Merge commit '581281e242609a222233a2e5538b89dfb88fb18e'
[ffmpeg.git] / libavformat / matroskadec.c
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 <stdio.h>
32 #include "avformat.h"
33 #include "internal.h"
34 #include "avio_internal.h"
35 /* For ff_codec_get_id(). */
36 #include "riff.h"
37 #include "isom.h"
38 #include "rm.h"
39 #include "matroska.h"
40 #include "libavcodec/mpeg4audio.h"
41 #include "libavutil/intfloat.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/avstring.h"
44 #include "libavutil/lzo.h"
45 #include "libavutil/dict.h"
46 #if CONFIG_ZLIB
47 #include <zlib.h>
48 #endif
49 #if CONFIG_BZLIB
50 #include <bzlib.h>
51 #endif
52
53 typedef enum {
54     EBML_NONE,
55     EBML_UINT,
56     EBML_FLOAT,
57     EBML_STR,
58     EBML_UTF8,
59     EBML_BIN,
60     EBML_NEST,
61     EBML_PASS,
62     EBML_STOP,
63     EBML_TYPE_COUNT
64 } EbmlType;
65
66 typedef const struct EbmlSyntax {
67     uint32_t id;
68     EbmlType type;
69     int list_elem_size;
70     int data_offset;
71     union {
72         uint64_t    u;
73         double      f;
74         const char *s;
75         const struct EbmlSyntax *n;
76     } def;
77 } EbmlSyntax;
78
79 typedef struct {
80     int nb_elem;
81     void *elem;
82 } EbmlList;
83
84 typedef struct {
85     int      size;
86     uint8_t *data;
87     int64_t  pos;
88 } EbmlBin;
89
90 typedef struct {
91     uint64_t version;
92     uint64_t max_size;
93     uint64_t id_length;
94     char    *doctype;
95     uint64_t doctype_version;
96 } Ebml;
97
98 typedef struct {
99     uint64_t algo;
100     EbmlBin  settings;
101 } MatroskaTrackCompression;
102
103 typedef struct {
104     uint64_t scope;
105     uint64_t type;
106     MatroskaTrackCompression compression;
107 } MatroskaTrackEncoding;
108
109 typedef struct {
110     double   frame_rate;
111     uint64_t display_width;
112     uint64_t display_height;
113     uint64_t pixel_width;
114     uint64_t pixel_height;
115     EbmlBin color_space;
116     uint64_t stereo_mode;
117 } MatroskaTrackVideo;
118
119 typedef struct {
120     double   samplerate;
121     double   out_samplerate;
122     uint64_t bitdepth;
123     uint64_t channels;
124
125     /* real audio header (extracted from extradata) */
126     int      coded_framesize;
127     int      sub_packet_h;
128     int      frame_size;
129     int      sub_packet_size;
130     int      sub_packet_cnt;
131     int      pkt_cnt;
132     uint64_t buf_timecode;
133     uint8_t *buf;
134 } MatroskaTrackAudio;
135
136 typedef struct {
137     uint64_t uid;
138     uint64_t type;
139 } MatroskaTrackPlane;
140
141 typedef struct {
142     EbmlList combine_planes;
143 } MatroskaTrackOperation;
144
145 typedef struct {
146     uint64_t num;
147     uint64_t uid;
148     uint64_t type;
149     char    *name;
150     char    *codec_id;
151     EbmlBin  codec_priv;
152     char    *language;
153     double time_scale;
154     uint64_t default_duration;
155     uint64_t flag_default;
156     uint64_t flag_forced;
157     MatroskaTrackVideo video;
158     MatroskaTrackAudio audio;
159     MatroskaTrackOperation operation;
160     EbmlList encodings;
161
162     AVStream *stream;
163     int64_t end_timecode;
164     int ms_compat;
165 } MatroskaTrack;
166
167 typedef struct {
168     uint64_t uid;
169     char *filename;
170     char *mime;
171     EbmlBin bin;
172
173     AVStream *stream;
174 } MatroskaAttachement;
175
176 typedef struct {
177     uint64_t start;
178     uint64_t end;
179     uint64_t uid;
180     char    *title;
181
182     AVChapter *chapter;
183 } MatroskaChapter;
184
185 typedef struct {
186     uint64_t track;
187     uint64_t pos;
188 } MatroskaIndexPos;
189
190 typedef struct {
191     uint64_t time;
192     EbmlList pos;
193 } MatroskaIndex;
194
195 typedef struct {
196     char *name;
197     char *string;
198     char *lang;
199     uint64_t def;
200     EbmlList sub;
201 } MatroskaTag;
202
203 typedef struct {
204     char    *type;
205     uint64_t typevalue;
206     uint64_t trackuid;
207     uint64_t chapteruid;
208     uint64_t attachuid;
209 } MatroskaTagTarget;
210
211 typedef struct {
212     MatroskaTagTarget target;
213     EbmlList tag;
214 } MatroskaTags;
215
216 typedef struct {
217     uint64_t id;
218     uint64_t pos;
219 } MatroskaSeekhead;
220
221 typedef struct {
222     uint64_t start;
223     uint64_t length;
224 } MatroskaLevel;
225
226 typedef struct {
227     uint64_t timecode;
228     EbmlList blocks;
229 } MatroskaCluster;
230
231 typedef struct {
232     AVFormatContext *ctx;
233
234     /* EBML stuff */
235     int num_levels;
236     MatroskaLevel levels[EBML_MAX_DEPTH];
237     int level_up;
238     uint32_t current_id;
239
240     uint64_t time_scale;
241     double   duration;
242     char    *title;
243     EbmlBin date_utc;
244     EbmlList tracks;
245     EbmlList attachments;
246     EbmlList chapters;
247     EbmlList index;
248     EbmlList tags;
249     EbmlList seekhead;
250
251     /* byte position of the segment inside the stream */
252     int64_t segment_start;
253
254     /* the packet queue */
255     AVPacket **packets;
256     int num_packets;
257     AVPacket *prev_pkt;
258
259     int done;
260
261     /* What to skip before effectively reading a packet. */
262     int skip_to_keyframe;
263     uint64_t skip_to_timecode;
264
265     /* File has a CUES element, but we defer parsing until it is needed. */
266     int cues_parsing_deferred;
267
268     int current_cluster_num_blocks;
269     int64_t current_cluster_pos;
270     MatroskaCluster current_cluster;
271
272     /* File has SSA subtitles which prevent incremental cluster parsing. */
273     int contains_ssa;
274 } MatroskaDemuxContext;
275
276 typedef struct {
277     uint64_t duration;
278     int64_t  reference;
279     uint64_t non_simple;
280     EbmlBin  bin;
281 } MatroskaBlock;
282
283 static EbmlSyntax ebml_header[] = {
284     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
285     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
286     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
287     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
288     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
289     { EBML_ID_EBMLVERSION,            EBML_NONE },
290     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
291     { 0 }
292 };
293
294 static EbmlSyntax ebml_syntax[] = {
295     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
296     { 0 }
297 };
298
299 static EbmlSyntax matroska_info[] = {
300     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
301     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
302     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
303     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
304     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
305     { MATROSKA_ID_DATEUTC,            EBML_BIN,  0, offsetof(MatroskaDemuxContext,date_utc) },
306     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
307     { 0 }
308 };
309
310 static EbmlSyntax matroska_track_video[] = {
311     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
312     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
313     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
314     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
315     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
316     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_BIN,  0, offsetof(MatroskaTrackVideo,color_space) },
317     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,stereo_mode) },
318     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
319     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
320     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
321     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
322     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
323     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
324     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
325     { 0 }
326 };
327
328 static EbmlSyntax matroska_track_audio[] = {
329     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
330     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
331     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
332     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
333     { 0 }
334 };
335
336 static EbmlSyntax matroska_track_encoding_compression[] = {
337     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
338     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
339     { 0 }
340 };
341
342 static EbmlSyntax matroska_track_encoding[] = {
343     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
344     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
345     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
346     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
347     { 0 }
348 };
349
350 static EbmlSyntax matroska_track_encodings[] = {
351     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
352     { 0 }
353 };
354
355 static EbmlSyntax matroska_track_plane[] = {
356     { MATROSKA_ID_TRACKPLANEUID,  EBML_UINT, 0, offsetof(MatroskaTrackPlane,uid) },
357     { MATROSKA_ID_TRACKPLANETYPE, EBML_UINT, 0, offsetof(MatroskaTrackPlane,type) },
358     { 0 }
359 };
360
361 static EbmlSyntax matroska_track_combine_planes[] = {
362     { MATROSKA_ID_TRACKPLANE, EBML_NEST, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n=matroska_track_plane} },
363     { 0 }
364 };
365
366 static EbmlSyntax matroska_track_operation[] = {
367     { MATROSKA_ID_TRACKCOMBINEPLANES, EBML_NEST, 0, 0, {.n=matroska_track_combine_planes} },
368     { 0 }
369 };
370
371 static EbmlSyntax matroska_track[] = {
372     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
373     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
374     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
375     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
376     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
377     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
378     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
379     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
380     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
381     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
382     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_UINT, 0, offsetof(MatroskaTrack,flag_forced), {.u=0} },
383     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
384     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
385     { MATROSKA_ID_TRACKOPERATION,       EBML_NEST, 0, offsetof(MatroskaTrack,operation), {.n=matroska_track_operation} },
386     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
387     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
388     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
389     { MATROSKA_ID_CODECNAME,            EBML_NONE },
390     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
391     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
392     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
393     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
394     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
395     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
396     { 0 }
397 };
398
399 static EbmlSyntax matroska_tracks[] = {
400     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
401     { 0 }
402 };
403
404 static EbmlSyntax matroska_attachment[] = {
405     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
406     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
407     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
408     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
409     { MATROSKA_ID_FILEDESC,           EBML_NONE },
410     { 0 }
411 };
412
413 static EbmlSyntax matroska_attachments[] = {
414     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
415     { 0 }
416 };
417
418 static EbmlSyntax matroska_chapter_display[] = {
