FFmpeg
matroskadec.c
Go to the documentation of this file.
1 /*
2  * Matroska file demuxer
3  * Copyright (c) 2003-2008 The FFmpeg Project
4  *
5  * This file is part of FFmpeg.
6  *
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.
11  *
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.
16  *
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
20  */
21 
22 /**
23  * @file
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/
29  */
30 
31 #include "config.h"
32 
33 #include <inttypes.h>
34 #include <stdio.h>
35 
36 #include "libavutil/avstring.h"
37 #include "libavutil/base64.h"
38 #include "libavutil/bprint.h"
39 #include "libavutil/dict.h"
40 #include "libavutil/intfloat.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavutil/lzo.h"
44 #include "libavutil/mathematics.h"
45 #include "libavutil/opt.h"
47 #include "libavutil/spherical.h"
48 
49 #include "libavcodec/bytestream.h"
50 #include "libavcodec/flac.h"
51 #include "libavcodec/mpeg4audio.h"
53 
54 #include "avformat.h"
55 #include "avio_internal.h"
56 #include "dovi_isom.h"
57 #include "internal.h"
58 #include "isom.h"
59 #include "matroska.h"
60 #include "oggdec.h"
61 /* For ff_codec_get_id(). */
62 #include "riff.h"
63 #include "rmsipr.h"
64 
65 #if CONFIG_BZLIB
66 #include <bzlib.h>
67 #endif
68 #if CONFIG_ZLIB
69 #include <zlib.h>
70 #endif
71 
72 #include "qtpalette.h"
73 
74 #define EBML_UNKNOWN_LENGTH UINT64_MAX /* EBML unknown length, in uint64_t */
75 #define NEEDS_CHECKING 2 /* Indicates that some error checks
76  * still need to be performed */
77 #define LEVEL_ENDED 3 /* return value of ebml_parse when the
78  * syntax level used for parsing ended. */
79 #define SKIP_THRESHOLD 1024 * 1024 /* In non-seekable mode, if more than SKIP_THRESHOLD
80  * of unkown, potentially damaged data is encountered,
81  * it is considered an error. */
82 #define UNKNOWN_EQUIV 50 * 1024 /* An unknown element is considered equivalent
83  * to this many bytes of unknown data for the
84  * SKIP_THRESHOLD check. */
85 
86 typedef enum {
92  EBML_UTF8,
93  EBML_BIN,
99 
100 typedef struct CountedElement {
101  union {
102  uint64_t u;
103  int64_t i;
104  double f;
105  char *s;
106  } el;
107  unsigned count;
109 
110 typedef const struct EbmlSyntax {
111  uint32_t id;
112  uint8_t type;
113  uint8_t is_counted;
115  size_t data_offset;
116  union {
117  int64_t i;
118  uint64_t u;
119  double f;
120  const char *s;
121  const struct EbmlSyntax *n;
122  } def;
123 } EbmlSyntax;
124 
125 typedef struct EbmlList {
126  int nb_elem;
127  unsigned int alloc_elem_size;
128  void *elem;
130 
131 typedef struct EbmlBin {
132  int size;
134  uint8_t *data;
135  int64_t pos;
137 
138 typedef struct Ebml {
139  uint64_t version;
140  uint64_t max_size;
141  uint64_t id_length;
142  char *doctype;
143  uint64_t doctype_version;
144 } Ebml;
145 
146 typedef struct MatroskaTrackCompression {
147  uint64_t algo;
150 
151 typedef struct MatroskaTrackEncryption {
152  uint64_t algo;
155 
156 typedef struct MatroskaTrackEncoding {
157  uint64_t scope;
158  uint64_t type;
162 
163 typedef struct MatroskaMasteringMeta {
164  double r_x;
165  double r_y;
166  double g_x;
167  double g_y;
168  double b_x;
169  double b_y;
170  double white_x;
171  double white_y;
175 
176 typedef struct MatroskaTrackVideoColor {
179  uint64_t chroma_sub_horz;
180  uint64_t chroma_sub_vert;
181  uint64_t cb_sub_horz;
182  uint64_t cb_sub_vert;
185  uint64_t range;
186  uint64_t transfer_characteristics;
187  uint64_t primaries;
188  uint64_t max_cll;
189  uint64_t max_fall;
192 
193 typedef struct MatroskaTrackVideoProjection {
194  uint64_t type;
195  EbmlBin private;
196  double yaw;
197  double pitch;
198  double roll;
200 
201 typedef struct MatroskaTrackVideo {
202  double frame_rate;
203  uint64_t display_width;
204  uint64_t display_height;
205  uint64_t pixel_width;
206  uint64_t pixel_height;
208  uint64_t display_unit;
209  uint64_t interlaced;
210  uint64_t field_order;
211  uint64_t stereo_mode;
212  uint64_t alpha_mode;
216 
217 typedef struct MatroskaTrackAudio {
218  double samplerate;
220  uint64_t bitdepth;
221  uint64_t channels;
222 
223  /* real audio header (extracted from extradata) */
226  int frame_size;
227  int sub_packet_size;
229  int pkt_cnt;
230  uint64_t buf_timecode;
231  uint8_t *buf;
233 
234 typedef struct MatroskaTrackPlane {
235  uint64_t uid;
236  uint64_t type;
238 
239 typedef struct MatroskaTrackOperation {
242 
243 typedef struct MatroskaBlockAdditionMapping {
244  uint64_t value;
245  char *name;
246  uint64_t type;
249 
250 typedef struct MatroskaTrack {
251  uint64_t num;
252  uint64_t uid;
253  uint64_t type;
254  char *name;
255  char *codec_id;
257  char *language;
258  double time_scale;
260  uint64_t flag_default;
261  uint64_t flag_forced;
262  uint64_t flag_comment;
267  uint64_t seek_preroll;
272  uint64_t codec_delay;
274 
275  AVStream *stream;
276  int64_t end_timecode;
278  int needs_decoding;
279  uint64_t max_block_additional_id;
281 
285 
286 typedef struct MatroskaAttachment {
287  uint64_t uid;
288  char *filename;
289  char *description;
290  char *mime;
292 
295 
296 typedef struct MatroskaChapter {
297  uint64_t start;
298  uint64_t end;
299  uint64_t uid;
300  char *title;
301 
304 
305 typedef struct MatroskaIndexPos {
306  uint64_t track;
307  uint64_t pos;
309 
310 typedef struct MatroskaIndex {
311  uint64_t time;
314 
315 typedef struct MatroskaTag {
316  char *name;
317  char *string;
318  char *lang;
319  uint64_t def;
322 
323 typedef struct MatroskaTagTarget {
324  char *type;
325  uint64_t typevalue;
326  uint64_t trackuid;
327  uint64_t chapteruid;
328  uint64_t attachuid;
330 
331 typedef struct MatroskaTags {
333  EbmlList tag;
334 } MatroskaTags;
335 
336 typedef struct MatroskaSeekhead {
337  uint64_t id;
338  uint64_t pos;
340 
341 typedef struct MatroskaLevel {
342  uint64_t start;
343  uint64_t length;
345 
346 typedef struct MatroskaBlock {
347  uint64_t duration;
349  uint64_t non_simple;
351  uint64_t additional_id;
354 } MatroskaBlock;
355 
356 typedef struct MatroskaCluster {
358  uint64_t timecode;
359  int64_t pos;
361 
362 typedef struct MatroskaLevel1Element {
363  int64_t pos;
364  uint32_t id;
365  int parsed;
367 
368 typedef struct MatroskaDemuxContext {
369  const AVClass *class;
371 
372  /* EBML stuff */
375  uint32_t current_id;
376  int64_t resync_pos;
378 
379  uint64_t time_scale;
380  double duration;
381  char *title;
382  char *muxingapp;
387  EbmlList index;
388  EbmlList tags;
390 
391  /* byte position of the segment inside the stream */
392  int64_t segment_start;
393 
394  /* This packet coincides with FFFormatContext.parse_pkt
395  * and is not owned by us. */
396  AVPacket *pkt;
397 
398  /* the packet queue */
400 
401  int done;
402 
403  /* What to skip before effectively reading a packet. */
404  int skip_to_keyframe;
406 
407  /* File has a CUES element, but we defer parsing until it is needed. */
409 
410  /* Level1 elements and whether they were read yet */
412  int num_level1_elems;
413 
415 
416  /* WebM DASH Manifest live flag */
417  int is_live;
418 
419  /* Bandwidth value for WebM DASH Manifest */
420  int bandwidth;
422 
423 #define CHILD_OF(parent) { .def = { .n = parent } }
424 
425 // The following forward declarations need their size because
426 // a tentative definition with internal linkage must not be an
427 // incomplete type (6.7.2 in C90, 6.9.2 in C99).
428 // Removing the sizes breaks MSVC.
435 
436 static EbmlSyntax ebml_header[] = {
437  { EBML_ID_EBMLREADVERSION, EBML_UINT, 0, 0, offsetof(Ebml, version), { .u = EBML_VERSION } },
438  { EBML_ID_EBMLMAXSIZELENGTH, EBML_UINT, 0, 0, offsetof(Ebml, max_size), { .u = 8 } },
439  { EBML_ID_EBMLMAXIDLENGTH, EBML_UINT, 0, 0, offsetof(Ebml, id_length), { .u = 4 } },
440  { EBML_ID_DOCTYPE, EBML_STR, 0, 0, offsetof(Ebml, doctype), { .s = "(none)" } },
441  { EBML_ID_DOCTYPEREADVERSION, EBML_UINT, 0, 0, offsetof(Ebml, doctype_version), { .u = 1 } },
445 };
446 
448  { EBML_ID_HEADER, EBML_NEST, 0, 0, 0, { .n = ebml_header } },
450  { 0 }
451 };
452 
453 static EbmlSyntax matroska_info[] = {
454  { MATROSKA_ID_TIMECODESCALE, EBML_UINT, 0, 0, offsetof(MatroskaDemuxContext, time_scale), { .u = 1000000 } },
456  { MATROSKA_ID_TITLE, EBML_UTF8, 0, 0, offsetof(MatroskaDemuxContext, title) },
458  { MATROSKA_ID_MUXINGAPP, EBML_UTF8, 0, 0, offsetof(MatroskaDemuxContext, muxingapp) },
459  { MATROSKA_ID_DATEUTC, EBML_BIN, 0, 0, offsetof(MatroskaDemuxContext, date_utc) },
462 };
463 
471  { MATROSKA_ID_VIDEOCOLOR_WHITEX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, white_x) },
472  { MATROSKA_ID_VIDEOCOLOR_WHITEY, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, white_y) },
473  { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMIN, EBML_FLOAT, 1, 0, offsetof(MatroskaMasteringMeta, min_luminance) },
474  { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMAX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, max_luminance) },
476 };
477 
479  { MATROSKA_ID_VIDEOCOLORMATRIXCOEFF, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, matrix_coefficients), { .u = AVCOL_SPC_UNSPECIFIED } },
480  { MATROSKA_ID_VIDEOCOLORBITSPERCHANNEL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, bits_per_channel), { .u = 0 } },
481  { MATROSKA_ID_VIDEOCOLORCHROMASUBHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_horz) },
482  { MATROSKA_ID_VIDEOCOLORCHROMASUBVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_vert) },
483  { MATROSKA_ID_VIDEOCOLORCBSUBHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, cb_sub_horz) },
484  { MATROSKA_ID_VIDEOCOLORCBSUBVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, cb_sub_vert) },
491  { MATROSKA_ID_VIDEOCOLORMAXFALL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, max_fall) },
494 };
495 
499  { MATROSKA_ID_VIDEOPROJECTIONPOSEYAW, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, yaw), { .f = 0.0 } },
500  { MATROSKA_ID_VIDEOPROJECTIONPOSEPITCH, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, pitch), { .f = 0.0 } },
501  { MATROSKA_ID_VIDEOPROJECTIONPOSEROLL, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, roll), { .f = 0.0 } },
503 };
504 
505 static EbmlSyntax matroska_track_video[] = {
506  { MATROSKA_ID_VIDEOFRAMERATE, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideo, frame_rate) },
507  { MATROSKA_ID_VIDEODISPLAYWIDTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_width), { .u=-1 } },
508  { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_height), { .u=-1 } },
509  { MATROSKA_ID_VIDEOPIXELWIDTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, pixel_width) },
510  { MATROSKA_ID_VIDEOPIXELHEIGHT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, pixel_height) },
511  { MATROSKA_ID_VIDEOCOLORSPACE, EBML_BIN, 0, 0, offsetof(MatroskaTrackVideo, color_space) },
512  { MATROSKA_ID_VIDEOALPHAMODE, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, alpha_mode), { .u = 0 } },
525 };
526 
527 static EbmlSyntax matroska_track_audio[] = {
528  { MATROSKA_ID_AUDIOSAMPLINGFREQ, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackAudio, samplerate), { .f = 8000.0 } },
529  { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackAudio, out_samplerate) },
530  { MATROSKA_ID_AUDIOBITDEPTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackAudio, bitdepth) },
531  { MATROSKA_ID_AUDIOCHANNELS, EBML_UINT, 0, 0, offsetof(MatroskaTrackAudio, channels), { .u = 1 } },
533 };
534 
539 };
540 
542  { MATROSKA_ID_ENCODINGENCALGO, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncryption,algo), {.u = 0} },
550 };
552  { MATROSKA_ID_ENCODINGSCOPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncoding, scope), { .u = 1 } },
553  { MATROSKA_ID_ENCODINGTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncoding, type), { .u = 0 } },
558 };
559 
563 };
564 
565 static EbmlSyntax matroska_track_plane[] = {
569 };
570 
572  { MATROSKA_ID_TRACKPLANE, EBML_NEST, 0, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n = matroska_track_plane} },
574 };
575 
579 };
580 
587 };
588 
589 static EbmlSyntax matroska_track[] = {
590  { MATROSKA_ID_TRACKNUMBER, EBML_UINT, 0, 0, offsetof(MatroskaTrack, num) },
591  { MATROSKA_ID_TRACKNAME, EBML_UTF8, 0, 0, offsetof(MatroskaTrack, name) },
592  { MATROSKA_ID_TRACKUID, EBML_UINT, 0, 0, offsetof(MatroskaTrack, uid) },
593  { MATROSKA_ID_TRACKTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrack, type) },
594  { MATROSKA_ID_CODECID, EBML_STR, 0, 0, offsetof(MatroskaTrack, codec_id) },
595  { MATROSKA_ID_CODECPRIVATE, EBML_BIN, 0, 0, offsetof(MatroskaTrack, codec_priv) },
596  { MATROSKA_ID_CODECDELAY, EBML_UINT, 0, 0, offsetof(MatroskaTrack, codec_delay), { .u = 0 } },
597  { MATROSKA_ID_TRACKLANGUAGE, EBML_STR, 0, 0, offsetof(MatroskaTrack, language), { .s = "eng" } },
598  { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, 0, offsetof(MatroskaTrack, default_duration) },
599  { MATROSKA_ID_TRACKTIMECODESCALE, EBML_FLOAT, 0, 0, offsetof(MatroskaTrack, time_scale), { .f = 1.0 } },
600  { MATROSKA_ID_TRACKFLAGCOMMENTARY, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_comment), { .u = 0 } },
601  { MATROSKA_ID_TRACKFLAGDEFAULT, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_default), { .u = 1 } },
602  { MATROSKA_ID_TRACKFLAGFORCED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_forced), { .u = 0 } },
603  { MATROSKA_ID_TRACKFLAGHEARINGIMPAIRED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_hearingimpaired), { .u = 0 } },
604  { MATROSKA_ID_TRACKFLAGVISUALIMPAIRED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_visualimpaired), { .u = 0 } },
605  { MATROSKA_ID_TRACKFLAGTEXTDESCRIPTIONS, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_textdescriptions), { .u = 0 } },
606  { MATROSKA_ID_TRACKFLAGORIGINAL, EBML_UINT, 1, 0, offsetof(MatroskaTrack, flag_original), {.u = 0 } },
608  { MATROSKA_ID_TRACKAUDIO, EBML_NEST, 0, 0, offsetof(MatroskaTrack, audio), { .n = matroska_track_audio } },
609  { MATROSKA_ID_TRACKOPERATION, EBML_NEST, 0, 0, offsetof(MatroskaTrack, operation), { .n = matroska_track_operation } },
611  { MATROSKA_ID_TRACKMAXBLKADDID, EBML_UINT, 0, 0, offsetof(MatroskaTrack, max_block_additional_id), { .u = 0 } },
612  { MATROSKA_ID_TRACKBLKADDMAPPING, EBML_NEST, 0, sizeof(MatroskaBlockAdditionMapping), offsetof(MatroskaTrack, block_addition_mappings), { .n = matroska_block_addition_mapping } },
613  { MATROSKA_ID_SEEKPREROLL, EBML_UINT, 0, 0, offsetof(MatroskaTrack, seek_preroll), { .u = 0 } },
623 };
624 
625 static EbmlSyntax matroska_tracks[] = {
626  { MATROSKA_ID_TRACKENTRY, EBML_NEST, 0, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext, tracks), { .n = matroska_track } },
628 };
629 
630 static EbmlSyntax matroska_attachment[] = {
631  { MATROSKA_ID_FILEUID, EBML_UINT, 0, 0, offsetof(MatroskaAttachment, uid) },
632  { MATROSKA_ID_FILENAME, EBML_UTF8, 0, 0, offsetof(MatroskaAttachment, filename) },
633  { MATROSKA_ID_FILEMIMETYPE, EBML_STR, 0, 0, offsetof(MatroskaAttachment, mime) },
634  { MATROSKA_ID_FILEDATA, EBML_BIN, 0, 0, offsetof(MatroskaAttachment, bin) },
637 };
638 
639 static EbmlSyntax matroska_attachments[] = {
640  { MATROSKA_ID_ATTACHEDFILE, EBML_NEST, 0, sizeof(MatroskaAttachment), offsetof(MatroskaDemuxContext, attachments), { .n = matroska_attachment } },
642 };
643 
645  { MATROSKA_ID_CHAPSTRING, EBML_UTF8, 0, 0, offsetof(MatroskaChapter, title) },
649 };
650 
652  { MATROSKA_ID_CHAPTERTIMESTART, EBML_UINT, 0, 0, offsetof(MatroskaChapter, start), { .u = AV_NOPTS_VALUE } },
653  { MATROSKA_ID_CHAPTERTIMEEND, EBML_UINT, 0, 0, offsetof(MatroskaChapter, end), { .u = AV_NOPTS_VALUE } },
654  { MATROSKA_ID_CHAPTERUID, EBML_UINT, 0, 0, offsetof(MatroskaChapter, uid) },
661 };
662 
663 static EbmlSyntax matroska_chapter[] = {
664  { MATROSKA_ID_CHAPTERATOM, EBML_NEST, 0, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext, chapters), { .n = matroska_chapter_entry } },
670 };
671 
672 static EbmlSyntax matroska_chapters[] = {
673  { MATROSKA_ID_EDITIONENTRY, EBML_NEST, 0, 0, 0, { .n = matroska_chapter } },
675 };
676 
677 static EbmlSyntax matroska_index_pos[] = {
678  { MATROSKA_ID_CUETRACK, EBML_UINT, 0, 0, offsetof(MatroskaIndexPos, track) },
684 };
685 
686 static EbmlSyntax matroska_index_entry[] = {
687  { MATROSKA_ID_CUETIME, EBML_UINT, 0, 0, offsetof(MatroskaIndex, time) },
690 };
691 
692 static EbmlSyntax matroska_index[] = {
695 };
696 
697 static EbmlSyntax matroska_simpletag[] = {
698  { MATROSKA_ID_TAGNAME, EBML_UTF8, 0, 0, offsetof(MatroskaTag, name) },
699  { MATROSKA_ID_TAGSTRING, EBML_UTF8, 0, 0, offsetof(MatroskaTag, string) },
700  { MATROSKA_ID_TAGLANG, EBML_STR, 0, 0, offsetof(MatroskaTag, lang), { .s = "und" } },
701  { MATROSKA_ID_TAGDEFAULT, EBML_UINT, 0, 0, offsetof(MatroskaTag, def) },
702  { MATROSKA_ID_TAGDEFAULT_BUG, EBML_UINT, 0, 0, offsetof(MatroskaTag, def) },
703  { MATROSKA_ID_SIMPLETAG, EBML_NEST, 0, sizeof(MatroskaTag), offsetof(MatroskaTag, sub), { .n = matroska_simpletag } },
705 };
706 
707 static EbmlSyntax matroska_tagtargets[] = {
709  { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, typevalue), { .u = 50 } },
710  { MATROSKA_ID_TAGTARGETS_TRACKUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, trackuid), { .u = 0 } },
711  { MATROSKA_ID_TAGTARGETS_CHAPTERUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, chapteruid), { .u = 0 } },
712  { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, attachuid), { .u = 0 } },
714 };
715 
716 static EbmlSyntax matroska_tag[] = {
717  { MATROSKA_ID_SIMPLETAG, EBML_NEST, 0, sizeof(MatroskaTag), offsetof(MatroskaTags, tag), { .n = matroska_simpletag } },
718  { MATROSKA_ID_TAGTARGETS, EBML_NEST, 0, 0, offsetof(MatroskaTags, target), { .n = matroska_tagtargets } },
720 };
721 
722 static EbmlSyntax matroska_tags[] = {
723  { MATROSKA_ID_TAG, EBML_NEST, 0, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext, tags), { .n = matroska_tag } },
725 };
726 
728  { MATROSKA_ID_SEEKID, EBML_UINT, 0, 0, offsetof(MatroskaSeekhead, id) },
729  { MATROSKA_ID_SEEKPOSITION, EBML_UINT, 0, 0, offsetof(MatroskaSeekhead, pos), { .u = -1 } },
731 };
732 
733 static EbmlSyntax matroska_seekhead[] = {
734  { MATROSKA_ID_SEEKENTRY, EBML_NEST, 0, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext, seekhead), { .n = matroska_seekhead_entry } },
736 };
737 
738 static EbmlSyntax matroska_segment[] = {
740  { MATROSKA_ID_INFO, EBML_LEVEL1, 0, 0, 0, { .n = matroska_info } },
741  { MATROSKA_ID_TRACKS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_tracks } },
743  { MATROSKA_ID_CHAPTERS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_chapters } },
744  { MATROSKA_ID_CUES, EBML_LEVEL1, 0, 0, 0, { .n = matroska_index } },
745  { MATROSKA_ID_TAGS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_tags } },
746  { MATROSKA_ID_SEEKHEAD, EBML_LEVEL1, 0, 0, 0, { .n = matroska_seekhead } },
747  { 0 } /* We don't want to go back to level 0, so don't add the parent. */
748 };
749 
750 static EbmlSyntax matroska_segments[] = {
751  { MATROSKA_ID_SEGMENT, EBML_NEST, 0, 0, 0, { .n = matroska_segment } },
752  { 0 }
753 };
754 
755 static EbmlSyntax matroska_blockmore[] = {
756  { MATROSKA_ID_BLOCKADDID, EBML_UINT, 0, 0, offsetof(MatroskaBlock,additional_id), { .u = 1 } },
757  { MATROSKA_ID_BLOCKADDITIONAL, EBML_BIN, 0, 0, offsetof(MatroskaBlock,additional) },
759 };
760 
762  { MATROSKA_ID_BLOCKMORE, EBML_NEST, 0, 0, 0, {.n = matroska_blockmore} },
764 };
765 
766 static EbmlSyntax matroska_blockgroup[] = {
767  { MATROSKA_ID_BLOCK, EBML_BIN, 0, 0, offsetof(MatroskaBlock, bin) },
770  { MATROSKA_ID_DISCARDPADDING, EBML_SINT, 0, 0, offsetof(MatroskaBlock, discard_padding) },
771  { MATROSKA_ID_BLOCKREFERENCE, EBML_SINT, 1, 0, offsetof(MatroskaBlock, reference) },
773  { 1, EBML_UINT, 0, 0, offsetof(MatroskaBlock, non_simple), { .u = 1 } },
775 };
776 
777 // The following array contains SimpleBlock and BlockGroup twice
778 // in order to reuse the other values for matroska_cluster_enter.
