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"
51 #include "libavcodec/packet_internal.h"
54 #include "avio_internal.h"
59 /* For ff_codec_get_id(). */
70 #include "qtpalette.h"
72 #define EBML_UNKNOWN_LENGTH UINT64_MAX /* EBML unknown length, in uint64_t */
73 #define NEEDS_CHECKING 2 /* Indicates that some error checks
74 * still need to be performed */
75 #define LEVEL_ENDED 3 /* return value of ebml_parse when the
76 * syntax level used for parsing ended. */
77 #define SKIP_THRESHOLD 1024 * 1024 /* In non-seekable mode, if more than SKIP_THRESHOLD
78 * of unkown, potentially damaged data is encountered,
79 * it is considered an error. */
80 #define UNKNOWN_EQUIV 50 * 1024 /* An unknown element is considered equivalent
81 * to this many bytes of unknown data for the
82 * SKIP_THRESHOLD check. */
98 typedef struct CountedElement {
108 typedef const struct EbmlSyntax {
112 size_t list_elem_size;
119 const struct EbmlSyntax *n;
123 typedef struct EbmlList {
125 unsigned int alloc_elem_size;
129 typedef struct EbmlBin {
136 typedef struct Ebml {
141 uint64_t doctype_version;
144 typedef struct MatroskaTrackCompression {
147 } MatroskaTrackCompression;
149 typedef struct MatroskaTrackEncryption {
152 } MatroskaTrackEncryption;
154 typedef struct MatroskaTrackEncoding {
157 MatroskaTrackCompression compression;
158 MatroskaTrackEncryption encryption;
159 } MatroskaTrackEncoding;
161 typedef struct MatroskaMasteringMeta {
170 double max_luminance;
171 CountedElement min_luminance;
172 } MatroskaMasteringMeta;
174 typedef struct MatroskaTrackVideoColor {
175 uint64_t matrix_coefficients;
176 uint64_t bits_per_channel;
177 uint64_t chroma_sub_horz;
178 uint64_t chroma_sub_vert;
179 uint64_t cb_sub_horz;
180 uint64_t cb_sub_vert;
181 uint64_t chroma_siting_horz;
182 uint64_t chroma_siting_vert;
184 uint64_t transfer_characteristics;
188 MatroskaMasteringMeta mastering_meta;
189 } MatroskaTrackVideoColor;
191 typedef struct MatroskaTrackVideoProjection {
197 } MatroskaTrackVideoProjection;
199 typedef struct MatroskaTrackVideo {
201 uint64_t display_width;
202 uint64_t display_height;
203 uint64_t pixel_width;
204 uint64_t pixel_height;
206 uint64_t display_unit;
208 uint64_t field_order;
209 uint64_t stereo_mode;
212 MatroskaTrackVideoProjection projection;
213 } MatroskaTrackVideo;
215 typedef struct MatroskaTrackAudio {
217 double out_samplerate;
221 /* real audio header (extracted from extradata) */
228 uint64_t buf_timecode;
230 } MatroskaTrackAudio;
232 typedef struct MatroskaTrackPlane {
235 } MatroskaTrackPlane;
237 typedef struct MatroskaTrackOperation {
238 EbmlList combine_planes;
239 } MatroskaTrackOperation;
241 typedef struct MatroskaTrack {
250 uint64_t default_duration;
251 uint64_t flag_default;
252 uint64_t flag_forced;
253 uint64_t seek_preroll;
254 MatroskaTrackVideo video;
255 MatroskaTrackAudio audio;
256 MatroskaTrackOperation operation;
258 uint64_t codec_delay;
259 uint64_t codec_delay_in_track_tb;
262 int64_t end_timecode;
265 uint64_t max_block_additional_id;
267 uint32_t palette[AVPALETTE_COUNT];
271 typedef struct MatroskaAttachment {
279 } MatroskaAttachment;
281 typedef struct MatroskaChapter {
290 typedef struct MatroskaIndexPos {
295 typedef struct MatroskaIndex {
300 typedef struct MatroskaTag {
308 typedef struct MatroskaTagTarget {
316 typedef struct MatroskaTags {
317 MatroskaTagTarget target;
321 typedef struct MatroskaSeekhead {
326 typedef struct MatroskaLevel {
331 typedef struct MatroskaBlock {
333 CountedElement reference;
336 uint64_t additional_id;
338 int64_t discard_padding;
341 typedef struct MatroskaCluster {
347 typedef struct MatroskaLevel1Element {
351 } MatroskaLevel1Element;
353 typedef struct MatroskaDemuxContext {
354 const AVClass *class;
355 AVFormatContext *ctx;
358 MatroskaLevel levels[EBML_MAX_DEPTH];
370 EbmlList attachments;
376 /* byte position of the segment inside the stream */
377 int64_t segment_start;
379 /* the packet queue */
381 AVPacketList *queue_end;
385 /* What to skip before effectively reading a packet. */
386 int skip_to_keyframe;
387 uint64_t skip_to_timecode;
389 /* File has a CUES element, but we defer parsing until it is needed. */
390 int cues_parsing_deferred;
392 /* Level1 elements and whether they were read yet */
393 MatroskaLevel1Element level1_elems[64];
394 int num_level1_elems;
396 MatroskaCluster current_cluster;
398 /* WebM DASH Manifest live flag */
401 /* Bandwidth value for WebM DASH Manifest */
403 } MatroskaDemuxContext;
405 #define CHILD_OF(parent) { .def = { .n = parent } }
407 // The following forward declarations need their size because
408 // a tentative definition with internal linkage must not be an
409 // incomplete type (6.7.2 in C90, 6.9.2 in C99).
410 // Removing the sizes breaks MSVC.
411 static EbmlSyntax ebml_syntax[3], matroska_segment[9], matroska_track_video_color[15], matroska_track_video[19],
412 matroska_track[27], matroska_track_encoding[6], matroska_track_encodings[2],
413 matroska_track_combine_planes[2], matroska_track_operation[2], matroska_tracks[2],
414 matroska_attachments[2], matroska_chapter_entry[9], matroska_chapter[6], matroska_chapters[2],
415 matroska_index_entry[3], matroska_index[2], matroska_tag[3], matroska_tags[2], matroska_seekhead[2],
416 matroska_blockadditions[2], matroska_blockgroup[8], matroska_cluster_parsing[8];
418 static EbmlSyntax ebml_header[] = {
419 { EBML_ID_EBMLREADVERSION, EBML_UINT, 0, 0, offsetof(Ebml, version), { .u = EBML_VERSION } },
420 { EBML_ID_EBMLMAXSIZELENGTH, EBML_UINT, 0, 0, offsetof(Ebml, max_size), { .u = 8 } },
421 { EBML_ID_EBMLMAXIDLENGTH, EBML_UINT, 0, 0, offsetof(Ebml, id_length), { .u = 4 } },
422 { EBML_ID_DOCTYPE, EBML_STR, 0, 0, offsetof(Ebml, doctype), { .s = "(none)" } },
423 { EBML_ID_DOCTYPEREADVERSION, EBML_UINT, 0, 0, offsetof(Ebml, doctype_version), { .u = 1 } },
424 { EBML_ID_EBMLVERSION, EBML_NONE },
425 { EBML_ID_DOCTYPEVERSION, EBML_NONE },
426 CHILD_OF(ebml_syntax)
429 static EbmlSyntax ebml_syntax[] = {
430 { EBML_ID_HEADER, EBML_NEST, 0, 0, 0, { .n = ebml_header } },
431 { MATROSKA_ID_SEGMENT, EBML_STOP },
435 static EbmlSyntax matroska_info[] = {
436 { MATROSKA_ID_TIMECODESCALE, EBML_UINT, 0, 0, offsetof(MatroskaDemuxContext, time_scale), { .u = 1000000 } },
437 { MATROSKA_ID_DURATION, EBML_FLOAT, 0, 0, offsetof(MatroskaDemuxContext, duration) },
438 { MATROSKA_ID_TITLE, EBML_UTF8, 0, 0, offsetof(MatroskaDemuxContext, title) },
439 { MATROSKA_ID_WRITINGAPP, EBML_NONE },
440 { MATROSKA_ID_MUXINGAPP, EBML_UTF8, 0, 0, offsetof(MatroskaDemuxContext, muxingapp) },
441 { MATROSKA_ID_DATEUTC, EBML_BIN, 0, 0, offsetof(MatroskaDemuxContext, date_utc) },
442 { MATROSKA_ID_SEGMENTUID, EBML_NONE },
443 CHILD_OF(matroska_segment)
446 static EbmlSyntax matroska_mastering_meta[] = {
447 { MATROSKA_ID_VIDEOCOLOR_RX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, r_x) },
448 { MATROSKA_ID_VIDEOCOLOR_RY, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, r_y) },
449 { MATROSKA_ID_VIDEOCOLOR_GX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, g_x) },
450 { MATROSKA_ID_VIDEOCOLOR_GY, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, g_y) },
451 { MATROSKA_ID_VIDEOCOLOR_BX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, b_x) },
452 { MATROSKA_ID_VIDEOCOLOR_BY, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, b_y) },
453 { MATROSKA_ID_VIDEOCOLOR_WHITEX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, white_x) },
454 { MATROSKA_ID_VIDEOCOLOR_WHITEY, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, white_y) },
455 { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMIN, EBML_FLOAT, 1, 0, offsetof(MatroskaMasteringMeta, min_luminance) },
456 { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMAX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, max_luminance) },
457 CHILD_OF(matroska_track_video_color)
460 static EbmlSyntax matroska_track_video_color[] = {
461 { MATROSKA_ID_VIDEOCOLORMATRIXCOEFF, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, matrix_coefficients), { .u = AVCOL_SPC_UNSPECIFIED } },
462 { MATROSKA_ID_VIDEOCOLORBITSPERCHANNEL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, bits_per_channel), { .u = 0 } },
463 { MATROSKA_ID_VIDEOCOLORCHROMASUBHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_horz) },
464 { MATROSKA_ID_VIDEOCOLORCHROMASUBVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_vert) },
465 { MATROSKA_ID_VIDEOCOLORCBSUBHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, cb_sub_horz) },
466 { MATROSKA_ID_VIDEOCOLORCBSUBVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, cb_sub_vert) },
467 { MATROSKA_ID_VIDEOCOLORCHROMASITINGHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_siting_horz), { .u = MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED } },
468 { MATROSKA_ID_VIDEOCOLORCHROMASITINGVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_siting_vert), { .u = MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED } },
469 { MATROSKA_ID_VIDEOCOLORRANGE, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, range), { .u = AVCOL_RANGE_UNSPECIFIED } },
470 { MATROSKA_ID_VIDEOCOLORTRANSFERCHARACTERISTICS, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, transfer_characteristics), { .u = AVCOL_TRC_UNSPECIFIED } },
471 { MATROSKA_ID_VIDEOCOLORPRIMARIES, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, primaries), { .u = AVCOL_PRI_UNSPECIFIED } },
472 { MATROSKA_ID_VIDEOCOLORMAXCLL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, max_cll) },
473 { MATROSKA_ID_VIDEOCOLORMAXFALL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, max_fall) },
474 { MATROSKA_ID_VIDEOCOLORMASTERINGMETA, EBML_NEST, 0, 0, offsetof(MatroskaTrackVideoColor, mastering_meta), { .n = matroska_mastering_meta } },
475 CHILD_OF(matroska_track_video)
478 static EbmlSyntax matroska_track_video_projection[] = {
479 { MATROSKA_ID_VIDEOPROJECTIONTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoProjection, type), { .u = MATROSKA_VIDEO_PROJECTION_TYPE_RECTANGULAR } },
480 { MATROSKA_ID_VIDEOPROJECTIONPRIVATE, EBML_BIN, 0, 0, offsetof(MatroskaTrackVideoProjection, private) },
481 { MATROSKA_ID_VIDEOPROJECTIONPOSEYAW, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, yaw), { .f = 0.0 } },
482 { MATROSKA_ID_VIDEOPROJECTIONPOSEPITCH, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, pitch), { .f = 0.0 } },
483 { MATROSKA_ID_VIDEOPROJECTIONPOSEROLL, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, roll), { .f = 0.0 } },
484 CHILD_OF(matroska_track_video)
487 static EbmlSyntax matroska_track_video[] = {
488 { MATROSKA_ID_VIDEOFRAMERATE, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideo, frame_rate) },
489 { MATROSKA_ID_VIDEODISPLAYWIDTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_width), { .u=-1 } },
490 { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_height), { .u=-1 } },
491 { MATROSKA_ID_VIDEOPIXELWIDTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, pixel_width) },
492 { MATROSKA_ID_VIDEOPIXELHEIGHT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, pixel_height) },
493 { MATROSKA_ID_VIDEOCOLORSPACE, EBML_BIN, 0, 0, offsetof(MatroskaTrackVideo, color_space) },
494 { MATROSKA_ID_VIDEOALPHAMODE, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, alpha_mode), { .u = 0 } },
495 { MATROSKA_ID_VIDEOCOLOR, EBML_NEST, 0, sizeof(MatroskaTrackVideoColor), offsetof(MatroskaTrackVideo, color), { .n = matroska_track_video_color } },
496 { MATROSKA_ID_VIDEOPROJECTION, EBML_NEST, 0, 0, offsetof(MatroskaTrackVideo, projection), { .n = matroska_track_video_projection } },
497 { MATROSKA_ID_VIDEOPIXELCROPB, EBML_NONE },
498 { MATROSKA_ID_VIDEOPIXELCROPT, EBML_NONE },
499 { MATROSKA_ID_VIDEOPIXELCROPL, EBML_NONE },
500 { MATROSKA_ID_VIDEOPIXELCROPR, EBML_NONE },
501 { MATROSKA_ID_VIDEODISPLAYUNIT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_unit), { .u= MATROSKA_VIDEO_DISPLAYUNIT_PIXELS } },
502 { MATROSKA_ID_VIDEOFLAGINTERLACED, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, interlaced), { .u = MATROSKA_VIDEO_INTERLACE_FLAG_UNDETERMINED } },
503 { MATROSKA_ID_VIDEOFIELDORDER, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, field_order), { .u = MATROSKA_VIDEO_FIELDORDER_UNDETERMINED } },
504 { MATROSKA_ID_VIDEOSTEREOMODE, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, stereo_mode), { .u = MATROSKA_VIDEO_STEREOMODE_TYPE_NB } },
505 { MATROSKA_ID_VIDEOASPECTRATIO, EBML_NONE },
506 CHILD_OF(matroska_track)
509 static EbmlSyntax matroska_track_audio[] = {
510 { MATROSKA_ID_AUDIOSAMPLINGFREQ, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackAudio, samplerate), { .f = 8000.0 } },
511 { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackAudio, out_samplerate) },
512 { MATROSKA_ID_AUDIOBITDEPTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackAudio, bitdepth) },
513 { MATROSKA_ID_AUDIOCHANNELS, EBML_UINT, 0, 0, offsetof(MatroskaTrackAudio, channels), { .u = 1 } },
514 CHILD_OF(matroska_track)
517 static EbmlSyntax matroska_track_encoding_compression[] = {
518 { MATROSKA_ID_ENCODINGCOMPALGO, EBML_UINT, 0, 0, offsetof(MatroskaTrackCompression, algo), { .u = MATROSKA_TRACK_ENCODING_COMP_ZLIB } },
519 { MATROSKA_ID_ENCODINGCOMPSETTINGS, EBML_BIN, 0, 0, offsetof(MatroskaTrackCompression, settings) },
520 CHILD_OF(matroska_track_encoding)
523 static EbmlSyntax matroska_track_encoding_encryption[] = {
524 { MATROSKA_ID_ENCODINGENCALGO, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncryption,algo), {.u = 0} },
525 { MATROSKA_ID_ENCODINGENCKEYID, EBML_BIN, 0, 0, offsetof(MatroskaTrackEncryption,key_id) },
526 { MATROSKA_ID_ENCODINGENCAESSETTINGS, EBML_NONE },
527 { MATROSKA_ID_ENCODINGSIGALGO, EBML_NONE },
528 { MATROSKA_ID_ENCODINGSIGHASHALGO, EBML_NONE },
529 { MATROSKA_ID_ENCODINGSIGKEYID, EBML_NONE },
530 { MATROSKA_ID_ENCODINGSIGNATURE, EBML_NONE },
531 CHILD_OF(matroska_track_encoding)
533 static EbmlSyntax matroska_track_encoding[] = {
534 { MATROSKA_ID_ENCODINGSCOPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncoding, scope), { .u = 1 } },
535 { MATROSKA_ID_ENCODINGTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncoding, type), { .u = 0 } },
536 { MATROSKA_ID_ENCODINGCOMPRESSION, EBML_NEST, 0, 0, offsetof(MatroskaTrackEncoding, compression), { .n = matroska_track_encoding_compression } },
537 { MATROSKA_ID_ENCODINGENCRYPTION, EBML_NEST, 0, 0, offsetof(MatroskaTrackEncoding, encryption), { .n = matroska_track_encoding_encryption } },
538 { MATROSKA_ID_ENCODINGORDER, EBML_NONE },
539 CHILD_OF(matroska_track_encodings)
542 static EbmlSyntax matroska_track_encodings[] = {
543 { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, 0, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack, encodings), { .n = matroska_track_encoding } },
544 CHILD_OF(matroska_track)
547 static EbmlSyntax matroska_track_plane[] = {
548 { MATROSKA_ID_TRACKPLANEUID, EBML_UINT, 0, 0, offsetof(MatroskaTrackPlane,uid) },
549 { MATROSKA_ID_TRACKPLANETYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackPlane,type) },
550 CHILD_OF(matroska_track_combine_planes)
553 static EbmlSyntax matroska_track_combine_planes[] = {
