Add some basic metadata conversion tables for matroska and asf.
[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 libavformat/matroskadec.c
24  * Matroska file demuxer
25  * by Ronald Bultje <rbultje@ronald.bitfreak.net>
26  * with a little help from Moritz Bunkus <moritz@bunkus.org>
27  * totally reworked by Aurelien Jacobs <aurel@gnuage.org>
28  * Specs available on the Matroska project page: http://www.matroska.org/.
29  */
30
31 #include <stdio.h>
32 #include "avformat.h"
33 /* For codec_get_id(). */
34 #include "riff.h"
35 #include "isom.h"
36 #include "matroska.h"
37 #include "libavcodec/mpeg4audio.h"
38 #include "libavutil/intfloat_readwrite.h"
39 #include "libavutil/intreadwrite.h"
40 #include "libavutil/avstring.h"
41 #include "libavutil/lzo.h"
42 #if CONFIG_ZLIB
43 #include <zlib.h>
44 #endif
45 #if CONFIG_BZLIB
46 #include <bzlib.h>
47 #endif
48
49 typedef enum {
50     EBML_NONE,
51     EBML_UINT,
52     EBML_FLOAT,
53     EBML_STR,
54     EBML_UTF8,
55     EBML_BIN,
56     EBML_NEST,
57     EBML_PASS,
58     EBML_STOP,
59 } EbmlType;
60
61 typedef const struct EbmlSyntax {
62     uint32_t id;
63     EbmlType type;
64     int list_elem_size;
65     int data_offset;
66     union {
67         uint64_t    u;
68         double      f;
69         const char *s;
70         const struct EbmlSyntax *n;
71     } def;
72 } EbmlSyntax;
73
74 typedef struct {
75     int nb_elem;
76     void *elem;
77 } EbmlList;
78
79 typedef struct {
80     int      size;
81     uint8_t *data;
82     int64_t  pos;
83 } EbmlBin;
84
85 typedef struct {
86     uint64_t version;
87     uint64_t max_size;
88     uint64_t id_length;
89     char    *doctype;
90     uint64_t doctype_version;
91 } Ebml;
92
93 typedef struct {
94     uint64_t algo;
95     EbmlBin  settings;
96 } MatroskaTrackCompression;
97
98 typedef struct {
99     uint64_t scope;
100     uint64_t type;
101     MatroskaTrackCompression compression;
102 } MatroskaTrackEncoding;
103
104 typedef struct {
105     double   frame_rate;
106     uint64_t display_width;
107     uint64_t display_height;
108     uint64_t pixel_width;
109     uint64_t pixel_height;
110     uint64_t fourcc;
111 } MatroskaTrackVideo;
112
113 typedef struct {
114     double   samplerate;
115     double   out_samplerate;
116     uint64_t bitdepth;
117     uint64_t channels;
118
119     /* real audio header (extracted from extradata) */
120     int      coded_framesize;
121     int      sub_packet_h;
122     int      frame_size;
123     int      sub_packet_size;
124     int      sub_packet_cnt;
125     int      pkt_cnt;
126     uint8_t *buf;
127 } MatroskaTrackAudio;
128
129 typedef struct {
130     uint64_t num;
131     uint64_t uid;
132     uint64_t type;
133     char    *name;
134     char    *codec_id;
135     EbmlBin  codec_priv;
136     char    *language;
137     double time_scale;
138     uint64_t default_duration;
139     uint64_t flag_default;
140     MatroskaTrackVideo video;
141     MatroskaTrackAudio audio;
142     EbmlList encodings;
143
144     AVStream *stream;
145     int64_t end_timecode;
146 } MatroskaTrack;
147
148 typedef struct {
149     uint64_t uid;
150     char *filename;
151     char *mime;
152     EbmlBin bin;
153
154     AVStream *stream;
155 } MatroskaAttachement;
156
157 typedef struct {
158     uint64_t start;
159     uint64_t end;
160     uint64_t uid;
161     char    *title;
162
163     AVChapter *chapter;
164 } MatroskaChapter;
165
166 typedef struct {
167     uint64_t track;
168     uint64_t pos;
169 } MatroskaIndexPos;
170
171 typedef struct {
172     uint64_t time;
173     EbmlList pos;
174 } MatroskaIndex;
175
176 typedef struct {
177     char *name;
178     char *string;
179     char *lang;
180     uint64_t def;
181     EbmlList sub;
182 } MatroskaTag;
183
184 typedef struct {
185     char    *type;
186     uint64_t typevalue;
187     uint64_t trackuid;
188     uint64_t chapteruid;
189     uint64_t attachuid;
190 } MatroskaTagTarget;
191
192 typedef struct {
193     MatroskaTagTarget target;
194     EbmlList tag;
195 } MatroskaTags;
196
197 typedef struct {
198     uint64_t id;
199     uint64_t pos;
200 } MatroskaSeekhead;
201
202 typedef struct {
203     uint64_t start;
204     uint64_t length;
205 } MatroskaLevel;
206
207 typedef struct {
208     AVFormatContext *ctx;
209
210     /* EBML stuff */
211     int num_levels;
212     MatroskaLevel levels[EBML_MAX_DEPTH];
213     int level_up;
214
215     uint64_t time_scale;
216     double   duration;
217     char    *title;
218     EbmlList tracks;
219     EbmlList attachments;
220     EbmlList chapters;
221     EbmlList index;
222     EbmlList tags;
223     EbmlList seekhead;
224
225     /* byte position of the segment inside the stream */
226     int64_t segment_start;
227
228     /* the packet queue */
229     AVPacket **packets;
230     int num_packets;
231     AVPacket *prev_pkt;
232
233     int done;
234     int has_cluster_id;
235
236     /* What to skip before effectively reading a packet. */
237     int skip_to_keyframe;
238     uint64_t skip_to_timecode;
239 } MatroskaDemuxContext;
240
241 typedef struct {
242     uint64_t duration;
243     int64_t  reference;
244     EbmlBin  bin;
245 } MatroskaBlock;
246
247 typedef struct {
248     uint64_t timecode;
249     EbmlList blocks;
250 } MatroskaCluster;
251
252 static EbmlSyntax ebml_header[] = {
253     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
254     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
255     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
256     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
257     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
258     { EBML_ID_EBMLVERSION,            EBML_NONE },
259     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
260     { 0 }
261 };
262
263 static EbmlSyntax ebml_syntax[] = {
264     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
265     { 0 }
266 };
267
268 static EbmlSyntax matroska_info[] = {
269     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
270     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
271     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
272     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
273     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
274     { MATROSKA_ID_DATEUTC,            EBML_NONE },
275     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
276     { 0 }
277 };
278
279 static EbmlSyntax matroska_track_video[] = {
280     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
281     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
282     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
283     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
284     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
285     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,fourcc) },
286     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
287     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
288     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
289     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
290     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
291     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
292     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_NONE },
293     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
294     { 0 }
295 };
296
297 static EbmlSyntax matroska_track_audio[] = {
298     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
299     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
300     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
301     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
302     { 0 }
303 };
304
305 static EbmlSyntax matroska_track_encoding_compression[] = {
306     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
307     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
308     { 0 }
309 };
310
311 static EbmlSyntax matroska_track_encoding[] = {
312     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
313     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
314     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
315     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
316     { 0 }
317 };
318
319 static EbmlSyntax matroska_track_encodings[] = {
320     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
321     { 0 }
322 };
323
324 static EbmlSyntax matroska_track[] = {
325     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
326     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
327     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
328     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
329     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
330     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
331     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
332     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
333     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
334     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
335     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
336     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
337     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
338     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
339     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_NONE },
340     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
341     { MATROSKA_ID_CODECNAME,            EBML_NONE },
