrupt int4 RM-style audio packet size\n");
2361 return AVERROR_INVALIDDATA;
2363 for (x = 0; x < h / 2; x++)
2364 memcpy(track->audio.buf + x * 2 * w + y * cfs,
2365 data + x * cfs, cfs);
2366 } else if (st->codec->codec_id == AV_CODEC_ID_SIPR) {
2368 av_log(matroska->ctx, AV_LOG_ERROR,
2369 "Corrupt sipr RM-style audio packet size\n");
2370 return AVERROR_INVALIDDATA;
2372 memcpy(track->audio.buf + y * w, data, w);
2374 if (size < sps * w / sps || h<=0 || w%sps) {
2375 av_log(matroska->ctx, AV_LOG_ERROR,
2376 "Corrupt generic RM-style audio packet size\n");
2377 return AVERROR_INVALIDDATA;
2379 for (x = 0; x < w / sps; x++)
2380 memcpy(track->audio.buf +
2381 sps * (h * x + ((h + 1) / 2) * (y & 1) + (y >> 1)),
2382 data + x * sps, sps);
2385 if (++track->audio.sub_packet_cnt >= h) {
2386 if (st->codec->codec_id == AV_CODEC_ID_SIPR)
2387 ff_rm_reorder_sipr_data(track->audio.buf, h, w);
2388 track->audio.sub_packet_cnt = 0;
2389 track->audio.pkt_cnt = h * w / a;
2393 while (track->audio.pkt_cnt) {
2394 AVPacket *pkt = NULL;
2395 if (!(pkt = av_mallocz(sizeof(AVPacket))) || av_new_packet(pkt, a) < 0) {
2397 return AVERROR(ENOMEM);
2400 track->audio.buf + a * (h * w / a - track->audio.pkt_cnt--),
2402 pkt->pts = track->audio.buf_timecode;
2403 track->audio.buf_timecode = AV_NOPTS_VALUE;
2405 pkt->stream_index = st->index;
2406 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2412 /* reconstruct full wavpack blocks from mangled matroska ones */
2413 static int matroska_parse_wavpack(MatroskaTrack *track, uint8_t *src,
2414 uint8_t **pdst, int *size)
2416 uint8_t *dst = NULL;
2421 int ret, offset = 0;
2423 if (srclen < 12 || track->stream->codec->extradata_size < 2)
2424 return AVERROR_INVALIDDATA;
2426 ver = AV_RL16(track->stream->codec->extradata);
2428 samples = AV_RL32(src);
2432 while (srclen >= 8) {
2437 uint32_t flags = AV_RL32(src);
2438 uint32_t crc = AV_RL32(src + 4);
2442 multiblock = (flags & 0x1800) != 0x1800;
2445 ret = AVERROR_INVALIDDATA;
2448 blocksize = AV_RL32(src);
2454 if (blocksize > srclen) {
2455 ret = AVERROR_INVALIDDATA;
2459 tmp = av_realloc(dst, dstlen + blocksize + 32);
2461 ret = AVERROR(ENOMEM);
2465 dstlen += blocksize + 32;
2467 AV_WL32(dst + offset, MKTAG('w', 'v', 'p', 'k')); // tag
2468 AV_WL32(dst + offset + 4, blocksize + 24); // blocksize - 8
2469 AV_WL16(dst + offset + 8, ver); // version
2470 AV_WL16(dst + offset + 10, 0); // track/index_no
2471 AV_WL32(dst + offset + 12, 0); // total samples
2472 AV_WL32(dst + offset + 16, 0); // block index
2473 AV_WL32(dst + offset + 20, samples); // number of samples
2474 AV_WL32(dst + offset + 24, flags); // flags
2475 AV_WL32(dst + offset + 28, crc); // crc
2476 memcpy(dst + offset + 32, src, blocksize); // block data
2479 srclen -= blocksize;
2480 offset += blocksize + 32;
2493 static int matroska_parse_webvtt(MatroskaDemuxContext *matroska,
2494 MatroskaTrack *track,
2496 uint8_t *data, int data_len,
2502 uint8_t *id, *settings, *text, *buf;
2503 int id_len, settings_len, text_len;
2508 return AVERROR_INVALIDDATA;
2511 q = data + data_len;
2516 if (*p == '\r' || *p == '\n') {
2525 if (p >= q || *p != '\n')
2526 return AVERROR_INVALIDDATA;
2532 if (*p == '\r' || *p == '\n') {
2533 settings_len = p - settings;
2541 if (p >= q || *p != '\n')
2542 return AVERROR_INVALIDDATA;
2547 while (text_len > 0) {
2548 const int len = text_len - 1;
2549 const uint8_t c = p[len];
2550 if (c != '\r' && c != '\n')
2556 return AVERROR_INVALIDDATA;
2558 pkt = av_mallocz(sizeof(*pkt));
2559 err = av_new_packet(pkt, text_len);
2562 return AVERROR(err);
2565 memcpy(pkt->data, text, text_len);
2568 buf = av_packet_new_side_data(pkt,
2569 AV_PKT_DATA_WEBVTT_IDENTIFIER,
2573 return AVERROR(ENOMEM);
2575 memcpy(buf, id, id_len);
2578 if (settings_len > 0) {
2579 buf = av_packet_new_side_data(pkt,
2580 AV_PKT_DATA_WEBVTT_SETTINGS,
2584 return AVERROR(ENOMEM);
2586 memcpy(buf, settings, settings_len);
2589 // Do we need this for subtitles?
