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/
36 #include "libavutil/avstring.h"
37 #include "libavutil/base64.h"
38 #include "libavutil/dict.h"
39 #include "libavutil/intfloat.h"
40 #include "libavutil/intreadwrite.h"
41 #include "libavutil/lzo.h"
42 #include "libavutil/mathematics.h"
43 #include "libavutil/time_internal.h"
45 #include "libavcodec/bytestream.h"
46 #include "libavcodec/flac.h"
47 #include "libavcodec/mpeg4audio.h"
50 #include "avio_internal.h"
55 /* For ff_codec_get_id(). */
80 typedef const struct EbmlSyntax {
89 const struct EbmlSyntax *n;
109 uint64_t doctype_version;
115 } MatroskaTrackCompression;
120 } MatroskaTrackEncryption;
125 MatroskaTrackCompression compression;
126 MatroskaTrackEncryption encryption;
127 } MatroskaTrackEncoding;
131 uint64_t display_width;
132 uint64_t display_height;
133 uint64_t pixel_width;
134 uint64_t pixel_height;
136 uint64_t stereo_mode;
138 } MatroskaTrackVideo;
142 double out_samplerate;
146 /* real audio header (extracted from extradata) */
153 uint64_t buf_timecode;
155 } MatroskaTrackAudio;
160 } MatroskaTrackPlane;
163 EbmlList combine_planes;
164 } MatroskaTrackOperation;
175 uint64_t default_duration;
176 uint64_t flag_default;
177 uint64_t flag_forced;
178 uint64_t seek_preroll;
179 MatroskaTrackVideo video;
180 MatroskaTrackAudio audio;
181 MatroskaTrackOperation operation;
183 uint64_t codec_delay;
186 int64_t end_timecode;
188 uint64_t max_block_additional_id;
198 } MatroskaAttachment;
236 MatroskaTagTarget target;
256 AVFormatContext *ctx;
260 MatroskaLevel levels[EBML_MAX_DEPTH];
270 EbmlList attachments;
276 /* byte position of the segment inside the stream */
277 int64_t segment_start;
279 /* the packet queue */
286 /* What to skip before effectively reading a packet. */
287 int skip_to_keyframe;
288 uint64_t skip_to_timecode;
290 /* File has a CUES element, but we defer parsing until it is needed. */
291 int cues_parsing_deferred;
293 int current_cluster_num_blocks;
294 int64_t current_cluster_pos;
295 MatroskaCluster current_cluster;
297 /* File has SSA subtitles which prevent incremental cluster parsing. */
299 } MatroskaDemuxContext;
306 uint64_t additional_id;
308 int64_t discard_padding;
311 static EbmlSyntax ebml_header[] = {
312 { EBML_ID_EBMLREADVERSION, EBML_UINT, 0, offsetof(Ebml, version), { .u = EBML_VERSION } },
313 { EBML_ID_EBMLMAXSIZELENGTH, EBML_UINT, 0, offsetof(Ebml, max_size), { .u = 8 } },
314 { EBML_ID_EBMLMAXIDLENGTH, EBML_UINT, 0, offsetof(Ebml, id_length), { .u = 4 } },
315 { EBML_ID_DOCTYPE, EBML_STR, 0, offsetof(Ebml, doctype), { .s = "(none)" } },
316 { EBML_ID_DOCTYPEREADVERSION, EBML_UINT, 0, offsetof(Ebml, doctype_version), { .u = 1 } },
317 { EBML_ID_EBMLVERSION, EBML_NONE },
318 { EBML_ID_DOCTYPEVERSION, EBML_NONE },
322 static EbmlSyntax ebml_syntax[] = {
323 { EBML_ID_HEADER, EBML_NEST, 0, 0, { .n = ebml_header } },
327 static EbmlSyntax matroska_info[] = {
328 { MATROSKA_ID_TIMECODESCALE, EBML_UINT, 0, offsetof(MatroskaDemuxContext, time_scale), { .u = 1000000 } },
329 { MATROSKA_ID_DURATION, EBML_FLOAT, 0, offsetof(MatroskaDemuxContext, duration) },
330 { MATROSKA_ID_TITLE, EBML_UTF8, 0, offsetof(MatroskaDemuxContext, title) },
331 { MATROSKA_ID_WRITINGAPP, EBML_NONE },
332 { MATROSKA_ID_MUXINGAPP, EBML_UTF8, 0, offsetof(MatroskaDemuxContext, muxingapp) },
333 { MATROSKA_ID_DATEUTC, EBML_BIN, 0, offsetof(MatroskaDemuxContext, date_utc) },
334 { MATROSKA_ID_SEGMENTUID, EBML_NONE },
338 static EbmlSyntax matroska_track_video[] = {
339 { MATROSKA_ID_VIDEOFRAMERATE, EBML_FLOAT, 0, offsetof(MatroskaTrackVideo, frame_rate) },
340 { MATROSKA_ID_VIDEODISPLAYWIDTH, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_width), { .u=-1 } },
341 { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_height), { .u=-1 } },
342 { MATROSKA_ID_VIDEOPIXELWIDTH, EBML_UINT, 0, offsetof(MatroskaTrackVideo, pixel_width) },
343 { MATROSKA_ID_VIDEOPIXELHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, pixel_height) },
344 { MATROSKA_ID_VIDEOCOLORSPACE, EBML_BIN, 0, offsetof(MatroskaTrackVideo, color_space) },
345 { MATROSKA_ID_VIDEOALPHAMODE, EBML_UINT, 0, offsetof(MatroskaTrackVideo, alpha_mode) },
346 { MATROSKA_ID_VIDEOPIXELCROPB, EBML_NONE },
347 { MATROSKA_ID_VIDEOPIXELCROPT, EBML_NONE },
348 { MATROSKA_ID_VIDEOPIXELCROPL, EBML_NONE },
349 { MATROSKA_ID_VIDEOPIXELCROPR, EBML_NONE },
350 { MATROSKA_ID_VIDEODISPLAYUNIT, EBML_NONE },
351 { MATROSKA_ID_VIDEOFLAGINTERLACED, EBML_NONE },
352 { MATROSKA_ID_VIDEOSTEREOMODE, EBML_UINT, 0, offsetof(MatroskaTrackVideo, stereo_mode), { .u = MATROSKA_VIDEO_STEREOMODE_TYPE_NB } },
353 { MATROSKA_ID_VIDEOASPECTRATIO, EBML_NONE },
357 static EbmlSyntax matroska_track_audio[] = {
358 { MATROSKA_ID_AUDIOSAMPLINGFREQ, EBML_FLOAT, 0, offsetof(MatroskaTrackAudio, samplerate), { .f = 8000.0 } },
359 { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, EBML_FLOAT, 0, offsetof(MatroskaTrackAudio, out_samplerate) },
360 { MATROSKA_ID_AUDIOBITDEPTH, EBML_UINT, 0, offsetof(MatroskaTrackAudio, bitdepth) },
361 { MATROSKA_ID_AUDIOCHANNELS, EBML_UINT, 0, offsetof(MatroskaTrackAudio, channels), { .u = 1 } },
365 static EbmlSyntax matroska_track_encoding_compression[] = {
366 { MATROSKA_ID_ENCODINGCOMPALGO, EBML_UINT, 0, offsetof(MatroskaTrackCompression, algo), { .u = 0 } },
367 { MATROSKA_ID_ENCODINGCOMPSETTINGS, EBML_BIN, 0, offsetof(MatroskaTrackCompression, settings) },
371 static EbmlSyntax matroska_track_encoding_encryption[] = {
372 { MATROSKA_ID_ENCODINGENCALGO, EBML_UINT, 0, offsetof(MatroskaTrackEncryption,algo), {.u = 0} },
373 { MATROSKA_ID_ENCODINGENCKEYID, EBML_BIN, 0, offsetof(MatroskaTrackEncryption,key_id) },
374 { MATROSKA_ID_ENCODINGENCAESSETTINGS, EBML_NONE },
375 { MATROSKA_ID_ENCODINGSIGALGO, EBML_NONE },
376 { MATROSKA_ID_ENCODINGSIGHASHALGO, EBML_NONE },
377 { MATROSKA_ID_ENCODINGSIGKEYID, EBML_NONE },
378 { MATROSKA_ID_ENCODINGSIGNATURE, EBML_NONE },
381 static EbmlSyntax matroska_track_encoding[] = {
382 { MATROSKA_ID_ENCODINGSCOPE, EBML_UINT, 0, offsetof(MatroskaTrackEncoding, scope), { .u = 1 } },
383 { MATROSKA_ID_ENCODINGTYPE, EBML_UINT, 0, offsetof(MatroskaTrackEncoding, type), { .u = 0 } },
384 { MATROSKA_ID_ENCODINGCOMPRESSION, EBML_NEST, 0, offsetof(MatroskaTrackEncoding, compression), { .n = matroska_track_encoding_compression } },
385 { MATROSKA_ID_ENCODINGENCRYPTION, EBML_NEST, 0, offsetof(MatroskaTrackEncoding, encryption), { .n = matroska_track_encoding_encryption } },
386 { MATROSKA_ID_ENCODINGORDER, EBML_NONE },
390 static EbmlSyntax matroska_track_encodings[] = {
391 { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack, encodings), { .n = matroska_track_encoding } },
395 static EbmlSyntax matroska_track_plane[] = {
396 { MATROSKA_ID_TRACKPLANEUID, EBML_UINT, 0, offsetof(MatroskaTrackPlane,uid) },
397 { MATROSKA_ID_TRACKPLANETYPE, EBML_UINT, 0, offsetof(MatroskaTrackPlane,type) },
401 static EbmlSyntax matroska_track_combine_planes[] = {
402 { MATROSKA_ID_TRACKPLANE, EBML_NEST, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n = matroska_track_plane} },
406 static EbmlSyntax matroska_track_operation[] = {
407 { MATROSKA_ID_TRACKCOMBINEPLANES, EBML_NEST, 0, 0, {.n = matroska_track_combine_planes} },
411 static EbmlSyntax matroska_track[] = {
412 { MATROSKA_ID_TRACKNUMBER, EBML_UINT, 0, offsetof(MatroskaTrack, num) },
413 { MATROSKA_ID_TRACKNAME, EBML_UTF8, 0, offsetof(MatroskaTrack, name) },
414 { MATROSKA_ID_TRACKUID, EBML_UINT, 0, offsetof(MatroskaTrack, uid) },
415 { MATROSKA_ID_TRACKTYPE, EBML_UINT, 0, offsetof(MatroskaTrack, type) },
416 { MATROSKA_ID_CODECID, EBML_STR, 0, offsetof(MatroskaTrack, codec_id) },
417 { MATROSKA_ID_CODECPRIVATE, EBML_BIN, 0, offsetof(MatroskaTrack, codec_priv) },
418 { MATROSKA_ID_CODECDELAY, EBML_UINT, 0, offsetof(MatroskaTrack, codec_delay) },
419 { MATROSKA_ID_TRACKLANGUAGE, EBML_UTF8, 0, offsetof(MatroskaTrack, language), { .s = "eng" } },
420 { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack, default_duration) },
421 { MATROSKA_ID_TRACKTIMECODESCALE, EBML_FLOAT, 0, offsetof(MatroskaTrack, time_scale), { .f = 1.0 } },
422 { MATROSKA_ID_TRACKFLAGDEFAULT, EBML_UINT, 0, offsetof(MatroskaTrack, flag_default), { .u = 1 } },
423 { MATROSKA_ID_TRACKFLAGFORCED, EBML_UINT, 0, offsetof(MatroskaTrack, flag_forced), { .u = 0 } },
424 { MATROSKA_ID_TRACKVIDEO, EBML_NEST, 0, offsetof(MatroskaTrack, video), { .n = matroska_track_video } },
425 { MATROSKA_ID_TRACKAUDIO, EBML_NEST, 0, offsetof(MatroskaTrack, audio), { .n = matroska_track_audio } },
426 { MATROSKA_ID_TRACKOPERATION, EBML_NEST, 0, offsetof(MatroskaTrack, operation), { .n = matroska_track_operation } },
427 { MATROSKA_ID_TRACKCONTENTENCODINGS, EBML_NEST, 0, 0, { .n = matroska_track_encodings } },
428 { MATROSKA_ID_TRACKMAXBLKADDID, EBML_UINT, 0, offsetof(MatroskaTrack, max_block_additional_id) },
429 { MATROSKA_ID_SEEKPREROLL, EBML_UINT, 0, offsetof(MatroskaTrack, seek_preroll) },
430 { MATROSKA_ID_TRACKFLAGENABLED, EBML_NONE },
431 { MATROSKA_ID_TRACKFLAGLACING, EBML_NONE },
432 { MATROSKA_ID_CODECNAME, EBML_NONE },
433 { MATROSKA_ID_CODECDECODEALL, EBML_NONE },
434 { MATROSKA_ID_CODECINFOURL, EBML_NONE },
435 { MATROSKA_ID_CODECDOWNLOADURL, EBML_NONE },
436 { MATROSKA_ID_TRACKMINCACHE, EBML_NONE },
437 { MATROSKA_ID_TRACKMAXCACHE, EBML_NONE },
441 static EbmlSyntax matroska_tracks[] = {
442 { MATROSKA_ID_TRACKENTRY, EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext, tracks), { .n = matroska_track } },
446 static EbmlSyntax matroska_attachment[] = {
447 { MATROSKA_ID_FILEUID, EBML_UINT, 0, offsetof(MatroskaAttachment, uid) },
448 { MATROSKA_ID_FILENAME, EBML_UTF8, 0, offsetof(MatroskaAttachment, filename) },
449 { MATROSKA_ID_FILEMIMETYPE, EBML_STR, 0, offsetof(MatroskaAttachment, mime) },
450 { MATROSKA_ID_FILEDATA, EBML_BIN, 0, offsetof(MatroskaAttachment, bin) },
451 { MATROSKA_ID_FILEDESC, EBML_NONE },
455 static EbmlSyntax matroska_attachments[] = {
456 { MATROSKA_ID_ATTACHEDFILE, EBML_NEST, sizeof(MatroskaAttachment), offsetof(MatroskaDemuxContext, attachments), { .n = matroska_attachment } },
460 static EbmlSyntax matroska_chapter_display[] = {
461 { MATROSKA_ID_CHAPSTRING, EBML_UTF8, 0, offsetof(MatroskaChapter, title) },
462 { MATROSKA_ID_CHAPLANG, EBML_NONE },
466 static EbmlSyntax matroska_chapter_entry[] = {
467 { MATROSKA_ID_CHAPTERTIMESTART, EBML_UINT, 0, offsetof(MatroskaChapter, start), { .u = AV_NOPTS_VALUE } },
468 { MATROSKA_ID_CHAPTERTIMEEND, EBML_UINT, 0, offsetof(MatroskaChapter, end), { .u = AV_NOPTS_VALUE } },
469 { MATROSKA_ID_CHAPTERUID, EBML_UINT, 0, offsetof(MatroskaChapter, uid) },