554 { MATROSKA_ID_TRACKPLANE, EBML_NEST, 0, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n = matroska_track_plane} },
555 CHILD_OF(matroska_track_operation)
558 static EbmlSyntax matroska_track_operation[] = {
559 { MATROSKA_ID_TRACKCOMBINEPLANES, EBML_NEST, 0, 0, 0, {.n = matroska_track_combine_planes} },
560 CHILD_OF(matroska_track)
563 static EbmlSyntax matroska_track[] = {
564 { MATROSKA_ID_TRACKNUMBER, EBML_UINT, 0, 0, offsetof(MatroskaTrack, num) },
565 { MATROSKA_ID_TRACKNAME, EBML_UTF8, 0, 0, offsetof(MatroskaTrack, name) },
566 { MATROSKA_ID_TRACKUID, EBML_UINT, 0, 0, offsetof(MatroskaTrack, uid) },
567 { MATROSKA_ID_TRACKTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrack, type) },
568 { MATROSKA_ID_CODECID, EBML_STR, 0, 0, offsetof(MatroskaTrack, codec_id) },
569 { MATROSKA_ID_CODECPRIVATE, EBML_BIN, 0, 0, offsetof(MatroskaTrack, codec_priv) },
570 { MATROSKA_ID_CODECDELAY, EBML_UINT, 0, 0, offsetof(MatroskaTrack, codec_delay), { .u = 0 } },
571 { MATROSKA_ID_TRACKLANGUAGE, EBML_STR, 0, 0, offsetof(MatroskaTrack, language), { .s = "eng" } },
572 { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, 0, offsetof(MatroskaTrack, default_duration) },
573 { MATROSKA_ID_TRACKTIMECODESCALE, EBML_FLOAT, 0, 0, offsetof(MatroskaTrack, time_scale), { .f = 1.0 } },
574 { MATROSKA_ID_TRACKFLAGDEFAULT, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_default), { .u = 1 } },
575 { MATROSKA_ID_TRACKFLAGFORCED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_forced), { .u = 0 } },
576 { MATROSKA_ID_TRACKVIDEO, EBML_NEST, 0, 0, offsetof(MatroskaTrack, video), { .n = matroska_track_video } },
577 { MATROSKA_ID_TRACKAUDIO, EBML_NEST, 0, 0, offsetof(MatroskaTrack, audio), { .n = matroska_track_audio } },
578 { MATROSKA_ID_TRACKOPERATION, EBML_NEST, 0, 0, offsetof(MatroskaTrack, operation), { .n = matroska_track_operation } },
579 { MATROSKA_ID_TRACKCONTENTENCODINGS, EBML_NEST, 0, 0, 0, { .n = matroska_track_encodings } },
580 { MATROSKA_ID_TRACKMAXBLKADDID, EBML_UINT, 0, 0, offsetof(MatroskaTrack, max_block_additional_id), { .u = 0 } },
581 { MATROSKA_ID_SEEKPREROLL, EBML_UINT, 0, 0, offsetof(MatroskaTrack, seek_preroll), { .u = 0 } },
582 { MATROSKA_ID_TRACKFLAGENABLED, EBML_NONE },
583 { MATROSKA_ID_TRACKFLAGLACING, EBML_NONE },
584 { MATROSKA_ID_CODECNAME, EBML_NONE },
585 { MATROSKA_ID_CODECDECODEALL, EBML_NONE },
586 { MATROSKA_ID_CODECINFOURL, EBML_NONE },
587 { MATROSKA_ID_CODECDOWNLOADURL, EBML_NONE },
588 { MATROSKA_ID_TRACKMINCACHE, EBML_NONE },
589 { MATROSKA_ID_TRACKMAXCACHE, EBML_NONE },
590 CHILD_OF(matroska_tracks)
593 static EbmlSyntax matroska_tracks[] = {
594 { MATROSKA_ID_TRACKENTRY, EBML_NEST, 0, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext, tracks), { .n = matroska_track } },
595 CHILD_OF(matroska_segment)
598 static EbmlSyntax matroska_attachment[] = {
599 { MATROSKA_ID_FILEUID, EBML_UINT, 0, 0, offsetof(MatroskaAttachment, uid) },
600 { MATROSKA_ID_FILENAME, EBML_UTF8, 0, 0, offsetof(MatroskaAttachment, filename) },
601 { MATROSKA_ID_FILEMIMETYPE, EBML_STR, 0, 0, offsetof(MatroskaAttachment, mime) },
602 { MATROSKA_ID_FILEDATA, EBML_BIN, 0, 0, offsetof(MatroskaAttachment, bin) },
603 { MATROSKA_ID_FILEDESC, EBML_UTF8, 0, 0, offsetof(MatroskaAttachment, description) },
604 CHILD_OF(matroska_attachments)
607 static EbmlSyntax matroska_attachments[] = {
608 { MATROSKA_ID_ATTACHEDFILE, EBML_NEST, 0, sizeof(MatroskaAttachment), offsetof(MatroskaDemuxContext, attachments), { .n = matroska_attachment } },
609 CHILD_OF(matroska_segment)
612 static EbmlSyntax matroska_chapter_display[] = {
613 { MATROSKA_ID_CHAPSTRING, EBML_UTF8, 0, 0, offsetof(MatroskaChapter, title) },
614 { MATROSKA_ID_CHAPLANG, EBML_NONE },
615 { MATROSKA_ID_CHAPCOUNTRY, EBML_NONE },
616 CHILD_OF(matroska_chapter_entry)
619 static EbmlSyntax matroska_chapter_entry[] = {
620 { MATROSKA_ID_CHAPTERTIMESTART, EBML_UINT, 0, 0, offsetof(MatroskaChapter, start), { .u = AV_NOPTS_VALUE } },
621 { MATROSKA_ID_CHAPTERTIMEEND, EBML_UINT, 0, 0, offsetof(MatroskaChapter, end), { .u = AV_NOPTS_VALUE } },
622 { MATROSKA_ID_CHAPTERUID, EBML_UINT, 0, 0, offsetof(MatroskaChapter, uid) },
623 { MATROSKA_ID_CHAPTERDISPLAY, EBML_NEST, 0, 0, 0, { .n = matroska_chapter_display } },
624 { MATROSKA_ID_CHAPTERFLAGHIDDEN, EBML_NONE },
625 { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
626 { MATROSKA_ID_CHAPTERPHYSEQUIV, EBML_NONE },
627 { MATROSKA_ID_CHAPTERATOM, EBML_NONE },
628 CHILD_OF(matroska_chapter)
631 static EbmlSyntax matroska_chapter[] = {
632 { MATROSKA_ID_CHAPTERATOM, EBML_NEST, 0, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext, chapters), { .n = matroska_chapter_entry } },
633 { MATROSKA_ID_EDITIONUID, EBML_NONE },
634 { MATROSKA_ID_EDITIONFLAGHIDDEN, EBML_NONE },
635 { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
636 { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
637 CHILD_OF(matroska_chapters)
640 static EbmlSyntax matroska_chapters[] = {
641 { MATROSKA_ID_EDITIONENTRY, EBML_NEST, 0, 0, 0, { .n = matroska_chapter } },
642 CHILD_OF(matroska_segment)
645 static EbmlSyntax matroska_index_pos[] = {
646 { MATROSKA_ID_CUETRACK, EBML_UINT, 0, 0, offsetof(MatroskaIndexPos, track) },
647 { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, 0, offsetof(MatroskaIndexPos, pos) },
648 { MATROSKA_ID_CUERELATIVEPOSITION,EBML_NONE },
649 { MATROSKA_ID_CUEDURATION, EBML_NONE },
650 { MATROSKA_ID_CUEBLOCKNUMBER, EBML_NONE },
651 CHILD_OF(matroska_index_entry)
654 static EbmlSyntax matroska_index_entry[] = {
655 { MATROSKA_ID_CUETIME, EBML_UINT, 0, 0, offsetof(MatroskaIndex, time) },
656 { MATROSKA_ID_CUETRACKPOSITION, EBML_NEST, 0, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex, pos), { .n = matroska_index_pos } },
657 CHILD_OF(matroska_index)
660 static EbmlSyntax matroska_index[] = {
661 { MATROSKA_ID_POINTENTRY, EBML_NEST, 0, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext, index), { .n = matroska_index_entry } },
662 CHILD_OF(matroska_segment)
665 static EbmlSyntax matroska_simpletag[] = {
666 { MATROSKA_ID_TAGNAME, EBML_UTF8, 0, 0, offsetof(MatroskaTag, name) },
667 { MATROSKA_ID_TAGSTRING, EBML_UTF8, 0, 0, offsetof(MatroskaTag, string) },
668 { MATROSKA_ID_TAGLANG, EBML_STR, 0, 0, offsetof(MatroskaTag, lang), { .s = "und" } },
669 { MATROSKA_ID_TAGDEFAULT, EBML_UINT, 0, 0, offsetof(MatroskaTag, def) },
670 { MATROSKA_ID_TAGDEFAULT_BUG, EBML_UINT, 0, 0, offsetof(MatroskaTag, def) },
671 { MATROSKA_ID_SIMPLETAG, EBML_NEST, 0, sizeof(MatroskaTag), offsetof(MatroskaTag, sub), { .n = matroska_simpletag } },
672 CHILD_OF(matroska_tag)
675 static EbmlSyntax matroska_tagtargets[] = {
676 { MATROSKA_ID_TAGTARGETS_TYPE, EBML_STR, 0, 0, offsetof(MatroskaTagTarget, type) },
677 { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, typevalue), { .u = 50 } },
678 { MATROSKA_ID_TAGTARGETS_TRACKUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, trackuid), { .u = 0 } },
679 { MATROSKA_ID_TAGTARGETS_CHAPTERUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, chapteruid), { .u = 0 } },
680 { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, attachuid), { .u = 0 } },
681 CHILD_OF(matroska_tag)
684 static EbmlSyntax matroska_tag[] = {
685 { MATROSKA_ID_SIMPLETAG, EBML_NEST, 0, sizeof(MatroskaTag), offsetof(MatroskaTags, tag), { .n = matroska_simpletag } },
686 { MATROSKA_ID_TAGTARGETS, EBML_NEST, 0, 0, offsetof(MatroskaTags, target), { .n = matroska_tagtargets } },
687 CHILD_OF(matroska_tags)
690 static EbmlSyntax matroska_tags[] = {
691 { MATROSKA_ID_TAG, EBML_NEST, 0, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext, tags), { .n = matroska_tag } },
692 CHILD_OF(matroska_segment)
695 static EbmlSyntax matroska_seekhead_entry[] = {
696 { MATROSKA_ID_SEEKID, EBML_UINT, 0, 0, offsetof(MatroskaSeekhead, id) },
697 { MATROSKA_ID_SEEKPOSITION, EBML_UINT, 0, 0, offsetof(MatroskaSeekhead, pos), { .u = -1 } },
698 CHILD_OF(matroska_seekhead)
701 static EbmlSyntax matroska_seekhead[] = {
702 { MATROSKA_ID_SEEKENTRY, EBML_NEST, 0, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext, seekhead), { .n = matroska_seekhead_entry } },
703 CHILD_OF(matroska_segment)
706 static EbmlSyntax matroska_segment[] = {
707 { MATROSKA_ID_CLUSTER, EBML_STOP },
708 { MATROSKA_ID_INFO, EBML_LEVEL1, 0, 0, 0, { .n = matroska_info } },
709 { MATROSKA_ID_TRACKS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_tracks } },
710 { MATROSKA_ID_ATTACHMENTS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_attachments } },
711 { MATROSKA_ID_CHAPTERS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_chapters } },
712 { MATROSKA_ID_CUES, EBML_LEVEL1, 0, 0, 0, { .n = matroska_index } },
713 { MATROSKA_ID_TAGS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_tags } },
714 { MATROSKA_ID_SEEKHEAD, EBML_LEVEL1, 0, 0, 0, { .n = matroska_seekhead } },
715 { 0 } /* We don't want to go back to level 0, so don't add the parent. */
718 static EbmlSyntax matroska_segments[] = {
719 { MATROSKA_ID_SEGMENT, EBML_NEST, 0, 0, 0, { .n = matroska_segment } },
723 static EbmlSyntax matroska_blockmore[] = {
724 { MATROSKA_ID_BLOCKADDID, EBML_UINT, 0, 0, offsetof(MatroskaBlock,additional_id), { .u = 1 } },
725 { MATROSKA_ID_BLOCKADDITIONAL, EBML_BIN, 0, 0, offsetof(MatroskaBlock,additional) },
726 CHILD_OF(matroska_blockadditions)
729 static EbmlSyntax matroska_blockadditions[] = {
730 { MATROSKA_ID_BLOCKMORE, EBML_NEST, 0, 0, 0, {.n = matroska_blockmore} },
731 CHILD_OF(matroska_blockgroup)
734 static EbmlSyntax matroska_blockgroup[] = {
735 { MATROSKA_ID_BLOCK, EBML_BIN, 0, 0, offsetof(MatroskaBlock, bin) },
736 { MATROSKA_ID_BLOCKADDITIONS, EBML_NEST, 0, 0, 0, { .n = matroska_blockadditions} },
737 { MATROSKA_ID_BLOCKDURATION, EBML_UINT, 0, 0, offsetof(MatroskaBlock, duration) },
738 { MATROSKA_ID_DISCARDPADDING, EBML_SINT, 0, 0, offsetof(MatroskaBlock, discard_padding) },
739 { MATROSKA_ID_BLOCKREFERENCE, EBML_SINT, 1, 0, offsetof(MatroskaBlock, reference) },
740 { MATROSKA_ID_CODECSTATE, EBML_NONE },
741 { 1, EBML_UINT, 0, 0, offsetof(MatroskaBlock, non_simple), { .u = 1 } },
742 CHILD_OF(matroska_cluster_parsing)
745 // The following array contains SimpleBlock and BlockGroup twice
746 // in order to reuse the other values for matroska_cluster_enter.
747 static EbmlSyntax matroska_cluster_parsing[] = {
748 { MATROSKA_ID_SIMPLEBLOCK, EBML_BIN, 0, 0, offsetof(MatroskaBlock, bin) },
749 { MATROSKA_ID_BLOCKGROUP, EBML_NEST, 0, 0, 0, { .n = matroska_blockgroup } },
750 { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, 0, offsetof(MatroskaCluster, timecode) },
751 { MATROSKA_ID_SIMPLEBLOCK, EBML_STOP },
752 { MATROSKA_ID_BLOCKGROUP, EBML_STOP },
753 { MATROSKA_ID_CLUSTERPOSITION, EBML_NONE },
754 { MATROSKA_ID_CLUSTERPREVSIZE, EBML_NONE },
755 CHILD_OF(matroska_segment)
758 static EbmlSyntax matroska_cluster_enter[] = {
759 { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, 0, { .n = &matroska_cluster_parsing[2] } },
764 static const CodecMime mkv_image_mime_tags[] = {
765 {"image/gif" , AV_CODEC_ID_GIF},
766 {"image/jpeg" , AV_CODEC_ID_MJPEG},
767 {"image/png" , AV_CODEC_ID_PNG},
768 {"image/tiff" , AV_CODEC_ID_TIFF},
770 {"" , AV_CODEC_ID_NONE}
773 static const CodecMime mkv_mime_tags[] = {
774 {"text/plain" , AV_CODEC_ID_TEXT},
775 {"application/x-truetype-font", AV_CODEC_ID_TTF},
776 {"application/x-font" , AV_CODEC_ID_TTF},
777 {"application/vnd.ms-opentype", AV_CODEC_ID_OTF},
778 {"binary" , AV_CODEC_ID_BIN_DATA},
780 {"" , AV_CODEC_ID_NONE}
783 static const char *const matroska_doctypes[] = { "matroska", "webm" };
785 static int matroska_read_close(AVFormatContext *s);
788 * This function prepares the status for parsing of level 1 elements.
790 static int matroska_reset_status(MatroskaDemuxContext *matroska,
791 uint32_t id, int64_t position)
794 int64_t err = avio_seek(matroska->ctx->pb, position, SEEK_SET);
799 matroska->current_id = id;
800 matroska->num_levels = 1;
801 matroska->unknown_count = 0;
802 matroska->resync_pos = avio_tell(matroska->ctx->pb);
804 matroska->resync_pos -= (av_log2(id) + 7) / 8;
809 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
811 AVIOContext *pb = matroska->ctx->pb;
814 /* Try to seek to the last position to resync from. If this doesn't work,
815 * we resync from the earliest position available: The start of the buffer. */
816 if (last_pos < avio_tell(pb) && avio_seek(pb, last_pos + 1, SEEK_SET) < 0) {
817 av_log(matroska->ctx, AV_LOG_WARNING,
818 "Seek to desired resync point failed. Seeking to "
819 "earliest point available instead.\n");
820 avio_seek(pb, FFMAX(avio_tell(pb) + (pb->buffer - pb->buf_ptr),
821 last_pos + 1), SEEK_SET);
826 // try to find a toplevel element
827 while (!avio_feof(pb)) {
828 if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
829 id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
830 id == MATROSKA_ID_SEEKHEAD || id == MATROSKA_ID_ATTACHMENTS ||
831 id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS) {
832 /* Prepare the context for parsing of a level 1 element. */
833 matroska_reset_status(matroska, id, -1);
834 /* Given that we are here means that an error has occurred,
835 * so treat the segment as unknown length in order not to
836 * discard valid data that happens to be beyond the designated
837 * end of the segment. */
838 matroska->levels[0].length = EBML_UNKNOWN_LENGTH;
841 id = (id << 8) | avio_r8(pb);
845 return pb->error ? pb->error : AVERROR_EOF;
849 * Read: an "EBML number", which is defined as a variable-length
850 * array of bytes. The first byte indicates the length by giving a
851 * number of 0-bits followed by a one. The position of the first
852 * "one" bit inside the first byte indicates the length of this
854 * Returns: number of bytes read, < 0 on error
856 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
857 int max_size, uint64_t *number, int eof_forbidden)
863 /* The first byte tells us the length in bytes - except when it is zero. */
868 /* get the length of the EBML number */
869 read = 8 - ff_log2_tab[total];
871 if (!total || read > max_size) {
872 pos = avio_tell(pb) - 1;
874 av_log(matroska->ctx, AV_LOG_ERROR,
875 "0x00 at pos %"PRId64" (0x%"PRIx64") invalid as first byte "
876 "of an EBML number\n", pos, pos);
878 av_log(matroska->ctx, AV_LOG_ERROR,
879 "Length %d indicated by an EBML number's first byte 0x%02x "
880 "at pos %"PRId64" (0x%"PRIx64") exceeds max length %d.\n",
881 read, (uint8_t) total, pos, pos, max_size);
883 return AVERROR_INVALIDDATA;
886 /* read out length */
887 total ^= 1 << ff_log2_tab[total];
889 total = (total << 8) | avio_r8(pb);
891 if (pb->eof_reached) {
903 av_log(matroska->ctx, AV_LOG_ERROR,
904 "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
909 av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
910 "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
917 * Read a EBML length value.
918 * This needs special handling for the "unknown length" case which has multiple
921 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
924 int res = ebml_read_num(matroska, pb, 8, number, 1);
925 if (res > 0 && *number + 1 == 1ULL << (7 * res))
926 *number = EBML_UNKNOWN_LENGTH;
931 * Read the next element as an unsigned int.