2590 // pkt->flags = AV_PKT_FLAG_KEY;
2592 pkt->stream_index = st->index;
2593 pkt->pts = timecode;
2595 // Do we need this for subtitles?
2596 // pkt->dts = timecode;
2598 pkt->duration = duration;
2601 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2602 matroska->prev_pkt = pkt;
2607 static int matroska_parse_frame(MatroskaDemuxContext *matroska,
2608 MatroskaTrack *track, AVStream *st,
2609 uint8_t *data, int pkt_size,
2610 uint64_t timecode, uint64_t lace_duration,
2611 int64_t pos, int is_keyframe,
2612 uint8_t *additional, uint64_t additional_id, int additional_size,
2613 int64_t discard_padding)
2615 MatroskaTrackEncoding *encodings = track->encodings.elem;
2616 uint8_t *pkt_data = data;
2617 int offset = 0, res;
2620 if (encodings && !encodings->type && encodings->scope & 1) {
2621 res = matroska_decode_buffer(&pkt_data, &pkt_size, track);
2626 if (st->codec->codec_id == AV_CODEC_ID_WAVPACK) {
2628 res = matroska_parse_wavpack(track, pkt_data, &wv_data, &pkt_size);
2630 av_log(matroska->ctx, AV_LOG_ERROR,
2631 "Error parsing a wavpack block.\n");
2634 if (pkt_data != data)
2635 av_freep(&pkt_data);
2639 if (st->codec->codec_id == AV_CODEC_ID_PRORES &&
2640 AV_RB32(&data[4]) != MKBETAG('i', 'c', 'p', 'f'))
2643 pkt = av_mallocz(sizeof(AVPacket));
2644 /* XXX: prevent data copy... */
2645 if (av_new_packet(pkt, pkt_size + offset) < 0) {
2647 res = AVERROR(ENOMEM);
2651 if (st->codec->codec_id == AV_CODEC_ID_PRORES && offset == 8) {
2652 uint8_t *buf = pkt->data;
2653 bytestream_put_be32(&buf, pkt_size);
2654 bytestream_put_be32(&buf, MKBETAG('i', 'c', 'p', 'f'));
2657 memcpy(pkt->data + offset, pkt_data, pkt_size);
2659 if (pkt_data != data)
2660 av_freep(&pkt_data);
2662 pkt->flags = is_keyframe;
2663 pkt->stream_index = st->index;
2665 if (additional_size > 0) {
2666 uint8_t *side_data = av_packet_new_side_data(pkt,
2667 AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL,
2668 additional_size + 8);
2670 av_free_packet(pkt);
2672 return AVERROR(ENOMEM);
2674 AV_WB64(side_data, additional_id);
2675 memcpy(side_data + 8, additional, additional_size);
2678 if (discard_padding) {
2679 uint8_t *side_data = av_packet_new_side_data(pkt,
2680 AV_PKT_DATA_SKIP_SAMPLES,
2683 av_free_packet(pkt);
2685 return AVERROR(ENOMEM);
2687 AV_WL32(side_data, 0);
2688 AV_WL32(side_data + 4, av_rescale_q(discard_padding,
2689 (AVRational){1, 1000000000},
2690 (AVRational){1, st->codec->sample_rate}));
2693 if (track->ms_compat)
2694 pkt->dts = timecode;
2696 pkt->pts = timecode;
2698 if (st->codec->codec_id == AV_CODEC_ID_SUBRIP) {
2700 * For backward compatibility.