780  { MATROSKA_ID_SIMPLEBLOCK, EBML_BIN, 0, 0, offsetof(MatroskaBlock, bin) },
781  { MATROSKA_ID_BLOCKGROUP, EBML_NEST, 0, 0, 0, { .n = matroska_blockgroup } },
782  { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, 0, offsetof(MatroskaCluster, timecode) },
788 };
789 
791  { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, 0, { .n = &matroska_cluster_parsing[2] } },
792  { 0 }
793 };
794 #undef CHILD_OF
795 
796 static const CodecMime mkv_image_mime_tags[] = {
797  {"image/gif" , AV_CODEC_ID_GIF},
798  {"image/jpeg" , AV_CODEC_ID_MJPEG},
799  {"image/png" , AV_CODEC_ID_PNG},
800  {"image/tiff" , AV_CODEC_ID_TIFF},
801 
802  {"" , AV_CODEC_ID_NONE}
803 };
804 
805 static const CodecMime mkv_mime_tags[] = {
806  {"text/plain" , AV_CODEC_ID_TEXT},
807  {"application/x-truetype-font", AV_CODEC_ID_TTF},
808  {"application/x-font" , AV_CODEC_ID_TTF},
809  {"application/vnd.ms-opentype", AV_CODEC_ID_OTF},
810  {"binary" , AV_CODEC_ID_BIN_DATA},
811 
812  {"" , AV_CODEC_ID_NONE}
813 };
814 
815 static const char *const matroska_doctypes[] = { "matroska", "webm" };
816 
817 /*
818  * This function prepares the status for parsing of level 1 elements.
819  */
820 static int matroska_reset_status(MatroskaDemuxContext *matroska,
821  uint32_t id, int64_t position)
822 {
823  int64_t err = 0;
824  if (position >= 0) {
825  err = avio_seek(matroska->ctx->pb, position, SEEK_SET);
826  if (err > 0)
827  err = 0;
828  } else
829  position = avio_tell(matroska->ctx->pb);
830 
831  matroska->current_id = id;
832  matroska->num_levels = 1;
833  matroska->unknown_count = 0;
834  matroska->resync_pos = position;
835  if (id)
836  matroska->resync_pos -= (av_log2(id) + 7) / 8;
837 
838  return err;
839 }
840 
841 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
842 {
843  AVIOContext *pb = matroska->ctx->pb;
844  uint32_t id;
845 
846  /* Try to seek to the last position to resync from. If this doesn't work,
847  * we resync from the earliest position available: The start of the buffer. */
848  if (last_pos < avio_tell(pb) && avio_seek(pb, last_pos + 1, SEEK_SET) < 0) {
849  av_log(matroska->ctx, AV_LOG_WARNING,
850  "Seek to desired resync point failed. Seeking to "
851  "earliest point available instead.\n");
852  avio_seek(pb, FFMAX(avio_tell(pb) + (pb->buffer - pb->buf_ptr),
853  last_pos + 1), SEEK_SET);
854  }
855 
856  id = avio_rb32(pb);
857 
858  // try to find a toplevel element
859  while (!avio_feof(pb)) {
860  if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
861  id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
863  id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS) {
864  /* Prepare the context for parsing of a level 1 element. */
865  matroska_reset_status(matroska, id, -1);
866  /* Given that we are here means that an error has occurred,
867  * so treat the segment as unknown length in order not to
868  * discard valid data that happens to be beyond the designated
869  * end of the segment. */
870  matroska->levels[0].length = EBML_UNKNOWN_LENGTH;
871  return 0;
872  }
873  id = (id << 8) | avio_r8(pb);
874  }
875 
876  matroska->done = 1;
877  return pb->error ? pb->error : AVERROR_EOF;
878 }
879 
880 /*
881  * Read: an "EBML number", which is defined as a variable-length
882  * array of bytes. The first byte indicates the length by giving a
883  * number of 0-bits followed by a one. The position of the first
884  * "one" bit inside the first byte indicates the length of this
885  * number.
886  * Returns: number of bytes read, < 0 on error
887  */
888 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
889  int max_size, uint64_t *number, int eof_forbidden)
890 {
891  int read, n = 1;
892  uint64_t total;
893  int64_t pos;
894 
895  /* The first byte tells us the length in bytes - except when it is zero. */
896  total = avio_r8(pb);
897  if (pb->eof_reached)
898  goto err;
899 
900  /* get the length of the EBML number */
901  read = 8 - ff_log2_tab[total];
902 
903  if (!total || read > max_size) {
904  pos = avio_tell(pb) - 1;
905  if (!total) {
906  av_log(matroska->ctx, AV_LOG_ERROR,
907  "0x00 at pos %"PRId64" (0x%"PRIx64") invalid as first byte "
908  "of an EBML number\n", pos, pos);
909  } else {
910  av_log(matroska->ctx, AV_LOG_ERROR,
911  "Length %d indicated by an EBML number's first byte 0x%02x "
912  "at pos %"PRId64" (0x%"PRIx64") exceeds max length %d.\n",
913  read, (uint8_t) total, pos, pos, max_size);
914  }
915  return AVERROR_INVALIDDATA;
916  }
917 
918  /* read out length */
919  total ^= 1 << ff_log2_tab[total];
920  while (n++ < read)
921  total = (total << 8) | avio_r8(pb);
922 
923  if (pb->eof_reached) {
924  eof_forbidden = 1;
925  goto err;
926  }
927 
928  *number = total;
929 
930  return read;
931 
932 err:
933  pos = avio_tell(pb);
934  if (pb->error) {
935  av_log(matroska->ctx, AV_LOG_ERROR,
936  "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
937  pos, pos);
938  return pb->error;
939  }
940  if (eof_forbidden) {
941  av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
942  "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
943  return AVERROR(EIO);
944  }
945  return AVERROR_EOF;
946 }
947 
948 /**
949  * Read a EBML length value.
950  * This needs special handling for the "unknown length" case which has multiple
951  * encodings.
952  */
953 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
954  uint64_t *number)
955 {
956  int res = ebml_read_num(matroska, pb, 8, number, 1);
957  if (res > 0 && *number + 1 == 1ULL << (7 * res))
958  *number = EBML_UNKNOWN_LENGTH;
959  return res;
960 }
961 
962 /*
963  * Read the next element as an unsigned int.
964  * Returns NEEDS_CHECKING unless size == 0.
965  */
966 static int ebml_read_uint(AVIOContext *pb, int size,
967  uint64_t default_value, uint64_t *num)
968 {
969  int n = 0;
970 
971  if (size == 0) {
972  *num = default_value;
973  return 0;
974  }
975  /* big-endian ordering; build up number */
976  *num = 0;
977  while (n++ < size)
978  *num = (*num << 8) | avio_r8(pb);
979 
980  return NEEDS_CHECKING;
981 }
982 
983 /*
984  * Read the next element as a signed int.
985  * Returns NEEDS_CHECKING unless size == 0.
986  */
987 static int ebml_read_sint(AVIOContext *pb, int size,
988  int64_t default_value, int64_t *num)
989 {
990  int n = 1;
991 
992  if (size == 0) {
993  *num = default_value;
994  return 0;
995  } else {
996  *num = sign_extend(avio_r8(pb), 8);
997 
998  /* big-endian ordering; build up number */
999  while (n++ < size)
1000  *num = ((uint64_t)*num << 8) | avio_r8(pb);
1001  }
1002 
1003  return NEEDS_CHECKING;
1005 
1006 /*
1007  * Read the next element as a float.
1008  * Returns 0 if size == 0, NEEDS_CHECKING or < 0 on obvious failure.
1009  */
1010 static int ebml_read_float(AVIOContext *pb, int size,
1011  double default_value, double *num)
1012 {
1013  if (size == 0) {
1014  *num = default_value;
1015  return 0;
1016  } else if (size == 4) {
1017  *num = av_int2float(avio_rb32(pb));
1018  } else if (size == 8) {
1019  *num = av_int2double(avio_rb64(pb));
1020  } else
1021  return AVERROR_INVALIDDATA;
1022 
1023  return NEEDS_CHECKING;
1025 
1026 /*
1027  * Read the next element as an ASCII string.
1028  * 0 is success, < 0 or NEEDS_CHECKING is failure.
1029  */
1030 static int ebml_read_ascii(AVIOContext *pb, int size,
1031  const char *default_value, char **str)
1032 {
1033  char *res;
1034  int ret;
1035 
1036  if (size == 0 && default_value) {
1037  res = av_strdup(default_value);
1038  if (!res)
1039  return AVERROR(ENOMEM);
1040  } else {
1041  /* EBML strings are usually not 0-terminated, so we allocate one
1042  * byte more, read the string and NUL-terminate it ourselves. */
1043  if (!(res = av_malloc(size + 1)))
1044  return AVERROR(ENOMEM);
1045  if ((ret = avio_read(pb, (uint8_t *) res, size)) != size) {
1046  av_free(res);
1047  return ret < 0 ? ret : NEEDS_CHECKING;
1048  }
1049  (res)[size] = '\0';
1050  }
1051  av_free(*str);
1052  *str = res;
1053 
1054  return 0;
1056 
1057 /*
1058  * Read the next element as binary data.
1059  * 0 is success, < 0 or NEEDS_CHECKING is failure.
1060  */
1061 static int ebml_read_binary(AVIOContext *pb, int length,
1062  int64_t pos, EbmlBin *bin)
1063 {
1064  int ret;
1065 
1067  if (ret < 0)
1068  return ret;
1069  memset(bin->buf->data + length, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1070 
1071  bin->data = bin->buf->data;
1072  bin->size = length;
1073  bin->pos = pos;
1074  if ((ret = avio_read(pb, bin->data, length)) != length) {
1075  av_buffer_unref(&bin->buf);
1076  bin->data = NULL;
1077  bin->size = 0;
1078  return ret < 0 ? ret : NEEDS_CHECKING;
1079  }
1080 
1081  return 0;
1082 }
1084 /*
1085  * Read the next element, but only the header. The contents
1086  * are supposed to be sub-elements which can be read separately.
1087  * 0 is success, < 0 is failure.
1088  */
1089 static int ebml_read_master(MatroskaDemuxContext *matroska,
1090  uint64_t length, int64_t pos)
1091 {
1093 
1094  if (matroska->num_levels >= EBML_MAX_DEPTH) {
1095  av_log(matroska->ctx, AV_LOG_ERROR,
1096  "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
1097  return AVERROR(ENOSYS);
1098  }
1099 
1100  level = &matroska->levels[matroska->num_levels++];
1101  level->start = pos;
1102  level->length = length;
1103 
1104  return 0;
1106 
1107 /*
1108  * Read a signed "EBML number"
1109  * Return: number of bytes processed, < 0 on error
1110  */
1111 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
1112  AVIOContext *pb, int64_t *num)
1113 {
1114  uint64_t unum;
1115  int res;
1116 
1117  /* read as unsigned number first */
1118  if ((res = ebml_read_num(matroska, pb, 8, &unum, 1)) < 0)
1119  return res;
1120 
1121  /* make signed (weird way) */
1122  *num = unum - ((1LL << (7 * res - 1)) - 1);
1123 
1124  return res;
1125 }
1126 
1127 static int ebml_parse(MatroskaDemuxContext *matroska,
1128  EbmlSyntax *syntax, void *data);
1129 
1130 static EbmlSyntax *ebml_parse_id(EbmlSyntax *syntax, uint32_t id)
1131 {
1132  int i;
1133 
1134  // Whoever touches this should be aware of the duplication
1135  // existing in matroska_cluster_parsing.
1136  for (i = 0; syntax[i].id; i++)
1137  if (id == syntax[i].id)
1138  break;
1139 
1140  return &syntax[i];
1141 }
1142 
1143 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
1144  void *data)
1145 {
1146  int res;
1147 
1148  if (data) {
1149  for (int i = 0; syntax[i].id; i++) {
1150  void *dst = (char *)data + syntax[i].data_offset;
1151  switch (syntax[i].type) {
1152  case EBML_UINT:
1153  *(uint64_t *)dst = syntax[i].def.u;
1154  break;
1155  case EBML_SINT:
1156  *(int64_t *) dst = syntax[i].def.i;
1157  break;
1158  case EBML_FLOAT:
1159  *(double *) dst = syntax[i].def.f;
1160  break;
1161  case EBML_STR:
1162  case EBML_UTF8:
1163  // the default may be NULL
1164  if (syntax[i].def.s) {
1165  *(char**)dst = av_strdup(syntax[i].def.s);
1166  if (!*(char**)dst)
1167  return AVERROR(ENOMEM);
1168  }
1169  break;
1170  }
1171  }
1172 
1173  if (!matroska->levels[matroska->num_levels - 1].length) {
1174  matroska->num_levels--;
1175  return 0;
1176  }
1177  }
1178 
1179  do {
1180  res = ebml_parse(matroska, syntax, data);
1181  } while (!res);
1182 
1183  return res == LEVEL_ENDED ? 0 : res;
1184 }
1185 
1186 static int is_ebml_id_valid(uint32_t id)
1187 {
1188  // Due to endian nonsense in Matroska, the highest byte with any bits set
1189  // will contain the leading length bit. This bit in turn identifies the
1190  // total byte length of the element by its position within the byte.
1191  unsigned int bits = av_log2(id);
1192  return id && (bits + 7) / 8 == (8 - bits % 8);
1194 
1195 /*
1196  * Allocate and return the entry for the level1 element with the given ID. If
1197  * an entry already exists, return the existing entry.
1198  */
1200  uint32_t id, int64_t pos)
1201 {
1202  int i;
1203  MatroskaLevel1Element *elem;
1204 
1205  if (!is_ebml_id_valid(id))
1206  return NULL;
1207 
1208  // Some files link to all clusters; useless.
1209  if (id == MATROSKA_ID_CLUSTER)
1210  return NULL;
1211 
1212  // There can be multiple SeekHeads and Tags.
1213  for (i = 0; i < matroska->num_level1_elems; i++) {
1214  if (matroska->level1_elems[i].id == id) {
1215  if (matroska->level1_elems[i].pos == pos ||
1216  id != MATROSKA_ID_SEEKHEAD && id != MATROSKA_ID_TAGS)
1217  return &matroska->level1_elems[i];
1218  }
1219  }
1220 
1221  // Only a completely broken file would have more elements.
1222  if (matroska->num_level1_elems >= FF_ARRAY_ELEMS(matroska->level1_elems)) {
1223  av_log(matroska->ctx, AV_LOG_ERROR, "Too many level1 elements.\n");
1224  return NULL;
1225  }
1226 
1227  elem = &matroska->level1_elems[matroska->num_level1_elems++];
1228  *elem = (MatroskaLevel1Element){.id = id};
1229 
1230  return elem;
1231 }
1232 
1233 static int ebml_parse(MatroskaDemuxContext *matroska,
1234  EbmlSyntax *syntax, void *data)
1235 {
1236  static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
1237  // Forbid unknown-length EBML_NONE elements.