470 { MATROSKA_ID_CHAPTERDISPLAY, EBML_NEST, 0, 0, { .n = matroska_chapter_display } },
471 { MATROSKA_ID_CHAPTERFLAGHIDDEN, EBML_NONE },
472 { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
473 { MATROSKA_ID_CHAPTERPHYSEQUIV, EBML_NONE },
474 { MATROSKA_ID_CHAPTERATOM, EBML_NONE },
478 static EbmlSyntax matroska_chapter[] = {
479 { MATROSKA_ID_CHAPTERATOM, EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext, chapters), { .n = matroska_chapter_entry } },
480 { MATROSKA_ID_EDITIONUID, EBML_NONE },
481 { MATROSKA_ID_EDITIONFLAGHIDDEN, EBML_NONE },
482 { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
483 { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
487 static EbmlSyntax matroska_chapters[] = {
488 { MATROSKA_ID_EDITIONENTRY, EBML_NEST, 0, 0, { .n = matroska_chapter } },
492 static EbmlSyntax matroska_index_pos[] = {
493 { MATROSKA_ID_CUETRACK, EBML_UINT, 0, offsetof(MatroskaIndexPos, track) },
494 { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos, pos) },
495 { MATROSKA_ID_CUERELATIVEPOSITION,EBML_NONE },
496 { MATROSKA_ID_CUEDURATION, EBML_NONE },
497 { MATROSKA_ID_CUEBLOCKNUMBER, EBML_NONE },
501 static EbmlSyntax matroska_index_entry[] = {
502 { MATROSKA_ID_CUETIME, EBML_UINT, 0, offsetof(MatroskaIndex, time) },
503 { MATROSKA_ID_CUETRACKPOSITION, EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex, pos), { .n = matroska_index_pos } },
507 static EbmlSyntax matroska_index[] = {
508 { MATROSKA_ID_POINTENTRY, EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext, index), { .n = matroska_index_entry } },
512 static EbmlSyntax matroska_simpletag[] = {
513 { MATROSKA_ID_TAGNAME, EBML_UTF8, 0, offsetof(MatroskaTag, name) },
514 { MATROSKA_ID_TAGSTRING, EBML_UTF8, 0, offsetof(MatroskaTag, string) },
515 { MATROSKA_ID_TAGLANG, EBML_STR, 0, offsetof(MatroskaTag, lang), { .s = "und" } },
516 { MATROSKA_ID_TAGDEFAULT, EBML_UINT, 0, offsetof(MatroskaTag, def) },
517 { MATROSKA_ID_TAGDEFAULT_BUG, EBML_UINT, 0, offsetof(MatroskaTag, def) },
518 { MATROSKA_ID_SIMPLETAG, EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag, sub), { .n = matroska_simpletag } },
522 static EbmlSyntax matroska_tagtargets[] = {
523 { MATROSKA_ID_TAGTARGETS_TYPE, EBML_STR, 0, offsetof(MatroskaTagTarget, type) },
524 { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget, typevalue), { .u = 50 } },
525 { MATROSKA_ID_TAGTARGETS_TRACKUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, trackuid) },
526 { MATROSKA_ID_TAGTARGETS_CHAPTERUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, chapteruid) },
527 { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, attachuid) },
531 static EbmlSyntax matroska_tag[] = {
532 { MATROSKA_ID_SIMPLETAG, EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags, tag), { .n = matroska_simpletag } },
533 { MATROSKA_ID_TAGTARGETS, EBML_NEST, 0, offsetof(MatroskaTags, target), { .n = matroska_tagtargets } },
537 static EbmlSyntax matroska_tags[] = {
538 { MATROSKA_ID_TAG, EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext, tags), { .n = matroska_tag } },
542 static EbmlSyntax matroska_seekhead_entry[] = {
543 { MATROSKA_ID_SEEKID, EBML_UINT, 0, offsetof(MatroskaSeekhead, id) },
544 { MATROSKA_ID_SEEKPOSITION, EBML_UINT, 0, offsetof(MatroskaSeekhead, pos), { .u = -1 } },
548 static EbmlSyntax matroska_seekhead[] = {
549 { MATROSKA_ID_SEEKENTRY, EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext, seekhead), { .n = matroska_seekhead_entry } },
553 static EbmlSyntax matroska_segment[] = {
554 { MATROSKA_ID_INFO, EBML_NEST, 0, 0, { .n = matroska_info } },
555 { MATROSKA_ID_TRACKS, EBML_NEST, 0, 0, { .n = matroska_tracks } },
556 { MATROSKA_ID_ATTACHMENTS, EBML_NEST, 0, 0, { .n = matroska_attachments } },
557 { MATROSKA_ID_CHAPTERS, EBML_NEST, 0, 0, { .n = matroska_chapters } },
558 { MATROSKA_ID_CUES, EBML_NEST, 0, 0, { .n = matroska_index } },
559 { MATROSKA_ID_TAGS, EBML_NEST, 0, 0, { .n = matroska_tags } },
560 { MATROSKA_ID_SEEKHEAD, EBML_NEST, 0, 0, { .n = matroska_seekhead } },
561 { MATROSKA_ID_CLUSTER, EBML_STOP },
565 static EbmlSyntax matroska_segments[] = {
566 { MATROSKA_ID_SEGMENT, EBML_NEST, 0, 0, { .n = matroska_segment } },
570 static EbmlSyntax matroska_blockmore[] = {
571 { MATROSKA_ID_BLOCKADDID, EBML_UINT, 0, offsetof(MatroskaBlock,additional_id) },
572 { MATROSKA_ID_BLOCKADDITIONAL, EBML_BIN, 0, offsetof(MatroskaBlock,additional) },
576 static EbmlSyntax matroska_blockadditions[] = {
577 { MATROSKA_ID_BLOCKMORE, EBML_NEST, 0, 0, {.n = matroska_blockmore} },
581 static EbmlSyntax matroska_blockgroup[] = {
582 { MATROSKA_ID_BLOCK, EBML_BIN, 0, offsetof(MatroskaBlock, bin) },
583 { MATROSKA_ID_BLOCKADDITIONS, EBML_NEST, 0, 0, { .n = matroska_blockadditions} },
584 { MATROSKA_ID_SIMPLEBLOCK, EBML_BIN, 0, offsetof(MatroskaBlock, bin) },
585 { MATROSKA_ID_BLOCKDURATION, EBML_UINT, 0, offsetof(MatroskaBlock, duration) },
586 { MATROSKA_ID_DISCARDPADDING, EBML_SINT, 0, offsetof(MatroskaBlock, discard_padding) },
587 { MATROSKA_ID_BLOCKREFERENCE, EBML_SINT, 0, offsetof(MatroskaBlock, reference) },
588 { MATROSKA_ID_CODECSTATE, EBML_NONE },
589 { 1, EBML_UINT, 0, offsetof(MatroskaBlock, non_simple), { .u = 1 } },
593 static EbmlSyntax matroska_cluster[] = {
594 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
595 { MATROSKA_ID_BLOCKGROUP, EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
596 { MATROSKA_ID_SIMPLEBLOCK, EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
597 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
598 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
602 static EbmlSyntax matroska_clusters[] = {
603 { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, { .n = matroska_cluster } },
604 { MATROSKA_ID_INFO, EBML_NONE },
605 { MATROSKA_ID_CUES, EBML_NONE },
606 { MATROSKA_ID_TAGS, EBML_NONE },
607 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
611 static EbmlSyntax matroska_cluster_incremental_parsing[] = {
612 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
613 { MATROSKA_ID_BLOCKGROUP, EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
614 { MATROSKA_ID_SIMPLEBLOCK, EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster, blocks), { .n = matroska_blockgroup } },
615 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
616 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
617 { MATROSKA_ID_INFO, EBML_NONE },
618 { MATROSKA_ID_CUES, EBML_NONE },
619 { MATROSKA_ID_TAGS, EBML_NONE },
620 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
621 { MATROSKA_ID_CLUSTER, EBML_STOP },
625 static EbmlSyntax matroska_cluster_incremental[] = {
626 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
627 { MATROSKA_ID_BLOCKGROUP, EBML_STOP },
628 { MATROSKA_ID_SIMPLEBLOCK, EBML_STOP },
629 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
630 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
634 static EbmlSyntax matroska_clusters_incremental[] = {
635 { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, { .n = matroska_cluster_incremental } },
636 { MATROSKA_ID_INFO, EBML_NONE },
637 { MATROSKA_ID_CUES, EBML_NONE },
638 { MATROSKA_ID_TAGS, EBML_NONE },
639 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
643 static const char *const matroska_doctypes[] = { "matroska", "webm" };
645 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
647 AVIOContext *pb = matroska->ctx->pb;
649 matroska->current_id = 0;
650 matroska->num_levels = 0;
652 /* seek to next position to resync from */
653 if (avio_seek(pb, last_pos + 1, SEEK_SET) < 0)
658 // try to find a toplevel element
659 while (!avio_feof(pb)) {
660 if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
661 id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
662 id == MATROSKA_ID_SEEKHEAD || id == MATROSKA_ID_ATTACHMENTS ||
663 id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS) {
664 matroska->current_id = id;
667 id = (id << 8) | avio_r8(pb);
676 * Return: Whether we reached the end of a level in the hierarchy or not.
678 static int ebml_level_end(MatroskaDemuxContext *matroska)
680 AVIOContext *pb = matroska->ctx->pb;
681 int64_t pos = avio_tell(pb);
683 if (matroska->num_levels > 0) {
684 MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
685 if (pos - level->start >= level->length || matroska->current_id) {
686 matroska->num_levels--;
694 * Read: an "EBML number", which is defined as a variable-length
695 * array of bytes. The first byte indicates the length by giving a
696 * number of 0-bits followed by a one. The position of the first
697 * "one" bit inside the first byte indicates the length of this
699 * Returns: number of bytes read, < 0 on error
701 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
702 int max_size, uint64_t *number)
707 /* The first byte tells us the length in bytes - avio_r8() can normally
708 * return 0, but since that's not a valid first ebmlID byte, we can
709 * use it safely here to catch EOS. */
710 if (!(total = avio_r8(pb))) {
711 /* we might encounter EOS here */
712 if (!avio_feof(pb)) {
713 int64_t pos = avio_tell(pb);
714 av_log(matroska->ctx, AV_LOG_ERROR,
715 "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
717 return pb->error ? pb->error : AVERROR(EIO);
722 /* get the length of the EBML number */
723 read = 8 - ff_log2_tab[total];
724 if (read > max_size) {
725 int64_t pos = avio_tell(pb) - 1;
726 av_log(matroska->ctx, AV_LOG_ERROR,
727 "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
728 (uint8_t) total, pos, pos);
729 return AVERROR_INVALIDDATA;
732 /* read out length */
733 total ^= 1 << ff_log2_tab[total];
735 total = (total << 8) | avio_r8(pb);
743 * Read a EBML length value.
744 * This needs special handling for the "unknown length" case which has multiple
747 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
750 int res = ebml_read_num(matroska, pb, 8, number);
751 if (res > 0 && *number + 1 == 1ULL << (7 * res))
752 *number = 0xffffffffffffffULL;
757 * Read the next element as an unsigned int.
758 * 0 is success, < 0 is failure.
760 static int ebml_read_uint(AVIOContext *pb, int size, uint64_t *num)
765 return AVERROR_INVALIDDATA;
767 /* big-endian ordering; build up number */
770 *num = (*num << 8) | avio_r8(pb);
776 * Read the next element as a signed int.
777 * 0 is success, < 0 is failure.
779 static int ebml_read_sint(AVIOContext *pb, int size, int64_t *num)
784 return AVERROR_INVALIDDATA;
789 *num = sign_extend(avio_r8(pb), 8);
791 /* big-endian ordering; build up number */
793 *num = (*num << 8) | avio_r8(pb);
800 * Read the next element as a float.