419     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
420     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
421     { 0 }
422 };
423
424 static EbmlSyntax matroska_chapter_entry[] = {
425     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
426     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
427     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
428     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
429     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
430     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
431     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
432     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
433     { 0 }
434 };
435
436 static EbmlSyntax matroska_chapter[] = {
437     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
438     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
439     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
440     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
441     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
442     { 0 }
443 };
444
445 static EbmlSyntax matroska_chapters[] = {
446     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
447     { 0 }
448 };
449
450 static EbmlSyntax matroska_index_pos[] = {
451     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
452     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
453     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
454     { 0 }
455 };
456
457 static EbmlSyntax matroska_index_entry[] = {
458     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
459     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
460     { 0 }
461 };
462
463 static EbmlSyntax matroska_index[] = {
464     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
465     { 0 }
466 };
467
468 static EbmlSyntax matroska_simpletag[] = {
469     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
470     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
471     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
472     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
473     { MATROSKA_ID_TAGDEFAULT_BUG,     EBML_UINT, 0, offsetof(MatroskaTag,def) },
474     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
475     { 0 }
476 };
477
478 static EbmlSyntax matroska_tagtargets[] = {
479     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
480     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
481     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
482     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
483     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
484     { 0 }
485 };
486
487 static EbmlSyntax matroska_tag[] = {
488     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
489     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
490     { 0 }
491 };
492
493 static EbmlSyntax matroska_tags[] = {
494     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
495     { 0 }
496 };
497
498 static EbmlSyntax matroska_seekhead_entry[] = {
499     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
500     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
501     { 0 }
502 };
503
504 static EbmlSyntax matroska_seekhead[] = {
505     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
506     { 0 }
507 };
508
509 static EbmlSyntax matroska_segment[] = {
510     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
511     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
512     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
513     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
514     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
515     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
516     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
517     { MATROSKA_ID_CLUSTER,        EBML_STOP },
518     { 0 }
519 };
520
521 static EbmlSyntax matroska_segments[] = {
522     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
523     { 0 }
524 };
525
526 static EbmlSyntax matroska_blockgroup[] = {
527     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
528     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
529     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration) },
530     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
531     { 1,                          EBML_UINT, 0, offsetof(MatroskaBlock,non_simple), {.u=1} },
532     { 0 }
533 };
534
535 static EbmlSyntax matroska_cluster[] = {
536     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
537     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
538     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
539     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
540     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
541     { 0 }
542 };
543
544 static EbmlSyntax matroska_clusters[] = {
545     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
546     { MATROSKA_ID_INFO,           EBML_NONE },
547     { MATROSKA_ID_CUES,           EBML_NONE },
548     { MATROSKA_ID_TAGS,           EBML_NONE },
549     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
550     { 0 }
551 };
552
553 static EbmlSyntax matroska_cluster_incremental_parsing[] = {
554     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
555     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
556     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
557     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
558     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
559     { MATROSKA_ID_INFO,           EBML_NONE },
560     { MATROSKA_ID_CUES,           EBML_NONE },
561     { MATROSKA_ID_TAGS,           EBML_NONE },
562     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
563     { MATROSKA_ID_CLUSTER,        EBML_STOP },
564     { 0 }
565 };
566
567 static EbmlSyntax matroska_cluster_incremental[] = {
568     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
569     { MATROSKA_ID_BLOCKGROUP,     EBML_STOP },
570     { MATROSKA_ID_SIMPLEBLOCK,    EBML_STOP },
571     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
572     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
573     { 0 }
574 };
575
576 static EbmlSyntax matroska_clusters_incremental[] = {
577     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster_incremental} },
578     { MATROSKA_ID_INFO,           EBML_NONE },
579     { MATROSKA_ID_CUES,           EBML_NONE },
580     { MATROSKA_ID_TAGS,           EBML_NONE },
581     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
582     { 0 }
583 };
584
585 static const char *const matroska_doctypes[] = { "matroska", "webm" };
586
587 static int matroska_resync(MatroskaDemuxContext *matroska, int64_t last_pos)
588 {
589     AVIOContext *pb = matroska->ctx->pb;
590     uint32_t id;
591     matroska->current_id = 0;
592     matroska->num_levels = 0;
593
594     // seek to next position to resync from
595     if (avio_seek(pb, last_pos + 1, SEEK_SET) < 0 || avio_tell(pb) <= last_pos)
596         goto eof;
597
598     id = avio_rb32(pb);
599
600     // try to find a toplevel element
601     while (!url_feof(pb)) {
602         if (id == MATROSKA_ID_INFO || id == MATROSKA_ID_TRACKS ||
603             id == MATROSKA_ID_CUES || id == MATROSKA_ID_TAGS ||
604             id == MATROSKA_ID_SEEKHEAD || id == MATROSKA_ID_ATTACHMENTS ||
605             id == MATROSKA_ID_CLUSTER || id == MATROSKA_ID_CHAPTERS)
606         {
607             matroska->current_id = id;
608             return 0;
609         }
610         id = (id << 8) | avio_r8(pb);
611     }
612 eof:
613     matroska->done = 1;
614     return AVERROR_EOF;
615 }
616
617 /*
618  * Return: Whether we reached the end of a level in the hierarchy or not.
619  */
620 static int ebml_level_end(MatroskaDemuxContext *matroska)
621 {
622     AVIOContext *pb = matroska->ctx->pb;
623     int64_t pos = avio_tell(pb);
624
625     if (matroska->num_levels > 0) {
626         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
627         if (pos - level->start >= level->length || matroska->current_id) {
628             matroska->num_levels--;
629             return 1;
630         }
631     }
632     return 0;
633 }
634
635 /*
636  * Read: an "EBML number", which is defined as a variable-length
637  * array of bytes. The first byte indicates the length by giving a
638  * number of 0-bits followed by a one. The position of the first
639  * "one" bit inside the first byte indicates the length of this
640  * number.
641  * Returns: number of bytes read, < 0 on error
642  */
643 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
644                          int max_size, uint64_t *number)
645 {
646     int read = 1, n = 1;
647     uint64_t total = 0;
648
649     /* The first byte tells us the length in bytes - avio_r8() can normally
650      * return 0, but since that's not a valid first ebmlID byte, we can
651      * use it safely here to catch EOS. */
652     if (!(total = avio_r8(pb))) {
653         /* we might encounter EOS here */
654         if (!url_feof(pb)) {
655             int64_t pos = avio_tell(pb);
656             av_log(matroska->ctx, AV_LOG_ERROR,
657                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
658                    pos, pos);
659             return pb->error ? pb->error : AVERROR(EIO);
660         }
661         return AVERROR_EOF;
662     }
663
664     /* get the length of the EBML number */
665     read = 8 - ff_log2_tab[total];
666     if (read > max_size) {
667         int64_t pos = avio_tell(pb) - 1;
668         av_log(matroska->ctx, AV_LOG_ERROR,
669                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
670                (uint8_t) total, pos, pos);
671         return AVERROR_INVALIDDATA;
672     }
673
674     /* read out length */
675     total ^= 1 << ff_log2_tab[total];
676     while (n++ < read)
677         total = (total << 8) | avio_r8(pb);
678
679     *number = total;
680
681     return read;
682 }
683
684 /**
685  * Read a EBML length value.
686  * This needs special handling for the "unknown length" case which has multiple
687  * encodings.
688  */
689 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
690                             uint64_t *number)
691 {
692     int res = ebml_read_num(matroska, pb, 8, number);
693     if (res > 0 && *number + 1 == 1ULL << (7 * res))
694         *number = 0xffffffffffffffULL;
695     return res;
696 }
697
698 /*
699  * Read the next element as an unsigned int.
700  * 0 is success, < 0 is failure.
701  */
702 static int ebml_read_uint(AVIOContext *pb, int size, uint64_t *num)
703 {
704     int n = 0;
705
706     if (size > 8)
707         return AVERROR_INVALIDDATA;
708
709     /* big-endian ordering; build up number */
710     *num = 0;
711     while (n++ < size)
712         *num = (*num << 8) | avio_r8(pb);
713
714     return 0;
715 }
716
717 /*
718  * Read the next element as a float.
719  * 0 is success, < 0 is failure.
720  */
721 static int ebml_read_float(AVIOContext *pb, int size, double *num)
722 {
723     if (size == 0) {
724         *num = 0;
725     } else if (size == 4) {
726         *num = av_int2float(avio_rb32(pb));
727     } else if (size == 8){
728         *num = av_int2double(avio_rb64(pb));
729     } else
730         return AVERROR_INVALIDDATA;
731
732     return 0;
733 }
734
735 /*
736  * Read the next element as an ASCII string.
737  * 0 is success, < 0 is failure.
738  */
739 static int ebml_read_ascii(AVIOContext *pb, int size, char **str)
740 {
741     char *res;
742
743     /* EBML strings are usually not 0-terminated, so we allocate one
744      * byte more, read the string and NULL-terminate it ourselves. */
745     if (!(res = av_malloc(size + 1)))
746         return AVERROR(ENOMEM);
747     if (avio_read(pb, (uint8_t *) res, size) != size) {
748         av_free(res);
749         return AVERROR(EIO);
750     }
751     (res)[size] = '\0';
752     av_free(*str);
753     *str = res;
754
755     return 0;
756 }
757
758 /*
759  * Read the next element as binary data.
760  * 0 is success, < 0 is failure.
