Fix the MATROSKA_ID_TAGDEFAULT to match the specification (instead of
[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  * 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 #include "internal.h"
34 /* For ff_codec_get_id(). */
35 #include "riff.h"
36 #include "isom.h"
37 #include "rm.h"
38 #include "matroska.h"
39 #include "libavcodec/mpeg4audio.h"
40 #include "libavutil/intfloat_readwrite.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavutil/avstring.h"
43 #include "libavutil/lzo.h"
44 #if CONFIG_ZLIB
45 #include <zlib.h>
46 #endif
47 #if CONFIG_BZLIB
48 #include <bzlib.h>
49 #endif
50
51 typedef enum {
52     EBML_NONE,
53     EBML_UINT,
54     EBML_FLOAT,
55     EBML_STR,
56     EBML_UTF8,
57     EBML_BIN,
58     EBML_NEST,
59     EBML_PASS,
60     EBML_STOP,
61 } EbmlType;
62
63 typedef const struct EbmlSyntax {
64     uint32_t id;
65     EbmlType type;
66     int list_elem_size;
67     int data_offset;
68     union {
69         uint64_t    u;
70         double      f;
71         const char *s;
72         const struct EbmlSyntax *n;
73     } def;
74 } EbmlSyntax;
75
76 typedef struct {
77     int nb_elem;
78     void *elem;
79 } EbmlList;
80
81 typedef struct {
82     int      size;
83     uint8_t *data;
84     int64_t  pos;
85 } EbmlBin;
86
87 typedef struct {
88     uint64_t version;
89     uint64_t max_size;
90     uint64_t id_length;
91     char    *doctype;
92     uint64_t doctype_version;
93 } Ebml;
94
95 typedef struct {
96     uint64_t algo;
97     EbmlBin  settings;
98 } MatroskaTrackCompression;
99
100 typedef struct {
101     uint64_t scope;
102     uint64_t type;
103     MatroskaTrackCompression compression;
104 } MatroskaTrackEncoding;
105
106 typedef struct {
107     double   frame_rate;
108     uint64_t display_width;
109     uint64_t display_height;
110     uint64_t pixel_width;
111     uint64_t pixel_height;
112     uint64_t fourcc;
113 } MatroskaTrackVideo;
114
115 typedef struct {
116     double   samplerate;
117     double   out_samplerate;
118     uint64_t bitdepth;
119     uint64_t channels;
120
121     /* real audio header (extracted from extradata) */
122     int      coded_framesize;
123     int      sub_packet_h;
124     int      frame_size;
125     int      sub_packet_size;
126     int      sub_packet_cnt;
127     int      pkt_cnt;
128     uint8_t *buf;
129 } MatroskaTrackAudio;
130
131 typedef struct {
132     uint64_t num;
133     uint64_t uid;
134     uint64_t type;
135     char    *name;
136     char    *codec_id;
137     EbmlBin  codec_priv;
138     char    *language;
139     double time_scale;
140     uint64_t default_duration;
141     uint64_t flag_default;
142     MatroskaTrackVideo video;
143     MatroskaTrackAudio audio;
144     EbmlList encodings;
145
146     AVStream *stream;
147     int64_t end_timecode;
148     int ms_compat;
149 } MatroskaTrack;
150
151 typedef struct {
152     uint64_t uid;
153     char *filename;
154     char *mime;
155     EbmlBin bin;
156
157     AVStream *stream;
158 } MatroskaAttachement;
159
160 typedef struct {
161     uint64_t start;
162     uint64_t end;
163     uint64_t uid;
164     char    *title;
165
166     AVChapter *chapter;
167 } MatroskaChapter;
168
169 typedef struct {
170     uint64_t track;
171     uint64_t pos;
172 } MatroskaIndexPos;
173
174 typedef struct {
175     uint64_t time;
176     EbmlList pos;
177 } MatroskaIndex;
178
179 typedef struct {
180     char *name;
181     char *string;
182     char *lang;
183     uint64_t def;
184     EbmlList sub;
185 } MatroskaTag;
186
187 typedef struct {
188     char    *type;
189     uint64_t typevalue;
190     uint64_t trackuid;
191     uint64_t chapteruid;
192     uint64_t attachuid;
193 } MatroskaTagTarget;
194
195 typedef struct {
196     MatroskaTagTarget target;
197     EbmlList tag;
198 } MatroskaTags;
199
200 typedef struct {
201     uint64_t id;
202     uint64_t pos;
203 } MatroskaSeekhead;
204
205 typedef struct {
206     uint64_t start;
207     uint64_t length;
208 } MatroskaLevel;
209
210 typedef struct {
211     AVFormatContext *ctx;
212
213     /* EBML stuff */
214     int num_levels;
215     MatroskaLevel levels[EBML_MAX_DEPTH];
216     int level_up;
217
218     uint64_t time_scale;
219     double   duration;
220     char    *title;
221     EbmlList tracks;
222     EbmlList attachments;
223     EbmlList chapters;
224     EbmlList index;
225     EbmlList tags;
226     EbmlList seekhead;
227
228     /* byte position of the segment inside the stream */
229     int64_t segment_start;
230
231     /* the packet queue */
232     AVPacket **packets;
233     int num_packets;
234     AVPacket *prev_pkt;
235
236     int done;
237     int has_cluster_id;
238
239     /* What to skip before effectively reading a packet. */
240     int skip_to_keyframe;
241     uint64_t skip_to_timecode;
242 } MatroskaDemuxContext;
243
244 typedef struct {
245     uint64_t duration;
246     int64_t  reference;
247     uint64_t non_simple;
248     EbmlBin  bin;
249 } MatroskaBlock;
250
251 typedef struct {
252     uint64_t timecode;
253     EbmlList blocks;
254 } MatroskaCluster;
255
256 static EbmlSyntax ebml_header[] = {
257     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
258     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
259     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
260     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
261     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
262     { EBML_ID_EBMLVERSION,            EBML_NONE },
263     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
264     { 0 }
265 };
266
267 static EbmlSyntax ebml_syntax[] = {
268     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
269     { 0 }
270 };
271
272 static EbmlSyntax matroska_info[] = {
273     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
274     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
275     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
276     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
277     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
278     { MATROSKA_ID_DATEUTC,            EBML_NONE },
279     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
280     { 0 }
281 };
282
283 static EbmlSyntax matroska_track_video[] = {
284     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
285     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
286     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
287     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
288     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
289     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,fourcc) },
290     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
291     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
292     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
293     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
294     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
295     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
296     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_NONE },
297     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
298     { 0 }
299 };
300
301 static EbmlSyntax matroska_track_audio[] = {
302     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
303     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
304     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
305     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
306     { 0 }
307 };
308
309 static EbmlSyntax matroska_track_encoding_compression[] = {
310     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
311     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
312     { 0 }
313 };
314
315 static EbmlSyntax matroska_track_encoding[] = {
316     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
317     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
318     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
319     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
320     { 0 }
321 };
322
323 static EbmlSyntax matroska_track_encodings[] = {
324     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
325     { 0 }
326 };
327
328 static EbmlSyntax matroska_track[] = {
329     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
330     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
331     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
332     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
333     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
334     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
335     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
336     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
337     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
338     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
339     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
340     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
341     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
342     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
