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 #include "config_components.h"
33 
34 #include <inttypes.h>
35 #include <stdio.h>
36 
37 #include "libavutil/avstring.h"
38 #include "libavutil/base64.h"
39 #include "libavutil/bprint.h"
40 #include "libavutil/dict.h"
42 #include "libavutil/display.h"
43 #include "libavutil/intfloat.h"
44 #include "libavutil/intreadwrite.h"
45 #include "libavutil/lzo.h"
47 #include "libavutil/mathematics.h"
48 #include "libavutil/opt.h"
49 #include "libavutil/pixdesc.h"
51 #include "libavutil/spherical.h"
52 
53 #include "libavcodec/bytestream.h"
54 #include "libavcodec/flac.h"
55 #include "libavcodec/mpeg4audio.h"
57 
58 #include "avformat.h"
59 #include "avio_internal.h"
60 #include "demux.h"
61 #include "dovi_isom.h"
62 #include "internal.h"
63 #include "isom.h"
64 #include "matroska.h"
65 #include "oggdec.h"
66 /* For ff_codec_get_id(). */
67 #include "riff.h"
68 #include "rmsipr.h"
69 
70 #if CONFIG_BZLIB
71 #include <bzlib.h>
72 #endif
73 #if CONFIG_ZLIB
74 #include <zlib.h>
75 #endif
76 
77 #include "qtpalette.h"
78 
79 #define EBML_UNKNOWN_LENGTH UINT64_MAX /* EBML unknown length, in uint64_t */
80 #define NEEDS_CHECKING 2 /* Indicates that some error checks
81  * still need to be performed */
82 #define LEVEL_ENDED 3 /* return value of ebml_parse when the
83  * syntax level used for parsing ended. */
84 #define SKIP_THRESHOLD 1024 * 1024 /* In non-seekable mode, if more than SKIP_THRESHOLD
85  * of unkown, potentially damaged data is encountered,
86  * it is considered an error. */
87 #define UNKNOWN_EQUIV 50 * 1024 /* An unknown element is considered equivalent
88  * to this many bytes of unknown data for the
89  * SKIP_THRESHOLD check. */
90 
91 typedef enum {
97  EBML_UTF8,
98  EBML_BIN,
100  EBML_LEVEL1,
104 
105 typedef struct CountedElement {
106  union {
107  uint64_t u;
108  int64_t i;
109  double f;
110  char *s;
111  } el;
112  unsigned count;
114 
115 typedef const struct EbmlSyntax {
116  uint32_t id;
117  uint8_t type;
118  uint8_t is_counted;
120  size_t data_offset;
121  union {
122  int64_t i;
123  uint64_t u;
124  double f;
125  const char *s;
126  const struct EbmlSyntax *n;
127  } def;
128 } EbmlSyntax;
129 
130 typedef struct EbmlList {
131  int nb_elem;
132  unsigned int alloc_elem_size;
133  void *elem;
135 
136 typedef struct EbmlBin {
137  int size;
139  uint8_t *data;
140  int64_t pos;
142 
143 typedef struct Ebml {
144  uint64_t version;
145  uint64_t max_size;
146  uint64_t id_length;
147  char *doctype;
148  uint64_t doctype_version;
149 } Ebml;
150 
151 typedef struct MatroskaTrackCompression {
152  uint64_t algo;
155 
156 typedef struct MatroskaTrackEncryption {
157  uint64_t algo;
160 
161 typedef struct MatroskaTrackEncoding {
162  uint64_t scope;
163  uint64_t type;
167 
168 typedef struct MatroskaMasteringMeta {
169  double r_x;
170  double r_y;
171  double g_x;
172  double g_y;
173  double b_x;
174  double b_y;
175  double white_x;
176  double white_y;
180 
181 typedef struct MatroskaTrackVideoColor {
184  uint64_t chroma_sub_horz;
185  uint64_t chroma_sub_vert;
186  uint64_t cb_sub_horz;
187  uint64_t cb_sub_vert;
190  uint64_t range;
191  uint64_t transfer_characteristics;
192  uint64_t primaries;
193  uint64_t max_cll;
194  uint64_t max_fall;
197 
198 typedef struct MatroskaTrackVideoProjection {
199  uint64_t type;
200  EbmlBin private;
201  double yaw;
202  double pitch;
203  double roll;
205 
206 typedef struct MatroskaTrackVideo {
207  double frame_rate;
208  uint64_t display_width;
209  uint64_t display_height;
210  uint64_t pixel_width;
211  uint64_t pixel_height;
213  uint64_t display_unit;
214  uint64_t interlaced;
215  uint64_t field_order;
216  uint64_t stereo_mode;
217  uint64_t alpha_mode;
221 
222 typedef struct MatroskaTrackAudio {
223  double samplerate;
225  uint64_t bitdepth;
226  uint64_t channels;
227 
228  /* real audio header (extracted from extradata) */
231  int frame_size;
232  int sub_packet_size;
234  int pkt_cnt;
235  uint64_t buf_timecode;
236  uint8_t *buf;
238 
239 typedef struct MatroskaTrackPlane {
240  uint64_t uid;
241  uint64_t type;
243 
244 typedef struct MatroskaTrackOperation {
247 
248 typedef struct MatroskaBlockAdditionMapping {
249  uint64_t value;
250  char *name;
251  uint64_t type;
254 
255 typedef struct MatroskaTrack {
256  uint64_t num;
257  uint64_t uid;
258  uint64_t type;
259  char *name;
260  char *codec_id;
262  char *language;
263  double time_scale;
265  uint64_t flag_default;
266  uint64_t flag_forced;
267  uint64_t flag_comment;
272  uint64_t seek_preroll;
277  uint64_t codec_delay;
279 
280  AVStream *stream;
281  int64_t end_timecode;
283  int needs_decoding;
284  uint64_t max_block_additional_id;
286 
290 
291 typedef struct MatroskaAttachment {
292  uint64_t uid;
293  char *filename;
294  char *description;
295  char *mime;
297 
300 
301 typedef struct MatroskaChapter {
302  uint64_t start;
303  uint64_t end;
304  uint64_t uid;
305  char *title;
306 
309 
310 typedef struct MatroskaIndexPos {
311  uint64_t track;
312  uint64_t pos;
314 
315 typedef struct MatroskaIndex {
316  uint64_t time;
319 
320 typedef struct MatroskaTag {
321  char *name;
322  char *string;
323  char *lang;
324  uint64_t def;
327 
328 typedef struct MatroskaTagTarget {
329  char *type;
330  uint64_t typevalue;
331  uint64_t trackuid;
332  uint64_t chapteruid;
333  uint64_t attachuid;
335 
336 typedef struct MatroskaTags {
338  EbmlList tag;
339 } MatroskaTags;
340 
341 typedef struct MatroskaSeekhead {
342  uint64_t id;
343  uint64_t pos;
345 
346 typedef struct MatroskaLevel {
347  uint64_t start;
348  uint64_t length;
350 
351 typedef struct MatroskaBlock {
352  uint64_t duration;
354  uint64_t non_simple;
356  uint64_t additional_id;
359 } MatroskaBlock;
360 
361 typedef struct MatroskaCluster {
363  uint64_t timecode;
364  int64_t pos;
366 
367 typedef struct MatroskaLevel1Element {
368  int64_t pos;
369  uint32_t id;
370  int parsed;
372 
373 typedef struct MatroskaDemuxContext {
374  const AVClass *class;
376 
377  /* EBML stuff */
380  uint32_t current_id;
381  int64_t resync_pos;
383 
384  uint64_t time_scale;
385  double duration;
386  char *title;
387  char *muxingapp;
392  EbmlList index;
393  EbmlList tags;
395 
396  /* byte position of the segment inside the stream */
397  int64_t segment_start;
398 
399  /* This packet coincides with FFFormatContext.parse_pkt
400  * and is not owned by us. */
401  AVPacket *pkt;
402 
403  /* the packet queue */
405 
406  int done;
407 
408  /* What to skip before effectively reading a packet. */
409  int skip_to_keyframe;
411 
412  /* File has a CUES element, but we defer parsing until it is needed. */
414 
415  /* Level1 elements and whether they were read yet */
417  int num_level1_elems;
418 
420 
421  /* WebM DASH Manifest live flag */
422  int is_live;
423 
424  /* Bandwidth value for WebM DASH Manifest */
425  int bandwidth;
427 
428 #define CHILD_OF(parent) { .def = { .n = parent } }
429 
430 // The following forward declarations need their size because
431 // a tentative definition with internal linkage must not be an
432 // incomplete type (6.7.2 in C90, 6.9.2 in C99).
433 // Removing the sizes breaks MSVC.
440 
441 static EbmlSyntax ebml_header[] = {
442  { EBML_ID_EBMLREADVERSION, EBML_UINT, 0, 0, offsetof(Ebml, version), { .u = EBML_VERSION } },
443  { EBML_ID_EBMLMAXSIZELENGTH, EBML_UINT, 0, 0, offsetof(Ebml, max_size), { .u = 8 } },
444  { EBML_ID_EBMLMAXIDLENGTH, EBML_UINT, 0, 0, offsetof(Ebml, id_length), { .u = 4 } },
445  { EBML_ID_DOCTYPE, EBML_STR, 0, 0, offsetof(Ebml, doctype), { .s = "(none)" } },
446  { EBML_ID_DOCTYPEREADVERSION, EBML_UINT, 0, 0, offsetof(Ebml, doctype_version), { .u = 1 } },
450 };
451 
453  { EBML_ID_HEADER, EBML_NEST, 0, 0, 0, { .n = ebml_header } },
455  { 0 }
456 };
457 
458 static EbmlSyntax matroska_info[] = {
459  { MATROSKA_ID_TIMECODESCALE, EBML_UINT, 0, 0, offsetof(MatroskaDemuxContext, time_scale), { .u = 1000000 } },
461  { MATROSKA_ID_TITLE, EBML_UTF8, 0, 0, offsetof(MatroskaDemuxContext, title) },
463  { MATROSKA_ID_MUXINGAPP, EBML_UTF8, 0, 0, offsetof(MatroskaDemuxContext, muxingapp) },
464  { MATROSKA_ID_DATEUTC, EBML_BIN, 0, 0, offsetof(MatroskaDemuxContext, date_utc) },
467 };
468 
476  { MATROSKA_ID_VIDEOCOLOR_WHITEX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, white_x) },
477  { MATROSKA_ID_VIDEOCOLOR_WHITEY, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, white_y) },
478  { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMIN, EBML_FLOAT, 1, 0, offsetof(MatroskaMasteringMeta, min_luminance) },
479  { MATROSKA_ID_VIDEOCOLOR_LUMINANCEMAX, EBML_FLOAT, 0, 0, offsetof(MatroskaMasteringMeta, max_luminance) },
481 };
482 
484  { MATROSKA_ID_VIDEOCOLORMATRIXCOEFF, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, matrix_coefficients), { .u = AVCOL_SPC_UNSPECIFIED } },
485  { MATROSKA_ID_VIDEOCOLORBITSPERCHANNEL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, bits_per_channel), { .u = 0 } },
486  { MATROSKA_ID_VIDEOCOLORCHROMASUBHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_horz) },
487  { MATROSKA_ID_VIDEOCOLORCHROMASUBVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, chroma_sub_vert) },
488  { MATROSKA_ID_VIDEOCOLORCBSUBHORZ, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, cb_sub_horz) },
489  { MATROSKA_ID_VIDEOCOLORCBSUBVERT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, cb_sub_vert) },
496  { MATROSKA_ID_VIDEOCOLORMAXFALL, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideoColor, max_fall) },
499 };
500 
504  { MATROSKA_ID_VIDEOPROJECTIONPOSEYAW, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, yaw), { .f = 0.0 } },
505  { MATROSKA_ID_VIDEOPROJECTIONPOSEPITCH, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, pitch), { .f = 0.0 } },
506  { MATROSKA_ID_VIDEOPROJECTIONPOSEROLL, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideoProjection, roll), { .f = 0.0 } },
508 };
509 
510 static EbmlSyntax matroska_track_video[] = {
511  { MATROSKA_ID_VIDEOFRAMERATE, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackVideo, frame_rate) },
512  { MATROSKA_ID_VIDEODISPLAYWIDTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_width), { .u=-1 } },
513  { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, display_height), { .u=-1 } },
514  { MATROSKA_ID_VIDEOPIXELWIDTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, pixel_width) },
515  { MATROSKA_ID_VIDEOPIXELHEIGHT, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, pixel_height) },
516  { MATROSKA_ID_VIDEOCOLORSPACE, EBML_BIN, 0, 0, offsetof(MatroskaTrackVideo, color_space) },
517  { MATROSKA_ID_VIDEOALPHAMODE, EBML_UINT, 0, 0, offsetof(MatroskaTrackVideo, alpha_mode), { .u = 0 } },
530 };
531 
532 static EbmlSyntax matroska_track_audio[] = {
533  { MATROSKA_ID_AUDIOSAMPLINGFREQ, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackAudio, samplerate), { .f = 8000.0 } },
534  { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ, EBML_FLOAT, 0, 0, offsetof(MatroskaTrackAudio, out_samplerate) },
535  { MATROSKA_ID_AUDIOBITDEPTH, EBML_UINT, 0, 0, offsetof(MatroskaTrackAudio, bitdepth) },
536  { MATROSKA_ID_AUDIOCHANNELS, EBML_UINT, 0, 0, offsetof(MatroskaTrackAudio, channels), { .u = 1 } },
538 };
539 
544 };
545 
547  { MATROSKA_ID_ENCODINGENCALGO, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncryption,algo), {.u = 0} },
555 };
557  { MATROSKA_ID_ENCODINGSCOPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncoding, scope), { .u = 1 } },
558  { MATROSKA_ID_ENCODINGTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrackEncoding, type), { .u = 0 } },
563 };
564 
568 };
569 
570 static EbmlSyntax matroska_track_plane[] = {
574 };
575 
577  { MATROSKA_ID_TRACKPLANE, EBML_NEST, 0, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n = matroska_track_plane} },
579 };
580 
584 };
585 
592 };
593 
594 static EbmlSyntax matroska_track[] = {
595  { MATROSKA_ID_TRACKNUMBER, EBML_UINT, 0, 0, offsetof(MatroskaTrack, num) },
596  { MATROSKA_ID_TRACKNAME, EBML_UTF8, 0, 0, offsetof(MatroskaTrack, name) },
597  { MATROSKA_ID_TRACKUID, EBML_UINT, 0, 0, offsetof(MatroskaTrack, uid) },
598  { MATROSKA_ID_TRACKTYPE, EBML_UINT, 0, 0, offsetof(MatroskaTrack, type) },
599  { MATROSKA_ID_CODECID, EBML_STR, 0, 0, offsetof(MatroskaTrack, codec_id) },
600  { MATROSKA_ID_CODECPRIVATE, EBML_BIN, 0, 0, offsetof(MatroskaTrack, codec_priv) },
601  { MATROSKA_ID_CODECDELAY, EBML_UINT, 0, 0, offsetof(MatroskaTrack, codec_delay), { .u = 0 } },
602  { MATROSKA_ID_TRACKLANGUAGE, EBML_STR, 0, 0, offsetof(MatroskaTrack, language), { .s = "eng" } },
603  { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, 0, offsetof(MatroskaTrack, default_duration) },
604  { MATROSKA_ID_TRACKTIMECODESCALE, EBML_FLOAT, 0, 0, offsetof(MatroskaTrack, time_scale), { .f = 1.0 } },
605  { MATROSKA_ID_TRACKFLAGCOMMENTARY, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_comment), { .u = 0 } },
606  { MATROSKA_ID_TRACKFLAGDEFAULT, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_default), { .u = 1 } },
607  { MATROSKA_ID_TRACKFLAGFORCED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_forced), { .u = 0 } },
608  { MATROSKA_ID_TRACKFLAGHEARINGIMPAIRED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_hearingimpaired), { .u = 0 } },
609  { MATROSKA_ID_TRACKFLAGVISUALIMPAIRED, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_visualimpaired), { .u = 0 } },
610  { MATROSKA_ID_TRACKFLAGTEXTDESCRIPTIONS, EBML_UINT, 0, 0, offsetof(MatroskaTrack, flag_textdescriptions), { .u = 0 } },
611  { MATROSKA_ID_TRACKFLAGORIGINAL, EBML_UINT, 1, 0, offsetof(MatroskaTrack, flag_original), {.u = 0 } },
613  { MATROSKA_ID_TRACKAUDIO, EBML_NEST, 0, 0, offsetof(MatroskaTrack, audio), { .n = matroska_track_audio } },
614  { MATROSKA_ID_TRACKOPERATION, EBML_NEST, 0, 0, offsetof(MatroskaTrack, operation), { .n = matroska_track_operation } },
616  { MATROSKA_ID_TRACKMAXBLKADDID, EBML_UINT, 0, 0, offsetof(MatroskaTrack, max_block_additional_id), { .u = 0 } },
617  { MATROSKA_ID_TRACKBLKADDMAPPING, EBML_NEST, 0, sizeof(MatroskaBlockAdditionMapping), offsetof(MatroskaTrack, block_addition_mappings), { .n = matroska_block_addition_mapping } },
618  { MATROSKA_ID_SEEKPREROLL, EBML_UINT, 0, 0, offsetof(MatroskaTrack, seek_preroll), { .u = 0 } },
628 };
629 
630 static EbmlSyntax matroska_tracks[] = {
631  { MATROSKA_ID_TRACKENTRY, EBML_NEST, 0, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext, tracks), { .n = matroska_track } },
633 };
634 
635 static EbmlSyntax matroska_attachment[] = {
636  { MATROSKA_ID_FILEUID, EBML_UINT, 0, 0, offsetof(MatroskaAttachment, uid) },
637  { MATROSKA_ID_FILENAME, EBML_UTF8, 0, 0, offsetof(MatroskaAttachment, filename) },
638  { MATROSKA_ID_FILEMIMETYPE, EBML_STR, 0, 0, offsetof(MatroskaAttachment, mime) },
639  { MATROSKA_ID_FILEDATA, EBML_BIN, 0, 0, offsetof(MatroskaAttachment, bin) },
642 };
643 
644 static EbmlSyntax matroska_attachments[] = {
645  { MATROSKA_ID_ATTACHEDFILE, EBML_NEST, 0, sizeof(MatroskaAttachment), offsetof(MatroskaDemuxContext, attachments), { .n = matroska_attachment } },
647 };
648 
650  { MATROSKA_ID_CHAPSTRING, EBML_UTF8, 0, 0, offsetof(MatroskaChapter, title) },
654 };
655 
657  { MATROSKA_ID_CHAPTERTIMESTART, EBML_UINT, 0, 0, offsetof(MatroskaChapter, start), { .u = AV_NOPTS_VALUE } },
658  { MATROSKA_ID_CHAPTERTIMEEND, EBML_UINT, 0, 0, offsetof(MatroskaChapter, end), { .u = AV_NOPTS_VALUE } },
659  { MATROSKA_ID_CHAPTERUID, EBML_UINT, 0, 0, offsetof(MatroskaChapter, uid) },
666 };
667 
668 static EbmlSyntax matroska_chapter[] = {
669  { MATROSKA_ID_CHAPTERATOM, EBML_NEST, 0, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext, chapters), { .n = matroska_chapter_entry } },
675 };
676 
677 static EbmlSyntax matroska_chapters[] = {
678  { MATROSKA_ID_EDITIONENTRY, EBML_NEST, 0, 0, 0, { .n = matroska_chapter } },
680 };
681 
682 static EbmlSyntax matroska_index_pos[] = {
683  { MATROSKA_ID_CUETRACK, EBML_UINT, 0, 0, offsetof(MatroskaIndexPos, track) },
689 };
690 
691 static EbmlSyntax matroska_index_entry[] = {
692  { MATROSKA_ID_CUETIME, EBML_UINT, 0, 0, offsetof(MatroskaIndex, time) },
695 };
696 
697 static EbmlSyntax matroska_index[] = {
700 };
701 
702 static EbmlSyntax matroska_simpletag[] = {
703  { MATROSKA_ID_TAGNAME, EBML_UTF8, 0, 0, offsetof(MatroskaTag, name) },
704  { MATROSKA_ID_TAGSTRING, EBML_UTF8, 0, 0, offsetof(MatroskaTag, string) },
705  { MATROSKA_ID_TAGLANG, EBML_STR, 0, 0, offsetof(MatroskaTag, lang), { .s = "und" } },
706  { MATROSKA_ID_TAGDEFAULT, EBML_UINT, 0, 0, offsetof(MatroskaTag, def) },
707  { MATROSKA_ID_TAGDEFAULT_BUG, EBML_UINT, 0, 0, offsetof(MatroskaTag, def) },
708  { MATROSKA_ID_SIMPLETAG, EBML_NEST, 0, sizeof(MatroskaTag), offsetof(MatroskaTag, sub), { .n = matroska_simpletag } },
710 };
711 
712 static EbmlSyntax matroska_tagtargets[] = {
714  { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, typevalue), { .u = 50 } },
715  { MATROSKA_ID_TAGTARGETS_TRACKUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, trackuid), { .u = 0 } },
716  { MATROSKA_ID_TAGTARGETS_CHAPTERUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, chapteruid), { .u = 0 } },
717  { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, 0, offsetof(MatroskaTagTarget, attachuid), { .u = 0 } },
719 };
720 
721 static EbmlSyntax matroska_tag[] = {
722  { MATROSKA_ID_SIMPLETAG, EBML_NEST, 0, sizeof(MatroskaTag), offsetof(MatroskaTags, tag), { .n = matroska_simpletag } },
723  { MATROSKA_ID_TAGTARGETS, EBML_NEST, 0, 0, offsetof(MatroskaTags, target), { .n = matroska_tagtargets } },
725 };
726 
727 static EbmlSyntax matroska_tags[] = {
728  { MATROSKA_ID_TAG, EBML_NEST, 0, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext, tags), { .n = matroska_tag } },
730 };
731 
733  { MATROSKA_ID_SEEKID, EBML_UINT, 0, 0, offsetof(MatroskaSeekhead, id) },
734  { MATROSKA_ID_SEEKPOSITION, EBML_UINT, 0, 0, offsetof(MatroskaSeekhead, pos), { .u = -1 } },
736 };
737 
738 static EbmlSyntax matroska_seekhead[] = {
739  { MATROSKA_ID_SEEKENTRY, EBML_NEST, 0, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext, seekhead), { .n = matroska_seekhead_entry } },
741 };
742 
743 static EbmlSyntax matroska_segment[] = {
745  { MATROSKA_ID_INFO, EBML_LEVEL1, 0, 0, 0, { .n = matroska_info } },
746  { MATROSKA_ID_TRACKS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_tracks } },
748  { MATROSKA_ID_CHAPTERS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_chapters } },
749  { MATROSKA_ID_CUES, EBML_LEVEL1, 0, 0, 0, { .n = matroska_index } },
750  { MATROSKA_ID_TAGS, EBML_LEVEL1, 0, 0, 0, { .n = matroska_tags } },
751  { MATROSKA_ID_SEEKHEAD, EBML_LEVEL1, 0, 0, 0, { .n = matroska_seekhead } },
752  { 0 } /* We don't want to go back to level 0, so don't add the parent. */
753 };
754 
755 static EbmlSyntax matroska_segments[] = {
756  { MATROSKA_ID_SEGMENT, EBML_NEST, 0, 0, 0, { .n = matroska_segment } },
757  { 0 }
758 };
759 
760 static EbmlSyntax matroska_blockmore[] = {
761  { MATROSKA_ID_BLOCKADDID, EBML_UINT, 0, 0, offsetof(MatroskaBlock,additional_id), { .u = 1 } },
762  { MATROSKA_ID_BLOCKADDITIONAL, EBML_BIN, 0, 0, offsetof(MatroskaBlock,additional) },
764 };
765 
767  { MATROSKA_ID_BLOCKMORE, EBML_NEST, 0, 0, 0, {.n = matroska_blockmore} },
769 };
770 
771 static EbmlSyntax matroska_blockgroup[] = {
772  { MATROSKA_ID_BLOCK, EBML_BIN, 0, 0, offsetof(MatroskaBlock, bin) },
775  { MATROSKA_ID_DISCARDPADDING, EBML_SINT, 0, 0, offsetof(MatroskaBlock, discard_padding) },
776  { MATROSKA_ID_BLOCKREFERENCE, EBML_SINT, 1, 0, offsetof(MatroskaBlock, reference) },
778  { 1, EBML_UINT, 0, 0, offsetof(MatroskaBlock, non_simple), { .u = 1 } },
780 };
781 
782 // The following array contains SimpleBlock and BlockGroup twice
783 // in order to reuse the other values for matroska_cluster_enter.