342     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
343     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
344     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
345     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
346     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
347     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
348     { 0 }
349 };
350
351 static EbmlSyntax matroska_tracks[] = {
352     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
353     { 0 }
354 };
355
356 static EbmlSyntax matroska_attachment[] = {
357     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
358     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
359     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
360     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
361     { MATROSKA_ID_FILEDESC,           EBML_NONE },
362     { 0 }
363 };
364
365 static EbmlSyntax matroska_attachments[] = {
366     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
367     { 0 }
368 };
369
370 static EbmlSyntax matroska_chapter_display[] = {
371     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
372     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
373     { 0 }
374 };
375
376 static EbmlSyntax matroska_chapter_entry[] = {
377     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
378     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
379     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
380     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
381     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
382     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
383     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
384     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
385     { 0 }
386 };
387
388 static EbmlSyntax matroska_chapter[] = {
389     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
390     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
391     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
392     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
393     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
394     { 0 }
395 };
396
397 static EbmlSyntax matroska_chapters[] = {
398     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
399     { 0 }
400 };
401
402 static EbmlSyntax matroska_index_pos[] = {
403     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
404     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
405     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
406     { 0 }
407 };
408
409 static EbmlSyntax matroska_index_entry[] = {
410     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
411     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
412     { 0 }
413 };
414
415 static EbmlSyntax matroska_index[] = {
416     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
417     { 0 }
418 };
419
420 static EbmlSyntax matroska_simpletag[] = {
421     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
422     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
423     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
424     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
425     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
426     { 0 }
427 };
428
429 static EbmlSyntax matroska_tagtargets[] = {
430     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
431     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
432     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
433     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
434     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
435     { 0 }
436 };
437
438 static EbmlSyntax matroska_tag[] = {
439     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
440     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
441     { 0 }
442 };
443
444 static EbmlSyntax matroska_tags[] = {
445     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
446     { 0 }
447 };
448
449 static EbmlSyntax matroska_seekhead_entry[] = {
450     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
451     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
452     { 0 }
453 };
454
455 static EbmlSyntax matroska_seekhead[] = {
456     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
457     { 0 }
458 };
459
460 static EbmlSyntax matroska_segment[] = {
461     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
462     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
463     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
464     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
465     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
466     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
467     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
468     { MATROSKA_ID_CLUSTER,        EBML_STOP, 0, offsetof(MatroskaDemuxContext,has_cluster_id) },
469     { 0 }
470 };
471
472 static EbmlSyntax matroska_segments[] = {
473     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
474     { 0 }
475 };
476
477 static EbmlSyntax matroska_blockgroup[] = {
478     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
479     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
480     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration), {.u=AV_NOPTS_VALUE} },
481     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
482     { 0 }
483 };
484
485 static EbmlSyntax matroska_cluster[] = {
486     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
487     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
488     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
489     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
490     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
491     { 0 }
492 };
493
494 static EbmlSyntax matroska_clusters[] = {
495     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
496     { MATROSKA_ID_INFO,           EBML_NONE },
497     { MATROSKA_ID_CUES,           EBML_NONE },
498     { MATROSKA_ID_TAGS,           EBML_NONE },
499     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
500     { 0 }
501 };
502
503 /*
504  * Return: Whether we reached the end of a level in the hierarchy or not.
505  */
506 static int ebml_level_end(MatroskaDemuxContext *matroska)
507 {
508     ByteIOContext *pb = matroska->ctx->pb;
509     int64_t pos = url_ftell(pb);
510
511     if (matroska->num_levels > 0) {
512         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
513         if (pos - level->start >= level->length) {
514             matroska->num_levels--;
515             return 1;
516         }
517     }
518     return 0;
519 }
520
521 /*
522  * Read: an "EBML number", which is defined as a variable-length
523  * array of bytes. The first byte indicates the length by giving a
524  * number of 0-bits followed by a one. The position of the first
525  * "one" bit inside the first byte indicates the length of this
526  * number.
527  * Returns: number of bytes read, < 0 on error
528  */
529 static int ebml_read_num(MatroskaDemuxContext *matroska, ByteIOContext *pb,
530                          int max_size, uint64_t *number)
531 {
532     int len_mask = 0x80, read = 1, n = 1;
533     int64_t total = 0;
534
535     /* The first byte tells us the length in bytes - get_byte() can normally
536      * return 0, but since that's not a valid first ebmlID byte, we can
537      * use it safely here to catch EOS. */
538     if (!(total = get_byte(pb))) {
539         /* we might encounter EOS here */
540         if (!url_feof(pb)) {
541             int64_t pos = url_ftell(pb);
542             av_log(matroska->ctx, AV_LOG_ERROR,
543                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
544                    pos, pos);
545         }
546         return AVERROR(EIO); /* EOS or actual I/O error */
547     }
548
549     /* get the length of the EBML number */
550     while (read <= max_size && !(total & len_mask)) {
551         read++;
552         len_mask >>= 1;
553     }
554     if (read > max_size) {
555         int64_t pos = url_ftell(pb) - 1;
556         av_log(matroska->ctx, AV_LOG_ERROR,
557                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
558                (uint8_t) total, pos, pos);
559         return AVERROR_INVALIDDATA;
560     }
561
562     /* read out length */
563     total &= ~len_mask;
564     while (n++ < read)
565         total = (total << 8) | get_byte(pb);
566
567     *number = total;
568
569     return read;
570 }
571
572 /*
573  * Read the next element as an unsigned int.
574  * 0 is success, < 0 is failure.
575  */
576 static int ebml_read_uint(ByteIOContext *pb, int size, uint64_t *num)
577 {
578     int n = 0;
579
580     if (size < 1 || size > 8)
581         return AVERROR_INVALIDDATA;
582
583     /* big-endian ordering; build up number */
584     *num = 0;
585     while (n++ < size)
586         *num = (*num << 8) | get_byte(pb);
587
588     return 0;
589 }
590
591 /*
592  * Read the next element as a float.
593  * 0 is success, < 0 is failure.
594  */
595 static int ebml_read_float(ByteIOContext *pb, int size, double *num)
596 {
597     if (size == 4) {
598         *num= av_int2flt(get_be32(pb));
599     } else if(size==8){
600         *num= av_int2dbl(get_be64(pb));
601     } else
602         return AVERROR_INVALIDDATA;
603
604     return 0;
605 }
606
607 /*
608  * Read the next element as an ASCII string.