343     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_NONE },
344     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
345     { MATROSKA_ID_CODECNAME,            EBML_NONE },
346     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
347     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
348     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
349     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
350     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
351     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
352     { 0 }
353 };
354
355 static EbmlSyntax matroska_tracks[] = {
356     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
357     { 0 }
358 };
359
360 static EbmlSyntax matroska_attachment[] = {
361     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
362     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
363     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
364     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
365     { MATROSKA_ID_FILEDESC,           EBML_NONE },
366     { 0 }
367 };
368
369 static EbmlSyntax matroska_attachments[] = {
370     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
371     { 0 }
372 };
373
374 static EbmlSyntax matroska_chapter_display[] = {
375     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
376     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
377     { 0 }
378 };
379
380 static EbmlSyntax matroska_chapter_entry[] = {
381     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
382     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
383     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
384     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
385     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
386     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
387     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
388     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
389     { 0 }
390 };
391
392 static EbmlSyntax matroska_chapter[] = {
393     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
394     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
395     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
396     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
397     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
398     { 0 }
399 };
400
401 static EbmlSyntax matroska_chapters[] = {
402     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
403     { 0 }
404 };
405
406 static EbmlSyntax matroska_index_pos[] = {
407     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
408     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
409     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
410     { 0 }
411 };
412
413 static EbmlSyntax matroska_index_entry[] = {
414     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
415     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
416     { 0 }
417 };
418
419 static EbmlSyntax matroska_index[] = {
420     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
421     { 0 }
422 };
423
424 static EbmlSyntax matroska_simpletag[] = {
425     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
426     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
427     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
428     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
429     { MATROSKA_ID_TAGDEFAULT_BUG,     EBML_UINT, 0, offsetof(MatroskaTag,def) },
430     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
431     { 0 }
432 };
433
434 static EbmlSyntax matroska_tagtargets[] = {
435     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
436     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
437     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
438     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
439     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
440     { 0 }
441 };
442
443 static EbmlSyntax matroska_tag[] = {
444     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
445     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
446     { 0 }
447 };
448
449 static EbmlSyntax matroska_tags[] = {
450     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
451     { 0 }
452 };
453
454 static EbmlSyntax matroska_seekhead_entry[] = {
455     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
456     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
457     { 0 }
458 };
459
460 static EbmlSyntax matroska_seekhead[] = {
461     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
462     { 0 }
463 };
464
465 static EbmlSyntax matroska_segment[] = {
466     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
467     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
468     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
469     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
470     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
471     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
472     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
473     { MATROSKA_ID_CLUSTER,        EBML_STOP, 0, offsetof(MatroskaDemuxContext,has_cluster_id) },
474     { 0 }
475 };
476
477 static EbmlSyntax matroska_segments[] = {
478     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
479     { 0 }
480 };
481
482 static EbmlSyntax matroska_blockgroup[] = {
483     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
484     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
485     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration), {.u=AV_NOPTS_VALUE} },
486     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
487     { 1,                          EBML_UINT, 0, offsetof(MatroskaBlock,non_simple), {.u=1} },
488     { 0 }
489 };
490
491 static EbmlSyntax matroska_cluster[] = {
492     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
493     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
494     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
495     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
496     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
497     { 0 }
498 };
499
500 static EbmlSyntax matroska_clusters[] = {
501     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
502     { MATROSKA_ID_INFO,           EBML_NONE },
503     { MATROSKA_ID_CUES,           EBML_NONE },
504     { MATROSKA_ID_TAGS,           EBML_NONE },
505     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
506     { 0 }
507 };
508
509 static const char *matroska_doctypes[] = { "matroska", "webm" };
510
511 /*
512  * Return: Whether we reached the end of a level in the hierarchy or not.
513  */
514 static int ebml_level_end(MatroskaDemuxContext *matroska)
515 {
516     ByteIOContext *pb = matroska->ctx->pb;
517     int64_t pos = url_ftell(pb);
518
519     if (matroska->num_levels > 0) {
520         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
521         if (pos - level->start >= level->length) {
522             matroska->num_levels--;
523             return 1;
524         }
525     }
526     return 0;
527 }
528
529 /*
530  * Read: an "EBML number", which is defined as a variable-length
531  * array of bytes. The first byte indicates the length by giving a
532  * number of 0-bits followed by a one. The position of the first
533  * "one" bit inside the first byte indicates the length of this
534  * number.
535  * Returns: number of bytes read, < 0 on error
536  */
537 static int ebml_read_num(MatroskaDemuxContext *matroska, ByteIOContext *pb,
538                          int max_size, uint64_t *number)
539 {
540     int len_mask = 0x80, read = 1, n = 1;
541     int64_t total = 0;
542
543     /* The first byte tells us the length in bytes - get_byte() can normally
544      * return 0, but since that's not a valid first ebmlID byte, we can
545      * use it safely here to catch EOS. */
546     if (!(total = get_byte(pb))) {
547         /* we might encounter EOS here */
548         if (!url_feof(pb)) {
549             int64_t pos = url_ftell(pb);
550             av_log(matroska->ctx, AV_LOG_ERROR,
551                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
552                    pos, pos);
553         }
554         return AVERROR(EIO); /* EOS or actual I/O error */
555     }
556
557     /* get the length of the EBML number */
558     while (read <= max_size && !(total & len_mask)) {
559         read++;
560         len_mask >>= 1;
561     }
562     if (read > max_size) {
563         int64_t pos = url_ftell(pb) - 1;
564         av_log(matroska->ctx, AV_LOG_ERROR,
565                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
566                (uint8_t) total, pos, pos);
567         return AVERROR_INVALIDDATA;
568     }
569
570     /* read out length */
571     total &= ~len_mask;
572     while (n++ < read)
573         total = (total << 8) | get_byte(pb);
574
575     *number = total;
576
577     return read;
578 }
579
580 /*
581  * Read the next element as an unsigned int.