1239  [EBML_UINT] = 8,
1240  [EBML_SINT] = 8,
1241  [EBML_FLOAT] = 8,
1242  // max. 16 MB for strings
1243  [EBML_STR] = 0x1000000,
1244  [EBML_UTF8] = 0x1000000,
1245  // max. 256 MB for binary data
1246  [EBML_BIN] = 0x10000000,
1247  // no limits for anything else
1248  };
1249  AVIOContext *pb = matroska->ctx->pb;
1250  uint32_t id;
1251  uint64_t length;
1252  int64_t pos = avio_tell(pb), pos_alt;
1253  int res, update_pos = 1, level_check;
1254  MatroskaLevel1Element *level1_elem;
1255  MatroskaLevel *level = matroska->num_levels ? &matroska->levels[matroska->num_levels - 1] : NULL;
1256 
1257  if (!matroska->current_id) {
1258  uint64_t id;
1259  res = ebml_read_num(matroska, pb, 4, &id, 0);
1260  if (res < 0) {
1261  if (pb->eof_reached && res == AVERROR_EOF) {
1262  if (matroska->is_live)
1263  // in live mode, finish parsing if EOF is reached.
1264  return 1;
1265  if (level && pos == avio_tell(pb)) {
1266  if (level->length == EBML_UNKNOWN_LENGTH) {
1267  // Unknown-length levels automatically end at EOF.
1268  matroska->num_levels--;
1269  return LEVEL_ENDED;
1270  } else {
1271  av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
1272  "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
1273  }
1274  }
1275  }
1276  return res;
1277  }
1278  matroska->current_id = id | 1 << 7 * res;
1279  pos_alt = pos + res;
1280  } else {
1281  pos_alt = pos;
1282  pos -= (av_log2(matroska->current_id) + 7) / 8;
1283  }
1284 
1285  id = matroska->current_id;
1286 
1287  syntax = ebml_parse_id(syntax, id);
1288  if (!syntax->id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
1289  if (level && level->length == EBML_UNKNOWN_LENGTH) {
1290  // Unknown-length levels end when an element from an upper level
1291  // in the hierarchy is encountered.
1292  while (syntax->def.n) {
1293  syntax = ebml_parse_id(syntax->def.n, id);
1294  if (syntax->id) {
1295  matroska->num_levels--;
1296  return LEVEL_ENDED;
1297  }
1298  };
1299  }
1300 
1301  av_log(matroska->ctx, AV_LOG_DEBUG, "Unknown entry 0x%"PRIX32" at pos. "
1302  "%"PRId64"\n", id, pos);
1303  update_pos = 0; /* Don't update resync_pos as an error might have happened. */
1304  }
1305 
1306  if (data) {
1307  data = (char *) data + syntax->data_offset;
1308  if (syntax->list_elem_size) {
1309  EbmlList *list = data;
1310  void *newelem;
1311 
1312  if ((unsigned)list->nb_elem + 1 >= UINT_MAX / syntax->list_elem_size)
1313  return AVERROR(ENOMEM);
1314  newelem = av_fast_realloc(list->elem,
1315  &list->alloc_elem_size,
1316  (list->nb_elem + 1) * syntax->list_elem_size);
1317  if (!newelem)
1318  return AVERROR(ENOMEM);
1319  list->elem = newelem;
1320  data = (char *) list->elem + list->nb_elem * syntax->list_elem_size;
1321  memset(data, 0, syntax->list_elem_size);
1322  list->nb_elem++;
1323  }
1324  }
1325 
1326  if (syntax->type != EBML_STOP) {
1327  matroska->current_id = 0;
1328  if ((res = ebml_read_length(matroska, pb, &length)) < 0)
1329  return res;
1330 
1331  pos_alt += res;
1332 
1333  if (matroska->num_levels > 0) {
1334  if (length != EBML_UNKNOWN_LENGTH &&
1335  level->length != EBML_UNKNOWN_LENGTH) {
1336  uint64_t elem_end = pos_alt + length,
1337  level_end = level->start + level->length;
1338 
1339  if (elem_end < level_end) {
1340  level_check = 0;
1341  } else if (elem_end == level_end) {
1342  level_check = LEVEL_ENDED;
1343  } else {
1344  av_log(matroska->ctx, AV_LOG_ERROR,
1345  "Element at 0x%"PRIx64" ending at 0x%"PRIx64" exceeds "
1346  "containing master element ending at 0x%"PRIx64"\n",
1347  pos, elem_end, level_end);
1348  return AVERROR_INVALIDDATA;
1349  }
1350  } else if (length != EBML_UNKNOWN_LENGTH) {
1351  level_check = 0;
1352  } else if (level->length != EBML_UNKNOWN_LENGTH) {
1353  av_log(matroska->ctx, AV_LOG_ERROR, "Unknown-sized element "
1354  "at 0x%"PRIx64" inside parent with finite size\n", pos);
1355  return AVERROR_INVALIDDATA;
1356  } else {
1357  level_check = 0;
1358  if (id != MATROSKA_ID_CLUSTER && (syntax->type == EBML_LEVEL1
1359  || syntax->type == EBML_NEST)) {
1360  // According to the current specifications only clusters and
1361  // segments are allowed to be unknown-length. We also accept
1362  // other unknown-length master elements.
1363  av_log(matroska->ctx, AV_LOG_WARNING,
1364  "Found unknown-length element 0x%"PRIX32" other than "
1365  "a cluster at 0x%"PRIx64". Spec-incompliant, but "
1366  "parsing will nevertheless be attempted.\n", id, pos);
1367  update_pos = -1;
1368  }
1369  }
1370  } else
1371  level_check = 0;
1372 
1373  if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
1374  if (length != EBML_UNKNOWN_LENGTH) {
1375  av_log(matroska->ctx, AV_LOG_ERROR,
1376  "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for element "
1377  "with ID 0x%"PRIX32" at 0x%"PRIx64"\n",
1378  length, max_lengths[syntax->type], id, pos);
1379  } else if (syntax->type != EBML_NONE) {
1380  av_log(matroska->ctx, AV_LOG_ERROR,
1381  "Element with ID 0x%"PRIX32" at pos. 0x%"PRIx64" has "
1382  "unknown length, yet the length of an element of its "
1383  "type must be known.\n", id, pos);
1384  } else {
1385  av_log(matroska->ctx, AV_LOG_ERROR,
1386  "Found unknown-length element with ID 0x%"PRIX32" at "
1387  "pos. 0x%"PRIx64" for which no syntax for parsing is "
1388  "available.\n", id, pos);
1389  }
1390  return AVERROR_INVALIDDATA;
1391  }
1392 
1393  if (!(pb->seekable & AVIO_SEEKABLE_NORMAL)) {
1394  // Loosing sync will likely manifest itself as encountering unknown
1395  // elements which are not reliably distinguishable from elements
1396  // belonging to future extensions of the format.
1397  // We use a heuristic to detect such situations: If the current
1398  // element is not expected at the current syntax level and there
1399  // were only a few unknown elements in a row, then the element is
1400  // skipped or considered defective based upon the length of the
1401  // current element (i.e. how much would be skipped); if there were
1402  // more than a few skipped elements in a row and skipping the current
1403  // element would lead us more than SKIP_THRESHOLD away from the last
1404  // known good position, then it is inferred that an error occurred.
1405  // The dependency on the number of unknown elements in a row exists
1406  // because the distance to the last known good position is
1407  // automatically big if the last parsed element was big.
1408  // In both cases, each unknown element is considered equivalent to
1409  // UNKNOWN_EQUIV of skipped bytes for the check.
1410  // The whole check is only done for non-seekable output, because
1411  // in this situation skipped data can't simply be rechecked later.
1412  // This is especially important when using unkown length elements
1413  // as the check for whether a child exceeds its containing master
1414  // element is not effective in this situation.
1415  if (update_pos) {
1416  matroska->unknown_count = 0;
1417  } else {
1418  int64_t dist = length + UNKNOWN_EQUIV * matroska->unknown_count++;
1419 
1420  if (matroska->unknown_count > 3)
1421  dist += pos_alt - matroska->resync_pos;
1422 
1423  if (dist > SKIP_THRESHOLD) {
1424  av_log(matroska->ctx, AV_LOG_ERROR,
1425  "Unknown element %"PRIX32" at pos. 0x%"PRIx64" with "
1426  "length 0x%"PRIx64" considered as invalid data. Last "
1427  "known good position 0x%"PRIx64", %d unknown elements"
1428  " in a row\n", id, pos, length, matroska->resync_pos,
1429  matroska->unknown_count);
1430  return AVERROR_INVALIDDATA;
1431  }
1432  }
1433  }
1434 
1435  if (update_pos > 0) {
1436  // We have found an element that is allowed at this place
1437  // in the hierarchy and it passed all checks, so treat the beginning
1438  // of the element as the "last known good" position.
1439  matroska->resync_pos = pos;
1440  }
1441 
1442  if (!data && length != EBML_UNKNOWN_LENGTH)
1443  goto skip;
1444  }
1445 
1446  switch (syntax->type) {
1447  case EBML_UINT:
1448  res = ebml_read_uint(pb, length, syntax->def.u, data);
1449  break;
1450  case EBML_SINT:
1451  res = ebml_read_sint(pb, length, syntax->def.i, data);
1452  break;
1453  case EBML_FLOAT:
1454  res = ebml_read_float(pb, length, syntax->def.f, data);
1455  break;
1456  case EBML_STR:
1457  case EBML_UTF8:
1458  res = ebml_read_ascii(pb, length, syntax->def.s, data);
1459  break;
1460  case EBML_BIN:
1461  res = ebml_read_binary(pb, length, pos_alt, data);
1462  break;
1463  case EBML_LEVEL1:
1464  case EBML_NEST:
1465  if ((res = ebml_read_master(matroska, length, pos_alt)) < 0)
1466  return res;
1467  if (id == MATROSKA_ID_SEGMENT)
1468  matroska->segment_start = pos_alt;
1469  if (id == MATROSKA_ID_CUES)
1470  matroska->cues_parsing_deferred = 0;
1471  if (syntax->type == EBML_LEVEL1 &&
1472  (level1_elem = matroska_find_level1_elem(matroska, syntax->id, pos))) {
1473  if (!level1_elem->pos) {
1474  // Zero is not a valid position for a level 1 element.
1475  level1_elem->pos = pos;
1476  } else if (level1_elem->pos != pos)
1477  av_log(matroska->ctx, AV_LOG_ERROR, "Duplicate element\n");
1478  level1_elem->parsed = 1;
1479  }
1480  if (res = ebml_parse_nest(matroska, syntax->def.n, data))
1481  return res;
1482  break;
1483  case EBML_STOP:
1484  return 1;
1485  skip:
1486  default:
1487  if (length) {
1488  int64_t res2;
1489  if (ffio_limit(pb, length) != length) {
1490  // ffio_limit emits its own error message,
1491  // so we don't have to.
1492  return AVERROR(EIO);
1493  }
1494  if ((res2 = avio_skip(pb, length - 1)) >= 0) {
1495  // avio_skip might take us past EOF. We check for this
1496  // by skipping only length - 1 bytes, reading a byte and
1497  // checking the error flags. This is done in order to check
1498  // that the element has been properly skipped even when
1499  // no filesize (that ffio_limit relies on) is available.
1500  avio_r8(pb);
1501  res = NEEDS_CHECKING;
1502  } else
1503  res = res2;
1504  } else
1505  res = 0;
1506  }
1507  if (res) {
1508  if (res == NEEDS_CHECKING) {
1509  if (pb->eof_reached) {
1510  if (pb->error)
1511  res = pb->error;
1512  else
1513  res = AVERROR_EOF;
1514  } else
1515  goto level_check;
1516  }
1517 
1518  if (res == AVERROR_INVALIDDATA)
1519  av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
1520  else if (res == AVERROR(EIO))
1521  av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
1522  else if (res == AVERROR_EOF) {
1523  av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely\n");
1524  res = AVERROR(EIO);
1525  }
1526 
1527  return res;
1528  }
1529 
1530 level_check:
1531  if (syntax->is_counted && data) {
1532  CountedElement *elem = data;
1533  if (elem->count != UINT_MAX)
1534  elem->count++;
1535  }
1536 
1537  if (level_check == LEVEL_ENDED && matroska->num_levels) {
1538  level = &matroska->levels[matroska->num_levels - 1];
1539  pos = avio_tell(pb);
1540 
1541  // Given that pos >= level->start no check for
1542  // level->length != EBML_UNKNOWN_LENGTH is necessary.
1543  while (matroska->num_levels && pos == level->start + level->length) {
1544  matroska->num_levels--;
1545  level--;
1546  }
1547  }
1548 
1549  return level_check;
1550 }
1551 
1552 static void ebml_free(EbmlSyntax *syntax, void *data)
1553 {
1554  int i, j;
1555  for (i = 0; syntax[i].id; i++) {
1556  void *data_off = (char *) data + syntax[i].data_offset;
1557  switch (syntax[i].type) {
1558  case EBML_STR:
1559  case EBML_UTF8:
1560  av_freep(data_off);
1561  break;
1562  case EBML_BIN:
1563  av_buffer_unref(&((EbmlBin *) data_off)->buf);
1564  break;
1565  case EBML_LEVEL1:
1566  case EBML_NEST:
1567  if (syntax[i].list_elem_size) {
1568  EbmlList *list = data_off;
1569  char *ptr = list->elem;
1570  for (j = 0; j < list->nb_elem;
1571  j++, ptr += syntax[i].list_elem_size)
1572  ebml_free(syntax[i].def.n, ptr);
1573  av_freep(&list->elem);
1574  list->nb_elem = 0;
1575  list->alloc_elem_size = 0;
1576  } else
1577  ebml_free(syntax[i].def.n, data_off);
1578  default:
1579  break;
1580  }
1581  }
1582 }
1583 
1584 /*
1585  * Autodetecting...
1586  */
1587 static int matroska_probe(const AVProbeData *p)
1588 {
1589  uint64_t total = 0;
1590  int len_mask = 0x80, size = 1, n = 1, i;
1591 
1592  /* EBML header? */
1593  if (AV_RB32(p->buf) != EBML_ID_HEADER)
1594  return 0;
1595 
1596  /* length of header */
1597  total = p->buf[4];
1598  while (size <= 8 && !(total & len_mask)) {
1599  size++;
1600  len_mask >>= 1;
1601  }
1602  if (size > 8)
1603  return 0;
1604  total &= (len_mask - 1);
1605  while (n < size)
1606  total = (total << 8) | p->buf[4 + n++];
1607 
1608  if (total + 1 == 1ULL << (7 * size)){
1609  /* Unknown-length header - simply parse the whole buffer. */
1610  total = p->buf_size - 4 - size;
1611  } else {
1612  /* Does the probe data contain the whole header? */
1613  if (p->buf_size < 4 + size + total)
1614  return 0;
1615  }
1616 
1617  /* The header should contain a known document type. For now,
1618  * we don't parse the whole header but simply check for the
1619  * availability of that array of characters inside the header.