932 * Returns NEEDS_CHECKING unless size == 0.
934 static int ebml_read_uint(AVIOContext *pb, int size,
935 uint64_t default_value, uint64_t *num)
940 *num = default_value;
943 /* big-endian ordering; build up number */
946 *num = (*num << 8) | avio_r8(pb);
948 return NEEDS_CHECKING;
952 * Read the next element as a signed int.
953 * Returns NEEDS_CHECKING unless size == 0.
955 static int ebml_read_sint(AVIOContext *pb, int size,
956 int64_t default_value, int64_t *num)
961 *num = default_value;
964 *num = sign_extend(avio_r8(pb), 8);
966 /* big-endian ordering; build up number */
968 *num = ((uint64_t)*num << 8) | avio_r8(pb);
971 return NEEDS_CHECKING;
975 * Read the next element as a float.
976 * Returns 0 if size == 0, NEEDS_CHECKING or < 0 on obvious failure.
978 static int ebml_read_float(AVIOContext *pb, int size,
979 double default_value, double *num)
982 *num = default_value;
984 } else if (size == 4) {
985 *num = av_int2float(avio_rb32(pb));
986 } else if (size == 8) {
987 *num = av_int2double(avio_rb64(pb));
989 return AVERROR_INVALIDDATA;
991 return NEEDS_CHECKING;
995 * Read the next element as an ASCII string.
996 * 0 is success, < 0 or NEEDS_CHECKING is failure.
998 static int ebml_read_ascii(AVIOContext *pb, int size,
999 const char *default_value, char **str)
1004 if (size == 0 && default_value) {
1005 res = av_strdup(default_value);
1007 return AVERROR(ENOMEM);
1009 /* EBML strings are usually not 0-terminated, so we allocate one
1010 * byte more, read the string and NULL-terminate it ourselves. */
1011 if (!(res = av_malloc(size + 1)))
1012 return AVERROR(ENOMEM);
1013 if ((ret = avio_read(pb, (uint8_t *) res, size)) != size) {
1015 return ret < 0 ? ret : NEEDS_CHECKING;
1026 * Read the next element as binary data.
1027 * 0 is success, < 0 or NEEDS_CHECKING is failure.
1029 static int ebml_read_binary(AVIOContext *pb, int length,
1030 int64_t pos, EbmlBin *bin)
1034 ret = av_buffer_realloc(&bin->buf, length + AV_INPUT_BUFFER_PADDING_SIZE);
1037 memset(bin->buf->data + length, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1039 bin->data = bin->buf->data;
1042 if ((ret = avio_read(pb, bin->data, length)) != length) {
1043 av_buffer_unref(&bin->buf);
1046 return ret < 0 ? ret : NEEDS_CHECKING;
1053 * Read the next element, but only the header. The contents
1054 * are supposed to be sub-elements which can be read separately.
1055 * 0 is success, < 0 is failure.
1057 static int ebml_read_master(MatroskaDemuxContext *matroska,
1058 uint64_t length, int64_t pos)
1060 MatroskaLevel *level;
1062 if (matroska->num_levels >= EBML_MAX_DEPTH) {
1063 av_log(matroska->ctx, AV_LOG_ERROR,
1064 "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
1065 return AVERROR(ENOSYS);
1068 level = &matroska->levels[matroska->num_levels++];
1070 level->length = length;
1076 * Read a signed "EBML number"
1077 * Return: number of bytes processed, < 0 on error
1079 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
1080 AVIOContext *pb, int64_t *num)
1085 /* read as unsigned number first */
1086 if ((res = ebml_read_num(matroska, pb, 8, &unum, 1)) < 0)
1089 /* make signed (weird way) */
1090 *num = unum - ((1LL << (7 * res - 1)) - 1);
1095 static int ebml_parse(MatroskaDemuxContext *matroska,
1096 EbmlSyntax *syntax, void *data);
1098 static EbmlSyntax *ebml_parse_id(EbmlSyntax *syntax, uint32_t id)
1102 // Whoever touches this should be aware of the duplication
1103 // existing in matroska_cluster_parsing.
1104 for (i = 0; syntax[i].id; i++)
1105 if (id == syntax[i].id)
1111 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
1117 for (int i = 0; syntax[i].id; i++)
1118 switch (syntax[i].type) {
1120 *(uint64_t *) ((char *) data + syntax[i].data_offset) = syntax[i].def.u;
1123 *(int64_t *) ((char *) data + syntax[i].data_offset) = syntax[i].def.i;
1126 *(double *) ((char *) data + syntax[i].data_offset) = syntax[i].def.f;
1130 // the default may be NULL
1131 if (syntax[i].def.s) {
1132 uint8_t **dst = (uint8_t **) ((uint8_t *) data + syntax[i].data_offset);
1133 *dst = av_strdup(syntax[i].def.s);
1135 return AVERROR(ENOMEM);
1140 if (!matroska->levels[matroska->num_levels - 1].length) {
1141 matroska->num_levels--;
1147 res = ebml_parse(matroska, syntax, data);
1150 return res == LEVEL_ENDED ? 0 : res;
1153 static int is_ebml_id_valid(uint32_t id)
1155 // Due to endian nonsense in Matroska, the highest byte with any bits set
1156 // will contain the leading length bit. This bit in turn identifies the
1157 // total byte length of the element by its position within the byte.
1158 unsigned int bits = av_log2(id);
1159 return id && (bits + 7) / 8 == (8 - bits % 8);
1163 * Allocate and return the entry for the level1 element with the given ID. If
1164 * an entry already exists, return the existing entry.
1166 static MatroskaLevel1Element *matroska_find_level1_elem(MatroskaDemuxContext *matroska,
1167 uint32_t id, int64_t pos)
1170 MatroskaLevel1Element *elem;
1172 if (!is_ebml_id_valid(id))
1175 // Some files link to all clusters; useless.
1176 if (id == MATROSKA_ID_CLUSTER)
1179 // There can be multiple SeekHeads and Tags.
1180 for (i = 0; i < matroska->num_level1_elems; i++) {
1181 if (matroska->level1_elems[i].id == id) {
1182 if (matroska->level1_elems[i].pos == pos ||
1183 id != MATROSKA_ID_SEEKHEAD && id != MATROSKA_ID_TAGS)
1184 return &matroska->level1_elems[i];
1188 // Only a completely broken file would have more elements.
1189 if (matroska->num_level1_elems >= FF_ARRAY_ELEMS(matroska->level1_elems)) {
1190 av_log(matroska->ctx, AV_LOG_ERROR, "Too many level1 elements.\n");
1194 elem = &matroska->level1_elems[matroska->num_level1_elems++];
1195 *elem = (MatroskaLevel1Element){.id = id};
1200 static int ebml_parse(MatroskaDemuxContext *matroska,
1201 EbmlSyntax *syntax, void *data)
1203 static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
1204 // Forbid unknown-length EBML_NONE elements.
1205 [EBML_NONE] = EBML_UNKNOWN_LENGTH - 1,
1209 // max. 16 MB for strings
1210 [EBML_STR] = 0x1000000,
1211 [EBML_UTF8] = 0x1000000,
1212 // max. 256 MB for binary data
1213 [EBML_BIN] = 0x10000000,
1214 // no limits for anything else
1216 AVIOContext *pb = matroska->ctx->pb;
1219 int64_t pos = avio_tell(pb), pos_alt;
1220 int res, update_pos = 1, level_check;
1221 MatroskaLevel1Element *level1_elem;
1222 MatroskaLevel *level = matroska->num_levels ? &matroska->levels[matroska->num_levels - 1] : NULL;
1224 if (!matroska->current_id) {
1226 res = ebml_read_num(matroska, pb, 4, &id, 0);
1228 if (pb->eof_reached && res == AVERROR_EOF) {
1229 if (matroska->is_live)
1230 // in live mode, finish parsing if EOF is reached.
1232 if (level && pos == avio_tell(pb)) {
1233 if (level->length == EBML_UNKNOWN_LENGTH) {
1234 // Unknown-length levels automatically end at EOF.
1235 matroska->num_levels--;
1238 av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
1239 "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
1245 matroska->current_id = id | 1 << 7 * res;
1246 pos_alt = pos + res;
1249 pos -= (av_log2(matroska->current_id) + 7) / 8;
1252 id = matroska->current_id;
1254 syntax = ebml_parse_id(syntax, id);
1255 if (!syntax->id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
1256 if (level && level->length == EBML_UNKNOWN_LENGTH) {
1257 // Unknown-length levels end when an element from an upper level
1258 // in the hierarchy is encountered.
1259 while (syntax->def.n) {
1260 syntax = ebml_parse_id(syntax->def.n, id);
1262 matroska->num_levels--;
1268 av_log(matroska->ctx, AV_LOG_DEBUG, "Unknown entry 0x%"PRIX32" at pos. "
1269 "%"PRId64"\n", id, pos);
1270 update_pos = 0; /* Don't update resync_pos as an error might have happened. */
1274 data = (char *) data + syntax->data_offset;
1275 if (syntax->list_elem_size) {
1276 EbmlList *list = data;
1279 if ((unsigned)list->nb_elem + 1 >= UINT_MAX / syntax->list_elem_size)
1280 return AVERROR(ENOMEM);
1281 newelem = av_fast_realloc(list->elem,
1282 &list->alloc_elem_size,
1283 (list->nb_elem + 1) * syntax->list_elem_size);
1285 return AVERROR(ENOMEM);
1286 list->elem = newelem;
1287 data = (char *) list->elem + list->nb_elem * syntax->list_elem_size;
1288 memset(data, 0, syntax->list_elem_size);
1293 if (syntax->type != EBML_STOP) {
1294 matroska->current_id = 0;
1295 if ((res = ebml_read_length(matroska, pb, &length)) < 0)
1300 if (matroska->num_levels > 0) {
1301 if (length != EBML_UNKNOWN_LENGTH &&
1302 level->length != EBML_UNKNOWN_LENGTH) {
1303 uint64_t elem_end = pos_alt + length,
1304 level_end = level->start + level->length;
1306 if (elem_end < level_end) {
1308 } else if (elem_end == level_end) {
1309 level_check = LEVEL_ENDED;
1311 av_log(matroska->ctx, AV_LOG_ERROR,
1312 "Element at 0x%"PRIx64" ending at 0x%"PRIx64" exceeds "
1313 "containing master element ending at 0x%"PRIx64"\n",
1314 pos, elem_end, level_end);
1315 return AVERROR_INVALIDDATA;
1317 } else if (length != EBML_UNKNOWN_LENGTH) {
1319 } else if (level->length != EBML_UNKNOWN_LENGTH) {
1320 av_log(matroska->ctx, AV_LOG_ERROR, "Unknown-sized element "
1321 "at 0x%"PRIx64" inside parent with finite size\n", pos);
1322 return AVERROR_INVALIDDATA;
1325 if (id != MATROSKA_ID_CLUSTER && (syntax->type == EBML_LEVEL1
1326 || syntax->type == EBML_NEST)) {
1327 // According to the current specifications only clusters and
1328 // segments are allowed to be unknown-length. We also accept
1329 // other unknown-length master elements.
1330 av_log(matroska->ctx, AV_LOG_WARNING,
1331 "Found unknown-length element 0x%"PRIX32" other than "
1332 "a cluster at 0x%"PRIx64". Spec-incompliant, but "
1333 "parsing will nevertheless be attempted.\n", id, pos);
1340 if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
1341 if (length != EBML_UNKNOWN_LENGTH) {
1342 av_log(matroska->ctx, AV_LOG_ERROR,
1343 "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for element "
1344 "with ID 0x%"PRIX32" at 0x%"PRIx64"\n",
1345 length, max_lengths[syntax->type], id, pos);
1346 } else if (syntax->type != EBML_NONE) {
1347 av_log(matroska->ctx, AV_LOG_ERROR,
1348 "Element with ID 0x%"PRIX32" at pos. 0x%"PRIx64" has "
1349 "unknown length, yet the length of an element of its "
1350 "type must be known.\n", id, pos);
1352 av_log(matroska->ctx, AV_LOG_ERROR,
1353 "Found unknown-length element with ID 0x%"PRIX32" at "
1354 "pos. 0x%"PRIx64" for which no syntax for parsing is "
1355 "available.\n", id, pos);
1357 return AVERROR_INVALIDDATA;
1360 if (!(pb->seekable & AVIO_SEEKABLE_NORMAL)) {
1361 // Loosing sync will likely manifest itself as encountering unknown
1362 // elements which are not reliably distinguishable from elements
1363 // belonging to future extensions of the format.
1364 // We use a heuristic to detect such situations: If the current
1365 // element is not expected at the current syntax level and there
1366 // were only a few unknown elements in a row, then the element is
1367 // skipped or considered defective based upon the length of the
1368 // current element (i.e. how much would be skipped); if there were
1369 // more than a few skipped elements in a row and skipping the current
1370 // element would lead us more than SKIP_THRESHOLD away from the last
1371 // known good position, then it is inferred that an error occurred.
1372 // The dependency on the number of unknown elements in a row exists
1373 // because the distance to the last known good position is
1374 // automatically big if the last parsed element was big.
1375 // In both cases, each unknown element is considered equivalent to
1376 // UNKNOWN_EQUIV of skipped bytes for the check.
1377 // The whole check is only done for non-seekable output, because
1378 // in this situation skipped data can't simply be rechecked later.
1379 // This is especially important when using unkown length elements
1380 // as the check for whether a child exceeds its containing master
1381 // element is not effective in this situation.
1383 matroska->unknown_count = 0;
1385 int64_t dist = length + UNKNOWN_EQUIV * matroska->unknown_count++;
1387 if (matroska->unknown_count > 3)
1388 dist += pos_alt - matroska->resync_pos;
1390 if (dist > SKIP_THRESHOLD) {
1391 av_log(matroska->ctx, AV_LOG_ERROR,
1392 "Unknown element %"PRIX32" at pos. 0x%"PRIx64" with "
1393 "length 0x%"PRIx64" considered as invalid data. Last "
1394 "known good position 0x%"PRIx64", %d unknown elements"
1395 " in a row\n", id, pos, length, matroska->resync_pos,
1396 matroska->unknown_count);
1397 return AVERROR_INVALIDDATA;
1402 if (update_pos > 0) {
1403 // We have found an element that is allowed at this place
1404 // in the hierarchy and it passed all checks, so treat the beginning
1405 // of the element as the "last known good" position.
1406 matroska->resync_pos = pos;
1409 if (!data && length != EBML_UNKNOWN_LENGTH)
1413 switch (syntax->type) {
1415 res = ebml_read_uint(pb, length, syntax->def.u, data);
1418 res = ebml_read_sint(pb, length, syntax->def.i, data);
1421 res = ebml_read_float(pb, length, syntax->def.f, data);
1425 res = ebml_read_ascii(pb, length, syntax->def.s, data);
1428 res = ebml_read_binary(pb, length, pos_alt, data);
1432 if ((res = ebml_read_master(matroska, length, pos_alt)) < 0)
1434 if (id == MATROSKA_ID_SEGMENT)
1435 matroska->segment_start = pos_alt;
1436 if (id == MATROSKA_ID_CUES)
1437 matroska->cues_parsing_deferred = 0;
1438 if (syntax->type == EBML_LEVEL1 &&
1439 (level1_elem = matroska_find_level1_elem(matroska, syntax->id, pos))) {
1440 if (!level1_elem->pos) {
1441 // Zero is not a valid position for a level 1 element.
1442 level1_elem->pos = pos;
1443 } else if (level1_elem->pos != pos)
1444 av_log(matroska->ctx, AV_LOG_ERROR, "Duplicate element\n");
1445 level1_elem->parsed = 1;
1447 if (res = ebml_parse_nest(matroska, syntax->def.n, data))
1456 if (ffio_limit(pb, length) != length) {
1457 // ffio_limit emits its own error message,
1458 // so we don't have to.
1459 return AVERROR(EIO);
1461 if ((res2 = avio_skip(pb, length - 1)) >= 0) {
1462 // avio_skip might take us past EOF. We check for this
1463 // by skipping only length - 1 bytes, reading a byte and
1464 // checking the error flags. This is done in order to check
1465 // that the element has been properly skipped even when
1466 // no filesize (that ffio_limit relies on) is available.
1468 res = NEEDS_CHECKING;
1475 if (res == NEEDS_CHECKING) {
1476 if (pb->eof_reached) {
1485 if (res == AVERROR_INVALIDDATA)
1486 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
1487 else if (res == AVERROR(EIO))
1488 av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
1489 else if (res == AVERROR_EOF) {
1490 av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely\n");
1498 if (syntax->is_counted && data) {
1499 CountedElement *elem = data;
1500 if (elem->count != UINT_MAX)
1504 if (level_check == LEVEL_ENDED && matroska->num_levels) {
1505 level = &matroska->levels[matroska->num_levels - 1];
1506 pos = avio_tell(pb);
1508 // Given that pos >= level->start no check for
1509 // level->length != EBML_UNKNOWN_LENGTH is necessary.
1510 while (matroska->num_levels && pos == level->start + level->length) {
1511 matroska->num_levels--;
1519 static void ebml_free(EbmlSyntax *syntax, void *data)
1522 for (i = 0; syntax[i].id; i++) {
1523 void *data_off = (char *) data + syntax[i].data_offset;
1524 switch (syntax[i].type) {
1530 av_buffer_unref(&((EbmlBin *) data_off)->buf);
1534 if (syntax[i].list_elem_size) {
1535 EbmlList *list = data_off;
1536 char *ptr = list->elem;
1537 for (j = 0; j < list->nb_elem;
1538 j++, ptr += syntax[i].list_elem_size)
1539 ebml_free(syntax[i].def.n, ptr);
1540 av_freep(&list->elem);
1542 list->alloc_elem_size = 0;
1544 ebml_free(syntax[i].def.n, data_off);
1554 static int matroska_probe(const AVProbeData *p)
1557 int len_mask = 0x80, size = 1, n = 1, i;
1560 if (AV_RB32(p->buf) != EBML_ID_HEADER)
1563 /* length of header */
1565 while (size <= 8 && !(total & len_mask)) {
1571 total &= (len_mask - 1);
1573 total = (total << 8) | p->buf[4 + n++];
1575 if (total + 1 == 1ULL << (7 * size)){
1576 /* Unknown-length header - simply parse the whole buffer. */
1577 total = p->buf_size - 4 - size;
1579 /* Does the probe data contain the whole header? */
1580 if (p->buf_size < 4 + size + total)
1584 /* The header should contain a known document type. For now,
1585 * we don't parse the whole header but simply check for the
1586 * availability of that array of characters inside the header.