2701 * Historically, we have put subtitle duration
2702 * in convergence_duration, on the off chance
2703 * that the time_scale is less than 1us, which
2704 * could result in a 32bit overflow on the
2705 * normal duration field.
2707 pkt->convergence_duration = lace_duration;
2710 if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE ||
2711 lace_duration <= INT_MAX) {
2713 * For non subtitle tracks, just store the duration
2716 * If it's a subtitle track and duration value does
2717 * not overflow a uint32, then also store it normally.
2719 pkt->duration = lace_duration;
2723 if (st->codec->codec_id == AV_CODEC_ID_SSA)
2724 matroska_fix_ass_packet(matroska, pkt, lace_duration);
2726 if (matroska->prev_pkt &&
2727 timecode != AV_NOPTS_VALUE &&
2728 matroska->prev_pkt->pts == timecode &&
2729 matroska->prev_pkt->stream_index == st->index &&
2730 st->codec->codec_id == AV_CODEC_ID_SSA)
2731 matroska_merge_packets(matroska->prev_pkt, pkt);
2733 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2734 matroska->prev_pkt = pkt;
2737 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2738 matroska->prev_pkt = pkt;
2744 if (pkt_data != data)
2745 av_freep(&pkt_data);
2749 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
2750 int size, int64_t pos, uint64_t cluster_time,
2751 uint64_t block_duration, int is_keyframe,
2752 uint8_t *additional, uint64_t additional_id, int additional_size,
2753 int64_t cluster_pos, int64_t discard_padding)
2755 uint64_t timecode = AV_NOPTS_VALUE;
2756 MatroskaTrack *track;
2760 uint32_t *lace_size = NULL;
2761 int n, flags, laces = 0;
2763 int trust_default_duration = 1;
2765 if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
2766 av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
2772 track = matroska_find_track_by_num(matroska, num);
2773 if (!track || !track->stream) {
2774 av_log(matroska->ctx, AV_LOG_INFO,
2775 "Invalid stream %"PRIu64" or size %u\n", num, size);
2776 return AVERROR_INVALIDDATA;
2777 } else if (size <= 3)
2780 if (st->discard >= AVDISCARD_ALL)
2782 av_assert1(block_duration != AV_NOPTS_VALUE);
2784 block_time = sign_extend(AV_RB16(data), 16);
2788 if (is_keyframe == -1)
2789 is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
2791 if (cluster_time != (uint64_t) -1 &&
2792 (block_time >= 0 || cluster_time >= -block_time)) {
2793 timecode = cluster_time + block_time - track->codec_delay;
2794 if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE &&
2795 timecode < track->end_timecode)
2796 is_keyframe = 0; /* overlapping subtitles are not key frame */
2798 av_add_index_entry(st, cluster_pos, timecode, 0, 0,
2802 if (matroska->skip_to_keyframe &&
2803 track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
2804 if (timecode < matroska->skip_to_timecode)
2807 matroska->skip_to_keyframe = 0;
2808 else if (!st->skip_to_keyframe) {
2809 av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
2810 matroska->skip_to_keyframe = 0;
2814 res = matroska_parse_laces(matroska, &data, &size, (flags & 0x06) >> 1,
2815 &lace_size, &laces);
2820 if (track->audio.samplerate == 8000) {
2821 // If this is needed for more codecs, then add them here
2822 if (st->codec->codec_id == AV_CODEC_ID_AC3) {
2823 if (track->audio.samplerate != st->codec->sample_rate || !st->codec->frame_size)
2824 trust_default_duration = 0;
2828 if (!block_duration && trust_default_duration)
2829 block_duration = track->default_duration * laces / matroska->time_scale;
2831 if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
2832 track->end_timecode =
2833 FFMAX(track->end_timecode, timecode + block_duration);
2835 for (n = 0; n < laces; n++) {
2836 int64_t lace_duration = block_duration*(n+1) / laces - block_duration*n / laces;
2838 if (lace_size[n] > size) {
2839 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
2843 if ((st->codec->codec_id == AV_CODEC_ID_RA_288 ||
2844 st->codec->codec_id == AV_CODEC_ID_COOK ||
2845 st->codec->codec_id == AV_CODEC_ID_SIPR ||
2846 st->codec->codec_id == AV_CODEC_ID_ATRAC3) &&
2847 st->codec->block_align && track->audio.sub_packet_size) {
2848 res = matroska_parse_rm_audio(matroska, track, st, data,