582  * 0 is success, < 0 is failure.
583  */
584 static int ebml_read_uint(ByteIOContext *pb, int size, uint64_t *num)
585 {
586     int n = 0;
587
588     if (size < 1 || size > 8)
589         return AVERROR_INVALIDDATA;
590
591     /* big-endian ordering; build up number */
592     *num = 0;
593     while (n++ < size)
594         *num = (*num << 8) | get_byte(pb);
595
596     return 0;
597 }
598
599 /*
600  * Read the next element as a float.
601  * 0 is success, < 0 is failure.
602  */
603 static int ebml_read_float(ByteIOContext *pb, int size, double *num)
604 {
605     if (size == 4) {
606         *num= av_int2flt(get_be32(pb));
607     } else if(size==8){
608         *num= av_int2dbl(get_be64(pb));
609     } else
610         return AVERROR_INVALIDDATA;
611
612     return 0;
613 }
614
615 /*
616  * Read the next element as an ASCII string.
617  * 0 is success, < 0 is failure.
618  */
619 static int ebml_read_ascii(ByteIOContext *pb, int size, char **str)
620 {
621     av_free(*str);
622     /* EBML strings are usually not 0-terminated, so we allocate one
623      * byte more, read the string and NULL-terminate it ourselves. */
624     if (!(*str = av_malloc(size + 1)))
625         return AVERROR(ENOMEM);
626     if (get_buffer(pb, (uint8_t *) *str, size) != size) {
627         av_freep(str);
628         return AVERROR(EIO);
629     }
630     (*str)[size] = '\0';
631
632     return 0;
633 }
634
635 /*
636  * Read the next element as binary data.
637  * 0 is success, < 0 is failure.
638  */
639 static int ebml_read_binary(ByteIOContext *pb, int length, EbmlBin *bin)
640 {
641     av_free(bin->data);
642     if (!(bin->data = av_malloc(length)))
643         return AVERROR(ENOMEM);
644
645     bin->size = length;
646     bin->pos  = url_ftell(pb);
647     if (get_buffer(pb, bin->data, length) != length) {
648         av_freep(&bin->data);
649         return AVERROR(EIO);
650     }
651
652     return 0;
653 }
654
655 /*
656  * Read the next element, but only the header. The contents
657  * are supposed to be sub-elements which can be read separately.
658  * 0 is success, < 0 is failure.
659  */
660 static int ebml_read_master(MatroskaDemuxContext *matroska, int length)
661 {
662     ByteIOContext *pb = matroska->ctx->pb;
663     MatroskaLevel *level;
664
665     if (matroska->num_levels >= EBML_MAX_DEPTH) {
666         av_log(matroska->ctx, AV_LOG_ERROR,
667                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
668         return AVERROR(ENOSYS);
669     }
670
671     level = &matroska->levels[matroska->num_levels++];
672     level->start = url_ftell(pb);
673     level->length = length;
674
675     return 0;
676 }
677
678 /*
679  * Read signed/unsigned "EBML" numbers.
680  * Return: number of bytes processed, < 0 on error
681  */
682 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
683                                  uint8_t *data, uint32_t size, uint64_t *num)
684 {
685     ByteIOContext pb;
686     init_put_byte(&pb, data, size, 0, NULL, NULL, NULL, NULL);
687     return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
688 }
689
690 /*
691  * Same as above, but signed.
692  */
693 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
694                                  uint8_t *data, uint32_t size, int64_t *num)
695 {
696     uint64_t unum;
697     int res;
698
699     /* read as unsigned number first */
700     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
701         return res;
702
703     /* make signed (weird way) */
704     *num = unum - ((1LL << (7*res - 1)) - 1);
705
706     return res;
707 }
708
709 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
710                            EbmlSyntax *syntax, void *data);
711
712 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
713                          uint32_t id, void *data)
714 {
715     int i;
716     for (i=0; syntax[i].id; i++)
717         if (id == syntax[i].id)
718             break;
719     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32)
720         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
721     return ebml_parse_elem(matroska, &syntax[i], data);
722 }
723
724 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
725                       void *data)
726 {
727     uint64_t id;
728     int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
729     id |= 1 << 7*res;
730     return res < 0 ? res : ebml_parse_id(matroska, syntax, id, data);
731 }
732
733 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
734                            void *data)
735 {
736     int i, res = 0;
737
738     for (i=0; syntax[i].id; i++)
739         switch (syntax[i].type) {
740         case EBML_UINT:
741             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
742             break;
743         case EBML_FLOAT:
744             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
745             break;
746         case EBML_STR:
747         case EBML_UTF8:
748             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
749             break;
750         }
751
752     while (!res && !ebml_level_end(matroska))
753         res = ebml_parse(matroska, syntax, data);
754
755     return res;
756 }
757
758 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
759                            EbmlSyntax *syntax, void *data)
760 {
761     ByteIOContext *pb = matroska->ctx->pb;
762     uint32_t id = syntax->id;
763     uint64_t length;
764     int res;
765
766     data = (char *)data + syntax->data_offset;
767     if (syntax->list_elem_size) {
768         EbmlList *list = data;
769         list->elem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
770         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
771         memset(data, 0, syntax->list_elem_size);
772         list->nb_elem++;
773     }
774
775     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP)
776         if ((res = ebml_read_num(matroska, pb, 8, &length)) < 0)
777             return res;
778
779     switch (syntax->type) {
780     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
781     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
782     case EBML_STR:
783     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
784     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
785     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
786                          return res;
787                      if (id == MATROSKA_ID_SEGMENT)
788                          matroska->segment_start = url_ftell(matroska->ctx->pb);
789                      return ebml_parse_nest(matroska, syntax->def.n, data);
790     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
791     case EBML_STOP:  *(int *)data = 1;      return 1;
792     default:         return url_fseek(pb,length,SEEK_CUR)<0 ? AVERROR(EIO) : 0;
793     }
794     if (res == AVERROR_INVALIDDATA)
795         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
796     else if (res == AVERROR(EIO))
797         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
798     return res;
799 }
800
801 static void ebml_free(EbmlSyntax *syntax, void *data)
802 {
803     int i, j;
804     for (i=0; syntax[i].id; i++) {
805         void *data_off = (char *)data + syntax[i].data_offset;
806         switch (syntax[i].type) {
807         case EBML_STR:
808         case EBML_UTF8:  av_freep(data_off);                      break;
809         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
810         case EBML_NEST:
811             if (syntax[i].list_elem_size) {
812                 EbmlList *list = data_off;
813                 char *ptr = list->elem;
814                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
815                     ebml_free(syntax[i].def.n, ptr);
816                 av_free(list->elem);
817             } else
818                 ebml_free(syntax[i].def.n, data_off);
819         default:  break;
820         }
821     }
822 }
823
824
825 /*
826  * Autodetecting...
