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/mastering_display_metadata.h"
43 #include "libavutil/mathematics.h"
44 #include "libavutil/opt.h"
45 #include "libavutil/time_internal.h"
46 #include "libavutil/spherical.h"
48 #include "libavcodec/bytestream.h"
49 #include "libavcodec/flac.h"
50 #include "libavcodec/mpeg4audio.h"
53 #include "avio_internal.h"
58 /* For ff_codec_get_id(). */
69 #include "qtpalette.h"
71 #define EBML_UNKNOWN_LENGTH UINT64_MAX /* EBML unknown length, in uint64_t */
72 #define NEEDS_CHECKING 2 /* Indicates that some error checks
73 * still need to be performed */
74 #define LEVEL_ENDED 3 /* return value of ebml_parse when the
75 * syntax level used for parsing ended. */
76 #define SKIP_THRESHOLD 1024 * 1024 /* In non-seekable mode, if more than SKIP_THRESHOLD
77 * of unkown, potentially damaged data is encountered,
78 * it is considered an error. */
79 #define UNKNOWN_EQUIV 50 * 1024 /* An unknown element is considered equivalent
80 * to this many bytes of unknown data for the
81 * SKIP_THRESHOLD check. */
97 typedef const struct EbmlSyntax {
107 const struct EbmlSyntax *n;
111 typedef struct EbmlList {
116 typedef struct EbmlBin {
123 typedef struct Ebml {
128 uint64_t doctype_version;
131 typedef struct MatroskaTrackCompression {
134 } MatroskaTrackCompression;
136 typedef struct MatroskaTrackEncryption {
139 } MatroskaTrackEncryption;
141 typedef struct MatroskaTrackEncoding {
144 MatroskaTrackCompression compression;
145 MatroskaTrackEncryption encryption;
146 } MatroskaTrackEncoding;
148 typedef struct MatroskaMasteringMeta {
157 double max_luminance;
158 double min_luminance;
159 } MatroskaMasteringMeta;
161 typedef struct MatroskaTrackVideoColor {
162 uint64_t matrix_coefficients;
163 uint64_t bits_per_channel;
164 uint64_t chroma_sub_horz;
165 uint64_t chroma_sub_vert;
166 uint64_t cb_sub_horz;
167 uint64_t cb_sub_vert;
168 uint64_t chroma_siting_horz;
169 uint64_t chroma_siting_vert;
171 uint64_t transfer_characteristics;
175 MatroskaMasteringMeta mastering_meta;
176 } MatroskaTrackVideoColor;
178 typedef struct MatroskaTrackVideoProjection {
184 } MatroskaTrackVideoProjection;
186 typedef struct MatroskaTrackVideo {
188 uint64_t display_width;
189 uint64_t display_height;
190 uint64_t pixel_width;
191 uint64_t pixel_height;
193 uint64_t display_unit;
195 uint64_t field_order;
196 uint64_t stereo_mode;
199 MatroskaTrackVideoProjection projection;
200 } MatroskaTrackVideo;
202 typedef struct MatroskaTrackAudio {
204 double out_samplerate;
208 /* real audio header (extracted from extradata) */
215 uint64_t buf_timecode;
217 } MatroskaTrackAudio;
219 typedef struct MatroskaTrackPlane {
222 } MatroskaTrackPlane;
224 typedef struct MatroskaTrackOperation {
225 EbmlList combine_planes;
226 } MatroskaTrackOperation;
228 typedef struct MatroskaTrack {
237 uint64_t default_duration;
238 uint64_t flag_default;
239 uint64_t flag_forced;
240 uint64_t seek_preroll;
241 MatroskaTrackVideo video;
242 MatroskaTrackAudio audio;
243 MatroskaTrackOperation operation;
245 uint64_t codec_delay;
246 uint64_t codec_delay_in_track_tb;
249 int64_t end_timecode;
251 uint64_t max_block_additional_id;
253 uint32_t palette[AVPALETTE_COUNT];
257 typedef struct MatroskaAttachment {
264 } MatroskaAttachment;
266 typedef struct MatroskaChapter {
275 typedef struct MatroskaIndexPos {
280 typedef struct MatroskaIndex {
285 typedef struct MatroskaTag {
293 typedef struct MatroskaTagTarget {
301 typedef struct MatroskaTags {
302 MatroskaTagTarget target;
306 typedef struct MatroskaSeekhead {
311 typedef struct MatroskaLevel {
316 typedef struct MatroskaBlock {
321 uint64_t additional_id;
323 int64_t discard_padding;
326 typedef struct MatroskaCluster {
332 typedef struct MatroskaLevel1Element {
336 } MatroskaLevel1Element;
338 typedef struct MatroskaDemuxContext {
339 const AVClass *class;
340 AVFormatContext *ctx;
343 MatroskaLevel levels[EBML_MAX_DEPTH];
355 EbmlList attachments;
361 /* byte position of the segment inside the stream */
362 int64_t segment_start;
364 /* the packet queue */
366 AVPacketList *queue_end;
370 /* What to skip before effectively reading a packet. */
371 int skip_to_keyframe;
372 uint64_t skip_to_timecode;
374 /* File has a CUES element, but we defer parsing until it is needed. */
375 int cues_parsing_deferred;
377 /* Level1 elements and whether they were read yet */
378 MatroskaLevel1Element level1_elems[64];
379 int num_level1_elems;
381 MatroskaCluster current_cluster;
383 /* WebM DASH Manifest live flag */
386 /* Bandwidth value for WebM DASH Manifest */
388 } MatroskaDemuxContext;
390 #define CHILD_OF(parent) { .def = { .n = parent } }
392 static const EbmlSyntax ebml_syntax[], matroska_segment[], matroska_track_video_color[], matroska_track_video[],
393 matroska_track[], matroska_track_encoding[], matroska_track_encodings[],
394 matroska_track_combine_planes[], matroska_track_operation[], matroska_tracks[],
395 matroska_attachments[], matroska_chapter_entry[], matroska_chapter[], matroska_chapters[],
396 matroska_index_entry[], matroska_index[], matroska_tag[], matroska_tags[], matroska_seekhead[],
397 matroska_blockadditions[], matroska_blockgroup[], matroska_cluster_parsing[];
399 static const EbmlSyntax ebml_header[] = {
400 { EBML_ID_EBMLREADVERSION, EBML_UINT, 0, offsetof(Ebml, version), { .u = EBML_VERSION } },
401 { EBML_ID_EBMLMAXSIZELENGTH, EBML_UINT, 0, offsetof(Ebml, max_size), { .u = 8 } },
402 { EBML_ID_EBMLMAXIDLENGTH, EBML_UINT, 0, offsetof(Ebml, id_length), { .u = 4 } },
403 { EBML_ID_DOCTYPE, EBML_STR, 0, offsetof(Ebml, doctype), { .s = "(none)" } },
404 { EBML_ID_DOCTYPEREADVERSION, EBML_UINT, 0, offsetof(Ebml, doctype_version), { .u = 1 } },
405 { EBML_ID_EBMLVERSION, EBML_NONE },
406 { EBML_ID_DOCTYPEVERSION, EBML_NONE },
407 CHILD_OF(ebml_syntax)
410 static const EbmlSyntax ebml_syntax[] = {
411 { EBML_ID_HEADER, EBML_NEST, 0, 0, { .n = ebml_header } },
412 { MATROSKA_ID_SEGMENT, EBML_STOP },
416 static const EbmlSyntax matroska_info[] = {
417 { MATROSKA_ID_TIMECODESCALE, EBML_UINT, 0, offsetof(MatroskaDemuxContext, time_scale), { .u = 1000000 } },
418 { MATROSKA_ID_DURATION, EBML_FLOAT, 0, offsetof(MatroskaDemuxContext, duration) },
419 { MATROSKA_ID_TITLE, EBML_UTF8, 0, offsetof(MatroskaDemuxContext, title) },
420 { MATROSKA_ID_WRITINGAPP, EBML_NONE },
421 { MATROSKA_ID_MUXINGAPP, EBML_UTF8, 0, offsetof(MatroskaDemuxContext, muxingapp) },
422 { MATROSKA_ID_DATEUTC, EBML_BIN, 0, offsetof(MatroskaDemuxContext, date_utc) },
423 { MATROSKA_ID_SEGMENTUID, EBML_NONE },
424 CHILD_OF(matroska_segment)
427 static const EbmlSyntax matroska_mastering_meta[] = {
428 { MATROSKA_ID_VIDEOCOLOR_RX, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, r_x), { .f=-1 } },
429 { MATROSKA_ID_VIDEOCOLOR_RY, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, r_y), { .f=-1 } },
430 { MATROSKA_ID_VIDEOCOLOR_GX, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, g_x), { .f=-1 } },
431 { MATROSKA_ID_VIDEOCOLOR_GY, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, g_y), { .f=-1 } },
432 { MATROSKA_ID_VIDEOCOLOR_BX, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, b_x), { .f=-1 } },
433 { MATROSKA_ID_VIDEOCOLOR_BY, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, b_y), { .f=-1 } },
434 { MATROSKA_ID_VIDEOCOLOR_WHITEX, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, white_x), { .f=-1 } },
435 { MATROSKA_ID_VIDEOCOLOR_WHITEY, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, white_y), { .f=-1 } },
436 { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMIN, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, min_luminance), { .f=-1 } },
437 { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMAX, EBML_FLOAT, 0, offsetof(MatroskaMasteringMeta, max_luminance), { .f=-1 } },
438 CHILD_OF(matroska_track_video_color)
441 static const EbmlSyntax matroska_track_video_color[] = {
442 { MATROSKA_ID_VIDEOCOLORMATRIXCOEFF, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, matrix_coefficients), { .u = AVCOL_SPC_UNSPECIFIED } },
443 { MATROSKA_ID_VIDEOCOLORBITSPERCHANNEL, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, bits_per_channel), { .u=0 } },
444 { MATROSKA_ID_VIDEOCOLORCHROMASUBHORZ, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_horz), { .u=0 } },
445 { MATROSKA_ID_VIDEOCOLORCHROMASUBVERT, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_vert), { .u=0 } },
446 { MATROSKA_ID_VIDEOCOLORCBSUBHORZ, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, cb_sub_horz), { .u=0 } },
447 { MATROSKA_ID_VIDEOCOLORCBSUBVERT, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, cb_sub_vert), { .u=0 } },
448 { MATROSKA_ID_VIDEOCOLORCHROMASITINGHORZ, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, chroma_siting_horz), { .u = MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED } },
449 { MATROSKA_ID_VIDEOCOLORCHROMASITINGVERT, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, chroma_siting_vert), { .u = MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED } },
450 { MATROSKA_ID_VIDEOCOLORRANGE, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, range), { .u = AVCOL_RANGE_UNSPECIFIED } },
451 { MATROSKA_ID_VIDEOCOLORTRANSFERCHARACTERISTICS, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, transfer_characteristics), { .u = AVCOL_TRC_UNSPECIFIED } },
452 { MATROSKA_ID_VIDEOCOLORPRIMARIES, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, primaries), { .u = AVCOL_PRI_UNSPECIFIED } },
453 { MATROSKA_ID_VIDEOCOLORMAXCLL, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, max_cll), { .u=0 } },
454 { MATROSKA_ID_VIDEOCOLORMAXFALL, EBML_UINT, 0, offsetof(MatroskaTrackVideoColor, max_fall), { .u=0 } },
455 { MATROSKA_ID_VIDEOCOLORMASTERINGMETA, EBML_NEST, 0, offsetof(MatroskaTrackVideoColor, mastering_meta), { .n = matroska_mastering_meta } },
456 CHILD_OF(matroska_track_video)
459 static const EbmlSyntax matroska_track_video_projection[] = {
460 { MATROSKA_ID_VIDEOPROJECTIONTYPE, EBML_UINT, 0, offsetof(MatroskaTrackVideoProjection, type), { .u = MATROSKA_VIDEO_PROJECTION_TYPE_RECTANGULAR } },
461 { MATROSKA_ID_VIDEOPROJECTIONPRIVATE, EBML_BIN, 0, offsetof(MatroskaTrackVideoProjection, private) },
462 { MATROSKA_ID_VIDEOPROJECTIONPOSEYAW, EBML_FLOAT, 0, offsetof(MatroskaTrackVideoProjection, yaw), { .f=0.0 } },
463 { MATROSKA_ID_VIDEOPROJECTIONPOSEPITCH, EBML_FLOAT, 0, offsetof(MatroskaTrackVideoProjection, pitch), { .f=0.0 } },
464 { MATROSKA_ID_VIDEOPROJECTIONPOSEROLL, EBML_FLOAT, 0, offsetof(MatroskaTrackVideoProjection, roll), { .f=0.0 } },
465 CHILD_OF(matroska_track_video)
468 static const EbmlSyntax matroska_track_video[] = {
469 { MATROSKA_ID_VIDEOFRAMERATE, EBML_FLOAT, 0, offsetof(MatroskaTrackVideo, frame_rate) },
470 { MATROSKA_ID_VIDEODISPLAYWIDTH, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_width), { .u=-1 } },
471 { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_height), { .u=-1 } },
472 { MATROSKA_ID_VIDEOPIXELWIDTH, EBML_UINT, 0, offsetof(MatroskaTrackVideo, pixel_width) },
473 { MATROSKA_ID_VIDEOPIXELHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, pixel_height) },
474 { MATROSKA_ID_VIDEOCOLORSPACE, EBML_BIN, 0, offsetof(MatroskaTrackVideo, color_space) },
475 { MATROSKA_ID_VIDEOALPHAMODE, EBML_UINT, 0, offsetof(MatroskaTrackVideo, alpha_mode) },
476 { MATROSKA_ID_VIDEOCOLOR, EBML_NEST, sizeof(MatroskaTrackVideoColor), offsetof(MatroskaTrackVideo, color), { .n = matroska_track_video_color } },
477 { MATROSKA_ID_VIDEOPROJECTION, EBML_NEST, 0, offsetof(MatroskaTrackVideo, projection), { .n = matroska_track_video_projection } },
478 { MATROSKA_ID_VIDEOPIXELCROPB, EBML_NONE },
479 { MATROSKA_ID_VIDEOPIXELCROPT, EBML_NONE },
480 { MATROSKA_ID_VIDEOPIXELCROPL, EBML_NONE },
481 { MATROSKA_ID_VIDEOPIXELCROPR, EBML_NONE },
482 { MATROSKA_ID_VIDEODISPLAYUNIT, EBML_UINT, 0, offsetof(MatroskaTrackVideo, display_unit), { .u= MATROSKA_VIDEO_DISPLAYUNIT_PIXELS } },
483 { MATROSKA_ID_VIDEOFLAGINTERLACED, EBML_UINT, 0, offsetof(MatroskaTrackVideo, interlaced), { .u = MATROSKA_VIDEO_INTERLACE_FLAG_UNDETERMINED } },
484 { MATROSKA_ID_VIDEOFIELDORDER, EBML_UINT, 0, offsetof(MatroskaTrackVideo, field_order), { .u = MATROSKA_VIDEO_FIELDORDER_UNDETERMINED } },
485 { MATROSKA_ID_VIDEOSTEREOMODE, EBML_UINT, 0, offsetof(MatroskaTrackVideo, stereo_mode), { .u = MATROSKA_VIDEO_STEREOMODE_TYPE_NB } },
486 { MATROSKA_ID_VIDEOASPECTRATIO, EBML_NONE },
487 CHILD_OF(matroska_track)
490 static const EbmlSyntax matroska_track_audio[] = {
491 { MATROSKA_ID_AUDIOSAMPLINGFREQ, EBML_FLOAT, 0, offsetof(MatroskaTrackAudio, samplerate), { .f = 8000.0 } },
492 { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, EBML_FLOAT, 0, offsetof(MatroskaTrackAudio, out_samplerate) },
493 { MATROSKA_ID_AUDIOBITDEPTH, EBML_UINT, 0, offsetof(MatroskaTrackAudio, bitdepth) },
494 { MATROSKA_ID_AUDIOCHANNELS, EBML_UINT, 0, offsetof(MatroskaTrackAudio, channels), { .u = 1 } },
495 CHILD_OF(matroska_track)
498 static const EbmlSyntax matroska_track_encoding_compression[] = {
499 { MATROSKA_ID_ENCODINGCOMPALGO, EBML_UINT, 0, offsetof(MatroskaTrackCompression, algo), { .u = 0 } },
500 { MATROSKA_ID_ENCODINGCOMPSETTINGS, EBML_BIN, 0, offsetof(MatroskaTrackCompression, settings) },
501 CHILD_OF(matroska_track_encoding)
504 static const EbmlSyntax matroska_track_encoding_encryption[] = {
505 { MATROSKA_ID_ENCODINGENCALGO, EBML_UINT, 0, offsetof(MatroskaTrackEncryption,algo), {.u = 0} },
506 { MATROSKA_ID_ENCODINGENCKEYID, EBML_BIN, 0, offsetof(MatroskaTrackEncryption,key_id) },
507 { MATROSKA_ID_ENCODINGENCAESSETTINGS, EBML_NONE },
508 { MATROSKA_ID_ENCODINGSIGALGO, EBML_NONE },
509 { MATROSKA_ID_ENCODINGSIGHASHALGO, EBML_NONE },
510 { MATROSKA_ID_ENCODINGSIGKEYID, EBML_NONE },
511 { MATROSKA_ID_ENCODINGSIGNATURE, EBML_NONE },
512 CHILD_OF(matroska_track_encoding)
514 static const EbmlSyntax matroska_track_encoding[] = {
515 { MATROSKA_ID_ENCODINGSCOPE, EBML_UINT, 0, offsetof(MatroskaTrackEncoding, scope), { .u = 1 } },
516 { MATROSKA_ID_ENCODINGTYPE, EBML_UINT, 0, offsetof(MatroskaTrackEncoding, type), { .u = 0 } },
517 { MATROSKA_ID_ENCODINGCOMPRESSION, EBML_NEST, 0, offsetof(MatroskaTrackEncoding, compression), { .n = matroska_track_encoding_compression } },
518 { MATROSKA_ID_ENCODINGENCRYPTION, EBML_NEST, 0, offsetof(MatroskaTrackEncoding, encryption), { .n = matroska_track_encoding_encryption } },
519 { MATROSKA_ID_ENCODINGORDER, EBML_NONE },
520 CHILD_OF(matroska_track_encodings)
523 static const EbmlSyntax matroska_track_encodings[] = {
524 { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack, encodings), { .n = matroska_track_encoding } },
525 CHILD_OF(matroska_track)
528 static const EbmlSyntax matroska_track_plane[] = {