801 * 0 is success, < 0 is failure.
803 static int ebml_read_float(AVIOContext *pb, int size, double *num)
808 *num = av_int2float(avio_rb32(pb));
810 *num = av_int2double(avio_rb64(pb));
812 return AVERROR_INVALIDDATA;
818 * Read the next element as an ASCII string.
819 * 0 is success, < 0 is failure.
821 static int ebml_read_ascii(AVIOContext *pb, int size, char **str)
825 /* EBML strings are usually not 0-terminated, so we allocate one
826 * byte more, read the string and NULL-terminate it ourselves. */
827 if (!(res = av_malloc(size + 1)))
828 return AVERROR(ENOMEM);
829 if (avio_read(pb, (uint8_t *) res, size) != size) {
841 * Read the next element as binary data.
842 * 0 is success, < 0 is failure.
844 static int ebml_read_binary(AVIOContext *pb, int length, EbmlBin *bin)
846 av_fast_padded_malloc(&bin->data, &bin->size, length);
848 return AVERROR(ENOMEM);
851 bin->pos = avio_tell(pb);
852 if (avio_read(pb, bin->data, length) != length) {
853 av_freep(&bin->data);
862 * Read the next element, but only the header. The contents
863 * are supposed to be sub-elements which can be read separately.
864 * 0 is success, < 0 is failure.
866 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
868 AVIOContext *pb = matroska->ctx->pb;
869 MatroskaLevel *level;
871 if (matroska->num_levels >= EBML_MAX_DEPTH) {
872 av_log(matroska->ctx, AV_LOG_ERROR,
873 "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
874 return AVERROR(ENOSYS);
877 level = &matroska->levels[matroska->num_levels++];
878 level->start = avio_tell(pb);
879 level->length = length;
885 * Read signed/unsigned "EBML" numbers.
886 * Return: number of bytes processed, < 0 on error
888 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
889 uint8_t *data, uint32_t size, uint64_t *num)
892 ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
893 return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
897 * Same as above, but signed.
899 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
900 uint8_t *data, uint32_t size, int64_t *num)
905 /* read as unsigned number first */
906 if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
909 /* make signed (weird way) */
910 *num = unum - ((1LL << (7 * res - 1)) - 1);
915 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
916 EbmlSyntax *syntax, void *data);
918 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
919 uint32_t id, void *data)
922 for (i = 0; syntax[i].id; i++)
923 if (id == syntax[i].id)
925 if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
926 matroska->num_levels > 0 &&
927 matroska->levels[matroska->num_levels - 1].length == 0xffffffffffffff)
928 return 0; // we reached the end of an unknown size cluster
929 if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
930 av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%"PRIX32"\n", id);
931 if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
932 return AVERROR_INVALIDDATA;
934 return ebml_parse_elem(matroska, &syntax[i], data);
937 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
940 if (!matroska->current_id) {
942 int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
945 matroska->current_id = id | 1 << 7 * res;
947 return ebml_parse_id(matroska, syntax, matroska->current_id, data);
950 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
955 for (i = 0; syntax[i].id; i++)
956 switch (syntax[i].type) {
958 *(uint64_t *) ((char *) data + syntax[i].data_offset) = syntax[i].def.u;
961 *(double *) ((char *) data + syntax[i].data_offset) = syntax[i].def.f;
965 // the default may be NULL
966 if (syntax[i].def.s) {
967 uint8_t **dst = (uint8_t **) ((uint8_t *) data + syntax[i].data_offset);
968 *dst = av_strdup(syntax[i].def.s);
970 return AVERROR(ENOMEM);
975 while (!res && !ebml_level_end(matroska))
976 res = ebml_parse(matroska, syntax, data);
981 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
982 EbmlSyntax *syntax, void *data)
984 static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
987 // max. 16 MB for strings
988 [EBML_STR] = 0x1000000,
989 [EBML_UTF8] = 0x1000000,
990 // max. 256 MB for binary data
991 [EBML_BIN] = 0x10000000,
992 // no limits for anything else
994 AVIOContext *pb = matroska->ctx->pb;
995 uint32_t id = syntax->id;
1000 data = (char *) data + syntax->data_offset;
1001 if (syntax->list_elem_size) {
1002 EbmlList *list = data;
1003 newelem = av_realloc_array(list->elem, list->nb_elem + 1, syntax->list_elem_size);
1005 return AVERROR(ENOMEM);
1006 list->elem = newelem;
1007 data = (char *) list->elem + list->nb_elem * syntax->list_elem_size;
1008 memset(data, 0, syntax->list_elem_size);
1012 if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
1013 matroska->current_id = 0;
1014 if ((res = ebml_read_length(matroska, pb, &length)) < 0)
1016 if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
1017 av_log(matroska->ctx, AV_LOG_ERROR,
1018 "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for syntax element %i\n",
1019 length, max_lengths[syntax->type], syntax->type);
1020 return AVERROR_INVALIDDATA;
1024 switch (syntax->type) {
1026 res = ebml_read_uint(pb, length, data);
1029 res = ebml_read_sint(pb, length, data);
1032 res = ebml_read_float(pb, length, data);
1036 res = ebml_read_ascii(pb, length, data);
1039 res = ebml_read_binary(pb, length, data);
1042 if ((res = ebml_read_master(matroska, length)) < 0)
1044 if (id == MATROSKA_ID_SEGMENT)
1045 matroska->segment_start = avio_tell(matroska->ctx->pb);
1046 return ebml_parse_nest(matroska, syntax->def.n, data);
1048 return ebml_parse_id(matroska, syntax->def.n, id, data);
1052 if (ffio_limit(pb, length) != length)
1053 return AVERROR(EIO);
1054 return avio_skip(pb, length) < 0 ? AVERROR(EIO) : 0;
1056 if (res == AVERROR_INVALIDDATA)
1057 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
1058 else if (res == AVERROR(EIO))
1059 av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
1063 static void ebml_free(EbmlSyntax *syntax, void *data)
1066 for (i = 0; syntax[i].id; i++) {
1067 void *data_off = (char *) data + syntax[i].data_offset;
1068 switch (syntax[i].type) {
1074 av_freep(&((EbmlBin *) data_off)->data);
1077 if (syntax[i].list_elem_size) {
1078 EbmlList *list = data_off;
1079 char *ptr = list->elem;
1080 for (j = 0; j < list->nb_elem;
1081 j++, ptr += syntax[i].list_elem_size)
1082 ebml_free(syntax[i].def.n, ptr);
1083 av_free(list->elem);
1085 ebml_free(syntax[i].def.n, data_off);
1095 static int matroska_probe(AVProbeData *p)
1098 int len_mask = 0x80, size = 1, n = 1, i;
1101 if (AV_RB32(p->buf) != EBML_ID_HEADER)
1104 /* length of header */
1106 while (size <= 8 && !(total & len_mask)) {
1112 total &= (len_mask - 1);
1114 total = (total << 8) | p->buf[4 + n++];
1116 /* Does the probe data contain the whole header? */
1117 if (p->buf_size < 4 + size + total)
1120 /* The header should contain a known document type. For now,
1121 * we don't parse the whole header but simply check for the
1122 * availability of that array of characters inside the header.
1123 * Not fully fool-proof, but good enough. */
1124 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1125 int probelen = strlen(matroska_doctypes[i]);
1126 if (total < probelen)
1128 for (n = 4 + size; n <= 4 + size + total - probelen; n++)
1129 if (!memcmp(p->buf + n, matroska_doctypes[i], probelen))
1130 return AVPROBE_SCORE_MAX;
1133 // probably valid EBML header but no recognized doctype
1134 return AVPROBE_SCORE_EXTENSION;
1137 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
1140 MatroskaTrack *tracks = matroska->tracks.elem;
1143 for (i = 0; i < matroska->tracks.nb_elem; i++)
1144 if (tracks[i].num == num)
1147 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
1151 static int matroska_decode_buffer(uint8_t **buf, int *buf_size,
1152 MatroskaTrack *track)
1154 MatroskaTrackEncoding *encodings = track->encodings.elem;
1155 uint8_t *data = *buf;
1156 int isize = *buf_size;
1157 uint8_t *pkt_data = NULL;
1158 uint8_t av_unused *newpktdata;
1159 int pkt_size = isize;
1163 if (pkt_size >= 10000000U)
1164 return AVERROR_INVALIDDATA;
1166 switch (encodings[0].compression.algo) {
1167 case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
1169 int header_size = encodings[0].compression.settings.size;
1170 uint8_t *header = encodings[0].compression.settings.data;
1172 if (header_size && !header) {
1173 av_log(NULL, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1180 pkt_size = isize + header_size;
1181 pkt_data = av_malloc(pkt_size);
1183 return AVERROR(ENOMEM);
1185 memcpy(pkt_data, header, header_size);
1186 memcpy(pkt_data + header_size, data, isize);
1190 case MATROSKA_TRACK_ENCODING_COMP_LZO:
1192 olen = pkt_size *= 3;
1193 newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING);
1195 result = AVERROR(ENOMEM);
1198 pkt_data = newpktdata;
1199 result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
1200 } while (result == AV_LZO_OUTPUT_FULL && pkt_size < 10000000);
1202 result = AVERROR_INVALIDDATA;
1209 case MATROSKA_TRACK_ENCODING_COMP_ZLIB:
1211 z_stream zstream = { 0 };
1212 if (inflateInit(&zstream) != Z_OK)
1214 zstream.next_in = data;
1215 zstream.avail_in = isize;
1218 newpktdata = av_realloc(pkt_data, pkt_size);
1220 inflateEnd(&zstream);
1223 pkt_data = newpktdata;
1224 zstream.avail_out = pkt_size - zstream.total_out;
1225 zstream.next_out = pkt_data + zstream.total_out;
1227 result = inflate(&zstream, Z_NO_FLUSH);
1229 result = Z_MEM_ERROR;
1230 } while (result == Z_OK && pkt_size < 10000000);
1231 pkt_size = zstream.total_out;
1232 inflateEnd(&zstream);
1233 if (result != Z_STREAM_END) {
1234 if (result == Z_MEM_ERROR)
1235 result = AVERROR(ENOMEM);
1237 result = AVERROR_INVALIDDATA;
1244 case MATROSKA_TRACK_ENCODING_COMP_BZLIB:
1246 bz_stream bzstream = { 0 };
1247 if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1249 bzstream.next_in = data;
1250 bzstream.avail_in = isize;
1253 newpktdata = av_realloc(pkt_data, pkt_size);
1255 BZ2_bzDecompressEnd(&bzstream);
1258 pkt_data = newpktdata;
1259 bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1260 bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1262 result = BZ2_bzDecompress(&bzstream);
1264 result = BZ_MEM_ERROR;
1265 } while (result == BZ_OK && pkt_size < 10000000);
1266 pkt_size = bzstream.total_out_lo32;
1267 BZ2_bzDecompressEnd(&bzstream);
1268 if (result != BZ_STREAM_END) {
1269 if (result == BZ_MEM_ERROR)
1270 result = AVERROR(ENOMEM);
1272 result = AVERROR_INVALIDDATA;
1279 return AVERROR_INVALIDDATA;
1283 *buf_size = pkt_size;
1291 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1292 AVDictionary **metadata, char *prefix)
1294 MatroskaTag *tags = list->elem;
1298 for (i = 0; i < list->nb_elem; i++) {
1299 const char *lang = tags[i].lang &&
1300 strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1302 if (!tags[i].name) {
1303 av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1307 snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1309 av_strlcpy(key, tags[i].name, sizeof(key));
1310 if (tags[i].def || !lang) {
1311 av_dict_set(metadata, key, tags[i].string, 0);
1312 if (tags[i].sub.nb_elem)
1313 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1316 av_strlcat(key, "-", sizeof(key));
1317 av_strlcat(key, lang, sizeof(key));
1318 av_dict_set(metadata, key, tags[i].string, 0);
1319 if (tags[i].sub.nb_elem)
1320 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1323 ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1326 static void matroska_convert_tags(AVFormatContext *s)
1328 MatroskaDemuxContext *matroska = s->priv_data;
1329 MatroskaTags *tags = matroska->tags.elem;
1332 for (i = 0; i < matroska->tags.nb_elem; i++) {
1333 if (tags[i].target.attachuid) {
1334 MatroskaAttachment *attachment = matroska->attachments.elem;
1335 for (j = 0; j < matroska->attachments.nb_elem; j++)
1336 if (attachment[j].uid == tags[i].target.attachuid &&
1337 attachment[j].stream)
1338 matroska_convert_tag(s, &tags[i].tag,
1339 &attachment[j].stream->metadata, NULL);
1340 } else if (tags[i].target.chapteruid) {
1341 MatroskaChapter *chapter = matroska->chapters.elem;
1342 for (j = 0; j < matroska->chapters.nb_elem; j++)
1343 if (chapter[j].uid == tags[i].target.chapteruid &&
1345 matroska_convert_tag(s, &tags[i].tag,
1346 &chapter[j].chapter->metadata, NULL);
1347 } else if (tags[i].target.trackuid) {
1348 MatroskaTrack *track = matroska->tracks.elem;
1349 for (j = 0; j < matroska->tracks.nb_elem; j++)
1350 if (track[j].uid == tags[i].target.trackuid && track[j].stream)
1351 matroska_convert_tag(s, &tags[i].tag,
1352 &track[j].stream->metadata, NULL);
1354 matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1355 tags[i].target.type);
1360 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska,
1363 EbmlList *seekhead_list = &matroska->seekhead;
1364 uint32_t level_up = matroska->level_up;
1365 uint32_t saved_id = matroska->current_id;
1366 MatroskaSeekhead *seekhead = seekhead_list->elem;
1367 int64_t before_pos = avio_tell(matroska->ctx->pb);
1368 MatroskaLevel level;
1372 if (idx >= seekhead_list->nb_elem ||
1373 seekhead[idx].id == MATROSKA_ID_SEEKHEAD ||
1374 seekhead[idx].id == MATROSKA_ID_CLUSTER)
1378 offset = seekhead[idx].pos + matroska->segment_start;
1379 if (avio_seek(matroska->ctx->pb, offset, SEEK_SET) == offset) {
1380 /* We don't want to lose our seekhead level, so we add
1381 * a dummy. This is a crude hack. */
1382 if (matroska->num_levels == EBML_MAX_DEPTH) {
1383 av_log(matroska->ctx, AV_LOG_INFO,
1384 "Max EBML element depth (%d) reached, "
1385 "cannot parse further.\n", EBML_MAX_DEPTH);
1386 ret = AVERROR_INVALIDDATA;
1389 level.length = (uint64_t) -1;
1390 matroska->levels[matroska->num_levels] = level;
1391 matroska->num_levels++;
1392 matroska->current_id = 0;
1394 ret = ebml_parse(matroska, matroska_segment, matroska);
1396 /* remove dummy level */
1397 while (matroska->num_levels) {
1398 uint64_t length = matroska->levels[--matroska->num_levels].length;
1399 if (length == (uint64_t) -1)
1405 avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
1406 matroska->level_up = level_up;
1407 matroska->current_id = saved_id;
1412 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1414 EbmlList *seekhead_list = &matroska->seekhead;
1415 int64_t before_pos = avio_tell(matroska->ctx->pb);
1419 // we should not do any seeking in the streaming case
1420 if (!matroska->ctx->pb->seekable ||
1421 (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1424 // do not read entries that are added while parsing seekhead entries
1425 nb_elem = seekhead_list->nb_elem;
1427 for (i = 0; i < nb_elem; i++) {
1428 MatroskaSeekhead *seekhead = seekhead_list->elem;
1429 if (seekhead[i].pos <= before_pos)
1432 // defer cues parsing until we actually need cue data.