827  */
828 static int matroska_probe(AVProbeData *p)
829 {
830     uint64_t total = 0;
831     int len_mask = 0x80, size = 1, n = 1, i;
832
833     /* EBML header? */
834     if (AV_RB32(p->buf) != EBML_ID_HEADER)
835         return 0;
836
837     /* length of header */
838     total = p->buf[4];
839     while (size <= 8 && !(total & len_mask)) {
840         size++;
841         len_mask >>= 1;
842     }
843     if (size > 8)
844       return 0;
845     total &= (len_mask - 1);
846     while (n < size)
847         total = (total << 8) | p->buf[4 + n++];
848
849     /* Does the probe data contain the whole header? */
850     if (p->buf_size < 4 + size + total)
851       return 0;
852
853     /* The header should contain a known document type. For now,
854      * we don't parse the whole header but simply check for the
855      * availability of that array of characters inside the header.
856      * Not fully fool-proof, but good enough. */
857     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
858         int probelen = strlen(matroska_doctypes[i]);
859         for (n = 4+size; n <= 4+size+total-probelen; n++)
860             if (!memcmp(p->buf+n, matroska_doctypes[i], probelen))
861                 return AVPROBE_SCORE_MAX;
862     }
863
864     // probably valid EBML header but no recognized doctype
865     return AVPROBE_SCORE_MAX/2;
866 }
867
868 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
869                                                  int num)
870 {
871     MatroskaTrack *tracks = matroska->tracks.elem;
872     int i;
873
874     for (i=0; i < matroska->tracks.nb_elem; i++)
875         if (tracks[i].num == num)
876             return &tracks[i];
877
878     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
879     return NULL;
880 }
881
882 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
883                                   MatroskaTrack *track)
884 {
885     MatroskaTrackEncoding *encodings = track->encodings.elem;
886     uint8_t* data = *buf;
887     int isize = *buf_size;
888     uint8_t* pkt_data = NULL;
889     int pkt_size = isize;
890     int result = 0;
891     int olen;
892
893     switch (encodings[0].compression.algo) {
894     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
895         return encodings[0].compression.settings.size;
896     case MATROSKA_TRACK_ENCODING_COMP_LZO:
897         do {
898             olen = pkt_size *= 3;
899             pkt_data = av_realloc(pkt_data, pkt_size+AV_LZO_OUTPUT_PADDING);
900             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
901         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
902         if (result)
903             goto failed;
904         pkt_size -= olen;
905         break;
906 #if CONFIG_ZLIB
907     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
908         z_stream zstream = {0};
909         if (inflateInit(&zstream) != Z_OK)
910             return -1;
911         zstream.next_in = data;
912         zstream.avail_in = isize;
913         do {
914             pkt_size *= 3;
915             pkt_data = av_realloc(pkt_data, pkt_size);
916             zstream.avail_out = pkt_size - zstream.total_out;
917             zstream.next_out = pkt_data + zstream.total_out;
918             result = inflate(&zstream, Z_NO_FLUSH);
919         } while (result==Z_OK && pkt_size<10000000);
920         pkt_size = zstream.total_out;
921         inflateEnd(&zstream);
922         if (result != Z_STREAM_END)
923             goto failed;
924         break;
925     }
926 #endif
927 #if CONFIG_BZLIB
928     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
929         bz_stream bzstream = {0};
930         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
931             return -1;
932         bzstream.next_in = data;
933         bzstream.avail_in = isize;
934         do {
935             pkt_size *= 3;
936             pkt_data = av_realloc(pkt_data, pkt_size);
937             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
938             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
939             result = BZ2_bzDecompress(&bzstream);
940         } while (result==BZ_OK && pkt_size<10000000);
941         pkt_size = bzstream.total_out_lo32;
942         BZ2_bzDecompressEnd(&bzstream);
943         if (result != BZ_STREAM_END)
944             goto failed;
945         break;
946     }
947 #endif
948     default:
949         return -1;
950     }
951
952     *buf = pkt_data;
953     *buf_size = pkt_size;
954     return 0;
955  failed:
956     av_free(pkt_data);
957     return -1;
958 }
959
960 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
961                                     AVPacket *pkt, uint64_t display_duration)
962 {
963     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
964     for (; *ptr!=',' && ptr<end-1; ptr++);
965     if (*ptr == ',')
966         layer = ++ptr;
967     for (; *ptr!=',' && ptr<end-1; ptr++);
968     if (*ptr == ',') {
969         int64_t end_pts = pkt->pts + display_duration;
970         int sc = matroska->time_scale * pkt->pts / 10000000;
971         int ec = matroska->time_scale * end_pts  / 10000000;
972         int sh, sm, ss, eh, em, es, len;
973         sh = sc/360000;  sc -= 360000*sh;
974         sm = sc/  6000;  sc -=   6000*sm;
975         ss = sc/   100;  sc -=    100*ss;
976         eh = ec/360000;  ec -= 360000*eh;
977         em = ec/  6000;  ec -=   6000*em;
978         es = ec/   100;  ec -=    100*es;
979         *ptr++ = '\0';
980         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
981         if (!(line = av_malloc(len)))
982             return;
983         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
984                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
985         av_free(pkt->data);
986         pkt->data = line;
987         pkt->size = strlen(line);
988     }
989 }
990
991 static void matroska_merge_packets(AVPacket *out, AVPacket *in)
992 {
993     out->data = av_realloc(out->data, out->size+in->size);
994     memcpy(out->data+out->size, in->data, in->size);
995     out->size += in->size;
996     av_destruct_packet(in);
997     av_free(in);
998 }
999
1000 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1001                                  AVMetadata **metadata, char *prefix)
1002 {
1003     MatroskaTag *tags = list->elem;
1004     char key[1024];
1005     int i;
1006
1007     for (i=0; i < list->nb_elem; i++) {
1008         const char *lang = strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1009         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1010         else         av_strlcpy(key, tags[i].name, sizeof(key));
1011         if (tags[i].def || !lang) {
1012         av_metadata_set2(metadata, key, tags[i].string, 0);
1013         if (tags[i].sub.nb_elem)
1014             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1015         }
1016         if (lang) {
1017             av_strlcat(key, "-", sizeof(key));
1018             av_strlcat(key, lang, sizeof(key));
1019             av_metadata_set2(metadata, key, tags[i].string, 0);
1020             if (tags[i].sub.nb_elem)
1021                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1022         }
1023     }
1024 }
1025
1026 static void matroska_convert_tags(AVFormatContext *s)
1027 {