2854 } else if (st->codec->codec_id == AV_CODEC_ID_WEBVTT) {
2855 res = matroska_parse_webvtt(matroska, track, st,
2857 timecode, lace_duration,
2862 res = matroska_parse_frame(matroska, track, st, data, lace_size[n],
2863 timecode, lace_duration, pos,
2864 !n ? is_keyframe : 0,
2865 additional, additional_id, additional_size,
2871 if (timecode != AV_NOPTS_VALUE)
2872 timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
2873 data += lace_size[n];
2874 size -= lace_size[n];
2882 static int matroska_parse_cluster_incremental(MatroskaDemuxContext *matroska)
2884 EbmlList *blocks_list;
2885 MatroskaBlock *blocks;
2887 res = ebml_parse(matroska,
2888 matroska_cluster_incremental_parsing,
2889 &matroska->current_cluster);
2892 if (matroska->current_cluster_pos)
2893 ebml_level_end(matroska);
2894 ebml_free(matroska_cluster, &matroska->current_cluster);
2895 memset(&matroska->current_cluster, 0, sizeof(MatroskaCluster));
2896 matroska->current_cluster_num_blocks = 0;
2897 matroska->current_cluster_pos = avio_tell(matroska->ctx->pb);
2898 matroska->prev_pkt = NULL;
2899 /* sizeof the ID which was already read */
2900 if (matroska->current_id)
2901 matroska->current_cluster_pos -= 4;
2902 res = ebml_parse(matroska,
2903 matroska_clusters_incremental,
2904 &matroska->current_cluster);
2905 /* Try parsing the block again. */
2907 res = ebml_parse(matroska,
2908 matroska_cluster_incremental_parsing,
2909 &matroska->current_cluster);
2913 matroska->current_cluster_num_blocks <
2914 matroska->current_cluster.blocks.nb_elem) {
2915 blocks_list = &matroska->current_cluster.blocks;
2916 blocks = blocks_list->elem;
2918 matroska->current_cluster_num_blocks = blocks_list->nb_elem;
2919 i = blocks_list->nb_elem - 1;
2920 if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
2921 int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
2922 uint8_t* additional = blocks[i].additional.size > 0 ?
2923 blocks[i].additional.data : NULL;
2924 if (!blocks[i].non_simple)
2925 blocks[i].duration = 0;
2926 res = matroska_parse_block(matroska, blocks[i].bin.data,
2927 blocks[i].bin.size, blocks[i].bin.pos,
2928 matroska->current_cluster.timecode,
2929 blocks[i].duration, is_keyframe,
2930 additional, blocks[i].additional_id,
2931 blocks[i].additional.size,
2932 matroska->current_cluster_pos,
2933 blocks[i].discard_padding);
2940 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
2942 MatroskaCluster cluster = { 0 };
2943 EbmlList *blocks_list;
2944 MatroskaBlock *blocks;
2948 if (!matroska->contains_ssa)
2949 return matroska_parse_cluster_incremental(matroska);
2950 pos = avio_tell(matroska->ctx->pb);
2951 matroska->prev_pkt = NULL;
2952 if (matroska->current_id)
2953 pos -= 4; /* sizeof the ID which was already read */
2954 res = ebml_parse(matroska, matroska_clusters, &cluster);
2955 blocks_list = &cluster.blocks;
2956 blocks = blocks_list->elem;
2957 for (i = 0; i < blocks_list->nb_elem; i++)
2958 if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
2959 int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
2960 res = matroska_parse_block(matroska, blocks[i].bin.data,
2961 blocks[i].bin.size, blocks[i].bin.pos,
2962 cluster.timecode, blocks[i].duration,
2963 is_keyframe, NULL, 0, 0, pos,
2964 blocks[i].discard_padding);
2966 ebml_free(matroska_cluster, &cluster);
2970 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
2972 MatroskaDemuxContext *matroska = s->priv_data;
2974 while (matroska_deliver_packet(matroska, pkt)) {
2975 int64_t pos = avio_tell(matroska->ctx->pb);
2978 if (matroska_parse_cluster(matroska) < 0)
2979 matroska_resync(matroska, pos);
2985 static int matroska_read_seek(AVFormatContext *s, int stream_index,
2986 int64_t timestamp, int flags)
2988 MatroskaDemuxContext *matroska = s->priv_data;
2989 MatroskaTrack *tracks = matroska->tracks.elem;
2990 AVStream *st = s->streams[stream_index];
2991 int i, index, index_sub, index_min;
2993 /* Parse the CUES now since we need the index data to seek. */
2994 if (matroska->cues_parsing_deferred > 0) {
2995 matroska->cues_parsing_deferred = 0;
2996 matroska_parse_cues(matroska);
2999 if (!st->nb_index_entries)