1433 if (seekhead[i].id == MATROSKA_ID_CUES) {
1434 matroska->cues_parsing_deferred = 1;
1438 if (matroska_parse_seekhead_entry(matroska, i) < 0) {
1439 // mark index as broken
1440 matroska->cues_parsing_deferred = -1;
1444 if (nb_elem != seekhead_list->nb_elem) {
1445 avpriv_request_sample(matroska->ctx, "recursive SeekHead elements");
1449 static void matroska_add_index_entries(MatroskaDemuxContext *matroska)
1451 EbmlList *index_list;
1452 MatroskaIndex *index;
1453 int index_scale = 1;
1456 index_list = &matroska->index;
1457 index = index_list->elem;
1458 if (index_list->nb_elem &&
1459 index[0].time > 1E14 / matroska->time_scale) {
1460 av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1461 index_scale = matroska->time_scale;
1463 for (i = 0; i < index_list->nb_elem; i++) {
1464 EbmlList *pos_list = &index[i].pos;
1465 MatroskaIndexPos *pos = pos_list->elem;
1466 for (j = 0; j < pos_list->nb_elem; j++) {
1467 MatroskaTrack *track = matroska_find_track_by_num(matroska,
1469 if (track && track->stream)
1470 av_add_index_entry(track->stream,
1471 pos[j].pos + matroska->segment_start,
1472 index[i].time / index_scale, 0, 0,
1478 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1479 EbmlList *seekhead_list = &matroska->seekhead;
1480 MatroskaSeekhead *seekhead = seekhead_list->elem;
1483 for (i = 0; i < seekhead_list->nb_elem; i++)
1484 if (seekhead[i].id == MATROSKA_ID_CUES)
1486 av_assert1(i <= seekhead_list->nb_elem);
1488 if (matroska_parse_seekhead_entry(matroska, i) < 0)
1489 matroska->cues_parsing_deferred = -1;
1490 matroska_add_index_entries(matroska);
1493 static int matroska_aac_profile(char *codec_id)
1495 static const char *const aac_profiles[] = { "MAIN", "LC", "SSR" };
1498 for (profile = 0; profile < FF_ARRAY_ELEMS(aac_profiles); profile++)
1499 if (strstr(codec_id, aac_profiles[profile]))
1504 static int matroska_aac_sri(int samplerate)
1508 for (sri = 0; sri < FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1509 if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
1514 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
1517 /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
1518 time_t creation_time = date_utc / 1000000000 + 978307200;
1519 struct tm tmpbuf, *ptm = gmtime_r(&creation_time, &tmpbuf);
1521 if (strftime(buffer, sizeof(buffer), "%Y-%m-%d %H:%M:%S", ptm))
1522 av_dict_set(metadata, "creation_time", buffer, 0);
1525 static int matroska_parse_flac(AVFormatContext *s,
1526 MatroskaTrack *track,
1529 AVStream *st = track->stream;
1530 uint8_t *p = track->codec_priv.data;
1531 int size = track->codec_priv.size;
1533 if (size < 8 + FLAC_STREAMINFO_SIZE || p[4] & 0x7f) {
1534 av_log(s, AV_LOG_WARNING, "Invalid FLAC private data\n");
1535 track->codec_priv.size = 0;
1539 track->codec_priv.size = 8 + FLAC_STREAMINFO_SIZE;
1541 p += track->codec_priv.size;
1542 size -= track->codec_priv.size;
1544 /* parse the remaining metadata blocks if present */
1546 int block_last, block_type, block_size;
1548 flac_parse_block_header(p, &block_last, &block_type, &block_size);
1552 if (block_size > size)
1555 /* check for the channel mask */
1556 if (block_type == FLAC_METADATA_TYPE_VORBIS_COMMENT) {
1557 AVDictionary *dict = NULL;
1558 AVDictionaryEntry *chmask;
1560 ff_vorbis_comment(s, &dict, p, block_size, 0);
1561 chmask = av_dict_get(dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", NULL, 0);
1563 uint64_t mask = strtol(chmask->value, NULL, 0);
1564 if (!mask || mask & ~0x3ffffULL) {
1565 av_log(s, AV_LOG_WARNING,
1566 "Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK\n");
1568 st->codec->channel_layout = mask;
1570 av_dict_free(&dict);
1580 static int matroska_parse_tracks(AVFormatContext *s)
1582 MatroskaDemuxContext *matroska = s->priv_data;
1583 MatroskaTrack *tracks = matroska->tracks.elem;
1588 for (i = 0; i < matroska->tracks.nb_elem; i++) {
1589 MatroskaTrack *track = &tracks[i];
1590 enum AVCodecID codec_id = AV_CODEC_ID_NONE;
1591 EbmlList *encodings_list = &track->encodings;
1592 MatroskaTrackEncoding *encodings = encodings_list->elem;
1593 uint8_t *extradata = NULL;
1594 int extradata_size = 0;
1595 int extradata_offset = 0;
1596 uint32_t fourcc = 0;
1598 char* key_id_base64 = NULL;
1601 /* Apply some sanity checks. */
1602 if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1603 track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1604 track->type != MATROSKA_TRACK_TYPE_SUBTITLE &&
1605 track->type != MATROSKA_TRACK_TYPE_METADATA) {
1606 av_log(matroska->ctx, AV_LOG_INFO,
1607 "Unknown or unsupported track type %"PRIu64"\n",
1611 if (!track->codec_id)
1614 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1615 if (!track->default_duration && track->video.frame_rate > 0)
1616 track->default_duration = 1000000000 / track->video.frame_rate;
1617 if (track->video.display_width == -1)
1618 track->video.display_width = track->video.pixel_width;
1619 if (track->video.display_height == -1)
1620 track->video.display_height = track->video.pixel_height;
1621 if (track->video.color_space.size == 4)
1622 fourcc = AV_RL32(track->video.color_space.data);
1623 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1624 if (!track->audio.out_samplerate)
1625 track->audio.out_samplerate = track->audio.samplerate;
1627 if (encodings_list->nb_elem > 1) {
1628 av_log(matroska->ctx, AV_LOG_ERROR,
1629 "Multiple combined encodings not supported");
1630 } else if (encodings_list->nb_elem == 1) {
1631 if (encodings[0].type) {
1632 if (encodings[0].encryption.key_id.size > 0) {
1633 /* Save the encryption key id to be stored later as a
1635 const int b64_size = AV_BASE64_SIZE(encodings[0].encryption.key_id.size);
1636 key_id_base64 = av_malloc(b64_size);
1637 if (key_id_base64 == NULL)
1638 return AVERROR(ENOMEM);
1640 av_base64_encode(key_id_base64, b64_size,
1641 encodings[0].encryption.key_id.data,
1642 encodings[0].encryption.key_id.size);
1644 encodings[0].scope = 0;
1645 av_log(matroska->ctx, AV_LOG_ERROR,
1646 "Unsupported encoding type");
1650 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1653 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1656 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO &&
1658 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP) {
1659 encodings[0].scope = 0;
1660 av_log(matroska->ctx, AV_LOG_ERROR,
1661 "Unsupported encoding type");
1662 } else if (track->codec_priv.size && encodings[0].scope & 2) {
1663 uint8_t *codec_priv = track->codec_priv.data;
1664 int ret = matroska_decode_buffer(&track->codec_priv.data,
1665 &track->codec_priv.size,
1668 track->codec_priv.data = NULL;
1669 track->codec_priv.size = 0;
1670 av_log(matroska->ctx, AV_LOG_ERROR,
1671 "Failed to decode codec private data\n");
1674 if (codec_priv != track->codec_priv.data)
1675 av_free(codec_priv);
1679 for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
1680 if (!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1681 strlen(ff_mkv_codec_tags[j].str))) {
1682 codec_id = ff_mkv_codec_tags[j].id;
1687 st = track->stream = avformat_new_stream(s, NULL);
1689 av_free(key_id_base64);
1690 return AVERROR(ENOMEM);
1693 if (key_id_base64) {
1694 /* export encryption key id as base64 metadata tag */
1695 av_dict_set(&st->metadata, "enc_key_id", key_id_base64, 0);
1696 av_freep(&key_id_base64);
1699 if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC") &&
1700 track->codec_priv.size >= 40 &&
1701 track->codec_priv.data) {
1702 track->ms_compat = 1;
1703 bit_depth = AV_RL16(track->codec_priv.data + 14);
1704 fourcc = AV_RL32(track->codec_priv.data + 16);
1705 codec_id = ff_codec_get_id(ff_codec_bmp_tags,
1708 codec_id = ff_codec_get_id(ff_codec_movvideo_tags,
1710 extradata_offset = 40;
1711 } else if (!strcmp(track->codec_id, "A_MS/ACM") &&
1712 track->codec_priv.size >= 14 &&
1713 track->codec_priv.data) {
1715 ffio_init_context(&b, track->codec_priv.data,
1716 track->codec_priv.size,
1717 0, NULL, NULL, NULL, NULL);
1718 ret = ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1721 codec_id = st->codec->codec_id;
1722 extradata_offset = FFMIN(track->codec_priv.size, 18);
1723 } else if (!strcmp(track->codec_id, "A_QUICKTIME")
1724 && (track->codec_priv.size >= 86)
1725 && (track->codec_priv.data)) {
1726 fourcc = AV_RL32(track->codec_priv.data + 4);
1727 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
1728 if (ff_codec_get_id(ff_codec_movaudio_tags, AV_RL32(track->codec_priv.data))) {
1729 fourcc = AV_RL32(track->codec_priv.data);
1730 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
1732 } else if (!strcmp(track->codec_id, "V_QUICKTIME") &&
1733 (track->codec_priv.size >= 21) &&
1734 (track->codec_priv.data)) {
1735 fourcc = AV_RL32(track->codec_priv.data + 4);
1736 codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
1737 if (ff_codec_get_id(ff_codec_movvideo_tags, AV_RL32(track->codec_priv.data))) {
1738 fourcc = AV_RL32(track->codec_priv.data);
1739 codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
1741 if (codec_id == AV_CODEC_ID_NONE && AV_RL32(track->codec_priv.data+4) == AV_RL32("SMI "))
1742 codec_id = AV_CODEC_ID_SVQ3;
1743 } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
1744 switch (track->audio.bitdepth) {
1746 codec_id = AV_CODEC_ID_PCM_U8;
1749 codec_id = AV_CODEC_ID_PCM_S24BE;
1752 codec_id = AV_CODEC_ID_PCM_S32BE;
1755 } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
1756 switch (track->audio.bitdepth) {
1758 codec_id = AV_CODEC_ID_PCM_U8;
1761 codec_id = AV_CODEC_ID_PCM_S24LE;
1764 codec_id = AV_CODEC_ID_PCM_S32LE;
1767 } else if (codec_id == AV_CODEC_ID_PCM_F32LE &&
1768 track->audio.bitdepth == 64) {
1769 codec_id = AV_CODEC_ID_PCM_F64LE;
1770 } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
1771 int profile = matroska_aac_profile(track->codec_id);
1772 int sri = matroska_aac_sri(track->audio.samplerate);
1773 extradata = av_mallocz(5 + FF_INPUT_BUFFER_PADDING_SIZE);
1775 return AVERROR(ENOMEM);
1776 extradata[0] = (profile << 3) | ((sri & 0x0E) >> 1);
1777 extradata[1] = ((sri & 0x01) << 7) | (track->audio.channels << 3);
1778 if (strstr(track->codec_id, "SBR")) {
1779 sri = matroska_aac_sri(track->audio.out_samplerate);
1780 extradata[2] = 0x56;
1781 extradata[3] = 0xE5;
1782 extradata[4] = 0x80 | (sri << 3);
1786 } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX - 12 - FF_INPUT_BUFFER_PADDING_SIZE) {
1787 /* Only ALAC's magic cookie is stored in Matroska's track headers.