1620  * Not fully fool-proof, but good enough. */
1621  for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1622  size_t probelen = strlen(matroska_doctypes[i]);
1623  if (total < probelen)
1624  continue;
1625  for (n = 4 + size; n <= 4 + size + total - probelen; n++)
1626  if (!memcmp(p->buf + n, matroska_doctypes[i], probelen))
1627  return AVPROBE_SCORE_MAX;
1628  }
1629 
1630  // probably valid EBML header but no recognized doctype
1631  return AVPROBE_SCORE_EXTENSION;
1632 }
1633 
1635  uint64_t num)
1636 {
1637  MatroskaTrack *tracks = matroska->tracks.elem;
1638  int i;
1639 
1640  for (i = 0; i < matroska->tracks.nb_elem; i++)
1641  if (tracks[i].num == num)
1642  return &tracks[i];
1643 
1644  av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %"PRIu64"\n", num);
1645  return NULL;
1646 }
1647 
1648 static int matroska_decode_buffer(uint8_t **buf, int *buf_size,
1649  MatroskaTrack *track)
1650 {
1651  MatroskaTrackEncoding *encodings = track->encodings.elem;
1652  uint8_t *data = *buf;
1653  int isize = *buf_size;
1654  uint8_t *pkt_data = NULL;
1655  uint8_t av_unused *newpktdata;
1656  int pkt_size = isize;
1657  int result = 0;
1658  int olen;
1659 
1660  if (pkt_size >= 10000000U)
1661  return AVERROR_INVALIDDATA;
1662 
1663  switch (encodings[0].compression.algo) {
1665  {
1666  int header_size = encodings[0].compression.settings.size;
1667  uint8_t *header = encodings[0].compression.settings.data;
1668 
1669  if (header_size && !header) {
1670  av_log(NULL, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1671  return -1;
1672  }
1673 
1674  if (!header_size)
1675  return 0;
1676 
1677  pkt_size = isize + header_size;
1678  pkt_data = av_malloc(pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1679  if (!pkt_data)
1680  return AVERROR(ENOMEM);
1681 
1682  memcpy(pkt_data, header, header_size);
1683  memcpy(pkt_data + header_size, data, isize);
1684  break;
1685  }
1686 #if CONFIG_LZO
1688  do {
1689  int insize = isize;
1690  olen = pkt_size *= 3;
1691  newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING
1693  if (!newpktdata) {
1694  result = AVERROR(ENOMEM);
1695  goto failed;
1696  }
1697  pkt_data = newpktdata;
1698  result = av_lzo1x_decode(pkt_data, &olen, data, &insize);
1699  } while (result == AV_LZO_OUTPUT_FULL && pkt_size < 10000000);
1700  if (result) {
1702  goto failed;
1703  }
1704  pkt_size -= olen;
1705  break;
1706 #endif
1707 #if CONFIG_ZLIB
1709  {
1710  z_stream zstream = { 0 };
1711  if (!pkt_size || inflateInit(&zstream) != Z_OK)
1712  return -1;
1713  zstream.next_in = data;
1714  zstream.avail_in = isize;
1715  do {
1716  pkt_size *= 3;
1717  newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1718  if (!newpktdata) {
1719  inflateEnd(&zstream);
1720  result = AVERROR(ENOMEM);
1721  goto failed;
1722  }
1723  pkt_data = newpktdata;
1724  zstream.avail_out = pkt_size - zstream.total_out;
1725  zstream.next_out = pkt_data + zstream.total_out;
1726  result = inflate(&zstream, Z_NO_FLUSH);
1727  } while (result == Z_OK && pkt_size < 10000000);
1728  pkt_size = zstream.total_out;
1729  inflateEnd(&zstream);
1730  if (result != Z_STREAM_END) {
1731  if (result == Z_MEM_ERROR)
1732  result = AVERROR(ENOMEM);
1733  else
1735  goto failed;
1736  }
1737  break;
1738  }
1739 #endif
1740 #if CONFIG_BZLIB
1742  {
1743  bz_stream bzstream = { 0 };
1744  if (!pkt_size || BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1745  return -1;
1746  bzstream.next_in = data;
1747  bzstream.avail_in = isize;
1748  do {
1749  pkt_size *= 3;
1750  newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1751  if (!newpktdata) {
1752  BZ2_bzDecompressEnd(&bzstream);
1753  result = AVERROR(ENOMEM);
1754  goto failed;
1755  }
1756  pkt_data = newpktdata;
1757  bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1758  bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1759  result = BZ2_bzDecompress(&bzstream);
1760  } while (result == BZ_OK && pkt_size < 10000000);
1761  pkt_size = bzstream.total_out_lo32;
1762  BZ2_bzDecompressEnd(&bzstream);
1763  if (result != BZ_STREAM_END) {
1764  if (result == BZ_MEM_ERROR)
1765  result = AVERROR(ENOMEM);
1766  else
1768  goto failed;
1769  }
1770  break;
1771  }
1772 #endif
1773  default:
1774  return AVERROR_INVALIDDATA;
1775  }
1776 
1777  memset(pkt_data + pkt_size, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1778 
1779  *buf = pkt_data;
1780  *buf_size = pkt_size;
1781  return 0;
1783 failed:
1784  av_free(pkt_data);
1785  return result;
1786 }
1787 
1789  AVDictionary **metadata, char *prefix)
1790 {
1791  MatroskaTag *tags = list->elem;
1792  char key[1024];
1793  int i;
1794 
1795  for (i = 0; i < list->nb_elem; i++) {
1796  const char *lang = tags[i].lang &&
1797  strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1798 
1799  if (!tags[i].name) {
1800  av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1801  continue;
1802  }
1803  if (prefix)
1804  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1805  else
1806  av_strlcpy(key, tags[i].name, sizeof(key));
1807  if (tags[i].def || !lang) {
1808  av_dict_set(metadata, key, tags[i].string, 0);
1809  if (tags[i].sub.nb_elem)
1810  matroska_convert_tag(s, &tags[i].sub, metadata, key);
1811  }
1812  if (lang) {
1813  av_strlcat(key, "-", sizeof(key));
1814  av_strlcat(key, lang, sizeof(key));
1815  av_dict_set(metadata, key, tags[i].string, 0);
1816  if (tags[i].sub.nb_elem)
1817  matroska_convert_tag(s, &tags[i].sub, metadata, key);
1818  }
1819  }
1821 }
1822 
1824 {
1825  MatroskaDemuxContext *matroska = s->priv_data;
1826  MatroskaTags *tags = matroska->tags.elem;
1827  int i, j;
1828 
1829  for (i = 0; i < matroska->tags.nb_elem; i++) {
1830  if (tags[i].target.attachuid) {
1831  MatroskaAttachment *attachment = matroska->attachments.elem;
1832  int found = 0;
1833  for (j = 0; j < matroska->attachments.nb_elem; j++) {
1834  if (attachment[j].uid == tags[i].target.attachuid &&
1835  attachment[j].stream) {
1836  matroska_convert_tag(s, &tags[i].tag,
1837  &attachment[j].stream->metadata, NULL);
1838  found = 1;
1839  }
1840  }
1841  if (!found) {
1843  "The tags at index %d refer to a "
1844  "non-existent attachment %"PRId64".\n",
1845  i, tags[i].target.attachuid);
1846  }
1847  } else if (tags[i].target.chapteruid) {
1848  MatroskaChapter *chapter = matroska->chapters.elem;
1849  int found = 0;
1850  for (j = 0; j < matroska->chapters.nb_elem; j++) {
1851  if (chapter[j].uid == tags[i].target.chapteruid &&
1852  chapter[j].chapter) {
1853  matroska_convert_tag(s, &tags[i].tag,
1854  &chapter[j].chapter->metadata, NULL);
1855  found = 1;
1856  }
1857  }
1858  if (!found) {
1860  "The tags at index %d refer to a non-existent chapter "
1861  "%"PRId64".\n",
1862  i, tags[i].target.chapteruid);
1863  }
1864  } else if (tags[i].target.trackuid) {
1865  MatroskaTrack *track = matroska->tracks.elem;
1866  int found = 0;
1867  for (j = 0; j < matroska->tracks.nb_elem; j++) {
1868  if (track[j].uid == tags[i].target.trackuid &&
1869  track[j].stream) {
1870  matroska_convert_tag(s, &tags[i].tag,
1871  &track[j].stream->metadata, NULL);
1872  found = 1;
1873  }
1874  }
1875  if (!found) {
1877  "The tags at index %d refer to a non-existent track "
1878  "%"PRId64".\n",
1879  i, tags[i].target.trackuid);
1880  }
1881  } else {
1882  matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1883  tags[i].target.type);
1884  }
1885  }
1886 }
1887 
1889  int64_t pos)
1890 {
1891  uint32_t saved_id = matroska->current_id;
1892  int64_t before_pos = avio_tell(matroska->ctx->pb);
1893  int ret = 0;
1894  int ret2;
1895 
1896  /* seek */
1897  if (avio_seek(matroska->ctx->pb, pos, SEEK_SET) == pos) {
1898  /* We don't want to lose our seekhead level, so we add
1899  * a dummy. This is a crude hack. */
1900  if (matroska->num_levels == EBML_MAX_DEPTH) {
1901  av_log(matroska->ctx, AV_LOG_INFO,
1902  "Max EBML element depth (%d) reached, "
1903  "cannot parse further.\n", EBML_MAX_DEPTH);
1905  } else {
1906  matroska->levels[matroska->num_levels] = (MatroskaLevel) { 0, EBML_UNKNOWN_LENGTH };
1907  matroska->num_levels++;
1908  matroska->current_id = 0;
1909 
1910  ret = ebml_parse(matroska, matroska_segment, matroska);
1911  if (ret == LEVEL_ENDED) {
1912  /* This can only happen if the seek brought us beyond EOF. */
1913  ret = AVERROR_EOF;
1914  }
1915  }
1916  }
1917  /* Seek back - notice that in all instances where this is used
1918  * it is safe to set the level to 1. */
1919  ret2 = matroska_reset_status(matroska, saved_id, before_pos);
1920  if (ret >= 0)
1921  ret = ret2;
1922 
1923  return ret;
1924 }
1925 
1926 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1927 {
1928  EbmlList *seekhead_list = &matroska->seekhead;
1929  int i;
1930 
1931  // we should not do any seeking in the streaming case
1932  if (!(matroska->ctx->pb->seekable & AVIO_SEEKABLE_NORMAL))
1933  return;
1934 
1935  for (i = 0; i < seekhead_list->nb_elem; i++) {
1936  MatroskaSeekhead *seekheads = seekhead_list->elem;
1937  uint32_t id = seekheads[i].id;
1938  int64_t pos = seekheads[i].pos + matroska->segment_start;
1939  MatroskaLevel1Element *elem;
1940 
1941  if (id != seekheads[i].id || pos < matroska->segment_start)
1942  continue;
1943 
1944  elem = matroska_find_level1_elem(matroska, id, pos);
1945  if (!elem || elem->parsed)
1946  continue;
1947 
1948  elem->pos = pos;
1949 
1950  // defer cues parsing until we actually need cue data.
1951  if (id == MATROSKA_ID_CUES)
1952  continue;
1953 
1954  if (matroska_parse_seekhead_entry(matroska, pos) < 0) {
1955  // mark index as broken
1956  matroska->cues_parsing_deferred = -1;
1957  break;
1958  }
1959 
1960  elem->parsed = 1;
1961  }
1962 }
1963 
1964 static void matroska_add_index_entries(MatroskaDemuxContext *matroska)
1965 {
1966  EbmlList *index_list;
1968  uint64_t index_scale = 1;
1969  int i, j;
1970 
1971  if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
1972  return;
1973 
1974  index_list = &matroska->index;
1975  index = index_list->elem;
1976  if (index_list->nb_elem < 2)
1977  return;
1978  if (index[1].time > 1E14 / matroska->time_scale) {
1979  av_log(matroska->ctx, AV_LOG_WARNING, "Dropping apparently-broken index.\n");
1980  return;
1981  }
1982  for (i = 0; i < index_list->nb_elem; i++) {
1983  EbmlList *pos_list = &index[i].pos;
1984  MatroskaIndexPos *pos = pos_list->elem;
1985  for (j = 0; j < pos_list->nb_elem; j++) {
1986  MatroskaTrack *track = matroska_find_track_by_num(matroska,
1987  pos[j].track);
1988  if (track && track->stream)
1989  av_add_index_entry(track->stream,
1990  pos[j].pos + matroska->segment_start,
1991  index[i].time / index_scale, 0, 0,
1993  }
1994  }
1995 }
1996 
1997 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1998  int i;
1999 
2000  if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
2001  return;
2002 
2003  for (i = 0; i < matroska->num_level1_elems; i++) {
2004  MatroskaLevel1Element *elem = &matroska->level1_elems[i];
2005  if (elem->id == MATROSKA_ID_CUES && !elem->parsed) {
2006  if (matroska_parse_seekhead_entry(matroska, elem->pos) < 0)
2007  matroska->cues_parsing_deferred = -1;
2008  elem->parsed = 1;
2009  break;
2010  }
2011  }
2012 
2013  matroska_add_index_entries(matroska);
2014 }
2015 
2016 static int matroska_aac_profile(char *codec_id)
2017 {
2018  static const char *const aac_profiles[] = { "MAIN", "LC", "SSR" };
2019  int profile;
2020 
2022  if (strstr(codec_id, aac_profiles[profile]))
2023  break;
2024  return profile + 1;
2025 }
2026 
2027 static int matroska_aac_sri(int samplerate)
2028 {
2029  int sri;
2030 
2031  for (sri = 0; sri < FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
2032  if (ff_mpeg4audio_sample_rates[sri] == samplerate)
2033  break;
2034  return sri;
2035 }
2036 
2037 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
2038 {
2039  /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
2040  avpriv_dict_set_timestamp(metadata, "creation_time", date_utc / 1000 + 978307200000000LL);
2041 }
2042 
2044  MatroskaTrack *track,
2045  int *offset)
2046 {
2047  AVStream *st = track->stream;
2048  uint8_t *p = track->codec_priv.data;
2049  int size = track->codec_priv.size;
2050 
2051  if (size < 8 + FLAC_STREAMINFO_SIZE || p[4] & 0x7f) {
2052  av_log(s, AV_LOG_WARNING, "Invalid FLAC private data\n");
2053  track->codec_priv.size = 0;
2054  return 0;
2055  }
2056  *offset = 8;
2057  track->codec_priv.size = 8 + FLAC_STREAMINFO_SIZE;
2058 
2059  p += track->codec_priv.size;
2060  size -= track->codec_priv.size;
2061 
2062  /* parse the remaining metadata blocks if present */
2063  while (size >= 4) {
2064  int block_last, block_type, block_size;
2065 
2066  flac_parse_block_header(p, &block_last, &block_type, &block_size);
2067 
2068  p += 4;
2069  size -= 4;
2070  if (block_size > size)
2071  return 0;
2072 
2073  /* check for the channel mask */
2074  if (block_type == FLAC_METADATA_TYPE_VORBIS_COMMENT) {
2075  AVDictionary *dict = NULL;
2076  AVDictionaryEntry *chmask;
2077 
2078  ff_vorbis_comment(s, &dict, p, block_size, 0);
2079  chmask = av_dict_get(dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", NULL, 0);
2080  if (chmask) {
2081  uint64_t mask = strtol(chmask->value, NULL, 0);
2082  if (!mask || mask & ~0x3ffffULL) {
2084  "Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK\n");
2085  } else
2086  st->codecpar->channel_layout = mask;
2087  }
2088  av_dict_free(&dict);
2089  }
2090 
2091  p += block_size;
2092  size -= block_size;
2093  }
2094 
2095  return 0;
2096 }
2097 
2098 static int mkv_field_order(MatroskaDemuxContext *matroska, int64_t field_order)
2099 {
2100  int minor, micro, bttb = 0;
2101 
2102  /* workaround a bug in our Matroska muxer, introduced in version 57.36 alongside
2103  * this function, and fixed in 57.52 */
2104  if (matroska->muxingapp && sscanf(matroska->muxingapp, "Lavf57.%d.%d", &minor, &micro) == 2)
2105  bttb = (minor >= 36 && minor <= 51 && micro >= 100);
2106 
2107  switch (field_order) {
2109  return AV_FIELD_PROGRESSIVE;
2111  return AV_FIELD_UNKNOWN;
2113  return AV_FIELD_TT;
2115  return AV_FIELD_BB;
2117  return bttb ? AV_FIELD_TB : AV_FIELD_BT;
2119  return bttb ? AV_FIELD_BT : AV_FIELD_TB;
2120  default:
2121  return AV_FIELD_UNKNOWN;
2122  }
2123 }
2124 
2125 static void mkv_stereo_mode_display_mul(int stereo_mode,
2126  int *h_width, int *h_height)
2127 {
2128  switch (stereo_mode) {
2134  break;
2139  *h_width = 2;
2140  break;
2145  *h_height = 2;
2146  break;
2147  }
2148 }
2149 
2150 static int mkv_parse_video_color(AVStream *st, const MatroskaTrack *track) {
2151  const MatroskaTrackVideoColor *color = track->video.color.elem;
2152  const MatroskaMasteringMeta *mastering_meta;
2153  int has_mastering_primaries, has_mastering_luminance;
2154 
2155  if (!track->video.color.nb_elem)
2156  return 0;
2157 
2158  mastering_meta = &color->mastering_meta;
2159  // Mastering primaries are CIE 1931 coords, and must be > 0.
2160  has_mastering_primaries =
2161  mastering_meta->r_x > 0 && mastering_meta->r_y > 0 &&
2162  mastering_meta->g_x > 0 && mastering_meta->g_y > 0 &&
2163  mastering_meta->b_x > 0 && mastering_meta->b_y > 0 &&
2164  mastering_meta->white_x > 0 && mastering_meta->white_y > 0;
2165  has_mastering_luminance = mastering_meta->max_luminance >
2166  mastering_meta->min_luminance.el.f &&
2167  mastering_meta->min_luminance.el.f >= 0 &&
2168  mastering_meta->min_luminance.count;
2169 
2170  if (color->matrix_coefficients != AVCOL_SPC_RESERVED)
2171  st->codecpar->color_space = color->matrix_coefficients;
2172  if (color->primaries != AVCOL_PRI_RESERVED &&
2173  color->primaries != AVCOL_PRI_RESERVED0)
2174  st->codecpar->color_primaries = color->primaries;
2175  if (color->transfer_characteristics != AVCOL_TRC_RESERVED &&
2176  color->transfer_characteristics != AVCOL_TRC_RESERVED0)
2177  st->codecpar->color_trc = color->transfer_characteristics;
2178  if (color->range != AVCOL_RANGE_UNSPECIFIED &&
2179  color->range <= AVCOL_RANGE_JPEG)
2180  st->codecpar->color_range = color->range;
2181  if (color->chroma_siting_horz != MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED &&
2182  color->chroma_siting_vert != MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED &&
2183  color->chroma_siting_horz < MATROSKA_COLOUR_CHROMASITINGHORZ_NB &&
2184  color->chroma_siting_vert < MATROSKA_COLOUR_CHROMASITINGVERT_NB) {
2185  st->codecpar->chroma_location =
2186  avcodec_chroma_pos_to_enum((color->chroma_siting_horz - 1) << 7,
2187  (color->chroma_siting_vert - 1) << 7);
2188  }
2189  if (color->max_cll && color->max_fall) {
2190  size_t size = 0;
2191  int ret;
2193  if (!metadata)
2194  return AVERROR(ENOMEM);
2196  (uint8_t *)metadata, size);
2197  if (ret < 0) {
2198  av_freep(&metadata);
2199  return ret;
2200  }
2201  metadata->MaxCLL = color->max_cll;
2202  metadata->MaxFALL = color->max_fall;
2203  }
2204 
2205  if (has_mastering_primaries || has_mastering_luminance) {
2206  AVMasteringDisplayMetadata *metadata =
2209  sizeof(AVMasteringDisplayMetadata));
2210  if (!metadata) {
2211  return AVERROR(ENOMEM);
2212  }
2213  memset(metadata, 0, sizeof(AVMasteringDisplayMetadata));
2214  if (has_mastering_primaries) {
2215  metadata->display_primaries[0][0] = av_d2q(mastering_meta->r_x, INT_MAX);
2216  metadata->display_primaries[0][1] = av_d2q(mastering_meta->r_y, INT_MAX);
2217  metadata->display_primaries[1][0] = av_d2q(mastering_meta->g_x, INT_MAX);
2218  metadata->display_primaries[1][1] = av_d2q(mastering_meta->g_y, INT_MAX);
2219  metadata->display_primaries[2][0] = av_d2q(mastering_meta->b_x, INT_MAX);
2220  metadata->display_primaries[2][1] = av_d2q(mastering_meta->b_y, INT_MAX);
2221  metadata->white_point[0] = av_d2q(mastering_meta->white_x, INT_MAX);
2222  metadata->white_point[1] = av_d2q(mastering_meta->white_y, INT_MAX);
2223  metadata->has_primaries = 1;
2224  }
2225  if (has_mastering_luminance) {
2226  metadata->max_luminance = av_d2q(mastering_meta->max_luminance, INT_MAX);
2227  metadata->min_luminance = av_d2q(mastering_meta->min_luminance.el.f, INT_MAX);
2228  metadata->has_luminance = 1;
2229  }
2230  }
2231  return 0;
2232 }
2233 
2234 static int mkv_parse_video_projection(AVStream *st, const MatroskaTrack *track,
2235  void *logctx)
2236 {
2237  AVSphericalMapping *spherical;
2238  const MatroskaTrackVideoProjection *mkv_projection = &track->video.projection;
2239  const uint8_t *priv_data = mkv_projection->private.data;
2240  enum AVSphericalProjection projection;
2241  size_t spherical_size;
2242  uint32_t l = 0, t = 0, r = 0, b = 0;
2243  uint32_t padding = 0;
2244  int ret;
2245 
2246  if (mkv_projection->private.size && priv_data[0] != 0) {
2247  av_log(logctx, AV_LOG_WARNING, "Unknown spherical metadata\n");
2248  return 0;
2249  }
2250 
2251  switch (track->video.projection.type) {
2253  if (track->video.projection.private.size == 20) {
2254  t = AV_RB32(priv_data + 4);
2255  b = AV_RB32(priv_data + 8);
2256  l = AV_RB32(priv_data + 12);
2257  r = AV_RB32(priv_data + 16);
2258 
2259  if (b >= UINT_MAX - t || r >= UINT_MAX - l) {
2260  av_log(logctx, AV_LOG_ERROR,
2261  "Invalid bounding rectangle coordinates "
2262  "%"PRIu32",%"PRIu32",%"PRIu32",%"PRIu32"\n",
2263  l, t, r, b);
2264  return AVERROR_INVALIDDATA;
2265  }
2266  } else if (track->video.projection.private.size != 0) {
2267  av_log(logctx, AV_LOG_ERROR, "Unknown spherical metadata\n");
2268  return AVERROR_INVALIDDATA;
2269  }
2270 
2271  if (l || t || r || b)
2272  projection = AV_SPHERICAL_EQUIRECTANGULAR_TILE;
2273  else
2274  projection = AV_SPHERICAL_EQUIRECTANGULAR;
2275  break;
2277  if (track->video.projection.private.size < 4) {
2278  av_log(logctx, AV_LOG_ERROR, "Missing projection private properties\n");
2279  return AVERROR_INVALIDDATA;
2280  } else if (track->video.projection.private.size == 12) {
2281  uint32_t layout = AV_RB32(priv_data + 4);
2282  if (layout) {
2283  av_log(logctx, AV_LOG_WARNING,
2284  "Unknown spherical cubemap layout %"PRIu32"\n", layout);
2285  return 0;
2286  }
2287  projection = AV_SPHERICAL_CUBEMAP;
2288  padding = AV_RB32(priv_data + 8);
2289  } else {
2290  av_log(logctx, AV_LOG_ERROR, "Unknown spherical metadata\n");
2291  return AVERROR_INVALIDDATA;
2292  }
2293  break;
2295  /* No Spherical metadata */
2296  return 0;
2297  default:
2298  av_log(logctx, AV_LOG_WARNING,
2299  "Unknown spherical metadata type %"PRIu64"\n",
2300  track->video.projection.type);
2301  return 0;
2302  }
2303 
2304  spherical = av_spherical_alloc(&spherical_size);
2305  if (!spherical)
2306  return AVERROR(ENOMEM);
2307 
2308  spherical->projection = projection;
2309 
2310  spherical->yaw = (int32_t) (track->video.projection.yaw * (1 << 16));
2311  spherical->pitch = (int32_t) (track->video.projection.pitch * (1 << 16));
2312  spherical->roll = (int32_t) (track->video.projection.roll * (1 << 16));
2313 
2314  spherical->padding = padding;
2315 
2316  spherical->bound_left = l;
2317  spherical->bound_top = t;
2318  spherical->bound_right = r;
2319  spherical->bound_bottom = b;
2320 
2321  ret = av_stream_add_side_data(st, AV_PKT_DATA_SPHERICAL, (uint8_t *)spherical,
2322  spherical_size);
2323  if (ret < 0) {
2324  av_freep(&spherical);
2325  return ret;
2326  }
2327 
2328  return 0;
2329 }
2330 
2332  EbmlBin *bin)
2333 {
2334  return ff_isom_parse_dvcc_dvvc(s, st, bin->data, bin->size);
2335 }
2336 
2338 {
2339  const EbmlList *mappings_list = &track->block_addition_mappings;
2340  MatroskaBlockAdditionMapping *mappings = mappings_list->elem;
2341  int ret;
2342 
2343  for (int i = 0; i < mappings_list->nb_elem; i++) {
2344  MatroskaBlockAdditionMapping *mapping = &mappings[i];
2345 
2346  switch (mapping->type) {
2347  case MKBETAG('d','v','c','C'):
2348  case MKBETAG('d','v','v','C'):
2349  if ((ret = mkv_parse_dvcc_dvvc(s, st, track, &mapping->extradata)) < 0)
2350  return ret;
2351 
2352  break;
2353  default:
2355  "Unknown block additional mapping type 0x%"PRIx64", value %"PRIu64", name \"%s\"\n",
2356  mapping->type, mapping->value, mapping->name ? mapping->name : "");
2357  }
2358  }
2359 
2360  return 0;
2361 }
2362 
2363 static int get_qt_codec(MatroskaTrack *track, uint32_t *fourcc, enum AVCodecID *codec_id)
2364 {
2365  const AVCodecTag *codec_tags;
2366 
2367  codec_tags = track->type == MATROSKA_TRACK_TYPE_VIDEO ?