761  */
762 static int ebml_read_binary(AVIOContext *pb, int length, EbmlBin *bin)
763 {
764     av_free(bin->data);
765     if (!(bin->data = av_malloc(length)))
766         return AVERROR(ENOMEM);
767
768     bin->size = length;
769     bin->pos  = avio_tell(pb);
770     if (avio_read(pb, bin->data, length) != length) {
771         av_freep(&bin->data);
772         return AVERROR(EIO);
773     }
774
775     return 0;
776 }
777
778 /*
779  * Read the next element, but only the header. The contents
780  * are supposed to be sub-elements which can be read separately.
781  * 0 is success, < 0 is failure.
782  */
783 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
784 {
785     AVIOContext *pb = matroska->ctx->pb;
786     MatroskaLevel *level;
787
788     if (matroska->num_levels >= EBML_MAX_DEPTH) {
789         av_log(matroska->ctx, AV_LOG_ERROR,
790                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
791         return AVERROR(ENOSYS);
792     }
793
794     level = &matroska->levels[matroska->num_levels++];
795     level->start = avio_tell(pb);
796     level->length = length;
797
798     return 0;
799 }
800
801 /*
802  * Read signed/unsigned "EBML" numbers.
803  * Return: number of bytes processed, < 0 on error
804  */
805 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
806                                  uint8_t *data, uint32_t size, uint64_t *num)
807 {
808     AVIOContext pb;
809     ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
810     return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
811 }
812
813 /*
814  * Same as above, but signed.
815  */
816 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
817                                  uint8_t *data, uint32_t size, int64_t *num)
818 {
819     uint64_t unum;
820     int res;
821
822     /* read as unsigned number first */
823     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
824         return res;
825
826     /* make signed (weird way) */
827     *num = unum - ((1LL << (7*res - 1)) - 1);
828
829     return res;
830 }
831
832 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
833                            EbmlSyntax *syntax, void *data);
834
835 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
836                          uint32_t id, void *data)
837 {
838     int i;
839     for (i=0; syntax[i].id; i++)
840         if (id == syntax[i].id)
841             break;
842     if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
843         matroska->num_levels > 0 &&
844         matroska->levels[matroska->num_levels-1].length == 0xffffffffffffff)
845         return 0;  // we reached the end of an unknown size cluster
846     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32) {
847         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
848         if (matroska->ctx->error_recognition & AV_EF_EXPLODE)
849             return AVERROR_INVALIDDATA;
850     }
851     return ebml_parse_elem(matroska, &syntax[i], data);
852 }
853
854 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
855                       void *data)
856 {
857     if (!matroska->current_id) {
858         uint64_t id;
859         int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
860         if (res < 0)
861             return res;
862         matroska->current_id = id | 1 << 7*res;
863     }
864     return ebml_parse_id(matroska, syntax, matroska->current_id, data);
865 }
866
867 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
868                            void *data)
869 {
870     int i, res = 0;
871
872     for (i=0; syntax[i].id; i++)
873         switch (syntax[i].type) {
874         case EBML_UINT:
875             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
876             break;
877         case EBML_FLOAT:
878             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
879             break;
880         case EBML_STR:
881         case EBML_UTF8:
882             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
883             break;
884         }
885
886     while (!res && !ebml_level_end(matroska))
887         res = ebml_parse(matroska, syntax, data);
888
889     return res;
890 }
891
892 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
893                            EbmlSyntax *syntax, void *data)
894 {
895     static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
896         [EBML_UINT]  = 8,
897         [EBML_FLOAT] = 8,
898         // max. 16 MB for strings
899         [EBML_STR]   = 0x1000000,
900         [EBML_UTF8]  = 0x1000000,
901         // max. 256 MB for binary data
902         [EBML_BIN]   = 0x10000000,
903         // no limits for anything else
904     };
905     AVIOContext *pb = matroska->ctx->pb;
906     uint32_t id = syntax->id;
907     uint64_t length;
908     int res;
909     void *newelem;
910
911     data = (char *)data + syntax->data_offset;
912     if (syntax->list_elem_size) {
913         EbmlList *list = data;
914         newelem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
915         if (!newelem)
916             return AVERROR(ENOMEM);
917         list->elem = newelem;
918         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
919         memset(data, 0, syntax->list_elem_size);
920         list->nb_elem++;
921     }
922
923     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
924         matroska->current_id = 0;
925         if ((res = ebml_read_length(matroska, pb, &length)) < 0)
926             return res;
927         if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
928             av_log(matroska->ctx, AV_LOG_ERROR,
929                    "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for syntax element %i\n",
930                    length, max_lengths[syntax->type], syntax->type);
931             return AVERROR_INVALIDDATA;
932         }
933     }
934
935     switch (syntax->type) {
936     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
937     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
938     case EBML_STR:
939     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
940     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
941     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
942                          return res;
943                      if (id == MATROSKA_ID_SEGMENT)
944                          matroska->segment_start = avio_tell(matroska->ctx->pb);
945                      return ebml_parse_nest(matroska, syntax->def.n, data);
946     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
947     case EBML_STOP:  return 1;
948     default:
949         if(ffio_limit(pb, length) != length)
950             return AVERROR(EIO);
951         return avio_skip(pb,length)<0 ? AVERROR(EIO) : 0;
952     }
953     if (res == AVERROR_INVALIDDATA)
954         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
955     else if (res == AVERROR(EIO))
956         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
957     return res;
958 }
959
960 static void ebml_free(EbmlSyntax *syntax, void *data)
961 {
962     int i, j;
963     for (i=0; syntax[i].id; i++) {
964         void *data_off = (char *)data + syntax[i].data_offset;
965         switch (syntax[i].type) {
966         case EBML_STR:
967         case EBML_UTF8:  av_freep(data_off);                      break;
968         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
969         case EBML_NEST:
970             if (syntax[i].list_elem_size) {
971                 EbmlList *list = data_off;
972                 char *ptr = list->elem;
973                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
974                     ebml_free(syntax[i].def.n, ptr);
975                 av_free(list->elem);
976             } else
977                 ebml_free(syntax[i].def.n, data_off);
978         default:  break;
979         }
980     }
981 }
982
983
984 /*
985  * Autodetecting...
986  */
987 static int matroska_probe(AVProbeData *p)
988 {
989     uint64_t total = 0;
990     int len_mask = 0x80, size = 1, n = 1, i;
991
992     /* EBML header? */
993     if (AV_RB32(p->buf) != EBML_ID_HEADER)
994         return 0;
995
996     /* length of header */
997     total = p->buf[4];
998     while (size <= 8 && !(total & len_mask)) {
999         size++;
1000         len_mask >>= 1;
1001     }
1002     if (size > 8)
1003       return 0;
1004     total &= (len_mask - 1);
1005     while (n < size)
1006         total = (total << 8) | p->buf[4 + n++];
1007
1008     /* Does the probe data contain the whole header? */
1009     if (p->buf_size < 4 + size + total)
1010       return 0;
1011
1012     /* The header should contain a known document type. For now,
1013      * we don't parse the whole header but simply check for the
1014      * availability of that array of characters inside the header.