785  { MATROSKA_ID_SIMPLEBLOCK, EBML_BIN, 0, 0, offsetof(MatroskaBlock, bin) },
786  { MATROSKA_ID_BLOCKGROUP, EBML_NEST, 0, 0, 0, { .n = matroska_blockgroup } },
787  { MATROSKA_ID_CLUSTERTIMECODE, EBML_UINT, 0, 0, offsetof(MatroskaCluster, timecode) },
793 };
794 
796  { MATROSKA_ID_CLUSTER, EBML_NEST, 0, 0, 0, { .n = &matroska_cluster_parsing[2] } },
797  { 0 }
798 };
799 #undef CHILD_OF
800 
801 static const CodecMime mkv_image_mime_tags[] = {
802  {"image/gif" , AV_CODEC_ID_GIF},
803  {"image/jpeg" , AV_CODEC_ID_MJPEG},
804  {"image/png" , AV_CODEC_ID_PNG},
805  {"image/tiff" , AV_CODEC_ID_TIFF},
806 
807  {"" , AV_CODEC_ID_NONE}
808 };
809 
810 static const CodecMime mkv_mime_tags[] = {
811  {"application/x-truetype-font", AV_CODEC_ID_TTF},
812  {"application/x-font" , AV_CODEC_ID_TTF},
813  {"application/vnd.ms-opentype", AV_CODEC_ID_OTF},
814  {"binary" , AV_CODEC_ID_BIN_DATA},
815 
816  {"" , AV_CODEC_ID_NONE}
817 };
818 
819 static const char *const matroska_doctypes[] = { "matroska", "webm" };
820 
821 /*
822  * This function prepares the status for parsing of level 1 elements.
823  */
824 static int matroska_reset_status(MatroskaDemuxContext *matroska,
825  uint32_t id, int64_t position)
826 {
827  int64_t err = 0;
828  if (position >= 0) {
829  err = avio_seek(matroska->ctx->pb, position, SEEK_SET);
830  if (err > 0)
831  err = 0;
832  } else
833  position = avio_tell(matroska->ctx->pb);
834 
835  matroska->current_id = id;
836  matroska->num_levels = 1;
837  matroska->unknown_count = 0;
838  matroska->resync_pos = position;
839  if (id)
840  matroska->resync_pos -= (av_log2(id) + 7) / 8;
841 
842  return err;
843 }
844 
845 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
846 {
847  AVIOContext *pb = matroska->ctx->pb;
848  uint32_t id;
849 
850  /* Try to seek to the last position to resync from. If this doesn't work,
851  * we resync from the earliest position available: The start of the buffer. */
852  if (last_pos < avio_tell(pb) && avio_seek(pb, last_pos + 1, SEEK_SET) < 0) {
853  av_log(matroska->ctx, AV_LOG_WARNING,
854  "Seek to desired resync point failed. Seeking to "
855  "earliest point available instead.\n");
856  avio_seek(pb, FFMAX(avio_tell(pb) + (pb->buffer - pb->buf_ptr),
857  last_pos + 1), SEEK_SET);
858  }
859 
860  id = avio_rb32(pb);
861 
862  // try to find a toplevel element
863  while (!avio_feof(pb)) {
864  if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
865  id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
867  id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS) {
868  /* Prepare the context for parsing of a level 1 element. */
869  matroska_reset_status(matroska, id, -1);
870  /* Given that we are here means that an error has occurred,
871  * so treat the segment as unknown length in order not to
872  * discard valid data that happens to be beyond the designated
873  * end of the segment. */
874  matroska->levels[0].length = EBML_UNKNOWN_LENGTH;
875  return 0;
876  }
877  id = (id << 8) | avio_r8(pb);
878  }
879 
880  matroska->done = 1;
881  return pb->error ? pb->error : AVERROR_EOF;
882 }
883 
884 /*
885  * Read: an "EBML number", which is defined as a variable-length
886  * array of bytes. The first byte indicates the length by giving a
887  * number of 0-bits followed by a one. The position of the first
888  * "one" bit inside the first byte indicates the length of this
889  * number.
890  * Returns: number of bytes read, < 0 on error
891  */
892 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
893  int max_size, uint64_t *number, int eof_forbidden)
894 {
895  int read, n = 1;
896  uint64_t total;
897  int64_t pos;
898 
899  /* The first byte tells us the length in bytes - except when it is zero. */
900  total = avio_r8(pb);
901  if (pb->eof_reached)
902  goto err;
903 
904  /* get the length of the EBML number */
905  read = 8 - ff_log2_tab[total];
906 
907  if (!total || read > max_size) {
908  pos = avio_tell(pb) - 1;
909  if (!total) {
910  av_log(matroska->ctx, AV_LOG_ERROR,
911  "0x00 at pos %"PRId64" (0x%"PRIx64") invalid as first byte "
912  "of an EBML number\n", pos, pos);
913  } else {
914  av_log(matroska->ctx, AV_LOG_ERROR,
915  "Length %d indicated by an EBML number's first byte 0x%02x "
916  "at pos %"PRId64" (0x%"PRIx64") exceeds max length %d.\n",
917  read, (uint8_t) total, pos, pos, max_size);
918  }
919  return AVERROR_INVALIDDATA;
920  }
921 
922  /* read out length */
923  total ^= 1 << ff_log2_tab[total];
924  while (n++ < read)
925  total = (total << 8) | avio_r8(pb);
926 
927  if (pb->eof_reached) {
928  eof_forbidden = 1;
929  goto err;
930  }
931 
932  *number = total;
933 
934  return read;
935 
936 err:
937  pos = avio_tell(pb);
938  if (pb->error) {
939  av_log(matroska->ctx, AV_LOG_ERROR,
940  "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
941  pos, pos);
942  return pb->error;
943  }
944  if (eof_forbidden) {
945  av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
946  "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
947  return AVERROR(EIO);
948  }
949  return AVERROR_EOF;
950 }
951 
952 /**
953  * Read a EBML length value.
954  * This needs special handling for the "unknown length" case which has multiple
955  * encodings.
956  */
957 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
958  uint64_t *number)
959 {
960  int res = ebml_read_num(matroska, pb, 8, number, 1);
961  if (res > 0 && *number + 1 == 1ULL << (7 * res))
962  *number = EBML_UNKNOWN_LENGTH;
963  return res;
964 }
965 
966 /*
967  * Read the next element as an unsigned int.
968  * Returns NEEDS_CHECKING unless size == 0.
969  */
970 static int ebml_read_uint(AVIOContext *pb, int size,
971  uint64_t default_value, uint64_t *num)
972 {
973  int n = 0;
974 
975  if (size == 0) {
976  *num = default_value;
977  return 0;
978  }
979  /* big-endian ordering; build up number */
980  *num = 0;
981  while (n++ < size)
982  *num = (*num << 8) | avio_r8(pb);
983 
984  return NEEDS_CHECKING;
985 }
986 
987 /*
988  * Read the next element as a signed int.
989  * Returns NEEDS_CHECKING unless size == 0.
990  */
991 static int ebml_read_sint(AVIOContext *pb, int size,
992  int64_t default_value, int64_t *num)
993 {
994  int n = 1;
995 
996  if (size == 0) {
997  *num = default_value;
998  return 0;
999  } else {
1000  *num = sign_extend(avio_r8(pb), 8);
1001 
1002  /* big-endian ordering; build up number */
1003  while (n++ < size)
1004  *num = ((uint64_t)*num << 8) | avio_r8(pb);
1005  }
1006 
1007  return NEEDS_CHECKING;
1009 
1010 /*
1011  * Read the next element as a float.
1012  * Returns 0 if size == 0, NEEDS_CHECKING or < 0 on obvious failure.
1013  */
1014 static int ebml_read_float(AVIOContext *pb, int size,
1015  double default_value, double *num)
1016 {
1017  if (size == 0) {
1018  *num = default_value;
1019  return 0;
1020  } else if (size == 4) {
1021  *num = av_int2float(avio_rb32(pb));
1022  } else if (size == 8) {
1023  *num = av_int2double(avio_rb64(pb));
1024  } else
1025  return AVERROR_INVALIDDATA;
1026 
1027  return NEEDS_CHECKING;
1029 
1030 /*
1031  * Read the next element as an ASCII string.
1032  * 0 is success, < 0 or NEEDS_CHECKING is failure.
1033  */
1034 static int ebml_read_ascii(AVIOContext *pb, int size,
1035  const char *default_value, char **str)
1036 {
1037  char *res;
1038  int ret;
1039 
1040  if (size == 0 && default_value) {
1041  res = av_strdup(default_value);
1042  if (!res)
1043  return AVERROR(ENOMEM);
1044  } else {
1045  /* EBML strings are usually not 0-terminated, so we allocate one
1046  * byte more, read the string and NUL-terminate it ourselves. */
1047  if (!(res = av_malloc(size + 1)))
1048  return AVERROR(ENOMEM);
1049  if ((ret = avio_read(pb, (uint8_t *) res, size)) != size) {
1050  av_free(res);
1051  return ret < 0 ? ret : NEEDS_CHECKING;
1052  }
1053  (res)[size] = '\0';
1054  }
1055  av_free(*str);
1056  *str = res;
1057 
1058  return 0;
1060 
1061 /*
1062  * Read the next element as binary data.
1063  * 0 is success, < 0 or NEEDS_CHECKING is failure.
1064  */
1065 static int ebml_read_binary(AVIOContext *pb, int length,
1066  int64_t pos, EbmlBin *bin)
1067 {
1068  int ret;
1069 
1071  if (ret < 0)
1072  return ret;
1073  memset(bin->buf->data + length, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1074 
1075  bin->data = bin->buf->data;
1076  bin->size = length;
1077  bin->pos = pos;
1078  if ((ret = avio_read(pb, bin->data, length)) != length) {
1079  av_buffer_unref(&bin->buf);
1080  bin->data = NULL;
1081  bin->size = 0;
1082  return ret < 0 ? ret : NEEDS_CHECKING;
1083  }
1084 
1085  return 0;
1086 }
1088 /*
1089  * Read the next element, but only the header. The contents
1090  * are supposed to be sub-elements which can be read separately.
1091  * 0 is success, < 0 is failure.
1092  */
1093 static int ebml_read_master(MatroskaDemuxContext *matroska,
1094  uint64_t length, int64_t pos)
1095 {
1097 
1098  if (matroska->num_levels >= EBML_MAX_DEPTH) {
1099  av_log(matroska->ctx, AV_LOG_ERROR,
1100  "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
1101  return AVERROR(ENOSYS);
1102  }
1103 
1104  level = &matroska->levels[matroska->num_levels++];
1105  level->start = pos;
1106  level->length = length;
1107 
1108  return 0;
1110 
1111 /*
1112  * Read a signed "EBML number"
1113  * Return: number of bytes processed, < 0 on error
1114  */
1115 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
1116  AVIOContext *pb, int64_t *num)
1117 {
1118  uint64_t unum;
1119  int res;
1120 
1121  /* read as unsigned number first */
1122  if ((res = ebml_read_num(matroska, pb, 8, &unum, 1)) < 0)
1123  return res;
1124 
1125  /* make signed (weird way) */
1126  *num = unum - ((1LL << (7 * res - 1)) - 1);
1127 
1128  return res;
1129 }
1130 
1131 static int ebml_parse(MatroskaDemuxContext *matroska,
1132  EbmlSyntax *syntax, void *data);
1133 
1134 static EbmlSyntax *ebml_parse_id(EbmlSyntax *syntax, uint32_t id)
1135 {
1136  int i;
1137 
1138  // Whoever touches this should be aware of the duplication
1139  // existing in matroska_cluster_parsing.
1140  for (i = 0; syntax[i].id; i++)
1141  if (id == syntax[i].id)
1142  break;
1143 
1144  return &syntax[i];
1145 }
1146 
1147 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
1148  void *data)
1149 {
1150  int res;
1151 
1152  if (data) {
1153  for (int i = 0; syntax[i].id; i++) {
1154  void *dst = (char *)data + syntax[i].data_offset;
1155  switch (syntax[i].type) {
1156  case EBML_UINT:
1157  *(uint64_t *)dst = syntax[i].def.u;
1158  break;
1159  case EBML_SINT:
1160  *(int64_t *) dst = syntax[i].def.i;
1161  break;
1162  case EBML_FLOAT:
1163  *(double *) dst = syntax[i].def.f;
1164  break;
1165  case EBML_STR:
1166  case EBML_UTF8:
1167  // the default may be NULL
1168  if (syntax[i].def.s) {
1169  *(char**)dst = av_strdup(syntax[i].def.s);
1170  if (!*(char**)dst)
1171  return AVERROR(ENOMEM);
1172  }
1173  break;
1174  }
1175  }
1176 
1177  if (!matroska->levels[matroska->num_levels - 1].length) {
1178  matroska->num_levels--;
1179  return 0;
1180  }
1181  }
1182 
1183  do {
1184  res = ebml_parse(matroska, syntax, data);
1185  } while (!res);
1186 
1187  return res == LEVEL_ENDED ? 0 : res;
1188 }
1189 
1190 static int is_ebml_id_valid(uint32_t id)
1191 {
1192  // Due to endian nonsense in Matroska, the highest byte with any bits set
1193  // will contain the leading length bit. This bit in turn identifies the
1194  // total byte length of the element by its position within the byte.
1195  unsigned int bits = av_log2(id);
1196  return id && (bits + 7) / 8 == (8 - bits % 8);
1198 
1199 /*
1200  * Allocate and return the entry for the level1 element with the given ID. If
1201  * an entry already exists, return the existing entry.
1202  */
1204  uint32_t id, int64_t pos)
1205 {
1206  int i;
1207  MatroskaLevel1Element *elem;
1208 
1209  if (!is_ebml_id_valid(id))
1210  return NULL;
1211 
1212  // Some files link to all clusters; useless.
1213  if (id == MATROSKA_ID_CLUSTER)
1214  return NULL;
1215 
1216  // There can be multiple SeekHeads and Tags.
1217  for (i = 0; i < matroska->num_level1_elems; i++) {
1218  if (matroska->level1_elems[i].id == id) {
1219  if (matroska->level1_elems[i].pos == pos ||
1220  id != MATROSKA_ID_SEEKHEAD && id != MATROSKA_ID_TAGS)
1221  return &matroska->level1_elems[i];
1222  }
1223  }
1224 
1225  // Only a completely broken file would have more elements.
1226  if (matroska->num_level1_elems >= FF_ARRAY_ELEMS(matroska->level1_elems)) {
1227  av_log(matroska->ctx, AV_LOG_ERROR, "Too many level1 elements.\n");
1228  return NULL;
1229  }
1230 
1231  elem = &matroska->level1_elems[matroska->num_level1_elems++];
1232  *elem = (MatroskaLevel1Element){.id = id};
1233 
1234  return elem;
1235 }
1236 
1237 static int ebml_parse(MatroskaDemuxContext *matroska,
1238  EbmlSyntax *syntax, void *data)
1239 {
1240  static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
1241  // Forbid unknown-length EBML_NONE elements.