1788 * Create the "atom size", "tag", and "tag version" fields the
1789 * decoder expects manually. */
1790 extradata_size = 12 + track->codec_priv.size;
1791 extradata = av_mallocz(extradata_size +
1792 FF_INPUT_BUFFER_PADDING_SIZE);
1794 return AVERROR(ENOMEM);
1795 AV_WB32(extradata, extradata_size);
1796 memcpy(&extradata[4], "alac", 4);
1797 AV_WB32(&extradata[8], 0);
1798 memcpy(&extradata[12], track->codec_priv.data,
1799 track->codec_priv.size);
1800 } else if (codec_id == AV_CODEC_ID_TTA) {
1801 extradata_size = 30;
1802 extradata = av_mallocz(extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
1804 return AVERROR(ENOMEM);
1805 ffio_init_context(&b, extradata, extradata_size, 1,
1806 NULL, NULL, NULL, NULL);
1807 avio_write(&b, "TTA1", 4);
1809 avio_wl16(&b, track->audio.channels);
1810 avio_wl16(&b, track->audio.bitdepth);
1811 if (track->audio.out_samplerate < 0 || track->audio.out_samplerate > INT_MAX)
1812 return AVERROR_INVALIDDATA;
1813 avio_wl32(&b, track->audio.out_samplerate);
1814 avio_wl32(&b, av_rescale((matroska->duration * matroska->time_scale),
1815 track->audio.out_samplerate,
1816 AV_TIME_BASE * 1000));
1817 } else if (codec_id == AV_CODEC_ID_RV10 ||
1818 codec_id == AV_CODEC_ID_RV20 ||
1819 codec_id == AV_CODEC_ID_RV30 ||
1820 codec_id == AV_CODEC_ID_RV40) {
1821 extradata_offset = 26;
1822 } else if (codec_id == AV_CODEC_ID_RA_144) {
1823 track->audio.out_samplerate = 8000;
1824 track->audio.channels = 1;
1825 } else if ((codec_id == AV_CODEC_ID_RA_288 ||
1826 codec_id == AV_CODEC_ID_COOK ||
1827 codec_id == AV_CODEC_ID_ATRAC3 ||
1828 codec_id == AV_CODEC_ID_SIPR)
1829 && track->codec_priv.data) {
1832 ffio_init_context(&b, track->codec_priv.data,
1833 track->codec_priv.size,
1834 0, NULL, NULL, NULL, NULL);
1836 flavor = avio_rb16(&b);
1837 track->audio.coded_framesize = avio_rb32(&b);
1839 track->audio.sub_packet_h = avio_rb16(&b);
1840 track->audio.frame_size = avio_rb16(&b);
1841 track->audio.sub_packet_size = avio_rb16(&b);
1843 track->audio.coded_framesize <= 0 ||
1844 track->audio.sub_packet_h <= 0 ||
1845 track->audio.frame_size <= 0 ||
1846 track->audio.sub_packet_size <= 0)
1847 return AVERROR_INVALIDDATA;
1848 track->audio.buf = av_malloc_array(track->audio.sub_packet_h,
1849 track->audio.frame_size);
1850 if (!track->audio.buf)
1851 return AVERROR(ENOMEM);
1852 if (codec_id == AV_CODEC_ID_RA_288) {
1853 st->codec->block_align = track->audio.coded_framesize;
1854 track->codec_priv.size = 0;
1856 if (codec_id == AV_CODEC_ID_SIPR && flavor < 4) {
1857 static const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1858 track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1859 st->codec->bit_rate = sipr_bit_rate[flavor];
1861 st->codec->block_align = track->audio.sub_packet_size;
1862 extradata_offset = 78;
1864 } else if (codec_id == AV_CODEC_ID_FLAC && track->codec_priv.size) {
1865 ret = matroska_parse_flac(s, track, &extradata_offset);
1868 } else if (codec_id == AV_CODEC_ID_PRORES && track->codec_priv.size == 4) {
1869 fourcc = AV_RL32(track->codec_priv.data);
1871 track->codec_priv.size -= extradata_offset;
1873 if (codec_id == AV_CODEC_ID_NONE)
1874 av_log(matroska->ctx, AV_LOG_INFO,
1875 "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
1877 if (track->time_scale < 0.01)
1878 track->time_scale = 1.0;
1879 avpriv_set_pts_info(st, 64, matroska->time_scale * track->time_scale,
1880 1000 * 1000 * 1000); /* 64 bit pts in ns */
1882 /* convert the delay from ns to the track timebase */
1883 track->codec_delay = av_rescale_q(track->codec_delay,
1884 (AVRational){ 1, 1000000000 },
1887 st->codec->codec_id = codec_id;
1889 if (strcmp(track->language, "und"))
1890 av_dict_set(&st->metadata, "language", track->language, 0);
1891 av_dict_set(&st->metadata, "title", track->name, 0);
1893 if (track->flag_default)
1894 st->disposition |= AV_DISPOSITION_DEFAULT;
1895 if (track->flag_forced)
1896 st->disposition |= AV_DISPOSITION_FORCED;
1898 if (!st->codec->extradata) {
1900 st->codec->extradata = extradata;
1901 st->codec->extradata_size = extradata_size;
1902 } else if (track->codec_priv.data && track->codec_priv.size > 0) {
1903 if (ff_alloc_extradata(st->codec, track->codec_priv.size))
1904 return AVERROR(ENOMEM);
1905 memcpy(st->codec->extradata,
1906 track->codec_priv.data + extradata_offset,
1907 track->codec_priv.size);
1911 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1912 MatroskaTrackPlane *planes = track->operation.combine_planes.elem;
1914 st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1915 st->codec->codec_tag = fourcc;
1917 st->codec->bits_per_coded_sample = bit_depth;
1918 st->codec->width = track->video.pixel_width;
1919 st->codec->height = track->video.pixel_height;
1920 av_reduce(&st->sample_aspect_ratio.num,
1921 &st->sample_aspect_ratio.den,
1922 st->codec->height * track->video.display_width,
1923 st->codec->width * track->video.display_height,
1925 if (st->codec->codec_id != AV_CODEC_ID_HEVC)
1926 st->need_parsing = AVSTREAM_PARSE_HEADERS;
1928 if (track->default_duration) {
1929 av_reduce(&st->avg_frame_rate.num, &st->avg_frame_rate.den,
1930 1000000000, track->default_duration, 30000);
1931 #if FF_API_R_FRAME_RATE
1932 if ( st->avg_frame_rate.num < st->avg_frame_rate.den * 1000L
1933 && st->avg_frame_rate.num > st->avg_frame_rate.den * 5L)
1934 st->r_frame_rate = st->avg_frame_rate;
1938 /* export stereo mode flag as metadata tag */
1939 if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
1940 av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
1942 /* export alpha mode flag as metadata tag */
1943 if (track->video.alpha_mode)
1944 av_dict_set(&st->metadata, "alpha_mode", "1", 0);
1946 /* if we have virtual track, mark the real tracks */
1947 for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
1949 if (planes[j].type >= MATROSKA_VIDEO_STEREO_PLANE_COUNT)
1951 snprintf(buf, sizeof(buf), "%s_%d",
1952 ff_matroska_video_stereo_plane[planes[j].type], i);
1953 for (k=0; k < matroska->tracks.nb_elem; k++)
1954 if (planes[j].uid == tracks[k].uid) {
1955 av_dict_set(&s->streams[k]->metadata,
1956 "stereo_mode", buf, 0);
1960 // add stream level stereo3d side data if it is a supported format
1961 if (track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB &&
1962 track->video.stereo_mode != 10 && track->video.stereo_mode != 12) {
1963 int ret = ff_mkv_stereo3d_conv(st, track->video.stereo_mode);
1967 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1968 st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1969 st->codec->sample_rate = track->audio.out_samplerate;
1970 st->codec->channels = track->audio.channels;
1971 if (!st->codec->bits_per_coded_sample)
1972 st->codec->bits_per_coded_sample = track->audio.bitdepth;
1973 if (st->codec->codec_id != AV_CODEC_ID_AAC)
1974 st->need_parsing = AVSTREAM_PARSE_HEADERS;
1975 if (track->codec_delay > 0) {
1976 st->codec->delay = av_rescale_q(track->codec_delay,
1978 (AVRational){1, st->codec->sample_rate});
1980 if (track->seek_preroll > 0) {
1981 av_codec_set_seek_preroll(st->codec,
1982 av_rescale_q(track->seek_preroll,
1983 (AVRational){1, 1000000000},
1984 (AVRational){1, st->codec->sample_rate}));
1986 } else if (codec_id == AV_CODEC_ID_WEBVTT) {
1987 st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1989 if (!strcmp(track->codec_id, "D_WEBVTT/CAPTIONS")) {
1990 st->disposition |= AV_DISPOSITION_CAPTIONS;
1991 } else if (!strcmp(track->codec_id, "D_WEBVTT/DESCRIPTIONS")) {
1992 st->disposition |= AV_DISPOSITION_DESCRIPTIONS;
1993 } else if (!strcmp(track->codec_id, "D_WEBVTT/METADATA")) {
1994 st->disposition |= AV_DISPOSITION_METADATA;
1996 } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1997 st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1998 if (st->codec->codec_id == AV_CODEC_ID_ASS)
1999 matroska->contains_ssa = 1;
2006 static int matroska_read_header(AVFormatContext *s)
2008 MatroskaDemuxContext *matroska = s->priv_data;
2009 EbmlList *attachments_list = &matroska->attachments;
2010 EbmlList *chapters_list = &matroska->chapters;
2011 MatroskaAttachment *attachments;
2012 MatroskaChapter *chapters;
2013 uint64_t max_start = 0;
2020 /* First read the EBML header. */
2021 if (ebml_parse(matroska, ebml_syntax, &ebml) ||
2022 ebml.version > EBML_VERSION ||
2023 ebml.max_size > sizeof(uint64_t) ||
2024 ebml.id_length > sizeof(uint32_t) ||
2025 ebml.doctype_version > 3 ||
2027 av_log(matroska->ctx, AV_LOG_ERROR,
2028 "EBML header using unsupported features\n"
2029 "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2030 ebml.version, ebml.doctype, ebml.doctype_version);
2031 ebml_free(ebml_syntax, &ebml);
2032 return AVERROR_PATCHWELCOME;
2033 } else if (ebml.doctype_version == 3) {
2034 av_log(matroska->ctx, AV_LOG_WARNING,
2035 "EBML header using unsupported features\n"
2036 "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2037 ebml.version, ebml.doctype, ebml.doctype_version);
2039 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
2040 if (!strcmp(ebml.doctype, matroska_doctypes[i]))
2042 if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
2043 av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
2044 if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
2045 ebml_free(ebml_syntax, &ebml);
2046 return AVERROR_INVALIDDATA;
2049 ebml_free(ebml_syntax, &ebml);
2051 /* The next thing is a segment. */
2052 pos = avio_tell(matroska->ctx->pb);
2053 res = ebml_parse(matroska, matroska_segments, matroska);
2054 // try resyncing until we find a EBML_STOP type element.
2056 res = matroska_resync(matroska, pos);
2059 pos = avio_tell(matroska->ctx->pb);
2060 res = ebml_parse(matroska, matroska_segment, matroska);
2062 matroska_execute_seekhead(matroska);
2064 if (!matroska->time_scale)
2065 matroska->time_scale = 1000000;
2066 if (matroska->duration)
2067 matroska->ctx->duration = matroska->duration * matroska->time_scale *
2068 1000 / AV_TIME_BASE;
2069 av_dict_set(&s->metadata, "title", matroska->title, 0);
2070 av_dict_set(&s->metadata, "encoder", matroska->muxingapp, 0);
2072 if (matroska->date_utc.size == 8)
2073 matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
2075 res = matroska_parse_tracks(s);
2079 attachments = attachments_list->elem;
2080 for (j = 0; j < attachments_list->nb_elem; j++) {
2081 if (!(attachments[j].filename && attachments[j].mime &&
2082 attachments[j].bin.data && attachments[j].bin.size > 0)) {
2083 av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
2085 AVStream *st = avformat_new_stream(s, NULL);
2088 av_dict_set(&st->metadata, "filename", attachments[j].filename, 0);
2089 av_dict_set(&st->metadata, "mimetype", attachments[j].mime, 0);
2090 st->codec->codec_id = AV_CODEC_ID_NONE;
2091 st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
2092 if (ff_alloc_extradata(st->codec, attachments[j].bin.size))
2094 memcpy(st->codec->extradata, attachments[j].bin.data,
2095 attachments[j].bin.size);
2097 for (i = 0; ff_mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
2098 if (!strncmp(ff_mkv_mime_tags[i].str, attachments[j].mime,
2099 strlen(ff_mkv_mime_tags[i].str))) {
2100 st->codec->codec_id = ff_mkv_mime_tags[i].id;
2104 attachments[j].stream = st;
2108 chapters = chapters_list->elem;
2109 for (i = 0; i < chapters_list->nb_elem; i++)
2110 if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid &&
2111 (max_start == 0 || chapters[i].start > max_start)) {
2112 chapters[i].chapter =
2113 avpriv_new_chapter(s, chapters[i].uid,
2114 (AVRational) { 1, 1000000000 },
2115 chapters[i].start, chapters[i].end,
2117 if (chapters[i].chapter) {
2118 av_dict_set(&chapters[i].chapter->metadata,
2119 "title", chapters[i].title, 0);
2121 max_start = chapters[i].start;
2124 matroska_add_index_entries(matroska);
2126 matroska_convert_tags(s);
2132 * Put one packet in an application-supplied AVPacket struct.