1015      * Not fully fool-proof, but good enough. */
1016     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
1017         int probelen = strlen(matroska_doctypes[i]);
1018         if (total < probelen)
1019             continue;
1020         for (n = 4+size; n <= 4+size+total-probelen; n++)
1021             if (!memcmp(p->buf+n, matroska_doctypes[i], probelen))
1022                 return AVPROBE_SCORE_MAX;
1023     }
1024
1025     // probably valid EBML header but no recognized doctype
1026     return AVPROBE_SCORE_MAX/2;
1027 }
1028
1029 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
1030                                                  int num)
1031 {
1032     MatroskaTrack *tracks = matroska->tracks.elem;
1033     int i;
1034
1035     for (i=0; i < matroska->tracks.nb_elem; i++)
1036         if (tracks[i].num == num)
1037             return &tracks[i];
1038
1039     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
1040     return NULL;
1041 }
1042
1043 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
1044                                   MatroskaTrack *track)
1045 {
1046     MatroskaTrackEncoding *encodings = track->encodings.elem;
1047     uint8_t* data = *buf;
1048     int isize = *buf_size;
1049     uint8_t* pkt_data = NULL;
1050     uint8_t av_unused *newpktdata;
1051     int pkt_size = isize;
1052     int result = 0;
1053     int olen;
1054
1055     if (pkt_size >= 10000000U)
1056         return -1;
1057
1058     switch (encodings[0].compression.algo) {
1059     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
1060         if (encodings[0].compression.settings.size && !encodings[0].compression.settings.data) {
1061             av_log(0, AV_LOG_ERROR, "Compression size but no data in headerstrip\n");
1062             return -1;
1063         }
1064         return encodings[0].compression.settings.size;
1065     case MATROSKA_TRACK_ENCODING_COMP_LZO:
1066         do {
1067             olen = pkt_size *= 3;
1068             newpktdata = av_realloc(pkt_data, pkt_size + AV_LZO_OUTPUT_PADDING);
1069             if (!newpktdata) {
1070                 goto failed;
1071             }
1072             pkt_data = newpktdata;
1073             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
1074         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
1075         if (result)
1076             goto failed;
1077         pkt_size -= olen;
1078         break;
1079 #if CONFIG_ZLIB
1080     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
1081         z_stream zstream = {0};
1082         if (inflateInit(&zstream) != Z_OK)
1083             return -1;
1084         zstream.next_in = data;
1085         zstream.avail_in = isize;
1086         do {
1087             pkt_size *= 3;
1088             newpktdata = av_realloc(pkt_data, pkt_size);
1089             if (!newpktdata) {
1090                 inflateEnd(&zstream);
1091                 goto failed;
1092             }
1093             pkt_data = newpktdata;
1094             zstream.avail_out = pkt_size - zstream.total_out;
1095             zstream.next_out = pkt_data + zstream.total_out;
1096             if (pkt_data) {
1097                 result = inflate(&zstream, Z_NO_FLUSH);
1098             } else
1099                 result = Z_MEM_ERROR;
1100         } while (result==Z_OK && pkt_size<10000000);
1101         pkt_size = zstream.total_out;
1102         inflateEnd(&zstream);
1103         if (result != Z_STREAM_END)
1104             goto failed;
1105         break;
1106     }
1107 #endif
1108 #if CONFIG_BZLIB
1109     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
1110         bz_stream bzstream = {0};
1111         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1112             return -1;
1113         bzstream.next_in = data;
1114         bzstream.avail_in = isize;
1115         do {
1116             pkt_size *= 3;
1117             newpktdata = av_realloc(pkt_data, pkt_size);
1118             if (!newpktdata) {
1119                 BZ2_bzDecompressEnd(&bzstream);
1120                 goto failed;
1121             }
1122             pkt_data = newpktdata;
1123             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1124             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1125             if (pkt_data) {
1126                 result = BZ2_bzDecompress(&bzstream);
1127             } else
1128                 result = BZ_MEM_ERROR;
1129         } while (result==BZ_OK && pkt_size<10000000);
1130         pkt_size = bzstream.total_out_lo32;
1131         BZ2_bzDecompressEnd(&bzstream);
1132         if (result != BZ_STREAM_END)
1133             goto failed;
1134         break;
1135     }
1136 #endif
1137     default:
1138         return -1;
1139     }
1140
1141     *buf = pkt_data;
1142     *buf_size = pkt_size;
1143     return 0;
1144  failed:
1145     av_free(pkt_data);
1146     return -1;
1147 }
1148
1149 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
1150                                     AVPacket *pkt, uint64_t display_duration)
1151 {
1152     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
1153     for (; *ptr!=',' && ptr<end-1; ptr++);
1154     if (*ptr == ',')
1155         ptr++;
1156     layer = ptr;
1157     for (; *ptr!=',' && ptr<end-1; ptr++);
1158     if (*ptr == ',') {
1159         int64_t end_pts = pkt->pts + display_duration;
1160         int sc = matroska->time_scale * pkt->pts / 10000000;
1161         int ec = matroska->time_scale * end_pts  / 10000000;
1162         int sh, sm, ss, eh, em, es, len;
1163         sh = sc/360000;  sc -= 360000*sh;
1164         sm = sc/  6000;  sc -=   6000*sm;
1165         ss = sc/   100;  sc -=    100*ss;
1166         eh = ec/360000;  ec -= 360000*eh;
1167         em = ec/  6000;  ec -=   6000*em;
1168         es = ec/   100;  ec -=    100*es;
1169         *ptr++ = '\0';
1170         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
1171         if (!(line = av_malloc(len)))
1172             return;
1173         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
1174                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
1175         av_free(pkt->data);
1176         pkt->data = line;
1177         pkt->size = strlen(line);
1178     }
1179 }
1180
1181 static int matroska_merge_packets(AVPacket *out, AVPacket *in)
1182 {
1183     int ret = av_grow_packet(out, in->size);
1184     if (ret < 0)
1185         return ret;
1186     memcpy(out->data + out->size - in->size, in->data, in->size);
1187     av_destruct_packet(in);
1188     av_free(in);
1189     return 0;
1190 }
1191
1192 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1193                                  AVDictionary **metadata, char *prefix)
1194 {
1195     MatroskaTag *tags = list->elem;
1196     char key[1024];
1197     int i;
1198
1199     for (i=0; i < list->nb_elem; i++) {
1200         const char *lang= (tags[i].lang && strcmp(tags[i].lang, "und")) ? tags[i].lang : NULL;
1201
1202         if (!tags[i].name) {
1203             av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1204             continue;
1205         }
1206         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1207         else         av_strlcpy(key, tags[i].name, sizeof(key));
1208         if (tags[i].def || !lang) {
1209         av_dict_set(metadata, key, tags[i].string, 0);
1210         if (tags[i].sub.nb_elem)
1211             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1212         }
1213         if (lang) {
1214             av_strlcat(key, "-", sizeof(key));
1215             av_strlcat(key, lang, sizeof(key));
1216             av_dict_set(metadata, key, tags[i].string, 0);
1217             if (tags[i].sub.nb_elem)
1218                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1219         }
1220     }
1221     ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1222 }
1223
1224 static void matroska_convert_tags(AVFormatContext *s)
1225 {
1226     MatroskaDemuxContext *matroska = s->priv_data;
1227     MatroskaTags *tags = matroska->tags.elem;
1228     int i, j;
1229
1230     for (i=0; i < matroska->tags.nb_elem; i++) {
1231         if (tags[i].target.attachuid) {
1232             MatroskaAttachement *attachment = matroska->attachments.elem;
1233             for (j=0; j<matroska->attachments.nb_elem; j++)
1234                 if (attachment[j].uid == tags[i].target.attachuid
1235                     && attachment[j].stream)
1236                     matroska_convert_tag(s, &tags[i].tag,
1237                                          &attachment[j].stream->metadata, NULL);
1238         } else if (tags[i].target.chapteruid) {
1239             MatroskaChapter *chapter = matroska->chapters.elem;
1240             for (j=0; j<matroska->chapters.nb_elem; j++)
1241                 if (chapter[j].uid == tags[i].target.chapteruid
1242                     && chapter[j].chapter)
1243                     matroska_convert_tag(s, &tags[i].tag,
1244                                          &chapter[j].chapter->metadata, NULL);
1245         } else if (tags[i].target.trackuid) {
1246             MatroskaTrack *track = matroska->tracks.elem;
1247             for (j=0; j<matroska->tracks.nb_elem; j++)
1248                 if (track[j].uid == tags[i].target.trackuid && track[j].stream)
1249                     matroska_convert_tag(s, &tags[i].tag,
1250                                          &track[j].stream->metadata, NULL);
1251         } else {
1252             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1253                                  tags[i].target.type);
1254         }
1255     }
1256 }
1257
1258 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska, int idx)
1259 {
1260     EbmlList *seekhead_list = &matroska->seekhead;
1261     MatroskaSeekhead *seekhead = seekhead_list->elem;
1262     uint32_t level_up = matroska->level_up;
1263     int64_t before_pos = avio_tell(matroska->ctx->pb);
1264     uint32_t saved_id = matroska->current_id;
1265     MatroskaLevel level;
1266     int64_t offset;
1267     int ret = 0;
1268
1269     if (idx >= seekhead_list->nb_elem
1270             || seekhead[idx].id == MATROSKA_ID_SEEKHEAD
1271             || seekhead[idx].id == MATROSKA_ID_CLUSTER)
1272         return 0;
1273
1274     /* seek */
1275     offset = seekhead[idx].pos + matroska->segment_start;
1276     if (avio_seek(matroska->ctx->pb, offset, SEEK_SET) == offset) {
1277         /* We don't want to lose our seekhead level, so we add
1278          * a dummy. This is a crude hack. */
1279         if (matroska->num_levels == EBML_MAX_DEPTH) {
1280             av_log(matroska->ctx, AV_LOG_INFO,
1281                    "Max EBML element depth (%d) reached, "
1282                    "cannot parse further.\n", EBML_MAX_DEPTH);
1283             ret = AVERROR_INVALIDDATA;
1284         } else {
1285             level.start = 0;
1286             level.length = (uint64_t)-1;
1287             matroska->levels[matroska->num_levels] = level;
1288             matroska->num_levels++;
1289             matroska->current_id = 0;
1290
1291             ret = ebml_parse(matroska, matroska_segment, matroska);
1292
1293             /* remove dummy level */
1294             while (matroska->num_levels) {
1295                 uint64_t length = matroska->levels[--matroska->num_levels].length;
1296                 if (length == (uint64_t)-1)
1297                     break;
1298             }
1299         }
1300     }
1301     /* seek back */
1302     avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
1303     matroska->level_up = level_up;
1304     matroska->current_id = saved_id;
1305
1306     return ret;
1307 }
1308
1309 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1310 {
1311     EbmlList *seekhead_list = &matroska->seekhead;
1312     int64_t before_pos = avio_tell(matroska->ctx->pb);
1313     int i;
1314
1315     // we should not do any seeking in the streaming case
1316     if (!matroska->ctx->pb->seekable ||
1317         (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1318         return;
1319
1320     for (i = 0; i < seekhead_list->nb_elem; i++) {
1321         MatroskaSeekhead *seekhead = seekhead_list->elem;
1322         if (seekhead[i].pos <= before_pos)
1323             continue;
1324
1325         // defer cues parsing until we actually need cue data.