609  * 0 is success, < 0 is failure.
610  */
611 static int ebml_read_ascii(ByteIOContext *pb, int size, char **str)
612 {
613     av_free(*str);
614     /* EBML strings are usually not 0-terminated, so we allocate one
615      * byte more, read the string and NULL-terminate it ourselves. */
616     if (!(*str = av_malloc(size + 1)))
617         return AVERROR(ENOMEM);
618     if (get_buffer(pb, (uint8_t *) *str, size) != size) {
619         av_free(*str);
620         return AVERROR(EIO);
621     }
622     (*str)[size] = '\0';
623
624     return 0;
625 }
626
627 /*
628  * Read the next element as binary data.
629  * 0 is success, < 0 is failure.
630  */
631 static int ebml_read_binary(ByteIOContext *pb, int length, EbmlBin *bin)
632 {
633     av_free(bin->data);
634     if (!(bin->data = av_malloc(length)))
635         return AVERROR(ENOMEM);
636
637     bin->size = length;
638     bin->pos  = url_ftell(pb);
639     if (get_buffer(pb, bin->data, length) != length)
640         return AVERROR(EIO);
641
642     return 0;
643 }
644
645 /*
646  * Read the next element, but only the header. The contents
647  * are supposed to be sub-elements which can be read separately.
648  * 0 is success, < 0 is failure.
649  */
650 static int ebml_read_master(MatroskaDemuxContext *matroska, int length)
651 {
652     ByteIOContext *pb = matroska->ctx->pb;
653     MatroskaLevel *level;
654
655     if (matroska->num_levels >= EBML_MAX_DEPTH) {
656         av_log(matroska->ctx, AV_LOG_ERROR,
657                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
658         return AVERROR(ENOSYS);
659     }
660
661     level = &matroska->levels[matroska->num_levels++];
662     level->start = url_ftell(pb);
663     level->length = length;
664
665     return 0;
666 }
667
668 /*
669  * Read signed/unsigned "EBML" numbers.
670  * Return: number of bytes processed, < 0 on error
671  */
672 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
673                                  uint8_t *data, uint32_t size, uint64_t *num)
674 {
675     ByteIOContext pb;
676     init_put_byte(&pb, data, size, 0, NULL, NULL, NULL, NULL);
677     return ebml_read_num(matroska, &pb, 8, num);
678 }
679
680 /*
681  * Same as above, but signed.
682  */
683 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
684                                  uint8_t *data, uint32_t size, int64_t *num)
685 {
686     uint64_t unum;
687     int res;
688
689     /* read as unsigned number first */
690     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
691         return res;
692
693     /* make signed (weird way) */
694     *num = unum - ((1LL << (7*res - 1)) - 1);
695
696     return res;
697 }
698
699 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
700                            EbmlSyntax *syntax, void *data);
701
702 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
703                          uint32_t id, void *data)
704 {
705     int i;
706     for (i=0; syntax[i].id; i++)
707         if (id == syntax[i].id)
708             break;
709     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32)
710         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
711     return ebml_parse_elem(matroska, &syntax[i], data);
712 }
713
714 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
715                       void *data)
716 {
717     uint64_t id;
718     int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
719     id |= 1 << 7*res;
720     return res < 0 ? res : ebml_parse_id(matroska, syntax, id, data);
721 }
722
723 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
724                            void *data)
725 {
726     int i, res = 0;
727
728     for (i=0; syntax[i].id; i++)
729         switch (syntax[i].type) {
730         case EBML_UINT:
731             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
732             break;
733         case EBML_FLOAT:
734             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
735             break;
736         case EBML_STR:
737         case EBML_UTF8:
738             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
739             break;
740         }
741
742     while (!res && !ebml_level_end(matroska))
743         res = ebml_parse(matroska, syntax, data);
744
745     return res;
746 }
747
748 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
749                            EbmlSyntax *syntax, void *data)
750 {
751     ByteIOContext *pb = matroska->ctx->pb;
752     uint32_t id = syntax->id;
753     uint64_t length;
754     int res;
755
756     data = (char *)data + syntax->data_offset;
757     if (syntax->list_elem_size) {
758         EbmlList *list = data;
759         list->elem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
760         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
761         memset(data, 0, syntax->list_elem_size);
762         list->nb_elem++;
763     }
764
765     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP)
766         if ((res = ebml_read_num(matroska, pb, 8, &length)) < 0)
767             return res;
768
769     switch (syntax->type) {
770     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
771     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
772     case EBML_STR:
773     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
774     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
775     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
776                          return res;
777                      if (id == MATROSKA_ID_SEGMENT)
778                          matroska->segment_start = url_ftell(matroska->ctx->pb);
779                      return ebml_parse_nest(matroska, syntax->def.n, data);
780     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
781     case EBML_STOP:  *(int *)data = 1;      return 1;
782     default:         return url_fseek(pb,length,SEEK_CUR)<0 ? AVERROR(EIO) : 0;
783     }
784     if (res == AVERROR_INVALIDDATA)
785         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
786     else if (res == AVERROR(EIO))
787         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
788     return res;
789 }
790
791 static void ebml_free(EbmlSyntax *syntax, void *data)
792 {
793     int i, j;
794     for (i=0; syntax[i].id; i++) {
795         void *data_off = (char *)data + syntax[i].data_offset;
796         switch (syntax[i].type) {
797         case EBML_STR:
798         case EBML_UTF8:  av_freep(data_off);                      break;
799         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
800         case EBML_NEST:
801             if (syntax[i].list_elem_size) {
802                 EbmlList *list = data_off;
803                 char *ptr = list->elem;
804                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
805                     ebml_free(syntax[i].def.n, ptr);
806                 av_free(list->elem);
807             } else
808                 ebml_free(syntax[i].def.n, data_off);
809         default:  break;
810         }
811     }
812 }
813
814
815 /*
816  * Autodetecting...
817  */
818 static int matroska_probe(AVProbeData *p)
819 {
820     uint64_t total = 0;
821     int len_mask = 0x80, size = 1, n = 1;
822     static const char probe_data[] = "matroska";
823
824     /* EBML header? */
825     if (AV_RB32(p->buf) != EBML_ID_HEADER)
826         return 0;
827
828     /* length of header */
829     total = p->buf[4];
830     while (size <= 8 && !(total & len_mask)) {
831         size++;
832         len_mask >>= 1;
833     }
834     if (size > 8)
835       return 0;
836     total &= (len_mask - 1);
837     while (n < size)
838         total = (total << 8) | p->buf[4 + n++];
839
840     /* Does the probe data contain the whole header? */
841     if (p->buf_size < 4 + size + total)
842       return 0;
843
844     /* The header must contain the document type 'matroska'. For now,
845      * we don't parse the whole header but simply check for the
846      * availability of that array of characters inside the header.