1028     MatroskaDemuxContext *matroska = s->priv_data;
1029     MatroskaTags *tags = matroska->tags.elem;
1030     int i, j;
1031
1032     for (i=0; i < matroska->tags.nb_elem; i++) {
1033         if (tags[i].target.attachuid) {
1034             MatroskaAttachement *attachment = matroska->attachments.elem;
1035             for (j=0; j<matroska->attachments.nb_elem; j++)
1036                 if (attachment[j].uid == tags[i].target.attachuid)
1037                     matroska_convert_tag(s, &tags[i].tag,
1038                                          &attachment[j].stream->metadata, NULL);
1039         } else if (tags[i].target.chapteruid) {
1040             MatroskaChapter *chapter = matroska->chapters.elem;
1041             for (j=0; j<matroska->chapters.nb_elem; j++)
1042                 if (chapter[j].uid == tags[i].target.chapteruid)
1043                     matroska_convert_tag(s, &tags[i].tag,
1044                                          &chapter[j].chapter->metadata, NULL);
1045         } else if (tags[i].target.trackuid) {
1046             MatroskaTrack *track = matroska->tracks.elem;
1047             for (j=0; j<matroska->tracks.nb_elem; j++)
1048                 if (track[j].uid == tags[i].target.trackuid)
1049                     matroska_convert_tag(s, &tags[i].tag,
1050                                          &track[j].stream->metadata, NULL);
1051         } else {
1052             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1053                                  tags[i].target.type);
1054         }
1055     }
1056 }
1057
1058 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1059 {
1060     EbmlList *seekhead_list = &matroska->seekhead;
1061     MatroskaSeekhead *seekhead = seekhead_list->elem;
1062     uint32_t level_up = matroska->level_up;
1063     int64_t before_pos = url_ftell(matroska->ctx->pb);
1064     MatroskaLevel level;
1065     int i;
1066
1067     for (i=0; i<seekhead_list->nb_elem; i++) {
1068         int64_t offset = seekhead[i].pos + matroska->segment_start;
1069
1070         if (seekhead[i].pos <= before_pos
1071             || seekhead[i].id == MATROSKA_ID_SEEKHEAD
1072             || seekhead[i].id == MATROSKA_ID_CLUSTER)
1073             continue;
1074
1075         /* seek */
1076         if (url_fseek(matroska->ctx->pb, offset, SEEK_SET) != offset)
1077             continue;
1078
1079         /* We don't want to lose our seekhead level, so we add
1080          * a dummy. This is a crude hack. */
1081         if (matroska->num_levels == EBML_MAX_DEPTH) {
1082             av_log(matroska->ctx, AV_LOG_INFO,
1083                    "Max EBML element depth (%d) reached, "
1084                    "cannot parse further.\n", EBML_MAX_DEPTH);
1085             break;
1086         }
1087
1088         level.start = 0;
1089         level.length = (uint64_t)-1;
1090         matroska->levels[matroska->num_levels] = level;
1091         matroska->num_levels++;
1092
1093         ebml_parse(matroska, matroska_segment, matroska);
1094
1095         /* remove dummy level */
1096         while (matroska->num_levels) {
1097             uint64_t length = matroska->levels[--matroska->num_levels].length;
1098             if (length == (uint64_t)-1)
1099                 break;
1100         }
1101     }
1102
1103     /* seek back */
1104     url_fseek(matroska->ctx->pb, before_pos, SEEK_SET);
1105     matroska->level_up = level_up;
1106 }
1107
1108 static int matroska_aac_profile(char *codec_id)
1109 {
1110     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1111     int profile;
1112
1113     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1114         if (strstr(codec_id, aac_profiles[profile]))
1115             break;
1116     return profile + 1;
1117 }
1118
1119 static int matroska_aac_sri(int samplerate)
1120 {
1121     int sri;
1122
1123     for (sri=0; sri<FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
1124         if (ff_mpeg4audio_sample_rates[sri] == samplerate)
1125             break;
1126     return sri;
1127 }
1128
1129 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
1130 {
1131     MatroskaDemuxContext *matroska = s->priv_data;
1132     EbmlList *attachements_list = &matroska->attachments;
1133     MatroskaAttachement *attachements;
1134     EbmlList *chapters_list = &matroska->chapters;
1135     MatroskaChapter *chapters;
1136     MatroskaTrack *tracks;
1137     EbmlList *index_list;
1138     MatroskaIndex *index;
1139     int index_scale = 1;
1140     uint64_t max_start = 0;
1141     Ebml ebml = { 0 };
1142     AVStream *st;
1143     int i, j;
1144
1145     matroska->ctx = s;
1146
1147     /* First read the EBML header. */
1148     if (ebml_parse(matroska, ebml_syntax, &ebml)
1149         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1150         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 2) {
1151         av_log(matroska->ctx, AV_LOG_ERROR,
1152                "EBML header using unsupported features\n"
1153                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1154                ebml.version, ebml.doctype, ebml.doctype_version);
1155         ebml_free(ebml_syntax, &ebml);
1156         return AVERROR_PATCHWELCOME;
1157     }
1158     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1159         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1160             break;
1161     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1162         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1163     }
1164     ebml_free(ebml_syntax, &ebml);
1165
1166     /* The next thing is a segment. */
1167     if (ebml_parse(matroska, matroska_segments, matroska) < 0)
1168         return -1;
1169     matroska_execute_seekhead(matroska);
1170
1171     if (!matroska->time_scale)
1172         matroska->time_scale = 1000000;
1173     if (matroska->duration)
1174         matroska->ctx->duration = matroska->duration * matroska->time_scale
1175                                   * 1000 / AV_TIME_BASE;
1176     av_metadata_set2(&s->metadata, "title", matroska->title, 0);
1177
1178     tracks = matroska->tracks.elem;
1179     for (i=0; i < matroska->tracks.nb_elem; i++) {
1180         MatroskaTrack *track = &tracks[i];
1181         enum CodecID codec_id = CODEC_ID_NONE;
1182         EbmlList *encodings_list = &tracks->encodings;
1183         MatroskaTrackEncoding *encodings = encodings_list->elem;
1184         uint8_t *extradata = NULL;
1185         int extradata_size = 0;
1186         int extradata_offset = 0;
1187         ByteIOContext b;
1188
1189         /* Apply some sanity checks. */
1190         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1191             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1192             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1193             av_log(matroska->ctx, AV_LOG_INFO,
1194                    "Unknown or unsupported track type %"PRIu64"\n",
1195                    track->type);
1196             continue;
1197         }
1198         if (track->codec_id == NULL)
1199             continue;
1200
1201         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1202             if (!track->default_duration)
1203                 track->default_duration = 1000000000/track->video.frame_rate;
1204             if (!track->video.display_width)
1205                 track->video.display_width = track->video.pixel_width;