2369 
2370  /* Normalize noncompliant private data that starts with the fourcc
2371  * by expanding/shifting the data by 4 bytes and storing the data
2372  * size at the start. */
2373  if (ff_codec_get_id(codec_tags, AV_RL32(track->codec_priv.data))) {
2374  int ret = av_buffer_realloc(&track->codec_priv.buf,
2376  if (ret < 0)
2377  return ret;
2378 
2379  track->codec_priv.data = track->codec_priv.buf->data;
2380  memmove(track->codec_priv.data + 4, track->codec_priv.data, track->codec_priv.size);
2381  track->codec_priv.size += 4;
2382  AV_WB32(track->codec_priv.data, track->codec_priv.size);
2383  }
2384 
2385  *fourcc = AV_RL32(track->codec_priv.data + 4);
2386  *codec_id = ff_codec_get_id(codec_tags, *fourcc);
2387 
2388  return 0;
2389 }
2390 
2392 {
2393  MatroskaDemuxContext *matroska = s->priv_data;
2394  MatroskaTrack *tracks = matroska->tracks.elem;
2395  int i, j, ret;
2396  int k;
2397 
2398  for (i = 0; i < matroska->tracks.nb_elem; i++) {
2399  MatroskaTrack *track = &tracks[i];
2401  EbmlList *encodings_list = &track->encodings;
2402  MatroskaTrackEncoding *encodings = encodings_list->elem;
2403  uint8_t *extradata = NULL;
2404  int extradata_size = 0;
2405  int extradata_offset = 0;
2406  uint32_t fourcc = 0;
2407  FFIOContext b;
2408  AVStream *st;
2409  FFStream *sti;
2410  char* key_id_base64 = NULL;
2411  int bit_depth = -1;
2412 
2413  /* Apply some sanity checks. */
2414  if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
2415  track->type != MATROSKA_TRACK_TYPE_AUDIO &&
2416  track->type != MATROSKA_TRACK_TYPE_SUBTITLE &&
2417  track->type != MATROSKA_TRACK_TYPE_METADATA) {
2418  av_log(matroska->ctx, AV_LOG_INFO,
2419  "Unknown or unsupported track type %"PRIu64"\n",
2420  track->type);
2421  continue;
2422  }
2423  if (!track->codec_id)
2424  continue;
2425 
2426  if ( track->type == MATROSKA_TRACK_TYPE_AUDIO && track->codec_id[0] != 'A'
2427  || track->type == MATROSKA_TRACK_TYPE_VIDEO && track->codec_id[0] != 'V'
2428  || track->type == MATROSKA_TRACK_TYPE_SUBTITLE && track->codec_id[0] != 'D' && track->codec_id[0] != 'S'
2429  || track->type == MATROSKA_TRACK_TYPE_METADATA && track->codec_id[0] != 'D' && track->codec_id[0] != 'S'
2430  ) {
2431  av_log(matroska->ctx, AV_LOG_INFO, "Inconsistent track type\n");
2432  continue;
2433  }
2434 
2435  if (track->audio.samplerate < 0 || track->audio.samplerate > INT_MAX ||
2436  isnan(track->audio.samplerate)) {
2437  av_log(matroska->ctx, AV_LOG_WARNING,
2438  "Invalid sample rate %f, defaulting to 8000 instead.\n",
2439  track->audio.samplerate);
2440  track->audio.samplerate = 8000;
2441  }
2442 
2443  if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2444  if (!track->default_duration && track->video.frame_rate > 0) {
2445  double default_duration = 1000000000 / track->video.frame_rate;
2446  if (default_duration > UINT64_MAX || default_duration < 0) {
2447  av_log(matroska->ctx, AV_LOG_WARNING,
2448  "Invalid frame rate %e. Cannot calculate default duration.\n",
2449  track->video.frame_rate);
2450  } else {
2451  track->default_duration = default_duration;
2452  }
2453  }
2454  if (track->video.display_width == -1)
2455  track->video.display_width = track->video.pixel_width;
2456  if (track->video.display_height == -1)
2457  track->video.display_height = track->video.pixel_height;
2458  if (track->video.color_space.size == 4)
2459  fourcc = AV_RL32(track->video.color_space.data);
2460  } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2461  if (!track->audio.out_samplerate)
2462  track->audio.out_samplerate = track->audio.samplerate;
2463  }
2464  if (encodings_list->nb_elem > 1) {
2465  av_log(matroska->ctx, AV_LOG_ERROR,
2466  "Multiple combined encodings not supported");
2467  } else if (encodings_list->nb_elem == 1) {
2468  if (encodings[0].type) {
2469  if (encodings[0].encryption.key_id.size > 0) {
2470  /* Save the encryption key id to be stored later as a
2471  metadata tag. */
2472  const int b64_size = AV_BASE64_SIZE(encodings[0].encryption.key_id.size);
2473  key_id_base64 = av_malloc(b64_size);
2474  if (key_id_base64 == NULL)
2475  return AVERROR(ENOMEM);
2476 
2477  av_base64_encode(key_id_base64, b64_size,
2478  encodings[0].encryption.key_id.data,
2479  encodings[0].encryption.key_id.size);
2480  } else {
2481  encodings[0].scope = 0;
2482  av_log(matroska->ctx, AV_LOG_ERROR,
2483  "Unsupported encoding type");
2484  }
2485  } else if (
2486 #if CONFIG_ZLIB
2487  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
2488 #endif
2489 #if CONFIG_BZLIB
2490  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
2491 #endif
2492 #if CONFIG_LZO
2493  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO &&
2494 #endif
2495  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP) {
2496  encodings[0].scope = 0;
2497  av_log(matroska->ctx, AV_LOG_ERROR,
2498  "Unsupported encoding type");
2499  } else if (track->codec_priv.size && encodings[0].scope & 2) {
2500  uint8_t *codec_priv = track->codec_priv.data;
2501  int ret = matroska_decode_buffer(&track->codec_priv.data,
2502  &track->codec_priv.size,
2503  track);
2504  if (ret < 0) {
2505  track->codec_priv.data = NULL;
2506  track->codec_priv.size = 0;
2507  av_log(matroska->ctx, AV_LOG_ERROR,
2508  "Failed to decode codec private data\n");
2509  }
2510 
2511  if (codec_priv != track->codec_priv.data) {
2512  av_buffer_unref(&track->codec_priv.buf);
2513  if (track->codec_priv.data) {
2514  track->codec_priv.buf = av_buffer_create(track->codec_priv.data,
2516  NULL, NULL, 0);
2517  if (!track->codec_priv.buf) {
2518  av_freep(&track->codec_priv.data);
2519  track->codec_priv.size = 0;
2520  return AVERROR(ENOMEM);
2521  }
2522  }
2523  }
2524  }
2525  }
2526  track->needs_decoding = encodings && !encodings[0].type &&
2527  encodings[0].scope & 1 &&
2528  (encodings[0].compression.algo !=
2530  encodings[0].compression.settings.size);
2531 
2532  for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
2533  if (av_strstart(track->codec_id, ff_mkv_codec_tags[j].str, NULL)) {
2535  break;
2536  }
2537  }
2538 
2539  st = track->stream = avformat_new_stream(s, NULL);
2540  if (!st) {
2541  av_free(key_id_base64);
2542  return AVERROR(ENOMEM);
2543  }
2544  sti = ffstream(st);
2545 
2546  if (key_id_base64) {
2547  /* export encryption key id as base64 metadata tag */
2548  av_dict_set(&st->metadata, "enc_key_id", key_id_base64,
2550  }
2551 
2552  if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC") &&
2553  track->codec_priv.size >= 40 &&
2554  track->codec_priv.data) {
2555  track->ms_compat = 1;
2556  bit_depth = AV_RL16(track->codec_priv.data + 14);
2557  fourcc = AV_RL32(track->codec_priv.data + 16);
2559  fourcc);
2560  if (!codec_id)
2562  fourcc);
2563  extradata_offset = 40;
2564  } else if (!strcmp(track->codec_id, "A_MS/ACM") &&
2565  track->codec_priv.size >= 14 &&
2566  track->codec_priv.data) {
2567  int ret;
2569  track->codec_priv.size,
2570  0, NULL, NULL, NULL, NULL);
2571  ret = ff_get_wav_header(s, &b.pub, st->codecpar,
2572  track->codec_priv.size, 0);
2573  if (ret < 0)
2574  return ret;
2575  codec_id = st->codecpar->codec_id;
2576  fourcc = st->codecpar->codec_tag;
2577  extradata_offset = FFMIN(track->codec_priv.size, 18);
2578  } else if (!strcmp(track->codec_id, "A_QUICKTIME")
2579  /* Normally 36, but allow noncompliant private data */
2580  && (track->codec_priv.size >= 32)
2581  && (track->codec_priv.data)) {
2582  uint16_t sample_size;
2583  int ret = get_qt_codec(track, &fourcc, &codec_id);
2584  if (ret < 0)
2585  return ret;
2586  sample_size = AV_RB16(track->codec_priv.data + 26);
2587  if (fourcc == 0) {
2588  if (sample_size == 8) {
2589  fourcc = MKTAG('r','a','w',' ');
2591  } else if (sample_size == 16) {
2592  fourcc = MKTAG('t','w','o','s');
2594  }
2595  }
2596  if ((fourcc == MKTAG('t','w','o','s') ||
2597  fourcc == MKTAG('s','o','w','t')) &&
2598  sample_size == 8)
2600  } else if (!strcmp(track->codec_id, "V_QUICKTIME") &&
2601  (track->codec_priv.size >= 21) &&
2602  (track->codec_priv.data)) {
2603  int ret = get_qt_codec(track, &fourcc, &codec_id);
2604  if (ret < 0)
2605  return ret;
2606  if (codec_id == AV_CODEC_ID_NONE && AV_RL32(track->codec_priv.data+4) == AV_RL32("SMI ")) {
2607  fourcc = MKTAG('S','V','Q','3');
2609  }
2610  if (codec_id == AV_CODEC_ID_NONE)
2611  av_log(matroska->ctx, AV_LOG_ERROR,
2612  "mov FourCC not found %s.\n", av_fourcc2str(fourcc));
2613  if (track->codec_priv.size >= 86) {
2614  bit_depth = AV_RB16(track->codec_priv.data + 82);
2616  track->codec_priv.size,
2617  0, NULL, NULL, NULL, NULL);
2618  if (ff_get_qtpalette(codec_id, &b.pub, track->palette)) {
2619  bit_depth &= 0x1F;
2620  track->has_palette = 1;
2621  }
2622  }
2623  } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
2624  switch (track->audio.bitdepth) {
2625  case 8:
2627  break;
2628  case 24:
2630  break;
2631  case 32:
2633  break;
2634  }
2635  } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
2636  switch (track->audio.bitdepth) {
2637  case 8:
2639  break;
2640  case 24:
2642  break;
2643  case 32:
2645  break;
2646  }
2647  } else if (codec_id == AV_CODEC_ID_PCM_F32LE &&
2648  track->audio.bitdepth == 64) {
2650  } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
2651  int profile = matroska_aac_profile(track->codec_id);
2652  int sri = matroska_aac_sri(track->audio.samplerate);
2653  extradata = av_mallocz(5 + AV_INPUT_BUFFER_PADDING_SIZE);
2654  if (!extradata)
2655  return AVERROR(ENOMEM);
2656  extradata[0] = (profile << 3) | ((sri & 0x0E) >> 1);
2657  extradata[1] = ((sri & 0x01) << 7) | (track->audio.channels << 3);
2658  if (strstr(track->codec_id, "SBR")) {
2659  sri = matroska_aac_sri(track->audio.out_samplerate);
2660  extradata[2] = 0x56;
2661  extradata[3] = 0xE5;
2662  extradata[4] = 0x80 | (sri << 3);
2663  extradata_size = 5;
2664  } else
2665  extradata_size = 2;
2666  } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX - 12 - AV_INPUT_BUFFER_PADDING_SIZE) {
2667  /* Only ALAC's magic cookie is stored in Matroska's track headers.
2668  * Create the "atom size", "tag", and "tag version" fields the
2669  * decoder expects manually. */
2670  extradata_size = 12 + track->codec_priv.size;
2671  extradata = av_mallocz(extradata_size +
2673  if (!extradata)
2674  return AVERROR(ENOMEM);
2675  AV_WB32(extradata, extradata_size);
2676  memcpy(&extradata[4], "alac", 4);
2677  AV_WB32(&extradata[8], 0);
2678  memcpy(&extradata[12], track->codec_priv.data,
2679  track->codec_priv.size);
2680  } else if (codec_id == AV_CODEC_ID_TTA) {
2681  uint8_t *ptr;
2682  if (track->audio.channels > UINT16_MAX ||
2683  track->audio.bitdepth > UINT16_MAX) {
2684  av_log(matroska->ctx, AV_LOG_WARNING,
2685  "Too large audio channel number %"PRIu64
2686  " or bitdepth %"PRIu64". Skipping track.\n",
2687  track->audio.channels, track->audio.bitdepth);
2688  if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
2689  return AVERROR_INVALIDDATA;
2690  else
2691  continue;
2692  }
2693  if (track->audio.out_samplerate < 0 || track->audio.out_samplerate > INT_MAX)
2694  return AVERROR_INVALIDDATA;
2695  extradata_size = 22;
2696  extradata = av_mallocz(extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
2697  if (!extradata)
2698  return AVERROR(ENOMEM);
2699  ptr = extradata;
2700  bytestream_put_be32(&ptr, AV_RB32("TTA1"));
2701  bytestream_put_le16(&ptr, 1);
2702  bytestream_put_le16(&ptr, track->audio.channels);
2703  bytestream_put_le16(&ptr, track->audio.bitdepth);
2704  bytestream_put_le32(&ptr, track->audio.out_samplerate);
2705  bytestream_put_le32(&ptr, av_rescale(matroska->duration * matroska->time_scale,
2706  track->audio.out_samplerate,
2707  AV_TIME_BASE * 1000));
2708  } else if (codec_id == AV_CODEC_ID_RV10 ||
2712  extradata_offset = 26;
2713  } else if (codec_id == AV_CODEC_ID_RA_144) {
2714  track->audio.out_samplerate = 8000;
2715  track->audio.channels = 1;
2716  } else if ((codec_id == AV_CODEC_ID_RA_288 ||
2720  && track->codec_priv.data) {
2721  const uint8_t *ptr = track->codec_priv.data;
2722  int flavor;
2723 
2724  if (track->codec_priv.size < 46)
2725  return AVERROR_INVALIDDATA;
2726  ptr += 22;
2727  flavor = bytestream_get_be16(&ptr);
2728  track->audio.coded_framesize = bytestream_get_be32(&ptr);
2729  ptr += 12;
2730  track->audio.sub_packet_h = bytestream_get_be16(&ptr);
2731  track->audio.frame_size = bytestream_get_be16(&ptr);
2732  track->audio.sub_packet_size = bytestream_get_be16(&ptr);
2733  if (track->audio.coded_framesize <= 0 ||
2734  track->audio.sub_packet_h <= 0 ||
2735  track->audio.frame_size <= 0)
2736  return AVERROR_INVALIDDATA;
2737 
2738  if (codec_id == AV_CODEC_ID_RA_288) {
2739  if (track->audio.sub_packet_h & 1 || 2 * track->audio.frame_size
2740  != (int64_t)track->audio.sub_packet_h * track->audio.coded_framesize)
2741  return AVERROR_INVALIDDATA;
2742  st->codecpar->block_align = track->audio.coded_framesize;
2743  track->codec_priv.size = 0;
2744  } else {
2745  if (codec_id == AV_CODEC_ID_SIPR) {
2746  static const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
2747  if (flavor > 3)
2748  return AVERROR_INVALIDDATA;
2749  track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
2750  st->codecpar->bit_rate = sipr_bit_rate[flavor];
2751  } else if (track->audio.sub_packet_size <= 0 ||
2752  track->audio.frame_size % track->audio.sub_packet_size)
2753  return AVERROR_INVALIDDATA;
2754  st->codecpar->block_align = track->audio.sub_packet_size;
2755  extradata_offset = 78;
2756  }
2757  track->audio.buf = av_malloc_array(track->audio.sub_packet_h,
2758  track->audio.frame_size);
2759  if (!track->audio.buf)
2760  return AVERROR(ENOMEM);
2761  } else if (codec_id == AV_CODEC_ID_FLAC && track->codec_priv.size) {
2762  ret = matroska_parse_flac(s, track, &extradata_offset);
2763  if (ret < 0)
2764  return ret;
2765  } else if (codec_id == AV_CODEC_ID_WAVPACK && track->codec_priv.size < 2) {
2766  av_log(matroska->ctx, AV_LOG_INFO, "Assuming WavPack version 4.10 "
2767  "in absence of valid CodecPrivate.\n");
2768  extradata_size = 2;
2769  extradata = av_mallocz(2 + AV_INPUT_BUFFER_PADDING_SIZE);
2770  if (!extradata)
2771  return AVERROR(ENOMEM);
2772  AV_WL16(extradata, 0x410);
2773  } else if (codec_id == AV_CODEC_ID_PRORES && track->codec_priv.size == 4) {
2774  fourcc = AV_RL32(track->codec_priv.data);
2775  } else if (codec_id == AV_CODEC_ID_VP9 && track->codec_priv.size) {
2776  /* we don't need any value stored in CodecPrivate.