847      * Not fully fool-proof, but good enough. */
848     for (n = 4+size; n <= 4+size+total-(sizeof(probe_data)-1); n++)
849         if (!memcmp(p->buf+n, probe_data, sizeof(probe_data)-1))
850             return AVPROBE_SCORE_MAX;
851
852     return 0;
853 }
854
855 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
856                                                  int num)
857 {
858     MatroskaTrack *tracks = matroska->tracks.elem;
859     int i;
860
861     for (i=0; i < matroska->tracks.nb_elem; i++)
862         if (tracks[i].num == num)
863             return &tracks[i];
864
865     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
866     return NULL;
867 }
868
869 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
870                                   MatroskaTrack *track)
871 {
872     MatroskaTrackEncoding *encodings = track->encodings.elem;
873     uint8_t* data = *buf;
874     int isize = *buf_size;
875     uint8_t* pkt_data = NULL;
876     int pkt_size = isize;
877     int result = 0;
878     int olen;
879
880     switch (encodings[0].compression.algo) {
881     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
882         return encodings[0].compression.settings.size;
883     case MATROSKA_TRACK_ENCODING_COMP_LZO:
884         do {
885             olen = pkt_size *= 3;
886             pkt_data = av_realloc(pkt_data, pkt_size+AV_LZO_OUTPUT_PADDING);
887             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
888         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
889         if (result)
890             goto failed;
891         pkt_size -= olen;
892         break;
893 #if CONFIG_ZLIB
894     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
895         z_stream zstream = {0};
896         if (inflateInit(&zstream) != Z_OK)
897             return -1;
898         zstream.next_in = data;
899         zstream.avail_in = isize;
900         do {
901             pkt_size *= 3;
902             pkt_data = av_realloc(pkt_data, pkt_size);
903             zstream.avail_out = pkt_size - zstream.total_out;
904             zstream.next_out = pkt_data + zstream.total_out;
905             result = inflate(&zstream, Z_NO_FLUSH);
906         } while (result==Z_OK && pkt_size<10000000);
907         pkt_size = zstream.total_out;
908         inflateEnd(&zstream);
909         if (result != Z_STREAM_END)
910             goto failed;
911         break;
912     }
913 #endif
914 #if CONFIG_BZLIB
915     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
916         bz_stream bzstream = {0};
917         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
918             return -1;
919         bzstream.next_in = data;
920         bzstream.avail_in = isize;
921         do {
922             pkt_size *= 3;
923             pkt_data = av_realloc(pkt_data, pkt_size);
924             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
925             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
926             result = BZ2_bzDecompress(&bzstream);
927         } while (result==BZ_OK && pkt_size<10000000);
928         pkt_size = bzstream.total_out_lo32;
929         BZ2_bzDecompressEnd(&bzstream);
930         if (result != BZ_STREAM_END)
931             goto failed;
932         break;
933     }
934 #endif
935     default:
936         return -1;
937     }
938
939     *buf = pkt_data;
940     *buf_size = pkt_size;
941     return 0;
942  failed:
943     av_free(pkt_data);
944     return -1;
945 }
946
947 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
948                                     AVPacket *pkt, uint64_t display_duration)
949 {
950     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
951     for (; *ptr!=',' && ptr<end-1; ptr++);
952     if (*ptr == ',')
953         layer = ++ptr;
954     for (; *ptr!=',' && ptr<end-1; ptr++);
955     if (*ptr == ',') {
956         int64_t end_pts = pkt->pts + display_duration;
957         int sc = matroska->time_scale * pkt->pts / 10000000;
958         int ec = matroska->time_scale * end_pts  / 10000000;
959         int sh, sm, ss, eh, em, es, len;
960         sh = sc/360000;  sc -= 360000*sh;
961         sm = sc/  6000;  sc -=   6000*sm;
962         ss = sc/   100;  sc -=    100*ss;
963         eh = ec/360000;  ec -= 360000*eh;
964         em = ec/  6000;  ec -=   6000*em;
965         es = ec/   100;  ec -=    100*es;
966         *ptr++ = '\0';
967         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
968         if (!(line = av_malloc(len)))
969             return;
970         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
971                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
972         av_free(pkt->data);
973         pkt->data = line;
974         pkt->size = strlen(line);
975     }
976 }
977
978 static void matroska_merge_packets(AVPacket *out, AVPacket *in)
979 {
980     out->data = av_realloc(out->data, out->size+in->size);
981     memcpy(out->data+out->size, in->data, in->size);
982     out->size += in->size;
983     av_destruct_packet(in);
984     av_free(in);
985 }
986
987 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
988                                  AVMetadata **metadata, char *prefix)
989 {
990     MatroskaTag *tags = list->elem;
991     char key[1024];
992     int i;
993
994     for (i=0; i < list->nb_elem; i++) {
995         const char *lang = strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
996         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
997         else         av_strlcpy(key, tags[i].name, sizeof(key));
998         if (tags[i].def || !lang) {
999         av_metadata_set(metadata, key, tags[i].string);
1000         if (tags[i].sub.nb_elem)
1001             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1002         }
1003         if (lang) {
1004             av_strlcat(key, "-", sizeof(key));
1005             av_strlcat(key, lang, sizeof(key));
1006             av_metadata_set(metadata, key, tags[i].string);
1007             if (tags[i].sub.nb_elem)
1008                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1009         }
1010     }
1011 }
1012
1013 static void matroska_convert_tags(AVFormatContext *s)
1014 {
1015     MatroskaDemuxContext *matroska = s->priv_data;
1016     MatroskaTags *tags = matroska->tags.elem;
1017     int i, j;
1018
1019     for (i=0; i < matroska->tags.nb_elem; i++) {
1020         if (tags[i].target.attachuid) {
1021             MatroskaAttachement *attachment = matroska->attachments.elem;
1022             for (j=0; j<matroska->attachments.nb_elem; j++)
1023                 if (attachment[j].uid == tags[i].target.attachuid)
1024                     matroska_convert_tag(s, &tags[i].tag,
1025                                          &attachment[j].stream->metadata, NULL);
1026         } else if (tags[i].target.chapteruid) {
1027             MatroskaChapter *chapter = matroska->chapters.elem;
1028             for (j=0; j<matroska->chapters.nb_elem; j++)
1029                 if (chapter[j].uid == tags[i].target.chapteruid)
1030                     matroska_convert_tag(s, &tags[i].tag,
1031                                          &chapter[j].chapter->metadata, NULL);
1032         } else if (tags[i].target.trackuid) {
1033             MatroskaTrack *track = matroska->tracks.elem;
1034             for (j=0; j<matroska->tracks.nb_elem; j++)
1035                 if (track[j].uid == tags[i].target.trackuid)
1036                     matroska_convert_tag(s, &tags[i].tag,
1037                                          &track[j].stream->metadata, NULL);
1038         } else {
1039             matroska_convert_tag(s, &tags[i].tag, &s->metadata, NULL);
1040         }
1041     }
1042 }
1043
1044 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1045 {
1046     EbmlList *seekhead_list = &matroska->seekhead;
1047     MatroskaSeekhead *seekhead = seekhead_list->elem;
1048     uint32_t level_up = matroska->level_up;
1049     int64_t before_pos = url_ftell(matroska->ctx->pb);
1050     MatroskaLevel level;
1051     int i;
1052
1053     for (i=0; i<seekhead_list->nb_elem; i++) {
1054         int64_t offset = seekhead[i].pos + matroska->segment_start;
1055
1056         if (seekhead[i].pos <= before_pos
1057             || seekhead[i].id == MATROSKA_ID_SEEKHEAD
1058             || seekhead[i].id == MATROSKA_ID_CLUSTER)
1059             continue;
1060
1061         /* seek */
1062         if (url_fseek(matroska->ctx->pb, offset, SEEK_SET) != offset)
1063             continue;
1064
1065         /* We don't want to lose our seekhead level, so we add
1066          * a dummy. This is a crude hack. */
1067         if (matroska->num_levels == EBML_MAX_DEPTH) {
1068             av_log(matroska->ctx, AV_LOG_INFO,
1069                    "Max EBML element depth (%d) reached, "
1070                    "cannot parse further.\n", EBML_MAX_DEPTH);
1071             break;
1072         }
1073
1074         level.start = 0;
1075         level.length = (uint64_t)-1;