529 { MATROSKA_ID_TRACKPLANEUID, EBML_UINT, 0, offsetof(MatroskaTrackPlane,uid) },
530 { MATROSKA_ID_TRACKPLANETYPE, EBML_UINT, 0, offsetof(MatroskaTrackPlane,type) },
531 CHILD_OF(matroska_track_combine_planes)
534 static const EbmlSyntax matroska_track_combine_planes[] = {
535 { MATROSKA_ID_TRACKPLANE, EBML_NEST, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n = matroska_track_plane} },
536 CHILD_OF(matroska_track_operation)
539 static const EbmlSyntax matroska_track_operation[] = {
540 { MATROSKA_ID_TRACKCOMBINEPLANES, EBML_NEST, 0, 0, {.n = matroska_track_combine_planes} },
541 CHILD_OF(matroska_track)
544 static const EbmlSyntax matroska_track[] = {
545 { MATROSKA_ID_TRACKNUMBER, EBML_UINT, 0, offsetof(MatroskaTrack, num) },
546 { MATROSKA_ID_TRACKNAME, EBML_UTF8, 0, offsetof(MatroskaTrack, name) },
547 { MATROSKA_ID_TRACKUID, EBML_UINT, 0, offsetof(MatroskaTrack, uid) },
548 { MATROSKA_ID_TRACKTYPE, EBML_UINT, 0, offsetof(MatroskaTrack, type) },
549 { MATROSKA_ID_CODECID, EBML_STR, 0, offsetof(MatroskaTrack, codec_id) },
550 { MATROSKA_ID_CODECPRIVATE, EBML_BIN, 0, offsetof(MatroskaTrack, codec_priv) },
551 { MATROSKA_ID_CODECDELAY, EBML_UINT, 0, offsetof(MatroskaTrack, codec_delay) },
552 { MATROSKA_ID_TRACKLANGUAGE, EBML_UTF8, 0, offsetof(MatroskaTrack, language), { .s = "eng" } },
553 { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack, default_duration) },
554 { MATROSKA_ID_TRACKTIMECODESCALE, EBML_FLOAT, 0, offsetof(MatroskaTrack, time_scale), { .f = 1.0 } },
555 { MATROSKA_ID_TRACKFLAGDEFAULT, EBML_UINT, 0, offsetof(MatroskaTrack, flag_default), { .u = 1 } },
556 { MATROSKA_ID_TRACKFLAGFORCED, EBML_UINT, 0, offsetof(MatroskaTrack, flag_forced), { .u = 0 } },
557 { MATROSKA_ID_TRACKVIDEO, EBML_NEST, 0, offsetof(MatroskaTrack, video), { .n = matroska_track_video } },
558 { MATROSKA_ID_TRACKAUDIO, EBML_NEST, 0, offsetof(MatroskaTrack, audio), { .n = matroska_track_audio } },
559 { MATROSKA_ID_TRACKOPERATION, EBML_NEST, 0, offsetof(MatroskaTrack, operation), { .n = matroska_track_operation } },
560 { MATROSKA_ID_TRACKCONTENTENCODINGS, EBML_NEST, 0, 0, { .n = matroska_track_encodings } },
561 { MATROSKA_ID_TRACKMAXBLKADDID, EBML_UINT, 0, offsetof(MatroskaTrack, max_block_additional_id) },
562 { MATROSKA_ID_SEEKPREROLL, EBML_UINT, 0, offsetof(MatroskaTrack, seek_preroll) },
563 { MATROSKA_ID_TRACKFLAGENABLED, EBML_NONE },
564 { MATROSKA_ID_TRACKFLAGLACING, EBML_NONE },
565 { MATROSKA_ID_CODECNAME, EBML_NONE },
566 { MATROSKA_ID_CODECDECODEALL, EBML_NONE },
567 { MATROSKA_ID_CODECINFOURL, EBML_NONE },
568 { MATROSKA_ID_CODECDOWNLOADURL, EBML_NONE },
569 { MATROSKA_ID_TRACKMINCACHE, EBML_NONE },
570 { MATROSKA_ID_TRACKMAXCACHE, EBML_NONE },
571 CHILD_OF(matroska_tracks)
574 static const EbmlSyntax matroska_tracks[] = {
575 { MATROSKA_ID_TRACKENTRY, EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext, tracks), { .n = matroska_track } },
576 CHILD_OF(matroska_segment)
579 static const EbmlSyntax matroska_attachment[] = {
580 { MATROSKA_ID_FILEUID, EBML_UINT, 0, offsetof(MatroskaAttachment, uid) },
581 { MATROSKA_ID_FILENAME, EBML_UTF8, 0, offsetof(MatroskaAttachment, filename) },
582 { MATROSKA_ID_FILEMIMETYPE, EBML_STR, 0, offsetof(MatroskaAttachment, mime) },
583 { MATROSKA_ID_FILEDATA, EBML_BIN, 0, offsetof(MatroskaAttachment, bin) },
584 { MATROSKA_ID_FILEDESC, EBML_NONE },
585 CHILD_OF(matroska_attachments)
588 static const EbmlSyntax matroska_attachments[] = {
589 { MATROSKA_ID_ATTACHEDFILE, EBML_NEST, sizeof(MatroskaAttachment), offsetof(MatroskaDemuxContext, attachments), { .n = matroska_attachment } },
590 CHILD_OF(matroska_segment)
593 static const EbmlSyntax matroska_chapter_display[] = {
594 { MATROSKA_ID_CHAPSTRING, EBML_UTF8, 0, offsetof(MatroskaChapter, title) },
595 { MATROSKA_ID_CHAPLANG, EBML_NONE },
596 { MATROSKA_ID_CHAPCOUNTRY, EBML_NONE },
597 CHILD_OF(matroska_chapter_entry)
600 static const EbmlSyntax matroska_chapter_entry[] = {
601 { MATROSKA_ID_CHAPTERTIMESTART, EBML_UINT, 0, offsetof(MatroskaChapter, start), { .u = AV_NOPTS_VALUE } },
602 { MATROSKA_ID_CHAPTERTIMEEND, EBML_UINT, 0, offsetof(MatroskaChapter, end), { .u = AV_NOPTS_VALUE } },
603 { MATROSKA_ID_CHAPTERUID, EBML_UINT, 0, offsetof(MatroskaChapter, uid) },
604 { MATROSKA_ID_CHAPTERDISPLAY, EBML_NEST, 0, 0, { .n = matroska_chapter_display } },
605 { MATROSKA_ID_CHAPTERFLAGHIDDEN, EBML_NONE },
606 { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
607 { MATROSKA_ID_CHAPTERPHYSEQUIV, EBML_NONE },
608 { MATROSKA_ID_CHAPTERATOM, EBML_NONE },
609 CHILD_OF(matroska_chapter)
612 static const EbmlSyntax matroska_chapter[] = {
613 { MATROSKA_ID_CHAPTERATOM, EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext, chapters), { .n = matroska_chapter_entry } },
614 { MATROSKA_ID_EDITIONUID, EBML_NONE },
615 { MATROSKA_ID_EDITIONFLAGHIDDEN, EBML_NONE },
616 { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
617 { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
618 CHILD_OF(matroska_chapters)
621 static const EbmlSyntax matroska_chapters[] = {
622 { MATROSKA_ID_EDITIONENTRY, EBML_NEST, 0, 0, { .n = matroska_chapter } },
623 CHILD_OF(matroska_segment)
626 static const EbmlSyntax matroska_index_pos[] = {
627 { MATROSKA_ID_CUETRACK, EBML_UINT, 0, offsetof(MatroskaIndexPos, track) },
628 { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos, pos) },
629 { MATROSKA_ID_CUERELATIVEPOSITION,EBML_NONE },
630 { MATROSKA_ID_CUEDURATION, EBML_NONE },
631 { MATROSKA_ID_CUEBLOCKNUMBER, EBML_NONE },
632 CHILD_OF(matroska_index_entry)
635 static const EbmlSyntax matroska_index_entry[] = {
636 { MATROSKA_ID_CUETIME, EBML_UINT, 0, offsetof(MatroskaIndex, time) },
637 { MATROSKA_ID_CUETRACKPOSITION, EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex, pos), { .n = matroska_index_pos } },
638 CHILD_OF(matroska_index)
641 static const EbmlSyntax matroska_index[] = {
642 { MATROSKA_ID_POINTENTRY, EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext, index), { .n = matroska_index_entry } },
643 CHILD_OF(matroska_segment)
646 static const EbmlSyntax matroska_simpletag[] = {
647 { MATROSKA_ID_TAGNAME, EBML_UTF8, 0, offsetof(MatroskaTag, name) },
648 { MATROSKA_ID_TAGSTRING, EBML_UTF8, 0, offsetof(MatroskaTag, string) },
649 { MATROSKA_ID_TAGLANG, EBML_STR, 0, offsetof(MatroskaTag, lang), { .s = "und" } },
650 { MATROSKA_ID_TAGDEFAULT, EBML_UINT, 0, offsetof(MatroskaTag, def) },
651 { MATROSKA_ID_TAGDEFAULT_BUG, EBML_UINT, 0, offsetof(MatroskaTag, def) },
652 { MATROSKA_ID_SIMPLETAG, EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag, sub), { .n = matroska_simpletag } },
653 CHILD_OF(matroska_tag)
656 static const EbmlSyntax matroska_tagtargets[] = {
657 { MATROSKA_ID_TAGTARGETS_TYPE, EBML_STR, 0, offsetof(MatroskaTagTarget, type) },
658 { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget, typevalue), { .u = 50 } },
659 { MATROSKA_ID_TAGTARGETS_TRACKUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, trackuid) },
660 { MATROSKA_ID_TAGTARGETS_CHAPTERUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, chapteruid) },
661 { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget, attachuid) },
662 CHILD_OF(matroska_tag)
665 static const EbmlSyntax matroska_tag[] = {
666 { MATROSKA_ID_SIMPLETAG, EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags, tag), { .n = matroska_simpletag } },
667 { MATROSKA_ID_TAGTARGETS, EBML_NEST, 0, offsetof(MatroskaTags, target), { .n = matroska_tagtargets } },
668 CHILD_OF(matroska_tags)
671 static const EbmlSyntax matroska_tags[] = {
672 { MATROSKA_ID_TAG, EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext, tags), { .n = matroska_tag } },
673 CHILD_OF(matroska_segment)
676 static const EbmlSyntax matroska_seekhead_entry[] = {
677 { MATROSKA_ID_SEEKID, EBML_UINT, 0, offsetof(MatroskaSeekhead, id) },
678 { MATROSKA_ID_SEEKPOSITION, EBML_UINT, 0, offsetof(MatroskaSeekhead, pos), { .u = -1 } },
679 CHILD_OF(matroska_seekhead)
682 static const EbmlSyntax matroska_seekhead[] = {
683 { MATROSKA_ID_SEEKENTRY, EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext, seekhead), { .n = matroska_seekhead_entry } },
684 CHILD_OF(matroska_segment)
687 static const EbmlSyntax matroska_segment[] = {
688 { MATROSKA_ID_INFO, EBML_LEVEL1, 0, 0, { .n = matroska_info } },
689 { MATROSKA_ID_TRACKS, EBML_LEVEL1, 0, 0, { .n = matroska_tracks } },
690 { MATROSKA_ID_ATTACHMENTS, EBML_LEVEL1, 0, 0, { .n = matroska_attachments } },
691 { MATROSKA_ID_CHAPTERS, EBML_LEVEL1, 0, 0, { .n = matroska_chapters } },
692 { MATROSKA_ID_CUES, EBML_LEVEL1, 0, 0, { .n = matroska_index } },
693 { MATROSKA_ID_TAGS, EBML_LEVEL1, 0, 0, { .n = matroska_tags } },
694 { MATROSKA_ID_SEEKHEAD, EBML_LEVEL1, 0, 0, { .n = matroska_seekhead } },
695 { MATROSKA_ID_CLUSTER, EBML_STOP },
696 { 0 } /* We don't want to go back to level 0, so don't add the parent. */
699 static const EbmlSyntax matroska_segments[] = {
700 { MATROSKA_ID_SEGMENT, EBML_NEST, 0, 0, { .n = matroska_segment } },
704 static const EbmlSyntax matroska_blockmore[] = {
705 { MATROSKA_ID_BLOCKADDID, EBML_UINT, 0, offsetof(MatroskaBlock,additional_id) },
706 { MATROSKA_ID_BLOCKADDITIONAL, EBML_BIN, 0, offsetof(MatroskaBlock,additional) },
707 CHILD_OF(matroska_blockadditions)
710 static const EbmlSyntax matroska_blockadditions[] = {
711 { MATROSKA_ID_BLOCKMORE, EBML_NEST, 0, 0, {.n = matroska_blockmore} },
712 CHILD_OF(matroska_blockgroup)
715 static const EbmlSyntax matroska_blockgroup[] = {
716 { MATROSKA_ID_BLOCK, EBML_BIN, 0, offsetof(MatroskaBlock, bin) },
717 { MATROSKA_ID_BLOCKADDITIONS, EBML_NEST, 0, 0, { .n = matroska_blockadditions} },
718 { MATROSKA_ID_BLOCKDURATION, EBML_UINT, 0, offsetof(MatroskaBlock, duration) },
719 { MATROSKA_ID_DISCARDPADDING, EBML_SINT, 0, offsetof(MatroskaBlock, discard_padding) },
720 { MATROSKA_ID_BLOCKREFERENCE, EBML_SINT, 0, offsetof(MatroskaBlock, reference), { .i = INT64_MIN } },
721 { MATROSKA_ID_CODECSTATE, EBML_NONE },
722 { 1, EBML_UINT, 0, offsetof(MatroskaBlock, non_simple), { .u = 1 } },
723 CHILD_OF(matroska_cluster_parsing)
726 // The following array contains SimpleBlock and BlockGroup twice
727 // in order to reuse the other values for matroska_cluster_enter.
728 static const EbmlSyntax matroska_cluster_parsing[] = {
729 { MATROSKA_ID_SIMPLEBLOCK, EBML_BIN, 0, offsetof(MatroskaBlock, bin) },
730 { MATROSKA_ID_BLOCKGROUP, EBML_NEST, 0, 0, { .n = matroska_blockgroup } },
731 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, offsetof(MatroskaCluster, timecode) },
732 { MATROSKA_ID_SIMPLEBLOCK, EBML_STOP },
733 { MATROSKA_ID_BLOCKGROUP, EBML_STOP },
734 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
735 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
736 CHILD_OF(matroska_segment)
739 static const EbmlSyntax matroska_cluster_enter[] = {
740 { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, { .n = &matroska_cluster_parsing[2] } },
744 static const EbmlSyntax matroska_clusters[] = {
745 { MATROSKA_ID_CLUSTER, EBML_STOP },
746 { MATROSKA_ID_CUES, EBML_NONE },
747 { MATROSKA_ID_TAGS, EBML_NONE },
748 { MATROSKA_ID_INFO, EBML_NONE },
749 { MATROSKA_ID_TRACKS, EBML_NONE },
750 { MATROSKA_ID_ATTACHMENTS, EBML_NONE },
751 { MATROSKA_ID_CHAPTERS, EBML_NONE },
752 { MATROSKA_ID_SEEKHEAD, EBML_NONE },
753 { 0 } /* We don't want to go back to level 0, so don't add the parent. */
757 static const char *const matroska_doctypes[] = { "matroska", "webm" };
759 static int matroska_read_close(AVFormatContext *s);
762 * This function prepares the status for parsing of level 1 elements.
764 static int matroska_reset_status(MatroskaDemuxContext *matroska,
765 uint32_t id, int64_t position)
768 int err = avio_seek(matroska->ctx->pb, position, SEEK_SET);
773 matroska->current_id = id;
774 matroska->num_levels = 1;
775 matroska->unknown_count = 0;
776 matroska->resync_pos = avio_tell(matroska->ctx->pb);
778 matroska->resync_pos -= (av_log2(id) + 7) / 8;
783 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
785 AVIOContext *pb = matroska->ctx->pb;
788 /* Try to seek to the last position to resync from. If this doesn't work,
789 * we resync from the earliest position available: The start of the buffer. */
790 if (last_pos < avio_tell(pb) && avio_seek(pb, last_pos + 1, SEEK_SET) < 0) {
791 av_log(matroska->ctx, AV_LOG_WARNING,
792 "Seek to desired resync point failed. Seeking to "
793 "earliest point available instead.\n");
794 avio_seek(pb, FFMAX(avio_tell(pb) + (pb->buffer - pb->buf_ptr),
795 last_pos + 1), SEEK_SET);
800 // try to find a toplevel element
801 while (!avio_feof(pb)) {
802 if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
803 id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
804 id == MATROSKA_ID_SEEKHEAD || id == MATROSKA_ID_ATTACHMENTS ||
805 id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS) {
806 /* Prepare the context for parsing of a level 1 element. */
807 matroska_reset_status(matroska, id, -1);
808 /* Given that we are here means that an error has occured,
809 * so treat the segment as unknown length in order not to
810 * discard valid data that happens to be beyond the designated
811 * end of the segment. */
812 matroska->levels[0].length = EBML_UNKNOWN_LENGTH;
815 id = (id << 8) | avio_r8(pb);
819 return pb->error ? pb->error : AVERROR_EOF;
823 * Read: an "EBML number", which is defined as a variable-length
824 * array of bytes. The first byte indicates the length by giving a
825 * number of 0-bits followed by a one. The position of the first
826 * "one" bit inside the first byte indicates the length of this
828 * Returns: number of bytes read, < 0 on error
830 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
831 int max_size, uint64_t *number, int eof_forbidden)
837 /* The first byte tells us the length in bytes - except when it is zero. */
842 /* get the length of the EBML number */
843 read = 8 - ff_log2_tab[total];
845 if (!total || read > max_size) {
846 pos = avio_tell(pb) - 1;
848 av_log(matroska->ctx, AV_LOG_ERROR,
849 "0x00 at pos %"PRId64" (0x%"PRIx64") invalid as first byte "
850 "of an EBML number\n", pos, pos);
852 av_log(matroska->ctx, AV_LOG_ERROR,
853 "Length %d indicated by an EBML number's first byte 0x%02x "
854 "at pos %"PRId64" (0x%"PRIx64") exceeds max length %d.\n",
855 read, (uint8_t) total, pos, pos, max_size);
857 return AVERROR_INVALIDDATA;
860 /* read out length */
861 total ^= 1 << ff_log2_tab[total];
863 total = (total << 8) | avio_r8(pb);
865 if (pb->eof_reached) {
877 av_log(matroska->ctx, AV_LOG_ERROR,
878 "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
883 av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
884 "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
891 * Read a EBML length value.