2777  make sure that it's not exported as extradata. */
2778  track->codec_priv.size = 0;
2779  }
2780  track->codec_priv.size -= extradata_offset;
2781 
2782  if (codec_id == AV_CODEC_ID_NONE)
2783  av_log(matroska->ctx, AV_LOG_INFO,
2784  "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
2785 
2786  if (track->time_scale < 0.01) {
2787  av_log(matroska->ctx, AV_LOG_WARNING,
2788  "Track TimestampScale too small %f, assuming 1.0.\n",
2789  track->time_scale);
2790  track->time_scale = 1.0;
2791  }
2792  avpriv_set_pts_info(st, 64, matroska->time_scale * track->time_scale,
2793  1000 * 1000 * 1000); /* 64 bit pts in ns */
2794 
2795  /* convert the delay from ns to the track timebase */
2797  (AVRational){ 1, 1000000000 },
2798  st->time_base);
2799 
2800  st->codecpar->codec_id = codec_id;
2801 
2802  if (strcmp(track->language, "und"))
2803  av_dict_set(&st->metadata, "language", track->language, 0);
2804  av_dict_set(&st->metadata, "title", track->name, 0);
2805 
2806  if (track->flag_default)
2808  if (track->flag_forced)
2810  if (track->flag_comment)
2812  if (track->flag_hearingimpaired)
2814  if (track->flag_visualimpaired)
2816  if (track->flag_original.count > 0)
2819 
2820  if (!st->codecpar->extradata) {
2821  if (extradata) {
2822  st->codecpar->extradata = extradata;
2823  st->codecpar->extradata_size = extradata_size;
2824  } else if (track->codec_priv.data && track->codec_priv.size > 0) {
2825  if (ff_alloc_extradata(st->codecpar, track->codec_priv.size))
2826  return AVERROR(ENOMEM);
2827  memcpy(st->codecpar->extradata,
2828  track->codec_priv.data + extradata_offset,
2829  track->codec_priv.size);
2830  }
2831  }
2832 
2833  if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2835  int display_width_mul = 1;
2836  int display_height_mul = 1;
2837 
2839  st->codecpar->codec_tag = fourcc;
2840  if (bit_depth >= 0)
2842  st->codecpar->width = track->video.pixel_width;
2843  st->codecpar->height = track->video.pixel_height;
2844 
2846  st->codecpar->field_order = mkv_field_order(matroska, track->video.field_order);
2849 
2851  mkv_stereo_mode_display_mul(track->video.stereo_mode, &display_width_mul, &display_height_mul);
2852 
2855  &st->sample_aspect_ratio.den,
2856  st->codecpar->height * track->video.display_width * display_width_mul,
2857  st->codecpar->width * track->video.display_height * display_height_mul,
2858  INT_MAX);
2859  }
2860  if (st->codecpar->codec_id != AV_CODEC_ID_HEVC)
2862 
2863  if (track->default_duration) {
2864  int div = track->default_duration <= INT64_MAX ? 1 : 2;
2866  1000000000 / div, track->default_duration / div, 30000);
2867 #if FF_API_R_FRAME_RATE
2868  if ( st->avg_frame_rate.num < st->avg_frame_rate.den * 1000LL
2869  && st->avg_frame_rate.num > st->avg_frame_rate.den * 5LL)
2870  st->r_frame_rate = st->avg_frame_rate;
2871 #endif
2872  }
2873 
2874  /* export stereo mode flag as metadata tag */
2876  av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
2877 
2878  /* export alpha mode flag as metadata tag */
2879  if (track->video.alpha_mode)
2880  av_dict_set(&st->metadata, "alpha_mode", "1", 0);
2881 
2882  /* if we have virtual track, mark the real tracks */
2883  for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
2884  char buf[32];
2886  continue;
2887  snprintf(buf, sizeof(buf), "%s_%d",
2889  for (k=0; k < matroska->tracks.nb_elem; k++)
2890  if (planes[j].uid == tracks[k].uid && tracks[k].stream) {
2891  av_dict_set(&tracks[k].stream->metadata,
2892  "stereo_mode", buf, 0);
2893  break;
2894  }
2895  }
2896  // add stream level stereo3d side data if it is a supported format
2898  track->video.stereo_mode != 10 && track->video.stereo_mode != 12) {
2899  int ret = ff_mkv_stereo3d_conv(st, track->video.stereo_mode);
2900  if (ret < 0)
2901  return ret;
2902  }
2903 
2904  ret = mkv_parse_video_color(st, track);
2905  if (ret < 0)
2906  return ret;
2907  ret = mkv_parse_video_projection(st, track, matroska->ctx);
2908  if (ret < 0)
2909  return ret;
2910  } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2912  st->codecpar->codec_tag = fourcc;
2913  st->codecpar->sample_rate = track->audio.out_samplerate;
2914  st->codecpar->channels = track->audio.channels;
2915  if (!st->codecpar->bits_per_coded_sample)
2917  if (st->codecpar->codec_id == AV_CODEC_ID_MP3 ||
2918  st->codecpar->codec_id == AV_CODEC_ID_MLP ||
2921  else if (st->codecpar->codec_id != AV_CODEC_ID_AAC)
2923  if (track->codec_delay > 0) {
2925  (AVRational){1, 1000000000},
2926  (AVRational){1, st->codecpar->codec_id == AV_CODEC_ID_OPUS ?
2927  48000 : st->codecpar->sample_rate});
2928  }
2929  if (track->seek_preroll > 0) {
2931  (AVRational){1, 1000000000},
2932  (AVRational){1, st->codecpar->sample_rate});
2933  }
2934  } else if (codec_id == AV_CODEC_ID_WEBVTT) {
2936 
2937  if (!strcmp(track->codec_id, "D_WEBVTT/CAPTIONS")) {
2939  } else if (!strcmp(track->codec_id, "D_WEBVTT/DESCRIPTIONS")) {
2941  } else if (!strcmp(track->codec_id, "D_WEBVTT/METADATA")) {
2943  }
2944  } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2946 
2947  if (track->flag_textdescriptions)
2949  }
2950 
2951  ret = mkv_parse_block_addition_mappings(s, st, track);
2952  if (ret < 0)
2953  return ret;
2954  }
2955 
2956  return 0;
2957 }
2958 
2960 {
2961  FFFormatContext *const si = ffformatcontext(s);
2962  MatroskaDemuxContext *matroska = s->priv_data;
2963  EbmlList *attachments_list = &matroska->attachments;
2964  EbmlList *chapters_list = &matroska->chapters;
2965  MatroskaAttachment *attachments;
2966  MatroskaChapter *chapters;
2967  uint64_t max_start = 0;
2968  int64_t pos;
2969  Ebml ebml = { 0 };
2970  int i, j, res;
2971 
2972  matroska->ctx = s;
2973  matroska->cues_parsing_deferred = 1;
2974 
2975  /* First read the EBML header. */
2976  if (ebml_parse(matroska, ebml_syntax, &ebml) || !ebml.doctype) {
2977  av_log(matroska->ctx, AV_LOG_ERROR, "EBML header parsing failed\n");
2978  ebml_free(ebml_syntax, &ebml);
2979  return AVERROR_INVALIDDATA;
2980  }
2981  if (ebml.version > EBML_VERSION ||
2982  ebml.max_size > sizeof(uint64_t) ||
2983  ebml.id_length > sizeof(uint32_t) ||
2984  ebml.doctype_version > 3) {
2986  "EBML version %"PRIu64", doctype %s, doc version %"PRIu64,
2987  ebml.version, ebml.doctype, ebml.doctype_version);
2988  ebml_free(ebml_syntax, &ebml);
2989  return AVERROR_PATCHWELCOME;
2990  } else if (ebml.doctype_version == 3) {
2991  av_log(matroska->ctx, AV_LOG_WARNING,
2992  "EBML header using unsupported features\n"
2993  "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
2994  ebml.version, ebml.doctype, ebml.doctype_version);
2995  }
2996  for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
2997  if (!strcmp(ebml.doctype, matroska_doctypes[i]))
2998  break;
3000  av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
3001  if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
3002  ebml_free(ebml_syntax, &ebml);
3003  return AVERROR_INVALIDDATA;
3004  }
3005  }
3006  ebml_free(ebml_syntax, &ebml);
3007 
3008  matroska->pkt = si->parse_pkt;
3009 
3010  /* The next thing is a segment. */
3011  pos = avio_tell(matroska->ctx->pb);
3012  res = ebml_parse(matroska, matroska_segments, matroska);
3013  // Try resyncing until we find an EBML_STOP type element.
3014  while (res != 1) {
3015  res = matroska_resync(matroska, pos);
3016  if (res < 0)
3017  return res;
3018  pos = avio_tell(matroska->ctx->pb);
3019  res = ebml_parse(matroska, matroska_segment, matroska);
3020  if (res == AVERROR(EIO)) // EOF is translated to EIO, this exists the loop on EOF
3021  return res;
3022  }
3023  /* Set data_offset as it might be needed later by seek_frame_generic. */
3024  if (matroska->current_id == MATROSKA_ID_CLUSTER)
3025  si->data_offset = avio_tell(matroska->ctx->pb) - 4;
3026  matroska_execute_seekhead(matroska);
3027 
3028  if (!matroska->time_scale)
3029  matroska->time_scale = 1000000;
3030  if (isnan(matroska->duration))
3031  matroska->duration = 0;
3032  if (matroska->duration)
3033  matroska->ctx->duration = matroska->duration * matroska->time_scale *
3034  1000 / AV_TIME_BASE;
3035  av_dict_set(&s->metadata, "title", matroska->title, 0);
3036  av_dict_set(&s->metadata, "encoder", matroska->muxingapp, 0);
3037 
3038  if (matroska->date_utc.size == 8)
3039  matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
3040 
3041  res = matroska_parse_tracks(s);
3042  if (res < 0)
3043  return res;
3044 
3045  attachments = attachments_list->elem;
3046  for (j = 0; j < attachments_list->nb_elem; j++) {
3047  if (!(attachments[j].filename && attachments[j].mime &&
3048  attachments[j].bin.data && attachments[j].bin.size > 0)) {
3049  av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
3050  } else {
3052  if (!st)
3053  break;
3054  av_dict_set(&st->metadata, "filename", attachments[j].filename, 0);
3055  av_dict_set(&st->metadata, "mimetype", attachments[j].mime, 0);
3056  if (attachments[j].description)
3057  av_dict_set(&st->metadata, "title", attachments[j].description, 0);
3059 
3060  for (i = 0; mkv_image_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
3061  if (av_strstart(attachments[j].mime, mkv_image_mime_tags[i].str, NULL)) {
3063  break;
3064  }
3065  }
3066 
3067  attachments[j].stream = st;
3068 
3069  if (st->codecpar->codec_id != AV_CODEC_ID_NONE) {
3070  res = ff_add_attached_pic(s, st, NULL, &attachments[j].bin.buf, 0);
3071  if (res < 0)
3072  return res;
3073  } else {
3075  if (ff_alloc_extradata(st->codecpar, attachments[j].bin.size))
3076  break;
3077  memcpy(st->codecpar->extradata, attachments[j].bin.data,
3078  attachments[j].bin.size);
3079 
3080  for (i = 0; mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
3081  if (av_strstart(attachments[j].mime, mkv_mime_tags[i].str, NULL)) {
3083  break;
3084  }
3085  }
3086  }
3087  }
3088  }
3089 
3090  chapters = chapters_list->elem;
3091  for (i = 0; i < chapters_list->nb_elem; i++)
3092  if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid &&
3093  (max_start == 0 || chapters[i].start > max_start)) {
3094  chapters[i].chapter =
3095  avpriv_new_chapter(s, chapters[i].uid,
3096  (AVRational) { 1, 1000000000 },
3097  chapters[i].start, chapters[i].end,
3098  chapters[i].title);
3099  max_start = chapters[i].start;
3100  }
3101 
3102  matroska_add_index_entries(matroska);
3103 
3105 
3106  return 0;
3108 
3109 /*
3110  * Put one packet in an application-supplied AVPacket struct.
3111  * Returns 0 on success or -1 on failure.
3112  */
3113 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
3114  AVPacket *pkt)
3115 {
3116  if (matroska->queue.head) {
3117  MatroskaTrack *tracks = matroska->tracks.elem;
3118  MatroskaTrack *track;
3119 
3120  avpriv_packet_list_get(&matroska->queue, pkt);
3121  track = &tracks[pkt->stream_index];
3122  if (track->has_palette) {
3124  if (!pal) {
3125  av_log(matroska->ctx, AV_LOG_ERROR, "Cannot append palette to packet\n");
3126  } else {
3127  memcpy(pal, track->palette, AVPALETTE_SIZE);
3128  }
3129  track->has_palette = 0;
3130  }
3131  return 0;
3132  }
3133 
3134  return -1;
3135 }
3136 
3137 /*
3138  * Free all packets in our internal queue.
3139  */
3140 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
3141 {
3142  avpriv_packet_list_free(&matroska->queue);
3143 }
3144 
3145 static int matroska_parse_laces(MatroskaDemuxContext *matroska, uint8_t **buf,
3146  int size, int type, AVIOContext *pb,
3147  uint32_t lace_size[256], int *laces)
3148 {
3149  int n;
3150  uint8_t *data = *buf;
3151 
3152  if (!type) {
3153  *laces = 1;
3154  lace_size[0] = size;
3155  return 0;
3156  }
3157 
3158  if (size <= 0)
3159  return AVERROR_INVALIDDATA;
3160 
3161  *laces = *data + 1;
3162  data += 1;
3163  size -= 1;
3164 
3165  switch (type) {
3166  case 0x1: /* Xiph lacing */
3167  {
3168  uint8_t temp;
3169  uint32_t total = 0;
3170  for (n = 0; n < *laces - 1; n++) {
3171  lace_size[n] = 0;
3172 
3173  do {
3174  if (size <= total)
3175  return AVERROR_INVALIDDATA;
3176  temp = *data;
3177  total += temp;
3178  lace_size[n] += temp;
3179  data += 1;
3180  size -= 1;
3181  } while (temp == 0xff);
3182  }
3183  if (size < total)
3184  return AVERROR_INVALIDDATA;
3185 
3186  lace_size[n] = size - total;
3187  break;
3188  }
3189 
3190  case 0x2: /* fixed-size lacing */
3191  if (size % (*laces))
3192  return AVERROR_INVALIDDATA;
3193  for (n = 0; n < *laces; n++)
3194  lace_size[n] = size / *laces;
3195  break;
3196 
3197  case 0x3: /* EBML lacing */
3198  {
3199  uint64_t num;
3200  uint64_t total;
3201  int offset;
3202 
3203  avio_skip(pb, 4);
3204 
3205  n = ebml_read_num(matroska, pb, 8, &num, 1);
3206  if (n < 0)
3207  return n;
3208  if (num > INT_MAX)
3209  return AVERROR_INVALIDDATA;
3210 
3211  total = lace_size[0] = num;
3212  offset = n;
3213  for (n = 1; n < *laces - 1; n++) {
3214  int64_t snum;
3215  int r;
3216  r = matroska_ebmlnum_sint(matroska, pb, &snum);
3217  if (r < 0)
3218  return r;
3219  if (lace_size[n - 1] + snum > (uint64_t)INT_MAX)
3220  return AVERROR_INVALIDDATA;
3221 
3222  lace_size[n] = lace_size[n - 1] + snum;
3223  total += lace_size[n];
3224  offset += r;
3225  }
3226  data += offset;
3227  size -= offset;
3228  if (size < total)
3229  return AVERROR_INVALIDDATA;
3230 
3231  lace_size[*laces - 1] = size - total;
3232  break;
3233  }
3234  }
3236  *buf = data;
3237 
3238  return 0;
3239 }
3240 
3241 static int matroska_parse_rm_audio(MatroskaDemuxContext *matroska,
3242  MatroskaTrack *track, AVStream *st,
3243  uint8_t *data, int size, uint64_t timecode,
3244  int64_t pos)
3245 {
3246  const int a = st->codecpar->block_align;
3247  const int sps = track->audio.sub_packet_size;
3248  const int cfs = track->audio.coded_framesize;
3249  const int h = track->audio.sub_packet_h;
3250  const int w = track->audio.frame_size;
3251  int y = track->audio.sub_packet_cnt;
3252  int x;
3253 
3254  if (!track->audio.pkt_cnt) {
3255  if (track->audio.sub_packet_cnt == 0)
3256  track->audio.buf_timecode = timecode;
3257  if (st->codecpar->codec_id == AV_CODEC_ID_RA_288) {
3258  if (size < cfs * h / 2) {
3259  av_log(matroska->ctx, AV_LOG_ERROR,
3260  "Corrupt int4 RM-style audio packet size\n");
3261  return AVERROR_INVALIDDATA;
3262  }
3263  for (x = 0; x < h / 2; x++)
3264  memcpy(track->audio.buf + x * 2 * w + y * cfs,
3265  data + x * cfs, cfs);
3266  } else if (st->codecpar->codec_id == AV_CODEC_ID_SIPR) {
3267  if (size < w) {
3268  av_log(matroska->ctx, AV_LOG_ERROR,
3269  "Corrupt sipr RM-style audio packet size\n");
3270  return AVERROR_INVALIDDATA;
3271  }
3272  memcpy(track->audio.buf + y * w, data, w);
3273  } else {
3274  if (size < w) {
3275  av_log(matroska->ctx, AV_LOG_ERROR,
3276  "Corrupt generic RM-style audio packet size\n");
3277  return AVERROR_INVALIDDATA;
3278  }
3279  for (x = 0; x < w / sps; x++)
3280  memcpy(track->audio.buf +
3281  sps * (h * x + ((h + 1) / 2) * (y & 1) + (y >> 1)),
3282  data + x * sps, sps);
3283  }
3284 
3285  if (++track->audio.sub_packet_cnt >= h) {
3286  if (st->codecpar->codec_id == AV_CODEC_ID_SIPR)
3287  ff_rm_reorder_sipr_data(track->audio.buf, h, w);
3288  track->audio.sub_packet_cnt = 0;
3289  track->audio.pkt_cnt = h * w / a;
3290  }
3291  }
3292 
3293  while (track->audio.pkt_cnt) {
3294  int ret;
3295  AVPacket *pkt = matroska->pkt;
3296 
3297  ret = av_new_packet(pkt, a);
3298  if (ret < 0) {
3299  return ret;
3300  }
3301  memcpy(pkt->data,
3302  track->audio.buf + a * (h * w / a - track->audio.pkt_cnt--),
3303  a);
3304  pkt->pts = track->audio.buf_timecode;
3306  pkt->pos = pos;
3307  pkt->stream_index = st->index;
3308  ret = avpriv_packet_list_put(&matroska->queue, pkt, NULL, 0);
3309  if (ret < 0) {
3311  return AVERROR(ENOMEM);
3312  }
3313  }
3314 
3315  return 0;
3316 }
3317 
3318 /* reconstruct full wavpack blocks from mangled matroska ones */
3319 static int matroska_parse_wavpack(MatroskaTrack *track,
3320  uint8_t **data, int *size)
3321 {
3322  uint8_t *dst = NULL;
3323  uint8_t *src = *data;
3324  int dstlen = 0;
3325  int srclen = *size;
3326  uint32_t samples;
3327  uint16_t ver;
3328  int ret, offset = 0;
3329 
3330  if (srclen < 12)
3331  return AVERROR_INVALIDDATA;
3332 
3333  av_assert1(track->stream->codecpar->extradata_size >= 2);
3334  ver = AV_RL16(track->stream->codecpar->extradata);
3335 
3336  samples = AV_RL32(src);
3337  src += 4;
3338  srclen -= 4;
3339 
3340  while (srclen >= 8) {
3341  int multiblock;
3342  uint32_t blocksize;
3343  uint8_t *tmp;
3344 
3345  uint32_t flags = AV_RL32(src);
3346  uint32_t crc = AV_RL32(src + 4);
3347  src += 8;
3348  srclen -= 8;
3349 
3350  multiblock = (flags & 0x1800) != 0x1800;
3351  if (multiblock) {
3352  if (srclen < 4) {
3354  goto fail;
3355  }
3356  blocksize = AV_RL32(src);
3357  src += 4;
3358  srclen -= 4;
3359  } else
3360  blocksize = srclen;
3361 
3362  if (blocksize > srclen) {
3364  goto fail;
3365  }
3366 
3367  tmp = av_realloc(dst, dstlen + blocksize + 32 + AV_INPUT_BUFFER_PADDING_SIZE);
3368  if (!tmp) {
3369  ret = AVERROR(ENOMEM);
3370  goto fail;
3371  }
3372  dst = tmp;
3373  dstlen += blocksize + 32;
3374 
3375  AV_WL32(dst + offset, MKTAG('w', 'v', 'p', 'k')); // tag
3376  AV_WL32(dst + offset + 4, blocksize + 24); // blocksize - 8
3377  AV_WL16(dst + offset + 8, ver); // version
3378  AV_WL16(dst + offset + 10, 0); // track/index_no
3379  AV_WL32(dst + offset + 12, 0); // total samples
3380  AV_WL32(dst + offset + 16, 0); // block index
3381  AV_WL32(dst + offset + 20, samples); // number of samples
3382  AV_WL32(dst + offset + 24, flags); // flags
3383  AV_WL32(dst + offset + 28, crc); // crc
3384  memcpy(dst + offset + 32, src, blocksize); // block data
3385 
3386  src += blocksize;
3387  srclen -= blocksize;
3388  offset += blocksize + 32;
3389  }
3390 
3391  memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3392 
3393  *data = dst;
3394  *size = dstlen;
3395 
3396  return 0;
3398 fail:
3399  av_freep(&dst);
3400  return ret;
3401 }
3402 
3403 static int matroska_parse_prores(MatroskaTrack *track,
3404  uint8_t **data, int *size)
3405 {
3406  uint8_t *dst;
3407  int dstlen = *size + 8;
3408 
3409  dst = av_malloc(dstlen + AV_INPUT_BUFFER_PADDING_SIZE);
3410  if (!dst)
3411  return AVERROR(ENOMEM);
3412 
3413  AV_WB32(dst, dstlen);
3414  AV_WB32(dst + 4, MKBETAG('i', 'c', 'p', 'f'));
3415  memcpy(dst + 8, *data, dstlen - 8);
3416  memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3417 
3418  *data = dst;
3419  *size = dstlen;
3420 
3421  return 0;
3422 }
3423 
3424 static int matroska_parse_webvtt(MatroskaDemuxContext *matroska,
3425  MatroskaTrack *track,
3426  AVStream *st,
3427  uint8_t *data, int data_len,
3428  uint64_t timecode,
3429  uint64_t duration,
3430  int64_t pos)
3431 {
3432  AVPacket *pkt = matroska->pkt;
3433  uint8_t *id, *settings, *text, *buf;
3434  int id_len, settings_len, text_len;
3435  uint8_t *p, *q;
3436  int err;
3437 
3438  if (data_len <= 0)
3439  return AVERROR_INVALIDDATA;
3440 
3441  p = data;
3442  q = data + data_len;
3443 
3444  id = p;
3445  id_len = -1;
3446  while (p < q) {
3447  if (*p == '\r' || *p == '\n') {
3448  id_len = p - id;
3449  if (*p == '\r')
3450  p++;
3451  break;
3452  }
3453  p++;
3454  }
3455 
3456  if (p >= q || *p != '\n')
3457  return AVERROR_INVALIDDATA;
3458  p++;
3459 
3460  settings = p;
3461  settings_len = -1;
3462  while (p < q) {
3463  if (*p == '\r' || *p == '\n') {
3464  settings_len = p - settings;
3465  if (*p == '\r')
3466  p++;
3467  break;
3468  }
3469  p++;
3470  }
3471 
3472  if (p >= q || *p != '\n')
3473  return AVERROR_INVALIDDATA;
3474  p++;
3475 
3476  text = p;
3477  text_len = q - p;
3478  while (text_len > 0) {
3479  const int len = text_len - 1;
3480  const uint8_t c = p[len];
3481  if (c != '\r' && c != '\n')
3482  break;
3483  text_len = len;
3484  }
3485 
3486  if (text_len <= 0)
3487  return AVERROR_INVALIDDATA;
3488 
3489  err = av_new_packet(pkt, text_len);
3490  if (err < 0) {
3491  return err;
3492  }
3493 
3494  memcpy(pkt->data, text, text_len);
3495 
3496  if (id_len > 0) {
3499  id_len);
3500  if (!buf) {
3502  return AVERROR(ENOMEM);
3503  }
3504  memcpy(buf, id, id_len);
3505  }
3506 
3507  if (settings_len > 0) {
3510  settings_len);
3511  if (!buf) {
3513  return AVERROR(ENOMEM);
3514  }
3515  memcpy(buf, settings, settings_len);
3516  }
3517 
3518  // Do we need this for subtitles?