1076         matroska->levels[matroska->num_levels] = level;
1077         matroska->num_levels++;
1078
1079         ebml_parse(matroska, matroska_segment, matroska);
1080
1081         /* remove dummy level */
1082         while (matroska->num_levels) {
1083             uint64_t length = matroska->levels[--matroska->num_levels].length;
1084             if (length == (uint64_t)-1)
1085                 break;
1086         }
1087     }
1088
1089     /* seek back */
1090     url_fseek(matroska->ctx->pb, before_pos, SEEK_SET);
1091     matroska->level_up = level_up;
1092 }
1093
1094 static int matroska_aac_profile(char *codec_id)
1095 {
1096     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1097     int profile;
1098
1099     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1100         if (strstr(codec_id, aac_profiles[profile]))
1101             break;
1102     return profile + 1;
1103 }
1104
1105 static int matroska_aac_sri(int samplerate)
1106 {
1107     int sri;
1108
1109     for (sri=0; sri<FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
1110         if (ff_mpeg4audio_sample_rates[sri] == samplerate)
1111             break;
1112     return sri;
1113 }
1114
1115 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
1116 {
1117     MatroskaDemuxContext *matroska = s->priv_data;
1118     EbmlList *attachements_list = &matroska->attachments;
1119     MatroskaAttachement *attachements;
1120     EbmlList *chapters_list = &matroska->chapters;
1121     MatroskaChapter *chapters;
1122     MatroskaTrack *tracks;
1123     EbmlList *index_list;
1124     MatroskaIndex *index;
1125     int index_scale = 1;
1126     uint64_t max_start = 0;
1127     Ebml ebml = { 0 };
1128     AVStream *st;
1129     int i, j;
1130
1131     matroska->ctx = s;
1132
1133     /* First read the EBML header. */
1134     if (ebml_parse(matroska, ebml_syntax, &ebml)
1135         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1136         || ebml.id_length > sizeof(uint32_t) || strcmp(ebml.doctype, "matroska")
1137         || ebml.doctype_version > 2) {
1138         av_log(matroska->ctx, AV_LOG_ERROR,
1139                "EBML header using unsupported features\n"
1140                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1141                ebml.version, ebml.doctype, ebml.doctype_version);
1142         return AVERROR_NOFMT;
1143     }
1144     ebml_free(ebml_syntax, &ebml);
1145
1146     /* The next thing is a segment. */
1147     if (ebml_parse(matroska, matroska_segments, matroska) < 0)
1148         return -1;
1149     matroska_execute_seekhead(matroska);
1150
1151     if (matroska->duration)
1152         matroska->ctx->duration = matroska->duration * matroska->time_scale
1153                                   * 1000 / AV_TIME_BASE;
1154     av_metadata_set(&s->metadata, "title", matroska->title);
1155
1156     tracks = matroska->tracks.elem;
1157     for (i=0; i < matroska->tracks.nb_elem; i++) {
1158         MatroskaTrack *track = &tracks[i];
1159         enum CodecID codec_id = CODEC_ID_NONE;
1160         EbmlList *encodings_list = &tracks->encodings;
1161         MatroskaTrackEncoding *encodings = encodings_list->elem;
1162         uint8_t *extradata = NULL;
1163         int extradata_size = 0;
1164         int extradata_offset = 0;
1165         ByteIOContext b;
1166
1167         /* Apply some sanity checks. */
1168         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1169             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1170             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1171             av_log(matroska->ctx, AV_LOG_INFO,
1172                    "Unknown or unsupported track type %"PRIu64"\n",
1173                    track->type);
1174             continue;
1175         }
1176         if (track->codec_id == NULL)
1177             continue;
1178
1179         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1180             if (!track->default_duration)
1181                 track->default_duration = 1000000000/track->video.frame_rate;
1182             if (!track->video.display_width)
1183                 track->video.display_width = track->video.pixel_width;
1184             if (!track->video.display_height)
1185                 track->video.display_height = track->video.pixel_height;
1186         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1187             if (!track->audio.out_samplerate)
1188                 track->audio.out_samplerate = track->audio.samplerate;
1189         }
1190         if (encodings_list->nb_elem > 1) {
1191             av_log(matroska->ctx, AV_LOG_ERROR,
1192                    "Multiple combined encodings no supported");
1193         } else if (encodings_list->nb_elem == 1) {
1194             if (encodings[0].type ||
1195                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1196 #if CONFIG_ZLIB
1197                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1198 #endif
1199 #if CONFIG_BZLIB
1200                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1201 #endif
1202                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1203                 encodings[0].scope = 0;
1204                 av_log(matroska->ctx, AV_LOG_ERROR,
1205                        "Unsupported encoding type");
1206             } else if (track->codec_priv.size && encodings[0].scope&2) {
1207                 uint8_t *codec_priv = track->codec_priv.data;
1208                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1209                                                     &track->codec_priv.size,
1210                                                     track);
1211                 if (offset < 0) {
1212                     track->codec_priv.data = NULL;
1213                     track->codec_priv.size = 0;
1214                     av_log(matroska->ctx, AV_LOG_ERROR,
1215                            "Failed to decode codec private data\n");
1216                 } else if (offset > 0) {
1217                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1218                     memcpy(track->codec_priv.data,
1219                            encodings[0].compression.settings.data, offset);
1220                     memcpy(track->codec_priv.data+offset, codec_priv,
1221                            track->codec_priv.size);
1222                     track->codec_priv.size += offset;
1223                 }
1224                 if (codec_priv != track->codec_priv.data)
1225                     av_free(codec_priv);
1226             }
1227         }
1228
1229         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1230             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1231                         strlen(ff_mkv_codec_tags[j].str))){
1232                 codec_id= ff_mkv_codec_tags[j].id;
1233                 break;
1234             }
1235         }
1236
1237         st = track->stream = av_new_stream(s, 0);
1238         if (st == NULL)
1239             return AVERROR(ENOMEM);
1240
1241         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1242             && track->codec_priv.size >= 40
1243             && track->codec_priv.data != NULL) {
1244             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1245             codec_id = codec_get_id(codec_bmp_tags, track->video.fourcc);
1246         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1247                    && track->codec_priv.size >= 18
1248                    && track->codec_priv.data != NULL) {
1249             init_put_byte(&b, track->codec_priv.data, track->codec_priv.size,
1250                           URL_RDONLY, NULL, NULL, NULL, NULL);
1251             get_wav_header(&b, st->codec, track->codec_priv.size);
1252             codec_id = st->codec->codec_id;
1253             extradata_offset = 18;
1254             track->codec_priv.size -= extradata_offset;
1255         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1256                    && (track->codec_priv.size >= 86)
1257                    && (track->codec_priv.data != NULL)) {
1258             track->video.fourcc = AV_RL32(track->codec_priv.data);
1259             codec_id=codec_get_id(codec_movvideo_tags, track->video.fourcc);
1260         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1261             switch (track->audio.bitdepth) {
1262             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1263             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1264             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1265             }
1266         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1267             switch (track->audio.bitdepth) {
1268             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1269             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1270             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1271             }