1326         if (seekhead[i].id == MATROSKA_ID_CUES) {
1327             matroska->cues_parsing_deferred = 1;
1328             continue;
1329         }
1330
1331         if (matroska_parse_seekhead_entry(matroska, i) < 0) {
1332             // mark index as broken
1333             matroska->cues_parsing_deferred = -1;
1334             break;
1335         }
1336     }
1337 }
1338
1339 static void matroska_add_index_entries(MatroskaDemuxContext *matroska) {
1340     EbmlList *index_list;
1341     MatroskaIndex *index;
1342     int index_scale = 1;
1343     int i, j;
1344
1345     index_list = &matroska->index;
1346     index = index_list->elem;
1347     if (index_list->nb_elem
1348         && index[0].time > 1E14/matroska->time_scale) {
1349         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1350         index_scale = matroska->time_scale;
1351     }
1352     for (i = 0; i < index_list->nb_elem; i++) {
1353         EbmlList *pos_list = &index[i].pos;
1354         MatroskaIndexPos *pos = pos_list->elem;
1355         for (j = 0; j < pos_list->nb_elem; j++) {
1356             MatroskaTrack *track = matroska_find_track_by_num(matroska, pos[j].track);
1357             if (track && track->stream)
1358                 av_add_index_entry(track->stream,
1359                                    pos[j].pos + matroska->segment_start,
1360                                    index[i].time/index_scale, 0, 0,
1361                                    AVINDEX_KEYFRAME);
1362         }
1363     }
1364 }
1365
1366 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1367     EbmlList *seekhead_list = &matroska->seekhead;
1368     MatroskaSeekhead *seekhead = seekhead_list->elem;
1369     int i;
1370
1371     for (i = 0; i < seekhead_list->nb_elem; i++)
1372         if (seekhead[i].id == MATROSKA_ID_CUES)
1373             break;
1374     assert(i <= seekhead_list->nb_elem);
1375
1376     if (matroska_parse_seekhead_entry(matroska, i) < 0)
1377        matroska->cues_parsing_deferred = -1;
1378     matroska_add_index_entries(matroska);
1379 }
1380
1381 static int matroska_aac_profile(char *codec_id)
1382 {
1383     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1384     int profile;
1385
1386     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1387         if (strstr(codec_id, aac_profiles[profile]))
1388             break;
1389     return profile + 1;
1390 }
1391
1392 static int matroska_aac_sri(int samplerate)
1393 {
1394     int sri;
1395
1396     for (sri=0; sri<FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1397         if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
1398             break;
1399     return sri;
1400 }
1401
1402 static void matroska_metadata_creation_time(AVDictionary **metadata, int64_t date_utc)
1403 {
1404     char buffer[32];
1405     /* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
1406     time_t creation_time = date_utc / 1000000000 + 978307200;
1407     struct tm *ptm = gmtime(&creation_time);
1408     if (!ptm) return;
1409     strftime(buffer, sizeof(buffer), "%Y-%m-%d %H:%M:%S", ptm);
1410     av_dict_set(metadata, "creation_time", buffer, 0);
1411 }
1412
1413 static int matroska_read_header(AVFormatContext *s)
1414 {
1415     MatroskaDemuxContext *matroska = s->priv_data;
1416     EbmlList *attachements_list = &matroska->attachments;
1417     MatroskaAttachement *attachements;
1418     EbmlList *chapters_list = &matroska->chapters;
1419     MatroskaChapter *chapters;
1420     MatroskaTrack *tracks;
1421     uint64_t max_start = 0;
1422     int64_t pos;
1423     Ebml ebml = { 0 };
1424     AVStream *st;
1425     int i, j, k, res;
1426
1427     matroska->ctx = s;
1428
1429     /* First read the EBML header. */
1430     if (ebml_parse(matroska, ebml_syntax, &ebml)
1431         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1432         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 3 || !ebml.doctype) {
1433         av_log(matroska->ctx, AV_LOG_ERROR,
1434                "EBML header using unsupported features\n"
1435                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1436                ebml.version, ebml.doctype, ebml.doctype_version);
1437         ebml_free(ebml_syntax, &ebml);
1438         return AVERROR_PATCHWELCOME;
1439     } else if (ebml.doctype_version == 3) {
1440         av_log(matroska->ctx, AV_LOG_WARNING,
1441                "EBML header using unsupported features\n"
1442                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1443                ebml.version, ebml.doctype, ebml.doctype_version);
1444     }
1445     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1446         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1447             break;
1448     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1449         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1450         if (matroska->ctx->error_recognition & AV_EF_EXPLODE) {
1451             ebml_free(ebml_syntax, &ebml);
1452             return AVERROR_INVALIDDATA;
1453         }
1454     }
1455     ebml_free(ebml_syntax, &ebml);
1456
1457     /* The next thing is a segment. */
1458     pos = avio_tell(matroska->ctx->pb);
1459     res = ebml_parse(matroska, matroska_segments, matroska);
1460     // try resyncing until we find a EBML_STOP type element.
1461     while (res != 1) {
1462         res = matroska_resync(matroska, pos);
1463         if (res < 0)
1464             return res;
1465         pos = avio_tell(matroska->ctx->pb);
1466         res = ebml_parse(matroska, matroska_segment, matroska);
1467     }
1468     matroska_execute_seekhead(matroska);
1469
1470     if (!matroska->time_scale)
1471         matroska->time_scale = 1000000;
1472     if (matroska->duration)
1473         matroska->ctx->duration = matroska->duration * matroska->time_scale
1474                                   * 1000 / AV_TIME_BASE;
1475     av_dict_set(&s->metadata, "title", matroska->title, 0);
1476
1477     if (matroska->date_utc.size == 8)
1478         matroska_metadata_creation_time(&s->metadata, AV_RB64(matroska->date_utc.data));
1479
1480     tracks = matroska->tracks.elem;
1481     for (i=0; i < matroska->tracks.nb_elem; i++) {
1482         MatroskaTrack *track = &tracks[i];
1483         enum AVCodecID codec_id = AV_CODEC_ID_NONE;
1484         EbmlList *encodings_list = &track->encodings;
1485         MatroskaTrackEncoding *encodings = encodings_list->elem;
1486         uint8_t *extradata = NULL;
1487         int extradata_size = 0;
1488         int extradata_offset = 0;
1489         uint32_t fourcc = 0;
1490         AVIOContext b;
1491
1492         /* Apply some sanity checks. */
1493         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1494             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1495             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1496             av_log(matroska->ctx, AV_LOG_INFO,
1497                    "Unknown or unsupported track type %"PRIu64"\n",
1498                    track->type);
1499             continue;
1500         }
1501         if (track->codec_id == NULL)
1502             continue;
1503
1504         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1505             if (!track->default_duration && track->video.frame_rate > 0)
1506                 track->default_duration = 1000000000/track->video.frame_rate;
1507             if (!track->video.display_width)
1508                 track->video.display_width = track->video.pixel_width;
1509             if (!track->video.display_height)
1510                 track->video.display_height = track->video.pixel_height;
1511             if (track->video.color_space.size == 4)
1512                 fourcc = AV_RL32(track->video.color_space.data);
1513         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1514             if (!track->audio.out_samplerate)
1515                 track->audio.out_samplerate = track->audio.samplerate;
1516         }
1517         if (encodings_list->nb_elem > 1) {
1518             av_log(matroska->ctx, AV_LOG_ERROR,
1519                    "Multiple combined encodings not supported");
1520         } else if (encodings_list->nb_elem == 1) {
1521             if (encodings[0].type ||
1522                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1523 #if CONFIG_ZLIB
1524                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1525 #endif
1526 #if CONFIG_BZLIB
1527                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1528 #endif
1529                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1530                 encodings[0].scope = 0;
1531                 av_log(matroska->ctx, AV_LOG_ERROR,
1532                        "Unsupported encoding type");
1533             } else if (track->codec_priv.size && encodings[0].scope&2) {
1534                 uint8_t *codec_priv = track->codec_priv.data;
1535                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1536                                                     &track->codec_priv.size,
1537                                                     track);
1538                 if (offset < 0) {
1539                     track->codec_priv.data = NULL;
1540                     track->codec_priv.size = 0;
1541                     av_log(matroska->ctx, AV_LOG_ERROR,
1542                            "Failed to decode codec private data\n");
1543                 } else if (offset > 0) {
1544                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1545                     memcpy(track->codec_priv.data,
1546                            encodings[0].compression.settings.data, offset);
1547                     memcpy(track->codec_priv.data+offset, codec_priv,
1548                            track->codec_priv.size);
1549                     track->codec_priv.size += offset;
1550                 }
1551                 if (codec_priv != track->codec_priv.data)
1552                     av_free(codec_priv);
1553             }
1554         }
1555
1556         for(j=0; ff_mkv_codec_tags[j].id != AV_CODEC_ID_NONE; j++){
1557             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1558                         strlen(ff_mkv_codec_tags[j].str))){
1559                 codec_id= ff_mkv_codec_tags[j].id;
1560                 break;
1561             }
1562         }
1563
1564         st = track->stream = avformat_new_stream(s, NULL);
1565         if (st == NULL)
1566             return AVERROR(ENOMEM);
1567
1568         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1569             && track->codec_priv.size >= 40
1570             && track->codec_priv.data != NULL) {
1571             track->ms_compat = 1;
1572             fourcc = AV_RL32(track->codec_priv.data + 16);
1573             codec_id = ff_codec_get_id(ff_codec_bmp_tags, fourcc);
1574             extradata_offset = 40;
1575         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1576                    && track->codec_priv.size >= 14
1577                    && track->codec_priv.data != NULL) {
1578             int ret;
1579             ffio_init_context(&b, track->codec_priv.data, track->codec_priv.size,
1580                           AVIO_FLAG_READ, NULL, NULL, NULL, NULL);
1581             ret = ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1582             if (ret < 0)
1583                 return ret;
1584             codec_id = st->codec->codec_id;
1585             extradata_offset = FFMIN(track->codec_priv.size, 18);
1586         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1587                    && (track->codec_priv.size >= 86)
1588                    && (track->codec_priv.data != NULL)) {
1589             fourcc = AV_RL32(track->codec_priv.data);
1590             codec_id = ff_codec_get_id(ff_codec_movvideo_tags, fourcc);
1591         } else if (codec_id == AV_CODEC_ID_ALAC && track->codec_priv.size && track->codec_priv.size < INT_MAX-12) {
1592             /* Only ALAC's magic cookie is stored in Matroska's track headers.