892 * This needs special handling for the "unknown length" case which has multiple
895 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
898 int res = ebml_read_num(matroska, pb, 8, number, 1);
899 if (res > 0 && *number + 1 == 1ULL << (7 * res))
900 *number = EBML_UNKNOWN_LENGTH;
905 * Read the next element as an unsigned int.
906 * Returns NEEDS_CHECKING.
908 static int ebml_read_uint(AVIOContext *pb, int size, uint64_t *num)
912 /* big-endian ordering; build up number */
915 *num = (*num << 8) | avio_r8(pb);
917 return NEEDS_CHECKING;
921 * Read the next element as a signed int.
922 * Returns NEEDS_CHECKING.
924 static int ebml_read_sint(AVIOContext *pb, int size, int64_t *num)
931 *num = sign_extend(avio_r8(pb), 8);
933 /* big-endian ordering; build up number */
935 *num = ((uint64_t)*num << 8) | avio_r8(pb);
938 return NEEDS_CHECKING;
942 * Read the next element as a float.
943 * Returns NEEDS_CHECKING or < 0 on obvious failure.
945 static int ebml_read_float(AVIOContext *pb, int size, double *num)
950 *num = av_int2float(avio_rb32(pb));
952 *num = av_int2double(avio_rb64(pb));
954 return AVERROR_INVALIDDATA;
956 return NEEDS_CHECKING;
960 * Read the next element as an ASCII string.
961 * 0 is success, < 0 or NEEDS_CHECKING is failure.
963 static int ebml_read_ascii(AVIOContext *pb, int size, char **str)
968 /* EBML strings are usually not 0-terminated, so we allocate one
969 * byte more, read the string and NULL-terminate it ourselves. */
970 if (!(res = av_malloc(size + 1)))
971 return AVERROR(ENOMEM);
972 if ((ret = avio_read(pb, (uint8_t *) res, size)) != size) {
974 return ret < 0 ? ret : NEEDS_CHECKING;
984 * Read the next element as binary data.
985 * 0 is success, < 0 or NEEDS_CHECKING is failure.
987 static int ebml_read_binary(AVIOContext *pb, int length,
988 int64_t pos, EbmlBin *bin)
992 ret = av_buffer_realloc(&bin->buf, length + AV_INPUT_BUFFER_PADDING_SIZE);
995 memset(bin->buf->data + length, 0, AV_INPUT_BUFFER_PADDING_SIZE);
997 bin->data = bin->buf->data;
1000 if ((ret = avio_read(pb, bin->data, length)) != length) {
1001 av_buffer_unref(&bin->buf);
1004 return ret < 0 ? ret : NEEDS_CHECKING;
1011 * Read the next element, but only the header. The contents
1012 * are supposed to be sub-elements which can be read separately.
1013 * 0 is success, < 0 is failure.
1015 static int ebml_read_master(MatroskaDemuxContext *matroska,
1016 uint64_t length, int64_t pos)
1018 MatroskaLevel *level;
1020 if (matroska->num_levels >= EBML_MAX_DEPTH) {
1021 av_log(matroska->ctx, AV_LOG_ERROR,
1022 "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
1023 return AVERROR(ENOSYS);
1026 level = &matroska->levels[matroska->num_levels++];
1028 level->length = length;
1034 * Read signed/unsigned "EBML" numbers.
1035 * Return: number of bytes processed, < 0 on error
1037 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
1038 uint8_t *data, uint32_t size, uint64_t *num)
1041 ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
1042 return ebml_read_num(matroska, &pb, FFMIN(size, 8), num, 1);
1046 * Same as above, but signed.
1048 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
1049 uint8_t *data, uint32_t size, int64_t *num)
1054 /* read as unsigned number first */
1055 if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
1058 /* make signed (weird way) */
1059 *num = unum - ((1LL << (7 * res - 1)) - 1);
1064 static int ebml_parse(MatroskaDemuxContext *matroska,
1065 EbmlSyntax *syntax, void *data);
1067 static EbmlSyntax *ebml_parse_id(EbmlSyntax *syntax, uint32_t id)
1071 // Whoever touches this should be aware of the duplication
1072 // existing in matroska_cluster_parsing.
1073 for (i = 0; syntax[i].id; i++)
1074 if (id == syntax[i].id)
1080 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
1085 for (i = 0; syntax[i].id; i++)
1086 switch (syntax[i].type) {
1088 *(uint64_t *) ((char *) data + syntax[i].data_offset) = syntax[i].def.u;
1091 *(int64_t *) ((char *) data + syntax[i].data_offset) = syntax[i].def.i;
1094 *(double *) ((char *) data + syntax[i].data_offset) = syntax[i].def.f;
1098 // the default may be NULL
1099 if (syntax[i].def.s) {
1100 uint8_t **dst = (uint8_t **) ((uint8_t *) data + syntax[i].data_offset);
1101 *dst = av_strdup(syntax[i].def.s);
1103 return AVERROR(ENOMEM);
1108 if (!matroska->levels[matroska->num_levels - 1].length) {
1109 matroska->num_levels--;
1114 res = ebml_parse(matroska, syntax, data);
1117 return res == LEVEL_ENDED ? 0 : res;
1120 static int is_ebml_id_valid(uint32_t id)
1122 // Due to endian nonsense in Matroska, the highest byte with any bits set
1123 // will contain the leading length bit. This bit in turn identifies the
1124 // total byte length of the element by its position within the byte.
1125 unsigned int bits = av_log2(id);
1126 return id && (bits + 7) / 8 == (8 - bits % 8);
1130 * Allocate and return the entry for the level1 element with the given ID. If
1131 * an entry already exists, return the existing entry.
1133 static MatroskaLevel1Element *matroska_find_level1_elem(MatroskaDemuxContext *matroska,
1137 MatroskaLevel1Element *elem;
1139 if (!is_ebml_id_valid(id))
1142 // Some files link to all clusters; useless.
1143 if (id == MATROSKA_ID_CLUSTER)
1146 // There can be multiple seekheads.
1147 if (id != MATROSKA_ID_SEEKHEAD) {
1148 for (i = 0; i < matroska->num_level1_elems; i++) {
1149 if (matroska->level1_elems[i].id == id)
1150 return &matroska->level1_elems[i];
1154 // Only a completely broken file would have more elements.
1155 // It also provides a low-effort way to escape from circular seekheads
1156 // (every iteration will add a level1 entry).
1157 if (matroska->num_level1_elems >= FF_ARRAY_ELEMS(matroska->level1_elems)) {
1158 av_log(matroska->ctx, AV_LOG_ERROR, "Too many level1 elements or circular seekheads.\n");
1162 elem = &matroska->level1_elems[matroska->num_level1_elems++];
1163 *elem = (MatroskaLevel1Element){.id = id};
1168 static int ebml_parse(MatroskaDemuxContext *matroska,
1169 EbmlSyntax *syntax, void *data)
1171 static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
1172 // Forbid unknown-length EBML_NONE elements.
1173 [EBML_NONE] = EBML_UNKNOWN_LENGTH - 1,
1177 // max. 16 MB for strings
1178 [EBML_STR] = 0x1000000,
1179 [EBML_UTF8] = 0x1000000,
1180 // max. 256 MB for binary data
1181 [EBML_BIN] = 0x10000000,
1182 // no limits for anything else
1184 AVIOContext *pb = matroska->ctx->pb;
1187 int64_t pos = avio_tell(pb), pos_alt;
1188 int res, update_pos = 1, level_check;
1189 MatroskaLevel1Element *level1_elem;
1190 MatroskaLevel *level = matroska->num_levels ? &matroska->levels[matroska->num_levels - 1] : NULL;
1192 if (!matroska->current_id) {
1194 res = ebml_read_num(matroska, pb, 4, &id, 0);
1196 if (pb->eof_reached && res == AVERROR_EOF) {
1197 if (matroska->is_live)
1198 // in live mode, finish parsing if EOF is reached.
1200 if (level && pos == avio_tell(pb)) {
1201 if (level->length == EBML_UNKNOWN_LENGTH) {
1202 // Unknown-length levels automatically end at EOF.
1203 matroska->num_levels--;
1206 av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
1207 "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
1213 matroska->current_id = id | 1 << 7 * res;
1214 pos_alt = pos + res;
1217 pos -= (av_log2(matroska->current_id) + 7) / 8;
1220 id = matroska->current_id;
1222 syntax = ebml_parse_id(syntax, id);
1223 if (!syntax->id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
1224 if (level && level->length == EBML_UNKNOWN_LENGTH) {
1225 // Unknown-length levels end when an element from an upper level
1226 // in the hierarchy is encountered.
1227 while (syntax->def.n) {
1228 syntax = ebml_parse_id(syntax->def.n, id);
1230 matroska->num_levels--;
1236 av_log(matroska->ctx, AV_LOG_DEBUG, "Unknown entry 0x%"PRIX32" at pos. "
1237 "%"PRId64"\n", id, pos);
1238 update_pos = 0; /* Don't update resync_pos as an error might have happened. */
1241 data = (char *) data + syntax->data_offset;
1242 if (syntax->list_elem_size) {
1243 EbmlList *list = data;
1244 void *newelem = av_realloc_array(list->elem, list->nb_elem + 1,
1245 syntax->list_elem_size);
1247 return AVERROR(ENOMEM);
1248 list->elem = newelem;
1249 data = (char *) list->elem + list->nb_elem * syntax->list_elem_size;
1250 memset(data, 0, syntax->list_elem_size);
1254 if (syntax->type != EBML_STOP) {
1255 matroska->current_id = 0;
1256 if ((res = ebml_read_length(matroska, pb, &length)) < 0)
1261 if (matroska->num_levels > 0) {
1262 if (length != EBML_UNKNOWN_LENGTH &&
1263 level->length != EBML_UNKNOWN_LENGTH) {
1264 uint64_t elem_end = pos_alt + length,
1265 level_end = level->start + level->length;
1267 if (elem_end < level_end) {
1269 } else if (elem_end == level_end) {
1270 level_check = LEVEL_ENDED;
1272 av_log(matroska->ctx, AV_LOG_ERROR,
1273 "Element at 0x%"PRIx64" ending at 0x%"PRIx64" exceeds "
1274 "containing master element ending at 0x%"PRIx64"\n",
1275 pos, elem_end, level_end);
1276 return AVERROR_INVALIDDATA;
1278 } else if (length != EBML_UNKNOWN_LENGTH) {
1280 } else if (level->length != EBML_UNKNOWN_LENGTH) {
1281 av_log(matroska->ctx, AV_LOG_ERROR, "Unknown-sized element "
1282 "at 0x%"PRIx64" inside parent with finite size\n", pos);
1283 return AVERROR_INVALIDDATA;
1286 if (id != MATROSKA_ID_CLUSTER && (syntax->type == EBML_LEVEL1
1287 || syntax->type == EBML_NEST)) {
1288 // According to the current specifications only clusters and
1289 // segments are allowed to be unknown-length. We also accept
1290 // other unknown-length master elements.
1291 av_log(matroska->ctx, AV_LOG_WARNING,
1292 "Found unknown-length element 0x%"PRIX32" other than "
1293 "a cluster at 0x%"PRIx64". Spec-incompliant, but "
1294 "parsing will nevertheless be attempted.\n", id, pos);
1301 if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
1302 if (length != EBML_UNKNOWN_LENGTH) {
1303 av_log(matroska->ctx, AV_LOG_ERROR,
1304 "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for element "
1305 "with ID 0x%"PRIX32" at 0x%"PRIx64"\n",
1306 length, max_lengths[syntax->type], id, pos);
1307 } else if (syntax->type != EBML_NONE) {
1308 av_log(matroska->ctx, AV_LOG_ERROR,
1309 "Element with ID 0x%"PRIX32" at pos. 0x%"PRIx64" has "
1310 "unknown length, yet the length of an element of its "
1311 "type must be known.\n", id, pos);
1313 av_log(matroska->ctx, AV_LOG_ERROR,
1314 "Found unknown-length element with ID 0x%"PRIX32" at "
1315 "pos. 0x%"PRIx64" for which no syntax for parsing is "
1316 "available.\n", id, pos);
1318 return AVERROR_INVALIDDATA;
1321 if (!(pb->seekable & AVIO_SEEKABLE_NORMAL)) {
1322 // Loosing sync will likely manifest itself as encountering unknown
1323 // elements which are not reliably distinguishable from elements
1324 // belonging to future extensions of the format.
1325 // We use a heuristic to detect such situations: If the current
1326 // element is not expected at the current syntax level and there
1327 // were only a few unknown elements in a row, then the element is
1328 // skipped or considered defective based upon the length of the
1329 // current element (i.e. how much would be skipped); if there were
1330 // more than a few skipped elements in a row and skipping the current
1331 // element would lead us more than SKIP_THRESHOLD away from the last
1332 // known good position, then it is inferred that an error occured.
1333 // The dependency on the number of unknown elements in a row exists
1334 // because the distance to the last known good position is
1335 // automatically big if the last parsed element was big.
1336 // In both cases, each unknown element is considered equivalent to
1337 // UNKNOWN_EQUIV of skipped bytes for the check.
1338 // The whole check is only done for non-seekable output, because
1339 // in this situation skipped data can't simply be rechecked later.
1340 // This is especially important when using unkown length elements
1341 // as the check for whether a child exceeds its containing master
1342 // element is not effective in this situation.
1344 matroska->unknown_count = 0;
1346 int64_t dist = length + UNKNOWN_EQUIV * matroska->unknown_count++;
1348 if (matroska->unknown_count > 3)
1349 dist += pos_alt - matroska->resync_pos;
1351 if (dist > SKIP_THRESHOLD) {
1352 av_log(matroska->ctx, AV_LOG_ERROR,
1353 "Unknown element %"PRIX32" at pos. 0x%"PRIx64" with "
1354 "length 0x%"PRIx64" considered as invalid data. Last "
1355 "known good position 0x%"PRIx64", %d unknown elements"
1356 " in a row\n", id, pos, length, matroska->resync_pos,
1357 matroska->unknown_count);
1358 return AVERROR_INVALIDDATA;
1363 if (update_pos > 0) {
1364 // We have found an element that is allowed at this place
1365 // in the hierarchy and it passed all checks, so treat the beginning
1366 // of the element as the "last known good" position.
1367 matroska->resync_pos = pos;
1371 switch (syntax->type) {
1373 res = ebml_read_uint(pb, length, data);
1376 res = ebml_read_sint(pb, length, data);
1379 res = ebml_read_float(pb, length, data);
1383 res = ebml_read_ascii(pb, length, data);
1386 res = ebml_read_binary(pb, length, pos_alt, data);
1390 if ((res = ebml_read_master(matroska, length, pos_alt)) < 0)
1392 if (id == MATROSKA_ID_SEGMENT)
1393 matroska->segment_start = pos_alt;
1394 if (id == MATROSKA_ID_CUES)
1395 matroska->cues_parsing_deferred = 0;
1396 if (syntax->type == EBML_LEVEL1 &&
1397 (level1_elem = matroska_find_level1_elem(matroska, syntax->id))) {
1398 if (level1_elem->parsed)
1399 av_log(matroska->ctx, AV_LOG_ERROR, "Duplicate element\n");
1400 level1_elem->parsed = 1;
1402 if (res = ebml_parse_nest(matroska, syntax->def.n, data))
1410 if (ffio_limit(pb, length) != length) {
1411 // ffio_limit emits its own error message,
1412 // so we don't have to.
1413 return AVERROR(EIO);
1415 if ((res2 = avio_skip(pb, length - 1)) >= 0) {
1416 // avio_skip might take us past EOF. We check for this
1417 // by skipping only length - 1 bytes, reading a byte and
1418 // checking the error flags. This is done in order to check
1419 // that the element has been properly skipped even when
1420 // no filesize (that ffio_limit relies on) is available.
1422 res = NEEDS_CHECKING;
1429 if (res == NEEDS_CHECKING) {
1430 if (pb->eof_reached) {
1439 if (res == AVERROR_INVALIDDATA)
1440 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
1441 else if (res == AVERROR(EIO))
1442 av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
1443 else if (res == AVERROR_EOF) {
1444 av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely\n");
1452 if (level_check == LEVEL_ENDED && matroska->num_levels) {
1453 level = &matroska->levels[matroska->num_levels - 1];
1454 pos = avio_tell(pb);
1456 // Given that pos >= level->start no check for
1457 // level->length != EBML_UNKNOWN_LENGTH is necessary.
1458 while (matroska->num_levels && pos == level->start + level->length) {
1459 matroska->num_levels--;
1467 static void ebml_free(EbmlSyntax *syntax, void *data)
1470 for (i = 0; syntax[i].id; i++) {
1471 void *data_off = (char *) data + syntax[i].data_offset;
1472 switch (syntax[i].type) {
1478 av_buffer_unref(&((EbmlBin *) data_off)->buf);
1482 if (syntax[i].list_elem_size) {
1483 EbmlList *list = data_off;
1484 char *ptr = list->elem;
1485 for (j = 0; j < list->nb_elem;
1486 j++, ptr += syntax[i].list_elem_size)
1487 ebml_free(syntax[i].def.n, ptr);
1488 av_freep(&list->elem);
1491 ebml_free(syntax[i].def.n, data_off);
1501 static int matroska_probe(const AVProbeData *p)
1504 int len_mask = 0x80, size = 1, n = 1, i;
1507 if (AV_RB32(p->buf) != EBML_ID_HEADER)
1510 /* length of header */
1512 while (size <= 8 && !(total & len_mask)) {
1518 total &= (len_mask - 1);
1520 total = (total << 8) | p->buf[4 + n++];
1522 if (total + 1 == 1ULL << (7 * size)){
1523 /* Unknown-length header - simply parse the whole buffer. */
1524 total = p->buf_size - 4 - size;
1526 /* Does the probe data contain the whole header? */
1527 if (p->buf_size < 4 + size + total)
1531 /* The header should contain a known document type. For now,
1532 * we don't parse the whole header but simply check for the
1533 * availability of that array of characters inside the header.