1272         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1273             codec_id = CODEC_ID_PCM_F64LE;
1274         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1275             int profile = matroska_aac_profile(track->codec_id);
1276             int sri = matroska_aac_sri(track->audio.samplerate);
1277             extradata = av_malloc(5);
1278             if (extradata == NULL)
1279                 return AVERROR(ENOMEM);
1280             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1281             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1282             if (strstr(track->codec_id, "SBR")) {
1283                 sri = matroska_aac_sri(track->audio.out_samplerate);
1284                 extradata[2] = 0x56;
1285                 extradata[3] = 0xE5;
1286                 extradata[4] = 0x80 | (sri<<3);
1287                 extradata_size = 5;
1288             } else
1289                 extradata_size = 2;
1290         } else if (codec_id == CODEC_ID_TTA) {
1291             extradata_size = 30;
1292             extradata = av_mallocz(extradata_size);
1293             if (extradata == NULL)
1294                 return AVERROR(ENOMEM);
1295             init_put_byte(&b, extradata, extradata_size, 1,
1296                           NULL, NULL, NULL, NULL);
1297             put_buffer(&b, "TTA1", 4);
1298             put_le16(&b, 1);
1299             put_le16(&b, track->audio.channels);
1300             put_le16(&b, track->audio.bitdepth);
1301             put_le32(&b, track->audio.out_samplerate);
1302             put_le32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1303         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1304                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1305             extradata_offset = 26;
1306             track->codec_priv.size -= extradata_offset;
1307         } else if (codec_id == CODEC_ID_RA_144) {
1308             track->audio.out_samplerate = 8000;
1309             track->audio.channels = 1;
1310         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1311                    codec_id == CODEC_ID_ATRAC3) {
1312             init_put_byte(&b, track->codec_priv.data,track->codec_priv.size,
1313                           0, NULL, NULL, NULL, NULL);
1314             url_fskip(&b, 24);
1315             track->audio.coded_framesize = get_be32(&b);
1316             url_fskip(&b, 12);
1317             track->audio.sub_packet_h    = get_be16(&b);
1318             track->audio.frame_size      = get_be16(&b);
1319             track->audio.sub_packet_size = get_be16(&b);
1320             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1321             if (codec_id == CODEC_ID_RA_288) {
1322                 st->codec->block_align = track->audio.coded_framesize;
1323                 track->codec_priv.size = 0;
1324             } else {
1325                 st->codec->block_align = track->audio.sub_packet_size;
1326                 extradata_offset = 78;
1327                 track->codec_priv.size -= extradata_offset;
1328             }
1329         }
1330
1331         if (codec_id == CODEC_ID_NONE)
1332             av_log(matroska->ctx, AV_LOG_INFO,
1333                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1334
1335         if (track->time_scale < 0.01)
1336             track->time_scale = 1.0;
1337         av_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1338
1339         st->codec->codec_id = codec_id;
1340         st->start_time = 0;
1341         if (strcmp(track->language, "und"))
1342             av_metadata_set(&st->metadata, "language", track->language);
1343         av_metadata_set(&st->metadata, "description", track->name);
1344
1345         if (track->flag_default)
1346             st->disposition |= AV_DISPOSITION_DEFAULT;
1347
1348         if (track->default_duration)
1349             av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
1350                       track->default_duration, 1000000000, 30000);
1351
1352         if(extradata){
1353             st->codec->extradata = extradata;
1354             st->codec->extradata_size = extradata_size;
1355         } else if(track->codec_priv.data && track->codec_priv.size > 0){
1356             st->codec->extradata = av_mallocz(track->codec_priv.size +
1357                                               FF_INPUT_BUFFER_PADDING_SIZE);
1358             if(st->codec->extradata == NULL)
1359                 return AVERROR(ENOMEM);
1360             st->codec->extradata_size = track->codec_priv.size;
1361             memcpy(st->codec->extradata,
1362                    track->codec_priv.data + extradata_offset,
1363                    track->codec_priv.size);
1364         }
1365
1366         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1367             st->codec->codec_type = CODEC_TYPE_VIDEO;
1368             st->codec->codec_tag  = track->video.fourcc;
1369             st->codec->width  = track->video.pixel_width;
1370             st->codec->height = track->video.pixel_height;
1371             av_reduce(&st->sample_aspect_ratio.num,
1372                       &st->sample_aspect_ratio.den,
1373                       st->codec->height * track->video.display_width,
1374                       st->codec-> width * track->video.display_height,
1375                       255);
1376             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1377         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1378             st->codec->codec_type = CODEC_TYPE_AUDIO;
1379             st->codec->sample_rate = track->audio.out_samplerate;
1380             st->codec->channels = track->audio.channels;
1381         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1382             st->codec->codec_type = CODEC_TYPE_SUBTITLE;
1383         }
1384     }
1385
1386     attachements = attachements_list->elem;
1387     for (j=0; j<attachements_list->nb_elem; j++) {
1388         if (!(attachements[j].filename && attachements[j].mime &&
1389               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1390             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1391         } else {
1392             AVStream *st = av_new_stream(s, 0);
1393             if (st == NULL)
1394                 break;
1395             av_metadata_set(&st->metadata, "filename",attachements[j].filename);
1396             st->codec->codec_id = CODEC_ID_NONE;
1397             st->codec->codec_type = CODEC_TYPE_ATTACHMENT;
1398             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1399             if(st->codec->extradata == NULL)
1400                 break;
1401             st->codec->extradata_size = attachements[j].bin.size;
1402             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1403
1404             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1405                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1406                              strlen(ff_mkv_mime_tags[i].str))) {
1407                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1408                     break;
1409                 }
1410             }
1411             attachements[j].stream = st;
1412         }
1413     }
1414
1415     chapters = chapters_list->elem;
1416     for (i=0; i<chapters_list->nb_elem; i++)
1417         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1418             && (max_start==0 || chapters[i].start > max_start)) {
1419             chapters[i].chapter =
1420             ff_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1421                            chapters[i].start, chapters[i].end,
1422                            chapters[i].title);
1423             av_metadata_set(&chapters[i].chapter->metadata,
1424                             "title", chapters[i].title);
1425             max_start = chapters[i].start;
1426         }
1427
1428     index_list = &matroska->index;
1429     index = index_list->elem;
1430     if (index_list->nb_elem
1431         && index[0].time > 100000000000000/matroska->time_scale) {
1432         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1433         index_scale = matroska->time_scale;
1434     }
1435     for (i=0; i<index_list->nb_elem; i++) {
1436         EbmlList *pos_list = &index[i].pos;
1437         MatroskaIndexPos *pos = pos_list->elem;
1438         for (j=0; j<pos_list->nb_elem; j++) {
1439             MatroskaTrack *track = matroska_find_track_by_num(matroska,
1440                                                               pos[j].track);
1441             if (track && track->stream)
1442                 av_add_index_entry(track->stream,
1443                                    pos[j].pos + matroska->segment_start,
1444                                    index[i].time/index_scale, 0, 0,
1445                                    AVINDEX_KEYFRAME);
1446         }
1447     }
1448
1449     matroska_convert_tags(s);
1450
1451     return 0;
1452 }
1453
1454 /*
1455  * Put one packet in an application-supplied AVPacket struct.