2133 * Returns 0 on success or -1 on failure.
2135 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
2138 if (matroska->num_packets > 0) {
2139 memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
2140 av_free(matroska->packets[0]);
2141 if (matroska->num_packets > 1) {
2143 memmove(&matroska->packets[0], &matroska->packets[1],
2144 (matroska->num_packets - 1) * sizeof(AVPacket *));
2145 newpackets = av_realloc(matroska->packets,
2146 (matroska->num_packets - 1) *
2147 sizeof(AVPacket *));
2149 matroska->packets = newpackets;
2151 av_freep(&matroska->packets);
2152 matroska->prev_pkt = NULL;
2154 matroska->num_packets--;
2162 * Free all packets in our internal queue.
2164 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
2166 matroska->prev_pkt = NULL;
2167 if (matroska->packets) {
2169 for (n = 0; n < matroska->num_packets; n++) {
2170 av_free_packet(matroska->packets[n]);
2171 av_free(matroska->packets[n]);
2173 av_freep(&matroska->packets);
2174 matroska->num_packets = 0;
2178 static int matroska_parse_laces(MatroskaDemuxContext *matroska, uint8_t **buf,
2179 int *buf_size, int type,
2180 uint32_t **lace_buf, int *laces)
2182 int res = 0, n, size = *buf_size;
2183 uint8_t *data = *buf;
2184 uint32_t *lace_size;
2188 *lace_buf = av_mallocz(sizeof(int));
2190 return AVERROR(ENOMEM);
2192 *lace_buf[0] = size;
2196 av_assert0(size > 0);
2200 lace_size = av_mallocz(*laces * sizeof(int));
2202 return AVERROR(ENOMEM);
2205 case 0x1: /* Xiph lacing */
2209 for (n = 0; res == 0 && n < *laces - 1; n++) {
2211 if (size <= total) {
2212 res = AVERROR_INVALIDDATA;
2217 lace_size[n] += temp;
2224 if (size <= total) {
2225 res = AVERROR_INVALIDDATA;
2229 lace_size[n] = size - total;
2233 case 0x2: /* fixed-size lacing */
2234 if (size % (*laces)) {
2235 res = AVERROR_INVALIDDATA;
2238 for (n = 0; n < *laces; n++)
2239 lace_size[n] = size / *laces;
2242 case 0x3: /* EBML lacing */
2246 n = matroska_ebmlnum_uint(matroska, data, size, &num);
2247 if (n < 0 || num > INT_MAX) {
2248 av_log(matroska->ctx, AV_LOG_INFO,
2249 "EBML block data error\n");
2250 res = n<0 ? n : AVERROR_INVALIDDATA;
2255 total = lace_size[0] = num;
2256 for (n = 1; res == 0 && n < *laces - 1; n++) {
2259 r = matroska_ebmlnum_sint(matroska, data, size, &snum);
2260 if (r < 0 || lace_size[n - 1] + snum > (uint64_t)INT_MAX) {
2261 av_log(matroska->ctx, AV_LOG_INFO,
2262 "EBML block data error\n");
2263 res = r<0 ? r : AVERROR_INVALIDDATA;
2268 lace_size[n] = lace_size[n - 1] + snum;
2269 total += lace_size[n];
2271 if (size <= total) {
2272 res = AVERROR_INVALIDDATA;
2275 lace_size[*laces - 1] = size - total;
2281 *lace_buf = lace_size;
2287 static int matroska_parse_rm_audio(MatroskaDemuxContext *matroska,
2288 MatroskaTrack *track, AVStream *st,
2289 uint8_t *data, int size, uint64_t timecode,
2292 int a = st->codec->block_align;
2293 int sps = track->audio.sub_packet_size;
2294 int cfs = track->audio.coded_framesize;
2295 int h = track->audio.sub_packet_h;
2296 int y = track->audio.sub_packet_cnt;
2297 int w = track->audio.frame_size;
2300 if (!track->audio.pkt_cnt) {
2301 if (track->audio.sub_packet_cnt == 0)
2302 track->audio.buf_timecode = timecode;
2303 if (st->codec->codec_id == AV_CODEC_ID_RA_288) {
2304 if (size < cfs * h / 2) {
2305 av_log(matroska->ctx, AV_LOG_ERROR,
2306 "Corrupt int4 RM-style audio packet size\n");
2307 return AVERROR_INVALIDDATA;
2309 for (x = 0; x < h / 2; x++)
2310 memcpy(track->audio.buf + x * 2 * w + y * cfs,
2311 data + x * cfs, cfs);
2312 } else if (st->codec->codec_id == AV_CODEC_ID_SIPR) {
2314 av_log(matroska->ctx, AV_LOG_ERROR,
2315 "Corrupt sipr RM-style audio packet size\n");
2316 return AVERROR_INVALIDDATA;
2318 memcpy(track->audio.buf + y * w, data, w);
2320 if (size < sps * w / sps || h<=0 || w%sps) {
2321 av_log(matroska->ctx, AV_LOG_ERROR,
2322 "Corrupt generic RM-style audio packet size\n");
2323 return AVERROR_INVALIDDATA;
2325 for (x = 0; x < w / sps; x++)
2326 memcpy(track->audio.buf +
2327 sps * (h * x + ((h + 1) / 2) * (y & 1) + (y >> 1)),
2328 data + x * sps, sps);
2331 if (++track->audio.sub_packet_cnt >= h) {
2332 if (st->codec->codec_id == AV_CODEC_ID_SIPR)
2333 ff_rm_reorder_sipr_data(track->audio.buf, h, w);
2334 track->audio.sub_packet_cnt = 0;
2335 track->audio.pkt_cnt = h * w / a;
2339 while (track->audio.pkt_cnt) {
2341 AVPacket *pkt = av_mallocz(sizeof(AVPacket));
2343 return AVERROR(ENOMEM);
2345 ret = av_new_packet(pkt, a);
2351 track->audio.buf + a * (h * w / a - track->audio.pkt_cnt--),
2353 pkt->pts = track->audio.buf_timecode;
2354 track->audio.buf_timecode = AV_NOPTS_VALUE;
2356 pkt->stream_index = st->index;
2357 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2363 /* reconstruct full wavpack blocks from mangled matroska ones */
2364 static int matroska_parse_wavpack(MatroskaTrack *track, uint8_t *src,
2365 uint8_t **pdst, int *size)
2367 uint8_t *dst = NULL;
2372 int ret, offset = 0;
2374 if (srclen < 12 || track->stream->codec->extradata_size < 2)
2375 return AVERROR_INVALIDDATA;
2377 ver = AV_RL16(track->stream->codec->extradata);
2379 samples = AV_RL32(src);
2383 while (srclen >= 8) {
2388 uint32_t flags = AV_RL32(src);
2389 uint32_t crc = AV_RL32(src + 4);
2393 multiblock = (flags & 0x1800) != 0x1800;
2396 ret = AVERROR_INVALIDDATA;
2399 blocksize = AV_RL32(src);
2405 if (blocksize > srclen) {
2406 ret = AVERROR_INVALIDDATA;
2410 tmp = av_realloc(dst, dstlen + blocksize + 32);
2412 ret = AVERROR(ENOMEM);
2416 dstlen += blocksize + 32;
2418 AV_WL32(dst + offset, MKTAG('w', 'v', 'p', 'k')); // tag
2419 AV_WL32(dst + offset + 4, blocksize + 24); // blocksize - 8
2420 AV_WL16(dst + offset + 8, ver); // version
2421 AV_WL16(dst + offset + 10, 0); // track/index_no
2422 AV_WL32(dst + offset + 12, 0); // total samples
2423 AV_WL32(dst + offset + 16, 0); // block index
2424 AV_WL32(dst + offset + 20, samples); // number of samples
2425 AV_WL32(dst + offset + 24, flags); // flags
2426 AV_WL32(dst + offset + 28, crc); // crc
2427 memcpy(dst + offset + 32, src, blocksize); // block data
2430 srclen -= blocksize;
2431 offset += blocksize + 32;
2444 static int matroska_parse_webvtt(MatroskaDemuxContext *matroska,
2445 MatroskaTrack *track,
2447 uint8_t *data, int data_len,
2453 uint8_t *id, *settings, *text, *buf;
2454 int id_len, settings_len, text_len;
2459 return AVERROR_INVALIDDATA;
2462 q = data + data_len;
2467 if (*p == '\r' || *p == '\n') {
2476 if (p >= q || *p != '\n')
2477 return AVERROR_INVALIDDATA;
2483 if (*p == '\r' || *p == '\n') {
2484 settings_len = p - settings;
2492 if (p >= q || *p != '\n')
2493 return AVERROR_INVALIDDATA;
2498 while (text_len > 0) {
2499 const int len = text_len - 1;
2500 const uint8_t c = p[len];
2501 if (c != '\r' && c != '\n')
2507 return AVERROR_INVALIDDATA;
2509 pkt = av_mallocz(sizeof(*pkt));
2510 err = av_new_packet(pkt, text_len);
2513 return AVERROR(err);
2516 memcpy(pkt->data, text, text_len);
2519 buf = av_packet_new_side_data(pkt,
2520 AV_PKT_DATA_WEBVTT_IDENTIFIER,
2524 return AVERROR(ENOMEM);
2526 memcpy(buf, id, id_len);
2529 if (settings_len > 0) {
2530 buf = av_packet_new_side_data(pkt,
2531 AV_PKT_DATA_WEBVTT_SETTINGS,
2535 return AVERROR(ENOMEM);
2537 memcpy(buf, settings, settings_len);
2540 // Do we need this for subtitles?
2541 // pkt->flags = AV_PKT_FLAG_KEY;
2543 pkt->stream_index = st->index;
2544 pkt->pts = timecode;
2546 // Do we need this for subtitles?
2547 // pkt->dts = timecode;
2549 pkt->duration = duration;
2552 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2553 matroska->prev_pkt = pkt;
2558 static int matroska_parse_frame(MatroskaDemuxContext *matroska,
2559 MatroskaTrack *track, AVStream *st,
2560 uint8_t *data, int pkt_size,
2561 uint64_t timecode, uint64_t lace_duration,
2562 int64_t pos, int is_keyframe,
2563 uint8_t *additional, uint64_t additional_id, int additional_size,
2564 int64_t discard_padding)
2566 MatroskaTrackEncoding *encodings = track->encodings.elem;
2567 uint8_t *pkt_data = data;
2568 int offset = 0, res;
2571 if (encodings && !encodings->type && encodings->scope & 1) {
2572 res = matroska_decode_buffer(&pkt_data, &pkt_size, track);
2577 if (st->codec->codec_id == AV_CODEC_ID_WAVPACK) {
2579 res = matroska_parse_wavpack(track, pkt_data, &wv_data, &pkt_size);
2581 av_log(matroska->ctx, AV_LOG_ERROR,
2582 "Error parsing a wavpack block.\n");
2585 if (pkt_data != data)
2586 av_freep(&pkt_data);
2590 if (st->codec->codec_id == AV_CODEC_ID_PRORES &&
2591 AV_RB32(&data[4]) != MKBETAG('i', 'c', 'p', 'f'))
2594 pkt = av_mallocz(sizeof(AVPacket));
2595 /* XXX: prevent data copy... */
2596 if (av_new_packet(pkt, pkt_size + offset) < 0) {
2598 res = AVERROR(ENOMEM);
2602 if (st->codec->codec_id == AV_CODEC_ID_PRORES && offset == 8) {
2603 uint8_t *buf = pkt->data;
2604 bytestream_put_be32(&buf, pkt_size);
2605 bytestream_put_be32(&buf, MKBETAG('i', 'c', 'p', 'f'));
2608 memcpy(pkt->data + offset, pkt_data, pkt_size);
2610 if (pkt_data != data)
2611 av_freep(&pkt_data);
2613 pkt->flags = is_keyframe;
2614 pkt->stream_index = st->index;
2616 if (additional_size > 0) {
2617 uint8_t *side_data = av_packet_new_side_data(pkt,
2618 AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL,
2619 additional_size + 8);
2621 av_free_packet(pkt);
2623 return AVERROR(ENOMEM);
2625 AV_WB64(side_data, additional_id);
2626 memcpy(side_data + 8, additional, additional_size);
2629 if (discard_padding) {
2630 uint8_t *side_data = av_packet_new_side_data(pkt,
2631 AV_PKT_DATA_SKIP_SAMPLES,
2634 av_free_packet(pkt);
2636 return AVERROR(ENOMEM);
2638 AV_WL32(side_data, 0);
2639 AV_WL32(side_data + 4, av_rescale_q(discard_padding,
2640 (AVRational){1, 1000000000},
2641 (AVRational){1, st->codec->sample_rate}));
2644 if (track->ms_compat)
2645 pkt->dts = timecode;
2647 pkt->pts = timecode;
2649 if (st->codec->codec_id == AV_CODEC_ID_SUBRIP) {
2651 * For backward compatibility.