1534 * Not fully fool-proof, but good enough. */
1535 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1536 size_t probelen = strlen(matroska_doctypes[i]);
1537 if (total < probelen)
1539 for (n = 4 + size; n <= 4 + size + total - probelen; n++)
1540 if (!memcmp(p->buf + n, matroska_doctypes[i], probelen))
1541 return AVPROBE_SCORE_MAX;
1544 // probably valid EBML header but no recognized doctype
1545 return AVPROBE_SCORE_EXTENSION;
1548 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
1551 MatroskaTrack *tracks = matroska->tracks.elem;
1554 for (i = 0; i < matroska->tracks.nb_elem; i++)
1555 if (tracks[i].num == num)
1558 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
1562 static int matroska_decode_buffer(uint8_t **buf, int *buf_size,
1563 MatroskaTrack *track)
1565 MatroskaTrackEncoding *encodings = track->encodings.elem;
1566 uint8_t *data = *buf;
1567 int isize = *buf_size;
1568 uint8_t *pkt_data = NULL;
1569 uint8_t av_unused *newpktdata;
1570 int pkt_size = isize;
1574 if (pkt_size >= 10000000U)
1575 return AVERROR_INVALIDDATA;
1577 switch (encodings[0].compression.algo) {
1578 case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
1580 int header_size = encodings[0].compression.settings.size;
1581 uint8_t *header = encodings[0].compression.settings.data;
1583 if (header_size && !header) {
1584 av_log(NULL, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1591 pkt_size = isize + header_size;
1592 pkt_data = av_malloc(pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1594 return AVERROR(ENOMEM);
1596 memcpy(pkt_data, header, header_size);
1597 memcpy(pkt_data + header_size, data, isize);
1601 case MATROSKA_TRACK_ENCODING_COMP_LZO:
1603 olen = pkt_size *= 3;
1604 newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING
1605 + AV_INPUT_BUFFER_PADDING_SIZE);
1607 result = AVERROR(ENOMEM);
1610 pkt_data = newpktdata;
1611 result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
1612 } while (result == AV_LZO_OUTPUT_FULL && pkt_size < 10000000);
1614 result = AVERROR_INVALIDDATA;
1621 case MATROSKA_TRACK_ENCODING_COMP_ZLIB:
1623 z_stream zstream = { 0 };
1624 if (inflateInit(&zstream) != Z_OK)
1626 zstream.next_in = data;
1627 zstream.avail_in = isize;
1630 newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1632 inflateEnd(&zstream);
1633 result = AVERROR(ENOMEM);
1636 pkt_data = newpktdata;
1637 zstream.avail_out = pkt_size - zstream.total_out;
1638 zstream.next_out = pkt_data + zstream.total_out;
1639 result = inflate(&zstream, Z_NO_FLUSH);
1640 } while (result == Z_OK && pkt_size < 10000000);
1641 pkt_size = zstream.total_out;
1642 inflateEnd(&zstream);
1643 if (result != Z_STREAM_END) {
1644 if (result == Z_MEM_ERROR)
1645 result = AVERROR(ENOMEM);
1647 result = AVERROR_INVALIDDATA;
1654 case MATROSKA_TRACK_ENCODING_COMP_BZLIB:
1656 bz_stream bzstream = { 0 };
1657 if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1659 bzstream.next_in = data;
1660 bzstream.avail_in = isize;
1663 newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1665 BZ2_bzDecompressEnd(&bzstream);
1666 result = AVERROR(ENOMEM);
1669 pkt_data = newpktdata;
1670 bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1671 bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1672 result = BZ2_bzDecompress(&bzstream);
1673 } while (result == BZ_OK && pkt_size < 10000000);
1674 pkt_size = bzstream.total_out_lo32;
1675 BZ2_bzDecompressEnd(&bzstream);
1676 if (result != BZ_STREAM_END) {
1677 if (result == BZ_MEM_ERROR)
1678 result = AVERROR(ENOMEM);
1680 result = AVERROR_INVALIDDATA;
1687 return AVERROR_INVALIDDATA;
1690 memset(pkt_data + pkt_size, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1693 *buf_size = pkt_size;
1701 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1702 AVDictionary **metadata, char *prefix)
1704 MatroskaTag *tags = list->elem;
1708 for (i = 0; i < list->nb_elem; i++) {
1709 const char *lang = tags[i].lang &&
1710 strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1712 if (!tags[i].name) {
1713 av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1717 snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1719 av_strlcpy(key, tags[i].name, sizeof(key));
1720 if (tags[i].def || !lang) {
1721 av_dict_set(metadata, key, tags[i].string, 0);
1722 if (tags[i].sub.nb_elem)
1723 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1726 av_strlcat(key, "-", sizeof(key));
1727 av_strlcat(key, lang, sizeof(key));
1728 av_dict_set(metadata, key, tags[i].string, 0);
1729 if (tags[i].sub.nb_elem)
1730 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1733 ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1736 static void matroska_convert_tags(AVFormatContext *s)
1738 MatroskaDemuxContext *matroska = s->priv_data;
1739 MatroskaTags *tags = matroska->tags.elem;
1742 for (i = 0; i < matroska->tags.nb_elem; i++) {
1743 if (tags[i].target.attachuid) {
1744 MatroskaAttachment *attachment = matroska->attachments.elem;
1746 for (j = 0; j < matroska->attachments.nb_elem; j++) {
1747 if (attachment[j].uid == tags[i].target.attachuid &&
1748 attachment[j].stream) {
1749 matroska_convert_tag(s, &tags[i].tag,
1750 &attachment[j].stream->metadata, NULL);
1755 av_log(NULL, AV_LOG_WARNING,
1756 "The tags at index %d refer to a "
1757 "non-existent attachment %"PRId64".\n",
1758 i, tags[i].target.attachuid);
1760 } else if (tags[i].target.chapteruid) {
1761 MatroskaChapter *chapter = matroska->chapters.elem;
1763 for (j = 0; j < matroska->chapters.nb_elem; j++) {
1764 if (chapter[j].uid == tags[i].target.chapteruid &&
1765 chapter[j].chapter) {
1766 matroska_convert_tag(s, &tags[i].tag,
1767 &chapter[j].chapter->metadata, NULL);
1772 av_log(NULL, AV_LOG_WARNING,
1773 "The tags at index %d refer to a non-existent chapter "
1775 i, tags[i].target.chapteruid);
1777 } else if (tags[i].target.trackuid) {
1778 MatroskaTrack *track = matroska->tracks.elem;
1780 for (j = 0; j < matroska->tracks.nb_elem; j++) {
1781 if (track[j].uid == tags[i].target.trackuid &&
1783 matroska_convert_tag(s, &tags[i].tag,
1784 &track[j].stream->metadata, NULL);
1789 av_log(NULL, AV_LOG_WARNING,
1790 "The tags at index %d refer to a non-existent track "
1792 i, tags[i].target.trackuid);
1795 matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1796 tags[i].target.type);
1801 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska,
1804 uint32_t saved_id = matroska->current_id;
1805 int64_t before_pos = avio_tell(matroska->ctx->pb);
1810 offset = pos + matroska->segment_start;
1811 if (avio_seek(matroska->ctx->pb, offset, SEEK_SET) == offset) {
1812 /* We don't want to lose our seekhead level, so we add
1813 * a dummy. This is a crude hack. */
1814 if (matroska->num_levels == EBML_MAX_DEPTH) {
1815 av_log(matroska->ctx, AV_LOG_INFO,
1816 "Max EBML element depth (%d) reached, "
1817 "cannot parse further.\n", EBML_MAX_DEPTH);
1818 ret = AVERROR_INVALIDDATA;
1820 matroska->levels[matroska->num_levels] = (MatroskaLevel) { 0, EBML_UNKNOWN_LENGTH };
1821 matroska->num_levels++;
1822 matroska->current_id = 0;
1824 ret = ebml_parse(matroska, matroska_segment, matroska);
1825 if (ret == LEVEL_ENDED) {
1826 /* This can only happen if the seek brought us beyond EOF. */
1831 /* Seek back - notice that in all instances where this is used
1832 * it is safe to set the level to 1. */
1833 matroska_reset_status(matroska, saved_id, before_pos);
1838 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1840 EbmlList *seekhead_list = &matroska->seekhead;
1843 // we should not do any seeking in the streaming case
1844 if (!(matroska->ctx->pb->seekable & AVIO_SEEKABLE_NORMAL))
1847 for (i = 0; i < seekhead_list->nb_elem; i++) {
1848 MatroskaSeekhead *seekheads = seekhead_list->elem;
1849 uint32_t id = seekheads[i].id;
1850 uint64_t pos = seekheads[i].pos;
1852 MatroskaLevel1Element *elem = matroska_find_level1_elem(matroska, id);
1853 if (!elem || elem->parsed)
1858 // defer cues parsing until we actually need cue data.
1859 if (id == MATROSKA_ID_CUES)
1862 if (matroska_parse_seekhead_entry(matroska, pos) < 0) {
1863 // mark index as broken
1864 matroska->cues_parsing_deferred = -1;
1872 static void matroska_add_index_entries(MatroskaDemuxContext *matroska)
1874 EbmlList *index_list;
1875 MatroskaIndex *index;
1876 uint64_t index_scale = 1;
1879 if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
1882 index_list = &matroska->index;
1883 index = index_list->elem;
1884 if (index_list->nb_elem < 2)
1886 if (index[1].time > 1E14 / matroska->time_scale) {
1887 av_log(matroska->ctx, AV_LOG_WARNING, "Dropping apparently-broken index.\n");
1890 for (i = 0; i < index_list->nb_elem; i++) {
1891 EbmlList *pos_list = &index[i].pos;
1892 MatroskaIndexPos *pos = pos_list->elem;
1893 for (j = 0; j < pos_list->nb_elem; j++) {
1894 MatroskaTrack *track = matroska_find_track_by_num(matroska,
1896 if (track && track->stream)
1897 av_add_index_entry(track->stream,
1898 pos[j].pos + matroska->segment_start,
1899 index[i].time / index_scale, 0, 0,
1905 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1908 if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
1911 for (i = 0; i < matroska->num_level1_elems; i++) {
1912 MatroskaLevel1Element *elem = &matroska->level1_elems[i];
1913 if (elem->id == MATROSKA_ID_CUES && !elem->parsed) {
1914 if (matroska_parse_seekhead_entry(matroska, elem->pos) < 0)
1915 matroska->cues_parsing_deferred = -1;
1921 matroska_add_index_entries(matroska);
1924 static int matroska_aac_profile(char *codec_id)
1926 static const char *const aac_profiles[] = { "MAIN", "LC", "SSR" };
1929 for (profile = 0; profile < FF_ARRAY_ELEMS(aac_profiles); profile++)
1930 if (strstr(codec_id, aac_profiles[profile]))
1935 static int matroska_aac_sri(int samplerate)
1939 for (sri = 0; sri < FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1940 if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
1945 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
1947 /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
1948 avpriv_dict_set_timestamp(metadata, "creation_time", date_utc / 1000 + 978307200000000LL);
1951 static int matroska_parse_flac(AVFormatContext *s,
1952 MatroskaTrack *track,
1955 AVStream *st = track->stream;
1956 uint8_t *p = track->codec_priv.data;
1957 int size = track->codec_priv.size;
1959 if (size < 8 + FLAC_STREAMINFO_SIZE || p[4] & 0x7f) {
1960 av_log(s, AV_LOG_WARNING, "Invalid FLAC private data\n");
1961 track->codec_priv.size = 0;
1965 track->codec_priv.size = 8 + FLAC_STREAMINFO_SIZE;
1967 p += track->codec_priv.size;
1968 size -= track->codec_priv.size;
1970 /* parse the remaining metadata blocks if present */
1972 int block_last, block_type, block_size;
1974 flac_parse_block_header(p, &block_last, &block_type, &block_size);
1978 if (block_size > size)
1981 /* check for the channel mask */
1982 if (block_type == FLAC_METADATA_TYPE_VORBIS_COMMENT) {
1983 AVDictionary *dict = NULL;
1984 AVDictionaryEntry *chmask;
1986 ff_vorbis_comment(s, &dict, p, block_size, 0);
1987 chmask = av_dict_get(dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", NULL, 0);
1989 uint64_t mask = strtol(chmask->value, NULL, 0);
1990 if (!mask || mask & ~0x3ffffULL) {
1991 av_log(s, AV_LOG_WARNING,
1992 "Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK\n");
1994 st->codecpar->channel_layout = mask;
1996 av_dict_free(&dict);
2006 static int mkv_field_order(MatroskaDemuxContext *matroska, int64_t field_order)
2008 int major, minor, micro, bttb = 0;
2010 /* workaround a bug in our Matroska muxer, introduced in version 57.36 alongside
2011 * this function, and fixed in 57.52 */
2012 if (matroska->muxingapp && sscanf(matroska->muxingapp, "Lavf%d.%d.%d", &major, &minor, µ) == 3)
2013 bttb = (major == 57 && minor >= 36 && minor <= 51 && micro >= 100);
2015 switch (field_order) {
2016 case MATROSKA_VIDEO_FIELDORDER_PROGRESSIVE:
2017 return AV_FIELD_PROGRESSIVE;
2018 case MATROSKA_VIDEO_FIELDORDER_UNDETERMINED:
2019 return AV_FIELD_UNKNOWN;
2020 case MATROSKA_VIDEO_FIELDORDER_TT:
2022 case MATROSKA_VIDEO_FIELDORDER_BB:
2024 case MATROSKA_VIDEO_FIELDORDER_BT:
2025 return bttb ? AV_FIELD_TB : AV_FIELD_BT;
2026 case MATROSKA_VIDEO_FIELDORDER_TB:
2027 return bttb ? AV_FIELD_BT : AV_FIELD_TB;
2029 return AV_FIELD_UNKNOWN;
2033 static void mkv_stereo_mode_display_mul(int stereo_mode,
2034 int *h_width, int *h_height)
2036 switch (stereo_mode) {
2037 case MATROSKA_VIDEO_STEREOMODE_TYPE_MONO:
2038 case MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_RL:
2039 case MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_LR:
2040 case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_RL:
2041 case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_LR:
2043 case MATROSKA_VIDEO_STEREOMODE_TYPE_RIGHT_LEFT:
2044 case MATROSKA_VIDEO_STEREOMODE_TYPE_LEFT_RIGHT:
2045 case MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_RL:
2046 case MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_LR:
2049 case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTTOM_TOP:
2050 case MATROSKA_VIDEO_STEREOMODE_TYPE_TOP_BOTTOM:
2051 case MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_RL:
2052 case MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_LR:
2058 static int mkv_parse_video_color(AVStream *st, const MatroskaTrack *track) {
2059 const MatroskaTrackVideoColor *color = track->video.color.elem;
2060 const MatroskaMasteringMeta *mastering_meta;
2061 int has_mastering_primaries, has_mastering_luminance;
2063 if (!track->video.color.nb_elem)
2066 mastering_meta = &color->mastering_meta;
2067 // Mastering primaries are CIE 1931 coords, and must be > 0.
2068 has_mastering_primaries =
2069 mastering_meta->r_x > 0 && mastering_meta->r_y > 0 &&
2070 mastering_meta->g_x > 0 && mastering_meta->g_y > 0 &&
2071 mastering_meta->b_x > 0 && mastering_meta->b_y > 0 &&
2072 mastering_meta->white_x > 0 && mastering_meta->white_y > 0;
2073 has_mastering_luminance = mastering_meta->max_luminance > 0;
2075 if (color->matrix_coefficients != AVCOL_SPC_RESERVED)
2076 st->codecpar->color_space = color->matrix_coefficients;
2077 if (color->primaries != AVCOL_PRI_RESERVED &&
2078 color->primaries != AVCOL_PRI_RESERVED0)
2079 st->codecpar->color_primaries = color->primaries;
2080 if (color->transfer_characteristics != AVCOL_TRC_RESERVED &&
2081 color->transfer_characteristics != AVCOL_TRC_RESERVED0)
2082 st->codecpar->color_trc = color->transfer_characteristics;
2083 if (color->range != AVCOL_RANGE_UNSPECIFIED &&
2084 color->range <= AVCOL_RANGE_JPEG)
2085 st->codecpar->color_range = color->range;
2086 if (color->chroma_siting_horz != MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED &&
2087 color->chroma_siting_vert != MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED &&
2088 color->chroma_siting_horz < MATROSKA_COLOUR_CHROMASITINGHORZ_NB &&
2089 color->chroma_siting_vert < MATROSKA_COLOUR_CHROMASITINGVERT_NB) {
2090 st->codecpar->chroma_location =
2091 avcodec_chroma_pos_to_enum((color->chroma_siting_horz - 1) << 7,
2092 (color->chroma_siting_vert - 1) << 7);
2094 if (color->max_cll && color->max_fall) {
2097 AVContentLightMetadata *metadata = av_content_light_metadata_alloc(&size);
2099 return AVERROR(ENOMEM);
2100 ret = av_stream_add_side_data(st, AV_PKT_DATA_CONTENT_LIGHT_LEVEL,
2101 (uint8_t *)metadata, size);
2103 av_freep(&metadata);
2106 metadata->MaxCLL = color->max_cll;
2107 metadata->MaxFALL = color->max_fall;
2110 if (has_mastering_primaries || has_mastering_luminance) {
2111 // Use similar rationals as other standards.