1593                Create the "atom size", "tag", and "tag version" fields the
1594                decoder expects manually. */
1595             extradata_size = 12 + track->codec_priv.size;
1596             extradata = av_mallocz(extradata_size);
1597             if (extradata == NULL)
1598                 return AVERROR(ENOMEM);
1599             AV_WB32(extradata, extradata_size);
1600             memcpy(&extradata[4], "alac", 4);
1601             AV_WB32(&extradata[8], 0);
1602             memcpy(&extradata[12], track->codec_priv.data, track->codec_priv.size);
1603         } else if (codec_id == AV_CODEC_ID_PCM_S16BE) {
1604             switch (track->audio.bitdepth) {
1605             case  8:  codec_id = AV_CODEC_ID_PCM_U8;     break;
1606             case 24:  codec_id = AV_CODEC_ID_PCM_S24BE;  break;
1607             case 32:  codec_id = AV_CODEC_ID_PCM_S32BE;  break;
1608             }
1609         } else if (codec_id == AV_CODEC_ID_PCM_S16LE) {
1610             switch (track->audio.bitdepth) {
1611             case  8:  codec_id = AV_CODEC_ID_PCM_U8;     break;
1612             case 24:  codec_id = AV_CODEC_ID_PCM_S24LE;  break;
1613             case 32:  codec_id = AV_CODEC_ID_PCM_S32LE;  break;
1614             }
1615         } else if (codec_id==AV_CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1616             codec_id = AV_CODEC_ID_PCM_F64LE;
1617         } else if (codec_id == AV_CODEC_ID_AAC && !track->codec_priv.size) {
1618             int profile = matroska_aac_profile(track->codec_id);
1619             int sri = matroska_aac_sri(track->audio.samplerate);
1620             extradata = av_mallocz(5 + FF_INPUT_BUFFER_PADDING_SIZE);
1621             if (extradata == NULL)
1622                 return AVERROR(ENOMEM);
1623             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1624             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1625             if (strstr(track->codec_id, "SBR")) {
1626                 sri = matroska_aac_sri(track->audio.out_samplerate);
1627                 extradata[2] = 0x56;
1628                 extradata[3] = 0xE5;
1629                 extradata[4] = 0x80 | (sri<<3);
1630                 extradata_size = 5;
1631             } else
1632                 extradata_size = 2;
1633         } else if (codec_id == AV_CODEC_ID_TTA) {
1634             extradata_size = 30;
1635             extradata = av_mallocz(extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
1636             if (extradata == NULL)
1637                 return AVERROR(ENOMEM);
1638             ffio_init_context(&b, extradata, extradata_size, 1,
1639                           NULL, NULL, NULL, NULL);
1640             avio_write(&b, "TTA1", 4);
1641             avio_wl16(&b, 1);
1642             avio_wl16(&b, track->audio.channels);
1643             avio_wl16(&b, track->audio.bitdepth);
1644             avio_wl32(&b, track->audio.out_samplerate);
1645             avio_wl32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1646         } else if (codec_id == AV_CODEC_ID_RV10 || codec_id == AV_CODEC_ID_RV20 ||
1647                    codec_id == AV_CODEC_ID_RV30 || codec_id == AV_CODEC_ID_RV40) {
1648             extradata_offset = 26;
1649         } else if (codec_id == AV_CODEC_ID_RA_144) {
1650             track->audio.out_samplerate = 8000;
1651             track->audio.channels = 1;
1652         } else if ((codec_id == AV_CODEC_ID_RA_288 || codec_id == AV_CODEC_ID_COOK ||
1653                     codec_id == AV_CODEC_ID_ATRAC3 || codec_id == AV_CODEC_ID_SIPR)
1654                     && track->codec_priv.data) {
1655             int flavor;
1656
1657             ffio_init_context(&b, track->codec_priv.data,track->codec_priv.size,
1658                           0, NULL, NULL, NULL, NULL);
1659             avio_skip(&b, 22);
1660             flavor                       = avio_rb16(&b);
1661             track->audio.coded_framesize = avio_rb32(&b);
1662             avio_skip(&b, 12);
1663             track->audio.sub_packet_h    = avio_rb16(&b);
1664             track->audio.frame_size      = avio_rb16(&b);
1665             track->audio.sub_packet_size = avio_rb16(&b);
1666             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1667             if (codec_id == AV_CODEC_ID_RA_288) {
1668                 st->codec->block_align = track->audio.coded_framesize;
1669                 track->codec_priv.size = 0;
1670             } else {
1671                 if (codec_id == AV_CODEC_ID_SIPR && flavor < 4) {
1672                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1673                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1674                     st->codec->bit_rate = sipr_bit_rate[flavor];
1675                 }
1676                 st->codec->block_align = track->audio.sub_packet_size;
1677                 extradata_offset = 78;
1678             }
1679         }
1680         track->codec_priv.size -= extradata_offset;
1681
1682         if (codec_id == AV_CODEC_ID_NONE)
1683             av_log(matroska->ctx, AV_LOG_INFO,
1684                    "Unknown/unsupported AVCodecID %s.\n", track->codec_id);
1685
1686         if (track->time_scale < 0.01)
1687             track->time_scale = 1.0;
1688         avpriv_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1689
1690         st->codec->codec_id = codec_id;
1691         st->start_time = 0;
1692         if (strcmp(track->language, "und"))
1693             av_dict_set(&st->metadata, "language", track->language, 0);
1694         av_dict_set(&st->metadata, "title", track->name, 0);
1695
1696         if (track->flag_default)
1697             st->disposition |= AV_DISPOSITION_DEFAULT;
1698         if (track->flag_forced)
1699             st->disposition |= AV_DISPOSITION_FORCED;
1700
1701         if (!st->codec->extradata) {
1702             if(extradata){
1703                 st->codec->extradata = extradata;
1704                 st->codec->extradata_size = extradata_size;
1705             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1706                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1707                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1708                 if(st->codec->extradata == NULL)
1709                     return AVERROR(ENOMEM);
1710                 st->codec->extradata_size = track->codec_priv.size;
1711                 memcpy(st->codec->extradata,
1712                        track->codec_priv.data + extradata_offset,
1713                        track->codec_priv.size);
1714             }
1715         }
1716
1717         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1718             MatroskaTrackPlane *planes = track->operation.combine_planes.elem;
1719
1720             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1721             st->codec->codec_tag  = fourcc;
1722             st->codec->width  = track->video.pixel_width;
1723             st->codec->height = track->video.pixel_height;
1724             av_reduce(&st->sample_aspect_ratio.num,
1725                       &st->sample_aspect_ratio.den,
1726                       st->codec->height * track->video.display_width,
1727                       st->codec-> width * track->video.display_height,
1728                       255);
1729             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1730             if (track->default_duration) {
1731                 av_reduce(&st->avg_frame_rate.num, &st->avg_frame_rate.den,
1732                           1000000000, track->default_duration, 30000);
1733 #if FF_API_R_FRAME_RATE
1734                 st->r_frame_rate = st->avg_frame_rate;
1735 #endif
1736             }
1737
1738             /* export stereo mode flag as metadata tag */
1739             if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREO_MODE_COUNT)
1740                 av_dict_set(&st->metadata, "stereo_mode", ff_matroska_video_stereo_mode[track->video.stereo_mode], 0);
1741
1742             /* if we have virtual track, mark the real tracks */
1743             for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
1744                 char buf[32];
1745                 if (planes[j].type >= MATROSKA_VIDEO_STEREO_PLANE_COUNT)
1746                     continue;
1747                 snprintf(buf, sizeof(buf), "%s_%d",
1748                          ff_matroska_video_stereo_plane[planes[j].type], i);
1749                 for (k=0; k < matroska->tracks.nb_elem; k++)
1750                     if (planes[j].uid == tracks[k].uid) {
1751                         av_dict_set(&s->streams[k]->metadata,
1752                                     "stereo_mode", buf, 0);
1753                         break;
1754                     }
1755             }
1756         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1757             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1758             st->codec->sample_rate = track->audio.out_samplerate;
1759             st->codec->channels = track->audio.channels;
1760             if (st->codec->codec_id != AV_CODEC_ID_AAC)
1761             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1762         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1763             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1764             if (st->codec->codec_id == AV_CODEC_ID_SSA)
1765                 matroska->contains_ssa = 1;
1766         }
1767     }
1768
1769     attachements = attachements_list->elem;
1770     for (j=0; j<attachements_list->nb_elem; j++) {
1771         if (!(attachements[j].filename && attachements[j].mime &&
1772               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1773             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1774         } else {
1775             AVStream *st = avformat_new_stream(s, NULL);
1776             if (st == NULL)
1777                 break;
1778             av_dict_set(&st->metadata, "filename",attachements[j].filename, 0);
1779             av_dict_set(&st->metadata, "mimetype", attachements[j].mime, 0);
1780             st->codec->codec_id = AV_CODEC_ID_NONE;
1781             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1782             st->codec->extradata  = av_malloc(attachements[j].bin.size + FF_INPUT_BUFFER_PADDING_SIZE);
1783             if(st->codec->extradata == NULL)
1784                 break;
1785             st->codec->extradata_size = attachements[j].bin.size;
1786             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1787
1788             for (i=0; ff_mkv_mime_tags[i].id != AV_CODEC_ID_NONE; i++) {
1789                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1790                              strlen(ff_mkv_mime_tags[i].str))) {
1791                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1792                     break;
1793                 }
1794             }
1795             attachements[j].stream = st;
1796         }
1797     }
1798
1799     chapters = chapters_list->elem;
1800     for (i=0; i<chapters_list->nb_elem; i++)
1801         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1802             && (max_start==0 || chapters[i].start > max_start)) {
1803             chapters[i].chapter =
1804             avpriv_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1805                            chapters[i].start, chapters[i].end,
1806                            chapters[i].title);
1807             av_dict_set(&chapters[i].chapter->metadata,
1808                              "title", chapters[i].title, 0);
1809             max_start = chapters[i].start;
1810         }
1811
1812     matroska_add_index_entries(matroska);
1813
1814     matroska_convert_tags(s);
1815
1816     return 0;
1817 }
1818
1819 /*
1820  * Put one packet in an application-supplied AVPacket struct.
1821  * Returns 0 on success or -1 on failure.