1587 * Not fully fool-proof, but good enough. */
1588 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1589 size_t probelen = strlen(matroska_doctypes[i]);
1590 if (total < probelen)
1592 for (n = 4 + size; n <= 4 + size + total - probelen; n++)
1593 if (!memcmp(p->buf + n, matroska_doctypes[i], probelen))
1594 return AVPROBE_SCORE_MAX;
1597 // probably valid EBML header but no recognized doctype
1598 return AVPROBE_SCORE_EXTENSION;
1601 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
1604 MatroskaTrack *tracks = matroska->tracks.elem;
1607 for (i = 0; i < matroska->tracks.nb_elem; i++)
1608 if (tracks[i].num == num)
1611 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %"PRIu64"\n", num);
1615 static int matroska_decode_buffer(uint8_t **buf, int *buf_size,
1616 MatroskaTrack *track)
1618 MatroskaTrackEncoding *encodings = track->encodings.elem;
1619 uint8_t *data = *buf;
1620 int isize = *buf_size;
1621 uint8_t *pkt_data = NULL;
1622 uint8_t av_unused *newpktdata;
1623 int pkt_size = isize;
1627 if (pkt_size >= 10000000U)
1628 return AVERROR_INVALIDDATA;
1630 switch (encodings[0].compression.algo) {
1631 case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
1633 int header_size = encodings[0].compression.settings.size;
1634 uint8_t *header = encodings[0].compression.settings.data;
1636 if (header_size && !header) {
1637 av_log(NULL, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1644 pkt_size = isize + header_size;
1645 pkt_data = av_malloc(pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1647 return AVERROR(ENOMEM);
1649 memcpy(pkt_data, header, header_size);
1650 memcpy(pkt_data + header_size, data, isize);
1654 case MATROSKA_TRACK_ENCODING_COMP_LZO:
1657 olen = pkt_size *= 3;
1658 newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING
1659 + AV_INPUT_BUFFER_PADDING_SIZE);
1661 result = AVERROR(ENOMEM);
1664 pkt_data = newpktdata;
1665 result = av_lzo1x_decode(pkt_data, &olen, data, &insize);
1666 } while (result == AV_LZO_OUTPUT_FULL && pkt_size < 10000000);
1668 result = AVERROR_INVALIDDATA;
1675 case MATROSKA_TRACK_ENCODING_COMP_ZLIB:
1677 z_stream zstream = { 0 };
1678 if (inflateInit(&zstream) != Z_OK)
1680 zstream.next_in = data;
1681 zstream.avail_in = isize;
1684 newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1686 inflateEnd(&zstream);
1687 result = AVERROR(ENOMEM);
1690 pkt_data = newpktdata;
1691 zstream.avail_out = pkt_size - zstream.total_out;
1692 zstream.next_out = pkt_data + zstream.total_out;
1693 result = inflate(&zstream, Z_NO_FLUSH);
1694 } while (result == Z_OK && pkt_size < 10000000);
1695 pkt_size = zstream.total_out;
1696 inflateEnd(&zstream);
1697 if (result != Z_STREAM_END) {
1698 if (result == Z_MEM_ERROR)
1699 result = AVERROR(ENOMEM);
1701 result = AVERROR_INVALIDDATA;
1708 case MATROSKA_TRACK_ENCODING_COMP_BZLIB:
1710 bz_stream bzstream = { 0 };
1711 if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1713 bzstream.next_in = data;
1714 bzstream.avail_in = isize;
1717 newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1719 BZ2_bzDecompressEnd(&bzstream);
1720 result = AVERROR(ENOMEM);
1723 pkt_data = newpktdata;
1724 bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1725 bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1726 result = BZ2_bzDecompress(&bzstream);
1727 } while (result == BZ_OK && pkt_size < 10000000);
1728 pkt_size = bzstream.total_out_lo32;
1729 BZ2_bzDecompressEnd(&bzstream);
1730 if (result != BZ_STREAM_END) {
1731 if (result == BZ_MEM_ERROR)
1732 result = AVERROR(ENOMEM);
1734 result = AVERROR_INVALIDDATA;
1741 return AVERROR_INVALIDDATA;
1744 memset(pkt_data + pkt_size, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1747 *buf_size = pkt_size;
1755 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1756 AVDictionary **metadata, char *prefix)
1758 MatroskaTag *tags = list->elem;
1762 for (i = 0; i < list->nb_elem; i++) {
1763 const char *lang = tags[i].lang &&
1764 strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1766 if (!tags[i].name) {
1767 av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1771 snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1773 av_strlcpy(key, tags[i].name, sizeof(key));
1774 if (tags[i].def || !lang) {
1775 av_dict_set(metadata, key, tags[i].string, 0);
1776 if (tags[i].sub.nb_elem)
1777 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1780 av_strlcat(key, "-", sizeof(key));
1781 av_strlcat(key, lang, sizeof(key));
1782 av_dict_set(metadata, key, tags[i].string, 0);
1783 if (tags[i].sub.nb_elem)
1784 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1787 ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1790 static void matroska_convert_tags(AVFormatContext *s)
1792 MatroskaDemuxContext *matroska = s->priv_data;
1793 MatroskaTags *tags = matroska->tags.elem;
1796 for (i = 0; i < matroska->tags.nb_elem; i++) {
1797 if (tags[i].target.attachuid) {
1798 MatroskaAttachment *attachment = matroska->attachments.elem;
1800 for (j = 0; j < matroska->attachments.nb_elem; j++) {
1801 if (attachment[j].uid == tags[i].target.attachuid &&
1802 attachment[j].stream) {
1803 matroska_convert_tag(s, &tags[i].tag,
1804 &attachment[j].stream->metadata, NULL);
1809 av_log(s, AV_LOG_WARNING,
1810 "The tags at index %d refer to a "
1811 "non-existent attachment %"PRId64".\n",
1812 i, tags[i].target.attachuid);
1814 } else if (tags[i].target.chapteruid) {
1815 MatroskaChapter *chapter = matroska->chapters.elem;
1817 for (j = 0; j < matroska->chapters.nb_elem; j++) {
1818 if (chapter[j].uid == tags[i].target.chapteruid &&
1819 chapter[j].chapter) {
1820 matroska_convert_tag(s, &tags[i].tag,
1821 &chapter[j].chapter->metadata, NULL);
1826 av_log(s, AV_LOG_WARNING,
1827 "The tags at index %d refer to a non-existent chapter "
1829 i, tags[i].target.chapteruid);
1831 } else if (tags[i].target.trackuid) {
1832 MatroskaTrack *track = matroska->tracks.elem;
1834 for (j = 0; j < matroska->tracks.nb_elem; j++) {
1835 if (track[j].uid == tags[i].target.trackuid &&
1837 matroska_convert_tag(s, &tags[i].tag,
1838 &track[j].stream->metadata, NULL);
1843 av_log(s, AV_LOG_WARNING,
1844 "The tags at index %d refer to a non-existent track "
1846 i, tags[i].target.trackuid);
1849 matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1850 tags[i].target.type);
1855 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska,
1858 uint32_t saved_id = matroska->current_id;
1859 int64_t before_pos = avio_tell(matroska->ctx->pb);
1863 if (avio_seek(matroska->ctx->pb, pos, SEEK_SET) == pos) {
1864 /* We don't want to lose our seekhead level, so we add
1865 * a dummy. This is a crude hack. */
1866 if (matroska->num_levels == EBML_MAX_DEPTH) {
1867 av_log(matroska->ctx, AV_LOG_INFO,
1868 "Max EBML element depth (%d) reached, "
1869 "cannot parse further.\n", EBML_MAX_DEPTH);
1870 ret = AVERROR_INVALIDDATA;
1872 matroska->levels[matroska->num_levels] = (MatroskaLevel) { 0, EBML_UNKNOWN_LENGTH };
1873 matroska->num_levels++;
1874 matroska->current_id = 0;
1876 ret = ebml_parse(matroska, matroska_segment, matroska);
1877 if (ret == LEVEL_ENDED) {
1878 /* This can only happen if the seek brought us beyond EOF. */
1883 /* Seek back - notice that in all instances where this is used
1884 * it is safe to set the level to 1. */
1885 matroska_reset_status(matroska, saved_id, before_pos);
1890 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1892 EbmlList *seekhead_list = &matroska->seekhead;
1895 // we should not do any seeking in the streaming case
1896 if (!(matroska->ctx->pb->seekable & AVIO_SEEKABLE_NORMAL))
1899 for (i = 0; i < seekhead_list->nb_elem; i++) {
1900 MatroskaSeekhead *seekheads = seekhead_list->elem;
1901 uint32_t id = seekheads[i].id;
1902 int64_t pos = seekheads[i].pos + matroska->segment_start;
1903 MatroskaLevel1Element *elem;
1905 if (id != seekheads[i].id || pos < matroska->segment_start)
1908 elem = matroska_find_level1_elem(matroska, id, pos);
1909 if (!elem || elem->parsed)
1914 // defer cues parsing until we actually need cue data.
1915 if (id == MATROSKA_ID_CUES)
1918 if (matroska_parse_seekhead_entry(matroska, pos) < 0) {
1919 // mark index as broken
1920 matroska->cues_parsing_deferred = -1;
1928 static void matroska_add_index_entries(MatroskaDemuxContext *matroska)
1930 EbmlList *index_list;
1931 MatroskaIndex *index;
1932 uint64_t index_scale = 1;
1935 if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
1938 index_list = &matroska->index;
1939 index = index_list->elem;
1940 if (index_list->nb_elem < 2)
1942 if (index[1].time > 1E14 / matroska->time_scale) {
1943 av_log(matroska->ctx, AV_LOG_WARNING, "Dropping apparently-broken index.\n");
1946 for (i = 0; i < index_list->nb_elem; i++) {
1947 EbmlList *pos_list = &index[i].pos;
1948 MatroskaIndexPos *pos = pos_list->elem;
1949 for (j = 0; j < pos_list->nb_elem; j++) {
1950 MatroskaTrack *track = matroska_find_track_by_num(matroska,
1952 if (track && track->stream)
1953 av_add_index_entry(track->stream,
1954 pos[j].pos + matroska->segment_start,
1955 index[i].time / index_scale, 0, 0,
1961 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1964 if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
1967 for (i = 0; i < matroska->num_level1_elems; i++) {
1968 MatroskaLevel1Element *elem = &matroska->level1_elems[i];
1969 if (elem->id == MATROSKA_ID_CUES && !elem->parsed) {
1970 if (matroska_parse_seekhead_entry(matroska, elem->pos) < 0)
1971 matroska->cues_parsing_deferred = -1;
1977 matroska_add_index_entries(matroska);
1980 static int matroska_aac_profile(char *codec_id)
1982 static const char *const aac_profiles[] = { "MAIN", "LC", "SSR" };
1985 for (profile = 0; profile < FF_ARRAY_ELEMS(aac_profiles); profile++)
1986 if (strstr(codec_id, aac_profiles[profile]))
1991 static int matroska_aac_sri(int samplerate)
1995 for (sri = 0; sri < FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1996 if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
2001 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
2003 /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
2004 avpriv_dict_set_timestamp(metadata, "creation_time", date_utc / 1000 + 978307200000000LL);
2007 static int matroska_parse_flac(AVFormatContext *s,
2008 MatroskaTrack *track,
2011 AVStream *st = track->stream;
2012 uint8_t *p = track->codec_priv.data;
2013 int size = track->codec_priv.size;
2015 if (size < 8 + FLAC_STREAMINFO_SIZE || p[4] & 0x7f) {
2016 av_log(s, AV_LOG_WARNING, "Invalid FLAC private data\n");
2017 track->codec_priv.size = 0;
2021 track->codec_priv.size = 8 + FLAC_STREAMINFO_SIZE;
2023 p += track->codec_priv.size;
2024 size -= track->codec_priv.size;
2026 /* parse the remaining metadata blocks if present */
2028 int block_last, block_type, block_size;
2030 flac_parse_block_header(p, &block_last, &block_type, &block_size);
2034 if (block_size > size)
2037 /* check for the channel mask */
2038 if (block_type == FLAC_METADATA_TYPE_VORBIS_COMMENT) {
2039 AVDictionary *dict = NULL;
2040 AVDictionaryEntry *chmask;
2042 ff_vorbis_comment(s, &dict, p, block_size, 0);
2043 chmask = av_dict_get(dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", NULL, 0);
2045 uint64_t mask = strtol(chmask->value, NULL, 0);
2046 if (!mask || mask & ~0x3ffffULL) {
2047 av_log(s, AV_LOG_WARNING,
2048 "Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK\n");
2050 st->codecpar->channel_layout = mask;
2052 av_dict_free(&dict);
2062 static int mkv_field_order(MatroskaDemuxContext *matroska, int64_t field_order)
2064 int minor, micro, bttb = 0;
2066 /* workaround a bug in our Matroska muxer, introduced in version 57.36 alongside
2067 * this function, and fixed in 57.52 */
2068 if (matroska->muxingapp && sscanf(matroska->muxingapp, "Lavf57.%d.%d", &minor, µ) == 2)
2069 bttb = (minor >= 36 && minor <= 51 && micro >= 100);
2071 switch (field_order) {
2072 case MATROSKA_VIDEO_FIELDORDER_PROGRESSIVE:
2073 return AV_FIELD_PROGRESSIVE;
2074 case MATROSKA_VIDEO_FIELDORDER_UNDETERMINED:
2075 return AV_FIELD_UNKNOWN;
2076 case MATROSKA_VIDEO_FIELDORDER_TT:
2078 case MATROSKA_VIDEO_FIELDORDER_BB:
2080 case MATROSKA_VIDEO_FIELDORDER_BT:
2081 return bttb ? AV_FIELD_TB : AV_FIELD_BT;
2082 case MATROSKA_VIDEO_FIELDORDER_TB:
2083 return bttb ? AV_FIELD_BT : AV_FIELD_TB;
2085 return AV_FIELD_UNKNOWN;
2089 static void mkv_stereo_mode_display_mul(int stereo_mode,
2090 int *h_width, int *h_height)
2092 switch (stereo_mode) {
2093 case MATROSKA_VIDEO_STEREOMODE_TYPE_MONO:
2094 case MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_RL:
2095 case MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_LR:
2096 case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_RL:
2097 case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_LR:
2099 case MATROSKA_VIDEO_STEREOMODE_TYPE_RIGHT_LEFT:
2100 case MATROSKA_VIDEO_STEREOMODE_TYPE_LEFT_RIGHT:
2101 case MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_RL:
2102 case MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_LR:
2105 case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTTOM_TOP:
2106 case MATROSKA_VIDEO_STEREOMODE_TYPE_TOP_BOTTOM:
2107 case MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_RL:
2108 case MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_LR:
2114 static int mkv_parse_video_color(AVStream *st, const MatroskaTrack *track) {
2115 const MatroskaTrackVideoColor *color = track->video.color.elem;
2116 const MatroskaMasteringMeta *mastering_meta;
2117 int has_mastering_primaries, has_mastering_luminance;
2119 if (!track->video.color.nb_elem)
2122 mastering_meta = &color->mastering_meta;
2123 // Mastering primaries are CIE 1931 coords, and must be > 0.