1206             if (!track->video.display_height)
1207                 track->video.display_height = track->video.pixel_height;
1208         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1209             if (!track->audio.out_samplerate)
1210                 track->audio.out_samplerate = track->audio.samplerate;
1211         }
1212         if (encodings_list->nb_elem > 1) {
1213             av_log(matroska->ctx, AV_LOG_ERROR,
1214                    "Multiple combined encodings no supported");
1215         } else if (encodings_list->nb_elem == 1) {
1216             if (encodings[0].type ||
1217                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1218 #if CONFIG_ZLIB
1219                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1220 #endif
1221 #if CONFIG_BZLIB
1222                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1223 #endif
1224                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1225                 encodings[0].scope = 0;
1226                 av_log(matroska->ctx, AV_LOG_ERROR,
1227                        "Unsupported encoding type");
1228             } else if (track->codec_priv.size && encodings[0].scope&2) {
1229                 uint8_t *codec_priv = track->codec_priv.data;
1230                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1231                                                     &track->codec_priv.size,
1232                                                     track);
1233                 if (offset < 0) {
1234                     track->codec_priv.data = NULL;
1235                     track->codec_priv.size = 0;
1236                     av_log(matroska->ctx, AV_LOG_ERROR,
1237                            "Failed to decode codec private data\n");
1238                 } else if (offset > 0) {
1239                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1240                     memcpy(track->codec_priv.data,
1241                            encodings[0].compression.settings.data, offset);
1242                     memcpy(track->codec_priv.data+offset, codec_priv,
1243                            track->codec_priv.size);
1244                     track->codec_priv.size += offset;
1245                 }
1246                 if (codec_priv != track->codec_priv.data)
1247                     av_free(codec_priv);
1248             }
1249         }
1250
1251         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1252             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1253                         strlen(ff_mkv_codec_tags[j].str))){
1254                 codec_id= ff_mkv_codec_tags[j].id;
1255                 break;
1256             }
1257         }
1258
1259         st = track->stream = av_new_stream(s, 0);
1260         if (st == NULL)
1261             return AVERROR(ENOMEM);
1262
1263         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1264             && track->codec_priv.size >= 40
1265             && track->codec_priv.data != NULL) {
1266             track->ms_compat = 1;
1267             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1268             codec_id = ff_codec_get_id(ff_codec_bmp_tags, track->video.fourcc);
1269             extradata_offset = 40;
1270         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1271                    && track->codec_priv.size >= 14
1272                    && track->codec_priv.data != NULL) {
1273             init_put_byte(&b, track->codec_priv.data, track->codec_priv.size,
1274                           URL_RDONLY, NULL, NULL, NULL, NULL);
1275             ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1276             codec_id = st->codec->codec_id;
1277             extradata_offset = FFMIN(track->codec_priv.size, 18);
1278         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1279                    && (track->codec_priv.size >= 86)
1280                    && (track->codec_priv.data != NULL)) {
1281             track->video.fourcc = AV_RL32(track->codec_priv.data);
1282             codec_id=ff_codec_get_id(codec_movvideo_tags, track->video.fourcc);
1283         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1284             switch (track->audio.bitdepth) {
1285             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1286             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1287             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1288             }
1289         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1290             switch (track->audio.bitdepth) {
1291             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1292             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1293             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1294             }
1295         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1296             codec_id = CODEC_ID_PCM_F64LE;
1297         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1298             int profile = matroska_aac_profile(track->codec_id);
1299             int sri = matroska_aac_sri(track->audio.samplerate);
1300             extradata = av_malloc(5);
1301             if (extradata == NULL)
1302                 return AVERROR(ENOMEM);
1303             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1304             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1305             if (strstr(track->codec_id, "SBR")) {
1306                 sri = matroska_aac_sri(track->audio.out_samplerate);
1307                 extradata[2] = 0x56;
1308                 extradata[3] = 0xE5;
1309                 extradata[4] = 0x80 | (sri<<3);
1310                 extradata_size = 5;
1311             } else
1312                 extradata_size = 2;
1313         } else if (codec_id == CODEC_ID_TTA) {
1314             extradata_size = 30;
1315             extradata = av_mallocz(extradata_size);
1316             if (extradata == NULL)
1317                 return AVERROR(ENOMEM);
1318             init_put_byte(&b, extradata, extradata_size, 1,
1319                           NULL, NULL, NULL, NULL);
1320             put_buffer(&b, "TTA1", 4);
1321             put_le16(&b, 1);
1322             put_le16(&b, track->audio.channels);
1323             put_le16(&b, track->audio.bitdepth);
1324             put_le32(&b, track->audio.out_samplerate);
1325             put_le32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1326         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1327                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1328             extradata_offset = 26;
1329         } else if (codec_id == CODEC_ID_RA_144) {
1330             track->audio.out_samplerate = 8000;
1331             track->audio.channels = 1;
1332         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1333                    codec_id == CODEC_ID_ATRAC3 || codec_id == CODEC_ID_SIPR) {
1334             int flavor;
1335             init_put_byte(&b, track->codec_priv.data,track->codec_priv.size,
1336                           0, NULL, NULL, NULL, NULL);
1337             url_fskip(&b, 22);
1338             flavor                       = get_be16(&b);
1339             track->audio.coded_framesize = get_be32(&b);
1340             url_fskip(&b, 12);
1341             track->audio.sub_packet_h    = get_be16(&b);