2112 const int chroma_den = 50000;
2113 const int luma_den = 10000;
2114 AVMasteringDisplayMetadata *metadata =
2115 (AVMasteringDisplayMetadata*) av_stream_new_side_data(
2116 st, AV_PKT_DATA_MASTERING_DISPLAY_METADATA,
2117 sizeof(AVMasteringDisplayMetadata));
2119 return AVERROR(ENOMEM);
2121 memset(metadata, 0, sizeof(AVMasteringDisplayMetadata));
2122 if (has_mastering_primaries) {
2123 metadata->display_primaries[0][0] = av_make_q(
2124 round(mastering_meta->r_x * chroma_den), chroma_den);
2125 metadata->display_primaries[0][1] = av_make_q(
2126 round(mastering_meta->r_y * chroma_den), chroma_den);
2127 metadata->display_primaries[1][0] = av_make_q(
2128 round(mastering_meta->g_x * chroma_den), chroma_den);
2129 metadata->display_primaries[1][1] = av_make_q(
2130 round(mastering_meta->g_y * chroma_den), chroma_den);
2131 metadata->display_primaries[2][0] = av_make_q(
2132 round(mastering_meta->b_x * chroma_den), chroma_den);
2133 metadata->display_primaries[2][1] = av_make_q(
2134 round(mastering_meta->b_y * chroma_den), chroma_den);
2135 metadata->white_point[0] = av_make_q(
2136 round(mastering_meta->white_x * chroma_den), chroma_den);
2137 metadata->white_point[1] = av_make_q(
2138 round(mastering_meta->white_y * chroma_den), chroma_den);
2139 metadata->has_primaries = 1;
2141 if (has_mastering_luminance) {
2142 metadata->max_luminance = av_make_q(
2143 round(mastering_meta->max_luminance * luma_den), luma_den);
2144 metadata->min_luminance = av_make_q(
2145 round(mastering_meta->min_luminance * luma_den), luma_den);
2146 metadata->has_luminance = 1;
2152 static int mkv_parse_video_projection(AVStream *st, const MatroskaTrack *track) {
2153 AVSphericalMapping *spherical;
2154 enum AVSphericalProjection projection;
2155 size_t spherical_size;
2156 uint32_t l = 0, t = 0, r = 0, b = 0;
2157 uint32_t padding = 0;
2161 bytestream2_init(&gb, track->video.projection.private.data,
2162 track->video.projection.private.size);
2164 if (bytestream2_get_byte(&gb) != 0) {
2165 av_log(NULL, AV_LOG_WARNING, "Unknown spherical metadata\n");
2169 bytestream2_skip(&gb, 3); // flags
2171 switch (track->video.projection.type) {
2172 case MATROSKA_VIDEO_PROJECTION_TYPE_EQUIRECTANGULAR:
2173 if (track->video.projection.private.size == 20) {
2174 t = bytestream2_get_be32(&gb);
2175 b = bytestream2_get_be32(&gb);
2176 l = bytestream2_get_be32(&gb);
2177 r = bytestream2_get_be32(&gb);
2179 if (b >= UINT_MAX - t || r >= UINT_MAX - l) {
2180 av_log(NULL, AV_LOG_ERROR,
2181 "Invalid bounding rectangle coordinates "
2182 "%"PRIu32",%"PRIu32",%"PRIu32",%"PRIu32"\n",
2184 return AVERROR_INVALIDDATA;
2186 } else if (track->video.projection.private.size != 0) {
2187 av_log(NULL, AV_LOG_ERROR, "Unknown spherical metadata\n");
2188 return AVERROR_INVALIDDATA;
2191 if (l || t || r || b)
2192 projection = AV_SPHERICAL_EQUIRECTANGULAR_TILE;
2194 projection = AV_SPHERICAL_EQUIRECTANGULAR;
2196 case MATROSKA_VIDEO_PROJECTION_TYPE_CUBEMAP:
2197 if (track->video.projection.private.size < 4) {
2198 av_log(NULL, AV_LOG_ERROR, "Missing projection private properties\n");
2199 return AVERROR_INVALIDDATA;
2200 } else if (track->video.projection.private.size == 12) {
2201 uint32_t layout = bytestream2_get_be32(&gb);
2203 av_log(NULL, AV_LOG_WARNING,
2204 "Unknown spherical cubemap layout %"PRIu32"\n", layout);
2207 projection = AV_SPHERICAL_CUBEMAP;
2208 padding = bytestream2_get_be32(&gb);
2210 av_log(NULL, AV_LOG_ERROR, "Unknown spherical metadata\n");
2211 return AVERROR_INVALIDDATA;
2214 case MATROSKA_VIDEO_PROJECTION_TYPE_RECTANGULAR:
2215 /* No Spherical metadata */
2218 av_log(NULL, AV_LOG_WARNING,
2219 "Unknown spherical metadata type %"PRIu64"\n",
2220 track->video.projection.type);
2224 spherical = av_spherical_alloc(&spherical_size);
2226 return AVERROR(ENOMEM);
2228 spherical->projection = projection;
2230 spherical->yaw = (int32_t) (track->video.projection.yaw * (1 << 16));
2231 spherical->pitch = (int32_t) (track->video.projection.pitch * (1 << 16));
2232 spherical->roll = (int32_t) (track->video.projection.roll * (1 << 16));
2234 spherical->padding = padding;
2236 spherical->bound_left = l;
2237 spherical->bound_top = t;
2238 spherical->bound_right = r;
2239 spherical->bound_bottom = b;
2241 ret = av_stream_add_side_data(st, AV_PKT_DATA_SPHERICAL, (uint8_t *)spherical,
2244 av_freep(&spherical);
2251 static int get_qt_codec(MatroskaTrack *track, uint32_t *fourcc, enum AVCodecID *codec_id)
2253 const AVCodecTag *codec_tags;
2255 codec_tags = track->type == MATROSKA_TRACK_TYPE_VIDEO ?
2256 ff_codec_movvideo_tags : ff_codec_movaudio_tags;
2258 /* Normalize noncompliant private data that starts with the fourcc
2259 * by expanding/shifting the data by 4 bytes and storing the data
2260 * size at the start. */
2261 if (ff_codec_get_id(codec_tags, AV_RL32(track->codec_priv.data))) {
2262 int ret = av_buffer_realloc(&track->codec_priv.buf,
2263 track->codec_priv.size + 4 + AV_INPUT_BUFFER_PADDING_SIZE);
2267 track->codec_priv.data = track->codec_priv.buf->data;
2268 memmove(track->codec_priv.data + 4, track->codec_priv.data, track->codec_priv.size);
2269 track->codec_priv.size += 4;
2270 AV_WB32(track->codec_priv.data, track->codec_priv.size);
2273 *fourcc = AV_RL32(track->codec_priv.data + 4);
2274 *codec_id = ff_codec_get_id(codec_tags, *fourcc);
2279 static int matroska_parse_tracks(AVFormatContext *s)
2281 MatroskaDemuxContext *matroska = s->priv_data;
2282 MatroskaTrack *tracks = matroska->tracks.elem;
2287 for (i = 0; i < matroska->tracks.nb_elem; i++) {
2288 MatroskaTrack *track = &tracks[i];
2289 enum AVCodecID codec_id = AV_CODEC_ID_NONE;
2290 EbmlList *encodings_list = &track->encodings;
2291 MatroskaTrackEncoding *encodings = encodings_list->elem;
2292 uint8_t *extradata = NULL;
2293 int extradata_size = 0;
2294 int extradata_offset = 0;
2295 uint32_t fourcc = 0;
2297 char* key_id_base64 = NULL;
2300 /* Apply some sanity checks. */
2301 if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
2302 track->type != MATROSKA_TRACK_TYPE_AUDIO &&
2303 track->type != MATROSKA_TRACK_TYPE_SUBTITLE &&
2304 track->type != MATROSKA_TRACK_TYPE_METADATA) {
2305 av_log(matroska->ctx, AV_LOG_INFO,
2306 "Unknown or unsupported track type %"PRIu64"\n",
2310 if (!track->codec_id)
2313 if (track->audio.samplerate < 0 || track->audio.samplerate > INT_MAX ||
2314 isnan(track->audio.samplerate)) {
2315 av_log(matroska->ctx, AV_LOG_WARNING,
2316 "Invalid sample rate %f, defaulting to 8000 instead.\n",
2317 track->audio.samplerate);
2318 track->audio.samplerate = 8000;
2321 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2322 if (!track->default_duration && track->video.frame_rate > 0) {
2323 double default_duration = 1000000000 / track->video.frame_rate;
2324 if (default_duration > UINT64_MAX || default_duration < 0) {
2325 av_log(matroska->ctx, AV_LOG_WARNING,
2326 "Invalid frame rate %e. Cannot calculate default duration.\n",
2327 track->video.frame_rate);
2329 track->default_duration = default_duration;
2332 if (track->video.display_width == -1)
2333 track->video.display_width = track->video.pixel_width;
2334 if (track->video.display_height == -1)
2335 track->video.display_height = track->video.pixel_height;
2336 if (track->video.color_space.size == 4)
2337 fourcc = AV_RL32(track->video.color_space.data);
2338 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2339 if (!track->audio.out_samplerate)
2340 track->audio.out_samplerate = track->audio.samplerate;
2342 if (encodings_list->nb_elem > 1) {
2343 av_log(matroska->ctx, AV_LOG_ERROR,
2344 "Multiple combined encodings not supported");
2345 } else if (encodings_list->nb_elem == 1) {
2346 if (encodings[0].type) {
2347 if (encodings[0].encryption.key_id.size > 0) {
2348 /* Save the encryption key id to be stored later as a
2350 const int b64_size = AV_BASE64_SIZE(encodings[0].encryption.key_id.size);
2351 key_id_base64 = av_malloc(b64_size);
2352 if (key_id_base64 == NULL)
2353 return AVERROR(ENOMEM);
2355 av_base64_encode(key_id_base64, b64_size,
2356 encodings[0].encryption.key_id.data,
2357 encodings[0].encryption.key_id.size);
2359 encodings[0].scope = 0;
2360 av_log(matroska->ctx, AV_LOG_ERROR,
2361 "Unsupported encoding type");
2365 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
2368 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
2371 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO &&
2373 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP) {
2374 encodings[0].scope = 0;
2375 av_log(matroska->ctx, AV_LOG_ERROR,
2376 "Unsupported encoding type");
2377 } else if (track->codec_priv.size && encodings[0].scope & 2) {
2378 uint8_t *codec_priv = track->codec_priv.data;
2379 int ret = matroska_decode_buffer(&track->codec_priv.data,
2380 &track->codec_priv.size,
2383 track->codec_priv.data = NULL;
2384 track->codec_priv.size = 0;
2385 av_log(matroska->ctx, AV_LOG_ERROR,
2386 "Failed to decode codec private data\n");
2389 if (codec_priv != track->codec_priv.data) {
2390 av_buffer_unref(&track->codec_priv.buf);
2391 if (track->codec_priv.data) {
2392 track->codec_priv.buf = av_buffer_create(track->codec_priv.data,
2393 track->codec_priv.size + AV_INPUT_BUFFER_PADDING_SIZE,
2395 if (!track->codec_priv.buf) {
2396 av_freep(&track->codec_priv.data);
2397 track->codec_priv.size = 0;
2398 return AVERROR(ENOMEM);
2405 for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
2406 if (!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
2407 strlen(ff_mkv_codec_tags[j].str))) {
2408 codec_id = ff_mkv_codec_tags[j].id;
2413 st = track->stream = avformat_new_stream(s, NULL);
2415 av_free(key_id_base64);
2416 return AVERROR(ENOMEM);
2419 if (key_id_base64) {
2420 /* export encryption key id as base64 metadata tag */
2421 av_dict_set(&st->metadata, "enc_key_id", key_id_base64, 0);
2422 av_freep(&key_id_base64);
2425 if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC") &&
2426 track->codec_priv.size >= 40 &&
2427 track->codec_priv.data) {
2428 track->ms_compat = 1;
2429 bit_depth = AV_RL16(track->codec_priv.data + 14);
2430 fourcc = AV_RL32(track->codec_priv.data + 16);
2431 codec_id = ff_codec_get_id(ff_codec_bmp_tags,
2434 codec_id = ff_codec_get_id(ff_codec_movvideo_tags,
2436 extradata_offset = 40;
2437 } else if (!strcmp(track->codec_id, "A_MS/ACM") &&
2438 track->codec_priv.size >= 14 &&
2439 track->codec_priv.data) {
2441 ffio_init_context(&b, track->codec_priv.data,
2442 track->codec_priv.size,
2443 0, NULL, NULL, NULL, NULL);
2444 ret = ff_get_wav_header(s, &b, st->codecpar, track->codec_priv.size, 0);
2447 codec_id = st->codecpar->codec_id;
2448 fourcc = st->codecpar->codec_tag;
2449 extradata_offset = FFMIN(track->codec_priv.size, 18);
2450 } else if (!strcmp(track->codec_id, "A_QUICKTIME")
2451 /* Normally 36, but allow noncompliant private data */
2452 && (track->codec_priv.size >= 32)
2453 && (track->codec_priv.data)) {
2454 uint16_t sample_size;
2455 int ret = get_qt_codec(track, &fourcc, &codec_id);
2458 sample_size = AV_RB16(track->codec_priv.data + 26);
2460 if (sample_size == 8) {
2461 fourcc = MKTAG('r','a','w',' ');
2462 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
2463 } else if (sample_size == 16) {
2464 fourcc = MKTAG('t','w','o','s');
2465 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
2468 if ((fourcc == MKTAG('t','w','o','s') ||
2469 fourcc == MKTAG('s','o','w','t')) &&
2471 codec_id = AV_CODEC_ID_PCM_S8;
2472 } else if (!strcmp(track->codec_id, "V_QUICKTIME") &&
2473 (track->codec_priv.size >= 21) &&
2474 (track->codec_priv.data)) {
2475 int ret = get_qt_codec(track, &fourcc, &codec_id);
2478 if (codec_id == AV_CODEC_ID_NONE && AV_RL32(track->codec_priv.data+4) == AV_RL32("SMI ")) {
2479 fourcc = MKTAG('S','V','Q','3');
2480 codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
2482 if (codec_id == AV_CODEC_ID_NONE)
2483 av_log(matroska->ctx, AV_LOG_ERROR,
2484 "mov FourCC not found %s.\n", av_fourcc2str(fourcc));
2485 if (track->codec_priv.size >= 86) {
2486 bit_depth = AV_RB16(track->codec_priv.data + 82);
2487 ffio_init_context(&b, track->codec_priv.data,
2488 track->codec_priv.size,
2489 0, NULL, NULL, NULL, NULL);
2490 if (ff_get_qtpalette(codec_id, &b, track->palette)) {
2492 track->has_palette = 1;
2495 } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
2496 switch (track->audio.bitdepth) {
2498 codec_id = AV_CODEC_ID_PCM_U8;
2501 codec_id = AV_CODEC_ID_PCM_S24BE;
2504 codec_id = AV_CODEC_ID_PCM_S32BE;
2507 } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
2508 switch (track->audio.bitdepth) {
2510 codec_id = AV_CODEC_ID_PCM_U8;
2513 codec_id = AV_CODEC_ID_PCM_S24LE;
2516 codec_id = AV_CODEC_ID_PCM_S32LE;
2519 } else if (codec_id == AV_CODEC_ID_PCM_F32LE &&
2520 track->audio.bitdepth == 64) {
2521 codec_id = AV_CODEC_ID_PCM_F64LE;
2522 } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
2523 int profile = matroska_aac_profile(track->codec_id);
2524 int sri = matroska_aac_sri(track->audio.samplerate);
2525 extradata = av_mallocz(5 + AV_INPUT_BUFFER_PADDING_SIZE);
2527 return AVERROR(ENOMEM);
2528 extradata[0] = (profile << 3) | ((sri & 0x0E) >> 1);
2529 extradata[1] = ((sri & 0x01) << 7) | (track->audio.channels << 3);
2530 if (strstr(track->codec_id, "SBR")) {
2531 sri = matroska_aac_sri(track->audio.out_samplerate);
2532 extradata[2] = 0x56;
2533 extradata[3] = 0xE5;
2534 extradata[4] = 0x80 | (sri << 3);
2538 } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX - 12 - AV_INPUT_BUFFER_PADDING_SIZE) {
2539 /* Only ALAC's magic cookie is stored in Matroska's track headers.
2540 * Create the "atom size", "tag", and "tag version" fields the
2541 * decoder expects manually. */
2542 extradata_size = 12 + track->codec_priv.size;
2543 extradata = av_mallocz(extradata_size +
2544 AV_INPUT_BUFFER_PADDING_SIZE);
2546 return AVERROR(ENOMEM);
2547 AV_WB32(extradata, extradata_size);
2548 memcpy(&extradata[4], "alac", 4);
2549 AV_WB32(&extradata[8], 0);
2550 memcpy(&extradata[12], track->codec_priv.data,
2551 track->codec_priv.size);
2552 } else if (codec_id == AV_CODEC_ID_TTA) {
2553 extradata_size = 30;
2554 extradata = av_mallocz(extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
2556 return AVERROR(ENOMEM);
2557 ffio_init_context(&b, extradata, extradata_size, 1,
2558 NULL, NULL, NULL, NULL);
2559 avio_write(&b, "TTA1", 4);
2561 if (track->audio.channels > UINT16_MAX ||
2562 track->audio.bitdepth > UINT16_MAX) {
2563 av_log(matroska->ctx, AV_LOG_WARNING,
2564 "Too large audio channel number %"PRIu64
2565 " or bitdepth %"PRIu64". Skipping track.\n",
2566 track->audio.channels, track->audio.bitdepth);
2567 av_freep(&extradata);
2568 if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
2569 return AVERROR_INVALIDDATA;
2573 avio_wl16(&b, track->audio.channels);
2574 avio_wl16(&b, track->audio.bitdepth);
2575 if (track->audio.out_samplerate < 0 || track->audio.out_samplerate > INT_MAX)
2576 return AVERROR_INVALIDDATA;
2577 avio_wl32(&b, track->audio.out_samplerate);
2578 avio_wl32(&b, av_rescale((matroska->duration * matroska->time_scale),
2579 track->audio.out_samplerate,
2580 AV_TIME_BASE * 1000));
2581 } else if (codec_id == AV_CODEC_ID_RV10 ||
2582 codec_id == AV_CODEC_ID_RV20 ||
2583 codec_id == AV_CODEC_ID_RV30 ||
2584 codec_id == AV_CODEC_ID_RV40) {
2585 extradata_offset = 26;
2586 } else if (codec_id == AV_CODEC_ID_RA_144) {
2587 track->audio.out_samplerate = 8000;
2588 track->audio.channels = 1;
2589 } else if ((codec_id == AV_CODEC_ID_RA_288 ||
2590 codec_id == AV_CODEC_ID_COOK ||
2591 codec_id == AV_CODEC_ID_ATRAC3 ||
2592 codec_id == AV_CODEC_ID_SIPR)
2593 && track->codec_priv.data) {
2596 ffio_init_context(&b, track->codec_priv.data,
2597 track->codec_priv.size,
2598 0, NULL, NULL, NULL, NULL);
2600 flavor = avio_rb16(&b);
2601 track->audio.coded_framesize = avio_rb32(&b);
2603 track->audio.sub_packet_h = avio_rb16(&b);
2604 track->audio.frame_size = avio_rb16(&b);
2605 track->audio.sub_packet_size = avio_rb16(&b);
2607 track->audio.coded_framesize <= 0 ||
2608 track->audio.sub_packet_h <= 0 ||
2609 track->audio.frame_size <= 0 ||
2610 track->audio.sub_packet_size <= 0 && codec_id != AV_CODEC_ID_SIPR)
2611 return AVERROR_INVALIDDATA;
2612 track->audio.buf = av_malloc_array(track->audio.sub_packet_h,
2613 track->audio.frame_size);
2614 if (!track->audio.buf)
2615 return AVERROR(ENOMEM);
2616 if (codec_id == AV_CODEC_ID_RA_288) {
2617 st->codecpar->block_align = track->audio.coded_framesize;
2618 track->codec_priv.size = 0;
2620 if (codec_id == AV_CODEC_ID_SIPR && flavor < 4) {
2621 static const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
2622 track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
2623 st->codecpar->bit_rate = sipr_bit_rate[flavor];
2625 st->codecpar->block_align = track->audio.sub_packet_size;
2626 extradata_offset = 78;
2628 } else if (codec_id == AV_CODEC_ID_FLAC && track->codec_priv.size) {
2629 ret = matroska_parse_flac(s, track, &extradata_offset);
2632 } else if (codec_id == AV_CODEC_ID_PRORES && track->codec_priv.size == 4) {
2633 fourcc = AV_RL32(track->codec_priv.data);
2634 } else if (codec_id == AV_CODEC_ID_VP9 && track->codec_priv.size) {
2635 /* we don't need any value stored in CodecPrivate.