2124 has_mastering_primaries =
2125 mastering_meta->r_x > 0 && mastering_meta->r_y > 0 &&
2126 mastering_meta->g_x > 0 && mastering_meta->g_y > 0 &&
2127 mastering_meta->b_x > 0 && mastering_meta->b_y > 0 &&
2128 mastering_meta->white_x > 0 && mastering_meta->white_y > 0;
2129 has_mastering_luminance = mastering_meta->max_luminance >
2130 mastering_meta->min_luminance.el.f &&
2131 mastering_meta->min_luminance.el.f >= 0 &&
2132 mastering_meta->min_luminance.count;
2134 if (color->matrix_coefficients != AVCOL_SPC_RESERVED)
2135 st->codecpar->color_space = color->matrix_coefficients;
2136 if (color->primaries != AVCOL_PRI_RESERVED &&
2137 color->primaries != AVCOL_PRI_RESERVED0)
2138 st->codecpar->color_primaries = color->primaries;
2139 if (color->transfer_characteristics != AVCOL_TRC_RESERVED &&
2140 color->transfer_characteristics != AVCOL_TRC_RESERVED0)
2141 st->codecpar->color_trc = color->transfer_characteristics;
2142 if (color->range != AVCOL_RANGE_UNSPECIFIED &&
2143 color->range <= AVCOL_RANGE_JPEG)
2144 st->codecpar->color_range = color->range;
2145 if (color->chroma_siting_horz != MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED &&
2146 color->chroma_siting_vert != MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED &&
2147 color->chroma_siting_horz < MATROSKA_COLOUR_CHROMASITINGHORZ_NB &&
2148 color->chroma_siting_vert < MATROSKA_COLOUR_CHROMASITINGVERT_NB) {
2149 st->codecpar->chroma_location =
2150 avcodec_chroma_pos_to_enum((color->chroma_siting_horz - 1) << 7,
2151 (color->chroma_siting_vert - 1) << 7);
2153 if (color->max_cll && color->max_fall) {
2156 AVContentLightMetadata *metadata = av_content_light_metadata_alloc(&size);
2158 return AVERROR(ENOMEM);
2159 ret = av_stream_add_side_data(st, AV_PKT_DATA_CONTENT_LIGHT_LEVEL,
2160 (uint8_t *)metadata, size);
2162 av_freep(&metadata);
2165 metadata->MaxCLL = color->max_cll;
2166 metadata->MaxFALL = color->max_fall;
2169 if (has_mastering_primaries || has_mastering_luminance) {
2170 AVMasteringDisplayMetadata *metadata =
2171 (AVMasteringDisplayMetadata*) av_stream_new_side_data(
2172 st, AV_PKT_DATA_MASTERING_DISPLAY_METADATA,
2173 sizeof(AVMasteringDisplayMetadata));
2175 return AVERROR(ENOMEM);
2177 memset(metadata, 0, sizeof(AVMasteringDisplayMetadata));
2178 if (has_mastering_primaries) {
2179 metadata->display_primaries[0][0] = av_d2q(mastering_meta->r_x, INT_MAX);
2180 metadata->display_primaries[0][1] = av_d2q(mastering_meta->r_y, INT_MAX);
2181 metadata->display_primaries[1][0] = av_d2q(mastering_meta->g_x, INT_MAX);
2182 metadata->display_primaries[1][1] = av_d2q(mastering_meta->g_y, INT_MAX);
2183 metadata->display_primaries[2][0] = av_d2q(mastering_meta->b_x, INT_MAX);
2184 metadata->display_primaries[2][1] = av_d2q(mastering_meta->b_y, INT_MAX);
2185 metadata->white_point[0] = av_d2q(mastering_meta->white_x, INT_MAX);
2186 metadata->white_point[1] = av_d2q(mastering_meta->white_y, INT_MAX);
2187 metadata->has_primaries = 1;
2189 if (has_mastering_luminance) {
2190 metadata->max_luminance = av_d2q(mastering_meta->max_luminance, INT_MAX);
2191 metadata->min_luminance = av_d2q(mastering_meta->min_luminance.el.f, INT_MAX);
2192 metadata->has_luminance = 1;
2198 static int mkv_parse_video_projection(AVStream *st, const MatroskaTrack *track,
2201 AVSphericalMapping *spherical;
2202 const MatroskaTrackVideoProjection *mkv_projection = &track->video.projection;
2203 const uint8_t *priv_data = mkv_projection->private.data;
2204 enum AVSphericalProjection projection;
2205 size_t spherical_size;
2206 uint32_t l = 0, t = 0, r = 0, b = 0;
2207 uint32_t padding = 0;
2210 if (mkv_projection->private.size && priv_data[0] != 0) {
2211 av_log(logctx, AV_LOG_WARNING, "Unknown spherical metadata\n");
2215 switch (track->video.projection.type) {
2216 case MATROSKA_VIDEO_PROJECTION_TYPE_EQUIRECTANGULAR:
2217 if (track->video.projection.private.size == 20) {
2218 t = AV_RB32(priv_data + 4);
2219 b = AV_RB32(priv_data + 8);
2220 l = AV_RB32(priv_data + 12);
2221 r = AV_RB32(priv_data + 16);
2223 if (b >= UINT_MAX - t || r >= UINT_MAX - l) {
2224 av_log(logctx, AV_LOG_ERROR,
2225 "Invalid bounding rectangle coordinates "
2226 "%"PRIu32",%"PRIu32",%"PRIu32",%"PRIu32"\n",
2228 return AVERROR_INVALIDDATA;
2230 } else if (track->video.projection.private.size != 0) {
2231 av_log(logctx, AV_LOG_ERROR, "Unknown spherical metadata\n");
2232 return AVERROR_INVALIDDATA;
2235 if (l || t || r || b)
2236 projection = AV_SPHERICAL_EQUIRECTANGULAR_TILE;
2238 projection = AV_SPHERICAL_EQUIRECTANGULAR;
2240 case MATROSKA_VIDEO_PROJECTION_TYPE_CUBEMAP:
2241 if (track->video.projection.private.size < 4) {
2242 av_log(logctx, AV_LOG_ERROR, "Missing projection private properties\n");
2243 return AVERROR_INVALIDDATA;
2244 } else if (track->video.projection.private.size == 12) {
2245 uint32_t layout = AV_RB32(priv_data + 4);
2247 av_log(logctx, AV_LOG_WARNING,
2248 "Unknown spherical cubemap layout %"PRIu32"\n", layout);
2251 projection = AV_SPHERICAL_CUBEMAP;
2252 padding = AV_RB32(priv_data + 8);
2254 av_log(logctx, AV_LOG_ERROR, "Unknown spherical metadata\n");
2255 return AVERROR_INVALIDDATA;
2258 case MATROSKA_VIDEO_PROJECTION_TYPE_RECTANGULAR:
2259 /* No Spherical metadata */
2262 av_log(logctx, AV_LOG_WARNING,
2263 "Unknown spherical metadata type %"PRIu64"\n",
2264 track->video.projection.type);
2268 spherical = av_spherical_alloc(&spherical_size);
2270 return AVERROR(ENOMEM);
2272 spherical->projection = projection;
2274 spherical->yaw = (int32_t) (track->video.projection.yaw * (1 << 16));
2275 spherical->pitch = (int32_t) (track->video.projection.pitch * (1 << 16));
2276 spherical->roll = (int32_t) (track->video.projection.roll * (1 << 16));
2278 spherical->padding = padding;
2280 spherical->bound_left = l;
2281 spherical->bound_top = t;
2282 spherical->bound_right = r;
2283 spherical->bound_bottom = b;
2285 ret = av_stream_add_side_data(st, AV_PKT_DATA_SPHERICAL, (uint8_t *)spherical,
2288 av_freep(&spherical);
2295 static int get_qt_codec(MatroskaTrack *track, uint32_t *fourcc, enum AVCodecID *codec_id)
2297 const AVCodecTag *codec_tags;
2299 codec_tags = track->type == MATROSKA_TRACK_TYPE_VIDEO ?
2300 ff_codec_movvideo_tags : ff_codec_movaudio_tags;
2302 /* Normalize noncompliant private data that starts with the fourcc
2303 * by expanding/shifting the data by 4 bytes and storing the data
2304 * size at the start. */
2305 if (ff_codec_get_id(codec_tags, AV_RL32(track->codec_priv.data))) {
2306 int ret = av_buffer_realloc(&track->codec_priv.buf,
2307 track->codec_priv.size + 4 + AV_INPUT_BUFFER_PADDING_SIZE);
2311 track->codec_priv.data = track->codec_priv.buf->data;
2312 memmove(track->codec_priv.data + 4, track->codec_priv.data, track->codec_priv.size);
2313 track->codec_priv.size += 4;
2314 AV_WB32(track->codec_priv.data, track->codec_priv.size);
2317 *fourcc = AV_RL32(track->codec_priv.data + 4);
2318 *codec_id = ff_codec_get_id(codec_tags, *fourcc);
2323 static int matroska_parse_tracks(AVFormatContext *s)
2325 MatroskaDemuxContext *matroska = s->priv_data;
2326 MatroskaTrack *tracks = matroska->tracks.elem;
2331 for (i = 0; i < matroska->tracks.nb_elem; i++) {
2332 MatroskaTrack *track = &tracks[i];
2333 enum AVCodecID codec_id = AV_CODEC_ID_NONE;
2334 EbmlList *encodings_list = &track->encodings;
2335 MatroskaTrackEncoding *encodings = encodings_list->elem;
2336 uint8_t *extradata = NULL;
2337 int extradata_size = 0;
2338 int extradata_offset = 0;
2339 uint32_t fourcc = 0;
2341 char* key_id_base64 = NULL;
2344 /* Apply some sanity checks. */
2345 if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
2346 track->type != MATROSKA_TRACK_TYPE_AUDIO &&
2347 track->type != MATROSKA_TRACK_TYPE_SUBTITLE &&
2348 track->type != MATROSKA_TRACK_TYPE_METADATA) {
2349 av_log(matroska->ctx, AV_LOG_INFO,
2350 "Unknown or unsupported track type %"PRIu64"\n",
2354 if (!track->codec_id)
2357 if ( track->type == MATROSKA_TRACK_TYPE_AUDIO && track->codec_id[0] != 'A'
2358 || track->type == MATROSKA_TRACK_TYPE_VIDEO && track->codec_id[0] != 'V'
2359 || track->type == MATROSKA_TRACK_TYPE_SUBTITLE && track->codec_id[0] != 'D' && track->codec_id[0] != 'S'
2360 || track->type == MATROSKA_TRACK_TYPE_METADATA && track->codec_id[0] != 'D' && track->codec_id[0] != 'S'
2362 av_log(matroska->ctx, AV_LOG_INFO, "Inconsistent track type\n");
2366 if (track->audio.samplerate < 0 || track->audio.samplerate > INT_MAX ||
2367 isnan(track->audio.samplerate)) {
2368 av_log(matroska->ctx, AV_LOG_WARNING,
2369 "Invalid sample rate %f, defaulting to 8000 instead.\n",
2370 track->audio.samplerate);
2371 track->audio.samplerate = 8000;
2374 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2375 if (!track->default_duration && track->video.frame_rate > 0) {
2376 double default_duration = 1000000000 / track->video.frame_rate;
2377 if (default_duration > UINT64_MAX || default_duration < 0) {
2378 av_log(matroska->ctx, AV_LOG_WARNING,
2379 "Invalid frame rate %e. Cannot calculate default duration.\n",
2380 track->video.frame_rate);
2382 track->default_duration = default_duration;
2385 if (track->video.display_width == -1)
2386 track->video.display_width = track->video.pixel_width;
2387 if (track->video.display_height == -1)
2388 track->video.display_height = track->video.pixel_height;
2389 if (track->video.color_space.size == 4)
2390 fourcc = AV_RL32(track->video.color_space.data);
2391 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2392 if (!track->audio.out_samplerate)
2393 track->audio.out_samplerate = track->audio.samplerate;
2395 if (encodings_list->nb_elem > 1) {
2396 av_log(matroska->ctx, AV_LOG_ERROR,
2397 "Multiple combined encodings not supported");
2398 } else if (encodings_list->nb_elem == 1) {
2399 if (encodings[0].type) {
2400 if (encodings[0].encryption.key_id.size > 0) {
2401 /* Save the encryption key id to be stored later as a
2403 const int b64_size = AV_BASE64_SIZE(encodings[0].encryption.key_id.size);
2404 key_id_base64 = av_malloc(b64_size);
2405 if (key_id_base64 == NULL)
2406 return AVERROR(ENOMEM);
2408 av_base64_encode(key_id_base64, b64_size,
2409 encodings[0].encryption.key_id.data,
2410 encodings[0].encryption.key_id.size);
2412 encodings[0].scope = 0;
2413 av_log(matroska->ctx, AV_LOG_ERROR,
2414 "Unsupported encoding type");
2418 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
2421 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
2424 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO &&
2426 encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP) {
2427 encodings[0].scope = 0;
2428 av_log(matroska->ctx, AV_LOG_ERROR,
2429 "Unsupported encoding type");
2430 } else if (track->codec_priv.size && encodings[0].scope & 2) {
2431 uint8_t *codec_priv = track->codec_priv.data;
2432 int ret = matroska_decode_buffer(&track->codec_priv.data,
2433 &track->codec_priv.size,
2436 track->codec_priv.data = NULL;
2437 track->codec_priv.size = 0;
2438 av_log(matroska->ctx, AV_LOG_ERROR,
2439 "Failed to decode codec private data\n");
2442 if (codec_priv != track->codec_priv.data) {
2443 av_buffer_unref(&track->codec_priv.buf);
2444 if (track->codec_priv.data) {
2445 track->codec_priv.buf = av_buffer_create(track->codec_priv.data,
2446 track->codec_priv.size + AV_INPUT_BUFFER_PADDING_SIZE,
2448 if (!track->codec_priv.buf) {
2449 av_freep(&track->codec_priv.data);
2450 track->codec_priv.size = 0;
2451 return AVERROR(ENOMEM);
2457 track->needs_decoding = encodings && !encodings[0].type &&
2458 encodings[0].scope & 1 &&
2459 (encodings[0].compression.algo !=
2460 MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP ||
2461 encodings[0].compression.settings.size);
2463 for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
2464 if (!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
2465 strlen(ff_mkv_codec_tags[j].str))) {
2466 codec_id = ff_mkv_codec_tags[j].id;
2471 st = track->stream = avformat_new_stream(s, NULL);
2473 av_free(key_id_base64);
2474 return AVERROR(ENOMEM);
2477 if (key_id_base64) {
2478 /* export encryption key id as base64 metadata tag */
2479 av_dict_set(&st->metadata, "enc_key_id", key_id_base64,
2480 AV_DICT_DONT_STRDUP_VAL);
2483 if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC") &&
2484 track->codec_priv.size >= 40 &&
2485 track->codec_priv.data) {
2486 track->ms_compat = 1;
2487 bit_depth = AV_RL16(track->codec_priv.data + 14);
2488 fourcc = AV_RL32(track->codec_priv.data + 16);
2489 codec_id = ff_codec_get_id(ff_codec_bmp_tags,
2492 codec_id = ff_codec_get_id(ff_codec_movvideo_tags,
2494 extradata_offset = 40;
2495 } else if (!strcmp(track->codec_id, "A_MS/ACM") &&
2496 track->codec_priv.size >= 14 &&
2497 track->codec_priv.data) {
2499 ffio_init_context(&b, track->codec_priv.data,
2500 track->codec_priv.size,
2501 0, NULL, NULL, NULL, NULL);
2502 ret = ff_get_wav_header(s, &b, st->codecpar, track->codec_priv.size, 0);
2505 codec_id = st->codecpar->codec_id;
2506 fourcc = st->codecpar->codec_tag;
2507 extradata_offset = FFMIN(track->codec_priv.size, 18);
2508 } else if (!strcmp(track->codec_id, "A_QUICKTIME")
2509 /* Normally 36, but allow noncompliant private data */
2510 && (track->codec_priv.size >= 32)
2511 && (track->codec_priv.data)) {
2512 uint16_t sample_size;
2513 int ret = get_qt_codec(track, &fourcc, &codec_id);
2516 sample_size = AV_RB16(track->codec_priv.data + 26);
2518 if (sample_size == 8) {
2519 fourcc = MKTAG('r','a','w',' ');
2520 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
2521 } else if (sample_size == 16) {
2522 fourcc = MKTAG('t','w','o','s');
2523 codec_id = ff_codec_get_id(ff_codec_movaudio_tags, fourcc);
2526 if ((fourcc == MKTAG('t','w','o','s') ||
2527 fourcc == MKTAG('s','o','w','t')) &&
2529 codec_id = AV_CODEC_ID_PCM_S8;
2530 } else if (!strcmp(track->codec_id, "V_QUICKTIME") &&
2531 (track->codec_priv.size >= 21) &&
2532 (track->codec_priv.data)) {
2533 int ret = get_qt_codec(track, &fourcc, &codec_id);
2536 if (codec_id == AV_CODEC_ID_NONE && AV_RL32(track->codec_priv.data+4) == AV_RL32("SMI ")) {
2537 fourcc = MKTAG('S','V','Q','3');
2538 codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
2540 if (codec_id == AV_CODEC_ID_NONE)
2541 av_log(matroska->ctx, AV_LOG_ERROR,
2542 "mov FourCC not found %s.\n", av_fourcc2str(fourcc));
2543 if (track->codec_priv.size >= 86) {
2544 bit_depth = AV_RB16(track->codec_priv.data + 82);
2545 ffio_init_context(&b, track->codec_priv.data,
2546 track->codec_priv.size,
2547 0, NULL, NULL, NULL, NULL);
2548 if (ff_get_qtpalette(codec_id, &b, track->palette)) {
2550 track->has_palette = 1;
2553 } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
2554 switch (track->audio.bitdepth) {
2556 codec_id = AV_CODEC_ID_PCM_U8;
2559 codec_id = AV_CODEC_ID_PCM_S24BE;
2562 codec_id = AV_CODEC_ID_PCM_S32BE;
2565 } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
2566 switch (track->audio.bitdepth) {
2568 codec_id = AV_CODEC_ID_PCM_U8;
2571 codec_id = AV_CODEC_ID_PCM_S24LE;
2574 codec_id = AV_CODEC_ID_PCM_S32LE;
2577 } else if (codec_id == AV_CODEC_ID_PCM_F32LE &&
2578 track->audio.bitdepth == 64) {
2579 codec_id = AV_CODEC_ID_PCM_F64LE;
2580 } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
2581 int profile = matroska_aac_profile(track->codec_id);
2582 int sri = matroska_aac_sri(track->audio.samplerate);
2583 extradata = av_mallocz(5 + AV_INPUT_BUFFER_PADDING_SIZE);
2585 return AVERROR(ENOMEM);
2586 extradata[0] = (profile << 3) | ((sri & 0x0E) >> 1);
2587 extradata[1] = ((sri & 0x01) << 7) | (track->audio.channels << 3);
2588 if (strstr(track->codec_id, "SBR")) {
2589 sri = matroska_aac_sri(track->audio.out_samplerate);
2590 extradata[2] = 0x56;
2591 extradata[3] = 0xE5;
2592 extradata[4] = 0x80 | (sri << 3);
2596 } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX - 12 - AV_INPUT_BUFFER_PADDING_SIZE) {
2597 /* Only ALAC's magic cookie is stored in Matroska's track headers.