1342             track->audio.frame_size      = get_be16(&b);
1343             track->audio.sub_packet_size = get_be16(&b);
1344             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1345             if (codec_id == CODEC_ID_RA_288) {
1346                 st->codec->block_align = track->audio.coded_framesize;
1347                 track->codec_priv.size = 0;
1348             } else {
1349                 if (codec_id == CODEC_ID_SIPR && flavor < 4) {
1350                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1351                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1352                     st->codec->bit_rate = sipr_bit_rate[flavor];
1353                 }
1354                 st->codec->block_align = track->audio.sub_packet_size;
1355                 extradata_offset = 78;
1356             }
1357         }
1358         track->codec_priv.size -= extradata_offset;
1359
1360         if (codec_id == CODEC_ID_NONE)
1361             av_log(matroska->ctx, AV_LOG_INFO,
1362                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1363
1364         if (track->time_scale < 0.01)
1365             track->time_scale = 1.0;
1366         av_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1367
1368         st->codec->codec_id = codec_id;
1369         st->start_time = 0;
1370         if (strcmp(track->language, "und"))
1371             av_metadata_set2(&st->metadata, "language", track->language, 0);
1372         av_metadata_set2(&st->metadata, "title", track->name, 0);
1373
1374         if (track->flag_default)
1375             st->disposition |= AV_DISPOSITION_DEFAULT;
1376
1377         if (track->default_duration)
1378             av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
1379                       track->default_duration, 1000000000, 30000);
1380
1381         if (!st->codec->extradata) {
1382             if(extradata){
1383                 st->codec->extradata = extradata;
1384                 st->codec->extradata_size = extradata_size;
1385             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1386                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1387                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1388                 if(st->codec->extradata == NULL)
1389                     return AVERROR(ENOMEM);
1390                 st->codec->extradata_size = track->codec_priv.size;
1391                 memcpy(st->codec->extradata,
1392                        track->codec_priv.data + extradata_offset,
1393                        track->codec_priv.size);
1394             }
1395         }
1396
1397         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1398             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1399             st->codec->codec_tag  = track->video.fourcc;
1400             st->codec->width  = track->video.pixel_width;
1401             st->codec->height = track->video.pixel_height;
1402             av_reduce(&st->sample_aspect_ratio.num,
1403                       &st->sample_aspect_ratio.den,
1404                       st->codec->height * track->video.display_width,
1405                       st->codec-> width * track->video.display_height,
1406                       255);
1407             if (st->codec->codec_id != CODEC_ID_H264)
1408             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1409             if (track->default_duration)
1410                 st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
1411         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1412             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1413             st->codec->sample_rate = track->audio.out_samplerate;
1414             st->codec->channels = track->audio.channels;
1415             if (st->codec->codec_id != CODEC_ID_AAC)
1416             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1417         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1418             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1419         }
1420     }
1421
1422     attachements = attachements_list->elem;
1423     for (j=0; j<attachements_list->nb_elem; j++) {
1424         if (!(attachements[j].filename && attachements[j].mime &&
1425               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1426             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1427         } else {
1428             AVStream *st = av_new_stream(s, 0);
1429             if (st == NULL)
1430                 break;
1431             av_metadata_set2(&st->metadata, "filename",attachements[j].filename, 0);
1432             st->codec->codec_id = CODEC_ID_NONE;
1433             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1434             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1435             if(st->codec->extradata == NULL)
1436                 break;
1437             st->codec->extradata_size = attachements[j].bin.size;
1438             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1439
1440             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1441                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1442                              strlen(ff_mkv_mime_tags[i].str))) {
1443                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1444                     break;
1445                 }
1446             }
1447             attachements[j].stream = st;
1448         }
1449     }
1450
1451     chapters = chapters_list->elem;
1452     for (i=0; i<chapters_list->nb_elem; i++)
1453         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1454             && (max_start==0 || chapters[i].start > max_start)) {
1455             chapters[i].chapter =
1456             ff_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1457                            chapters[i].start, chapters[i].end,
1458                            chapters[i].title);
1459             av_metadata_set2(&chapters[i].chapter->metadata,
1460                              "title", chapters[i].title, 0);
1461             max_start = chapters[i].start;
1462         }
1463
1464     index_list = &matroska->index;
1465     index = index_list->elem;
1466     if (index_list->nb_elem
1467         && index[0].time > 100000000000000/matroska->time_scale) {
1468         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1469         index_scale = matroska->time_scale;
1470     }
1471     for (i=0; i<index_list->nb_elem; i++) {
1472         EbmlList *pos_list = &index[i].pos;
1473         MatroskaIndexPos *pos = pos_list->elem;
1474         for (j=0; j<pos_list->nb_elem; j++) {
1475             MatroskaTrack *track = matroska_find_track_by_num(matroska,
1476                                                               pos[j].track);
1477             if (track && track->stream)
1478                 av_add_index_entry(track->stream,
1479                                    pos[j].pos + matroska->segment_start,
1480                                    index[i].time/index_scale, 0, 0,
1481                                    AVINDEX_KEYFRAME);
1482         }
1483     }
1484
1485     matroska_convert_tags(s);
1486
1487     return 0;
1488 }
1489
1490 /*
1491  * Put one packet in an application-supplied AVPacket struct.
1492  * Returns 0 on success or -1 on failure.