2652 * Historically, we have put subtitle duration
2653 * in convergence_duration, on the off chance
2654 * that the time_scale is less than 1us, which
2655 * could result in a 32bit overflow on the
2656 * normal duration field.
2658 pkt->convergence_duration = lace_duration;
2661 if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE ||
2662 lace_duration <= INT_MAX) {
2664 * For non subtitle tracks, just store the duration
2667 * If it's a subtitle track and duration value does
2668 * not overflow a uint32, then also store it normally.
2670 pkt->duration = lace_duration;
2673 dynarray_add(&matroska->packets, &matroska->num_packets, pkt);
2674 matroska->prev_pkt = pkt;
2679 if (pkt_data != data)
2680 av_freep(&pkt_data);
2684 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
2685 int size, int64_t pos, uint64_t cluster_time,
2686 uint64_t block_duration, int is_keyframe,
2687 uint8_t *additional, uint64_t additional_id, int additional_size,
2688 int64_t cluster_pos, int64_t discard_padding)
2690 uint64_t timecode = AV_NOPTS_VALUE;
2691 MatroskaTrack *track;
2695 uint32_t *lace_size = NULL;
2696 int n, flags, laces = 0;
2698 int trust_default_duration = 1;
2700 if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
2701 av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
2707 track = matroska_find_track_by_num(matroska, num);
2708 if (!track || !track->stream) {
2709 av_log(matroska->ctx, AV_LOG_INFO,
2710 "Invalid stream %"PRIu64" or size %u\n", num, size);
2711 return AVERROR_INVALIDDATA;
2712 } else if (size <= 3)
2715 if (st->discard >= AVDISCARD_ALL)
2717 av_assert1(block_duration != AV_NOPTS_VALUE);
2719 block_time = sign_extend(AV_RB16(data), 16);
2723 if (is_keyframe == -1)
2724 is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
2726 if (cluster_time != (uint64_t) -1 &&
2727 (block_time >= 0 || cluster_time >= -block_time)) {
2728 timecode = cluster_time + block_time - track->codec_delay;
2729 if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE &&
2730 timecode < track->end_timecode)
2731 is_keyframe = 0; /* overlapping subtitles are not key frame */
2733 av_add_index_entry(st, cluster_pos, timecode, 0, 0,
2737 if (matroska->skip_to_keyframe &&
2738 track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
2739 if (timecode < matroska->skip_to_timecode)
2742 matroska->skip_to_keyframe = 0;
2743 else if (!st->skip_to_keyframe) {
2744 av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
2745 matroska->skip_to_keyframe = 0;
2749 res = matroska_parse_laces(matroska, &data, &size, (flags & 0x06) >> 1,
2750 &lace_size, &laces);
2755 if (track->audio.samplerate == 8000) {
2756 // If this is needed for more codecs, then add them here
2757 if (st->codec->codec_id == AV_CODEC_ID_AC3) {
2758 if (track->audio.samplerate != st->codec->sample_rate || !st->codec->frame_size)
2759 trust_default_duration = 0;
2763 if (!block_duration && trust_default_duration)
2764 block_duration = track->default_duration * laces / matroska->time_scale;
2766 if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
2767 track->end_timecode =
2768 FFMAX(track->end_timecode, timecode + block_duration);
2770 for (n = 0; n < laces; n++) {
2771 int64_t lace_duration = block_duration*(n+1) / laces - block_duration*n / laces;
2773 if (lace_size[n] > size) {
2774 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
2778 if ((st->codec->codec_id == AV_CODEC_ID_RA_288 ||
2779 st->codec->codec_id == AV_CODEC_ID_COOK ||
2780 st->codec->codec_id == AV_CODEC_ID_SIPR ||
2781 st->codec->codec_id == AV_CODEC_ID_ATRAC3) &&
2782 st->codec->block_align && track->audio.sub_packet_size) {
2783 res = matroska_parse_rm_audio(matroska, track, st, data,
2789 } else if (st->codec->codec_id == AV_CODEC_ID_WEBVTT) {
2790 res = matroska_parse_webvtt(matroska, track, st,
2792 timecode, lace_duration,
2797 res = matroska_parse_frame(matroska, track, st, data, lace_size[n],
2798 timecode, lace_duration, pos,
2799 !n ? is_keyframe : 0,
2800 additional, additional_id, additional_size,
2806 if (timecode != AV_NOPTS_VALUE)
2807 timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
2808 data += lace_size[n];
2809 size -= lace_size[n];
2817 static int matroska_parse_cluster_incremental(MatroskaDemuxContext *matroska)
2819 EbmlList *blocks_list;
2820 MatroskaBlock *blocks;
2822 res = ebml_parse(matroska,
2823 matroska_cluster_incremental_parsing,
2824 &matroska->current_cluster);
2827 if (matroska->current_cluster_pos)
2828 ebml_level_end(matroska);
2829 ebml_free(matroska_cluster, &matroska->current_cluster);
2830 memset(&matroska->current_cluster, 0, sizeof(MatroskaCluster));
2831 matroska->current_cluster_num_blocks = 0;
2832 matroska->current_cluster_pos = avio_tell(matroska->ctx->pb);
2833 matroska->prev_pkt = NULL;
2834 /* sizeof the ID which was already read */
2835 if (matroska->current_id)
2836 matroska->current_cluster_pos -= 4;
2837 res = ebml_parse(matroska,
2838 matroska_clusters_incremental,
2839 &matroska->current_cluster);
2840 /* Try parsing the block again. */
2842 res = ebml_parse(matroska,
2843 matroska_cluster_incremental_parsing,
2844 &matroska->current_cluster);
2848 matroska->current_cluster_num_blocks <
2849 matroska->current_cluster.blocks.nb_elem) {
2850 blocks_list = &matroska->current_cluster.blocks;
2851 blocks = blocks_list->elem;
2853 matroska->current_cluster_num_blocks = blocks_list->nb_elem;
2854 i = blocks_list->nb_elem - 1;
2855 if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
2856 int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
2857 uint8_t* additional = blocks[i].additional.size > 0 ?
2858 blocks[i].additional.data : NULL;
2859 if (!blocks[i].non_simple)
2860 blocks[i].duration = 0;
2861 res = matroska_parse_block(matroska, blocks[i].bin.data,
2862 blocks[i].bin.size, blocks[i].bin.pos,
2863 matroska->current_cluster.timecode,
2864 blocks[i].duration, is_keyframe,
2865 additional, blocks[i].additional_id,
2866 blocks[i].additional.size,
2867 matroska->current_cluster_pos,
2868 blocks[i].discard_padding);
2875 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
2877 MatroskaCluster cluster = { 0 };
2878 EbmlList *blocks_list;
2879 MatroskaBlock *blocks;
2883 if (!matroska->contains_ssa)
2884 return matroska_parse_cluster_incremental(matroska);
2885 pos = avio_tell(matroska->ctx->pb);
2886 matroska->prev_pkt = NULL;
2887 if (matroska->current_id)
2888 pos -= 4; /* sizeof the ID which was already read */
2889 res = ebml_parse(matroska, matroska_clusters, &cluster);
2890 blocks_list = &cluster.blocks;
2891 blocks = blocks_list->elem;
2892 for (i = 0; i < blocks_list->nb_elem; i++)
2893 if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
2894 int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
2895 res = matroska_parse_block(matroska, blocks[i].bin.data,
2896 blocks[i].bin.size, blocks[i].bin.pos,
2897 cluster.timecode, blocks[i].duration,
2898 is_keyframe, NULL, 0, 0, pos,
2899 blocks[i].discard_padding);
2901 ebml_free(matroska_cluster, &cluster);
2905 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
2907 MatroskaDemuxContext *matroska = s->priv_data;
2909 while (matroska_deliver_packet(matroska, pkt)) {
2910 int64_t pos = avio_tell(matroska->ctx->pb);
2913 if (matroska_parse_cluster(matroska) < 0)
2914 matroska_resync(matroska, pos);
2920 static int matroska_read_seek(AVFormatContext *s, int stream_index,
2921 int64_t timestamp, int flags)
2923 MatroskaDemuxContext *matroska = s->priv_data;
2924 MatroskaTrack *tracks = NULL;
2925 AVStream *st = s->streams[stream_index];
2926 int i, index, index_sub, index_min;
2928 /* Parse the CUES now since we need the index data to seek. */
2929 if (matroska->cues_parsing_deferred > 0) {
2930 matroska->cues_parsing_deferred = 0;
2931 matroska_parse_cues(matroska);
2934 if (!st->nb_index_entries)
2936 timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
2938 if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 || index == st->nb_index_entries - 1) {
2939 avio_seek(s->pb, st->index_entries[st->nb_index_entries - 1].pos,
2941 matroska->current_id = 0;
2942 while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 || index == st->nb_index_entries - 1) {
2943 matroska_clear_queue(matroska);
2944 if (matroska_parse_cluster(matroska) < 0)
2949 matroska_clear_queue(matroska);
2950 if (index < 0 || (matroska->cues_parsing_deferred < 0 && index == st->nb_index_entries - 1))
2954 tracks = matroska->tracks.elem;
2955 for (i = 0; i < matroska->tracks.nb_elem; i++) {
2956 tracks[i].audio.pkt_cnt = 0;
2957 tracks[i].audio.sub_packet_cnt = 0;
2958 tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
2959 tracks[i].end_timecode = 0;
2960 if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE &&
2961 tracks[i].stream->discard != AVDISCARD_ALL) {
2962 index_sub = av_index_search_timestamp(
2963 tracks[i].stream, st->index_entries[index].timestamp,
2964 AVSEEK_FLAG_BACKWARD);
2965 while (index_sub >= 0 &&
2967 tracks[i].stream->index_entries[index_sub].pos < st->index_entries[index_min].pos &&
2968 st->index_entries[index].timestamp - tracks[i].stream->index_entries[index_sub].timestamp < 30000000000 / matroska->time_scale)
2973 avio_seek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
2974 matroska->current_id = 0;
2975 if (flags & AVSEEK_FLAG_ANY) {
2976 st->skip_to_keyframe = 0;
2977 matroska->skip_to_timecode = timestamp;
2979 st->skip_to_keyframe = 1;
2980 matroska->skip_to_timecode = st->index_entries[index].timestamp;
2982 matroska->skip_to_keyframe = 1;
2984 matroska->num_levels = 0;
2985 ff_update_cur_dts(s, st, st->index_entries[index].timestamp);
2988 // slightly hackish but allows proper fallback to
2989 // the generic seeking code.
2990 matroska_clear_queue(matroska);
2991 matroska->current_id = 0;
2992 st->skip_to_keyframe =
2993 matroska->skip_to_keyframe = 0;
2995 matroska->num_levels = 0;
2999 static int matroska_read_close(AVFormatContext *s)
3001 MatroskaDemuxContext *matroska = s->priv_data;
3002 MatroskaTrack *tracks = matroska->tracks.elem;
3005 matroska_clear_queue(matroska);
3007 for (n = 0; n < matroska->tracks.nb_elem; n++)
3008 if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
3009 av_free(tracks[n].audio.buf);
3010 ebml_free(matroska_cluster, &matroska->current_cluster);
3011 ebml_free(matroska_segment, matroska);
3017 int64_t start_time_ns;
3018 int64_t end_time_ns;
3019 int64_t start_offset;
3023 /* This function searches all the Cues and returns the CueDesc corresponding the
3024 * the timestamp ts. Returned CueDesc will be such that start_time_ns <= ts <
3025 * end_time_ns. All 4 fields will be set to -1 if ts >= file's duration.
3027 static CueDesc get_cue_desc(AVFormatContext *s, int64_t ts, int64_t cues_start) {
3028 MatroskaDemuxContext *matroska = s->priv_data;
3031 int nb_index_entries = s->streams[0]->nb_index_entries;
3032 AVIndexEntry *index_entries = s->streams[0]->index_entries;
3033 if (ts >= matroska->duration * matroska->time_scale) return (CueDesc) {-1, -1, -1, -1};
3034 for (i = 1; i < nb_index_entries; i++) {
3035 if (index_entries[i - 1].timestamp * matroska->time_scale <= ts &&
3036 index_entries[i].timestamp * matroska->time_scale > ts) {
3041 cue_desc.start_time_ns = index_entries[i].timestamp * matroska->time_scale;
3042 cue_desc.start_offset = index_entries[i].pos - matroska->segment_start;
3043 if (i != nb_index_entries - 1) {
3044 cue_desc.end_time_ns = index_entries[i + 1].timestamp * matroska->time_scale;
3045 cue_desc.end_offset = index_entries[i + 1].pos - matroska->segment_start;
3047 cue_desc.end_time_ns = matroska->duration * matroska->time_scale;
3048 // FIXME: this needs special handling for files where Cues appear
3049 // before Clusters. the current logic assumes Cues appear after
3051 cue_desc.end_offset = cues_start - matroska->segment_start;
3056 static int webm_clusters_start_with_keyframe(AVFormatContext *s)
3058 MatroskaDemuxContext *matroska = s->priv_data;
3059 int64_t cluster_pos, before_pos;
3061 if (s->streams[0]->nb_index_entries <= 0) return 0;
3062 // seek to the first cluster using cues.