2598 * Create the "atom size", "tag", and "tag version" fields the
2599 * decoder expects manually. */
2600 extradata_size = 12 + track->codec_priv.size;
2601 extradata = av_mallocz(extradata_size +
2602 AV_INPUT_BUFFER_PADDING_SIZE);
2604 return AVERROR(ENOMEM);
2605 AV_WB32(extradata, extradata_size);
2606 memcpy(&extradata[4], "alac", 4);
2607 AV_WB32(&extradata[8], 0);
2608 memcpy(&extradata[12], track->codec_priv.data,
2609 track->codec_priv.size);
2610 } else if (codec_id == AV_CODEC_ID_TTA) {
2612 if (track->audio.channels > UINT16_MAX ||
2613 track->audio.bitdepth > UINT16_MAX) {
2614 av_log(matroska->ctx, AV_LOG_WARNING,
2615 "Too large audio channel number %"PRIu64
2616 " or bitdepth %"PRIu64". Skipping track.\n",
2617 track->audio.channels, track->audio.bitdepth);
2618 if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
2619 return AVERROR_INVALIDDATA;
2623 if (track->audio.out_samplerate < 0 || track->audio.out_samplerate > INT_MAX)
2624 return AVERROR_INVALIDDATA;
2625 extradata_size = 22;
2626 extradata = av_mallocz(extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
2628 return AVERROR(ENOMEM);
2630 bytestream_put_be32(&ptr, AV_RB32("TTA1"));
2631 bytestream_put_le16(&ptr, 1);
2632 bytestream_put_le16(&ptr, track->audio.channels);
2633 bytestream_put_le16(&ptr, track->audio.bitdepth);
2634 bytestream_put_le32(&ptr, track->audio.out_samplerate);
2635 bytestream_put_le32(&ptr, av_rescale(matroska->duration * matroska->time_scale,
2636 track->audio.out_samplerate,
2637 AV_TIME_BASE * 1000));
2638 } else if (codec_id == AV_CODEC_ID_RV10 ||
2639 codec_id == AV_CODEC_ID_RV20 ||
2640 codec_id == AV_CODEC_ID_RV30 ||
2641 codec_id == AV_CODEC_ID_RV40) {
2642 extradata_offset = 26;
2643 } else if (codec_id == AV_CODEC_ID_RA_144) {
2644 track->audio.out_samplerate = 8000;
2645 track->audio.channels = 1;
2646 } else if ((codec_id == AV_CODEC_ID_RA_288 ||
2647 codec_id == AV_CODEC_ID_COOK ||
2648 codec_id == AV_CODEC_ID_ATRAC3 ||
2649 codec_id == AV_CODEC_ID_SIPR)
2650 && track->codec_priv.data) {
2653 ffio_init_context(&b, track->codec_priv.data,
2654 track->codec_priv.size,
2655 0, NULL, NULL, NULL, NULL);
2657 flavor = avio_rb16(&b);
2658 track->audio.coded_framesize = avio_rb32(&b);
2660 track->audio.sub_packet_h = avio_rb16(&b);
2661 track->audio.frame_size = avio_rb16(&b);
2662 track->audio.sub_packet_size = avio_rb16(&b);
2663 if (track->audio.coded_framesize <= 0 ||
2664 track->audio.sub_packet_h <= 0 ||
2665 track->audio.frame_size <= 0)
2666 return AVERROR_INVALIDDATA;
2668 if (codec_id == AV_CODEC_ID_RA_288) {
2669 if (track->audio.sub_packet_h & 1 || 2 * track->audio.frame_size
2670 != (int64_t)track->audio.sub_packet_h * track->audio.coded_framesize)
2671 return AVERROR_INVALIDDATA;
2672 st->codecpar->block_align = track->audio.coded_framesize;
2673 track->codec_priv.size = 0;
2675 if (codec_id == AV_CODEC_ID_SIPR) {
2676 static const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
2678 return AVERROR_INVALIDDATA;
2679 track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
2680 st->codecpar->bit_rate = sipr_bit_rate[flavor];
2681 } else if (track->audio.sub_packet_size <= 0 ||
2682 track->audio.frame_size % track->audio.sub_packet_size)
2683 return AVERROR_INVALIDDATA;
2684 st->codecpar->block_align = track->audio.sub_packet_size;
2685 extradata_offset = 78;
2687 track->audio.buf = av_malloc_array(track->audio.sub_packet_h,
2688 track->audio.frame_size);
2689 if (!track->audio.buf)
2690 return AVERROR(ENOMEM);
2691 } else if (codec_id == AV_CODEC_ID_FLAC && track->codec_priv.size) {
2692 ret = matroska_parse_flac(s, track, &extradata_offset);
2695 } else if (codec_id == AV_CODEC_ID_WAVPACK && track->codec_priv.size < 2) {
2696 av_log(matroska->ctx, AV_LOG_INFO, "Assuming WavPack version 4.10 "
2697 "in absence of valid CodecPrivate.\n");
2699 extradata = av_mallocz(2 + AV_INPUT_BUFFER_PADDING_SIZE);
2701 return AVERROR(ENOMEM);
2702 AV_WL16(extradata, 0x410);
2703 } else if (codec_id == AV_CODEC_ID_PRORES && track->codec_priv.size == 4) {
2704 fourcc = AV_RL32(track->codec_priv.data);
2705 } else if (codec_id == AV_CODEC_ID_VP9 && track->codec_priv.size) {
2706 /* we don't need any value stored in CodecPrivate.
2707 make sure that it's not exported as extradata. */
2708 track->codec_priv.size = 0;
2709 } else if (codec_id == AV_CODEC_ID_AV1 && track->codec_priv.size) {
2710 /* For now, propagate only the OBUs, if any. Once libavcodec is
2711 updated to handle isobmff style extradata this can be removed. */
2712 extradata_offset = 4;
2714 track->codec_priv.size -= extradata_offset;
2716 if (codec_id == AV_CODEC_ID_NONE)
2717 av_log(matroska->ctx, AV_LOG_INFO,
2718 "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
2720 if (track->time_scale < 0.01) {
2721 av_log(matroska->ctx, AV_LOG_WARNING,
2722 "Track TimestampScale too small %f, assuming 1.0.\n",
2724 track->time_scale = 1.0;
2726 avpriv_set_pts_info(st, 64, matroska->time_scale * track->time_scale,
2727 1000 * 1000 * 1000); /* 64 bit pts in ns */
2729 /* convert the delay from ns to the track timebase */
2730 track->codec_delay_in_track_tb = av_rescale_q(track->codec_delay,
2731 (AVRational){ 1, 1000000000 },
2734 st->codecpar->codec_id = codec_id;
2736 if (strcmp(track->language, "und"))
2737 av_dict_set(&st->metadata, "language", track->language, 0);
2738 av_dict_set(&st->metadata, "title", track->name, 0);
2740 if (track->flag_default)
2741 st->disposition |= AV_DISPOSITION_DEFAULT;
2742 if (track->flag_forced)
2743 st->disposition |= AV_DISPOSITION_FORCED;
2745 if (!st->codecpar->extradata) {
2747 st->codecpar->extradata = extradata;
2748 st->codecpar->extradata_size = extradata_size;
2749 } else if (track->codec_priv.data && track->codec_priv.size > 0) {
2750 if (ff_alloc_extradata(st->codecpar, track->codec_priv.size))
2751 return AVERROR(ENOMEM);
2752 memcpy(st->codecpar->extradata,
2753 track->codec_priv.data + extradata_offset,
2754 track->codec_priv.size);
2758 if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2759 MatroskaTrackPlane *planes = track->operation.combine_planes.elem;
2760 int display_width_mul = 1;
2761 int display_height_mul = 1;
2763 st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
2764 st->codecpar->codec_tag = fourcc;
2766 st->codecpar->bits_per_coded_sample = bit_depth;
2767 st->codecpar->width = track->video.pixel_width;
2768 st->codecpar->height = track->video.pixel_height;
2770 if (track->video.interlaced == MATROSKA_VIDEO_INTERLACE_FLAG_INTERLACED)
2771 st->codecpar->field_order = mkv_field_order(matroska, track->video.field_order);
2772 else if (track->video.interlaced == MATROSKA_VIDEO_INTERLACE_FLAG_PROGRESSIVE)
2773 st->codecpar->field_order = AV_FIELD_PROGRESSIVE;
2775 if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
2776 mkv_stereo_mode_display_mul(track->video.stereo_mode, &display_width_mul, &display_height_mul);
2778 if (track->video.display_unit < MATROSKA_VIDEO_DISPLAYUNIT_UNKNOWN) {
2779 av_reduce(&st->sample_aspect_ratio.num,
2780 &st->sample_aspect_ratio.den,
2781 st->codecpar->height * track->video.display_width * display_width_mul,
2782 st->codecpar->width * track->video.display_height * display_height_mul,
2785 if (st->codecpar->codec_id != AV_CODEC_ID_HEVC)
2786 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2788 if (track->default_duration) {
2789 av_reduce(&st->avg_frame_rate.num, &st->avg_frame_rate.den,
2790 1000000000, track->default_duration, 30000);
2791 #if FF_API_R_FRAME_RATE
2792 if ( st->avg_frame_rate.num < st->avg_frame_rate.den * 1000LL
2793 && st->avg_frame_rate.num > st->avg_frame_rate.den * 5LL)
2794 st->r_frame_rate = st->avg_frame_rate;
2798 /* export stereo mode flag as metadata tag */
2799 if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB)
2800 av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
2802 /* export alpha mode flag as metadata tag */
2803 if (track->video.alpha_mode)
2804 av_dict_set(&st->metadata, "alpha_mode", "1", 0);
2806 /* if we have virtual track, mark the real tracks */
2807 for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
2809 if (planes[j].type >= MATROSKA_VIDEO_STEREO_PLANE_COUNT)
2811 snprintf(buf, sizeof(buf), "%s_%d",
2812 ff_matroska_video_stereo_plane[planes[j].type], i);
2813 for (k=0; k < matroska->tracks.nb_elem; k++)
2814 if (planes[j].uid == tracks[k].uid && tracks[k].stream) {
2815 av_dict_set(&tracks[k].stream->metadata,
2816 "stereo_mode", buf, 0);
2820 // add stream level stereo3d side data if it is a supported format
2821 if (track->video.stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB &&
2822 track->video.stereo_mode != 10 && track->video.stereo_mode != 12) {
2823 int ret = ff_mkv_stereo3d_conv(st, track->video.stereo_mode);
2828 ret = mkv_parse_video_color(st, track);
2831 ret = mkv_parse_video_projection(st, track, matroska->ctx);
2834 } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2835 st->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
2836 st->codecpar->codec_tag = fourcc;
2837 st->codecpar->sample_rate = track->audio.out_samplerate;
2838 st->codecpar->channels = track->audio.channels;
2839 if (!st->codecpar->bits_per_coded_sample)
2840 st->codecpar->bits_per_coded_sample = track->audio.bitdepth;
2841 if (st->codecpar->codec_id == AV_CODEC_ID_MP3 ||
2842 st->codecpar->codec_id == AV_CODEC_ID_MLP ||
2843 st->codecpar->codec_id == AV_CODEC_ID_TRUEHD)
2844 st->need_parsing = AVSTREAM_PARSE_FULL;
2845 else if (st->codecpar->codec_id != AV_CODEC_ID_AAC)
2846 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2847 if (track->codec_delay > 0) {
2848 st->codecpar->initial_padding = av_rescale_q(track->codec_delay,
2849 (AVRational){1, 1000000000},
2850 (AVRational){1, st->codecpar->codec_id == AV_CODEC_ID_OPUS ?
2851 48000 : st->codecpar->sample_rate});
2853 if (track->seek_preroll > 0) {
2854 st->codecpar->seek_preroll = av_rescale_q(track->seek_preroll,
2855 (AVRational){1, 1000000000},
2856 (AVRational){1, st->codecpar->sample_rate});
2858 } else if (codec_id == AV_CODEC_ID_WEBVTT) {
2859 st->codecpar->codec_type = AVMEDIA_TYPE_SUBTITLE;
2861 if (!strcmp(track->codec_id, "D_WEBVTT/CAPTIONS")) {
2862 st->disposition |= AV_DISPOSITION_CAPTIONS;
2863 } else if (!strcmp(track->codec_id, "D_WEBVTT/DESCRIPTIONS")) {
2864 st->disposition |= AV_DISPOSITION_DESCRIPTIONS;
2865 } else if (!strcmp(track->codec_id, "D_WEBVTT/METADATA")) {
2866 st->disposition |= AV_DISPOSITION_METADATA;
2868 } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2869 st->codecpar->codec_type = AVMEDIA_TYPE_SUBTITLE;
2876 static int matroska_read_header(AVFormatContext *s)
2878 MatroskaDemuxContext *matroska = s->priv_data;
2879 EbmlList *attachments_list = &matroska->attachments;
2880 EbmlList *chapters_list = &matroska->chapters;
2881 MatroskaAttachment *attachments;
2882 MatroskaChapter *chapters;
2883 uint64_t max_start = 0;
2889 matroska->cues_parsing_deferred = 1;
2891 /* First read the EBML header. */
2892 if (ebml_parse(matroska, ebml_syntax, &ebml) || !ebml.doctype) {
2893 av_log(matroska->ctx, AV_LOG_ERROR, "EBML header parsing failed\n");
2894 ebml_free(ebml_syntax, &ebml);
2895 return AVERROR_INVALIDDATA;
2897 if (ebml.version > EBML_VERSION ||
2898 ebml.max_size > sizeof(uint64_t) ||
2899 ebml.id_length > sizeof(uint32_t) ||
2900 ebml.doctype_version > 3) {
2901 avpriv_report_missing_feature(matroska->ctx,
2902 "EBML version %"PRIu64", doctype %s, doc version %"PRIu64,
2903 ebml.version, ebml.doctype, ebml.doctype_version);
2904 ebml_free(ebml_syntax, &ebml);
2905 return AVERROR_PATCHWELCOME;
2906 } else if (ebml.doctype_version == 3) {
2907 av_log(matroska->ctx, AV_LOG_WARNING,
2908 "EBML header using unsupported features\n"
2909 "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2910 ebml.version, ebml.doctype, ebml.doctype_version);
2912 for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
2913 if (!strcmp(ebml.doctype, matroska_doctypes[i]))
2915 if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
2916 av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
2917 if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
2918 ebml_free(ebml_syntax, &ebml);
2919 return AVERROR_INVALIDDATA;
2922 ebml_free(ebml_syntax, &ebml);
2924 /* The next thing is a segment. */
2925 pos = avio_tell(matroska->ctx->pb);
2926 res = ebml_parse(matroska, matroska_segments, matroska);
2927 // Try resyncing until we find an EBML_STOP type element.
2929 res = matroska_resync(matroska, pos);
2932 pos = avio_tell(matroska->ctx->pb);
2933 res = ebml_parse(matroska, matroska_segment, matroska);
2935 /* Set data_offset as it might be needed later by seek_frame_generic. */
2936 if (matroska->current_id == MATROSKA_ID_CLUSTER)
2937 s->internal->data_offset = avio_tell(matroska->ctx->pb) - 4;
2938 matroska_execute_seekhead(matroska);
2940 if (!matroska->time_scale)
2941 matroska->time_scale = 1000000;
2942 if (matroska->duration)
2943 matroska->ctx->duration = matroska->duration * matroska->time_scale *
2944 1000 / AV_TIME_BASE;
2945 av_dict_set(&s->metadata, "title", matroska->title, 0);
2946 av_dict_set(&s->metadata, "encoder", matroska->muxingapp, 0);
2948 if (matroska->date_utc.size == 8)
2949 matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
2951 res = matroska_parse_tracks(s);
2955 attachments = attachments_list->elem;
2956 for (j = 0; j < attachments_list->nb_elem; j++) {
2957 if (!(attachments[j].filename && attachments[j].mime &&
2958 attachments[j].bin.data && attachments[j].bin.size > 0)) {
2959 av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
2961 AVStream *st = avformat_new_stream(s, NULL);
2964 av_dict_set(&st->metadata, "filename", attachments[j].filename, 0);
2965 av_dict_set(&st->metadata, "mimetype", attachments[j].mime, 0);
2966 if (attachments[j].description)
2967 av_dict_set(&st->metadata, "title", attachments[j].description, 0);
2968 st->codecpar->codec_id = AV_CODEC_ID_NONE;
2970 for (i = 0; mkv_image_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
2971 if (!strncmp(mkv_image_mime_tags[i].str, attachments[j].mime,
2972 strlen(mkv_image_mime_tags[i].str))) {
2973 st->codecpar->codec_id = mkv_image_mime_tags[i].id;
2978 attachments[j].stream = st;
2980 if (st->codecpar->codec_id != AV_CODEC_ID_NONE) {
2981 AVPacket *pkt = &st->attached_pic;
2983 st->disposition |= AV_DISPOSITION_ATTACHED_PIC;
2984 st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
2986 av_init_packet(pkt);
2987 pkt->buf = attachments[j].bin.buf;
2988 attachments[j].bin.buf = NULL;
2989 pkt->data = attachments[j].bin.data;
2990 pkt->size = attachments[j].bin.size;
2991 pkt->stream_index = st->index;
2992 pkt->flags |= AV_PKT_FLAG_KEY;
2994 st->codecpar->codec_type = AVMEDIA_TYPE_ATTACHMENT;
2995 if (ff_alloc_extradata(st->codecpar, attachments[j].bin.size))
2997 memcpy(st->codecpar->extradata, attachments[j].bin.data,
2998 attachments[j].bin.size);
3000 for (i = 0; mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
3001 if (!strncmp(mkv_mime_tags[i].str, attachments[j].mime,
3002 strlen(mkv_mime_tags[i].str))) {
3003 st->codecpar->codec_id = mkv_mime_tags[i].id;
3011 chapters = chapters_list->elem;
3012 for (i = 0; i < chapters_list->nb_elem; i++)
3013 if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid &&
3014 (max_start == 0 || chapters[i].start > max_start)) {
3015 chapters[i].chapter =
3016 avpriv_new_chapter(s, chapters[i].uid,
3017 (AVRational) { 1, 1000000000 },
3018 chapters[i].start, chapters[i].end,
3020 max_start = chapters[i].start;
3023 matroska_add_index_entries(matroska);
3025 matroska_convert_tags(s);
3029 matroska_read_close(s);
3034 * Put one packet in an application-supplied AVPacket struct.
3035 * Returns 0 on success or -1 on failure.
3037 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
3040 if (matroska->queue) {
3041 MatroskaTrack *tracks = matroska->tracks.elem;
3042 MatroskaTrack *track;
3044 avpriv_packet_list_get(&matroska->queue, &matroska->queue_end, pkt);
3045 track = &tracks[pkt->stream_index];
3046 if (track->has_palette) {
3047 uint8_t *pal = av_packet_new_side_data(pkt, AV_PKT_DATA_PALETTE, AVPALETTE_SIZE);
3049 av_log(matroska->ctx, AV_LOG_ERROR, "Cannot append palette to packet\n");
3051 memcpy(pal, track->palette, AVPALETTE_SIZE);
3053 track->has_palette = 0;
3062 * Free all packets in our internal queue.