3001 timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
3003 if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
3004 avio_seek(s->pb, st->index_entries[st->nb_index_entries - 1].pos,
3006 matroska->current_id = 0;
3007 while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
3008 matroska_clear_queue(matroska);
3009 if (matroska_parse_cluster(matroska) < 0)
3014 matroska_clear_queue(matroska);
3015 if (index < 0 || (matroska->cues_parsing_deferred < 0 && index == st->nb_index_entries - 1))
3019 for (i = 0; i < matroska->tracks.nb_elem; i++) {
3020 tracks[i].audio.pkt_cnt = 0;
3021 tracks[i].audio.sub_packet_cnt = 0;
3022 tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
3023 tracks[i].end_timecode = 0;
3024 if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE &&
3025 tracks[i].stream->discard != AVDISCARD_ALL) {
3026 index_sub = av_index_search_timestamp(
3027 tracks[i].stream, st->index_entries[index].timestamp,
3028 AVSEEK_FLAG_BACKWARD);
3029 while (index_sub >= 0 &&
3031 tracks[i].stream->index_entries[index_sub].pos < st->index_entries[index_min].pos &&
3032 st->index_entries[index].timestamp - tracks[i].stream->index_entries[index_sub].timestamp < 30000000000 / matroska->time_scale)
3037 avio_seek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
3038 matroska->current_id = 0;
3039 if (flags & AVSEEK_FLAG_ANY) {
3040 st->skip_to_keyframe = 0;
3041 matroska->skip_to_timecode = timestamp;
3043 st->skip_to_keyframe = 1;
3044 matroska->skip_to_timecode = st->index_entries[index].timestamp;
3046 matroska->skip_to_keyframe = 1;
3048 matroska->num_levels = 0;
3049 ff_update_cur_dts(s, st, st->index_entries[index].timestamp);
3052 // slightly hackish but allows proper fallback to
3053 // the generic seeking code.
3054 matroska_clear_queue(matroska);
3055 matroska->current_id = 0;
3056 st->skip_to_keyframe =
3057 matroska->skip_to_keyframe = 0;
3059 matroska->num_levels = 0;
3063 static int matroska_read_close(AVFormatContext *s)
3065 MatroskaDemuxContext *matroska = s->priv_data;
3066 MatroskaTrack *tracks = matroska->tracks.elem;
3069 matroska_clear_queue(matroska);
3071 for (n = 0; n < matroska->tracks.nb_elem; n++)
3072 if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
3073 av_free(tracks[n].audio.buf);
3074 ebml_free(matroska_cluster, &matroska->current_cluster);
3075 ebml_free(matroska_segment, matroska);
3081 int64_t start_time_ns;
3082 int64_t end_time_ns;
3083 int64_t start_offset;
3087 /* This function searches all the Cues and returns the CueDesc corresponding the
3088 * the timestamp ts. Returned CueDesc will be such that start_time_ns <= ts <
3089 * end_time_ns. All 4 fields will be set to -1 if ts >= file's duration.
3091 static CueDesc get_cue_desc(AVFormatContext *s, int64_t ts, int64_t cues_start) {
3092 MatroskaDemuxContext *matroska = s->priv_data;
3095 int nb_index_entries = s->streams[0]->nb_index_entries;
3096 AVIndexEntry *index_entries = s->streams[0]->index_entries;
3097 if (ts >= matroska->duration * matroska->time_scale) return (CueDesc) {-1, -1, -1, -1};
3098 for (i = 1; i < nb_index_entries; i++) {
3099 if (index_entries[i - 1].timestamp * matroska->time_scale <= ts &&
3100 index_entries[i].timestamp * matroska->time_scale > ts) {
3105 cue_desc.start_time_ns = index_entries[i].timestamp * matroska->time_scale;
3106 cue_desc.start_offset = index_entries[i].pos - matroska->segment_start;
3107 if (i != nb_index_entries - 1) {
3108 cue_desc.end_time_ns = index_entries[i + 1].timestamp * matroska->time_scale;
3109 cue_desc.end_offset = index_entries[i + 1].pos - matroska->segment_start;
3111 cue_desc.end_time_ns = matroska->duration * matroska->time_scale;
3112 // FIXME: this needs special handling for files where Cues appear
3113 // before Clusters. the current logic assumes Cues appear after
3115 cue_desc.end_offset = cues_start - matroska->segment_start;
3120 static int webm_clusters_start_with_keyframe(AVFormatContext *s)
3122 MatroskaDemuxContext *matroska = s->priv_data;
3123 int64_t cluster_pos, before_pos;
3125 if (s->streams[0]->nb_index_entries <= 0) return 0;
3126 // seek to the first cluster using cues.