2636 make sure that it's not exported as extradata. */
2637 track->codec_priv.size = 0;
2638 } else if (codec_id == AV_CODEC_ID_AV1 && track->codec_priv.size) {
2639 /* For now, propagate only the OBUs, if any. Once libavcodec is
2640 updated to handle isobmff style extradata this can be removed. */
2641 extradata_offset = 4;
2643 track->codec_priv.size -= extradata_offset;
2645 if (codec_id == AV_CODEC_ID_NONE)
2646 av_log(matroska->ctx, AV_LOG_INFO,
2647 "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
2649 if (track->time_scale < 0.01)
2650 track->time_scale = 1.0;
2651 avpriv_set_pts_info(st, 64, matroska->time_scale * track->time_scale,
2652 1000 * 1000 * 1000); /* 64 bit pts in ns */
2654 /* convert the delay from ns to the track timebase */
2655 track->codec_delay_in_track_tb = av_rescale_q(track->codec_delay,
2656 (AVRational){ 1, 1000000000 },
2659 st->codecpar->codec_id = codec_id;
2661 if (strcmp(track->language, "und"))
2662 av_dict_set(&st->metadata, "language", track->language, 0);
2663 av_dict_set(&st->metadata, "title", track->name, 0);
2665 if (track->flag_default)
2666 st->disposition |= AV_DISPOSITION_DEFAULT;
2667 if (track->flag_forced)
2668 st->disposition |= AV_DISPOSITION_FORCED;
2670 if (!st->codecpar->extradata) {
2672 st->codecpar->extradata = extradata;
2673 st->codecpar->extradata_size = extradata_size;
2674 } else if (track->codec_priv.data && track->codec_priv.size > 0) {
2675 if (ff_alloc_extradata(st->codecpar, track->codec_priv.size))
2676 return AVERROR(ENOMEM);
2677 memcpy(st->codecpar->extradata,
2678 track->codec_priv.data + extradata_offset,
2679 track->codec_priv.size);
2683 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2684 MatroskaTrackPlane *planes = track->operation.combine_planes.elem;
2685 int display_width_mul = 1;
2686 int display_height_mul = 1;
2688 st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
2689 st->codecpar->codec_tag = fourcc;
2691 st->codecpar->bits_per_coded_sample = bit_depth;
2692 st->codecpar->width = track->video.pixel_width;
2693 st->codecpar->height = track->video.pixel_height;
2695 if (track->video.interlaced == MATROSKA_VIDEO_INTERLACE_FLAG_INTERLACED)
2696 st->codecpar->field_order = mkv_field_order(matroska, track->video.field_order);
2697 else if (track->video.interlaced == MATROSKA_VIDEO_INTERLACE_FLAG_PROGRESSIVE)
2698 st->codecpar->field_order = AV_FIELD_PROGRESSIVE;
2700 if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
2701 mkv_stereo_mode_display_mul(track->video.stereo_mode, &display_width_mul, &display_height_mul);
2703 if (track->video.display_unit < MATROSKA_VIDEO_DISPLAYUNIT_UNKNOWN) {
2704 av_reduce(&st->sample_aspect_ratio.num,
2705 &st->sample_aspect_ratio.den,
2706 st->codecpar->height * track->video.display_width * display_width_mul,
2707 st->codecpar->width * track->video.display_height * display_height_mul,
2710 if (st->codecpar->codec_id != AV_CODEC_ID_HEVC)
2711 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2713 if (track->default_duration) {
2714 av_reduce(&st->avg_frame_rate.num, &st->avg_frame_rate.den,
2715 1000000000, track->default_duration, 30000);
2716 #if FF_API_R_FRAME_RATE
2717 if ( st->avg_frame_rate.num < st->avg_frame_rate.den * 1000LL
2718 && st->avg_frame_rate.num > st->avg_frame_rate.den * 5LL)
2719 st->r_frame_rate = st->avg_frame_rate;
2723 /* export stereo mode flag as metadata tag */
2724 if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
2725 av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
2727 /* export alpha mode flag as metadata tag */
2728 if (track->video.alpha_mode)
2729 av_dict_set(&st->metadata, "alpha_mode", "1", 0);
2731 /* if we have virtual track, mark the real tracks */
2732 for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
2734 if (planes[j].type >= MATROSKA_VIDEO_STEREO_PLANE_COUNT)
2736 snprintf(buf, sizeof(buf), "%s_%d",
2737 ff_matroska_video_stereo_plane[planes[j].type], i);
2738 for (k=0; k < matroska->tracks.nb_elem; k++)
2739 if (planes[j].uid == tracks[k].uid && tracks[k].stream) {
2740 av_dict_set(&tracks[k].stream->metadata,
2741 "stereo_mode", buf, 0);
2745 // add stream level stereo3d side data if it is a supported format
2746 if (track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB &&
2747 track->video.stereo_mode != 10 && track->video.stereo_mode != 12) {
2748 int ret = ff_mkv_stereo3d_conv(st, track->video.stereo_mode);
2753 ret = mkv_parse_video_color(st, track);
2756 ret = mkv_parse_video_projection(st, track);
2759 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2760 st->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
2761 st->codecpar->codec_tag = fourcc;
2762 st->codecpar->sample_rate = track->audio.out_samplerate;
2763 st->codecpar->channels = track->audio.channels;
2764 if (!st->codecpar->bits_per_coded_sample)
2765 st->codecpar->bits_per_coded_sample = track->audio.bitdepth;
2766 if (st->codecpar->codec_id == AV_CODEC_ID_MP3 ||
2767 st->codecpar->codec_id == AV_CODEC_ID_MLP ||
2768 st->codecpar->codec_id == AV_CODEC_ID_TRUEHD)
2769 st->need_parsing = AVSTREAM_PARSE_FULL;
2770 else if (st->codecpar->codec_id != AV_CODEC_ID_AAC)
2771 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2772 if (track->codec_delay > 0) {
2773 st->codecpar->initial_padding = av_rescale_q(track->codec_delay,
2774 (AVRational){1, 1000000000},
2775 (AVRational){1, st->codecpar->codec_id == AV_CODEC_ID_OPUS ?
2776 48000 : st->codecpar->sample_rate});
2778 if (track->seek_preroll > 0) {
2779 st->codecpar->seek_preroll = av_rescale_q(track->seek_preroll,
2780 (AVRational){1, 1000000000},
2781 (AVRational){1, st->codecpar->sample_rate});
2783 } else if (codec_id == AV_CODEC_ID_WEBVTT) {
2784 st->codecpar->codec_type = AVMEDIA_TYPE_SUBTITLE;
2786 if (!strcmp(track->codec_id, "D_WEBVTT/CAPTIONS")) {
2787 st->disposition |= AV_DISPOSITION_CAPTIONS;
2788 } else if (!strcmp(track->codec_id, "D_WEBVTT/DESCRIPTIONS")) {
2789 st->disposition |= AV_DISPOSITION_DESCRIPTIONS;
2790 } else if (!strcmp(track->codec_id, "D_WEBVTT/METADATA")) {
2791 st->disposition |= AV_DISPOSITION_METADATA;
2793 } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2794 st->codecpar->codec_type = AVMEDIA_TYPE_SUBTITLE;
2801 static int matroska_read_header(AVFormatContext *s)
2803 MatroskaDemuxContext *matroska = s->priv_data;
2804 EbmlList *attachments_list = &matroska->attachments;
2805 EbmlList *chapters_list = &matroska->chapters;
2806 MatroskaAttachment *attachments;
2807 MatroskaChapter *chapters;
2808 uint64_t max_start = 0;
2814 matroska->cues_parsing_deferred = 1;
2816 /* First read the EBML header. */
2817 if (ebml_parse(matroska, ebml_syntax, &ebml) || !ebml.doctype) {
2818 av_log(matroska->ctx, AV_LOG_ERROR, "EBML header parsing failed\n");
2819 ebml_free(ebml_syntax, &ebml);
2820 return AVERROR_INVALIDDATA;
2822 if (ebml.version > EBML_VERSION ||
2823 ebml.max_size > sizeof(uint64_t) ||
2824 ebml.id_length > sizeof(uint32_t) ||
2825 ebml.doctype_version > 3) {
2826 avpriv_report_missing_feature(matroska->ctx,
2827 "EBML version %"PRIu64", doctype %s, doc version %"PRIu64,
2828 ebml.version, ebml.doctype, ebml.doctype_version);
2829 ebml_free(ebml_syntax, &ebml);
2830 return AVERROR_PATCHWELCOME;
2831 } else if (ebml.doctype_version == 3) {
2832 av_log(matroska->ctx, AV_LOG_WARNING,
2833 "EBML header using unsupported features\n"
2834 "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2835 ebml.version, ebml.doctype, ebml.doctype_version);
2837 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
2838 if (!strcmp(ebml.doctype, matroska_doctypes[i]))
2840 if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
2841 av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
2842 if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
2843 ebml_free(ebml_syntax, &ebml);
2844 return AVERROR_INVALIDDATA;
2847 ebml_free(ebml_syntax, &ebml);
2849 /* The next thing is a segment. */
2850 pos = avio_tell(matroska->ctx->pb);
2851 res = ebml_parse(matroska, matroska_segments, matroska);
2852 // Try resyncing until we find an EBML_STOP type element.
2854 res = matroska_resync(matroska, pos);
2857 pos = avio_tell(matroska->ctx->pb);
2858 res = ebml_parse(matroska, matroska_segment, matroska);
2860 /* Set data_offset as it might be needed later by seek_frame_generic. */
2861 if (matroska->current_id == MATROSKA_ID_CLUSTER)
2862 s->internal->data_offset = avio_tell(matroska->ctx->pb) - 4;
2863 matroska_execute_seekhead(matroska);
2865 if (!matroska->time_scale)
2866 matroska->time_scale = 1000000;
2867 if (matroska->duration)
2868 matroska->ctx->duration = matroska->duration * matroska->time_scale *
2869 1000 / AV_TIME_BASE;
2870 av_dict_set(&s->metadata, "title", matroska->title, 0);
2871 av_dict_set(&s->metadata, "encoder", matroska->muxingapp, 0);
2873 if (matroska->date_utc.size == 8)
2874 matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
2876 res = matroska_parse_tracks(s);
2880 attachments = attachments_list->elem;
2881 for (j = 0; j < attachments_list->nb_elem; j++) {
2882 if (!(attachments[j].filename && attachments[j].mime &&
2883 attachments[j].bin.data && attachments[j].bin.size > 0)) {
2884 av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
2886 AVStream *st = avformat_new_stream(s, NULL);
2889 av_dict_set(&st->metadata, "filename", attachments[j].filename, 0);
2890 av_dict_set(&st->metadata, "mimetype", attachments[j].mime, 0);
2891 st->codecpar->codec_id = AV_CODEC_ID_NONE;
2893 for (i = 0; ff_mkv_image_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
2894 if (!strncmp(ff_mkv_image_mime_tags[i].str, attachments[j].mime,
2895 strlen(ff_mkv_image_mime_tags[i].str))) {
2896 st->codecpar->codec_id = ff_mkv_image_mime_tags[i].id;
2901 attachments[j].stream = st;
2903 if (st->codecpar->codec_id != AV_CODEC_ID_NONE) {
2904 AVPacket *pkt = &st->attached_pic;
2906 st->disposition |= AV_DISPOSITION_ATTACHED_PIC;
2907 st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
2909 av_init_packet(pkt);
2910 pkt->buf = av_buffer_ref(attachments[j].bin.buf);
2912 return AVERROR(ENOMEM);
2913 pkt->data = attachments[j].bin.data;
2914 pkt->size = attachments[j].bin.size;
2915 pkt->stream_index = st->index;
2916 pkt->flags |= AV_PKT_FLAG_KEY;
2918 st->codecpar->codec_type = AVMEDIA_TYPE_ATTACHMENT;
2919 if (ff_alloc_extradata(st->codecpar, attachments[j].bin.size))
2921 memcpy(st->codecpar->extradata, attachments[j].bin.data,
2922 attachments[j].bin.size);
2924 for (i = 0; ff_mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
2925 if (!strncmp(ff_mkv_mime_tags[i].str, attachments[j].mime,
2926 strlen(ff_mkv_mime_tags[i].str))) {
2927 st->codecpar->codec_id = ff_mkv_mime_tags[i].id;
2935 chapters = chapters_list->elem;
2936 for (i = 0; i < chapters_list->nb_elem; i++)
2937 if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid &&
2938 (max_start == 0 || chapters[i].start > max_start)) {
2939 chapters[i].chapter =
2940 avpriv_new_chapter(s, chapters[i].uid,
2941 (AVRational) { 1, 1000000000 },
2942 chapters[i].start, chapters[i].end,
2944 if (chapters[i].chapter) {
2945 av_dict_set(&chapters[i].chapter->metadata,
2946 "title", chapters[i].title, 0);
2948 max_start = chapters[i].start;
2951 matroska_add_index_entries(matroska);
2953 matroska_convert_tags(s);
2957 matroska_read_close(s);
2962 * Put one packet in an application-supplied AVPacket struct.
2963 * Returns 0 on success or -1 on failure.
2965 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
2968 if (matroska->queue) {
2969 MatroskaTrack *tracks = matroska->tracks.elem;
2970 MatroskaTrack *track;
2972 ff_packet_list_get(&matroska->queue, &matroska->queue_end, pkt);
2973 track = &tracks[pkt->stream_index];
2974 if (track->has_palette) {
2975 uint8_t *pal = av_packet_new_side_data(pkt, AV_PKT_DATA_PALETTE, AVPALETTE_SIZE);
2977 av_log(matroska->ctx, AV_LOG_ERROR, "Cannot append palette to packet\n");
2979 memcpy(pal, track->palette, AVPALETTE_SIZE);
2981 track->has_palette = 0;
2990 * Free all packets in our internal queue.