1243  [EBML_UINT] = 8,
1244  [EBML_SINT] = 8,
1245  [EBML_FLOAT] = 8,
1246  // max. 16 MB for strings
1247  [EBML_STR] = 0x1000000,
1248  [EBML_UTF8] = 0x1000000,
1249  // max. 256 MB for binary data
1250  [EBML_BIN] = 0x10000000,
1251  // no limits for anything else
1252  };
1253  AVIOContext *pb = matroska->ctx->pb;
1254  uint32_t id;
1255  uint64_t length;
1256  int64_t pos = avio_tell(pb), pos_alt;
1257  int res, update_pos = 1, level_check;
1258  MatroskaLevel1Element *level1_elem;
1259  MatroskaLevel *level = matroska->num_levels ? &matroska->levels[matroska->num_levels - 1] : NULL;
1260 
1261  if (!matroska->current_id) {
1262  uint64_t id;
1263  res = ebml_read_num(matroska, pb, 4, &id, 0);
1264  if (res < 0) {
1265  if (pb->eof_reached && res == AVERROR_EOF) {
1266  if (matroska->is_live)
1267  // in live mode, finish parsing if EOF is reached.
1268  return 1;
1269  if (level && pos == avio_tell(pb)) {
1270  if (level->length == EBML_UNKNOWN_LENGTH) {
1271  // Unknown-length levels automatically end at EOF.
1272  matroska->num_levels--;
1273  return LEVEL_ENDED;
1274  } else {
1275  av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely "
1276  "at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
1277  }
1278  }
1279  }
1280  return res;
1281  }
1282  matroska->current_id = id | 1 << 7 * res;
1283  pos_alt = pos + res;
1284  } else {
1285  pos_alt = pos;
1286  pos -= (av_log2(matroska->current_id) + 7) / 8;
1287  }
1288 
1289  id = matroska->current_id;
1290 
1291  syntax = ebml_parse_id(syntax, id);
1292  if (!syntax->id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
1293  if (level && level->length == EBML_UNKNOWN_LENGTH) {
1294  // Unknown-length levels end when an element from an upper level
1295  // in the hierarchy is encountered.
1296  while (syntax->def.n) {
1297  syntax = ebml_parse_id(syntax->def.n, id);
1298  if (syntax->id) {
1299  matroska->num_levels--;
1300  return LEVEL_ENDED;
1301  }
1302  };
1303  }
1304 
1305  av_log(matroska->ctx, AV_LOG_DEBUG, "Unknown entry 0x%"PRIX32" at pos. "
1306  "%"PRId64"\n", id, pos);
1307  update_pos = 0; /* Don't update resync_pos as an error might have happened. */
1308  }
1309 
1310  if (data) {
1311  data = (char *) data + syntax->data_offset;
1312  if (syntax->list_elem_size) {
1313  EbmlList *list = data;
1314  void *newelem;
1315 
1316  if ((unsigned)list->nb_elem + 1 >= UINT_MAX / syntax->list_elem_size)
1317  return AVERROR(ENOMEM);
1318  newelem = av_fast_realloc(list->elem,
1319  &list->alloc_elem_size,
1320  (list->nb_elem + 1) * syntax->list_elem_size);
1321  if (!newelem)
1322  return AVERROR(ENOMEM);
1323  list->elem = newelem;
1324  data = (char *) list->elem + list->nb_elem * syntax->list_elem_size;
1325  memset(data, 0, syntax->list_elem_size);
1326  list->nb_elem++;
1327  }
1328  }
1329 
1330  if (syntax->type != EBML_STOP) {
1331  matroska->current_id = 0;
1332  if ((res = ebml_read_length(matroska, pb, &length)) < 0)
1333  return res;
1334 
1335  pos_alt += res;
1336 
1337  if (matroska->num_levels > 0) {
1338  if (length != EBML_UNKNOWN_LENGTH &&
1339  level->length != EBML_UNKNOWN_LENGTH) {
1340  uint64_t elem_end = pos_alt + length,
1341  level_end = level->start + level->length;
1342 
1343  if (elem_end < level_end) {
1344  level_check = 0;
1345  } else if (elem_end == level_end) {
1346  level_check = LEVEL_ENDED;
1347  } else {
1348  av_log(matroska->ctx, AV_LOG_ERROR,
1349  "Element at 0x%"PRIx64" ending at 0x%"PRIx64" exceeds "
1350  "containing master element ending at 0x%"PRIx64"\n",
1351  pos, elem_end, level_end);
1352  return AVERROR_INVALIDDATA;
1353  }
1354  } else if (length != EBML_UNKNOWN_LENGTH) {
1355  level_check = 0;
1356  } else if (level->length != EBML_UNKNOWN_LENGTH) {
1357  av_log(matroska->ctx, AV_LOG_ERROR, "Unknown-sized element "
1358  "at 0x%"PRIx64" inside parent with finite size\n", pos);
1359  return AVERROR_INVALIDDATA;
1360  } else {
1361  level_check = 0;
1362  if (id != MATROSKA_ID_CLUSTER && (syntax->type == EBML_LEVEL1
1363  || syntax->type == EBML_NEST)) {
1364  // According to the current specifications only clusters and
1365  // segments are allowed to be unknown-length. We also accept
1366  // other unknown-length master elements.
1367  av_log(matroska->ctx, AV_LOG_WARNING,
1368  "Found unknown-length element 0x%"PRIX32" other than "
1369  "a cluster at 0x%"PRIx64". Spec-incompliant, but "
1370  "parsing will nevertheless be attempted.\n", id, pos);
1371  update_pos = -1;
1372  }
1373  }
1374  } else
1375  level_check = 0;
1376 
1377  if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
1378  if (length != EBML_UNKNOWN_LENGTH) {
1379  av_log(matroska->ctx, AV_LOG_ERROR,
1380  "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for element "
1381  "with ID 0x%"PRIX32" at 0x%"PRIx64"\n",
1382  length, max_lengths[syntax->type], id, pos);
1383  } else if (syntax->type != EBML_NONE) {
1384  av_log(matroska->ctx, AV_LOG_ERROR,
1385  "Element with ID 0x%"PRIX32" at pos. 0x%"PRIx64" has "
1386  "unknown length, yet the length of an element of its "
1387  "type must be known.\n", id, pos);
1388  } else {
1389  av_log(matroska->ctx, AV_LOG_ERROR,
1390  "Found unknown-length element with ID 0x%"PRIX32" at "
1391  "pos. 0x%"PRIx64" for which no syntax for parsing is "
1392  "available.\n", id, pos);
1393  }
1394  return AVERROR_INVALIDDATA;
1395  }
1396 
1397  if (!(pb->seekable & AVIO_SEEKABLE_NORMAL)) {
1398  // Loosing sync will likely manifest itself as encountering unknown
1399  // elements which are not reliably distinguishable from elements
1400  // belonging to future extensions of the format.
1401  // We use a heuristic to detect such situations: If the current
1402  // element is not expected at the current syntax level and there
1403  // were only a few unknown elements in a row, then the element is
1404  // skipped or considered defective based upon the length of the
1405  // current element (i.e. how much would be skipped); if there were
1406  // more than a few skipped elements in a row and skipping the current
1407  // element would lead us more than SKIP_THRESHOLD away from the last
1408  // known good position, then it is inferred that an error occurred.
1409  // The dependency on the number of unknown elements in a row exists
1410  // because the distance to the last known good position is
1411  // automatically big if the last parsed element was big.
1412  // In both cases, each unknown element is considered equivalent to
1413  // UNKNOWN_EQUIV of skipped bytes for the check.
1414  // The whole check is only done for non-seekable output, because
1415  // in this situation skipped data can't simply be rechecked later.
1416  // This is especially important when using unkown length elements
1417  // as the check for whether a child exceeds its containing master
1418  // element is not effective in this situation.
1419  if (update_pos) {
1420  matroska->unknown_count = 0;
1421  } else {
1422  int64_t dist = length + UNKNOWN_EQUIV * matroska->unknown_count++;
1423 
1424  if (matroska->unknown_count > 3)
1425  dist += pos_alt - matroska->resync_pos;
1426 
1427  if (dist > SKIP_THRESHOLD) {
1428  av_log(matroska->ctx, AV_LOG_ERROR,
1429  "Unknown element %"PRIX32" at pos. 0x%"PRIx64" with "
1430  "length 0x%"PRIx64" considered as invalid data. Last "
1431  "known good position 0x%"PRIx64", %d unknown elements"
1432  " in a row\n", id, pos, length, matroska->resync_pos,
1433  matroska->unknown_count);
1434  return AVERROR_INVALIDDATA;
1435  }
1436  }
1437  }
1438 
1439  if (update_pos > 0) {
1440  // We have found an element that is allowed at this place
1441  // in the hierarchy and it passed all checks, so treat the beginning
1442  // of the element as the "last known good" position.
1443  matroska->resync_pos = pos;
1444  }
1445 
1446  if (!data && length != EBML_UNKNOWN_LENGTH)
1447  goto skip;
1448  }
1449 
1450  switch (syntax->type) {
1451  case EBML_UINT:
1452  res = ebml_read_uint(pb, length, syntax->def.u, data);
1453  break;
1454  case EBML_SINT:
1455  res = ebml_read_sint(pb, length, syntax->def.i, data);
1456  break;
1457  case EBML_FLOAT:
1458  res = ebml_read_float(pb, length, syntax->def.f, data);
1459  break;
1460  case EBML_STR:
1461  case EBML_UTF8:
1462  res = ebml_read_ascii(pb, length, syntax->def.s, data);
1463  break;
1464  case EBML_BIN:
1465  res = ebml_read_binary(pb, length, pos_alt, data);
1466  break;
1467  case EBML_LEVEL1:
1468  case EBML_NEST:
1469  if ((res = ebml_read_master(matroska, length, pos_alt)) < 0)
1470  return res;
1471  if (id == MATROSKA_ID_SEGMENT)
1472  matroska->segment_start = pos_alt;
1473  if (id == MATROSKA_ID_CUES)
1474  matroska->cues_parsing_deferred = 0;
1475  if (syntax->type == EBML_LEVEL1 &&
1476  (level1_elem = matroska_find_level1_elem(matroska, syntax->id, pos))) {
1477  if (!level1_elem->pos) {
1478  // Zero is not a valid position for a level 1 element.
1479  level1_elem->pos = pos;
1480  } else if (level1_elem->pos != pos)
1481  av_log(matroska->ctx, AV_LOG_ERROR, "Duplicate element\n");
1482  level1_elem->parsed = 1;
1483  }
1484  if (res = ebml_parse_nest(matroska, syntax->def.n, data))
1485  return res;
1486  break;
1487  case EBML_STOP:
1488  return 1;
1489  skip:
1490  default:
1491  if (length) {
1492  int64_t res2;
1493  if (ffio_limit(pb, length) != length) {
1494  // ffio_limit emits its own error message,
1495  // so we don't have to.
1496  return AVERROR(EIO);
1497  }
1498  if ((res2 = avio_skip(pb, length - 1)) >= 0) {
1499  // avio_skip might take us past EOF. We check for this
1500  // by skipping only length - 1 bytes, reading a byte and
1501  // checking the error flags. This is done in order to check
1502  // that the element has been properly skipped even when
1503  // no filesize (that ffio_limit relies on) is available.
1504  avio_r8(pb);
1505  res = NEEDS_CHECKING;
1506  } else
1507  res = res2;
1508  } else
1509  res = 0;
1510  }
1511  if (res) {
1512  if (res == NEEDS_CHECKING) {
1513  if (pb->eof_reached) {
1514  if (pb->error)
1515  res = pb->error;
1516  else
1517  res = AVERROR_EOF;
1518  } else
1519  goto level_check;
1520  }
1521 
1522  if (res == AVERROR_INVALIDDATA)
1523  av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
1524  else if (res == AVERROR(EIO))
1525  av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
1526  else if (res == AVERROR_EOF) {
1527  av_log(matroska->ctx, AV_LOG_ERROR, "File ended prematurely\n");
1528  res = AVERROR(EIO);
1529  }
1530 
1531  return res;
1532  }
1533 
1534 level_check:
1535  if (syntax->is_counted && data) {
1536  CountedElement *elem = data;
1537  if (elem->count != UINT_MAX)
1538  elem->count++;
1539  }
1540 
1541  if (level_check == LEVEL_ENDED && matroska->num_levels) {
1542  level = &matroska->levels[matroska->num_levels - 1];
1543  pos = avio_tell(pb);
1544 
1545  // Given that pos >= level->start no check for
1546  // level->length != EBML_UNKNOWN_LENGTH is necessary.
1547  while (matroska->num_levels && pos == level->start + level->length) {
1548  matroska->num_levels--;
1549  level--;
1550  }
1551  }
1552 
1553  return level_check;
1554 }
1555 
1556 static void ebml_free(EbmlSyntax *syntax, void *data)
1557 {
1558  int i, j;
1559  for (i = 0; syntax[i].id; i++) {
1560  void *data_off = (char *) data + syntax[i].data_offset;
1561  switch (syntax[i].type) {
1562  case EBML_STR:
1563  case EBML_UTF8:
1564  av_freep(data_off);
1565  break;
1566  case EBML_BIN:
1567  av_buffer_unref(&((EbmlBin *) data_off)->buf);
1568  break;
1569  case EBML_LEVEL1:
1570  case EBML_NEST:
1571  if (syntax[i].list_elem_size) {
1572  EbmlList *list = data_off;
1573  char *ptr = list->elem;
1574  for (j = 0; j < list->nb_elem;
1575  j++, ptr += syntax[i].list_elem_size)
1576  ebml_free(syntax[i].def.n, ptr);
1577  av_freep(&list->elem);
1578  list->nb_elem = 0;
1579  list->alloc_elem_size = 0;
1580  } else
1581  ebml_free(syntax[i].def.n, data_off);
1582  default:
1583  break;
1584  }
1585  }
1586 }
1587 
1588 /*
1589  * Autodetecting...
1590  */
1591 static int matroska_probe(const AVProbeData *p)
1592 {
1593  uint64_t total = 0;
1594  int len_mask = 0x80, size = 1, n = 1, i;
1595 
1596  /* EBML header? */
1597  if (AV_RB32(p->buf) != EBML_ID_HEADER)
1598  return 0;
1599 
1600  /* length of header */
1601  total = p->buf[4];
1602  while (size <= 8 && !(total & len_mask)) {
1603  size++;
1604  len_mask >>= 1;
1605  }
1606  if (size > 8)
1607  return 0;
1608  total &= (len_mask - 1);
1609  while (n < size)
1610  total = (total << 8) | p->buf[4 + n++];
1611 
1612  if (total + 1 == 1ULL << (7 * size)){
1613  /* Unknown-length header - simply parse the whole buffer. */
1614  total = p->buf_size - 4 - size;
1615  } else {
1616  /* Does the probe data contain the whole header? */
1617  if (p->buf_size < 4 + size + total)
1618  return 0;
1619  }
1620 
1621  /* The header should contain a known document type. For now,
1622  * we don't parse the whole header but simply check for the
1623  * availability of that array of characters inside the header.