3064 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
3066 avpriv_packet_list_free(&matroska->queue, &matroska->queue_end);
3069 static int matroska_parse_laces(MatroskaDemuxContext *matroska, uint8_t **buf,
3070 int size, int type, AVIOContext *pb,
3071 uint32_t lace_size[256], int *laces)
3074 uint8_t *data = *buf;
3078 lace_size[0] = size;
3083 return AVERROR_INVALIDDATA;
3090 case 0x1: /* Xiph lacing */
3094 for (n = 0; n < *laces - 1; n++) {
3099 return AVERROR_INVALIDDATA;
3102 lace_size[n] += temp;
3105 } while (temp == 0xff);
3108 return AVERROR_INVALIDDATA;
3110 lace_size[n] = size - total;
3114 case 0x2: /* fixed-size lacing */
3115 if (size % (*laces))
3116 return AVERROR_INVALIDDATA;
3117 for (n = 0; n < *laces; n++)
3118 lace_size[n] = size / *laces;
3121 case 0x3: /* EBML lacing */
3129 n = ebml_read_num(matroska, pb, 8, &num, 1);
3133 return AVERROR_INVALIDDATA;
3135 total = lace_size[0] = num;
3137 for (n = 1; n < *laces - 1; n++) {
3140 r = matroska_ebmlnum_sint(matroska, pb, &snum);
3143 if (lace_size[n - 1] + snum > (uint64_t)INT_MAX)
3144 return AVERROR_INVALIDDATA;
3146 lace_size[n] = lace_size[n - 1] + snum;
3147 total += lace_size[n];
3153 return AVERROR_INVALIDDATA;
3155 lace_size[*laces - 1] = size - total;
3165 static int matroska_parse_rm_audio(MatroskaDemuxContext *matroska,
3166 MatroskaTrack *track, AVStream *st,
3167 uint8_t *data, int size, uint64_t timecode,
3170 const int a = st->codecpar->block_align;
3171 const int sps = track->audio.sub_packet_size;
3172 const int cfs = track->audio.coded_framesize;
3173 const int h = track->audio.sub_packet_h;
3174 const int w = track->audio.frame_size;
3175 int y = track->audio.sub_packet_cnt;
3178 if (!track->audio.pkt_cnt) {
3179 if (track->audio.sub_packet_cnt == 0)
3180 track->audio.buf_timecode = timecode;
3181 if (st->codecpar->codec_id == AV_CODEC_ID_RA_288) {
3182 if (size < cfs * h / 2) {
3183 av_log(matroska->ctx, AV_LOG_ERROR,
3184 "Corrupt int4 RM-style audio packet size\n");
3185 return AVERROR_INVALIDDATA;
3187 for (x = 0; x < h / 2; x++)
3188 memcpy(track->audio.buf + x * 2 * w + y * cfs,
3189 data + x * cfs, cfs);
3190 } else if (st->codecpar->codec_id == AV_CODEC_ID_SIPR) {
3192 av_log(matroska->ctx, AV_LOG_ERROR,
3193 "Corrupt sipr RM-style audio packet size\n");
3194 return AVERROR_INVALIDDATA;
3196 memcpy(track->audio.buf + y * w, data, w);
3199 av_log(matroska->ctx, AV_LOG_ERROR,
3200 "Corrupt generic RM-style audio packet size\n");
3201 return AVERROR_INVALIDDATA;
3203 for (x = 0; x < w / sps; x++)
3204 memcpy(track->audio.buf +
3205 sps * (h * x + ((h + 1) / 2) * (y & 1) + (y >> 1)),
3206 data + x * sps, sps);
3209 if (++track->audio.sub_packet_cnt >= h) {
3210 if (st->codecpar->codec_id == AV_CODEC_ID_SIPR)
3211 ff_rm_reorder_sipr_data(track->audio.buf, h, w);
3212 track->audio.sub_packet_cnt = 0;
3213 track->audio.pkt_cnt = h * w / a;
3217 while (track->audio.pkt_cnt) {
3219 AVPacket pktl, *pkt = &pktl;
3221 ret = av_new_packet(pkt, a);
3226 track->audio.buf + a * (h * w / a - track->audio.pkt_cnt--),
3228 pkt->pts = track->audio.buf_timecode;
3229 track->audio.buf_timecode = AV_NOPTS_VALUE;
3231 pkt->stream_index = st->index;
3232 ret = avpriv_packet_list_put(&matroska->queue, &matroska->queue_end, pkt, NULL, 0);
3234 av_packet_unref(pkt);
3235 return AVERROR(ENOMEM);
3242 /* reconstruct full wavpack blocks from mangled matroska ones */
3243 static int matroska_parse_wavpack(MatroskaTrack *track,
3244 uint8_t **data, int *size)
3246 uint8_t *dst = NULL;
3247 uint8_t *src = *data;
3252 int ret, offset = 0;
3255 return AVERROR_INVALIDDATA;
3257 av_assert1(track->stream->codecpar->extradata_size >= 2);
3258 ver = AV_RL16(track->stream->codecpar->extradata);
3260 samples = AV_RL32(src);
3264 while (srclen >= 8) {
3269 uint32_t flags = AV_RL32(src);
3270 uint32_t crc = AV_RL32(src + 4);
3274 multiblock = (flags & 0x1800) != 0x1800;
3277 ret = AVERROR_INVALIDDATA;
3280 blocksize = AV_RL32(src);
3286 if (blocksize > srclen) {
3287 ret = AVERROR_INVALIDDATA;
3291 tmp = av_realloc(dst, dstlen + blocksize + 32 + AV_INPUT_BUFFER_PADDING_SIZE);
3293 ret = AVERROR(ENOMEM);
3297 dstlen += blocksize + 32;
3299 AV_WL32(dst + offset, MKTAG('w', 'v', 'p', 'k')); // tag
3300 AV_WL32(dst + offset + 4, blocksize + 24); // blocksize - 8
3301 AV_WL16(dst + offset + 8, ver); // version
3302 AV_WL16(dst + offset + 10, 0); // track/index_no
3303 AV_WL32(dst + offset + 12, 0); // total samples
3304 AV_WL32(dst + offset + 16, 0); // block index
3305 AV_WL32(dst + offset + 20, samples); // number of samples
3306 AV_WL32(dst + offset + 24, flags); // flags
3307 AV_WL32(dst + offset + 28, crc); // crc
3308 memcpy(dst + offset + 32, src, blocksize); // block data
3311 srclen -= blocksize;
3312 offset += blocksize + 32;
3315 memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3327 static int matroska_parse_prores(MatroskaTrack *track,
3328 uint8_t **data, int *size)
3331 int dstlen = *size + 8;
3333 dst = av_malloc(dstlen + AV_INPUT_BUFFER_PADDING_SIZE);
3335 return AVERROR(ENOMEM);
3337 AV_WB32(dst, dstlen);
3338 AV_WB32(dst + 4, MKBETAG('i', 'c', 'p', 'f'));
3339 memcpy(dst + 8, *data, dstlen - 8);
3340 memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3348 static int matroska_parse_webvtt(MatroskaDemuxContext *matroska,
3349 MatroskaTrack *track,
3351 uint8_t *data, int data_len,
3356 AVPacket pktl, *pkt = &pktl;
3357 uint8_t *id, *settings, *text, *buf;
3358 int id_len, settings_len, text_len;
3363 return AVERROR_INVALIDDATA;
3366 q = data + data_len;
3371 if (*p == '\r' || *p == '\n') {
3380 if (p >= q || *p != '\n')
3381 return AVERROR_INVALIDDATA;
3387 if (*p == '\r' || *p == '\n') {
3388 settings_len = p - settings;
3396 if (p >= q || *p != '\n')
3397 return AVERROR_INVALIDDATA;
3402 while (text_len > 0) {
3403 const int len = text_len - 1;
3404 const uint8_t c = p[len];
3405 if (c != '\r' && c != '\n')
3411 return AVERROR_INVALIDDATA;
3413 err = av_new_packet(pkt, text_len);
3418 memcpy(pkt->data, text, text_len);
3421 buf = av_packet_new_side_data(pkt,
3422 AV_PKT_DATA_WEBVTT_IDENTIFIER,
3425 av_packet_unref(pkt);
3426 return AVERROR(ENOMEM);
3428 memcpy(buf, id, id_len);
3431 if (settings_len > 0) {
3432 buf = av_packet_new_side_data(pkt,
3433 AV_PKT_DATA_WEBVTT_SETTINGS,
3436 av_packet_unref(pkt);
3437 return AVERROR(ENOMEM);
3439 memcpy(buf, settings, settings_len);
3442 // Do we need this for subtitles?
3443 // pkt->flags = AV_PKT_FLAG_KEY;
3445 pkt->stream_index = st->index;
3446 pkt->pts = timecode;
3448 // Do we need this for subtitles?
3449 // pkt->dts = timecode;
3451 pkt->duration = duration;
3454 err = avpriv_packet_list_put(&matroska->queue, &matroska->queue_end, pkt, NULL, 0);
3456 av_packet_unref(pkt);
3457 return AVERROR(ENOMEM);
3463 static int matroska_parse_frame(MatroskaDemuxContext *matroska,
3464 MatroskaTrack *track, AVStream *st,
3465 AVBufferRef *buf, uint8_t *data, int pkt_size,
3466 uint64_t timecode, uint64_t lace_duration,
3467 int64_t pos, int is_keyframe,
3468 uint8_t *additional, uint64_t additional_id, int additional_size,
3469 int64_t discard_padding)
3471 uint8_t *pkt_data = data;
3473 AVPacket pktl, *pkt = &pktl;
3475 if (st->codecpar->codec_id == AV_CODEC_ID_WAVPACK) {
3476 res = matroska_parse_wavpack(track, &pkt_data, &pkt_size);
3478 av_log(matroska->ctx, AV_LOG_ERROR,
3479 "Error parsing a wavpack block.\n");
3487 if (st->codecpar->codec_id == AV_CODEC_ID_PRORES &&
3488 AV_RB32(pkt_data + 4) != MKBETAG('i', 'c', 'p', 'f')) {
3489 res = matroska_parse_prores(track, &pkt_data, &pkt_size);
3491 av_log(matroska->ctx, AV_LOG_ERROR,
3492 "Error parsing a prores block.\n");
3500 if (!pkt_size && !additional_size)
3503 av_init_packet(pkt);
3505 pkt->buf = av_buffer_create(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE,
3508 pkt->buf = av_buffer_ref(buf);
3511 res = AVERROR(ENOMEM);
3515 pkt->data = pkt_data;
3516 pkt->size = pkt_size;
3517 pkt->flags = is_keyframe;
3518 pkt->stream_index = st->index;
3520 if (additional_size > 0) {
3521 uint8_t *side_data = av_packet_new_side_data(pkt,
3522 AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL,
3523 additional_size + 8);
3525 av_packet_unref(pkt);
3526 return AVERROR(ENOMEM);
3528 AV_WB64(side_data, additional_id);
3529 memcpy(side_data + 8, additional, additional_size);
3532 if (discard_padding) {
3533 uint8_t *side_data = av_packet_new_side_data(pkt,
3534 AV_PKT_DATA_SKIP_SAMPLES,
3537 av_packet_unref(pkt);
3538 return AVERROR(ENOMEM);
3540 discard_padding = av_rescale_q(discard_padding,
3541 (AVRational){1, 1000000000},
3542 (AVRational){1, st->codecpar->sample_rate});
3543 if (discard_padding > 0) {
3544 AV_WL32(side_data + 4, discard_padding);
3546 AV_WL32(side_data, -discard_padding);
3550 if (track->ms_compat)
3551 pkt->dts = timecode;
3553 pkt->pts = timecode;
3555 pkt->duration = lace_duration;
3557 #if FF_API_CONVERGENCE_DURATION
3558 FF_DISABLE_DEPRECATION_WARNINGS
3559 if (st->codecpar->codec_id == AV_CODEC_ID_SUBRIP) {
3560 pkt->convergence_duration = lace_duration;
3562 FF_ENABLE_DEPRECATION_WARNINGS
3565 res = avpriv_packet_list_put(&matroska->queue, &matroska->queue_end, pkt, NULL, 0);
3567 av_packet_unref(pkt);
3568 return AVERROR(ENOMEM);
3580 static int matroska_parse_block(MatroskaDemuxContext *matroska, AVBufferRef *buf, uint8_t *data,
3581 int size, int64_t pos, uint64_t cluster_time,
3582 uint64_t block_duration, int is_keyframe,
3583 uint8_t *additional, uint64_t additional_id, int additional_size,
3584 int64_t cluster_pos, int64_t discard_padding)
3586 uint64_t timecode = AV_NOPTS_VALUE;
3587 MatroskaTrack *track;
3592 uint32_t lace_size[256];
3593 int n, flags, laces = 0;
3595 int trust_default_duration;
3597 ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
3599 if ((n = ebml_read_num(matroska, &pb, 8, &num, 1)) < 0)
3604 track = matroska_find_track_by_num(matroska, num);
3605 if (!track || size < 3)
3606 return AVERROR_INVALIDDATA;
3608 if (!(st = track->stream)) {
3609 av_log(matroska->ctx, AV_LOG_VERBOSE,
3610 "No stream associated to TrackNumber %"PRIu64". "
3611 "Ignoring Block with this TrackNumber.\n", num);
3615 if (st->discard >= AVDISCARD_ALL)
3617 if (block_duration > INT64_MAX)
3618 block_duration = INT64_MAX;
3620 block_time = sign_extend(AV_RB16(data), 16);
3624 if (is_keyframe == -1)
3625 is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
3627 if (cluster_time != (uint64_t) -1 &&
3628 (block_time >= 0 || cluster_time >= -block_time)) {
3629 uint64_t timecode_cluster_in_track_tb = (double) cluster_time / track->time_scale;
3630 timecode = timecode_cluster_in_track_tb + block_time - track->codec_delay_in_track_tb;
3631 if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE &&
3632 timecode < track->end_timecode)
3633 is_keyframe = 0; /* overlapping subtitles are not key frame */
3635 ff_reduce_index(matroska->ctx, st->index);
3636 av_add_index_entry(st, cluster_pos, timecode, 0, 0,
3641 if (matroska->skip_to_keyframe &&
3642 track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
3643 // Compare signed timecodes. Timecode may be negative due to codec delay
3644 // offset. We don't support timestamps greater than int64_t anyway - see
3646 if ((int64_t)timecode < (int64_t)matroska->skip_to_timecode)
3649 matroska->skip_to_keyframe = 0;
3650 else if (!st->internal->skip_to_keyframe) {
3651 av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
3652 matroska->skip_to_keyframe = 0;
3656 res = matroska_parse_laces(matroska, &data, size, (flags & 0x06) >> 1,
3657 &pb, lace_size, &laces);
3659 av_log(matroska->ctx, AV_LOG_ERROR, "Error parsing frame sizes.\n");
3663 trust_default_duration = track->default_duration != 0;
3664 if (track->audio.samplerate == 8000 && trust_default_duration) {
3665 // If this is needed for more codecs, then add them here
3666 if (st->codecpar->codec_id == AV_CODEC_ID_AC3) {
3667 if (track->audio.samplerate != st->codecpar->sample_rate || !st->codecpar->frame_size)
3668 trust_default_duration = 0;
3672 if (!block_duration && trust_default_duration)
3673 block_duration = track->default_duration * laces / matroska->time_scale;
3675 if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
3676 track->end_timecode =
3677 FFMAX(track->end_timecode, timecode + block_duration);
3679 for (n = 0; n < laces; n++) {
3680 int64_t lace_duration = block_duration*(n+1) / laces - block_duration*n / laces;
3681 uint8_t *out_data = data;
3682 int out_size = lace_size[n];
3684 if (track->needs_decoding) {
3685 res = matroska_decode_buffer(&out_data, &out_size, track);
3688 /* Given that we are here means that out_data is no longer
3689 * owned by buf, so set it to NULL. This depends upon
3690 * zero-length header removal compression being ignored. */
3691 av_assert1(out_data != data);
3695 if (track->audio.buf) {
3696 res = matroska_parse_rm_audio(matroska, track, st,
3703 } else if (st->codecpar->codec_id == AV_CODEC_ID_WEBVTT) {
3704 res = matroska_parse_webvtt(matroska, track, st,
3706 timecode, lace_duration,
3713 res = matroska_parse_frame(matroska, track, st, buf, out_data,
3714 out_size, timecode, lace_duration,
3715 pos, !n ? is_keyframe : 0,
3716 additional, additional_id, additional_size,
3722 if (timecode != AV_NOPTS_VALUE)
3723 timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
3724 data += lace_size[n];
3730 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
3732 MatroskaCluster *cluster = &matroska->current_cluster;
3733 MatroskaBlock *block = &cluster->block;
3736 av_assert0(matroska->num_levels <= 2);
3738 if (matroska->num_levels == 1) {
3739 res = ebml_parse(matroska, matroska_segment, NULL);
3742 /* Found a cluster: subtract the size of the ID already read. */
3743 cluster->pos = avio_tell(matroska->ctx->pb) - 4;
3745 res = ebml_parse(matroska, matroska_cluster_enter, cluster);
3751 if (matroska->num_levels == 2) {
3752 /* We are inside a cluster. */
3753 res = ebml_parse(matroska, matroska_cluster_parsing, cluster);
3755 if (res >= 0 && block->bin.size > 0) {
3756 int is_keyframe = block->non_simple ? block->reference.count == 0 : -1;
3757 uint8_t* additional = block->additional.size > 0 ?
3758 block->additional.data : NULL;
3760 res = matroska_parse_block(matroska, block->bin.buf, block->bin.data,
3761 block->bin.size, block->bin.pos,
3762 cluster->timecode, block->duration,
3763 is_keyframe, additional, block->additional_id,
3764 block->additional.size, cluster->pos,
3765 block->discard_padding);
3768 ebml_free(matroska_blockgroup, block);
3769 memset(block, 0, sizeof(*block));
3770 } else if (!matroska->num_levels) {
3771 if (!avio_feof(matroska->ctx->pb)) {
3772 avio_r8(matroska->ctx->pb);
3773 if (!avio_feof(matroska->ctx->pb)) {
3774 av_log(matroska->ctx, AV_LOG_WARNING, "File extends beyond "
3775 "end of segment.\n");
3776 return AVERROR_INVALIDDATA;
3786 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
3788 MatroskaDemuxContext *matroska = s->priv_data;
3791 if (matroska->resync_pos == -1) {
3792 // This can only happen if generic seeking has been used.
3793 matroska->resync_pos = avio_tell(s->pb);
3796 while (matroska_deliver_packet(matroska, pkt)) {
3798 return (ret < 0) ? ret : AVERROR_EOF;
3799 if (matroska_parse_cluster(matroska) < 0 && !matroska->done)
3800 ret = matroska_resync(matroska, matroska->resync_pos);
3806 static int matroska_read_seek(AVFormatContext *s, int stream_index,
3807 int64_t timestamp, int flags)
3809 MatroskaDemuxContext *matroska = s->priv_data;
3810 MatroskaTrack *tracks = NULL;
3811 AVStream *st = s->streams[stream_index];
3814 /* Parse the CUES now since we need the index data to seek. */
3815 if (matroska->cues_parsing_deferred > 0) {
3816 matroska->cues_parsing_deferred = 0;
3817 matroska_parse_cues(matroska);
3820 if (!st->internal->nb_index_entries)
3822 timestamp = FFMAX(timestamp, st->internal->index_entries[0].timestamp);
3824 if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 || index == st->internal->nb_index_entries - 1) {
3825 matroska_reset_status(matroska, 0, st->internal->index_entries[st->internal->nb_index_entries - 1].pos);
3826 while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 || index == st->internal->nb_index_entries - 1) {
3827 matroska_clear_queue(matroska);
3828 if (matroska_parse_cluster(matroska) < 0)
3833 matroska_clear_queue(matroska);
3834 if (index < 0 || (matroska->cues_parsing_deferred < 0 && index == st->internal->nb_index_entries - 1))
3837 tracks = matroska->tracks.elem;
3838 for (i = 0; i < matroska->tracks.nb_elem; i++) {
3839 tracks[i].audio.pkt_cnt = 0;
3840 tracks[i].audio.sub_packet_cnt = 0;
3841 tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
3842 tracks[i].end_timecode = 0;
3845 /* We seek to a level 1 element, so set the appropriate status. */
3846 matroska_reset_status(matroska, 0, st->internal->index_entries[index].pos);
3847 if (flags & AVSEEK_FLAG_ANY) {
3848 st->internal->skip_to_keyframe = 0;
3849 matroska->skip_to_timecode = timestamp;
3851 st->internal->skip_to_keyframe = 1;
3852 matroska->skip_to_timecode = st->internal->index_entries[index].timestamp;
3854 matroska->skip_to_keyframe = 1;
3856 ff_update_cur_dts(s, st, st->internal->index_entries[index].timestamp);
3859 // slightly hackish but allows proper fallback to
3860 // the generic seeking code.