3519  // pkt->flags = AV_PKT_FLAG_KEY;
3520 
3521  pkt->stream_index = st->index;
3522  pkt->pts = timecode;
3523 
3524  // Do we need this for subtitles?
3525  // pkt->dts = timecode;
3526 
3527  pkt->duration = duration;
3528  pkt->pos = pos;
3529 
3530  err = avpriv_packet_list_put(&matroska->queue, pkt, NULL, 0);
3531  if (err < 0) {
3533  return AVERROR(ENOMEM);
3534  }
3535 
3536  return 0;
3537 }
3538 
3539 static int matroska_parse_frame(MatroskaDemuxContext *matroska,
3540  MatroskaTrack *track, AVStream *st,
3541  AVBufferRef *buf, uint8_t *data, int pkt_size,
3542  uint64_t timecode, uint64_t lace_duration,
3543  int64_t pos, int is_keyframe,
3544  uint8_t *additional, uint64_t additional_id, int additional_size,
3545  int64_t discard_padding)
3546 {
3547  uint8_t *pkt_data = data;
3548  int res = 0;
3549  AVPacket *pkt = matroska->pkt;
3550 
3551  if (st->codecpar->codec_id == AV_CODEC_ID_WAVPACK) {
3552  res = matroska_parse_wavpack(track, &pkt_data, &pkt_size);
3553  if (res < 0) {
3554  av_log(matroska->ctx, AV_LOG_ERROR,
3555  "Error parsing a wavpack block.\n");
3556  goto fail;
3557  }
3558  if (!buf)
3559  av_freep(&data);
3560  buf = NULL;
3561  }
3562 
3563  if (st->codecpar->codec_id == AV_CODEC_ID_PRORES &&
3564  AV_RB32(pkt_data + 4) != MKBETAG('i', 'c', 'p', 'f')) {
3565  res = matroska_parse_prores(track, &pkt_data, &pkt_size);
3566  if (res < 0) {
3567  av_log(matroska->ctx, AV_LOG_ERROR,
3568  "Error parsing a prores block.\n");
3569  goto fail;
3570  }
3571  if (!buf)
3572  av_freep(&data);
3573  buf = NULL;
3574  }
3575 
3576  if (!pkt_size && !additional_size)
3577  goto no_output;
3578 
3579  if (!buf)
3580  pkt->buf = av_buffer_create(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE,
3581  NULL, NULL, 0);
3582  else
3583  pkt->buf = av_buffer_ref(buf);
3584 
3585  if (!pkt->buf) {
3586  res = AVERROR(ENOMEM);
3587  goto fail;
3588  }
3589 
3590  pkt->data = pkt_data;
3591  pkt->size = pkt_size;
3592  pkt->flags = is_keyframe;
3593  pkt->stream_index = st->index;
3594 
3595  if (additional_size > 0) {
3596  uint8_t *side_data = av_packet_new_side_data(pkt,
3598  additional_size + 8);
3599  if (!side_data) {
3601  return AVERROR(ENOMEM);
3602  }
3603  AV_WB64(side_data, additional_id);
3604  memcpy(side_data + 8, additional, additional_size);
3605  }
3606 
3607  if (discard_padding) {
3608  uint8_t *side_data = av_packet_new_side_data(pkt,
3610  10);
3611  if (!side_data) {
3613  return AVERROR(ENOMEM);
3614  }
3615  discard_padding = av_rescale_q(discard_padding,
3616  (AVRational){1, 1000000000},
3617  (AVRational){1, st->codecpar->sample_rate});
3618  if (discard_padding > 0) {
3619  AV_WL32(side_data + 4, discard_padding);
3620  } else {
3621  AV_WL32(side_data, -discard_padding);
3622  }
3623  }
3624 
3625  if (track->ms_compat)
3626  pkt->dts = timecode;
3627  else
3628  pkt->pts = timecode;
3629  pkt->pos = pos;
3630  pkt->duration = lace_duration;
3631 
3632  res = avpriv_packet_list_put(&matroska->queue, pkt, NULL, 0);
3633  if (res < 0) {
3635  return AVERROR(ENOMEM);
3636  }
3637 
3638  return 0;
3639 
3640 no_output:
3642  if (!buf)
3643  av_free(pkt_data);
3644  return res;
3645 }
3646 
3647 static int matroska_parse_block(MatroskaDemuxContext *matroska, AVBufferRef *buf, uint8_t *data,
3648  int size, int64_t pos, uint64_t cluster_time,
3649  uint64_t block_duration, int is_keyframe,
3650  uint8_t *additional, uint64_t additional_id, int additional_size,
3651  int64_t cluster_pos, int64_t discard_padding)
3652 {
3653  uint64_t timecode = AV_NOPTS_VALUE;
3654  MatroskaTrack *track;
3655  FFIOContext pb;
3656  int res = 0;
3657  AVStream *st;
3658  int16_t block_time;
3659  uint32_t lace_size[256];
3660  int n, flags, laces = 0;
3661  uint64_t num;
3662  int trust_default_duration;
3663 
3664  ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
3665 
3666  if ((n = ebml_read_num(matroska, &pb.pub, 8, &num, 1)) < 0)
3667  return n;
3668  data += n;
3669  size -= n;
3670 
3671  track = matroska_find_track_by_num(matroska, num);
3672  if (!track || size < 3)
3673  return AVERROR_INVALIDDATA;
3674 
3675  if (!(st = track->stream)) {
3676  av_log(matroska->ctx, AV_LOG_VERBOSE,
3677  "No stream associated to TrackNumber %"PRIu64". "
3678  "Ignoring Block with this TrackNumber.\n", num);
3679  return 0;
3680  }
3681 
3682  if (st->discard >= AVDISCARD_ALL)
3683  return res;
3684  if (block_duration > INT64_MAX)
3685  block_duration = INT64_MAX;
3686 
3687  block_time = sign_extend(AV_RB16(data), 16);
3688  data += 2;
3689  flags = *data++;
3690  size -= 3;
3691  if (is_keyframe == -1)
3692  is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
3693 
3694  if (cluster_time != (uint64_t) -1 &&
3695  (block_time >= 0 || cluster_time >= -block_time)) {
3696  uint64_t timecode_cluster_in_track_tb = (double) cluster_time / track->time_scale;
3697  timecode = timecode_cluster_in_track_tb + block_time - track->codec_delay_in_track_tb;
3698  if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE &&
3699  timecode < track->end_timecode)
3700  is_keyframe = 0; /* overlapping subtitles are not key frame */
3701  if (is_keyframe) {
3702  ff_reduce_index(matroska->ctx, st->index);
3703  av_add_index_entry(st, cluster_pos, timecode, 0, 0,
3705  }
3706  }
3707 
3708  if (matroska->skip_to_keyframe &&
3709  track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
3710  // Compare signed timecodes. Timecode may be negative due to codec delay
3711  // offset. We don't support timestamps greater than int64_t anyway - see
3712  // AVPacket's pts.
3713  if ((int64_t)timecode < (int64_t)matroska->skip_to_timecode)
3714  return res;
3715  if (is_keyframe)
3716  matroska->skip_to_keyframe = 0;
3717  else if (!ffstream(st)->skip_to_keyframe) {
3718  av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
3719  matroska->skip_to_keyframe = 0;
3720  }
3721  }
3722 
3723  res = matroska_parse_laces(matroska, &data, size, (flags & 0x06) >> 1,
3724  &pb.pub, lace_size, &laces);
3725  if (res < 0) {
3726  av_log(matroska->ctx, AV_LOG_ERROR, "Error parsing frame sizes.\n");
3727  return res;
3728  }
3729 
3730  trust_default_duration = track->default_duration != 0;
3731  if (track->audio.samplerate == 8000 && trust_default_duration) {
3732  // If this is needed for more codecs, then add them here
3733  if (st->codecpar->codec_id == AV_CODEC_ID_AC3) {
3734  if (track->audio.samplerate != st->codecpar->sample_rate || !st->codecpar->frame_size)
3735  trust_default_duration = 0;
3736  }
3737  }
3738 
3739  if (!block_duration && trust_default_duration)
3740  block_duration = track->default_duration * laces / matroska->time_scale;
3741 
3742  if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
3743  track->end_timecode =
3744  FFMAX(track->end_timecode, timecode + block_duration);
3745 
3746  for (n = 0; n < laces; n++) {
3747  int64_t lace_duration = block_duration*(n+1) / laces - block_duration*n / laces;
3748  uint8_t *out_data = data;
3749  int out_size = lace_size[n];
3750 
3751  if (track->needs_decoding) {
3752  res = matroska_decode_buffer(&out_data, &out_size, track);
3753  if (res < 0)
3754  return res;
3755  /* Given that we are here means that out_data is no longer
3756  * owned by buf, so set it to NULL. This depends upon
3757  * zero-length header removal compression being ignored. */
3758  av_assert1(out_data != data);
3759  buf = NULL;
3760  }
3761 
3762  if (track->audio.buf) {
3763  res = matroska_parse_rm_audio(matroska, track, st,
3764  out_data, out_size,
3765  timecode, pos);
3766  if (!buf)
3767  av_free(out_data);
3768  if (res)
3769  return res;
3770  } else if (st->codecpar->codec_id == AV_CODEC_ID_WEBVTT) {
3771  res = matroska_parse_webvtt(matroska, track, st,
3772  out_data, out_size,
3773  timecode, lace_duration,
3774  pos);
3775  if (!buf)
3776  av_free(out_data);
3777  if (res)
3778  return res;
3779  } else {
3780  res = matroska_parse_frame(matroska, track, st, buf, out_data,
3781  out_size, timecode, lace_duration,
3782  pos, !n ? is_keyframe : 0,
3783  additional, additional_id, additional_size,
3784  discard_padding);
3785  if (res)
3786  return res;
3787  }
3788 
3789  if (timecode != AV_NOPTS_VALUE)
3790  timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
3791  data += lace_size[n];
3792  }
3793 
3794  return 0;
3795 }
3796 
3797 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
3798 {
3799  MatroskaCluster *cluster = &matroska->current_cluster;
3800  MatroskaBlock *block = &cluster->block;
3801  int res;
3802 
3803  av_assert0(matroska->num_levels <= 2);
3804 
3805  if (matroska->num_levels == 1) {
3806  res = ebml_parse(matroska, matroska_segment, NULL);
3807 
3808  if (res == 1) {
3809  /* Found a cluster: subtract the size of the ID already read. */
3810  cluster->pos = avio_tell(matroska->ctx->pb) - 4;
3811 
3812  res = ebml_parse(matroska, matroska_cluster_enter, cluster);
3813  if (res < 0)
3814  return res;
3815  }
3816  }
3817 
3818  if (matroska->num_levels == 2) {
3819  /* We are inside a cluster. */
3820  res = ebml_parse(matroska, matroska_cluster_parsing, cluster);
3821 
3822  if (res >= 0 && block->bin.size > 0) {
3823  int is_keyframe = block->non_simple ? block->reference.count == 0 : -1;
3824  uint8_t* additional = block->additional.size > 0 ?
3825  block->additional.data : NULL;
3826 
3827  res = matroska_parse_block(matroska, block->bin.buf, block->bin.data,
3828  block->bin.size, block->bin.pos,
3829  cluster->timecode, block->duration,
3830  is_keyframe, additional, block->additional_id,
3831  block->additional.size, cluster->pos,
3832  block->discard_padding);
3833  }
3834 
3836  memset(block, 0, sizeof(*block));
3837  } else if (!matroska->num_levels) {
3838  if (!avio_feof(matroska->ctx->pb)) {
3839  avio_r8(matroska->ctx->pb);
3840  if (!avio_feof(matroska->ctx->pb)) {
3841  av_log(matroska->ctx, AV_LOG_WARNING, "File extends beyond "
3842  "end of segment.\n");
3843  return AVERROR_INVALIDDATA;
3844  }
3845  }
3846  matroska->done = 1;
3847  return AVERROR_EOF;
3848  }
3849 
3850  return res;
3851 }
3852 
3854 {
3855  MatroskaDemuxContext *matroska = s->priv_data;
3856  int ret = 0;
3857 
3858  if (matroska->resync_pos == -1) {
3859  // This can only happen if generic seeking has been used.
3860  matroska->resync_pos = avio_tell(s->pb);
3861  }
3862 
3863  while (matroska_deliver_packet(matroska, pkt)) {
3864  if (matroska->done)
3865  return (ret < 0) ? ret : AVERROR_EOF;
3866  if (matroska_parse_cluster(matroska) < 0 && !matroska->done)
3867  ret = matroska_resync(matroska, matroska->resync_pos);
3868  }
3869 
3870  return 0;
3871 }
3872 
3873 static int matroska_read_seek(AVFormatContext *s, int stream_index,
3874  int64_t timestamp, int flags)
3875 {
3876  MatroskaDemuxContext *matroska = s->priv_data;
3877  MatroskaTrack *tracks = NULL;
3878  AVStream *st = s->streams[stream_index];
3879  FFStream *const sti = ffstream(st);
3880  int i, index;
3881 
3882  /* Parse the CUES now since we need the index data to seek. */
3883  if (matroska->cues_parsing_deferred > 0) {
3884  matroska->cues_parsing_deferred = 0;
3885  matroska_parse_cues(matroska);
3886  }
3887 
3888  if (!sti->nb_index_entries)
3889  goto err;
3890  timestamp = FFMAX(timestamp, sti->index_entries[0].timestamp);
3891 
3892  if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 ||
3893  index == sti->nb_index_entries - 1) {
3894  matroska_reset_status(matroska, 0, sti->index_entries[sti->nb_index_entries - 1].pos);
3895  while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 ||
3896  index == sti->nb_index_entries - 1) {
3897  matroska_clear_queue(matroska);
3898  if (matroska_parse_cluster(matroska) < 0)
3899  break;
3900  }
3901  }
3902 
3903  matroska_clear_queue(matroska);
3904  if (index < 0 || (matroska->cues_parsing_deferred < 0 &&
3905  index == sti->nb_index_entries - 1))
3906  goto err;
3907 
3908  tracks = matroska->tracks.elem;
3909  for (i = 0; i < matroska->tracks.nb_elem; i++) {
3910  tracks[i].audio.pkt_cnt = 0;
3911  tracks[i].audio.sub_packet_cnt = 0;
3912  tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
3913  tracks[i].end_timecode = 0;
3914  }
3915 
3916  /* We seek to a level 1 element, so set the appropriate status. */
3917  matroska_reset_status(matroska, 0, sti->index_entries[index].pos);
3918  if (flags & AVSEEK_FLAG_ANY) {
3919  sti->skip_to_keyframe = 0;
3920  matroska->skip_to_timecode = timestamp;
3921  } else {
3922  sti->skip_to_keyframe = 1;
3923  matroska->skip_to_timecode = sti->index_entries[index].timestamp;
3924  }
3925  matroska->skip_to_keyframe = 1;
3926  matroska->done = 0;
3928  return 0;
3929 err:
3930  // slightly hackish but allows proper fallback to
3931  // the generic seeking code.