2992 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
2994 ff_packet_list_free(&matroska->queue, &matroska->queue_end);
2997 static int matroska_parse_laces(MatroskaDemuxContext *matroska, uint8_t **buf,
2998 int *buf_size, int type,
2999 uint32_t **lace_buf, int *laces)
3001 int res = 0, n, size = *buf_size;
3002 uint8_t *data = *buf;
3003 uint32_t *lace_size;
3007 *lace_buf = av_malloc(sizeof(**lace_buf));
3009 return AVERROR(ENOMEM);
3011 *lace_buf[0] = size;
3015 av_assert0(size > 0);
3019 lace_size = av_malloc_array(*laces, sizeof(*lace_size));
3021 return AVERROR(ENOMEM);
3024 case 0x1: /* Xiph lacing */
3028 for (n = 0; res == 0 && n < *laces - 1; n++) {
3032 if (size <= total) {
3033 res = AVERROR_INVALIDDATA;
3038 lace_size[n] += temp;
3045 if (size <= total) {
3046 res = AVERROR_INVALIDDATA;
3050 lace_size[n] = size - total;
3054 case 0x2: /* fixed-size lacing */
3055 if (size % (*laces)) {
3056 res = AVERROR_INVALIDDATA;
3059 for (n = 0; n < *laces; n++)
3060 lace_size[n] = size / *laces;
3063 case 0x3: /* EBML lacing */
3067 n = matroska_ebmlnum_uint(matroska, data, size, &num);
3068 if (n < 0 || num > INT_MAX) {
3069 av_log(matroska->ctx, AV_LOG_INFO,
3070 "EBML block data error\n");
3071 res = n<0 ? n : AVERROR_INVALIDDATA;
3076 total = lace_size[0] = num;
3077 for (n = 1; res == 0 && n < *laces - 1; n++) {
3080 r = matroska_ebmlnum_sint(matroska, data, size, &snum);
3081 if (r < 0 || lace_size[n - 1] + snum > (uint64_t)INT_MAX) {
3082 av_log(matroska->ctx, AV_LOG_INFO,
3083 "EBML block data error\n");
3084 res = r<0 ? r : AVERROR_INVALIDDATA;
3089 lace_size[n] = lace_size[n - 1] + snum;
3090 total += lace_size[n];
3092 if (size <= total) {
3093 res = AVERROR_INVALIDDATA;
3096 lace_size[*laces - 1] = size - total;
3102 *lace_buf = lace_size;
3108 static int matroska_parse_rm_audio(MatroskaDemuxContext *matroska,
3109 MatroskaTrack *track, AVStream *st,
3110 uint8_t *data, int size, uint64_t timecode,
3113 int a = st->codecpar->block_align;
3114 int sps = track->audio.sub_packet_size;
3115 int cfs = track->audio.coded_framesize;
3116 int h = track->audio.sub_packet_h;
3117 int y = track->audio.sub_packet_cnt;
3118 int w = track->audio.frame_size;
3121 if (!track->audio.pkt_cnt) {
3122 if (track->audio.sub_packet_cnt == 0)
3123 track->audio.buf_timecode = timecode;
3124 if (st->codecpar->codec_id == AV_CODEC_ID_RA_288) {
3125 if (size < cfs * h / 2) {
3126 av_log(matroska->ctx, AV_LOG_ERROR,
3127 "Corrupt int4 RM-style audio packet size\n");
3128 return AVERROR_INVALIDDATA;
3130 for (x = 0; x < h / 2; x++)
3131 memcpy(track->audio.buf + x * 2 * w + y * cfs,
3132 data + x * cfs, cfs);
3133 } else if (st->codecpar->codec_id == AV_CODEC_ID_SIPR) {
3135 av_log(matroska->ctx, AV_LOG_ERROR,
3136 "Corrupt sipr RM-style audio packet size\n");
3137 return AVERROR_INVALIDDATA;
3139 memcpy(track->audio.buf + y * w, data, w);
3141 if (size < sps * w / sps || h<=0 || w%sps) {
3142 av_log(matroska->ctx, AV_LOG_ERROR,
3143 "Corrupt generic RM-style audio packet size\n");
3144 return AVERROR_INVALIDDATA;
3146 for (x = 0; x < w / sps; x++)
3147 memcpy(track->audio.buf +
3148 sps * (h * x + ((h + 1) / 2) * (y & 1) + (y >> 1)),
3149 data + x * sps, sps);
3152 if (++track->audio.sub_packet_cnt >= h) {
3153 if (st->codecpar->codec_id == AV_CODEC_ID_SIPR)
3154 ff_rm_reorder_sipr_data(track->audio.buf, h, w);
3155 track->audio.sub_packet_cnt = 0;
3156 track->audio.pkt_cnt = h * w / a;
3160 while (track->audio.pkt_cnt) {
3162 AVPacket pktl, *pkt = &pktl;
3164 ret = av_new_packet(pkt, a);
3169 track->audio.buf + a * (h * w / a - track->audio.pkt_cnt--),
3171 pkt->pts = track->audio.buf_timecode;
3172 track->audio.buf_timecode = AV_NOPTS_VALUE;
3174 pkt->stream_index = st->index;
3175 ret = ff_packet_list_put(&matroska->queue, &matroska->queue_end, pkt, 0);
3177 av_packet_unref(pkt);
3178 return AVERROR(ENOMEM);
3185 /* reconstruct full wavpack blocks from mangled matroska ones */
3186 static int matroska_parse_wavpack(MatroskaTrack *track, uint8_t *src,
3187 uint8_t **pdst, int *size)
3189 uint8_t *dst = NULL;
3194 int ret, offset = 0;
3196 if (srclen < 12 || track->stream->codecpar->extradata_size < 2)
3197 return AVERROR_INVALIDDATA;
3199 ver = AV_RL16(track->stream->codecpar->extradata);
3201 samples = AV_RL32(src);
3205 while (srclen >= 8) {
3210 uint32_t flags = AV_RL32(src);
3211 uint32_t crc = AV_RL32(src + 4);
3215 multiblock = (flags & 0x1800) != 0x1800;
3218 ret = AVERROR_INVALIDDATA;
3221 blocksize = AV_RL32(src);
3227 if (blocksize > srclen) {
3228 ret = AVERROR_INVALIDDATA;
3232 tmp = av_realloc(dst, dstlen + blocksize + 32 + AV_INPUT_BUFFER_PADDING_SIZE);
3234 ret = AVERROR(ENOMEM);
3238 dstlen += blocksize + 32;
3240 AV_WL32(dst + offset, MKTAG('w', 'v', 'p', 'k')); // tag
3241 AV_WL32(dst + offset + 4, blocksize + 24); // blocksize - 8
3242 AV_WL16(dst + offset + 8, ver); // version
3243 AV_WL16(dst + offset + 10, 0); // track/index_no
3244 AV_WL32(dst + offset + 12, 0); // total samples
3245 AV_WL32(dst + offset + 16, 0); // block index
3246 AV_WL32(dst + offset + 20, samples); // number of samples
3247 AV_WL32(dst + offset + 24, flags); // flags
3248 AV_WL32(dst + offset + 28, crc); // crc
3249 memcpy(dst + offset + 32, src, blocksize); // block data
3252 srclen -= blocksize;
3253 offset += blocksize + 32;
3256 memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3268 static int matroska_parse_prores(MatroskaTrack *track, uint8_t *src,
3269 uint8_t **pdst, int *size)
3274 if (AV_RB32(&src[4]) != MKBETAG('i', 'c', 'p', 'f')) {
3275 dst = av_malloc(dstlen + 8 + AV_INPUT_BUFFER_PADDING_SIZE);
3277 return AVERROR(ENOMEM);
3279 AV_WB32(dst, dstlen);
3280 AV_WB32(dst + 4, MKBETAG('i', 'c', 'p', 'f'));
3281 memcpy(dst + 8, src, dstlen);
3282 memset(dst + 8 + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3292 static int matroska_parse_webvtt(MatroskaDemuxContext *matroska,
3293 MatroskaTrack *track,
3295 uint8_t *data, int data_len,
3300 AVPacket pktl, *pkt = &pktl;
3301 uint8_t *id, *settings, *text, *buf;
3302 int id_len, settings_len, text_len;
3307 return AVERROR_INVALIDDATA;
3310 q = data + data_len;
3315 if (*p == '\r' || *p == '\n') {
3324 if (p >= q || *p != '\n')
3325 return AVERROR_INVALIDDATA;
3331 if (*p == '\r' || *p == '\n') {
3332 settings_len = p - settings;
3340 if (p >= q || *p != '\n')
3341 return AVERROR_INVALIDDATA;
3346 while (text_len > 0) {
3347 const int len = text_len - 1;
3348 const uint8_t c = p[len];
3349 if (c != '\r' && c != '\n')
3355 return AVERROR_INVALIDDATA;
3357 err = av_new_packet(pkt, text_len);
3362 memcpy(pkt->data, text, text_len);
3365 buf = av_packet_new_side_data(pkt,
3366 AV_PKT_DATA_WEBVTT_IDENTIFIER,
3369 av_packet_unref(pkt);
3370 return AVERROR(ENOMEM);
3372 memcpy(buf, id, id_len);
3375 if (settings_len > 0) {
3376 buf = av_packet_new_side_data(pkt,
3377 AV_PKT_DATA_WEBVTT_SETTINGS,
3380 av_packet_unref(pkt);
3381 return AVERROR(ENOMEM);
3383 memcpy(buf, settings, settings_len);
3386 // Do we need this for subtitles?
3387 // pkt->flags = AV_PKT_FLAG_KEY;
3389 pkt->stream_index = st->index;
3390 pkt->pts = timecode;
3392 // Do we need this for subtitles?
3393 // pkt->dts = timecode;
3395 pkt->duration = duration;
3398 err = ff_packet_list_put(&matroska->queue, &matroska->queue_end, pkt, 0);
3400 av_packet_unref(pkt);
3401 return AVERROR(ENOMEM);
3407 static int matroska_parse_frame(MatroskaDemuxContext *matroska,
3408 MatroskaTrack *track, AVStream *st,
3409 AVBufferRef *buf, uint8_t *data, int pkt_size,
3410 uint64_t timecode, uint64_t lace_duration,
3411 int64_t pos, int is_keyframe,
3412 uint8_t *additional, uint64_t additional_id, int additional_size,
3413 int64_t discard_padding)
3415 MatroskaTrackEncoding *encodings = track->encodings.elem;
3416 uint8_t *pkt_data = data;
3418 AVPacket pktl, *pkt = &pktl;
3420 if (encodings && !encodings->type && encodings->scope & 1) {
3421 res = matroska_decode_buffer(&pkt_data, &pkt_size, track);
3426 if (st->codecpar->codec_id == AV_CODEC_ID_WAVPACK) {
3428 res = matroska_parse_wavpack(track, pkt_data, &wv_data, &pkt_size);
3430 av_log(matroska->ctx, AV_LOG_ERROR,
3431 "Error parsing a wavpack block.\n");
3434 if (pkt_data != data)
3435 av_freep(&pkt_data);
3439 if (st->codecpar->codec_id == AV_CODEC_ID_PRORES) {
3441 res = matroska_parse_prores(track, pkt_data, &pr_data, &pkt_size);
3443 av_log(matroska->ctx, AV_LOG_ERROR,
3444 "Error parsing a prores block.\n");
3447 if (pkt_data != data)
3448 av_freep(&pkt_data);
3452 av_init_packet(pkt);
3453 if (pkt_data != data)
3454 pkt->buf = av_buffer_create(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE,
3457 pkt->buf = av_buffer_ref(buf);
3460 res = AVERROR(ENOMEM);
3464 pkt->data = pkt_data;
3465 pkt->size = pkt_size;
3466 pkt->flags = is_keyframe;
3467 pkt->stream_index = st->index;
3469 if (additional_size > 0) {
3470 uint8_t *side_data = av_packet_new_side_data(pkt,
3471 AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL,
3472 additional_size + 8);
3474 av_packet_unref(pkt);
3475 return AVERROR(ENOMEM);
3477 AV_WB64(side_data, additional_id);
3478 memcpy(side_data + 8, additional, additional_size);
3481 if (discard_padding) {
3482 uint8_t *side_data = av_packet_new_side_data(pkt,
3483 AV_PKT_DATA_SKIP_SAMPLES,
3486 av_packet_unref(pkt);
3487 return AVERROR(ENOMEM);
3489 discard_padding = av_rescale_q(discard_padding,
3490 (AVRational){1, 1000000000},
3491 (AVRational){1, st->codecpar->sample_rate});
3492 if (discard_padding > 0) {
3493 AV_WL32(side_data + 4, discard_padding);
3495 AV_WL32(side_data, -discard_padding);
3499 if (track->ms_compat)
3500 pkt->dts = timecode;
3502 pkt->pts = timecode;
3504 pkt->duration = lace_duration;
3506 #if FF_API_CONVERGENCE_DURATION
3507 FF_DISABLE_DEPRECATION_WARNINGS
3508 if (st->codecpar->codec_id == AV_CODEC_ID_SUBRIP) {
3509 pkt->convergence_duration = lace_duration;
3511 FF_ENABLE_DEPRECATION_WARNINGS
3514 res = ff_packet_list_put(&matroska->queue, &matroska->queue_end, pkt, 0);
3516 av_packet_unref(pkt);
3517 return AVERROR(ENOMEM);
3523 if (pkt_data != data)
3524 av_freep(&pkt_data);
3528 static int matroska_parse_block(MatroskaDemuxContext *matroska, AVBufferRef *buf, uint8_t *data,
3529 int size, int64_t pos, uint64_t cluster_time,
3530 uint64_t block_duration, int is_keyframe,
3531 uint8_t *additional, uint64_t additional_id, int additional_size,
3532 int64_t cluster_pos, int64_t discard_padding)
3534 uint64_t timecode = AV_NOPTS_VALUE;
3535 MatroskaTrack *track;
3539 uint32_t *lace_size = NULL;
3540 int n, flags, laces = 0;
3542 int trust_default_duration = 1;
3544 if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
3550 track = matroska_find_track_by_num(matroska, num);
3551 if (!track || !track->stream) {
3552 av_log(matroska->ctx, AV_LOG_INFO,
3553 "Invalid stream %"PRIu64"\n", num);
3554 return AVERROR_INVALIDDATA;
3555 } else if (size <= 3)
3558 if (st->discard >= AVDISCARD_ALL)
3560 av_assert1(block_duration != AV_NOPTS_VALUE);
3562 block_time = sign_extend(AV_RB16(data), 16);
3566 if (is_keyframe == -1)
3567 is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
3569 if (cluster_time != (uint64_t) -1 &&
3570 (block_time >= 0 || cluster_time >= -block_time)) {
3571 timecode = cluster_time + block_time - track->codec_delay_in_track_tb;
3572 if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE &&
3573 timecode < track->end_timecode)
3574 is_keyframe = 0; /* overlapping subtitles are not key frame */
3576 ff_reduce_index(matroska->ctx, st->index);
3577 av_add_index_entry(st, cluster_pos, timecode, 0, 0,
3582 if (matroska->skip_to_keyframe &&
3583 track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
3584 // Compare signed timecodes. Timecode may be negative due to codec delay
3585 // offset. We don't support timestamps greater than int64_t anyway - see
3587 if ((int64_t)timecode < (int64_t)matroska->skip_to_timecode)
3590 matroska->skip_to_keyframe = 0;
3591 else if (!st->skip_to_keyframe) {
3592 av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
3593 matroska->skip_to_keyframe = 0;
3597 res = matroska_parse_laces(matroska, &data, &size, (flags & 0x06) >> 1,
3598 &lace_size, &laces);
3603 if (track->audio.samplerate == 8000) {
3604 // If this is needed for more codecs, then add them here
3605 if (st->codecpar->codec_id == AV_CODEC_ID_AC3) {
3606 if (track->audio.samplerate != st->codecpar->sample_rate || !st->codecpar->frame_size)
3607 trust_default_duration = 0;
3611 if (!block_duration && trust_default_duration)
3612 block_duration = track->default_duration * laces / matroska->time_scale;
3614 if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
3615 track->end_timecode =
3616 FFMAX(track->end_timecode, timecode + block_duration);
3618 for (n = 0; n < laces; n++) {
3619 int64_t lace_duration = block_duration*(n+1) / laces - block_duration*n / laces;
3621 if (lace_size[n] > size) {
3622 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
3626 if ((st->codecpar->codec_id == AV_CODEC_ID_RA_288 ||
3627 st->codecpar->codec_id == AV_CODEC_ID_COOK ||
3628 st->codecpar->codec_id == AV_CODEC_ID_SIPR ||
3629 st->codecpar->codec_id == AV_CODEC_ID_ATRAC3) &&
3630 st->codecpar->block_align && track->audio.sub_packet_size) {
3631 res = matroska_parse_rm_audio(matroska, track, st, data,
3637 } else if (st->codecpar->codec_id == AV_CODEC_ID_WEBVTT) {
3638 res = matroska_parse_webvtt(matroska, track, st,
3640 timecode, lace_duration,
3645 res = matroska_parse_frame(matroska, track, st, buf, data, lace_size[n],
3646 timecode, lace_duration, pos,
3647 !n ? is_keyframe : 0,
3648 additional, additional_id, additional_size,
3654 if (timecode != AV_NOPTS_VALUE)
3655 timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
3656 data += lace_size[n];
3657 size -= lace_size[n];
3665 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
3667 MatroskaCluster *cluster = &matroska->current_cluster;
3668 MatroskaBlock *block = &cluster->block;
3671 av_assert0(matroska->num_levels <= 2);
3673 if (matroska->num_levels == 1) {
3674 res = ebml_parse(matroska, matroska_clusters, NULL);
3677 /* Found a cluster: subtract the size of the ID already read. */
3678 cluster->pos = avio_tell(matroska->ctx->pb) - 4;
3680 res = ebml_parse(matroska, matroska_cluster_enter, cluster);
3686 if (matroska->num_levels == 2) {
3687 /* We are inside a cluster. */
3688 res = ebml_parse(matroska, matroska_cluster_parsing, cluster);
3690 if (res >= 0 && block->bin.size > 0) {
3691 int is_keyframe = block->non_simple ? block->reference == INT64_MIN : -1;
3692 uint8_t* additional = block->additional.size > 0 ?
3693 block->additional.data : NULL;
3695 res = matroska_parse_block(matroska, block->bin.buf, block->bin.data,
3696 block->bin.size, block->bin.pos,
3697 cluster->timecode, block->duration,
3698 is_keyframe, additional, block->additional_id,
3699 block->additional.size, cluster->pos,
3700 block->discard_padding);
3703 ebml_free(matroska_blockgroup, block);
3704 memset(block, 0, sizeof(*block));
3705 } else if (!matroska->num_levels) {
3706 if (!avio_feof(matroska->ctx->pb)) {
3707 avio_r8(matroska->ctx->pb);
3708 if (!avio_feof(matroska->ctx->pb)) {
3709 av_log(matroska->ctx, AV_LOG_WARNING, "File extends beyond "
3710 "end of segment.\n");
3711 return AVERROR_INVALIDDATA;
3721 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
3723 MatroskaDemuxContext *matroska = s->priv_data;
3726 if (matroska->resync_pos == -1) {
3727 // This can only happen if generic seeking has been used.
3728 matroska->resync_pos = avio_tell(s->pb);
3731 while (matroska_deliver_packet(matroska, pkt)) {
3733 return (ret < 0) ? ret : AVERROR_EOF;
3734 if (matroska_parse_cluster(matroska) < 0 && !matroska->done)
3735 ret = matroska_resync(matroska, matroska->resync_pos);
3741 static int matroska_read_seek(AVFormatContext *s, int stream_index,
3742 int64_t timestamp, int flags)
3744 MatroskaDemuxContext *matroska = s->priv_data;
3745 MatroskaTrack *tracks = NULL;
3746 AVStream *st = s->streams[stream_index];
3749 /* Parse the CUES now since we need the index data to seek. */
3750 if (matroska->cues_parsing_deferred > 0) {
3751 matroska->cues_parsing_deferred = 0;
3752 matroska_parse_cues(matroska);
3755 if (!st->nb_index_entries)
3757 timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
3759 if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 || index == st->nb_index_entries - 1) {
3760 matroska_reset_status(matroska, 0, st->index_entries[st->nb_index_entries - 1].pos);
3761 while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 || index == st->nb_index_entries - 1) {
3762 matroska_clear_queue(matroska);
3763 if (matroska_parse_cluster(matroska) < 0)
3768 matroska_clear_queue(matroska);
3769 if (index < 0 || (matroska->cues_parsing_deferred < 0 && index == st->nb_index_entries - 1))
3772 tracks = matroska->tracks.elem;
3773 for (i = 0; i < matroska->tracks.nb_elem; i++) {
3774 tracks[i].audio.pkt_cnt = 0;
3775 tracks[i].audio.sub_packet_cnt = 0;
3776 tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
3777 tracks[i].end_timecode = 0;
3780 /* We seek to a level 1 element, so set the appropriate status. */
3781 matroska_reset_status(matroska, 0, st->index_entries[index].pos);
3782 if (flags & AVSEEK_FLAG_ANY) {
3783 st->skip_to_keyframe = 0;
3784 matroska->skip_to_timecode = timestamp;
3786 st->skip_to_keyframe = 1;
3787 matroska->skip_to_timecode = st->index_entries[index].timestamp;
3789 matroska->skip_to_keyframe = 1;
3791 ff_update_cur_dts(s, st, st->index_entries[index].timestamp);
3794 // slightly hackish but allows proper fallback to
3795 // the generic seeking code.