1624  * Not fully fool-proof, but good enough. */
1625  for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1626  size_t probelen = strlen(matroska_doctypes[i]);
1627  if (total < probelen)
1628  continue;
1629  for (n = 4 + size; n <= 4 + size + total - probelen; n++)
1630  if (!memcmp(p->buf + n, matroska_doctypes[i], probelen))
1631  return AVPROBE_SCORE_MAX;
1632  }
1633 
1634  // probably valid EBML header but no recognized doctype
1635  return AVPROBE_SCORE_EXTENSION;
1636 }
1637 
1639  uint64_t num)
1640 {
1641  MatroskaTrack *tracks = matroska->tracks.elem;
1642  int i;
1643 
1644  for (i = 0; i < matroska->tracks.nb_elem; i++)
1645  if (tracks[i].num == num)
1646  return &tracks[i];
1647 
1648  av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %"PRIu64"\n", num);
1649  return NULL;
1650 }
1651 
1652 static int matroska_decode_buffer(uint8_t **buf, int *buf_size,
1653  MatroskaTrack *track)
1654 {
1655  MatroskaTrackEncoding *encodings = track->encodings.elem;
1656  uint8_t *data = *buf;
1657  int isize = *buf_size;
1658  uint8_t *pkt_data = NULL;
1659  uint8_t av_unused *newpktdata;
1660  int pkt_size = isize;
1661  int result = 0;
1662  int olen;
1663 
1664  if (pkt_size >= 10000000U)
1665  return AVERROR_INVALIDDATA;
1666 
1667  switch (encodings[0].compression.algo) {
1669  {
1670  int header_size = encodings[0].compression.settings.size;
1671  uint8_t *header = encodings[0].compression.settings.data;
1672 
1673  if (header_size && !header) {
1674  av_log(NULL, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1675  return -1;
1676  }
1677 
1678  if (!header_size)
1679  return 0;
1680 
1681  pkt_size = isize + header_size;
1682  pkt_data = av_malloc(pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1683  if (!pkt_data)
1684  return AVERROR(ENOMEM);
1685 
1686  memcpy(pkt_data, header, header_size);
1687  memcpy(pkt_data + header_size, data, isize);
1688  break;
1689  }
1691  do {
1692  int insize = isize;
1693  olen = pkt_size *= 3;
1694  newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING
1696  if (!newpktdata) {
1697  result = AVERROR(ENOMEM);
1698  goto failed;
1699  }
1700  pkt_data = newpktdata;
1701  result = av_lzo1x_decode(pkt_data, &olen, data, &insize);
1702  } while (result == AV_LZO_OUTPUT_FULL && pkt_size < 10000000);
1703  if (result) {
1705  goto failed;
1706  }
1707  pkt_size -= olen;
1708  break;
1709 #if CONFIG_ZLIB
1711  {
1712  z_stream zstream = { 0 };
1713  if (!pkt_size || inflateInit(&zstream) != Z_OK)
1714  return -1;
1715  zstream.next_in = data;
1716  zstream.avail_in = isize;
1717  do {
1718  pkt_size *= 3;
1719  newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1720  if (!newpktdata) {
1721  inflateEnd(&zstream);
1722  result = AVERROR(ENOMEM);
1723  goto failed;
1724  }
1725  pkt_data = newpktdata;
1726  zstream.avail_out = pkt_size - zstream.total_out;
1727  zstream.next_out = pkt_data + zstream.total_out;
1728  result = inflate(&zstream, Z_NO_FLUSH);
1729  } while (result == Z_OK && pkt_size < 10000000);
1730  pkt_size = zstream.total_out;
1731  inflateEnd(&zstream);
1732  if (result != Z_STREAM_END) {
1733  if (result == Z_MEM_ERROR)
1734  result = AVERROR(ENOMEM);
1735  else
1737  goto failed;
1738  }
1739  break;
1740  }
1741 #endif
1742 #if CONFIG_BZLIB
1744  {
1745  bz_stream bzstream = { 0 };
1746  if (!pkt_size || BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1747  return -1;
1748  bzstream.next_in = data;
1749  bzstream.avail_in = isize;
1750  do {
1751  pkt_size *= 3;
1752  newpktdata = av_realloc(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE);
1753  if (!newpktdata) {
1754  BZ2_bzDecompressEnd(&bzstream);
1755  result = AVERROR(ENOMEM);
1756  goto failed;
1757  }
1758  pkt_data = newpktdata;
1759  bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1760  bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1761  result = BZ2_bzDecompress(&bzstream);
1762  } while (result == BZ_OK && pkt_size < 10000000);
1763  pkt_size = bzstream.total_out_lo32;
1764  BZ2_bzDecompressEnd(&bzstream);
1765  if (result != BZ_STREAM_END) {
1766  if (result == BZ_MEM_ERROR)
1767  result = AVERROR(ENOMEM);
1768  else
1770  goto failed;
1771  }
1772  break;
1773  }
1774 #endif
1775  default:
1776  return AVERROR_INVALIDDATA;
1777  }
1778 
1779  memset(pkt_data + pkt_size, 0, AV_INPUT_BUFFER_PADDING_SIZE);
1780 
1781  *buf = pkt_data;
1782  *buf_size = pkt_size;
1783  return 0;
1785 failed:
1786  av_free(pkt_data);
1787  return result;
1788 }
1789 
1791  AVDictionary **metadata, char *prefix)
1792 {
1793  MatroskaTag *tags = list->elem;
1794  char key[1024];
1795  int i;
1796 
1797  for (i = 0; i < list->nb_elem; i++) {
1798  const char *lang = tags[i].lang &&
1799  strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1800 
1801  if (!tags[i].name) {
1802  av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1803  continue;
1804  }
1805  if (prefix)
1806  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1807  else
1808  av_strlcpy(key, tags[i].name, sizeof(key));
1809  if (tags[i].def || !lang) {
1810  av_dict_set(metadata, key, tags[i].string, 0);
1811  if (tags[i].sub.nb_elem)
1812  matroska_convert_tag(s, &tags[i].sub, metadata, key);
1813  }
1814  if (lang) {
1815  av_strlcat(key, "-", sizeof(key));
1816  av_strlcat(key, lang, sizeof(key));
1817  av_dict_set(metadata, key, tags[i].string, 0);
1818  if (tags[i].sub.nb_elem)
1819  matroska_convert_tag(s, &tags[i].sub, metadata, key);
1820  }
1821  }
1823 }
1824 
1826 {
1827  MatroskaDemuxContext *matroska = s->priv_data;
1828  MatroskaTags *tags = matroska->tags.elem;
1829  int i, j;
1830 
1831  for (i = 0; i < matroska->tags.nb_elem; i++) {
1832  if (tags[i].target.attachuid) {
1833  MatroskaAttachment *attachment = matroska->attachments.elem;
1834  int found = 0;
1835  for (j = 0; j < matroska->attachments.nb_elem; j++) {
1836  if (attachment[j].uid == tags[i].target.attachuid &&
1837  attachment[j].stream) {
1838  matroska_convert_tag(s, &tags[i].tag,
1839  &attachment[j].stream->metadata, NULL);
1840  found = 1;
1841  }
1842  }
1843  if (!found) {
1845  "The tags at index %d refer to a "
1846  "non-existent attachment %"PRId64".\n",
1847  i, tags[i].target.attachuid);
1848  }
1849  } else if (tags[i].target.chapteruid) {
1850  MatroskaChapter *chapter = matroska->chapters.elem;
1851  int found = 0;
1852  for (j = 0; j < matroska->chapters.nb_elem; j++) {
1853  if (chapter[j].uid == tags[i].target.chapteruid &&
1854  chapter[j].chapter) {
1855  matroska_convert_tag(s, &tags[i].tag,
1856  &chapter[j].chapter->metadata, NULL);
1857  found = 1;
1858  }
1859  }
1860  if (!found) {
1862  "The tags at index %d refer to a non-existent chapter "
1863  "%"PRId64".\n",
1864  i, tags[i].target.chapteruid);
1865  }
1866  } else if (tags[i].target.trackuid) {
1867  MatroskaTrack *track = matroska->tracks.elem;
1868  int found = 0;
1869  for (j = 0; j < matroska->tracks.nb_elem; j++) {
1870  if (track[j].uid == tags[i].target.trackuid &&
1871  track[j].stream) {
1872  matroska_convert_tag(s, &tags[i].tag,
1873  &track[j].stream->metadata, NULL);
1874  found = 1;
1875  }
1876  }
1877  if (!found) {
1879  "The tags at index %d refer to a non-existent track "
1880  "%"PRId64".\n",
1881  i, tags[i].target.trackuid);
1882  }
1883  } else {
1884  matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1885  tags[i].target.type);
1886  }
1887  }
1888 }
1889 
1891  int64_t pos)
1892 {
1893  uint32_t saved_id = matroska->current_id;
1894  int64_t before_pos = avio_tell(matroska->ctx->pb);
1895  int ret = 0;
1896  int ret2;
1897 
1898  /* seek */
1899  if (avio_seek(matroska->ctx->pb, pos, SEEK_SET) == pos) {
1900  /* We don't want to lose our seekhead level, so we add
1901  * a dummy. This is a crude hack. */
1902  if (matroska->num_levels == EBML_MAX_DEPTH) {
1903  av_log(matroska->ctx, AV_LOG_INFO,
1904  "Max EBML element depth (%d) reached, "
1905  "cannot parse further.\n", EBML_MAX_DEPTH);
1907  } else {
1908  matroska->levels[matroska->num_levels] = (MatroskaLevel) { 0, EBML_UNKNOWN_LENGTH };
1909  matroska->num_levels++;
1910  matroska->current_id = 0;
1911 
1912  ret = ebml_parse(matroska, matroska_segment, matroska);
1913  if (ret == LEVEL_ENDED) {
1914  /* This can only happen if the seek brought us beyond EOF. */
1915  ret = AVERROR_EOF;
1916  }
1917  }
1918  }
1919  /* Seek back - notice that in all instances where this is used
1920  * it is safe to set the level to 1. */
1921  ret2 = matroska_reset_status(matroska, saved_id, before_pos);
1922  if (ret >= 0)
1923  ret = ret2;
1924 
1925  return ret;
1926 }
1927 
1928 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1929 {
1930  EbmlList *seekhead_list = &matroska->seekhead;
1931  int i;
1932 
1933  // we should not do any seeking in the streaming case
1934  if (!(matroska->ctx->pb->seekable & AVIO_SEEKABLE_NORMAL))
1935  return;
1936 
1937  for (i = 0; i < seekhead_list->nb_elem; i++) {
1938  MatroskaSeekhead *seekheads = seekhead_list->elem;
1939  uint32_t id = seekheads[i].id;
1940  int64_t pos = seekheads[i].pos + matroska->segment_start;
1941  MatroskaLevel1Element *elem;
1942 
1943  if (id != seekheads[i].id || pos < matroska->segment_start)
1944  continue;
1945 
1946  elem = matroska_find_level1_elem(matroska, id, pos);
1947  if (!elem || elem->parsed)
1948  continue;
1949 
1950  elem->pos = pos;
1951 
1952  // defer cues parsing until we actually need cue data.
1953  if (id == MATROSKA_ID_CUES)
1954  continue;
1955 
1956  if (matroska_parse_seekhead_entry(matroska, pos) < 0) {
1957  // mark index as broken
1958  matroska->cues_parsing_deferred = -1;
1959  break;
1960  }
1961 
1962  elem->parsed = 1;
1963  }
1964 }
1965 
1966 static void matroska_add_index_entries(MatroskaDemuxContext *matroska)
1967 {
1968  EbmlList *index_list;
1970  uint64_t index_scale = 1;
1971  int i, j;
1972 
1973  if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
1974  return;
1975 
1976  index_list = &matroska->index;
1977  index = index_list->elem;
1978  if (index_list->nb_elem < 2)
1979  return;
1980  if (index[1].time > 1E14 / matroska->time_scale) {
1981  av_log(matroska->ctx, AV_LOG_WARNING, "Dropping apparently-broken index.\n");
1982  return;
1983  }
1984  for (i = 0; i < index_list->nb_elem; i++) {
1985  EbmlList *pos_list = &index[i].pos;
1986  MatroskaIndexPos *pos = pos_list->elem;
1987  for (j = 0; j < pos_list->nb_elem; j++) {
1988  MatroskaTrack *track = matroska_find_track_by_num(matroska,
1989  pos[j].track);
1990  if (track && track->stream)
1991  av_add_index_entry(track->stream,
1992  pos[j].pos + matroska->segment_start,
1993  index[i].time / index_scale, 0, 0,
1995  }
1996  }
1997 }
1998 
1999 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
2000  int i;
2001 
2002  if (matroska->ctx->flags & AVFMT_FLAG_IGNIDX)
2003  return;
2004 
2005  for (i = 0; i < matroska->num_level1_elems; i++) {
2006  MatroskaLevel1Element *elem = &matroska->level1_elems[i];
2007  if (elem->id == MATROSKA_ID_CUES && !elem->parsed) {
2008  if (matroska_parse_seekhead_entry(matroska, elem->pos) < 0)
2009  matroska->cues_parsing_deferred = -1;
2010  elem->parsed = 1;
2011  break;
2012  }
2013  }
2014 
2015  matroska_add_index_entries(matroska);
2016 }
2017 
2018 static int matroska_aac_profile(char *codec_id)
2019 {
2020  static const char *const aac_profiles[] = { "MAIN", "LC", "SSR" };
2021  int profile;
2022 
2024  if (strstr(codec_id, aac_profiles[profile]))
2025  break;
2026  return profile + 1;
2027 }
2028 
2029 static int matroska_aac_sri(int samplerate)
2030 {
2031  int sri;
2032 
2033  for (sri = 0; sri < FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
2034  if (ff_mpeg4audio_sample_rates[sri] == samplerate)
2035  break;
2036  return sri;
2037 }
2038 
2039 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
2040 {
2041  /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
2042  avpriv_dict_set_timestamp(metadata, "creation_time", date_utc / 1000 + 978307200000000LL);
2043 }
2044 
2046  MatroskaTrack *track,
2047  int *offset)
2048 {
2049  AVStream *st = track->stream;
2050  uint8_t *p = track->codec_priv.data;
2051  int size = track->codec_priv.size;
2052 
2053  if (size < 8 + FLAC_STREAMINFO_SIZE || p[4] & 0x7f) {
2054  av_log(s, AV_LOG_WARNING, "Invalid FLAC private data\n");
2055  track->codec_priv.size = 0;
2056  return 0;
2057  }
2058  *offset = 8;
2059  track->codec_priv.size = 8 + FLAC_STREAMINFO_SIZE;
2060 
2061  p += track->codec_priv.size;
2062  size -= track->codec_priv.size;
2063 
2064  /* parse the remaining metadata blocks if present */
2065  while (size >= 4) {
2066  int block_last, block_type, block_size;
2067 
2068  flac_parse_block_header(p, &block_last, &block_type, &block_size);
2069 
2070  p += 4;
2071  size -= 4;
2072  if (block_size > size)
2073  return 0;
2074 
2075  /* check for the channel mask */
2076  if (block_type == FLAC_METADATA_TYPE_VORBIS_COMMENT) {
2077  AVDictionary *dict = NULL;
2078  AVDictionaryEntry *chmask;
2079 
2080  ff_vorbis_comment(s, &dict, p, block_size, 0);
2081  chmask = av_dict_get(dict, "WAVEFORMATEXTENSIBLE_CHANNEL_MASK", NULL, 0);
2082  if (chmask) {
2083  uint64_t mask = strtol(chmask->value, NULL, 0);
2084  if (!mask || mask & ~0x3ffffULL) {
2086  "Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK\n");
2087  } else
2089  }
2090  av_dict_free(&dict);
2091  }
2092 
2093  p += block_size;
2094  size -= block_size;
2095  }
2096 
2097  return 0;
2098 }
2099 
2100 static int mkv_field_order(MatroskaDemuxContext *matroska, uint64_t field_order)
2101 {
2102  int minor, micro, bttb = 0;
2103 
2104  /* workaround a bug in our Matroska muxer, introduced in version 57.36 alongside
2105  * this function, and fixed in 57.52 */
2106  if (matroska->muxingapp && sscanf(matroska->muxingapp, "Lavf57.%d.%d", &minor, &micro) == 2)
2107  bttb = (minor >= 36 && minor <= 51 && micro >= 100);
2108 
2109  switch (field_order) {
2111  return AV_FIELD_PROGRESSIVE;
2113  return AV_FIELD_UNKNOWN;
2115  return AV_FIELD_TT;
2117  return AV_FIELD_BB;
2119  return bttb ? AV_FIELD_TB : AV_FIELD_BT;
2121  return bttb ? AV_FIELD_BT : AV_FIELD_TB;
2122  default:
2123  return AV_FIELD_UNKNOWN;
2124  }
2125 }
2126 
2127 static void mkv_stereo_mode_display_mul(int stereo_mode,
2128  int *h_width, int *h_height)
2129 {
2130  switch (stereo_mode) {
2136  break;
2141  *h_width = 2;
2142  break;
2147  *h_height = 2;
2148  break;
2149  }
2150 }
2151 
2152 static int mkv_parse_video_color(AVStream *st, const MatroskaTrack *track) {
2153  const MatroskaTrackVideoColor *color = track->video.color.elem;
2154  const MatroskaMasteringMeta *mastering_meta;
2155  int has_mastering_primaries, has_mastering_luminance;
2156 
2157  if (!track->video.color.nb_elem)
2158  return 0;
2159 
2160  mastering_meta = &color->mastering_meta;
2161  // Mastering primaries are CIE 1931 coords, and must be > 0.
2162  has_mastering_primaries =
2163  mastering_meta->r_x > 0 && mastering_meta->r_y > 0 &&
2164  mastering_meta->g_x > 0 && mastering_meta->g_y > 0 &&
2165  mastering_meta->b_x > 0 && mastering_meta->b_y > 0 &&
2166  mastering_meta->white_x > 0 && mastering_meta->white_y > 0;
2167  has_mastering_luminance = mastering_meta->max_luminance >
2168  mastering_meta->min_luminance.el.f &&
2169  mastering_meta->min_luminance.el.f >= 0 &&
2170  mastering_meta->min_luminance.count;
2171 
2172  if (color->matrix_coefficients != AVCOL_SPC_RESERVED)
2173  st->codecpar->color_space = color->matrix_coefficients;
2174  if (color->primaries != AVCOL_PRI_RESERVED &&
2175  color->primaries != AVCOL_PRI_RESERVED0)
2176  st->codecpar->color_primaries = color->primaries;
2177  if (color->transfer_characteristics != AVCOL_TRC_RESERVED &&
2178  color->transfer_characteristics != AVCOL_TRC_RESERVED0)
2179  st->codecpar->color_trc = color->transfer_characteristics;
2180  if (color->range != AVCOL_RANGE_UNSPECIFIED &&
2181  color->range <= AVCOL_RANGE_JPEG)
2182  st->codecpar->color_range = color->range;
2183  if (color->chroma_siting_horz != MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED &&
2184  color->chroma_siting_vert != MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED &&
2185  color->chroma_siting_horz < MATROSKA_COLOUR_CHROMASITINGHORZ_NB &&
2186  color->chroma_siting_vert < MATROSKA_COLOUR_CHROMASITINGVERT_NB) {
2187  st->codecpar->chroma_location =
2188  av_chroma_location_pos_to_enum((color->chroma_siting_horz - 1) << 7,
2189  (color->chroma_siting_vert - 1) << 7);
2190  }
2191  if (color->max_cll && color->max_fall) {
2192  size_t size = 0;
2193  int ret;
2195  if (!metadata)
2196  return AVERROR(ENOMEM);
2198  (uint8_t *)metadata, size);
2199  if (ret < 0) {
2200  av_freep(&metadata);
2201  return ret;
2202  }
2203  metadata->MaxCLL = color->max_cll;
2204  metadata->MaxFALL = color->max_fall;
2205  }
2206 
2207  if (has_mastering_primaries || has_mastering_luminance) {
2208  AVMasteringDisplayMetadata *metadata =
2211  sizeof(AVMasteringDisplayMetadata));
2212  if (!metadata) {
2213  return AVERROR(ENOMEM);
2214  }
2215  memset(metadata, 0, sizeof(AVMasteringDisplayMetadata));
2216  if (has_mastering_primaries) {
2217  metadata->display_primaries[0][0] = av_d2q(mastering_meta->r_x, INT_MAX);
2218  metadata->display_primaries[0][1] = av_d2q(mastering_meta->r_y, INT_MAX);
2219  metadata->display_primaries[1][0] = av_d2q(mastering_meta->g_x, INT_MAX);
2220  metadata->display_primaries[1][1] = av_d2q(mastering_meta->g_y, INT_MAX);
2221  metadata->display_primaries[2][0] = av_d2q(mastering_meta->b_x, INT_MAX);
2222  metadata->display_primaries[2][1] = av_d2q(mastering_meta->b_y, INT_MAX);
2223  metadata->white_point[0] = av_d2q(mastering_meta->white_x, INT_MAX);
2224  metadata->white_point[1] = av_d2q(mastering_meta->white_y, INT_MAX);
2225  metadata->has_primaries = 1;
2226  }
2227  if (has_mastering_luminance) {
2228  metadata->max_luminance = av_d2q(mastering_meta->max_luminance, INT_MAX);
2229  metadata->min_luminance = av_d2q(mastering_meta->min_luminance.el.f, INT_MAX);
2230  metadata->has_luminance = 1;
2231  }
2232  }
2233  return 0;
2234 }
2235 
2236 static int mkv_create_display_matrix(AVStream *st,
2237  const MatroskaTrackVideoProjection *proj,
2238  void *logctx)
2239 {
2240  double pitch = proj->pitch, yaw = proj->yaw, roll = proj->roll;
2241  int32_t *matrix;
2242  int hflip;
2243 
2244  if (pitch == 0.0 && yaw == 0.0 && roll == 0.0)
2245  return 0;
2246 
2247  /* Note: The following constants are exactly representable
2248  * as floating-point numbers. */
2249  if (pitch != 0.0 || (yaw != 0.0 && yaw != 180.0 && yaw != -180.0) ||
2250  isnan(roll)) {
2251  av_log(logctx, AV_LOG_WARNING, "Ignoring non-2D rectangular "
2252  "projection in stream %u (yaw %f, pitch %f, roll %f)\n",
2253  st->index, yaw, pitch, roll);
2254  return 0;
2255  }
2257  9 * sizeof(*matrix));
2258  if (!matrix)
2259  return AVERROR(ENOMEM);
2260 
2261  hflip = yaw != 0.0;
2262  /* ProjectionPoseRoll is in the counter-clockwise direction
2263  * whereas av_display_rotation_set() expects its argument
2264  * to be oriented clockwise, so we need to negate roll.