3932  matroska_reset_status(matroska, 0, -1);
3933  matroska->resync_pos = -1;
3934  matroska_clear_queue(matroska);
3936  matroska->skip_to_keyframe = 0;
3937  matroska->done = 0;
3938  return -1;
3939 }
3940 
3942 {
3943  MatroskaDemuxContext *matroska = s->priv_data;
3944  MatroskaTrack *tracks = matroska->tracks.elem;
3945  int n;
3946 
3947  matroska_clear_queue(matroska);
3948 
3949  for (n = 0; n < matroska->tracks.nb_elem; n++)
3950  if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
3951  av_freep(&tracks[n].audio.buf);
3952  ebml_free(matroska_segment, matroska);
3953 
3954  return 0;
3955 }
3956 
3957 #if CONFIG_WEBM_DASH_MANIFEST_DEMUXER
3958 typedef struct {
3959  int64_t start_time_ns;
3960  int64_t end_time_ns;
3961  int64_t start_offset;
3962  int64_t end_offset;
3963 } CueDesc;
3964 
3965 /* This function searches all the Cues and returns the CueDesc corresponding to
3966  * the timestamp ts. Returned CueDesc will be such that start_time_ns <= ts <
3967  * end_time_ns. All 4 fields will be set to -1 if ts >= file's duration.
3968  */
3969 static CueDesc get_cue_desc(AVFormatContext *s, int64_t ts, int64_t cues_start) {
3970  MatroskaDemuxContext *matroska = s->priv_data;
3971  FFStream *const sti = ffstream(s->streams[0]);
3972  AVIndexEntry *const index_entries = sti->index_entries;
3973  int nb_index_entries = sti->nb_index_entries;
3974  CueDesc cue_desc;
3975  int i;
3976 
3977  if (ts >= (int64_t)(matroska->duration * matroska->time_scale))
3978  return (CueDesc) {-1, -1, -1, -1};
3979  for (i = 1; i < nb_index_entries; i++) {
3980  if (index_entries[i - 1].timestamp * matroska->time_scale <= ts &&
3981  index_entries[i].timestamp * matroska->time_scale > ts) {
3982  break;
3983  }
3984  }
3985  --i;
3986  cue_desc.start_time_ns = index_entries[i].timestamp * matroska->time_scale;
3987  cue_desc.start_offset = index_entries[i].pos - matroska->segment_start;
3988  if (i != nb_index_entries - 1) {
3989  cue_desc.end_time_ns = index_entries[i + 1].timestamp * matroska->time_scale;
3990  cue_desc.end_offset = index_entries[i + 1].pos - matroska->segment_start;
3991  } else {
3992  cue_desc.end_time_ns = matroska->duration * matroska->time_scale;
3993  // FIXME: this needs special handling for files where Cues appear
3994  // before Clusters. the current logic assumes Cues appear after
3995  // Clusters.
3996  cue_desc.end_offset = cues_start - matroska->segment_start;
3997  }
3998  return cue_desc;
3999 }
4000 
4001 static int webm_clusters_start_with_keyframe(AVFormatContext *s)
4002 {
4003  MatroskaDemuxContext *matroska = s->priv_data;
4004  AVStream *const st = s->streams[0];
4005  FFStream *const sti = ffstream(st);
4006  uint32_t id = matroska->current_id;
4007  int64_t cluster_pos, before_pos;
4008  int index, rv = 1;
4009 
4010  if (sti->nb_index_entries <= 0)
4011  return 0;
4012 
4013  // seek to the first cluster using cues.
4014  index = av_index_search_timestamp(st, 0, 0);
4015  if (index < 0)
4016  return 0;
4017  cluster_pos = sti->index_entries[index].pos;
4018  before_pos = avio_tell(s->pb);
4019  while (1) {
4020  uint64_t cluster_id, cluster_length;
4021  int read;
4022  AVPacket *pkt;
4023  avio_seek(s->pb, cluster_pos, SEEK_SET);
4024  // read cluster id and length
4025  read = ebml_read_num(matroska, matroska->ctx->pb, 4, &cluster_id, 1);
4026  if (read < 0 || cluster_id != 0xF43B675) // done with all clusters
4027  break;
4028  read = ebml_read_length(matroska, matroska->ctx->pb, &cluster_length);
4029  if (read < 0)
4030  break;
4031 
4032  matroska_reset_status(matroska, 0, cluster_pos);
4033  matroska_clear_queue(matroska);
4034  if (matroska_parse_cluster(matroska) < 0 ||
4035  !matroska->queue.head) {
4036  break;
4037  }
4038  pkt = &matroska->queue.head->pkt;
4039  // 4 + read is the length of the cluster id and the cluster length field.
4040  cluster_pos += 4 + read + cluster_length;
4041  if (!(pkt->flags & AV_PKT_FLAG_KEY)) {
4042  rv = 0;
4043  break;
4044  }
4045  }
4046 
4047  /* Restore the status after matroska_read_header: */
4048  matroska_reset_status(matroska, id, before_pos);
4049 
4050  return rv;
4051 }
4052 
4053 static int buffer_size_after_time_downloaded(int64_t time_ns, double search_sec, int64_t bps,
4054  double min_buffer, double* buffer,
4055  double* sec_to_download, AVFormatContext *s,
4056  int64_t cues_start)
4057 {
4058  double nano_seconds_per_second = 1000000000.0;
4059  double time_sec = time_ns / nano_seconds_per_second;
4060  int rv = 0;
4061  int64_t time_to_search_ns = (int64_t)(search_sec * nano_seconds_per_second);
4062  int64_t end_time_ns = time_ns + time_to_search_ns;
4063  double sec_downloaded = 0.0;
4064  CueDesc desc_curr = get_cue_desc(s, time_ns, cues_start);
4065  if (desc_curr.start_time_ns == -1)
4066  return -1;
4067  *sec_to_download = 0.0;
4068 
4069  // Check for non cue start time.
4070  if (time_ns > desc_curr.start_time_ns) {
4071  int64_t cue_nano = desc_curr.end_time_ns - time_ns;
4072  double percent = (double)(cue_nano) / (desc_curr.end_time_ns - desc_curr.start_time_ns);
4073  double cueBytes = (desc_curr.end_offset - desc_curr.start_offset) * percent;
4074  double timeToDownload = (cueBytes * 8.0) / bps;
4075 
4076  sec_downloaded += (cue_nano / nano_seconds_per_second) - timeToDownload;
4077  *sec_to_download += timeToDownload;
4078 
4079  // Check if the search ends within the first cue.
4080  if (desc_curr.end_time_ns >= end_time_ns) {
4081  double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
4082  double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
4083  sec_downloaded = percent_to_sub * sec_downloaded;
4084  *sec_to_download = percent_to_sub * *sec_to_download;
4085  }
4086 
4087  if ((sec_downloaded + *buffer) <= min_buffer) {
4088  return 1;
4089  }
4090 
4091  // Get the next Cue.
4092  desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
4093  }
4094 
4095  while (desc_curr.start_time_ns != -1) {
4096  int64_t desc_bytes = desc_curr.end_offset - desc_curr.start_offset;
4097  int64_t desc_ns = desc_curr.end_time_ns - desc_curr.start_time_ns;
4098  double desc_sec = desc_ns / nano_seconds_per_second;
4099  double bits = (desc_bytes * 8.0);
4100  double time_to_download = bits / bps;
4101 
4102  sec_downloaded += desc_sec - time_to_download;
4103  *sec_to_download += time_to_download;
4104 
4105  if (desc_curr.end_time_ns >= end_time_ns) {
4106  double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
4107  double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
4108  sec_downloaded = percent_to_sub * sec_downloaded;
4109  *sec_to_download = percent_to_sub * *sec_to_download;
4110 
4111  if ((sec_downloaded + *buffer) <= min_buffer)
4112  rv = 1;
4113  break;
4114  }
4115 
4116  if ((sec_downloaded + *buffer) <= min_buffer) {
4117  rv = 1;
4118  break;
4119  }
4120 
4121  desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
4122  }
4123  *buffer = *buffer + sec_downloaded;
4124  return rv;
4125 }
4126 
4127 /* This function computes the bandwidth of the WebM file with the help of
4128  * buffer_size_after_time_downloaded() function. Both of these functions are
4129  * adapted from WebM Tools project and are adapted to work with FFmpeg's
4130  * Matroska parsing mechanism.
4131  *
4132  * Returns the bandwidth of the file on success; -1 on error.
4133  * */
4134 static int64_t webm_dash_manifest_compute_bandwidth(AVFormatContext *s, int64_t cues_start)
4135 {
4136  MatroskaDemuxContext *matroska = s->priv_data;
4137  AVStream *st = s->streams[0];
4138  FFStream *const sti = ffstream(st);
4139  double bandwidth = 0.0;
4140 
4141  for (int i = 0; i < sti->nb_index_entries; i++) {
4142  int64_t prebuffer_ns = 1000000000;
4143  int64_t time_ns = sti->index_entries[i].timestamp * matroska->time_scale;
4144  double nano_seconds_per_second = 1000000000.0;
4145  int64_t prebuffered_ns = time_ns + prebuffer_ns;
4146  double prebuffer_bytes = 0.0;
4147  int64_t temp_prebuffer_ns = prebuffer_ns;
4148  int64_t pre_bytes, pre_ns;
4149  double pre_sec, prebuffer, bits_per_second;
4150  CueDesc desc_beg = get_cue_desc(s, time_ns, cues_start);
4151 
4152  // Start with the first Cue.
4153  CueDesc desc_end = desc_beg;
4154 
4155  // Figure out how much data we have downloaded for the prebuffer. This will
4156  // be used later to adjust the bits per sample to try.
4157  while (desc_end.start_time_ns != -1 && desc_end.end_time_ns < prebuffered_ns) {
4158  // Prebuffered the entire Cue.
4159  prebuffer_bytes += desc_end.end_offset - desc_end.start_offset;
4160  temp_prebuffer_ns -= desc_end.end_time_ns - desc_end.start_time_ns;
4161  desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4162  }
4163  if (desc_end.start_time_ns == -1) {
4164  // The prebuffer is larger than the duration.
4165  if (matroska->duration * matroska->time_scale >= prebuffered_ns)
4166  return -1;
4167  bits_per_second = 0.0;
4168  } else {
4169  // The prebuffer ends in the last Cue. Estimate how much data was
4170  // prebuffered.
4171  pre_bytes = desc_end.end_offset - desc_end.start_offset;
4172  pre_ns = desc_end.end_time_ns - desc_end.start_time_ns;
4173  if (pre_ns <= 0)
4174  return -1;
4175  pre_sec = pre_ns / nano_seconds_per_second;
4176  prebuffer_bytes +=
4177  pre_bytes * ((temp_prebuffer_ns / nano_seconds_per_second) / pre_sec);
4178 
4179  prebuffer = prebuffer_ns / nano_seconds_per_second;
4180 
4181  // Set this to 0.0 in case our prebuffer buffers the entire video.
4182  bits_per_second = 0.0;
4183  do {
4184  int64_t desc_bytes = desc_end.end_offset - desc_beg.start_offset;
4185  int64_t desc_ns = desc_end.end_time_ns - desc_beg.start_time_ns;
4186  double desc_sec, calc_bits_per_second, percent, mod_bits_per_second;
4187  if (desc_bytes <= 0)
4188  return -1;
4189 
4190  desc_sec = desc_ns / nano_seconds_per_second;
4191  calc_bits_per_second = (desc_bytes * 8) / desc_sec;
4192 
4193  // Drop the bps by the percentage of bytes buffered.
4194  percent = (desc_bytes - prebuffer_bytes) / desc_bytes;
4195  mod_bits_per_second = calc_bits_per_second * percent;
4196 
4197  if (prebuffer < desc_sec) {
4198  double search_sec =
4199  (double)(matroska->duration * matroska->time_scale) / nano_seconds_per_second;
4200 
4201  // Add 1 so the bits per second should be a little bit greater than file
4202  // datarate.
4203  int64_t bps = (int64_t)(mod_bits_per_second) + 1;
4204  const double min_buffer = 0.0;
4205  double buffer = prebuffer;
4206  double sec_to_download = 0.0;
4207 
4208  int rv = buffer_size_after_time_downloaded(prebuffered_ns, search_sec, bps,
4209  min_buffer, &buffer, &sec_to_download,
4210  s, cues_start);
4211  if (rv < 0) {
4212  return -1;
4213  } else if (rv == 0) {
4214  bits_per_second = (double)(bps);
4215  break;
4216  }
4217  }
4218 
4219  desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4220  } while (desc_end.start_time_ns != -1);
4221  }
4222  if (bandwidth < bits_per_second) bandwidth = bits_per_second;
4223  }
4224  return (int64_t)bandwidth;
4225 }
4226 
4227 static int webm_dash_manifest_cues(AVFormatContext *s, int64_t init_range)
4228 {
4229  MatroskaDemuxContext *matroska = s->priv_data;
4230  EbmlList *seekhead_list = &matroska->seekhead;
4231  MatroskaSeekhead *seekhead = seekhead_list->elem;
4232  AVStream *const st = s->streams[0];
4233  FFStream *const sti = ffstream(st);
4234  AVBPrint bprint;
4235  char *buf;
4236  int64_t cues_start = -1, cues_end = -1, before_pos, bandwidth;
4237  int i;
4238  int ret;
4239 
4240  // determine cues start and end positions
4241  for (i = 0; i < seekhead_list->nb_elem; i++)
4242  if (seekhead[i].id == MATROSKA_ID_CUES)
4243  break;
4244 
4245  if (i >= seekhead_list->nb_elem) return -1;
4246 
4247  before_pos = avio_tell(matroska->ctx->pb);
4248  cues_start = seekhead[i].pos + matroska->segment_start;
4249  if (avio_seek(matroska->ctx->pb, cues_start, SEEK_SET) == cues_start) {
4250  // cues_end is computed as cues_start + cues_length + length of the
4251  // Cues element ID (i.e. 4) + EBML length of the Cues element.
4252  // cues_end is inclusive and the above sum is reduced by 1.
4253  uint64_t cues_length, cues_id;
4254  int bytes_read;
4255  bytes_read = ebml_read_num (matroska, matroska->ctx->pb, 4, &cues_id, 1);
4256  if (bytes_read < 0 || cues_id != (MATROSKA_ID_CUES & 0xfffffff))
4257  return bytes_read < 0 ? bytes_read : AVERROR_INVALIDDATA;
4258  bytes_read = ebml_read_length(matroska, matroska->ctx->pb, &cues_length);
4259  if (bytes_read < 0)
4260  return bytes_read;
4261  cues_end = cues_start + 4 + bytes_read + cues_length - 1;
4262  }
4263  avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
4264  if (cues_start == -1 || cues_end == -1) return -1;
4265 
4266  // parse the cues
4267  matroska_parse_cues(matroska);
4268 
4269  if (!sti->nb_index_entries)
4270  return AVERROR_INVALIDDATA;
4271 
4272  // cues start
4273  av_dict_set_int(&s->streams[0]->metadata, CUES_START, cues_start, 0);
4274 
4275  // cues end
4276  av_dict_set_int(&s->streams[0]->metadata, CUES_END, cues_end, 0);
4277 
4278  // if the file has cues at the start, fix up the init range so that
4279  // it does not include it
4280  if (cues_start <= init_range)
4281  av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, cues_start - 1, 0);
4282 
4283  // bandwidth
4284  bandwidth = webm_dash_manifest_compute_bandwidth(s, cues_start);
4285  if (bandwidth < 0) return -1;
4286  av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH, bandwidth, 0);
4287 
4288  // check if all clusters start with key frames
4289  av_dict_set_int(&s->streams[0]->metadata, CLUSTER_KEYFRAME, webm_clusters_start_with_keyframe(s), 0);
4290 
4291  // Store cue point timestamps as a comma separated list
4292  // for checking subsegment alignment in the muxer.
4294  for (int i = 0; i < sti->nb_index_entries; i++)
4295  av_bprintf(&bprint, "%" PRId64",", sti->index_entries[i].timestamp);
4296  if (!av_bprint_is_complete(&bprint)) {
4297  av_bprint_finalize(&bprint, NULL);
4298  return AVERROR(ENOMEM);
4299  }
4300  // Remove the trailing ','
4301  bprint.str[--bprint.len] = '\0';
4302  if ((ret = av_bprint_finalize(&bprint, &buf)) < 0)
4303  return ret;
4304  av_dict_set(&s->streams[0]->metadata, CUE_TIMESTAMPS,
4306 
4307  return 0;
4308 }
4309 
4310 static int webm_dash_manifest_read_header(AVFormatContext *s)
4311 {
4312  char *buf;
4313  int ret = matroska_read_header(s);
4314  int64_t init_range;
4315  MatroskaTrack *tracks;
4316  MatroskaDemuxContext *matroska = s->priv_data;
4317  if (ret) {
4318  av_log(s, AV_LOG_ERROR, "Failed to read file headers\n");
4319  return -1;
4320  }
4321  if (!matroska->tracks.nb_elem || !s->nb_streams) {
4322  av_log(s, AV_LOG_ERROR, "No track found\n");
4323  return AVERROR_INVALIDDATA;
4324  }
4325 
4326  if (!matroska->is_live) {
4327  buf = av_asprintf("%g", matroska->duration);
4328  if (!buf)
4329  return AVERROR(ENOMEM);
4330  av_dict_set(&s->streams[0]->metadata, DURATION,
4332 
4333  // initialization range
4334  // 5 is the offset of Cluster ID.
4335  init_range = avio_tell(s->pb) - 5;
4336  av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, init_range, 0);
4337  }
4338 
4339  // basename of the file
4340  buf = strrchr(s->url, '/');
4341  av_dict_set(&s->streams[0]->metadata, FILENAME, buf ? ++buf : s->url, 0);
4342 
4343  // track number
4344  tracks = matroska->tracks.elem;
4345  av_dict_set_int(&s->streams[0]->metadata, TRACK_NUMBER, tracks[0].num, 0);
4346 
4347  // parse the cues and populate Cue related fields
4348  if (!matroska->is_live) {
4349  ret = webm_dash_manifest_cues(s, init_range);
4350  if (ret < 0) {
4351  av_log(s, AV_LOG_ERROR, "Error parsing Cues\n");
4352  return ret;
4353  }
4354  }
4355 
4356  // use the bandwidth from the command line if it was provided
4357  if (matroska->bandwidth > 0) {
4358  av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH,
4359  matroska->bandwidth, 0);
4360  }
4361  return 0;
4362 }
4363 
4364 static int webm_dash_manifest_read_packet(AVFormatContext *s, AVPacket *pkt)
4365 {
4366  return AVERROR_EOF;
4367 }
4368 
4369 #define OFFSET(x) offsetof(MatroskaDemuxContext, x)
4370 static const AVOption options[] = {
4371  { "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 },
4372  { "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 },
4373  { NULL },
4374 };
4375 
4376 static const AVClass webm_dash_class = {
4377  .class_name = "WebM DASH Manifest demuxer",
4378  .item_name = av_default_item_name,
4379  .option = options,
4380  .version = LIBAVUTIL_VERSION_INT,
4381 };
4382 
4384  .name = "webm_dash_manifest",
4385  .long_name = NULL_IF_CONFIG_SMALL("WebM DASH Manifest"),
4386  .priv_class = &webm_dash_class,
4387  .priv_data_size = sizeof(MatroskaDemuxContext),
4388  .flags_internal = FF_FMT_INIT_CLEANUP,
4389  .read_header = webm_dash_manifest_read_header,
4390  .read_packet = webm_dash_manifest_read_packet,
4392 };
4393 #endif
4394 
4396  .name = "matroska,webm",
4397  .long_name = NULL_IF_CONFIG_SMALL("Matroska / WebM"),
4398  .extensions = "mkv,mk3d,mka,mks,webm",
4399  .priv_data_size = sizeof(MatroskaDemuxContext),
4400  .flags_internal = FF_FMT_INIT_CLEANUP,
4403  .