1822  */
1823 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1824                                    AVPacket *pkt)
1825 {
1826     if (matroska->num_packets > 0) {
1827         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1828         av_free(matroska->packets[0]);
1829         if (matroska->num_packets > 1) {
1830             void *newpackets;
1831             memmove(&matroska->packets[0], &matroska->packets[1],
1832                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1833             newpackets = av_realloc(matroska->packets,
1834                             (matroska->num_packets - 1) * sizeof(AVPacket *));
1835             if (newpackets)
1836                 matroska->packets = newpackets;
1837         } else {
1838             av_freep(&matroska->packets);
1839             matroska->prev_pkt = NULL;
1840         }
1841         matroska->num_packets--;
1842         return 0;
1843     }
1844
1845     return -1;
1846 }
1847
1848 /*
1849  * Free all packets in our internal queue.
1850  */
1851 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1852 {
1853     if (matroska->packets) {
1854         int n;
1855         for (n = 0; n < matroska->num_packets; n++) {
1856             av_free_packet(matroska->packets[n]);
1857             av_free(matroska->packets[n]);
1858         }
1859         av_freep(&matroska->packets);
1860         matroska->num_packets = 0;
1861     }
1862 }
1863
1864 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1865                                 int size, int64_t pos, uint64_t cluster_time,
1866                                 uint64_t duration, int is_keyframe,
1867                                 int64_t cluster_pos)
1868 {
1869     uint64_t timecode = AV_NOPTS_VALUE;
1870     MatroskaTrack *track;
1871     int res = 0;
1872     AVStream *st;
1873     AVPacket *pkt;
1874     int16_t block_time;
1875     uint32_t *lace_size = NULL;
1876     int n, flags, laces = 0;
1877     uint64_t num;
1878
1879     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1880         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1881         return n;
1882     }
1883     data += n;
1884     size -= n;
1885
1886     track = matroska_find_track_by_num(matroska, num);
1887     if (!track || !track->stream) {
1888         av_log(matroska->ctx, AV_LOG_INFO,
1889                "Invalid stream %"PRIu64" or size %u\n", num, size);
1890         return AVERROR_INVALIDDATA;
1891     } else if (size <= 3)
1892         return 0;
1893     st = track->stream;
1894     if (st->discard >= AVDISCARD_ALL)
1895         return res;
1896     av_assert1(duration != AV_NOPTS_VALUE);
1897
1898     block_time = AV_RB16(data);
1899     data += 2;
1900     flags = *data++;
1901     size -= 3;
1902     if (is_keyframe == -1)
1903         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1904
1905     if (cluster_time != (uint64_t)-1
1906         && (block_time >= 0 || cluster_time >= -block_time)) {
1907         timecode = cluster_time + block_time;
1908         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1909             && timecode < track->end_timecode)
1910             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1911         if (is_keyframe)
1912             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1913     }
1914
1915     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1916         if (timecode < matroska->skip_to_timecode)
1917             return res;
1918         if (!st->skip_to_keyframe) {
1919             av_log(matroska->ctx, AV_LOG_ERROR, "File is broken, keyframes not correctly marked!\n");
1920             matroska->skip_to_keyframe = 0;
1921         }
1922         if (is_keyframe)
1923             matroska->skip_to_keyframe = 0;
1924     }
1925
1926     switch ((flags & 0x06) >> 1) {
1927         case 0x0: /* no lacing */
1928             laces = 1;
1929             lace_size = av_mallocz(sizeof(int));
1930             lace_size[0] = size;
1931             break;
1932
1933         case 0x1: /* Xiph lacing */
1934         case 0x2: /* fixed-size lacing */
1935         case 0x3: /* EBML lacing */
1936             av_assert0(size>0); // size <=3 is checked before size-=3 above
1937             laces = (*data) + 1;
1938             data += 1;
1939             size -= 1;
1940             lace_size = av_mallocz(laces * sizeof(int));
1941
1942             switch ((flags & 0x06) >> 1) {
1943                 case 0x1: /* Xiph lacing */ {
1944                     uint8_t temp;
1945                     uint32_t total = 0;
1946                     for (n = 0; res == 0 && n < laces - 1; n++) {
1947                         while (1) {
1948                             if (size == 0) {
1949                                 res = -1;
1950                                 break;
1951                             }
1952                             temp = *data;
1953                             lace_size[n] += temp;
1954                             data += 1;
1955                             size -= 1;
1956                             if (temp != 0xff)
1957                                 break;
1958                         }
1959                         total += lace_size[n];
1960                     }
1961                     lace_size[n] = size - total;
1962                     break;
1963                 }
1964
1965                 case 0x2: /* fixed-size lacing */
1966                     for (n = 0; n < laces; n++)
1967                         lace_size[n] = size / laces;
1968                     break;
1969
1970                 case 0x3: /* EBML lacing */ {
1971                     uint32_t total;
1972                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1973                     if (n < 0) {
1974                         av_log(matroska->ctx, AV_LOG_INFO,
1975                                "EBML block data error\n");
1976                         break;
1977                     }
1978                     data += n;
1979                     size -= n;
1980                     total = lace_size[0] = num;
1981                     for (n = 1; res == 0 && n < laces - 1; n++) {
1982                         int64_t snum;
1983                         int r;
1984                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1985                         if (r < 0) {
1986                             av_log(matroska->ctx, AV_LOG_INFO,
1987                                    "EBML block data error\n");
1988                             break;
1989                         }
1990                         data += r;
1991                         size -= r;
1992                         lace_size[n] = lace_size[n - 1] + snum;
1993                         total += lace_size[n];
1994                     }
1995                     lace_size[laces - 1] = size - total;
1996                     break;
1997                 }
1998             }
1999             break;
2000     }
2001
2002     if (res == 0) {
2003         if (!duration)
2004             duration = track->default_duration * laces / matroska->time_scale;
2005
2006         if (cluster_time != (uint64_t)-1 && (block_time >= 0 || cluster_time >= -block_time))
2007             track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
2008
2009         for (n = 0; n < laces; n++) {
2010             int64_t lace_duration = duration*(n+1) / laces - duration*n / laces;
2011
2012             if (lace_size[n] > size) {
2013                 av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
2014                 break;
2015             }
2016
2017             if ((st->codec->codec_id == AV_CODEC_ID_RA_288 ||
2018                  st->codec->codec_id == AV_CODEC_ID_COOK ||
2019                  st->codec->codec_id == AV_CODEC_ID_SIPR ||
2020                  st->codec->codec_id == AV_CODEC_ID_ATRAC3) &&
2021                  st->codec->block_align && track->audio.sub_packet_size) {
2022                 int a = st->codec->block_align;
2023                 int sps = track->audio.sub_packet_size;
2024                 int cfs = track->audio.coded_framesize;
2025                 int h = track->audio.sub_packet_h;
2026                 int y = track->audio.sub_packet_cnt;
2027                 int w = track->audio.frame_size;
2028                 int x;
2029
2030                 if (!track->audio.pkt_cnt) {
2031                     if (track->audio.sub_packet_cnt == 0)
2032                         track->audio.buf_timecode = timecode;
2033                     if (st->codec->codec_id == AV_CODEC_ID_RA_288) {
2034                         if (size < cfs * h / 2) {
2035                             av_log(matroska->ctx, AV_LOG_ERROR,
2036                                    "Corrupt int4 RM-style audio packet size\n");
2037                             res = AVERROR_INVALIDDATA;
2038                             goto end;
2039                         }
2040                         for (x=0; x<h/2; x++)
2041                             memcpy(track->audio.buf+x*2*w+y*cfs,
2042                                    data+x*cfs, cfs);
2043                     } else if (st->codec->codec_id == AV_CODEC_ID_SIPR) {
2044                         if (size < w) {
2045                             av_log(matroska->ctx, AV_LOG_ERROR,
2046                                    "Corrupt sipr RM-style audio packet size\n");
2047                             res = AVERROR_INVALIDDATA;
2048                             goto end;
2049                         }
2050                         memcpy(track->audio.buf + y*w, data, w);
2051                     } else {
2052                         if (size < sps * w / sps) {
2053                             av_log(matroska->ctx, AV_LOG_ERROR,
2054                                    "Corrupt generic RM-style audio packet size\n");
2055                             res = AVERROR_INVALIDDATA;
2056                             goto end;
2057                         }
2058                         for (x=0; x<w/sps; x++)
2059                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
2060                     }
2061
2062                     if (++track->audio.sub_packet_cnt >= h) {
2063                         if (st->codec->codec_id == AV_CODEC_ID_SIPR)
2064                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
2065                         track->audio.sub_packet_cnt = 0;
2066                         track->audio.pkt_cnt = h*w / a;
2067                     }
2068                 }
2069                 while (track->audio.pkt_cnt) {
2070                     pkt = av_mallocz(sizeof(AVPacket));
2071                     av_new_packet(pkt, a);
2072                     memcpy(pkt->data, track->audio.buf
2073                            + a * (h*w / a - track->audio.pkt_cnt--), a);
2074                     pkt->pts = track->audio.buf_timecode;
2075                     track->audio.buf_timecode = AV_NOPTS_VALUE;
2076                     pkt->pos = pos;
2077                     pkt->stream_index = st->index;
2078                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
2079                 }
2080             } else {
2081                 MatroskaTrackEncoding *encodings = track->encodings.elem;
2082                 int offset = 0;
2083                 uint32_t pkt_size = lace_size[n];
2084                 uint8_t *pkt_data = data;
2085
2086                 if (encodings && encodings->scope & 1) {
2087                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
2088                     if (offset < 0)
2089                         continue;
2090                     av_assert0(offset + pkt_size >= pkt_size);
2091                 }
2092
2093                 pkt = av_mallocz(sizeof(AVPacket));
2094                 /* XXX: prevent data copy... */
2095                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
2096                     av_free(pkt);
2097                     res = AVERROR(ENOMEM);
2098                     break;
2099                 }
2100                 if (offset)
2101                     memcpy (pkt->data, encodings->compression.settings.data, offset);
2102                 memcpy (pkt->data+offset, pkt_data, pkt_size);
2103
2104                 if (pkt_data != data)
2105                     av_free(pkt_data);
2106
2107                 if (n == 0)
2108                     pkt->flags = is_keyframe;
2109                 pkt->stream_index = st->index;
2110
2111                 if (track->ms_compat)
2112                     pkt->dts = timecode;
2113                 else
2114                     pkt->pts = timecode;
2115                 pkt->pos = pos;
2116                 if (st->codec->codec_id == AV_CODEC_ID_SUBRIP) {
2117                     /*
2118                      * For backward compatibility.