2265  * Furthermore, if hflip is set, we need to negate it again
2266  * to account for the fact that the Matroska specifications
2267  * require the yaw rotation to be applied first. */
2268  av_display_rotation_set(matrix, roll * (2 * hflip - 1));
2269  av_display_matrix_flip(matrix, hflip, 0);
2270 
2271  return 0;
2272 }
2273 
2274 static int mkv_parse_video_projection(AVStream *st, const MatroskaTrack *track,
2275  void *logctx)
2276 {
2277  AVSphericalMapping *spherical;
2278  const MatroskaTrackVideoProjection *mkv_projection = &track->video.projection;
2279  const uint8_t *priv_data = mkv_projection->private.data;
2280  enum AVSphericalProjection projection;
2281  size_t spherical_size;
2282  uint32_t l = 0, t = 0, r = 0, b = 0;
2283  uint32_t padding = 0;
2284  int ret;
2285 
2286  if (mkv_projection->private.size && priv_data[0] != 0) {
2287  av_log(logctx, AV_LOG_WARNING, "Unknown spherical metadata\n");
2288  return 0;
2289  }
2290 
2291  switch (track->video.projection.type) {
2293  return mkv_create_display_matrix(st, mkv_projection, logctx);
2295  if (track->video.projection.private.size == 20) {
2296  t = AV_RB32(priv_data + 4);
2297  b = AV_RB32(priv_data + 8);
2298  l = AV_RB32(priv_data + 12);
2299  r = AV_RB32(priv_data + 16);
2300 
2301  if (b >= UINT_MAX - t || r >= UINT_MAX - l) {
2302  av_log(logctx, AV_LOG_ERROR,
2303  "Invalid bounding rectangle coordinates "
2304  "%"PRIu32",%"PRIu32",%"PRIu32",%"PRIu32"\n",
2305  l, t, r, b);
2306  return AVERROR_INVALIDDATA;
2307  }
2308  } else if (track->video.projection.private.size != 0) {
2309  av_log(logctx, AV_LOG_ERROR, "Unknown spherical metadata\n");
2310  return AVERROR_INVALIDDATA;
2311  }
2312 
2313  if (l || t || r || b)
2314  projection = AV_SPHERICAL_EQUIRECTANGULAR_TILE;
2315  else
2316  projection = AV_SPHERICAL_EQUIRECTANGULAR;
2317  break;
2319  if (track->video.projection.private.size < 4) {
2320  av_log(logctx, AV_LOG_ERROR, "Missing projection private properties\n");
2321  return AVERROR_INVALIDDATA;
2322  } else if (track->video.projection.private.size == 12) {
2323  uint32_t layout = AV_RB32(priv_data + 4);
2324  if (layout) {
2325  av_log(logctx, AV_LOG_WARNING,
2326  "Unknown spherical cubemap layout %"PRIu32"\n", layout);
2327  return 0;
2328  }
2329  projection = AV_SPHERICAL_CUBEMAP;
2330  padding = AV_RB32(priv_data + 8);
2331  } else {
2332  av_log(logctx, AV_LOG_ERROR, "Unknown spherical metadata\n");
2333  return AVERROR_INVALIDDATA;
2334  }
2335  break;
2336  default:
2337  av_log(logctx, AV_LOG_WARNING,
2338  "Unknown spherical metadata type %"PRIu64"\n",
2339  track->video.projection.type);
2340  return 0;
2341  }
2342 
2343  spherical = av_spherical_alloc(&spherical_size);
2344  if (!spherical)
2345  return AVERROR(ENOMEM);
2346 
2347  spherical->projection = projection;
2348 
2349  spherical->yaw = (int32_t) (track->video.projection.yaw * (1 << 16));
2350  spherical->pitch = (int32_t) (track->video.projection.pitch * (1 << 16));
2351  spherical->roll = (int32_t) (track->video.projection.roll * (1 << 16));
2352 
2353  spherical->padding = padding;
2354 
2355  spherical->bound_left = l;
2356  spherical->bound_top = t;
2357  spherical->bound_right = r;
2358  spherical->bound_bottom = b;
2359 
2360  ret = av_stream_add_side_data(st, AV_PKT_DATA_SPHERICAL, (uint8_t *)spherical,
2361  spherical_size);
2362  if (ret < 0) {
2363  av_freep(&spherical);
2364  return ret;
2365  }
2366 
2367  return 0;
2368 }
2369 
2371  EbmlBin *bin)
2372 {
2373  return ff_isom_parse_dvcc_dvvc(s, st, bin->data, bin->size);
2374 }
2375 
2377 {
2378  const EbmlList *mappings_list = &track->block_addition_mappings;
2379  MatroskaBlockAdditionMapping *mappings = mappings_list->elem;
2380  int ret;
2381 
2382  for (int i = 0; i < mappings_list->nb_elem; i++) {
2383  MatroskaBlockAdditionMapping *mapping = &mappings[i];
2384 
2385  switch (mapping->type) {
2386  case MKBETAG('d','v','c','C'):
2387  case MKBETAG('d','v','v','C'):
2388  if ((ret = mkv_parse_dvcc_dvvc(s, st, track, &mapping->extradata)) < 0)
2389  return ret;
2390 
2391  break;
2392  default:
2394  "Unknown block additional mapping type 0x%"PRIx64", value %"PRIu64", name \"%s\"\n",
2395  mapping->type, mapping->value, mapping->name ? mapping->name : "");
2396  }
2397  }
2398 
2399  return 0;
2400 }
2401 
2402 static int get_qt_codec(MatroskaTrack *track, uint32_t *fourcc, enum AVCodecID *codec_id)
2403 {
2404  const AVCodecTag *codec_tags;
2405 
2406  codec_tags = track->type == MATROSKA_TRACK_TYPE_VIDEO ?
2408 
2409  /* Normalize noncompliant private data that starts with the fourcc
2410  * by expanding/shifting the data by 4 bytes and storing the data
2411  * size at the start. */
2412  if (ff_codec_get_id(codec_tags, AV_RL32(track->codec_priv.data))) {
2413  int ret = av_buffer_realloc(&track->codec_priv.buf,
2415  if (ret < 0)
2416  return ret;
2417 
2418  track->codec_priv.data = track->codec_priv.buf->data;
2419  memmove(track->codec_priv.data + 4, track->codec_priv.data, track->codec_priv.size);
2420  track->codec_priv.size += 4;
2421  AV_WB32(track->codec_priv.data, track->codec_priv.size);
2422  }
2423 
2424  *fourcc = AV_RL32(track->codec_priv.data + 4);
2425  *codec_id = ff_codec_get_id(codec_tags, *fourcc);
2426 
2427  return 0;
2428 }
2429 
2431 {
2432  MatroskaDemuxContext *matroska = s->priv_data;
2433  MatroskaTrack *tracks = matroska->tracks.elem;
2434  int i, j, ret;
2435  int k;
2436 
2437  for (i = 0; i < matroska->tracks.nb_elem; i++) {
2438  MatroskaTrack *track = &tracks[i];
2440  EbmlList *encodings_list = &track->encodings;
2441  MatroskaTrackEncoding *encodings = encodings_list->elem;
2442  uint8_t *extradata = NULL;
2443  int extradata_size = 0;
2444  int extradata_offset = 0;
2445  uint32_t fourcc = 0;
2446  FFIOContext b;
2447  AVStream *st;
2448  FFStream *sti;
2449  char* key_id_base64 = NULL;
2450  int bit_depth = -1;
2451 
2452  /* Apply some sanity checks. */
2453  if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
2454  track->type != MATROSKA_TRACK_TYPE_AUDIO &&
2455  track->type != MATROSKA_TRACK_TYPE_SUBTITLE &&
2456  track->type != MATROSKA_TRACK_TYPE_METADATA) {
2457  av_log(matroska->ctx, AV_LOG_INFO,
2458  "Unknown or unsupported track type %"PRIu64"\n",
2459  track->type);
2460  continue;
2461  }
2462  if (!track->codec_id)
2463  continue;
2464 
2465  if ( track->type == MATROSKA_TRACK_TYPE_AUDIO && track->codec_id[0] != 'A'
2466  || track->type == MATROSKA_TRACK_TYPE_VIDEO && track->codec_id[0] != 'V'
2467  || track->type == MATROSKA_TRACK_TYPE_SUBTITLE && track->codec_id[0] != 'D' && track->codec_id[0] != 'S'
2468  || track->type == MATROSKA_TRACK_TYPE_METADATA && track->codec_id[0] != 'D' && track->codec_id[0] != 'S'
2469  ) {
2470  av_log(matroska->ctx, AV_LOG_INFO, "Inconsistent track type\n");
2471  continue;
2472  }
2473 
2474  if (track->audio.samplerate < 0 || track->audio.samplerate > INT_MAX ||
2475  isnan(track->audio.samplerate)) {
2476  av_log(matroska->ctx, AV_LOG_WARNING,
2477  "Invalid sample rate %f, defaulting to 8000 instead.\n",
2478  track->audio.samplerate);
2479  track->audio.samplerate = 8000;
2480  }
2481 
2482  if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2483  if (!track->default_duration && track->video.frame_rate > 0) {
2484  double default_duration = 1000000000 / track->video.frame_rate;
2485  if (default_duration > UINT64_MAX || default_duration < 0) {
2486  av_log(matroska->ctx, AV_LOG_WARNING,
2487  "Invalid frame rate %e. Cannot calculate default duration.\n",
2488  track->video.frame_rate);
2489  } else {
2490  track->default_duration = default_duration;
2491  }
2492  }
2493  if (track->video.display_width == -1)
2494  track->video.display_width = track->video.pixel_width;
2495  if (track->video.display_height == -1)
2496  track->video.display_height = track->video.pixel_height;
2497  if (track->video.color_space.size == 4)
2498  fourcc = AV_RL32(track->video.color_space.data);
2499  } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2500  if (!track->audio.out_samplerate)
2501  track->audio.out_samplerate = track->audio.samplerate;
2502  }
2503  if (encodings_list->nb_elem > 1) {
2504  av_log(matroska->ctx, AV_LOG_ERROR,
2505  "Multiple combined encodings not supported");
2506  } else if (encodings_list->nb_elem == 1) {
2507  if (encodings[0].type) {
2508  if (encodings[0].encryption.key_id.size > 0) {
2509  /* Save the encryption key id to be stored later as a
2510  metadata tag. */
2511  const int b64_size = AV_BASE64_SIZE(encodings[0].encryption.key_id.size);
2512  key_id_base64 = av_malloc(b64_size);
2513  if (key_id_base64 == NULL)
2514  return AVERROR(ENOMEM);
2515 
2516  av_base64_encode(key_id_base64, b64_size,
2517  encodings[0].encryption.key_id.data,
2518  encodings[0].encryption.key_id.size);
2519  } else {
2520  encodings[0].scope = 0;
2521  av_log(matroska->ctx, AV_LOG_ERROR,
2522  "Unsupported encoding type");
2523  }
2524  } else if (
2525 #if CONFIG_ZLIB
2526  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
2527 #endif
2528 #if CONFIG_BZLIB
2529  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
2530 #endif
2531  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO &&
2532  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP) {
2533  encodings[0].scope = 0;
2534  av_log(matroska->ctx, AV_LOG_ERROR,
2535  "Unsupported encoding type");
2536  } else if (track->codec_priv.size && encodings[0].scope & 2) {
2537  uint8_t *codec_priv = track->codec_priv.data;
2538  int ret = matroska_decode_buffer(&track->codec_priv.data,
2539  &track->codec_priv.size,
2540  track);
2541  if (ret < 0) {
2542  track->codec_priv.data = NULL;
2543  track->codec_priv.size = 0;
2544  av_log(matroska->ctx, AV_LOG_ERROR,
2545  "Failed to decode codec private data\n");
2546  }
2547 
2548  if (codec_priv != track->codec_priv.data) {
2549  av_buffer_unref(&track->codec_priv.buf);
2550  if (track->codec_priv.data) {
2551  track->codec_priv.buf = av_buffer_create(track->codec_priv.data,
2553  NULL, NULL, 0);
2554  if (!track->codec_priv.buf) {
2555  av_freep(&track->codec_priv.data);
2556  track->codec_priv.size = 0;
2557  return AVERROR(ENOMEM);
2558  }
2559  }
2560  }
2561  }
2562  }
2563  track->needs_decoding = encodings && !encodings[0].type &&
2564  encodings[0].scope & 1 &&
2565  (encodings[0].compression.algo !=
2567  encodings[0].compression.settings.size);
2568 
2569  for (j = 0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++) {
2570  if (av_strstart(track->codec_id, ff_mkv_codec_tags[j].str, NULL)) {
2572  break;
2573  }
2574  }
2575 
2576  st = track->stream = avformat_new_stream(s, NULL);
2577  if (!st) {
2578  av_free(key_id_base64);
2579  return AVERROR(ENOMEM);
2580  }
2581  sti = ffstream(st);
2582 
2583  if (key_id_base64) {
2584  /* export encryption key id as base64 metadata tag */
2585  av_dict_set(&st->metadata, "enc_key_id", key_id_base64,
2587  }
2588 
2589  if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC") &&
2590  track->codec_priv.size >= 40 &&
2591  track->codec_priv.data) {
2592  track->ms_compat = 1;
2593  bit_depth = AV_RL16(track->codec_priv.data + 14);
2594  fourcc = AV_RL32(track->codec_priv.data + 16);
2596  fourcc);
2597  if (!codec_id)
2599  fourcc);
2600  extradata_offset = 40;
2601  } else if (!strcmp(track->codec_id, "A_MS/ACM") &&
2602  track->codec_priv.size >= 14 &&
2603  track->codec_priv.data) {
2604  int ret;
2606  track->codec_priv.size,
2607  0, NULL, NULL, NULL, NULL);
2608  ret = ff_get_wav_header(s, &b.pub, st->codecpar,
2609  track->codec_priv.size, 0);
2610  if (ret < 0)
2611  return ret;
2612  codec_id = st->codecpar->codec_id;
2613  fourcc = st->codecpar->codec_tag;
2614  extradata_offset = FFMIN(track->codec_priv.size, 18);
2615  } else if (!strcmp(track->codec_id, "A_QUICKTIME")
2616  /* Normally 36, but allow noncompliant private data */
2617  && (track->codec_priv.size >= 32)
2618  && (track->codec_priv.data)) {
2619  uint16_t sample_size;
2620  int ret = get_qt_codec(track, &fourcc, &codec_id);
2621  if (ret < 0)
2622  return ret;
2623  sample_size = AV_RB16(track->codec_priv.data + 26);
2624  if (fourcc == 0) {
2625  if (sample_size == 8) {
2626  fourcc = MKTAG('r','a','w',' ');
2628  } else if (sample_size == 16) {
2629  fourcc = MKTAG('t','w','o','s');
2631  }
2632  }
2633  if ((fourcc == MKTAG('t','w','o','s') ||
2634  fourcc == MKTAG('s','o','w','t')) &&
2635  sample_size == 8)
2637  } else if (!strcmp(track->codec_id, "V_QUICKTIME") &&
2638  (track->codec_priv.size >= 21) &&
2639  (track->codec_priv.data)) {
2640  int ret = get_qt_codec(track, &fourcc, &codec_id);
2641  if (ret < 0)
2642  return ret;
2643  if (codec_id == AV_CODEC_ID_NONE && AV_RL32(track->codec_priv.data+4) == AV_RL32("SMI ")) {
2644  fourcc = MKTAG('S','V','Q','3');
2646  }
2647  if (codec_id == AV_CODEC_ID_NONE)
2648  av_log(matroska->ctx, AV_LOG_ERROR,
2649  "mov FourCC not found %s.\n", av_fourcc2str(fourcc));
2650  if (track->codec_priv.size >= 86) {
2651  bit_depth = AV_RB16(track->codec_priv.data + 82);
2653  track->codec_priv.size,
2654  0, NULL, NULL, NULL, NULL);
2655  if (ff_get_qtpalette(codec_id, &b.pub, track->palette)) {
2656  bit_depth &= 0x1F;
2657  track->has_palette = 1;
2658  }
2659  }
2660  } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
2661  switch (track->audio.bitdepth) {
2662  case 8:
2664  break;
2665  case 24:
2667  break;
2668  case 32:
2670  break;
2671  }
2672  } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
2673  switch (track->audio.bitdepth) {
2674  case 8:
2676  break;
2677  case 24:
2679  break;
2680  case 32:
2682  break;
2683  }
2684  } else if (codec_id == AV_CODEC_ID_PCM_F32LE &&
2685  track->audio.bitdepth == 64) {
2687  } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
2688  int profile = matroska_aac_profile(track->codec_id);
2689  int sri = matroska_aac_sri(track->audio.samplerate);
2690  extradata = av_mallocz(5 + AV_INPUT_BUFFER_PADDING_SIZE);
2691  if (!extradata)
2692  return AVERROR(ENOMEM);
2693  extradata[0] = (profile << 3) | ((sri & 0x0E) >> 1);
2694  extradata[1] = ((sri & 0x01) << 7) | (track->audio.channels << 3);
2695  if (strstr(track->codec_id, "SBR")) {
2696  sri = matroska_aac_sri(track->audio.out_samplerate);
2697  extradata[2] = 0x56;
2698  extradata[3] = 0xE5;
2699  extradata[4] = 0x80 | (sri << 3);
2700  extradata_size = 5;
2701  } else
2702  extradata_size = 2;
2703  } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX - 12 - AV_INPUT_BUFFER_PADDING_SIZE) {
2704  /* Only ALAC's magic cookie is stored in Matroska's track headers.
2705  * Create the "atom size", "tag", and "tag version" fields the
2706  * decoder expects manually. */
2707  extradata_size = 12 + track->codec_priv.size;
2708  extradata = av_mallocz(extradata_size +
2710  if (!extradata)
2711  return AVERROR(ENOMEM);
2712  AV_WB32(extradata, extradata_size);
2713  memcpy(&extradata[4], "alac", 4);
2714  AV_WB32(&extradata[8], 0);
2715  memcpy(&extradata[12], track->codec_priv.data,
2716  track->codec_priv.size);
2717  } else if (codec_id == AV_CODEC_ID_TTA) {
2718  uint8_t *ptr;
2719  if (track->audio.channels > UINT16_MAX ||
2720  track->audio.bitdepth > UINT16_MAX) {
2721  av_log(matroska->ctx, AV_LOG_WARNING,
2722  "Too large audio channel number %"PRIu64
2723  " or bitdepth %"PRIu64". Skipping track.\n",
2724  track->audio.channels, track->audio.bitdepth);
2725  if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
2726  return AVERROR_INVALIDDATA;
2727  else
2728  continue;
2729  }
2730  if (track->audio.out_samplerate < 0 || track->audio.out_samplerate > INT_MAX)
2731  return AVERROR_INVALIDDATA;
2732  extradata_size = 22;
2733  extradata = av_mallocz(extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
2734  if (!extradata)
2735  return AVERROR(ENOMEM);
2736  ptr = extradata;
2737  bytestream_put_be32(&ptr, AV_RB32("TTA1"));
2738  bytestream_put_le16(&ptr, 1);
2739  bytestream_put_le16(&ptr, track->audio.channels);
2740  bytestream_put_le16(&ptr, track->audio.bitdepth);
2741  bytestream_put_le32(&ptr, track->audio.out_samplerate);
2742  bytestream_put_le32(&ptr, av_rescale(matroska->duration * matroska->time_scale,
2743  track->audio.out_samplerate,
2744  AV_TIME_BASE * 1000));
2745  } else if (codec_id == AV_CODEC_ID_RV10 ||
2749  extradata_offset = 26;
2750  } else if (codec_id == AV_CODEC_ID_RA_144) {
2751  track->audio.out_samplerate = 8000;
2752  track->audio.channels = 1;
2753  } else if ((codec_id == AV_CODEC_ID_RA_288 ||
2757  && track->codec_priv.data) {
2758  const uint8_t *ptr = track->codec_priv.data;
2759  int flavor;
2760 
2761  if (track->codec_priv.size < 46)
2762  return AVERROR_INVALIDDATA;
2763  ptr += 22;
2764  flavor = bytestream_get_be16(&ptr);
2765  track->audio.coded_framesize = bytestream_get_be32(&ptr);
2766  ptr += 12;
2767  track->audio.sub_packet_h = bytestream_get_be16(&ptr);
2768  track->audio.frame_size = bytestream_get_be16(&ptr);
2769  track->audio.sub_packet_size = bytestream_get_be16(&ptr);
2770  if (track->audio.coded_framesize <= 0 ||
2771  track->audio.sub_packet_h <= 0 ||
2772  track->audio.frame_size <= 0)
2773  return AVERROR_INVALIDDATA;
2774 
2775  if (codec_id == AV_CODEC_ID_RA_288) {
2776  if (track->audio.sub_packet_h & 1 || 2 * track->audio.frame_size
2777  != (int64_t)track->audio.sub_packet_h * track->audio.coded_framesize)
2778  return AVERROR_INVALIDDATA;
2779  st->codecpar->block_align = track->audio.coded_framesize;
2780  track->codec_priv.size = 0;
2781  } else {
2782  if (codec_id == AV_CODEC_ID_SIPR) {
2783  static const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
2784  if (flavor > 3)
2785  return AVERROR_INVALIDDATA;
2786  track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
2787  st->codecpar->bit_rate = sipr_bit_rate[flavor];
2788  } else if (track->audio.sub_packet_size <= 0 ||
2789  track->audio.frame_size % track->audio.sub_packet_size)
2790  return AVERROR_INVALIDDATA;
2791  st->codecpar->block_align = track->audio.sub_packet_size;
2792  extradata_offset = 78;
2793  }
2794  track->audio.buf = av_malloc_array(track->audio.sub_packet_h,
2795  track->audio.frame_size);
2796  if (!track->audio.buf)
2797  return AVERROR(ENOMEM);
2798  } else if (codec_id == AV_CODEC_ID_FLAC && track->codec_priv.size) {
2799  ret = matroska_parse_flac(s, track, &extradata_offset);
2800  if (ret < 0)
2801  return ret;
2802  } else if (codec_id == AV_CODEC_ID_WAVPACK && track->codec_priv.size < 2) {
2803  av_log(matroska->ctx, AV_LOG_INFO, "Assuming WavPack version 4.10 "
2804  "in absence of valid CodecPrivate.\n");
2805  extradata_size = 2;
2806  extradata = av_mallocz(2 + AV_INPUT_BUFFER_PADDING_SIZE);
2807  if (!extradata)
2808  return AVERROR(ENOMEM);
2809  AV_WL16(extradata, 0x410);
2810  } else if (codec_id == AV_CODEC_ID_PRORES && track->codec_priv.size == 4) {
2811  fourcc = AV_RL32(track->codec_priv.data);
2812  } else if (codec_id == AV_CODEC_ID_VP9 && track->codec_priv.size) {
2813  /* we don't need any value stored in CodecPrivate.