3127 index = av_index_search_timestamp(s->streams[0], 0, 0);
3128 if (index < 0) return 0;
3129 cluster_pos = s->streams[0]->index_entries[index].pos;
3130 before_pos = avio_tell(s->pb);
3132 int64_t cluster_id = 0, cluster_length = 0;
3134 avio_seek(s->pb, cluster_pos, SEEK_SET);
3135 // read cluster id and length
3136 ebml_read_num(matroska, matroska->ctx->pb, 4, &cluster_id);
3137 ebml_read_length(matroska, matroska->ctx->pb, &cluster_length);
3138 if (cluster_id != 0xF43B675) { // done with all clusters
3141 avio_seek(s->pb, cluster_pos, SEEK_SET);
3142 matroska->current_id = 0;
3143 matroska_clear_queue(matroska);
3144 if (matroska_parse_cluster(matroska) < 0 ||
3145 matroska->num_packets <= 0) {
3148 pkt = matroska->packets[0];
3149 cluster_pos += cluster_length + 12; // 12 is the offset of the cluster id and length.
3150 if (!(pkt->flags & AV_PKT_FLAG_KEY)) {
3155 avio_seek(s->pb, before_pos, SEEK_SET);
3159 static int buffer_size_after_time_downloaded(int64_t time_ns, double search_sec, int64_t bps,
3160 double min_buffer, double* buffer,
3161 double* sec_to_download, AVFormatContext *s,
3164 double nano_seconds_per_second = 1000000000.0;
3165 double time_sec = time_ns / nano_seconds_per_second;
3167 int64_t time_to_search_ns = (int64_t)(search_sec * nano_seconds_per_second);
3168 int64_t end_time_ns = time_ns + time_to_search_ns;
3169 double sec_downloaded = 0.0;
3170 CueDesc desc_curr = get_cue_desc(s, time_ns, cues_start);
3171 if (desc_curr.start_time_ns == -1)
3173 *sec_to_download = 0.0;
3175 // Check for non cue start time.
3176 if (time_ns > desc_curr.start_time_ns) {
3177 int64_t cue_nano = desc_curr.end_time_ns - time_ns;
3178 double percent = (double)(cue_nano) / (desc_curr.end_time_ns - desc_curr.start_time_ns);
3179 double cueBytes = (desc_curr.end_offset - desc_curr.start_offset) * percent;
3180 double timeToDownload = (cueBytes * 8.0) / bps;
3182 sec_downloaded += (cue_nano / nano_seconds_per_second) - timeToDownload;
3183 *sec_to_download += timeToDownload;
3185 // Check if the search ends within the first cue.
3186 if (desc_curr.end_time_ns >= end_time_ns) {
3187 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
3188 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
3189 sec_downloaded = percent_to_sub * sec_downloaded;
3190 *sec_to_download = percent_to_sub * *sec_to_download;
3193 if ((sec_downloaded + *buffer) <= min_buffer) {
3197 // Get the next Cue.
3198 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
3201 while (desc_curr.start_time_ns != -1) {
3202 int64_t desc_bytes = desc_curr.end_offset - desc_curr.start_offset;
3203 int64_t desc_ns = desc_curr.end_time_ns - desc_curr.start_time_ns;
3204 double desc_sec = desc_ns / nano_seconds_per_second;
3205 double bits = (desc_bytes * 8.0);
3206 double time_to_download = bits / bps;
3208 sec_downloaded += desc_sec - time_to_download;
3209 *sec_to_download += time_to_download;
3211 if (desc_curr.end_time_ns >= end_time_ns) {
3212 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
3213 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
3214 sec_downloaded = percent_to_sub * sec_downloaded;
3215 *sec_to_download = percent_to_sub * *sec_to_download;
3217 if ((sec_downloaded + *buffer) <= min_buffer)
3222 if ((sec_downloaded + *buffer) <= min_buffer) {
3227 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
3229 *buffer = *buffer + sec_downloaded;
3233 /* This function computes the bandwidth of the WebM file with the help of
3234 * buffer_size_after_time_downloaded() function. Both of these functions are
3235 * adapted from WebM Tools project and are adapted to work with FFmpeg's
3236 * Matroska parsing mechanism.