3063 index = av_index_search_timestamp(s->streams[0], 0, 0);
3064 if (index < 0) return 0;
3065 cluster_pos = s->streams[0]->index_entries[index].pos;
3066 before_pos = avio_tell(s->pb);
3068 int64_t cluster_id = 0, cluster_length = 0;
3070 avio_seek(s->pb, cluster_pos, SEEK_SET);
3071 // read cluster id and length
3072 ebml_read_num(matroska, matroska->ctx->pb, 4, &cluster_id);
3073 ebml_read_length(matroska, matroska->ctx->pb, &cluster_length);
3074 if (cluster_id != 0xF43B675) { // done with all clusters
3077 avio_seek(s->pb, cluster_pos, SEEK_SET);
3078 matroska->current_id = 0;
3079 matroska_clear_queue(matroska);
3080 if (matroska_parse_cluster(matroska) < 0 ||
3081 matroska->num_packets <= 0) {
3084 pkt = matroska->packets[0];
3085 cluster_pos += cluster_length + 12; // 12 is the offset of the cluster id and length.
3086 if (!(pkt->flags & AV_PKT_FLAG_KEY)) {
3091 avio_seek(s->pb, before_pos, SEEK_SET);
3095 static int buffer_size_after_time_downloaded(int64_t time_ns, double search_sec, int64_t bps,
3096 double min_buffer, double* buffer,
3097 double* sec_to_download, AVFormatContext *s,
3100 double nano_seconds_per_second = 1000000000.0;
3101 double time_sec = time_ns / nano_seconds_per_second;
3103 int64_t time_to_search_ns = (int64_t)(search_sec * nano_seconds_per_second);
3104 int64_t end_time_ns = time_ns + time_to_search_ns;
3105 double sec_downloaded = 0.0;
3106 CueDesc desc_curr = get_cue_desc(s, time_ns, cues_start);
3107 if (desc_curr.start_time_ns == -1)
3109 *sec_to_download = 0.0;
3111 // Check for non cue start time.
3112 if (time_ns > desc_curr.start_time_ns) {
3113 int64_t cue_nano = desc_curr.end_time_ns - time_ns;
3114 double percent = (double)(cue_nano) / (desc_curr.end_time_ns - desc_curr.start_time_ns);
3115 double cueBytes = (desc_curr.end_offset - desc_curr.start_offset) * percent;
3116 double timeToDownload = (cueBytes * 8.0) / bps;
3118 sec_downloaded += (cue_nano / nano_seconds_per_second) - timeToDownload;
3119 *sec_to_download += timeToDownload;
3121 // Check if the search ends within the first cue.
3122 if (desc_curr.end_time_ns >= end_time_ns) {
3123 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
3124 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
3125 sec_downloaded = percent_to_sub * sec_downloaded;
3126 *sec_to_download = percent_to_sub * *sec_to_download;
3129 if ((sec_downloaded + *buffer) <= min_buffer) {
3133 // Get the next Cue.
3134 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
3137 while (desc_curr.start_time_ns != -1) {
3138 int64_t desc_bytes = desc_curr.end_offset - desc_curr.start_offset;
3139 int64_t desc_ns = desc_curr.end_time_ns - desc_curr.start_time_ns;
3140 double desc_sec = desc_ns / nano_seconds_per_second;
3141 double bits = (desc_bytes * 8.0);
3142 double time_to_download = bits / bps;
3144 sec_downloaded += desc_sec - time_to_download;
3145 *sec_to_download += time_to_download;
3147 if (desc_curr.end_time_ns >= end_time_ns) {
3148 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
3149 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
3150 sec_downloaded = percent_to_sub * sec_downloaded;
3151 *sec_to_download = percent_to_sub * *sec_to_download;
3153 if ((sec_downloaded + *buffer) <= min_buffer)
3158 if ((sec_downloaded + *buffer) <= min_buffer) {
3163 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
3165 *buffer = *buffer + sec_downloaded;
3169 /* This function computes the bandwidth of the WebM file with the help of
3170 * buffer_size_after_time_downloaded() function. Both of these functions are
3171 * adapted from WebM Tools project and are adapted to work with FFmpeg's
3172 * Matroska parsing mechanism.
3174 * Returns the bandwidth of the file on success; -1 on error.
3176 static int64_t webm_dash_manifest_compute_bandwidth(AVFormatContext *s, int64_t cues_start)
3178 MatroskaDemuxContext *matroska = s->priv_data;
3179 AVStream *st = s->streams[0];
3180 double bandwidth = 0.0;
3183 for (i = 0; i < st->nb_index_entries; i++) {
3184 int64_t prebuffer_ns = 1000000000;
3185 int64_t time_ns = st->index_entries[i].timestamp * matroska->time_scale;
3186 double nano_seconds_per_second = 1000000000.0;
3187 int64_t prebuffered_ns = time_ns + prebuffer_ns;
3188 double prebuffer_bytes = 0.0;
3189 int64_t temp_prebuffer_ns = prebuffer_ns;
3190 int64_t pre_bytes, pre_ns;
3191 double pre_sec, prebuffer, bits_per_second;
3192 CueDesc desc_beg = get_cue_desc(s, time_ns, cues_start);
3194 // Start with the first Cue.
3195 CueDesc desc_end = desc_beg;
3197 // Figure out how much data we have downloaded for the prebuffer. This will
3198 // be used later to adjust the bits per sample to try.
3199 while (desc_end.start_time_ns != -1 && desc_end.end_time_ns < prebuffered_ns) {
3200 // Prebuffered the entire Cue.
3201 prebuffer_bytes += desc_end.end_offset - desc_end.start_offset;
3202 temp_prebuffer_ns -= desc_end.end_time_ns - desc_end.start_time_ns;
3203 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
3205 if (desc_end.start_time_ns == -1) {
3206 // The prebuffer is larger than the duration.
3207 if (matroska->duration * matroska->time_scale >= prebuffered_ns)
3209 bits_per_second = 0.0;
3211 // The prebuffer ends in the last Cue. Estimate how much data was
3213 pre_bytes = desc_end.end_offset - desc_end.start_offset;
3214 pre_ns = desc_end.end_time_ns - desc_end.start_time_ns;
3215 pre_sec = pre_ns / nano_seconds_per_second;
3217 pre_bytes * ((temp_prebuffer_ns / nano_seconds_per_second) / pre_sec);
3219 prebuffer = prebuffer_ns / nano_seconds_per_second;
3221 // Set this to 0.0 in case our prebuffer buffers the entire video.
3222 bits_per_second = 0.0;
3224 int64_t desc_bytes = desc_end.end_offset - desc_beg.start_offset;
3225 int64_t desc_ns = desc_end.end_time_ns - desc_beg.start_time_ns;
3226 double desc_sec = desc_ns / nano_seconds_per_second;
3227 double calc_bits_per_second = (desc_bytes * 8) / desc_sec;
3229 // Drop the bps by the percentage of bytes buffered.
3230 double percent = (desc_bytes - prebuffer_bytes) / desc_bytes;
3231 double mod_bits_per_second = calc_bits_per_second * percent;
3233 if (prebuffer < desc_sec) {
3235 (double)(matroska->duration * matroska->time_scale) / nano_seconds_per_second;
3237 // Add 1 so the bits per second should be a little bit greater than file
3239 int64_t bps = (int64_t)(mod_bits_per_second) + 1;
3240 const double min_buffer = 0.0;
3241 double buffer = prebuffer;
3242 double sec_to_download = 0.0;
3244 int rv = buffer_size_after_time_downloaded(prebuffered_ns, search_sec, bps,
3245 min_buffer, &buffer, &sec_to_download,
3249 } else if (rv == 0) {
3250 bits_per_second = (double)(bps);
3255 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
3256 } while (desc_end.start_time_ns != -1);
3258 if (bandwidth < bits_per_second) bandwidth = bits_per_second;
3260 return (int64_t)bandwidth;
3263 static int webm_dash_manifest_cues(AVFormatContext *s)
3265 MatroskaDemuxContext *matroska = s->priv_data;
3266 EbmlList *seekhead_list = &matroska->seekhead;
3267 MatroskaSeekhead *seekhead = seekhead_list->elem;
3269 int64_t cues_start = -1, cues_end = -1, before_pos, bandwidth;
3272 // determine cues start and end positions
3273 for (i = 0; i < seekhead_list->nb_elem; i++)
3274 if (seekhead[i].id == MATROSKA_ID_CUES)
3277 if (i >= seekhead_list->nb_elem) return -1;
3279 before_pos = avio_tell(matroska->ctx->pb);
3280 cues_start = seekhead[i].pos + matroska->segment_start;
3281 if (avio_seek(matroska->ctx->pb, cues_start, SEEK_SET) == cues_start) {
3282 // cues_end is computed as cues_start + cues_length + length of the
3283 // Cues element ID + EBML length of the Cues element. cues_end is
3284 // inclusive and the above sum is reduced by 1.
3285 uint64_t cues_length = 0, cues_id = 0, bytes_read = 0;
3286 bytes_read += ebml_read_num(matroska, matroska->ctx->pb, 4, &cues_id);
3287 bytes_read += ebml_read_length(matroska, matroska->ctx->pb, &cues_length);
3288 cues_end = cues_start + cues_length + bytes_read - 1;
3290 avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
3291 if (cues_start == -1 || cues_end == -1) return -1;
3294 matroska_parse_cues(matroska);
3297 av_dict_set_int(&s->streams[0]->metadata, CUES_START, cues_start, 0);
3300 av_dict_set_int(&s->streams[0]->metadata, CUES_END, cues_end, 0);
3303 bandwidth = webm_dash_manifest_compute_bandwidth(s, cues_start);
3304 if (bandwidth < 0) return -1;
3305 av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH, bandwidth, 0);
3307 // check if all clusters start with key frames
3308 av_dict_set_int(&s->streams[0]->metadata, CLUSTER_KEYFRAME, webm_clusters_start_with_keyframe(s), 0);
3310 // store cue point timestamps as a comma separated list for checking subsegment alignment in
3311 // the muxer. assumes that each timestamp cannot be more than 20 characters long.
3312 buf = av_malloc(s->streams[0]->nb_index_entries * 20 * sizeof(char));
3313 if (!buf) return -1;
3315 for (i = 0; i < s->streams[0]->nb_index_entries; i++) {
3316 snprintf(buf, (i + 1) * 20 * sizeof(char),
3317 "%s%" PRId64, buf, s->streams[0]->index_entries[i].timestamp);
3318 if (i != s->streams[0]->nb_index_entries - 1)
3319 strncat(buf, ",", sizeof(char));
3321 av_dict_set(&s->streams[0]->metadata, CUE_TIMESTAMPS, buf, 0);
3327 static int webm_dash_manifest_read_header(AVFormatContext *s)
3330 int ret = matroska_read_header(s);
3331 MatroskaTrack *tracks;
3332 MatroskaDemuxContext *matroska = s->priv_data;
3334 av_log(s, AV_LOG_ERROR, "Failed to read file headers\n");
3338 // initialization range
3339 // 5 is the offset of Cluster ID.
3340 av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, avio_tell(s->pb) - 5, 0);
3342 // basename of the file
3343 buf = strrchr(s->filename, '/');
3344 av_dict_set(&s->streams[0]->metadata, FILENAME, buf ? ++buf : s->filename, 0);
3347 buf = av_asprintf("%g", matroska->duration);
3348 if (!buf) return AVERROR(ENOMEM);
3349 av_dict_set(&s->streams[0]->metadata, DURATION, buf, 0);
3353 tracks = matroska->tracks.elem;
3354 av_dict_set_int(&s->streams[0]->metadata, TRACK_NUMBER, tracks[0].num, 0);
3356 // parse the cues and populate Cue related fields
3357 return webm_dash_manifest_cues(s);
3360 static int webm_dash_manifest_read_packet(AVFormatContext *s, AVPacket *pkt)
3365 AVInputFormat ff_matroska_demuxer = {
3366 .name = "matroska,webm",
3367 .long_name = NULL_IF_CONFIG_SMALL("Matroska / WebM"),
3368 .extensions = "mkv,mk3d,mka,mks",
3369 .priv_data_size = sizeof(MatroskaDemuxContext),
3370 .read_probe = matroska_probe,
3371 .read_header = matroska_read_header,
3372 .read_packet = matroska_read_packet,
3373 .read_close = matroska_read_close,
3374 .read_seek = matroska_read_seek,
3375 .mime_type = "audio/webm,audio/x-matroska,video/webm,video/x-matroska"
3378 AVInputFormat ff_webm_dash_manifest_demuxer = {
3379 .name = "webm_dash_manifest",
3380 .long_name = NULL_IF_CONFIG_SMALL("WebM DASH Manifest"),
3381 .priv_data_size = sizeof(MatroskaDemuxContext),
3382 .read_header = webm_dash_manifest_read_header,
3383 .read_packet = webm_dash_manifest_read_packet,
3384 .read_close = matroska_read_close,