2814  make sure that it's not exported as extradata. */
2815  track->codec_priv.size = 0;
2816  }
2817  track->codec_priv.size -= extradata_offset;
2818 
2819  if (codec_id == AV_CODEC_ID_NONE)
2820  av_log(matroska->ctx, AV_LOG_INFO,
2821  "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
2822 
2823  if (track->time_scale < 0.01) {
2824  av_log(matroska->ctx, AV_LOG_WARNING,
2825  "Track TimestampScale too small %f, assuming 1.0.\n",
2826  track->time_scale);
2827  track->time_scale = 1.0;
2828  }
2829  avpriv_set_pts_info(st, 64, matroska->time_scale * track->time_scale,
2830  1000 * 1000 * 1000); /* 64 bit pts in ns */
2831 
2832  /* convert the delay from ns to the track timebase */
2834  (AVRational){ 1, 1000000000 },
2835  st->time_base);
2836 
2837  st->codecpar->codec_id = codec_id;
2838 
2839  if (strcmp(track->language, "und"))
2840  av_dict_set(&st->metadata, "language", track->language, 0);
2841  av_dict_set(&st->metadata, "title", track->name, 0);
2842 
2843  if (track->flag_default)
2845  if (track->flag_forced)
2847  if (track->flag_comment)
2849  if (track->flag_hearingimpaired)
2851  if (track->flag_visualimpaired)
2853  if (track->flag_original.count > 0)
2856 
2857  if (!st->codecpar->extradata) {
2858  if (extradata) {
2859  st->codecpar->extradata = extradata;
2860  st->codecpar->extradata_size = extradata_size;
2861  } else if (track->codec_priv.data && track->codec_priv.size > 0) {
2862  if (ff_alloc_extradata(st->codecpar, track->codec_priv.size))
2863  return AVERROR(ENOMEM);
2864  memcpy(st->codecpar->extradata,
2865  track->codec_priv.data + extradata_offset,
2866  track->codec_priv.size);
2867  }
2868  }
2869 
2870  if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2872  int display_width_mul = 1;
2873  int display_height_mul = 1;
2874 
2876  st->codecpar->codec_tag = fourcc;
2877  if (bit_depth >= 0)
2879  st->codecpar->width = track->video.pixel_width;
2880  st->codecpar->height = track->video.pixel_height;
2881 
2883  st->codecpar->field_order = mkv_field_order(matroska, track->video.field_order);
2886 
2888  mkv_stereo_mode_display_mul(track->video.stereo_mode, &display_width_mul, &display_height_mul);
2889 
2891  if (track->video.display_width && track->video.display_height &&
2892  st->codecpar->height < INT64_MAX / track->video.display_width / display_width_mul &&
2893  st->codecpar->width < INT64_MAX / track->video.display_height / display_height_mul)
2895  &st->sample_aspect_ratio.den,
2896  st->codecpar->height * track->video.display_width * display_width_mul,
2897  st->codecpar->width * track->video.display_height * display_height_mul,
2898  INT_MAX);
2899  }
2900  if (st->codecpar->codec_id != AV_CODEC_ID_HEVC)
2902 
2903  if (track->default_duration) {
2904  int div = track->default_duration <= INT64_MAX ? 1 : 2;
2906  1000000000 / div, track->default_duration / div, 30000);
2907 #if FF_API_R_FRAME_RATE
2908  if ( st->avg_frame_rate.num < st->avg_frame_rate.den * 1000LL
2909  && st->avg_frame_rate.num > st->avg_frame_rate.den * 5LL)
2910  st->r_frame_rate = st->avg_frame_rate;
2911 #endif
2912  }
2913 
2914  /* export stereo mode flag as metadata tag */
2916  av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
2917 
2918  /* export alpha mode flag as metadata tag */
2919  if (track->video.alpha_mode)
2920  av_dict_set(&st->metadata, "alpha_mode", "1", 0);
2921 
2922  /* if we have virtual track, mark the real tracks */
2923  for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
2924  char buf[32];
2926  continue;
2927  snprintf(buf, sizeof(buf), "%s_%d",
2929  for (k=0; k < matroska->tracks.nb_elem; k++)
2930  if (planes[j].uid == tracks[k].uid && tracks[k].stream) {
2931  av_dict_set(&tracks[k].stream->metadata,
2932  "stereo_mode", buf, 0);
2933  break;
2934  }
2935  }
2936  // add stream level stereo3d side data if it is a supported format
2938  track->video.stereo_mode != 10 && track->video.stereo_mode != 12) {
2939  int ret = ff_mkv_stereo3d_conv(st, track->video.stereo_mode);
2940  if (ret < 0)
2941  return ret;
2942  }
2943 
2944  ret = mkv_parse_video_color(st, track);
2945  if (ret < 0)
2946  return ret;
2947  ret = mkv_parse_video_projection(st, track, matroska->ctx);
2948  if (ret < 0)
2949  return ret;
2950  } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2952  st->codecpar->codec_tag = fourcc;
2953  st->codecpar->sample_rate = track->audio.out_samplerate;
2954  // channel layout may be already set by codec private checks above
2958  }
2959  if (!st->codecpar->bits_per_coded_sample)
2961  if (st->codecpar->codec_id == AV_CODEC_ID_MP3 ||
2962  st->codecpar->codec_id == AV_CODEC_ID_MLP ||
2965  else if (st->codecpar->codec_id != AV_CODEC_ID_AAC)
2967  if (track->codec_delay > 0) {
2969  (AVRational){1, 1000000000},
2970  (AVRational){1, st->codecpar->codec_id == AV_CODEC_ID_OPUS ?
2971  48000 : st->codecpar->sample_rate});
2972  }
2973  if (track->seek_preroll > 0) {
2975  (AVRational){1, 1000000000},
2976  (AVRational){1, st->codecpar->sample_rate});
2977  }
2978  } else if (codec_id == AV_CODEC_ID_WEBVTT) {
2980 
2981  if (!strcmp(track->codec_id, "D_WEBVTT/CAPTIONS")) {
2983  } else if (!strcmp(track->codec_id, "D_WEBVTT/DESCRIPTIONS")) {
2985  } else if (!strcmp(track->codec_id, "D_WEBVTT/METADATA")) {
2987  }
2988  } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2990 
2991  if (track->flag_textdescriptions)
2993  }
2994 
2995  ret = mkv_parse_block_addition_mappings(s, st, track);
2996  if (ret < 0)
2997  return ret;
2998  }
2999 
3000  return 0;
3001 }
3002 
3004 {
3005  FFFormatContext *const si = ffformatcontext(s);
3006  MatroskaDemuxContext *matroska = s->priv_data;
3007  EbmlList *attachments_list = &matroska->attachments;
3008  EbmlList *chapters_list = &matroska->chapters;
3009  MatroskaAttachment *attachments;
3010  MatroskaChapter *chapters;
3011  uint64_t max_start = 0;
3012  int64_t pos;
3013  Ebml ebml = { 0 };
3014  int i, j, res;
3015 
3016  matroska->ctx = s;
3017  matroska->cues_parsing_deferred = 1;
3018 
3019  /* First read the EBML header. */
3020  if (ebml_parse(matroska, ebml_syntax, &ebml) || !ebml.doctype) {
3021  av_log(matroska->ctx, AV_LOG_ERROR, "EBML header parsing failed\n");
3022  ebml_free(ebml_syntax, &ebml);
3023  return AVERROR_INVALIDDATA;
3024  }
3025  if (ebml.version > EBML_VERSION ||
3026  ebml.max_size > sizeof(uint64_t) ||
3027  ebml.id_length > sizeof(uint32_t) ||
3028  ebml.doctype_version > 3) {
3030  "EBML version %"PRIu64", doctype %s, doc version %"PRIu64,
3031  ebml.version, ebml.doctype, ebml.doctype_version);
3032  ebml_free(ebml_syntax, &ebml);
3033  return AVERROR_PATCHWELCOME;
3034  } else if (ebml.doctype_version == 3) {
3035  av_log(matroska->ctx, AV_LOG_WARNING,
3036  "EBML header using unsupported features\n"
3037  "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
3038  ebml.version, ebml.doctype, ebml.doctype_version);
3039  }
3040  for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
3041  if (!strcmp(ebml.doctype, matroska_doctypes[i]))
3042  break;
3044  av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
3045  if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
3046  ebml_free(ebml_syntax, &ebml);
3047  return AVERROR_INVALIDDATA;
3048  }
3049  }
3050  ebml_free(ebml_syntax, &ebml);
3051 
3052  matroska->pkt = si->parse_pkt;
3053 
3054  /* The next thing is a segment. */
3055  pos = avio_tell(matroska->ctx->pb);
3056  res = ebml_parse(matroska, matroska_segments, matroska);
3057  // Try resyncing until we find an EBML_STOP type element.
3058  while (res != 1) {
3059  res = matroska_resync(matroska, pos);
3060  if (res < 0)
3061  return res;
3062  pos = avio_tell(matroska->ctx->pb);
3063  res = ebml_parse(matroska, matroska_segment, matroska);
3064  if (res == AVERROR(EIO)) // EOF is translated to EIO, this exists the loop on EOF
3065  return res;
3066  }
3067  /* Set data_offset as it might be needed later by seek_frame_generic. */
3068  if (matroska->current_id == MATROSKA_ID_CLUSTER)
3069  si->data_offset = avio_tell(matroska->ctx->pb) - 4;
3070  matroska_execute_seekhead(matroska);
3071 
3072  if (!matroska->time_scale)
3073  matroska->time_scale = 1000000;
3074  if (isnan(matroska->duration))
3075  matroska->duration = 0;
3076  if (matroska->duration)
3077  matroska->ctx->duration = matroska->duration * matroska->time_scale *
3078  1000 / AV_TIME_BASE;
3079  av_dict_set(&s->metadata, "title", matroska->title, 0);
3080  av_dict_set(&s->metadata, "encoder", matroska->muxingapp, 0);
3081 
3082  if (matroska->date_utc.size == 8)
3083  matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
3084 
3085  res = matroska_parse_tracks(s);
3086  if (res < 0)
3087  return res;
3088 
3089  attachments = attachments_list->elem;
3090  for (j = 0; j < attachments_list->nb_elem; j++) {
3091  if (!(attachments[j].filename && attachments[j].mime &&
3092  attachments[j].bin.data && attachments[j].bin.size > 0)) {
3093  av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
3094  } else {
3096  if (!st)
3097  break;
3098  av_dict_set(&st->metadata, "filename", attachments[j].filename, 0);
3099  av_dict_set(&st->metadata, "mimetype", attachments[j].mime, 0);
3100  if (attachments[j].description)
3101  av_dict_set(&st->metadata, "title", attachments[j].description, 0);
3103 
3104  for (i = 0; mkv_image_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
3105  if (av_strstart(attachments[j].mime, mkv_image_mime_tags[i].str, NULL)) {
3107  break;
3108  }
3109  }
3110 
3111  attachments[j].stream = st;
3112 
3113  if (st->codecpar->codec_id != AV_CODEC_ID_NONE) {
3114  res = ff_add_attached_pic(s, st, NULL, &attachments[j].bin.buf, 0);
3115  if (res < 0)
3116  return res;
3117  } else {
3119  if (ff_alloc_extradata(st->codecpar, attachments[j].bin.size))
3120  break;
3121  memcpy(st->codecpar->extradata, attachments[j].bin.data,
3122  attachments[j].bin.size);
3123 
3124  for (i = 0; mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
3125  if (av_strstart(attachments[j].mime, mkv_mime_tags[i].str, NULL)) {
3127  break;
3128  }
3129  }
3130  }
3131  }
3132  }
3133 
3134  chapters = chapters_list->elem;
3135  for (i = 0; i < chapters_list->nb_elem; i++)
3136  if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid &&
3137  (max_start == 0 || chapters[i].start > max_start)) {
3138  chapters[i].chapter =
3139  avpriv_new_chapter(s, chapters[i].uid,
3140  (AVRational) { 1, 1000000000 },
3141  chapters[i].start, chapters[i].end,
3142  chapters[i].title);
3143  max_start = chapters[i].start;
3144  }
3145 
3146  matroska_add_index_entries(matroska);
3147 
3149 
3150  return 0;
3152 
3153 /*
3154  * Put one packet in an application-supplied AVPacket struct.
3155  * Returns 0 on success or -1 on failure.
3156  */
3157 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
3158  AVPacket *pkt)
3159 {
3160  if (matroska->queue.head) {
3161  MatroskaTrack *tracks = matroska->tracks.elem;
3162  MatroskaTrack *track;
3163 
3164  avpriv_packet_list_get(&matroska->queue, pkt);
3165  track = &tracks[pkt->stream_index];
3166  if (track->has_palette) {
3168  if (!pal) {
3169  av_log(matroska->ctx, AV_LOG_ERROR, "Cannot append palette to packet\n");
3170  } else {
3171  memcpy(pal, track->palette, AVPALETTE_SIZE);
3172  }
3173  track->has_palette = 0;
3174  }
3175  return 0;
3176  }
3177 
3178  return -1;
3179 }
3180 
3181 /*
3182  * Free all packets in our internal queue.
3183  */
3184 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
3185 {
3186  avpriv_packet_list_free(&matroska->queue);
3187 }
3188 
3189 static int matroska_parse_laces(MatroskaDemuxContext *matroska, uint8_t **buf,
3190  int size, int type, AVIOContext *pb,
3191  uint32_t lace_size[256], int *laces)
3192 {
3193  int n;
3194  uint8_t *data = *buf;
3195 
3196  if (!type) {
3197  *laces = 1;
3198  lace_size[0] = size;
3199  return 0;
3200  }
3201 
3202  if (size <= 0)
3203  return AVERROR_INVALIDDATA;
3204 
3205  *laces = *data + 1;
3206  data += 1;
3207  size -= 1;
3208 
3209  switch (type) {
3210  case 0x1: /* Xiph lacing */
3211  {
3212  uint8_t temp;
3213  uint32_t total = 0;
3214  for (n = 0; n < *laces - 1; n++) {
3215  lace_size[n] = 0;
3216 
3217  do {
3218  if (size <= total)
3219  return AVERROR_INVALIDDATA;
3220  temp = *data;
3221  total += temp;
3222  lace_size[n] += temp;
3223  data += 1;
3224  size -= 1;
3225  } while (temp == 0xff);
3226  }
3227  if (size < total)
3228  return AVERROR_INVALIDDATA;
3229 
3230  lace_size[n] = size - total;
3231  break;
3232  }
3233 
3234  case 0x2: /* fixed-size lacing */
3235  if (size % (*laces))
3236  return AVERROR_INVALIDDATA;
3237  for (n = 0; n < *laces; n++)
3238  lace_size[n] = size / *laces;
3239  break;
3240 
3241  case 0x3: /* EBML lacing */
3242  {
3243  uint64_t num;
3244  uint64_t total;
3245  int offset;
3246 
3247  avio_skip(pb, 4);
3248 
3249  n = ebml_read_num(matroska, pb, 8, &num, 1);
3250  if (n < 0)
3251  return n;
3252  if (num > INT_MAX)
3253  return AVERROR_INVALIDDATA;
3254 
3255  total = lace_size[0] = num;
3256  offset = n;
3257  for (n = 1; n < *laces - 1; n++) {
3258  int64_t snum;
3259  int r;
3260  r = matroska_ebmlnum_sint(matroska, pb, &snum);
3261  if (r < 0)
3262  return r;
3263  if (lace_size[n - 1] + snum > (uint64_t)INT_MAX)
3264  return AVERROR_INVALIDDATA;
3265 
3266  lace_size[n] = lace_size[n - 1] + snum;
3267  total += lace_size[n];
3268  offset += r;
3269  }
3270  data += offset;
3271  size -= offset;
3272  if (size < total)
3273  return AVERROR_INVALIDDATA;
3274 
3275  lace_size[*laces - 1] = size - total;
3276  break;
3277  }
3278  }
3280  *buf = data;
3281 
3282  return 0;
3283 }
3284 
3285 static int matroska_parse_rm_audio(MatroskaDemuxContext *matroska,
3286  MatroskaTrack *track, AVStream *st,
3287  uint8_t *data, int size, uint64_t timecode,
3288  int64_t pos)
3289 {
3290  const int a = st->codecpar->block_align;
3291  const int sps = track->audio.sub_packet_size;
3292  const int cfs = track->audio.coded_framesize;
3293  const int h = track->audio.sub_packet_h;
3294  const int w = track->audio.frame_size;
3295  int y = track->audio.sub_packet_cnt;
3296  int x;
3297 
3298  if (!track->audio.pkt_cnt) {
3299  if (track->audio.sub_packet_cnt == 0)
3300  track->audio.buf_timecode = timecode;
3301  if (st->codecpar->codec_id == AV_CODEC_ID_RA_288) {
3302  if (size < cfs * h / 2) {
3303  av_log(matroska->ctx, AV_LOG_ERROR,
3304  "Corrupt int4 RM-style audio packet size\n");
3305  return AVERROR_INVALIDDATA;
3306  }
3307  for (x = 0; x < h / 2; x++)
3308  memcpy(track->audio.buf + x * 2 * w + y * cfs,
3309  data + x * cfs, cfs);
3310  } else if (st->codecpar->codec_id == AV_CODEC_ID_SIPR) {
3311  if (size < w) {
3312  av_log(matroska->ctx, AV_LOG_ERROR,
3313  "Corrupt sipr RM-style audio packet size\n");
3314  return AVERROR_INVALIDDATA;
3315  }
3316  memcpy(track->audio.buf + y * w, data, w);
3317  } else {
3318  if (size < w) {
3319  av_log(matroska->ctx, AV_LOG_ERROR,
3320  "Corrupt generic RM-style audio packet size\n");
3321  return AVERROR_INVALIDDATA;
3322  }
3323  for (x = 0; x < w / sps; x++)
3324  memcpy(track->audio.buf +
3325  sps * (h * x + ((h + 1) / 2) * (y & 1) + (y >> 1)),
3326  data + x * sps, sps);
3327  }
3328 
3329  if (++track->audio.sub_packet_cnt >= h) {
3330  if (st->codecpar->codec_id == AV_CODEC_ID_SIPR)
3331  ff_rm_reorder_sipr_data(track->audio.buf, h, w);
3332  track->audio.sub_packet_cnt = 0;
3333  track->audio.pkt_cnt = h * w / a;
3334  }
3335  }
3336 
3337  while (track->audio.pkt_cnt) {
3338  int ret;
3339  AVPacket *pkt = matroska->pkt;
3340 
3341  ret = av_new_packet(pkt, a);
3342  if (ret < 0) {
3343  return ret;
3344  }
3345  memcpy(pkt->data,
3346  track->audio.buf + a * (h * w / a - track->audio.pkt_cnt--),
3347  a);
3348  pkt->pts = track->audio.buf_timecode;
3350  pkt->pos = pos;
3351  pkt->stream_index = st->index;
3352  ret = avpriv_packet_list_put(&matroska->queue, pkt, NULL, 0);
3353  if (ret < 0) {
3355  return AVERROR(ENOMEM);
3356  }
3357  }
3358 
3359  return 0;
3360 }
3361 
3362 /* reconstruct full wavpack blocks from mangled matroska ones */
3363 static int matroska_parse_wavpack(MatroskaTrack *track,
3364  uint8_t **data, int *size)
3365 {
3366  uint8_t *dst = NULL;
3367  uint8_t *src = *data;
3368  int dstlen = 0;
3369  int srclen = *size;
3370  uint32_t samples;
3371  uint16_t ver;
3372  int ret, offset = 0;
3373 
3374  if (srclen < 12)
3375  return AVERROR_INVALIDDATA;
3376 
3377  av_assert1(track->stream->codecpar->extradata_size >= 2);
3378  ver = AV_RL16(track->stream->codecpar->extradata);
3379 
3380  samples = AV_RL32(src);
3381  src += 4;
3382  srclen -= 4;
3383 
3384  while (srclen >= 8) {
3385  int multiblock;