1456  * Returns 0 on success or -1 on failure.
1457  */
1458 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1459                                    AVPacket *pkt)
1460 {
1461     if (matroska->num_packets > 0) {
1462         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1463         av_free(matroska->packets[0]);
1464         if (matroska->num_packets > 1) {
1465             memmove(&matroska->packets[0], &matroska->packets[1],
1466                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1467             matroska->packets =
1468                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
1469                            sizeof(AVPacket *));
1470         } else {
1471             av_freep(&matroska->packets);
1472         }
1473         matroska->num_packets--;
1474         return 0;
1475     }
1476
1477     return -1;
1478 }
1479
1480 /*
1481  * Free all packets in our internal queue.
1482  */
1483 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1484 {
1485     if (matroska->packets) {
1486         int n;
1487         for (n = 0; n < matroska->num_packets; n++) {
1488             av_free_packet(matroska->packets[n]);
1489             av_free(matroska->packets[n]);
1490         }
1491         av_freep(&matroska->packets);
1492         matroska->num_packets = 0;
1493     }
1494 }
1495
1496 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1497                                 int size, int64_t pos, uint64_t cluster_time,
1498                                 uint64_t duration, int is_keyframe,
1499                                 int64_t cluster_pos)
1500 {
1501     uint64_t timecode = AV_NOPTS_VALUE;
1502     MatroskaTrack *track;
1503     int res = 0;
1504     AVStream *st;
1505     AVPacket *pkt;
1506     int16_t block_time;
1507     uint32_t *lace_size = NULL;
1508     int n, flags, laces = 0;
1509     uint64_t num;
1510
1511     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1512         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1513         return res;
1514     }
1515     data += n;
1516     size -= n;
1517
1518     track = matroska_find_track_by_num(matroska, num);
1519     if (size <= 3 || !track || !track->stream) {
1520         av_log(matroska->ctx, AV_LOG_INFO,
1521                "Invalid stream %"PRIu64" or size %u\n", num, size);
1522         return res;
1523     }
1524     st = track->stream;
1525     if (st->discard >= AVDISCARD_ALL)
1526         return res;
1527     if (duration == AV_NOPTS_VALUE)
1528         duration = track->default_duration / matroska->time_scale;
1529
1530     block_time = AV_RB16(data);
1531     data += 2;
1532     flags = *data++;
1533     size -= 3;
1534     if (is_keyframe == -1)
1535         is_keyframe = flags & 0x80 ? PKT_FLAG_KEY : 0;
1536
1537     if (cluster_time != (uint64_t)-1
1538         && (block_time >= 0 || cluster_time >= -block_time)) {
1539         timecode = cluster_time + block_time;
1540         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1541             && timecode < track->end_timecode)
1542             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1543         if (is_keyframe)
1544             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1545         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1546     }
1547
1548     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1549         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1550             return res;
1551         matroska->skip_to_keyframe = 0;
1552     }
1553
1554     switch ((flags & 0x06) >> 1) {
1555         case 0x0: /* no lacing */
1556             laces = 1;
1557             lace_size = av_mallocz(sizeof(int));
1558             lace_size[0] = size;
1559             break;
1560
1561         case 0x1: /* Xiph lacing */
1562         case 0x2: /* fixed-size lacing */
1563         case 0x3: /* EBML lacing */
1564             assert(size>0); // size <=3 is checked before size-=3 above
1565             laces = (*data) + 1;
1566             data += 1;
1567             size -= 1;
1568             lace_size = av_mallocz(laces * sizeof(int));
1569
1570             switch ((flags & 0x06) >> 1) {
1571                 case 0x1: /* Xiph lacing */ {
1572                     uint8_t temp;
1573                     uint32_t total = 0;
1574                     for (n = 0; res == 0 && n < laces - 1; n++) {
1575                         while (1) {
1576                             if (size == 0) {
1577                                 res = -1;
1578                                 break;
1579                             }
1580                             temp = *data;
1581                             lace_size[n] += temp;
1582                             data += 1;
1583                             size -= 1;
1584                             if (temp != 0xff)
1585                                 break;
1586                         }
1587                         total += lace_size[n];
1588                     }
1589                     lace_size[n] = size - total;
1590                     break;
1591                 }
1592
1593                 case 0x2: /* fixed-size lacing */
1594                     for (n = 0; n < laces; n++)
1595                         lace_size[n] = size / laces;
1596                     break;
1597
1598                 case 0x3: /* EBML lacing */ {
1599                     uint32_t total;
1600                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1601                     if (n < 0) {
1602                         av_log(matroska->ctx, AV_LOG_INFO,
1603                                "EBML block data error\n");
1604                         break;
1605                     }
1606                     data += n;
1607                     size -= n;
1608                     total = lace_size[0] = num;
1609                     for (n = 1; res == 0 && n < laces - 1; n++) {
1610                         int64_t snum;
1611                         int r;
1612                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1613                         if (r < 0) {
1614                             av_log(matroska->ctx, AV_LOG_INFO,
1615                                    "EBML block data error\n");
1616                             break;
1617                         }
1618                         data += r;
1619                         size -= r;
1620                         lace_size[n] = lace_size[n - 1] + snum;
1621                         total += lace_size[n];
1622                     }
1623                     lace_size[n] = size - total;
1624                     break;
1625                 }
1626             }
1627             break;
1628     }
1629
1630     if (res == 0) {
1631         for (n = 0; n < laces; n++) {
1632             if (st->codec->codec_id == CODEC_ID_RA_288 ||
1633                 st->codec->codec_id == CODEC_ID_COOK ||
1634                 st->codec->codec_id == CODEC_ID_ATRAC3) {
1635                 int a = st->codec->block_align;
1636                 int sps = track->audio.sub_packet_size;
1637                 int cfs = track->audio.coded_framesize;
1638                 int h = track->audio.sub_packet_h;
1639                 int y = track->audio.sub_packet_cnt;
1640                 int w = track->audio.frame_size;
1641                 int x;
1642
1643                 if (!track->audio.pkt_cnt) {
1644                     if (st->codec->codec_id == CODEC_ID_RA_288)
1645                         for (x=0; x<h/2; x++)
1646                             memcpy(track->audio.buf+x*2*w+y*cfs,
1647                                    data+x*cfs, cfs);
1648                     else
1649                         for (x=0; x<w/sps; x++)