3861 matroska_reset_status(matroska, 0, -1);
3862 matroska->resync_pos = -1;
3863 matroska_clear_queue(matroska);
3864 st->internal->skip_to_keyframe =
3865 matroska->skip_to_keyframe = 0;
3870 static int matroska_read_close(AVFormatContext *s)
3872 MatroskaDemuxContext *matroska = s->priv_data;
3873 MatroskaTrack *tracks = matroska->tracks.elem;
3876 matroska_clear_queue(matroska);
3878 for (n = 0; n < matroska->tracks.nb_elem; n++)
3879 if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
3880 av_freep(&tracks[n].audio.buf);
3881 ebml_free(matroska_segment, matroska);
3887 int64_t start_time_ns;
3888 int64_t end_time_ns;
3889 int64_t start_offset;
3893 /* This function searches all the Cues and returns the CueDesc corresponding to
3894 * the timestamp ts. Returned CueDesc will be such that start_time_ns <= ts <
3895 * end_time_ns. All 4 fields will be set to -1 if ts >= file's duration.
3897 static CueDesc get_cue_desc(AVFormatContext *s, int64_t ts, int64_t cues_start) {
3898 MatroskaDemuxContext *matroska = s->priv_data;
3901 int nb_index_entries = s->streams[0]->internal->nb_index_entries;
3902 AVIndexEntry *index_entries = s->streams[0]->internal->index_entries;
3903 if (ts >= matroska->duration * matroska->time_scale) return (CueDesc) {-1, -1, -1, -1};
3904 for (i = 1; i < nb_index_entries; i++) {
3905 if (index_entries[i - 1].timestamp * matroska->time_scale <= ts &&
3906 index_entries[i].timestamp * matroska->time_scale > ts) {
3911 cue_desc.start_time_ns = index_entries[i].timestamp * matroska->time_scale;
3912 cue_desc.start_offset = index_entries[i].pos - matroska->segment_start;
3913 if (i != nb_index_entries - 1) {
3914 cue_desc.end_time_ns = index_entries[i + 1].timestamp * matroska->time_scale;
3915 cue_desc.end_offset = index_entries[i + 1].pos - matroska->segment_start;
3917 cue_desc.end_time_ns = matroska->duration * matroska->time_scale;
3918 // FIXME: this needs special handling for files where Cues appear
3919 // before Clusters. the current logic assumes Cues appear after
3921 cue_desc.end_offset = cues_start - matroska->segment_start;
3926 static int webm_clusters_start_with_keyframe(AVFormatContext *s)
3928 MatroskaDemuxContext *matroska = s->priv_data;
3929 uint32_t id = matroska->current_id;
3930 int64_t cluster_pos, before_pos;
3932 if (s->streams[0]->internal->nb_index_entries <= 0) return 0;
3933 // seek to the first cluster using cues.
3934 index = av_index_search_timestamp(s->streams[0], 0, 0);
3935 if (index < 0) return 0;
3936 cluster_pos = s->streams[0]->internal->index_entries[index].pos;
3937 before_pos = avio_tell(s->pb);
3939 uint64_t cluster_id, cluster_length;
3942 avio_seek(s->pb, cluster_pos, SEEK_SET);
3943 // read cluster id and length
3944 read = ebml_read_num(matroska, matroska->ctx->pb, 4, &cluster_id, 1);
3945 if (read < 0 || cluster_id != 0xF43B675) // done with all clusters
3947 read = ebml_read_length(matroska, matroska->ctx->pb, &cluster_length);
3951 matroska_reset_status(matroska, 0, cluster_pos);
3952 matroska_clear_queue(matroska);
3953 if (matroska_parse_cluster(matroska) < 0 ||
3957 pkt = &matroska->queue->pkt;
3958 // 4 + read is the length of the cluster id and the cluster length field.
3959 cluster_pos += 4 + read + cluster_length;
3960 if (!(pkt->flags & AV_PKT_FLAG_KEY)) {
3966 /* Restore the status after matroska_read_header: */
3967 matroska_reset_status(matroska, id, before_pos);
3972 static int buffer_size_after_time_downloaded(int64_t time_ns, double search_sec, int64_t bps,
3973 double min_buffer, double* buffer,
3974 double* sec_to_download, AVFormatContext *s,
3977 double nano_seconds_per_second = 1000000000.0;
3978 double time_sec = time_ns / nano_seconds_per_second;
3980 int64_t time_to_search_ns = (int64_t)(search_sec * nano_seconds_per_second);
3981 int64_t end_time_ns = time_ns + time_to_search_ns;
3982 double sec_downloaded = 0.0;
3983 CueDesc desc_curr = get_cue_desc(s, time_ns, cues_start);
3984 if (desc_curr.start_time_ns == -1)
3986 *sec_to_download = 0.0;
3988 // Check for non cue start time.
3989 if (time_ns > desc_curr.start_time_ns) {
3990 int64_t cue_nano = desc_curr.end_time_ns - time_ns;
3991 double percent = (double)(cue_nano) / (desc_curr.end_time_ns - desc_curr.start_time_ns);
3992 double cueBytes = (desc_curr.end_offset - desc_curr.start_offset) * percent;
3993 double timeToDownload = (cueBytes * 8.0) / bps;
3995 sec_downloaded += (cue_nano / nano_seconds_per_second) - timeToDownload;
3996 *sec_to_download += timeToDownload;
3998 // Check if the search ends within the first cue.
3999 if (desc_curr.end_time_ns >= end_time_ns) {
4000 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
4001 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
4002 sec_downloaded = percent_to_sub * sec_downloaded;
4003 *sec_to_download = percent_to_sub * *sec_to_download;
4006 if ((sec_downloaded + *buffer) <= min_buffer) {
4010 // Get the next Cue.
4011 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
4014 while (desc_curr.start_time_ns != -1) {
4015 int64_t desc_bytes = desc_curr.end_offset - desc_curr.start_offset;
4016 int64_t desc_ns = desc_curr.end_time_ns - desc_curr.start_time_ns;
4017 double desc_sec = desc_ns / nano_seconds_per_second;
4018 double bits = (desc_bytes * 8.0);
4019 double time_to_download = bits / bps;
4021 sec_downloaded += desc_sec - time_to_download;
4022 *sec_to_download += time_to_download;
4024 if (desc_curr.end_time_ns >= end_time_ns) {
4025 double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
4026 double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
4027 sec_downloaded = percent_to_sub * sec_downloaded;
4028 *sec_to_download = percent_to_sub * *sec_to_download;
4030 if ((sec_downloaded + *buffer) <= min_buffer)
4035 if ((sec_downloaded + *buffer) <= min_buffer) {
4040 desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
4042 *buffer = *buffer + sec_downloaded;
4046 /* This function computes the bandwidth of the WebM file with the help of
4047 * buffer_size_after_time_downloaded() function. Both of these functions are
4048 * adapted from WebM Tools project and are adapted to work with FFmpeg's
4049 * Matroska parsing mechanism.
4051 * Returns the bandwidth of the file on success; -1 on error.
4053 static int64_t webm_dash_manifest_compute_bandwidth(AVFormatContext *s, int64_t cues_start)
4055 MatroskaDemuxContext *matroska = s->priv_data;
4056 AVStream *st = s->streams[0];
4057 double bandwidth = 0.0;
4060 for (i = 0; i < st->internal->nb_index_entries; i++) {
4061 int64_t prebuffer_ns = 1000000000;
4062 int64_t time_ns = st->internal->index_entries[i].timestamp * matroska->time_scale;
4063 double nano_seconds_per_second = 1000000000.0;
4064 int64_t prebuffered_ns = time_ns + prebuffer_ns;
4065 double prebuffer_bytes = 0.0;
4066 int64_t temp_prebuffer_ns = prebuffer_ns;
4067 int64_t pre_bytes, pre_ns;
4068 double pre_sec, prebuffer, bits_per_second;
4069 CueDesc desc_beg = get_cue_desc(s, time_ns, cues_start);
4071 // Start with the first Cue.
4072 CueDesc desc_end = desc_beg;
4074 // Figure out how much data we have downloaded for the prebuffer. This will
4075 // be used later to adjust the bits per sample to try.
4076 while (desc_end.start_time_ns != -1 && desc_end.end_time_ns < prebuffered_ns) {
4077 // Prebuffered the entire Cue.
4078 prebuffer_bytes += desc_end.end_offset - desc_end.start_offset;
4079 temp_prebuffer_ns -= desc_end.end_time_ns - desc_end.start_time_ns;
4080 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4082 if (desc_end.start_time_ns == -1) {
4083 // The prebuffer is larger than the duration.
4084 if (matroska->duration * matroska->time_scale >= prebuffered_ns)
4086 bits_per_second = 0.0;
4088 // The prebuffer ends in the last Cue. Estimate how much data was
4090 pre_bytes = desc_end.end_offset - desc_end.start_offset;
4091 pre_ns = desc_end.end_time_ns - desc_end.start_time_ns;
4092 pre_sec = pre_ns / nano_seconds_per_second;
4094 pre_bytes * ((temp_prebuffer_ns / nano_seconds_per_second) / pre_sec);
4096 prebuffer = prebuffer_ns / nano_seconds_per_second;
4098 // Set this to 0.0 in case our prebuffer buffers the entire video.
4099 bits_per_second = 0.0;
4101 int64_t desc_bytes = desc_end.end_offset - desc_beg.start_offset;
4102 int64_t desc_ns = desc_end.end_time_ns - desc_beg.start_time_ns;
4103 double desc_sec = desc_ns / nano_seconds_per_second;
4104 double calc_bits_per_second = (desc_bytes * 8) / desc_sec;
4106 // Drop the bps by the percentage of bytes buffered.
4107 double percent = (desc_bytes - prebuffer_bytes) / desc_bytes;
4108 double mod_bits_per_second = calc_bits_per_second * percent;
4110 if (prebuffer < desc_sec) {
4112 (double)(matroska->duration * matroska->time_scale) / nano_seconds_per_second;
4114 // Add 1 so the bits per second should be a little bit greater than file
4116 int64_t bps = (int64_t)(mod_bits_per_second) + 1;
4117 const double min_buffer = 0.0;
4118 double buffer = prebuffer;
4119 double sec_to_download = 0.0;
4121 int rv = buffer_size_after_time_downloaded(prebuffered_ns, search_sec, bps,
4122 min_buffer, &buffer, &sec_to_download,
4126 } else if (rv == 0) {
4127 bits_per_second = (double)(bps);
4132 desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4133 } while (desc_end.start_time_ns != -1);
4135 if (bandwidth < bits_per_second) bandwidth = bits_per_second;
4137 return (int64_t)bandwidth;
4140 static int webm_dash_manifest_cues(AVFormatContext *s, int64_t init_range)
4142 MatroskaDemuxContext *matroska = s->priv_data;
4143 EbmlList *seekhead_list = &matroska->seekhead;
4144 MatroskaSeekhead *seekhead = seekhead_list->elem;
4146 int64_t cues_start = -1, cues_end = -1, before_pos, bandwidth;
4150 // determine cues start and end positions
4151 for (i = 0; i < seekhead_list->nb_elem; i++)
4152 if (seekhead[i].id == MATROSKA_ID_CUES)
4155 if (i >= seekhead_list->nb_elem) return -1;
4157 before_pos = avio_tell(matroska->ctx->pb);
4158 cues_start = seekhead[i].pos + matroska->segment_start;
4159 if (avio_seek(matroska->ctx->pb, cues_start, SEEK_SET) == cues_start) {
4160 // cues_end is computed as cues_start + cues_length + length of the
4161 // Cues element ID (i.e. 4) + EBML length of the Cues element.
4162 // cues_end is inclusive and the above sum is reduced by 1.
4163 uint64_t cues_length, cues_id;
4165 bytes_read = ebml_read_num (matroska, matroska->ctx->pb, 4, &cues_id, 1);
4166 if (bytes_read < 0 || cues_id != (MATROSKA_ID_CUES & 0xfffffff))
4167 return bytes_read < 0 ? bytes_read : AVERROR_INVALIDDATA;
4168 bytes_read = ebml_read_length(matroska, matroska->ctx->pb, &cues_length);
4171 cues_end = cues_start + 4 + bytes_read + cues_length - 1;
4173 avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
4174 if (cues_start == -1 || cues_end == -1) return -1;
4177 matroska_parse_cues(matroska);
4180 av_dict_set_int(&s->streams[0]->metadata, CUES_START, cues_start, 0);
4183 av_dict_set_int(&s->streams[0]->metadata, CUES_END, cues_end, 0);
4185 // if the file has cues at the start, fix up the init range so that
4186 // it does not include it
4187 if (cues_start <= init_range)
4188 av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, cues_start - 1, 0);
4191 bandwidth = webm_dash_manifest_compute_bandwidth(s, cues_start);
4192 if (bandwidth < 0) return -1;
4193 av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH, bandwidth, 0);
4195 // check if all clusters start with key frames
4196 av_dict_set_int(&s->streams[0]->metadata, CLUSTER_KEYFRAME, webm_clusters_start_with_keyframe(s), 0);
4198 // store cue point timestamps as a comma separated list for checking subsegment alignment in
4199 // the muxer. assumes that each timestamp cannot be more than 20 characters long.
4200 buf = av_malloc_array(s->streams[0]->internal->nb_index_entries, 20);
4201 if (!buf) return -1;
4203 for (i = 0; i < s->streams[0]->internal->nb_index_entries; i++) {
4204 int ret = snprintf(buf + end, 20,
4205 "%" PRId64"%s", s->streams[0]->internal->index_entries[i].timestamp,
4206 i != s->streams[0]->internal->nb_index_entries - 1 ? "," : "");
4207 if (ret <= 0 || (ret == 20 && i == s->streams[0]->internal->nb_index_entries - 1)) {
4208 av_log(s, AV_LOG_ERROR, "timestamp too long.\n");
4210 return AVERROR_INVALIDDATA;
4214 av_dict_set(&s->streams[0]->metadata, CUE_TIMESTAMPS,
4215 buf, AV_DICT_DONT_STRDUP_VAL);
4220 static int webm_dash_manifest_read_header(AVFormatContext *s)
4223 int ret = matroska_read_header(s);
4225 MatroskaTrack *tracks;
4226 MatroskaDemuxContext *matroska = s->priv_data;
4228 av_log(s, AV_LOG_ERROR, "Failed to read file headers\n");
4231 if (!matroska->tracks.nb_elem || !s->nb_streams) {
4232 av_log(s, AV_LOG_ERROR, "No track found\n");
4233 ret = AVERROR_INVALIDDATA;
4237 if (!matroska->is_live) {
4238 buf = av_asprintf("%g", matroska->duration);
4240 ret = AVERROR(ENOMEM);
4243 av_dict_set(&s->streams[0]->metadata, DURATION,
4244 buf, AV_DICT_DONT_STRDUP_VAL);
4246 // initialization range
4247 // 5 is the offset of Cluster ID.
4248 init_range = avio_tell(s->pb) - 5;
4249 av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, init_range, 0);
4252 // basename of the file
4253 buf = strrchr(s->url, '/');
4254 av_dict_set(&s->streams[0]->metadata, FILENAME, buf ? ++buf : s->url, 0);
4257 tracks = matroska->tracks.elem;
4258 av_dict_set_int(&s->streams[0]->metadata, TRACK_NUMBER, tracks[0].num, 0);
4260 // parse the cues and populate Cue related fields
4261 if (!matroska->is_live) {
4262 ret = webm_dash_manifest_cues(s, init_range);
4264 av_log(s, AV_LOG_ERROR, "Error parsing Cues\n");
4269 // use the bandwidth from the command line if it was provided
4270 if (matroska->bandwidth > 0) {
4271 av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH,
4272 matroska->bandwidth, 0);
4276 matroska_read_close(s);
4280 static int webm_dash_manifest_read_packet(AVFormatContext *s, AVPacket *pkt)
4285 #define OFFSET(x) offsetof(MatroskaDemuxContext, x)
4286 static const AVOption options[] = {
4287 { "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 },
4288 { "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 },
4292 static const AVClass webm_dash_class = {
4293 .class_name = "WebM DASH Manifest demuxer",
4294 .item_name = av_default_item_name,
4296 .version = LIBAVUTIL_VERSION_INT,
4299 AVInputFormat ff_matroska_demuxer = {
4300 .name = "matroska,webm",
4301 .long_name = NULL_IF_CONFIG_SMALL("Matroska / WebM"),
4302 .extensions = "mkv,mk3d,mka,mks",
4303 .priv_data_size = sizeof(MatroskaDemuxContext),
4304 .read_probe = matroska_probe,
4305 .read_header = matroska_read_header,
4306 .read_packet = matroska_read_packet,
4307 .read_close = matroska_read_close,
4308 .read_seek = matroska_read_seek,
4309 .mime_type = "audio/webm,audio/x-matroska,video/webm,video/x-matroska"
4312 AVInputFormat ff_webm_dash_manifest_demuxer = {
4313 .name = "webm_dash_manifest",
4314 .long_name = NULL_IF_CONFIG_SMALL("WebM DASH Manifest"),
4315 .priv_data_size = sizeof(MatroskaDemuxContext),
4316 .read_header = webm_dash_manifest_read_header,
4317 .read_packet = webm_dash_manifest_read_packet,
4318 .read_close = matroska_read_close,
4319 .priv_class = &webm_dash_class,