3386  uint32_t blocksize;
3387  uint8_t *tmp;
3388 
3389  uint32_t flags = AV_RL32(src);
3390  uint32_t crc = AV_RL32(src + 4);
3391  src += 8;
3392  srclen -= 8;
3393 
3394  multiblock = (flags & 0x1800) != 0x1800;
3395  if (multiblock) {
3396  if (srclen < 4) {
3398  goto fail;
3399  }
3400  blocksize = AV_RL32(src);
3401  src += 4;
3402  srclen -= 4;
3403  } else
3404  blocksize = srclen;
3405 
3406  if (blocksize > srclen) {
3408  goto fail;
3409  }
3410 
3411  tmp = av_realloc(dst, dstlen + blocksize + 32 + AV_INPUT_BUFFER_PADDING_SIZE);
3412  if (!tmp) {
3413  ret = AVERROR(ENOMEM);
3414  goto fail;
3415  }
3416  dst = tmp;
3417  dstlen += blocksize + 32;
3418 
3419  AV_WL32(dst + offset, MKTAG('w', 'v', 'p', 'k')); // tag
3420  AV_WL32(dst + offset + 4, blocksize + 24); // blocksize - 8
3421  AV_WL16(dst + offset + 8, ver); // version
3422  AV_WL16(dst + offset + 10, 0); // track/index_no
3423  AV_WL32(dst + offset + 12, 0); // total samples
3424  AV_WL32(dst + offset + 16, 0); // block index
3425  AV_WL32(dst + offset + 20, samples); // number of samples
3426  AV_WL32(dst + offset + 24, flags); // flags
3427  AV_WL32(dst + offset + 28, crc); // crc
3428  memcpy(dst + offset + 32, src, blocksize); // block data
3429 
3430  src += blocksize;
3431  srclen -= blocksize;
3432  offset += blocksize + 32;
3433  }
3434 
3435  memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3436 
3437  *data = dst;
3438  *size = dstlen;
3439 
3440  return 0;
3442 fail:
3443  av_freep(&dst);
3444  return ret;
3445 }
3446 
3447 static int matroska_parse_prores(MatroskaTrack *track,
3448  uint8_t **data, int *size)
3449 {
3450  uint8_t *dst;
3451  int dstlen = *size + 8;
3452 
3453  dst = av_malloc(dstlen + AV_INPUT_BUFFER_PADDING_SIZE);
3454  if (!dst)
3455  return AVERROR(ENOMEM);
3456 
3457  AV_WB32(dst, dstlen);
3458  AV_WB32(dst + 4, MKBETAG('i', 'c', 'p', 'f'));
3459  memcpy(dst + 8, *data, dstlen - 8);
3460  memset(dst + dstlen, 0, AV_INPUT_BUFFER_PADDING_SIZE);
3461 
3462  *data = dst;
3463  *size = dstlen;
3464 
3465  return 0;
3466 }
3467 
3468 static int matroska_parse_webvtt(MatroskaDemuxContext *matroska,
3469  MatroskaTrack *track,
3470  AVStream *st,
3471  uint8_t *data, int data_len,
3472  uint64_t timecode,
3473  uint64_t duration,
3474  int64_t pos)
3475 {
3476  AVPacket *pkt = matroska->pkt;
3477  uint8_t *id, *settings, *text, *buf;
3478  int id_len, settings_len, text_len;
3479  uint8_t *p, *q;
3480  int err;
3481 
3482  if (data_len <= 0)
3483  return AVERROR_INVALIDDATA;
3484 
3485  p = data;
3486  q = data + data_len;
3487 
3488  id = p;
3489  id_len = -1;
3490  while (p < q) {
3491  if (*p == '\r' || *p == '\n') {
3492  id_len = p - id;
3493  if (*p == '\r')
3494  p++;
3495  break;
3496  }
3497  p++;
3498  }
3499 
3500  if (p >= q || *p != '\n')
3501  return AVERROR_INVALIDDATA;
3502  p++;
3503 
3504  settings = p;
3505  settings_len = -1;
3506  while (p < q) {
3507  if (*p == '\r' || *p == '\n') {
3508  settings_len = p - settings;
3509  if (*p == '\r')
3510  p++;
3511  break;
3512  }
3513  p++;
3514  }
3515 
3516  if (p >= q || *p != '\n')
3517  return AVERROR_INVALIDDATA;
3518  p++;
3519 
3520  text = p;
3521  text_len = q - p;
3522  while (text_len > 0) {
3523  const int len = text_len - 1;
3524  const uint8_t c = p[len];
3525  if (c != '\r' && c != '\n')
3526  break;
3527  text_len = len;
3528  }
3529 
3530  if (text_len <= 0)
3531  return AVERROR_INVALIDDATA;
3532 
3533  err = av_new_packet(pkt, text_len);
3534  if (err < 0) {
3535  return err;
3536  }
3537 
3538  memcpy(pkt->data, text, text_len);
3539 
3540  if (id_len > 0) {
3543  id_len);
3544  if (!buf) {
3546  return AVERROR(ENOMEM);
3547  }
3548  memcpy(buf, id, id_len);
3549  }
3550 
3551  if (settings_len > 0) {
3554  settings_len);
3555  if (!buf) {
3557  return AVERROR(ENOMEM);
3558  }
3559  memcpy(buf, settings, settings_len);
3560  }
3561 
3562  // Do we need this for subtitles?
3563  // pkt->flags = AV_PKT_FLAG_KEY;
3564 
3565  pkt->stream_index = st->index;
3566  pkt->pts = timecode;
3567 
3568  // Do we need this for subtitles?
3569  // pkt->dts = timecode;
3570 
3571  pkt->duration = duration;
3572  pkt->pos = pos;
3573 
3574  err = avpriv_packet_list_put(&matroska->queue, pkt, NULL, 0);
3575  if (err < 0) {
3577  return AVERROR(ENOMEM);
3578  }
3579 
3580  return 0;
3581 }
3582 
3583 static int matroska_parse_frame(MatroskaDemuxContext *matroska,
3584  MatroskaTrack *track, AVStream *st,
3585  AVBufferRef *buf, uint8_t *data, int pkt_size,
3586  uint64_t timecode, uint64_t lace_duration,
3587  int64_t pos, int is_keyframe,
3588  uint8_t *additional, uint64_t additional_id, int additional_size,
3589  int64_t discard_padding)
3590 {
3591  uint8_t *pkt_data = data;
3592  int res = 0;
3593  AVPacket *pkt = matroska->pkt;
3594 
3595  if (st->codecpar->codec_id == AV_CODEC_ID_WAVPACK) {
3596  res = matroska_parse_wavpack(track, &pkt_data, &pkt_size);
3597  if (res < 0) {
3598  av_log(matroska->ctx, AV_LOG_ERROR,
3599  "Error parsing a wavpack block.\n");
3600  goto fail;
3601  }
3602  if (!buf)
3603  av_freep(&data);
3604  buf = NULL;
3605  }
3606 
3607  if (st->codecpar->codec_id == AV_CODEC_ID_PRORES &&
3608  AV_RB32(pkt_data + 4) != MKBETAG('i', 'c', 'p', 'f')) {
3609  res = matroska_parse_prores(track, &pkt_data, &pkt_size);
3610  if (res < 0) {
3611  av_log(matroska->ctx, AV_LOG_ERROR,
3612  "Error parsing a prores block.\n");
3613  goto fail;
3614  }
3615  if (!buf)
3616  av_freep(&data);
3617  buf = NULL;
3618  }
3619 
3620  if (!pkt_size && !additional_size)
3621  goto no_output;
3622 
3623  if (!buf)
3624  pkt->buf = av_buffer_create(pkt_data, pkt_size + AV_INPUT_BUFFER_PADDING_SIZE,
3625  NULL, NULL, 0);
3626  else
3627  pkt->buf = av_buffer_ref(buf);
3628 
3629  if (!pkt->buf) {
3630  res = AVERROR(ENOMEM);
3631  goto fail;
3632  }
3633 
3634  pkt->data = pkt_data;
3635  pkt->size = pkt_size;
3636  pkt->flags = is_keyframe;
3637  pkt->stream_index = st->index;
3638 
3639  if (additional_size > 0) {
3640  uint8_t *side_data = av_packet_new_side_data(pkt,
3642  additional_size + 8);
3643  if (!side_data) {
3645  return AVERROR(ENOMEM);
3646  }
3647  AV_WB64(side_data, additional_id);
3648  memcpy(side_data + 8, additional, additional_size);
3649  }
3650 
3651  if (discard_padding) {
3652  uint8_t *side_data = av_packet_new_side_data(pkt,
3654  10);
3655  if (!side_data) {
3657  return AVERROR(ENOMEM);
3658  }
3659  discard_padding = av_rescale_q(discard_padding,
3660  (AVRational){1, 1000000000},
3661  (AVRational){1, st->codecpar->sample_rate});
3662  if (discard_padding > 0) {
3663  AV_WL32(side_data + 4, discard_padding);
3664  } else {
3665  AV_WL32(side_data, -discard_padding);
3666  }
3667  }
3668 
3669  if (track->ms_compat)
3670  pkt->dts = timecode;
3671  else
3672  pkt->pts = timecode;
3673  pkt->pos = pos;
3674  pkt->duration = lace_duration;
3675 
3676  res = avpriv_packet_list_put(&matroska->queue, pkt, NULL, 0);
3677  if (res < 0) {
3679  return AVERROR(ENOMEM);
3680  }
3681 
3682  return 0;
3683 
3684 no_output:
3686  if (!buf)
3687  av_free(pkt_data);
3688  return res;
3689 }
3690 
3691 static int matroska_parse_block(MatroskaDemuxContext *matroska, AVBufferRef *buf, uint8_t *data,
3692  int size, int64_t pos, uint64_t cluster_time,
3693  uint64_t block_duration, int is_keyframe,
3694  uint8_t *additional, uint64_t additional_id, int additional_size,
3695  int64_t cluster_pos, int64_t discard_padding)
3696 {
3697  uint64_t timecode = AV_NOPTS_VALUE;
3698  MatroskaTrack *track;
3699  FFIOContext pb;
3700  int res = 0;
3701  AVStream *st;
3702  int16_t block_time;
3703  uint32_t lace_size[256];
3704  int n, flags, laces = 0;
3705  uint64_t num;
3706  int trust_default_duration;
3707 
3708  av_assert1(buf);
3709 
3710  ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
3711 
3712  if ((n = ebml_read_num(matroska, &pb.pub, 8, &num, 1)) < 0)
3713  return n;
3714  data += n;
3715  size -= n;
3716 
3717  track = matroska_find_track_by_num(matroska, num);
3718  if (!track || size < 3)
3719  return AVERROR_INVALIDDATA;
3720 
3721  if (!(st = track->stream)) {
3722  av_log(matroska->ctx, AV_LOG_VERBOSE,
3723  "No stream associated to TrackNumber %"PRIu64". "
3724  "Ignoring Block with this TrackNumber.\n", num);
3725  return 0;
3726  }
3727 
3728  if (st->discard >= AVDISCARD_ALL)
3729  return res;
3730  if (block_duration > INT64_MAX)
3731  block_duration = INT64_MAX;
3732 
3733  block_time = sign_extend(AV_RB16(data), 16);
3734  data += 2;
3735  flags = *data++;
3736  size -= 3;
3737  if (is_keyframe == -1)
3738  is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
3739 
3740  if (cluster_time != (uint64_t) -1 &&
3741  (block_time >= 0 || cluster_time >= -block_time)) {
3742  uint64_t timecode_cluster_in_track_tb = (double) cluster_time / track->time_scale;
3743  timecode = timecode_cluster_in_track_tb + block_time - track->codec_delay_in_track_tb;
3744  if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE &&
3745  timecode < track->end_timecode)
3746  is_keyframe = 0; /* overlapping subtitles are not key frame */
3747  if (is_keyframe) {
3748  ff_reduce_index(matroska->ctx, st->index);
3749  av_add_index_entry(st, cluster_pos, timecode, 0, 0,
3751  }
3752  }
3753 
3754  if (matroska->skip_to_keyframe &&
3755  track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
3756  // Compare signed timecodes. Timecode may be negative due to codec delay
3757  // offset. We don't support timestamps greater than int64_t anyway - see
3758  // AVPacket's pts.
3759  if ((int64_t)timecode < (int64_t)matroska->skip_to_timecode)
3760  return res;
3761  if (is_keyframe)
3762  matroska->skip_to_keyframe = 0;
3763  else if (!ffstream(st)->skip_to_keyframe) {
3764  av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
3765  matroska->skip_to_keyframe = 0;
3766  }
3767  }
3768 
3769  res = matroska_parse_laces(matroska, &data, size, (flags & 0x06) >> 1,
3770  &pb.pub, lace_size, &laces);
3771  if (res < 0) {
3772  av_log(matroska->ctx, AV_LOG_ERROR, "Error parsing frame sizes.\n");
3773  return res;
3774  }
3775 
3776  trust_default_duration = track->default_duration != 0;
3777  if (track->audio.samplerate == 8000 && trust_default_duration) {
3778  // If this is needed for more codecs, then add them here
3779  if (st->codecpar->codec_id == AV_CODEC_ID_AC3) {
3780  if (track->audio.samplerate != st->codecpar->sample_rate || !st->codecpar->frame_size)
3781  trust_default_duration = 0;
3782  }
3783  }
3784 
3785  if (!block_duration && trust_default_duration)
3786  block_duration = track->default_duration * laces / matroska->time_scale;
3787 
3788  if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
3789  track->end_timecode =
3790  FFMAX(track->end_timecode, timecode + block_duration);
3791 
3792  for (n = 0; n < laces; n++) {
3793  int64_t lace_duration = block_duration*(n+1) / laces - block_duration*n / laces;
3794  uint8_t *out_data = data;
3795  int out_size = lace_size[n];
3796 
3797  if (track->needs_decoding) {
3798  res = matroska_decode_buffer(&out_data, &out_size, track);
3799  if (res < 0)
3800  return res;
3801  /* Given that we are here means that out_data is no longer
3802  * owned by buf, so set it to NULL. This depends upon
3803  * zero-length header removal compression being ignored. */
3804  av_assert1(out_data != data);
3805  buf = NULL;
3806  }
3807 
3808  if (track->audio.buf) {
3809  res = matroska_parse_rm_audio(matroska, track, st,
3810  out_data, out_size,
3811  timecode, pos);
3812  if (!buf)
3813  av_free(out_data);
3814  if (res)
3815  return res;
3816  } else if (st->codecpar->codec_id == AV_CODEC_ID_WEBVTT) {
3817  res = matroska_parse_webvtt(matroska, track, st,
3818  out_data, out_size,
3819  timecode, lace_duration,
3820  pos);
3821  if (!buf)
3822  av_free(out_data);
3823  if (res)
3824  return res;
3825  } else {
3826  res = matroska_parse_frame(matroska, track, st, buf, out_data,
3827  out_size, timecode, lace_duration,
3828  pos, !n ? is_keyframe : 0,
3829  additional, additional_id, additional_size,
3830  discard_padding);
3831  if (res)
3832  return res;
3833  }
3834 
3835  if (timecode != AV_NOPTS_VALUE)
3836  timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
3837  data += lace_size[n];
3838  }
3839 
3840  return 0;
3841 }
3842 
3843 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
3844 {
3845  MatroskaCluster *cluster = &matroska->current_cluster;
3846  MatroskaBlock *block = &cluster->block;
3847  int res;
3848 
3849  av_assert0(matroska->num_levels <= 2);
3850 
3851  if (matroska->num_levels == 1) {
3852  res = ebml_parse(matroska, matroska_segment, NULL);
3853 
3854  if (res == 1) {
3855  /* Found a cluster: subtract the size of the ID already read. */
3856  cluster->pos = avio_tell(matroska->ctx->pb) - 4;
3857 
3858  res = ebml_parse(matroska, matroska_cluster_enter, cluster);
3859  if (res < 0)
3860  return res;
3861  }
3862  }
3863 
3864  if (matroska->num_levels == 2) {
3865  /* We are inside a cluster. */
3866  res = ebml_parse(matroska, matroska_cluster_parsing, cluster);
3867 
3868  if (res >= 0 && block->bin.size > 0) {
3869  int is_keyframe = block->non_simple ? block->reference.count == 0 : -1;
3870  uint8_t* additional = block->additional.size > 0 ?
3871  block->additional.data : NULL;
3872 
3873  res = matroska_parse_block(matroska, block->bin.buf, block->bin.data,
3874  block->bin.size, block->bin.pos,
3875  cluster->timecode, block->duration,
3876  is_keyframe, additional, block->additional_id,
3877  block->additional.size, cluster->pos,
3878  block->discard_padding);
3879  }
3880 
3882  memset(block, 0, sizeof(*block));
3883  } else if (!matroska->num_levels) {
3884  if (!avio_feof(matroska->ctx->pb)) {
3885  avio_r8(matroska->ctx->pb);
3886  if (!avio_feof(matroska->ctx->pb)) {
3887  av_log(matroska->ctx, AV_LOG_WARNING, "File extends beyond "
3888  "end of segment.\n");
3889  return AVERROR_INVALIDDATA;
3890  }
3891  }
3892  matroska->done = 1;
3893  return AVERROR_EOF;
3894  }
3895 
3896  return res;
3897 }
3898 
3900 {
3901  MatroskaDemuxContext *matroska = s->priv_data;
3902  int ret = 0;
3903 
3904  if (matroska->resync_pos == -1) {
3905  // This can only happen if generic seeking has been used.
3906  matroska->resync_pos = avio_tell(s->pb);
3907  }
3908 
3909  while (matroska_deliver_packet(matroska, pkt)) {
3910  if (matroska->done)
3911  return (ret < 0) ? ret : AVERROR_EOF;
3912  if (matroska_parse_cluster(matroska) < 0 && !matroska->done)
3913  ret = matroska_resync(matroska, matroska->resync_pos);
3914  }
3915 
3916  return 0;
3917 }
3918 
3919 static int matroska_read_seek(AVFormatContext *s, int stream_index,
3920  int64_t timestamp, int flags)
3921 {
3922  MatroskaDemuxContext *matroska = s->priv_data;
3923  MatroskaTrack *tracks = NULL;
3924  AVStream *st = s->streams[stream_index];
3925  FFStream *const sti = ffstream(st);
3926  int i, index;
3927 
3928  /* Parse the CUES now since we need the index data to seek. */
3929  if (matroska->cues_parsing_deferred > 0) {
3930  matroska->cues_parsing_deferred = 0;
3931  matroska_parse_cues(matroska);
3932  }
3933 
3934  if (!sti->nb_index_entries)
3935  goto err;
3936  timestamp = FFMAX(timestamp, sti->index_entries[0].timestamp);
3937 
3938  if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 ||
3939  index == sti->nb_index_entries - 1) {
3940  matroska_reset_status(matroska, 0, sti->index_entries[sti->nb_index_entries - 1].pos);
3941  while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0 ||
3942  index == sti->nb_index_entries - 1) {
3943  matroska_clear_queue(matroska);
3944  if (matroska_parse_cluster(matroska) < 0)
3945  break;
3946  }
3947  }
3948 
3949  matroska_clear_queue(matroska);
3950  if (index < 0 || (matroska->cues_parsing_deferred < 0 &&
3951  index == sti->nb_index_entries - 1))
3952  goto err;
3953 
3954  tracks = matroska->tracks.elem;
3955  for (i = 0; i < matroska->tracks.nb_elem; i++) {
3956  tracks[i].audio.pkt_cnt = 0;
3957  tracks[i].audio.sub_packet_cnt = 0;
3958  tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
3959  tracks[i].end_timecode = 0;
3960  }
3961 
3962  /* We seek to a level 1 element, so set the appropriate status. */
3963  matroska_reset_status(matroska, 0, sti->index_entries[index].pos);
3964  if (flags & AVSEEK_FLAG_ANY) {
3965  sti->skip_to_keyframe = 0;
3966  matroska->skip_to_timecode = timestamp;
3967  } else {
3968  sti->skip_to_keyframe = 1;
3969  matroska->skip_to_timecode = sti->index_entries[index].timestamp;
3970  }
3971  matroska->skip_to_keyframe = 1;
3972  matroska->done = 0;
3974  return 0;
3975 err:
3976  // slightly hackish but allows proper fallback to
3977  // the generic seeking code.