1650                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1651
1652                     if (++track->audio.sub_packet_cnt >= h) {
1653                         track->audio.sub_packet_cnt = 0;
1654                         track->audio.pkt_cnt = h*w / a;
1655                     }
1656                 }
1657                 while (track->audio.pkt_cnt) {
1658                     pkt = av_mallocz(sizeof(AVPacket));
1659                     av_new_packet(pkt, a);
1660                     memcpy(pkt->data, track->audio.buf
1661                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1662                     pkt->pos = pos;
1663                     pkt->stream_index = st->index;
1664                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1665                 }
1666             } else {
1667                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1668                 int offset = 0, pkt_size = lace_size[n];
1669                 uint8_t *pkt_data = data;
1670
1671                 if (encodings && encodings->scope & 1) {
1672                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1673                     if (offset < 0)
1674                         continue;
1675                 }
1676
1677                 pkt = av_mallocz(sizeof(AVPacket));
1678                 /* XXX: prevent data copy... */
1679                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1680                     av_free(pkt);
1681                     res = AVERROR(ENOMEM);
1682                     n = laces-1;
1683                     break;
1684                 }
1685                 if (offset)
1686                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1687                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1688
1689                 if (pkt_data != data)
1690                     av_free(pkt_data);
1691
1692                 if (n == 0)
1693                     pkt->flags = is_keyframe;
1694                 pkt->stream_index = st->index;
1695
1696                 pkt->pts = timecode;
1697                 pkt->pos = pos;
1698                 if (st->codec->codec_id == CODEC_ID_TEXT)
1699                     pkt->convergence_duration = duration;
1700                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1701                     pkt->duration = duration;
1702
1703                 if (st->codec->codec_id == CODEC_ID_SSA)
1704                     matroska_fix_ass_packet(matroska, pkt, duration);
1705
1706                 if (matroska->prev_pkt &&
1707                     timecode != AV_NOPTS_VALUE &&
1708                     matroska->prev_pkt->pts == timecode &&
1709                     matroska->prev_pkt->stream_index == st->index)
1710                     matroska_merge_packets(matroska->prev_pkt, pkt);
1711                 else {
1712                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1713                     matroska->prev_pkt = pkt;
1714                 }
1715             }
1716
1717             if (timecode != AV_NOPTS_VALUE)
1718                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1719             data += lace_size[n];
1720         }
1721     }
1722
1723     av_free(lace_size);
1724     return res;
1725 }
1726
1727 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1728 {
1729     MatroskaCluster cluster = { 0 };
1730     EbmlList *blocks_list;
1731     MatroskaBlock *blocks;
1732     int i, res;
1733     int64_t pos = url_ftell(matroska->ctx->pb);
1734     matroska->prev_pkt = NULL;
1735     if (matroska->has_cluster_id){
1736         /* For the first cluster we parse, its ID was already read as
1737            part of matroska_read_header(), so don't read it again */
1738         res = ebml_parse_id(matroska, matroska_clusters,
1739                             MATROSKA_ID_CLUSTER, &cluster);
1740         pos -= 4;  /* sizeof the ID which was already read */
1741         matroska->has_cluster_id = 0;
1742     } else
1743         res = ebml_parse(matroska, matroska_clusters, &cluster);
1744     blocks_list = &cluster.blocks;
1745     blocks = blocks_list->elem;
1746     for (i=0; i<blocks_list->nb_elem; i++)
1747         if (blocks[i].bin.size > 0)
1748             res=matroska_parse_block(matroska,
1749                                      blocks[i].bin.data, blocks[i].bin.size,
1750                                      blocks[i].bin.pos,  cluster.timecode,
1751                                      blocks[i].duration, !blocks[i].reference,
1752                                      pos);
1753     ebml_free(matroska_cluster, &cluster);
1754     if (res < 0)  matroska->done = 1;
1755     return res;
1756 }
1757
1758 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1759 {
1760     MatroskaDemuxContext *matroska = s->priv_data;
1761
1762     while (matroska_deliver_packet(matroska, pkt)) {
1763         if (matroska->done)
1764             return AVERROR_EOF;
1765         matroska_parse_cluster(matroska);
1766     }
1767
1768     return 0;
1769 }
1770
1771 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1772                               int64_t timestamp, int flags)
1773 {
1774     MatroskaDemuxContext *matroska = s->priv_data;
1775     MatroskaTrack *tracks = matroska->tracks.elem;
1776     AVStream *st = s->streams[stream_index];
1777     int i, index, index_sub, index_min;
1778
1779     if (!st->nb_index_entries)
1780         return 0;
1781     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
1782
1783     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1784         url_fseek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
1785         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1786             matroska_clear_queue(matroska);
1787             if (matroska_parse_cluster(matroska) < 0)
1788                 break;
1789         }
1790     }
1791
1792     matroska_clear_queue(matroska);
1793     if (index < 0)
1794         return 0;
1795
1796     index_min = index;
1797     for (i=0; i < matroska->tracks.nb_elem; i++) {
1798         tracks[i].end_timecode = 0;
1799         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
1800             && !tracks[i].stream->discard != AVDISCARD_ALL) {
1801             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
1802             if (index_sub >= 0
1803                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
1804                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
1805                 index_min = index_sub;
1806         }
1807     }
1808
1809     url_fseek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
1810     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
1811     matroska->skip_to_timecode = st->index_entries[index].timestamp;
1812     matroska->done = 0;
1813     av_update_cur_dts(s, st, st->index_entries[index].timestamp);
1814     return 0;
1815 }
1816
1817 static int matroska_read_close(AVFormatContext *s)
1818 {
1819     MatroskaDemuxContext *matroska = s->priv_data;
1820     MatroskaTrack *tracks = matroska->tracks.elem;
1821     int n;
1822
1823     matroska_clear_queue(matroska);
1824
1825     for (n=0; n < matroska->tracks.nb_elem; n++)
1826         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
1827             av_free(tracks[n].audio.buf);
1828     ebml_free(matroska_segment, matroska);
1829
1830     return 0;
1831 }
1832
1833 AVInputFormat matroska_demuxer = {
1834     "matroska",
1835     NULL_IF_CONFIG_SMALL("Matroska file format"),
1836     sizeof(MatroskaDemuxContext),
1837     matroska_probe,
1838     matroska_read_header,
1839     matroska_read_packet,
1840     matroska_read_close,
1841     matroska_read_seek,
1842     .metadata_conv = ff_mkv_metadata_conv,
1843 };