3238 * Returns the bandwidth of the file on success; -1 on error.
3240 static int64_t webm_dash_manifest_compute_bandwidth(AVFormatContext *s, int64_t cues_start)
3242 MatroskaDemuxContext *matroska = s->priv_data;
3243 AVStream *st = s->streams[0];
3244 double bandwidth = 0.0;
3247 for (i = 0; i < st->nb_index_entries; i++) {
3248 int64_t prebuffer_ns = 1000000000;
3249 int64_t time_ns = st->index_entries[i].timestamp * matroska->time_scale;
3250 double nano_seconds_per_second = 1000000000.0;
3251 int64_t prebuffered_ns = time_ns + prebuffer_ns;
3252 double prebuffer_bytes = 0.0;
3253 int64_t temp_prebuffer_ns = prebuffer_ns;
3254 int64_t pre_bytes, pre_ns;
3255 double pre_sec, prebuffer, bits_per_second;
3256 CueDesc desc_beg = get_cue_desc(s, time_ns, cues_start);
3258 // Start with the first Cue.
3259 CueDesc desc_end = desc_beg;
3261 // Figure out how much data we have downloaded for the prebuffer. This will
3262 // be used later to adjust the bits per sample to try.
3263 while (desc_end.start_time_ns != -1 && desc_end.end_time_ns < prebuffered_ns) {
3264 // Prebuffered the entire Cue.
3265 prebuffer_bytes += desc_end.end_offset - desc_end.start_offset;
3266 temp_prebuffer_ns -= desc_end.end_time_ns - desc_end.start_time_ns;
3267 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
3269 if (desc_end.start_time_ns == -1) {
3270 // The prebuffer is larger than the duration.
3271 return (matroska->duration * matroska->time_scale >= prebuffered_ns) ? -1 : 0;
3274 // The prebuffer ends in the last Cue. Estimate how much data was
3276 pre_bytes = desc_end.end_offset - desc_end.start_offset;
3277 pre_ns = desc_end.end_time_ns - desc_end.start_time_ns;
3278 pre_sec = pre_ns / nano_seconds_per_second;
3280 pre_bytes * ((temp_prebuffer_ns / nano_seconds_per_second) / pre_sec);
3282 prebuffer = prebuffer_ns / nano_seconds_per_second;
3284 // Set this to 0.0 in case our prebuffer buffers the entire video.
3285 bits_per_second = 0.0;
3287 int64_t desc_bytes = desc_end.end_offset - desc_beg.start_offset;
3288 int64_t desc_ns = desc_end.end_time_ns - desc_beg.start_time_ns;
3289 double desc_sec = desc_ns / nano_seconds_per_second;
3290 double calc_bits_per_second = (desc_bytes * 8) / desc_sec;
3292 // Drop the bps by the percentage of bytes buffered.
3293 double percent = (desc_bytes - prebuffer_bytes) / desc_bytes;
3294 double mod_bits_per_second = calc_bits_per_second * percent;
3296 if (prebuffer < desc_sec) {
3298 (double)(matroska->duration * matroska->time_scale) / nano_seconds_per_second;
3300 // Add 1 so the bits per second should be a little bit greater than file
3302 int64_t bps = (int64_t)(mod_bits_per_second) + 1;
3303 const double min_buffer = 0.0;
3304 double buffer = prebuffer;
3305 double sec_to_download = 0.0;
3307 int rv = buffer_size_after_time_downloaded(prebuffered_ns, search_sec, bps,
3308 min_buffer, &buffer, &sec_to_download,
3312 } else if (rv == 0) {
3313 bits_per_second = (double)(bps);
3318 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
3319 } while (desc_end.start_time_ns != -1);
3320 if (bandwidth < bits_per_second) bandwidth = bits_per_second;
3322 return (int64_t)bandwidth;
3325 static int webm_dash_manifest_cues(AVFormatContext *s)
3327 MatroskaDemuxContext *matroska = s->priv_data;
3328 EbmlList *seekhead_list = &matroska->seekhead;
3329 MatroskaSeekhead *seekhead = seekhead_list->elem;
3331 int64_t cues_start = -1, cues_end = -1, before_pos, bandwidth;
3334 // determine cues start and end positions
3335 for (i = 0; i < seekhead_list->nb_elem; i++)
3336 if (seekhead[i].id == MATROSKA_ID_CUES)
3339 if (i >= seekhead_list->nb_elem) return -1;
3341 before_pos = avio_tell(matroska->ctx->pb);
3342 cues_start = seekhead[i].pos + matroska->segment_start;
3343 if (avio_seek(matroska->ctx->pb, cues_start, SEEK_SET) == cues_start) {
3344 uint64_t cues_length = 0, cues_id = 0;
3345 ebml_read_num(matroska, matroska->ctx->pb, 4, &cues_id);
3346 ebml_read_length(matroska, matroska->ctx->pb, &cues_length);
3347 cues_end = cues_start + cues_length + 11; // 11 is the offset of Cues ID.