2119                      * Historically, we have put subtitle duration
2120                      * in convergence_duration, on the off chance
2121                      * that the time_scale is less than 1us, which
2122                      * could result in a 32bit overflow on the
2123                      * normal duration field.
2124                      */
2125                     pkt->convergence_duration = lace_duration;
2126                 }
2127
2128                 if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE ||
2129                     lace_duration <= INT_MAX) {
2130                     /*
2131                      * For non subtitle tracks, just store the duration
2132                      * as normal.
2133                      *
2134                      * If it's a subtitle track and duration value does
2135                      * not overflow a uint32, then also store it normally.
2136                      */
2137                     pkt->duration = lace_duration;
2138                 }
2139
2140                 if (st->codec->codec_id == AV_CODEC_ID_SSA)
2141                     matroska_fix_ass_packet(matroska, pkt, lace_duration);
2142
2143                 if (matroska->prev_pkt &&
2144                     timecode != AV_NOPTS_VALUE &&
2145                     matroska->prev_pkt->pts == timecode &&
2146                     matroska->prev_pkt->stream_index == st->index &&
2147                     st->codec->codec_id == AV_CODEC_ID_SSA)
2148                     matroska_merge_packets(matroska->prev_pkt, pkt);
2149                 else {
2150                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
2151                     matroska->prev_pkt = pkt;
2152                 }
2153             }
2154
2155             if (timecode != AV_NOPTS_VALUE)
2156                 timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE;
2157             data += lace_size[n];
2158             size -= lace_size[n];
2159         }
2160     }
2161
2162 end:
2163     av_free(lace_size);
2164     return res;
2165 }
2166
2167 static int matroska_parse_cluster_incremental(MatroskaDemuxContext *matroska)
2168 {
2169     EbmlList *blocks_list;
2170     MatroskaBlock *blocks;
2171     int i, res;
2172     res = ebml_parse(matroska,
2173                      matroska_cluster_incremental_parsing,
2174                      &matroska->current_cluster);
2175     if (res == 1) {
2176         /* New Cluster */
2177         if (matroska->current_cluster_pos)
2178             ebml_level_end(matroska);
2179         ebml_free(matroska_cluster, &matroska->current_cluster);
2180         memset(&matroska->current_cluster, 0, sizeof(MatroskaCluster));
2181         matroska->current_cluster_num_blocks = 0;
2182         matroska->current_cluster_pos = avio_tell(matroska->ctx->pb);
2183         matroska->prev_pkt = NULL;
2184         /* sizeof the ID which was already read */
2185         if (matroska->current_id)
2186             matroska->current_cluster_pos -= 4;
2187         res = ebml_parse(matroska,
2188                          matroska_clusters_incremental,
2189                          &matroska->current_cluster);
2190         /* Try parsing the block again. */
2191         if (res == 1)
2192             res = ebml_parse(matroska,
2193                              matroska_cluster_incremental_parsing,
2194                              &matroska->current_cluster);
2195     }
2196
2197     if (!res &&
2198         matroska->current_cluster_num_blocks <
2199             matroska->current_cluster.blocks.nb_elem) {
2200         blocks_list = &matroska->current_cluster.blocks;
2201         blocks = blocks_list->elem;
2202
2203         matroska->current_cluster_num_blocks = blocks_list->nb_elem;
2204         i = blocks_list->nb_elem - 1;
2205         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
2206             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
2207             if (!blocks[i].non_simple)
2208                 blocks[i].duration = 0;
2209             res = matroska_parse_block(matroska,
2210                                        blocks[i].bin.data, blocks[i].bin.size,
2211                                        blocks[i].bin.pos,
2212                                        matroska->current_cluster.timecode,
2213                                        blocks[i].duration, is_keyframe,
2214                                        matroska->current_cluster_pos);
2215         }
2216     }
2217
2218     if (res < 0)  matroska->done = 1;
2219     return res;
2220 }
2221
2222 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
2223 {
2224     MatroskaCluster cluster = { 0 };
2225     EbmlList *blocks_list;
2226     MatroskaBlock *blocks;
2227     int i, res;
2228     int64_t pos;
2229     if (!matroska->contains_ssa)
2230         return matroska_parse_cluster_incremental(matroska);
2231     pos = avio_tell(matroska->ctx->pb);
2232     matroska->prev_pkt = NULL;
2233     if (matroska->current_id)
2234         pos -= 4;  /* sizeof the ID which was already read */
2235     res = ebml_parse(matroska, matroska_clusters, &cluster);
2236     blocks_list = &cluster.blocks;
2237     blocks = blocks_list->elem;
2238     for (i=0; i<blocks_list->nb_elem; i++)
2239         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
2240             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
2241             res=matroska_parse_block(matroska,
2242                                      blocks[i].bin.data, blocks[i].bin.size,
2243                                      blocks[i].bin.pos,  cluster.timecode,
2244                                      blocks[i].duration, is_keyframe,
2245                                      pos);
2246         }
2247     ebml_free(matroska_cluster, &cluster);
2248     return res;
2249 }
2250
2251 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
2252 {
2253     MatroskaDemuxContext *matroska = s->priv_data;
2254
2255     while (matroska_deliver_packet(matroska, pkt)) {
2256         int64_t pos = avio_tell(matroska->ctx->pb);
2257         if (matroska->done)
2258             return AVERROR_EOF;
2259         if (matroska_parse_cluster(matroska) < 0)
2260             matroska_resync(matroska, pos);
2261     }
2262
2263     return 0;
2264 }
2265
2266 static int matroska_read_seek(AVFormatContext *s, int stream_index,
2267                               int64_t timestamp, int flags)
2268 {
2269     MatroskaDemuxContext *matroska = s->priv_data;
2270     MatroskaTrack *tracks = matroska->tracks.elem;
2271     AVStream *st = s->streams[stream_index];
2272     int i, index, index_sub, index_min;
2273
2274     /* Parse the CUES now since we need the index data to seek. */
2275     if (matroska->cues_parsing_deferred > 0) {
2276         matroska->cues_parsing_deferred = 0;
2277         matroska_parse_cues(matroska);
2278     }
2279
2280     if (!st->nb_index_entries)
2281         goto err;
2282     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
2283
2284     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
2285         avio_seek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
2286         matroska->current_id = 0;
2287         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
2288             matroska->prev_pkt = NULL;
2289             matroska_clear_queue(matroska);
2290             if (matroska_parse_cluster(matroska) < 0)
2291                 break;
2292         }
2293     }
2294
2295     matroska_clear_queue(matroska);
2296     if (index < 0 || (matroska->cues_parsing_deferred < 0 && index == st->nb_index_entries - 1))
2297         goto err;
2298
2299     index_min = index;
2300     for (i=0; i < matroska->tracks.nb_elem; i++) {
2301         tracks[i].audio.pkt_cnt = 0;
2302         tracks[i].audio.sub_packet_cnt = 0;
2303         tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
2304         tracks[i].end_timecode = 0;
2305         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
2306             && !tracks[i].stream->discard != AVDISCARD_ALL) {
2307             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
2308             if (index_sub >= 0
2309                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
2310                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
2311                 index_min = index_sub;
2312         }
2313     }
2314
2315     avio_seek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
2316     matroska->current_id = 0;
2317     st->skip_to_keyframe =
2318     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
2319     matroska->skip_to_timecode = st->index_entries[index].timestamp;
2320     matroska->done = 0;
2321     matroska->num_levels = 0;
2322     ff_update_cur_dts(s, st, st->index_entries[index].timestamp);
2323     return 0;
2324 err:
2325     // slightly hackish but allows proper fallback to
2326     // the generic seeking code.
2327     matroska_clear_queue(matroska);
2328     matroska->current_id = 0;
2329     st->skip_to_keyframe =
2330     matroska->skip_to_keyframe = 0;
2331     matroska->done = 0;
2332     matroska->num_levels = 0;
2333     return -1;
2334 }
2335
2336 static int matroska_read_close(AVFormatContext *s)
2337 {
2338     MatroskaDemuxContext *matroska = s->priv_data;
2339     MatroskaTrack *tracks = matroska->tracks.elem;
2340     int n;
2341
2342     matroska_clear_queue(matroska);
2343
2344     for (n=0; n < matroska->tracks.nb_elem; n++)
2345         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
2346             av_free(tracks[n].audio.buf);
2347     ebml_free(matroska_cluster, &matroska->current_cluster);
2348     ebml_free(matroska_segment, matroska);
2349
2350     return 0;
2351 }
2352
2353 AVInputFormat ff_matroska_demuxer = {
2354     .name           = "matroska,webm",
2355     .long_name      = NULL_IF_CONFIG_SMALL("Matroska / WebM"),
2356     .priv_data_size = sizeof(MatroskaDemuxContext),
2357     .read_probe     = matroska_probe,
2358     .read_header    = matroska_read_header,
2359     .read_packet    = matroska_read_packet,
2360     .read_close     = matroska_read_close,
2361     .read_seek      = matroska_read_seek,
2362 };