3796 matroska_reset_status(matroska, 0, -1);
3797 matroska->resync_pos = -1;
3798 matroska_clear_queue(matroska);
3799 st->skip_to_keyframe =
3800 matroska->skip_to_keyframe = 0;
3805 static int matroska_read_close(AVFormatContext *s)
3807 MatroskaDemuxContext *matroska = s->priv_data;
3808 MatroskaTrack *tracks = matroska->tracks.elem;
3811 matroska_clear_queue(matroska);
3813 for (n = 0; n < matroska->tracks.nb_elem; n++)
3814 if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
3815 av_freep(&tracks[n].audio.buf);
3816 ebml_free(matroska_segment, matroska);
3822 int64_t start_time_ns;
3823 int64_t end_time_ns;
3824 int64_t start_offset;
3828 /* This function searches all the Cues and returns the CueDesc corresponding to
3829 * the timestamp ts. Returned CueDesc will be such that start_time_ns <= ts <
3830 * end_time_ns. All 4 fields will be set to -1 if ts >= file's duration.
3832 static CueDesc get_cue_desc(AVFormatContext *s, int64_t ts, int64_t cues_start) {
3833 MatroskaDemuxContext *matroska = s->priv_data;
3836 int nb_index_entries = s->streams[0]->nb_index_entries;
3837 AVIndexEntry *index_entries = s->streams[0]->index_entries;
3838 if (ts >= matroska->duration * matroska->time_scale) return (CueDesc) {-1, -1, -1, -1};
3839 for (i = 1; i < nb_index_entries; i++) {
3840 if (index_entries[i - 1].timestamp * matroska->time_scale <= ts &&
3841 index_entries[i].timestamp * matroska->time_scale > ts) {
3846 cue_desc.start_time_ns = index_entries[i].timestamp * matroska->time_scale;
3847 cue_desc.start_offset = index_entries[i].pos - matroska->segment_start;
3848 if (i != nb_index_entries - 1) {
3849 cue_desc.end_time_ns = index_entries[i + 1].timestamp * matroska->time_scale;
3850 cue_desc.end_offset = index_entries[i + 1].pos - matroska->segment_start;
3852 cue_desc.end_time_ns = matroska->duration * matroska->time_scale;
3853 // FIXME: this needs special handling for files where Cues appear
3854 // before Clusters. the current logic assumes Cues appear after
3856 cue_desc.end_offset = cues_start - matroska->segment_start;
3861 static int webm_clusters_start_with_keyframe(AVFormatContext *s)
3863 MatroskaDemuxContext *matroska = s->priv_data;
3864 uint32_t id = matroska->current_id;
3865 int64_t cluster_pos, before_pos;
3867 if (s->streams[0]->nb_index_entries <= 0) return 0;
3868 // seek to the first cluster using cues.
3869 index = av_index_search_timestamp(s->streams[0], 0, 0);
3870 if (index < 0) return 0;
3871 cluster_pos = s->streams[0]->index_entries[index].pos;
3872 before_pos = avio_tell(s->pb);
3874 uint64_t cluster_id, cluster_length;
3877 avio_seek(s->pb, cluster_pos, SEEK_SET);
3878 // read cluster id and length
3879 read = ebml_read_num(matroska, matroska->ctx->pb, 4, &cluster_id, 1);
3880 if (read < 0 || cluster_id != 0xF43B675) // done with all clusters
3882 read = ebml_read_length(matroska, matroska->ctx->pb, &cluster_length);
3886 matroska_reset_status(matroska, 0, cluster_pos);
3887 matroska_clear_queue(matroska);
3888 if (matroska_parse_cluster(matroska) < 0 ||
3892 pkt = &matroska->queue->pkt;
3893 // 4 + read is the length of the cluster id and the cluster length field.
3894 cluster_pos += 4 + read + cluster_length;
3895 if (!(pkt->flags & AV_PKT_FLAG_KEY)) {
3901 /* Restore the status after matroska_read_header: */
3902 matroska_reset_status(matroska, id, before_pos);
3907 static int buffer_size_after_time_downloaded(int64_t time_ns, double search_sec, int64_t bps,
3908 double min_buffer, double* buffer,
3909 double* sec_to_download, AVFormatContext *s,
3912 double nano_seconds_per_second = 1000000000.0;
3913 double time_sec = time_ns / nano_seconds_per_second;
3915 int64_t time_to_search_ns = (int64_t)(search_sec * nano_seconds_per_second);
3916 int64_t end_time_ns = time_ns + time_to_search_ns;
3917 double sec_downloaded = 0.0;
3918 CueDesc desc_curr = get_cue_desc(s, time_ns, cues_start);
3919 if (desc_curr.start_time_ns == -1)
3921 *sec_to_download = 0.0;
3923 // Check for non cue start time.
3924 if (time_ns > desc_curr.start_time_ns) {
3925 int64_t cue_nano = desc_curr.end_time_ns - time_ns;
3926 double percent = (double)(cue_nano) / (desc_curr.end_time_ns - desc_curr.start_time_ns);
3927 double cueBytes = (desc_curr.end_offset - desc_curr.start_offset) * percent;
3928 double timeToDownload = (cueBytes * 8.0) / bps;
3930 sec_downloaded += (cue_nano / nano_seconds_per_second) - timeToDownload;
3931 *sec_to_download += timeToDownload;
3933 // Check if the search ends within the first cue.
3934 if (desc_curr.end_time_ns >= end_time_ns) {
3935 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
3936 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
3937 sec_downloaded = percent_to_sub * sec_downloaded;
3938 *sec_to_download = percent_to_sub * *sec_to_download;
3941 if ((sec_downloaded + *buffer) <= min_buffer) {
3945 // Get the next Cue.
3946 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
3949 while (desc_curr.start_time_ns != -1) {
3950 int64_t desc_bytes = desc_curr.end_offset - desc_curr.start_offset;
3951 int64_t desc_ns = desc_curr.end_time_ns - desc_curr.start_time_ns;
3952 double desc_sec = desc_ns / nano_seconds_per_second;
3953 double bits = (desc_bytes * 8.0);
3954 double time_to_download = bits / bps;
3956 sec_downloaded += desc_sec - time_to_download;
3957 *sec_to_download += time_to_download;
3959 if (desc_curr.end_time_ns >= end_time_ns) {
3960 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
3961 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
3962 sec_downloaded = percent_to_sub * sec_downloaded;
3963 *sec_to_download = percent_to_sub * *sec_to_download;
3965 if ((sec_downloaded + *buffer) <= min_buffer)
3970 if ((sec_downloaded + *buffer) <= min_buffer) {
3975 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
3977 *buffer = *buffer + sec_downloaded;
3981 /* This function computes the bandwidth of the WebM file with the help of
3982 * buffer_size_after_time_downloaded() function. Both of these functions are
3983 * adapted from WebM Tools project and are adapted to work with FFmpeg's
3984 * Matroska parsing mechanism.
3986 * Returns the bandwidth of the file on success; -1 on error.
3988 static int64_t webm_dash_manifest_compute_bandwidth(AVFormatContext *s, int64_t cues_start)
3990 MatroskaDemuxContext *matroska = s->priv_data;
3991 AVStream *st = s->streams[0];
3992 double bandwidth = 0.0;
3995 for (i = 0; i < st->nb_index_entries; i++) {
3996 int64_t prebuffer_ns = 1000000000;
3997 int64_t time_ns = st->index_entries[i].timestamp * matroska->time_scale;
3998 double nano_seconds_per_second = 1000000000.0;
3999 int64_t prebuffered_ns = time_ns + prebuffer_ns;
4000 double prebuffer_bytes = 0.0;
4001 int64_t temp_prebuffer_ns = prebuffer_ns;
4002 int64_t pre_bytes, pre_ns;
4003 double pre_sec, prebuffer, bits_per_second;
4004 CueDesc desc_beg = get_cue_desc(s, time_ns, cues_start);
4006 // Start with the first Cue.
4007 CueDesc desc_end = desc_beg;
4009 // Figure out how much data we have downloaded for the prebuffer. This will
4010 // be used later to adjust the bits per sample to try.
4011 while (desc_end.start_time_ns != -1 && desc_end.end_time_ns < prebuffered_ns) {
4012 // Prebuffered the entire Cue.
4013 prebuffer_bytes += desc_end.end_offset - desc_end.start_offset;
4014 temp_prebuffer_ns -= desc_end.end_time_ns - desc_end.start_time_ns;
4015 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4017 if (desc_end.start_time_ns == -1) {
4018 // The prebuffer is larger than the duration.
4019 if (matroska->duration * matroska->time_scale >= prebuffered_ns)
4021 bits_per_second = 0.0;
4023 // The prebuffer ends in the last Cue. Estimate how much data was
4025 pre_bytes = desc_end.end_offset - desc_end.start_offset;
4026 pre_ns = desc_end.end_time_ns - desc_end.start_time_ns;
4027 pre_sec = pre_ns / nano_seconds_per_second;
4029 pre_bytes * ((temp_prebuffer_ns / nano_seconds_per_second) / pre_sec);
4031 prebuffer = prebuffer_ns / nano_seconds_per_second;
4033 // Set this to 0.0 in case our prebuffer buffers the entire video.
4034 bits_per_second = 0.0;
4036 int64_t desc_bytes = desc_end.end_offset - desc_beg.start_offset;
4037 int64_t desc_ns = desc_end.end_time_ns - desc_beg.start_time_ns;
4038 double desc_sec = desc_ns / nano_seconds_per_second;
4039 double calc_bits_per_second = (desc_bytes * 8) / desc_sec;
4041 // Drop the bps by the percentage of bytes buffered.
4042 double percent = (desc_bytes - prebuffer_bytes) / desc_bytes;
4043 double mod_bits_per_second = calc_bits_per_second * percent;
4045 if (prebuffer < desc_sec) {
4047 (double)(matroska->duration * matroska->time_scale) / nano_seconds_per_second;
4049 // Add 1 so the bits per second should be a little bit greater than file
4051 int64_t bps = (int64_t)(mod_bits_per_second) + 1;
4052 const double min_buffer = 0.0;
4053 double buffer = prebuffer;
4054 double sec_to_download = 0.0;
4056 int rv = buffer_size_after_time_downloaded(prebuffered_ns, search_sec, bps,
4057 min_buffer, &buffer, &sec_to_download,
4061 } else if (rv == 0) {
4062 bits_per_second = (double)(bps);
4067 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4068 } while (desc_end.start_time_ns != -1);
4070 if (bandwidth < bits_per_second) bandwidth = bits_per_second;
4072 return (int64_t)bandwidth;
4075 static int webm_dash_manifest_cues(AVFormatContext *s, int64_t init_range)
4077 MatroskaDemuxContext *matroska = s->priv_data;
4078 EbmlList *seekhead_list = &matroska->seekhead;
4079 MatroskaSeekhead *seekhead = seekhead_list->elem;
4081 int64_t cues_start = -1, cues_end = -1, before_pos, bandwidth;
4085 // determine cues start and end positions
4086 for (i = 0; i < seekhead_list->nb_elem; i++)
4087 if (seekhead[i].id == MATROSKA_ID_CUES)
4090 if (i >= seekhead_list->nb_elem) return -1;
4092 before_pos = avio_tell(matroska->ctx->pb);
4093 cues_start = seekhead[i].pos + matroska->segment_start;
4094 if (avio_seek(matroska->ctx->pb, cues_start, SEEK_SET) == cues_start) {
4095 // cues_end is computed as cues_start + cues_length + length of the
4096 // Cues element ID (i.e. 4) + EBML length of the Cues element.
4097 // cues_end is inclusive and the above sum is reduced by 1.
4098 uint64_t cues_length, cues_id;
4100 bytes_read = ebml_read_num (matroska, matroska->ctx->pb, 4, &cues_id, 1);
4101 if (bytes_read < 0 || cues_id != (MATROSKA_ID_CUES & 0xfffffff))
4102 return bytes_read < 0 ? bytes_read : AVERROR_INVALIDDATA;
4103 bytes_read = ebml_read_length(matroska, matroska->ctx->pb, &cues_length);
4106 cues_end = cues_start + 4 + bytes_read + cues_length - 1;
4108 avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
4109 if (cues_start == -1 || cues_end == -1) return -1;
4112 matroska_parse_cues(matroska);
4115 av_dict_set_int(&s->streams[0]->metadata, CUES_START, cues_start, 0);
4118 av_dict_set_int(&s->streams[0]->metadata, CUES_END, cues_end, 0);
4120 // if the file has cues at the start, fix up the init range so that
4121 // it does not include it
4122 if (cues_start <= init_range)
4123 av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, cues_start - 1, 0);
4126 bandwidth = webm_dash_manifest_compute_bandwidth(s, cues_start);
4127 if (bandwidth < 0) return -1;
4128 av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH, bandwidth, 0);
4130 // check if all clusters start with key frames
4131 av_dict_set_int(&s->streams[0]->metadata, CLUSTER_KEYFRAME, webm_clusters_start_with_keyframe(s), 0);
4133 // store cue point timestamps as a comma separated list for checking subsegment alignment in
4134 // the muxer. assumes that each timestamp cannot be more than 20 characters long.
4135 buf = av_malloc_array(s->streams[0]->nb_index_entries, 20);
4136 if (!buf) return -1;
4138 for (i = 0; i < s->streams[0]->nb_index_entries; i++) {
4139 int ret = snprintf(buf + end, 20,
4140 "%" PRId64"%s", s->streams[0]->index_entries[i].timestamp,
4141 i != s->streams[0]->nb_index_entries - 1 ? "," : "");
4142 if (ret <= 0 || (ret == 20 && i == s->streams[0]->nb_index_entries - 1)) {
4143 av_log(s, AV_LOG_ERROR, "timestamp too long.\n");
4145 return AVERROR_INVALIDDATA;
4149 av_dict_set(&s->streams[0]->metadata, CUE_TIMESTAMPS, buf, 0);
4155 static int webm_dash_manifest_read_header(AVFormatContext *s)
4158 int ret = matroska_read_header(s);
4160 MatroskaTrack *tracks;
4161 MatroskaDemuxContext *matroska = s->priv_data;
4163 av_log(s, AV_LOG_ERROR, "Failed to read file headers\n");
4166 if (!s->nb_streams) {
4167 matroska_read_close(s);
4168 av_log(s, AV_LOG_ERROR, "No streams found\n");
4169 return AVERROR_INVALIDDATA;
4172 if (!matroska->is_live) {
4173 buf = av_asprintf("%g", matroska->duration);
4174 if (!buf) return AVERROR(ENOMEM);
4175 av_dict_set(&s->streams[0]->metadata, DURATION, buf, 0);
4178 // initialization range
4179 // 5 is the offset of Cluster ID.
4180 init_range = avio_tell(s->pb) - 5;
4181 av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, init_range, 0);
4184 // basename of the file
4185 buf = strrchr(s->url, '/');
4186 av_dict_set(&s->streams[0]->metadata, FILENAME, buf ? ++buf : s->url, 0);
4189 tracks = matroska->tracks.elem;
4190 av_dict_set_int(&s->streams[0]->metadata, TRACK_NUMBER, tracks[0].num, 0);
4192 // parse the cues and populate Cue related fields
4193 if (!matroska->is_live) {
4194 ret = webm_dash_manifest_cues(s, init_range);
4196 av_log(s, AV_LOG_ERROR, "Error parsing Cues\n");
4201 // use the bandwidth from the command line if it was provided
4202 if (matroska->bandwidth > 0) {
4203 av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH,
4204 matroska->bandwidth, 0);
4209 static int webm_dash_manifest_read_packet(AVFormatContext *s, AVPacket *pkt)
4214 #define OFFSET(x) offsetof(MatroskaDemuxContext, x)
4215 static const AVOption options[] = {
4216 { "live", "flag indicating that the input is a live file that only has the headers.", OFFSET(is_live), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, AV_OPT_FLAG_DECODING_PARAM },
4217 { "bandwidth", "bandwidth of this stream to be specified in the DASH manifest.", OFFSET(bandwidth), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX, AV_OPT_FLAG_DECODING_PARAM },
4221 static const AVClass webm_dash_class = {
4222 .class_name = "WebM DASH Manifest demuxer",
4223 .item_name = av_default_item_name,
4225 .version = LIBAVUTIL_VERSION_INT,
4228 AVInputFormat ff_matroska_demuxer = {
4229 .name = "matroska,webm",
4230 .long_name = NULL_IF_CONFIG_SMALL("Matroska / WebM"),
4231 .extensions = "mkv,mk3d,mka,mks",
4232 .priv_data_size = sizeof(MatroskaDemuxContext),
4233 .read_probe = matroska_probe,
4234 .read_header = matroska_read_header,
4235 .read_packet = matroska_read_packet,
4236 .read_close = matroska_read_close,
4237 .read_seek = matroska_read_seek,
4238 .mime_type = "audio/webm,audio/x-matroska,video/webm,video/x-matroska"
4241 AVInputFormat ff_webm_dash_manifest_demuxer = {
4242 .name = "webm_dash_manifest",
4243 .long_name = NULL_IF_CONFIG_SMALL("WebM DASH Manifest"),
4244 .priv_data_size = sizeof(MatroskaDemuxContext),
4245 .read_header = webm_dash_manifest_read_header,
4246 .read_packet = webm_dash_manifest_read_packet,
4247 .read_close = matroska_read_close,
4248 .priv_class = &webm_dash_class,