3349 avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
3350 if (cues_start == -1 || cues_end == -1) return -1;
3353 matroska_parse_cues(matroska);
3356 av_dict_set_int(&s->streams[0]->metadata, CUES_START, cues_start, 0);
3359 av_dict_set_int(&s->streams[0]->metadata, CUES_END, cues_end, 0);
3362 bandwidth = webm_dash_manifest_compute_bandwidth(s, cues_start);
3363 if (bandwidth < 0) return -1;
3364 av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH, bandwidth, 0);
3366 // check if all clusters start with key frames
3367 av_dict_set_int(&s->streams[0]->metadata, CLUSTER_KEYFRAME, webm_clusters_start_with_keyframe(s), 0);
3369 // store cue point timestamps as a comma separated list for checking subsegment alignment in
3370 // the muxer. assumes that each timestamp cannot be more than 20 characters long.
3371 buf = av_malloc(s->streams[0]->nb_index_entries * 20 * sizeof(char));
3372 if (!buf) return -1;
3374 for (i = 0; i < s->streams[0]->nb_index_entries; i++) {
3375 snprintf(buf, (i + 1) * 20 * sizeof(char),
3376 "%s%" PRId64, buf, s->streams[0]->index_entries[i].timestamp);
3377 if (i != s->streams[0]->nb_index_entries - 1)
3378 strncat(buf, ",", sizeof(char));
3380 av_dict_set(&s->streams[0]->metadata, CUE_TIMESTAMPS, buf, 0);
3386 static int webm_dash_manifest_read_header(AVFormatContext *s)
3389 int ret = matroska_read_header(s);
3390 MatroskaTrack *tracks;
3391 MatroskaDemuxContext *matroska = s->priv_data;
3393 av_log(s, AV_LOG_ERROR, "Failed to read file headers\n");
3397 // initialization range
3398 // 5 is the offset of Cluster ID.
3399 av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, avio_tell(s->pb) - 5, 0);
3401 // basename of the file
3402 buf = strrchr(s->filename, '/');
3403 if (!buf) return -1;
3404 av_dict_set(&s->streams[0]->metadata, FILENAME, ++buf, 0);
3407 buf = av_asprintf("%g", matroska->duration);
3408 if (!buf) return AVERROR(ENOMEM);
3409 av_dict_set(&s->streams[0]->metadata, DURATION, buf, 0);
3413 tracks = matroska->tracks.elem;
3414 av_dict_set_int(&s->streams[0]->metadata, TRACK_NUMBER, tracks[0].num, 0);
3416 // parse the cues and populate Cue related fields
3417 return webm_dash_manifest_cues(s);
3420 static int webm_dash_manifest_read_packet(AVFormatContext *s, AVPacket *pkt)
3425 AVInputFormat ff_matroska_demuxer = {
3426 .name = "matroska,webm",
3427 .long_name = NULL_IF_CONFIG_SMALL("Matroska / WebM"),
3428 .extensions = "mkv,mk3d,mka,mks",
3429 .priv_data_size = sizeof(MatroskaDemuxContext),
3430 .read_probe = matroska_probe,
3431 .read_header = matroska_read_header,
3432 .read_packet = matroska_read_packet,
3433 .read_close = matroska_read_close,
3434 .read_seek = matroska_read_seek,
3435 .mime_type = "audio/webm,audio/x-matroska,video/webm,video/x-matroska"
3438 AVInputFormat ff_webm_dash_manifest_demuxer = {
3439 .name = "webm_dash_manifest",
3440 .long_name = NULL_IF_CONFIG_SMALL("WebM DASH Manifest"),
3441 .priv_data_size = sizeof(MatroskaDemuxContext),
3442 .read_header = webm_dash_manifest_read_header,
3443 .read_packet = webm_dash_manifest_read_packet,
3444 .read_close = matroska_read_close,