3978  matroska_reset_status(matroska, 0, -1);
3979  matroska->resync_pos = -1;
3980  matroska_clear_queue(matroska);
3982  matroska->skip_to_keyframe = 0;
3983  matroska->done = 0;
3984  return -1;
3985 }
3986 
3988 {
3989  MatroskaDemuxContext *matroska = s->priv_data;
3990  MatroskaTrack *tracks = matroska->tracks.elem;
3991  int n;
3992 
3993  matroska_clear_queue(matroska);
3994 
3995  for (n = 0; n < matroska->tracks.nb_elem; n++)
3996  if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
3997  av_freep(&tracks[n].audio.buf);
3998  ebml_free(matroska_segment, matroska);
3999 
4000  return 0;
4001 }
4002 
4003 #if CONFIG_WEBM_DASH_MANIFEST_DEMUXER
4004 typedef struct {
4005  int64_t start_time_ns;
4006  int64_t end_time_ns;
4007  int64_t start_offset;
4008  int64_t end_offset;
4009 } CueDesc;
4010 
4011 /* This function searches all the Cues and returns the CueDesc corresponding to
4012  * the timestamp ts. Returned CueDesc will be such that start_time_ns <= ts <
4013  * end_time_ns. All 4 fields will be set to -1 if ts >= file's duration or
4014  * if an error occurred.
4015  */
4016 static CueDesc get_cue_desc(AVFormatContext *s, int64_t ts, int64_t cues_start) {
4017  MatroskaDemuxContext *matroska = s->priv_data;
4018  FFStream *const sti = ffstream(s->streams[0]);
4019  AVIndexEntry *const index_entries = sti->index_entries;
4020  int nb_index_entries = sti->nb_index_entries;
4021  CueDesc cue_desc;
4022  int i;
4023 
4024  if (ts >= (int64_t)(matroska->duration * matroska->time_scale))
4025  return (CueDesc) {-1, -1, -1, -1};
4026  for (i = 1; i < nb_index_entries; i++) {
4027  if (index_entries[i - 1].timestamp * matroska->time_scale <= ts &&
4028  index_entries[i].timestamp * matroska->time_scale > ts) {
4029  break;
4030  }
4031  }
4032  --i;
4033  if (index_entries[i].timestamp > matroska->duration)
4034  return (CueDesc) {-1, -1, -1, -1};
4035  cue_desc.start_time_ns = index_entries[i].timestamp * matroska->time_scale;
4036  cue_desc.start_offset = index_entries[i].pos - matroska->segment_start;
4037  if (i != nb_index_entries - 1) {
4038  cue_desc.end_time_ns = index_entries[i + 1].timestamp * matroska->time_scale;
4039  cue_desc.end_offset = index_entries[i + 1].pos - matroska->segment_start;
4040  } else {
4041  cue_desc.end_time_ns = matroska->duration * matroska->time_scale;
4042  // FIXME: this needs special handling for files where Cues appear
4043  // before Clusters. the current logic assumes Cues appear after
4044  // Clusters.
4045  cue_desc.end_offset = cues_start - matroska->segment_start;
4046  }
4047  return cue_desc;
4048 }
4049 
4050 static int webm_clusters_start_with_keyframe(AVFormatContext *s)
4051 {
4052  MatroskaDemuxContext *matroska = s->priv_data;
4053  AVStream *const st = s->streams[0];
4054  FFStream *const sti = ffstream(st);
4055  uint32_t id = matroska->current_id;
4056  int64_t cluster_pos, before_pos;
4057  int index, rv = 1;
4058 
4059  if (sti->nb_index_entries <= 0)
4060  return 0;
4061 
4062  // seek to the first cluster using cues.
4063  index = av_index_search_timestamp(st, 0, 0);
4064  if (index < 0)
4065  return 0;
4066  cluster_pos = sti->index_entries[index].pos;
4067  before_pos = avio_tell(s->pb);
4068  while (1) {
4069  uint64_t cluster_id, cluster_length;
4070  int read;
4071  AVPacket *pkt;
4072  avio_seek(s->pb, cluster_pos, SEEK_SET);
4073  // read cluster id and length
4074  read = ebml_read_num(matroska, matroska->ctx->pb, 4, &cluster_id, 1);
4075  if (read < 0 || cluster_id != 0xF43B675) // done with all clusters
4076  break;
4077  read = ebml_read_length(matroska, matroska->ctx->pb, &cluster_length);
4078  if (read < 0)
4079  break;
4080 
4081  matroska_reset_status(matroska, 0, cluster_pos);
4082  matroska_clear_queue(matroska);
4083  if (matroska_parse_cluster(matroska) < 0 ||
4084  !matroska->queue.head) {
4085  break;
4086  }
4087  pkt = &matroska->queue.head->pkt;
4088  // 4 + read is the length of the cluster id and the cluster length field.
4089  cluster_pos += 4 + read + cluster_length;
4090  if (!(pkt->flags & AV_PKT_FLAG_KEY)) {
4091  rv = 0;
4092  break;
4093  }
4094  }
4095 
4096  /* Restore the status after matroska_read_header: */
4097  matroska_reset_status(matroska, id, before_pos);
4098 
4099  return rv;
4100 }
4101 
4102 static int buffer_size_after_time_downloaded(int64_t time_ns, double search_sec, int64_t bps,
4103  double min_buffer, double* buffer,
4104  double* sec_to_download, AVFormatContext *s,
4105  int64_t cues_start)
4106 {
4107  double nano_seconds_per_second = 1000000000.0;
4108  double time_sec = time_ns / nano_seconds_per_second;
4109  int rv = 0;
4110  int64_t time_to_search_ns = (int64_t)(search_sec * nano_seconds_per_second);
4111  int64_t end_time_ns = time_ns + time_to_search_ns;
4112  double sec_downloaded = 0.0;
4113  CueDesc desc_curr = get_cue_desc(s, time_ns, cues_start);
4114  if (desc_curr.start_time_ns == -1)
4115  return -1;
4116  *sec_to_download = 0.0;
4117 
4118  // Check for non cue start time.
4119  if (time_ns > desc_curr.start_time_ns) {
4120  int64_t cue_nano = desc_curr.end_time_ns - time_ns;
4121  double percent = (double)(cue_nano) / (desc_curr.end_time_ns - desc_curr.start_time_ns);
4122  double cueBytes = (desc_curr.end_offset - desc_curr.start_offset) * percent;
4123  double timeToDownload = (cueBytes * 8.0) / bps;
4124 
4125  sec_downloaded += (cue_nano / nano_seconds_per_second) - timeToDownload;
4126  *sec_to_download += timeToDownload;
4127 
4128  // Check if the search ends within the first cue.
4129  if (desc_curr.end_time_ns >= end_time_ns) {
4130  double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
4131  double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
4132  sec_downloaded = percent_to_sub * sec_downloaded;
4133  *sec_to_download = percent_to_sub * *sec_to_download;
4134  }
4135 
4136  if ((sec_downloaded + *buffer) <= min_buffer) {
4137  return 1;
4138  }
4139 
4140  // Get the next Cue.
4141  desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
4142  }
4143 
4144  while (desc_curr.start_time_ns != -1) {
4145  int64_t desc_bytes = desc_curr.end_offset - desc_curr.start_offset;
4146  int64_t desc_ns = desc_curr.end_time_ns - desc_curr.start_time_ns;
4147  double desc_sec = desc_ns / nano_seconds_per_second;
4148  double bits = (desc_bytes * 8.0);
4149  double time_to_download = bits / bps;
4150 
4151  sec_downloaded += desc_sec - time_to_download;
4152  *sec_to_download += time_to_download;
4153 
4154  if (desc_curr.end_time_ns >= end_time_ns) {
4155  double desc_end_time_sec = desc_curr.end_time_ns / nano_seconds_per_second;
4156  double percent_to_sub = search_sec / (desc_end_time_sec - time_sec);
4157  sec_downloaded = percent_to_sub * sec_downloaded;
4158  *sec_to_download = percent_to_sub * *sec_to_download;
4159 
4160  if ((sec_downloaded + *buffer) <= min_buffer)
4161  rv = 1;
4162  break;
4163  }
4164 
4165  if ((sec_downloaded + *buffer) <= min_buffer) {
4166  rv = 1;
4167  break;
4168  }
4169 
4170  desc_curr = get_cue_desc(s, desc_curr.end_time_ns, cues_start);
4171  }
4172  *buffer = *buffer + sec_downloaded;
4173  return rv;
4174 }
4175 
4176 /* This function computes the bandwidth of the WebM file with the help of
4177  * buffer_size_after_time_downloaded() function. Both of these functions are
4178  * adapted from WebM Tools project and are adapted to work with FFmpeg's
4179  * Matroska parsing mechanism.
4180  *
4181  * Returns the bandwidth of the file on success; -1 on error.
4182  * */
4183 static int64_t webm_dash_manifest_compute_bandwidth(AVFormatContext *s, int64_t cues_start)
4184 {
4185  MatroskaDemuxContext *matroska = s->priv_data;
4186  AVStream *st = s->streams[0];
4187  FFStream *const sti = ffstream(st);
4188  double bandwidth = 0.0;
4189 
4190  for (int i = 0; i < sti->nb_index_entries; i++) {
4191  int64_t prebuffer_ns = 1000000000;
4192  int64_t time_ns = sti->index_entries[i].timestamp * matroska->time_scale;
4193  double nano_seconds_per_second = 1000000000.0;
4194  int64_t prebuffered_ns = time_ns + prebuffer_ns;
4195  double prebuffer_bytes = 0.0;
4196  int64_t temp_prebuffer_ns = prebuffer_ns;
4197  int64_t pre_bytes, pre_ns;
4198  double pre_sec, prebuffer, bits_per_second;
4199  CueDesc desc_beg = get_cue_desc(s, time_ns, cues_start);
4200 
4201  // Start with the first Cue.
4202  CueDesc desc_end = desc_beg;
4203 
4204  // Figure out how much data we have downloaded for the prebuffer. This will
4205  // be used later to adjust the bits per sample to try.
4206  while (desc_end.start_time_ns != -1 && desc_end.end_time_ns < prebuffered_ns) {
4207  // Prebuffered the entire Cue.
4208  prebuffer_bytes += desc_end.end_offset - desc_end.start_offset;
4209  temp_prebuffer_ns -= desc_end.end_time_ns - desc_end.start_time_ns;
4210  desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4211  }
4212  if (desc_end.start_time_ns == -1) {
4213  // The prebuffer is larger than the duration.
4214  if (matroska->duration * matroska->time_scale >= prebuffered_ns)
4215  return -1;
4216  bits_per_second = 0.0;
4217  } else {
4218  // The prebuffer ends in the last Cue. Estimate how much data was
4219  // prebuffered.
4220  pre_bytes = desc_end.end_offset - desc_end.start_offset;
4221  pre_ns = desc_end.end_time_ns - desc_end.start_time_ns;
4222  if (pre_ns <= 0)
4223  return -1;
4224  pre_sec = pre_ns / nano_seconds_per_second;
4225  prebuffer_bytes +=
4226  pre_bytes * ((temp_prebuffer_ns / nano_seconds_per_second) / pre_sec);
4227 
4228  prebuffer = prebuffer_ns / nano_seconds_per_second;
4229 
4230  // Set this to 0.0 in case our prebuffer buffers the entire video.
4231  bits_per_second = 0.0;
4232  do {
4233  int64_t desc_bytes = desc_end.end_offset - desc_beg.start_offset;
4234  int64_t desc_ns = desc_end.end_time_ns - desc_beg.start_time_ns;
4235  double desc_sec, calc_bits_per_second, percent, mod_bits_per_second;
4236  if (desc_bytes <= 0)
4237  return -1;
4238 
4239  desc_sec = desc_ns / nano_seconds_per_second;
4240  calc_bits_per_second = (desc_bytes * 8) / desc_sec;
4241 
4242  // Drop the bps by the percentage of bytes buffered.
4243  percent = (desc_bytes - prebuffer_bytes) / desc_bytes;
4244  mod_bits_per_second = calc_bits_per_second * percent;
4245 
4246  if (prebuffer < desc_sec) {
4247  double search_sec =
4248  (double)(matroska->duration * matroska->time_scale) / nano_seconds_per_second;
4249 
4250  // Add 1 so the bits per second should be a little bit greater than file
4251  // datarate.
4252  int64_t bps = (int64_t)(mod_bits_per_second) + 1;
4253  const double min_buffer = 0.0;
4254  double buffer = prebuffer;
4255  double sec_to_download = 0.0;
4256 
4257  int rv = buffer_size_after_time_downloaded(prebuffered_ns, search_sec, bps,
4258  min_buffer, &buffer, &sec_to_download,
4259  s, cues_start);
4260  if (rv < 0) {
4261  return -1;
4262  } else if (rv == 0) {
4263  bits_per_second = (double)(bps);
4264  break;
4265  }
4266  }
4267 
4268  desc_end = get_cue_desc(s, desc_end.end_time_ns, cues_start);
4269  } while (desc_end.start_time_ns != -1);
4270  }
4271  if (bandwidth < bits_per_second) bandwidth = bits_per_second;
4272  }
4273  return (int64_t)bandwidth;
4274 }
4275 
4276 static int webm_dash_manifest_cues(AVFormatContext *s, int64_t init_range)
4277 {
4278  MatroskaDemuxContext *matroska = s->priv_data;
4279  EbmlList *seekhead_list = &matroska->seekhead;
4280  MatroskaSeekhead *seekhead = seekhead_list->elem;
4281  AVStream *const st = s->streams[0];
4282  FFStream *const sti = ffstream(st);
4283  AVBPrint bprint;
4284  char *buf;
4285  int64_t cues_start = -1, cues_end = -1, before_pos, bandwidth;
4286  int i;
4287  int ret;
4288 
4289  // determine cues start and end positions
4290  for (i = 0; i < seekhead_list->nb_elem; i++)
4291  if (seekhead[i].id == MATROSKA_ID_CUES)
4292  break;
4293 
4294  if (i >= seekhead_list->nb_elem) return -1;
4295 
4296  before_pos = avio_tell(matroska->ctx->pb);
4297  cues_start = seekhead[i].pos + matroska->segment_start;
4298  if (avio_seek(matroska->ctx->pb, cues_start, SEEK_SET) == cues_start) {
4299  // cues_end is computed as cues_start + cues_length + length of the
4300  // Cues element ID (i.e. 4) + EBML length of the Cues element.
4301  // cues_end is inclusive and the above sum is reduced by 1.
4302  uint64_t cues_length, cues_id;
4303  int bytes_read;
4304  bytes_read = ebml_read_num (matroska, matroska->ctx->pb, 4, &cues_id, 1);
4305  if (bytes_read < 0 || cues_id != (MATROSKA_ID_CUES & 0xfffffff))
4306  return bytes_read < 0 ? bytes_read : AVERROR_INVALIDDATA;
4307  bytes_read = ebml_read_length(matroska, matroska->ctx->pb, &cues_length);
4308  if (bytes_read < 0)
4309  return bytes_read;
4310  cues_end = cues_start + 4 + bytes_read + cues_length - 1;
4311  }
4312  avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
4313  if (cues_start == -1 || cues_end == -1) return -1;
4314 
4315  // parse the cues
4316  matroska_parse_cues(matroska);
4317 
4318  if (!sti->nb_index_entries)
4319  return AVERROR_INVALIDDATA;
4320 
4321  // cues start
4322  av_dict_set_int(&s->streams[0]->metadata, CUES_START, cues_start, 0);
4323 
4324  // cues end
4325  av_dict_set_int(&s->streams[0]->metadata, CUES_END, cues_end, 0);
4326 
4327  // if the file has cues at the start, fix up the init range so that
4328  // it does not include it
4329  if (cues_start <= init_range)
4330  av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, cues_start - 1, 0);
4331 
4332  // bandwidth
4333  bandwidth = webm_dash_manifest_compute_bandwidth(s, cues_start);
4334  if (bandwidth < 0) return -1;
4335  av_dict_set_int(&s->streams[0]->metadata, BANDWIDTH, bandwidth, 0);
4336 
4337  // check if all clusters start with key frames
4338  av_dict_set_int(&s->streams[0]->metadata, CLUSTER_KEYFRAME, webm_clusters_start_with_keyframe(s), 0);
4339 
4340  // Store cue point timestamps as a comma separated list
4341  // for checking subsegment alignment in the muxer.
4343  for (int i = 0; i < sti->nb_index_entries; i++)
4344  av_bprintf(&bprint, "%" PRId64",", sti->index_entries[i].timestamp);
4345  if (!av_bprint_is_complete(&bprint)) {
4346  av_bprint_finalize(&bprint, NULL);
4347  return AVERROR(ENOMEM);
4348  }
4349  // Remove the trailing ','
4350  bprint.str[--bprint.len] = '\0';
4351  if ((ret = av_bprint_finalize(&bprint, &buf)) < 0)
4352  return ret;
4353  av_dict_set(&s->streams[0]->metadata, CUE_TIMESTAMPS,
4355 
4356  return 0;
4357 }
4358 
4359 static int webm_dash_manifest_read_header(AVFormatContext *s)
4360 {
4361  char *buf;
4362  int ret = matroska_read_header(s);
4363  int64_t init_range;
4364  MatroskaTrack *tracks;
4365  MatroskaDemuxContext *matroska = s->priv_data;
4366  if (ret) {
4367  av_log(s, AV_LOG_ERROR, "Failed to read file headers\n");
4368  return -1;
4369  }
4370  if (!matroska->tracks.nb_elem || !s->nb_streams) {
4371  av_log(s, AV_LOG_ERROR, "No track found\n");
4372  return AVERROR_INVALIDDATA;
4373  }
4374 
4375  if (!matroska->is_live) {
4376  buf = av_asprintf("%g", matroska->duration);
4377  if (!buf)
4378  return AVERROR(ENOMEM);
4379  av_dict_set(&s->streams[0]->metadata, DURATION,
4381 
4382  // initialization range
4383  // 5 is the offset of Cluster ID.
4384  init_range = avio_tell(s->pb) - 5;
4385  av_dict_set_int(&s->streams[0]->metadata, INITIALIZATION_RANGE, init_range, 0);
4386  }
4387 
4388  // basename of the file
4389  buf = strrchr(s->url, '/');
4390  av_dict_set(&s->streams[0]->metadata, FILENAME, buf ? ++buf : s->url, 0);
4391 
4392  // track number
4393  tracks = matroska->tracks.elem;
4394  av_dict_set_int(&s->streams[0]->metadata, TRACK_NUMBER, tracks[0].num, 0);
4395 
4396  // parse the cues and populate Cue related fields
4397  if (!matroska->is_live) {
4398  ret = webm_dash_manifest_cues(s, init_range);
4399  if (