1493  */
1494 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1495                                    AVPacket *pkt)
1496 {
1497     if (matroska->num_packets > 0) {
1498         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1499         av_free(matroska->packets[0]);
1500         if (matroska->num_packets > 1) {
1501             memmove(&matroska->packets[0], &matroska->packets[1],
1502                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1503             matroska->packets =
1504                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
1505                            sizeof(AVPacket *));
1506         } else {
1507             av_freep(&matroska->packets);
1508         }
1509         matroska->num_packets--;
1510         return 0;
1511     }
1512
1513     return -1;
1514 }
1515
1516 /*
1517  * Free all packets in our internal queue.
1518  */
1519 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1520 {
1521     if (matroska->packets) {
1522         int n;
1523         for (n = 0; n < matroska->num_packets; n++) {
1524             av_free_packet(matroska->packets[n]);
1525             av_free(matroska->packets[n]);
1526         }
1527         av_freep(&matroska->packets);
1528         matroska->num_packets = 0;
1529     }
1530 }
1531
1532 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1533                                 int size, int64_t pos, uint64_t cluster_time,
1534                                 uint64_t duration, int is_keyframe,
1535                                 int64_t cluster_pos)
1536 {
1537     uint64_t timecode = AV_NOPTS_VALUE;
1538     MatroskaTrack *track;
1539     int res = 0;
1540     AVStream *st;
1541     AVPacket *pkt;
1542     int16_t block_time;
1543     uint32_t *lace_size = NULL;
1544     int n, flags, laces = 0;
1545     uint64_t num;
1546
1547     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1548         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1549         return res;
1550     }
1551     data += n;
1552     size -= n;
1553
1554     track = matroska_find_track_by_num(matroska, num);
1555     if (size <= 3 || !track || !track->stream) {
1556         av_log(matroska->ctx, AV_LOG_INFO,
1557                "Invalid stream %"PRIu64" or size %u\n", num, size);
1558         return res;
1559     }
1560     st = track->stream;
1561     if (st->discard >= AVDISCARD_ALL)
1562         return res;
1563     if (duration == AV_NOPTS_VALUE)
1564         duration = track->default_duration / matroska->time_scale;
1565
1566     block_time = AV_RB16(data);
1567     data += 2;
1568     flags = *data++;
1569     size -= 3;
1570     if (is_keyframe == -1)
1571         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1572
1573     if (cluster_time != (uint64_t)-1
1574         && (block_time >= 0 || cluster_time >= -block_time)) {
1575         timecode = cluster_time + block_time;
1576         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1577             && timecode < track->end_timecode)
1578             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1579         if (is_keyframe)
1580             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1581         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1582     }
1583
1584     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1585         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1586             return res;
1587         matroska->skip_to_keyframe = 0;
1588     }
1589
1590     switch ((flags & 0x06) >> 1) {
1591         case 0x0: /* no lacing */
1592             laces = 1;
1593             lace_size = av_mallocz(sizeof(int));
1594             lace_size[0] = size;
1595             break;
1596
1597         case 0x1: /* Xiph lacing */
1598         case 0x2: /* fixed-size lacing */
1599         case 0x3: /* EBML lacing */
1600             assert(size>0); // size <=3 is checked before size-=3 above
1601             laces = (*data) + 1;
1602             data += 1;
1603             size -= 1;
1604             lace_size = av_mallocz(laces * sizeof(int));
1605
1606             switch ((flags & 0x06) >> 1) {
1607                 case 0x1: /* Xiph lacing */ {
1608                     uint8_t temp;
1609                     uint32_t total = 0;
1610                     for (n = 0; res == 0 && n < laces - 1; n++) {
1611                         while (1) {
1612                             if (size == 0) {
1613                                 res = -1;
1614                                 break;
1615                             }
1616                             temp = *data;
1617                             lace_size[n] += temp;
1618                             data += 1;
1619                             size -= 1;
1620                             if (temp != 0xff)
1621                                 break;
1622                         }
1623                         total += lace_size[n];
1624                     }
1625                     lace_size[n] = size - total;
1626                     break;
1627                 }
1628
1629                 case 0x2: /* fixed-size lacing */
1630                     for (n = 0; n < laces; n++)
1631                         lace_size[n] = size / laces;
1632                     break;
1633
1634                 case 0x3: /* EBML lacing */ {
1635                     uint32_t total;
1636                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1637                     if (n < 0) {
1638                         av_log(matroska->ctx, AV_LOG_INFO,
1639                                "EBML block data error\n");
1640                         break;
1641                     }
1642                     data += n;
1643                     size -= n;
1644                     total = lace_size[0] = num;
1645                     for (n = 1; res == 0 && n < laces - 1; n++) {
1646                         int64_t snum;
1647                         int r;
1648                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1649                         if (r < 0) {
1650                             av_log(matroska->ctx, AV_LOG_INFO,
1651                                    "EBML block data error\n");
1652                             break;
1653                         }
1654                         data += r;
1655                         size -= r;
1656                         lace_size[n] = lace_size[n - 1] + snum;
1657                         total += lace_size[n];
1658                     }
1659                     lace_size[n] = size - total;
1660                     break;
1661                 }
1662             }
1663             break;
1664     }
1665
1666     if (res == 0) {
1667         for (n = 0; n < laces; n++) {
1668             if ((st->codec->codec_id == CODEC_ID_RA_288 ||
1669                  st->codec->codec_id == CODEC_ID_COOK ||
1670                  st->codec->codec_id == CODEC_ID_SIPR ||
1671                  st->codec->codec_id == CODEC_ID_ATRAC3) &&
1672                  st->codec->block_align && track->audio.sub_packet_size) {
1673                 int a = st->codec->block_align;
1674                 int sps = track->audio.sub_packet_size;
1675                 int cfs = track->audio.coded_framesize;
1676                 int h = track->audio.sub_packet_h;
1677                 int y = track->audio.sub_packet_cnt;
1678                 int w = track->audio.frame_size;
1679                 int x;
1680
1681                 if (!track->audio.pkt_cnt) {
1682                     if (st->codec->codec_id == CODEC_ID_RA_288)
1683                         for (x=0; x<h/2; x++)
1684                             memcpy(track->audio.buf+x*2*w+y*cfs,
1685                                    data+x*cfs, cfs);
1686                     else if (st->codec->codec_id == CODEC_ID_SIPR)
1687                         memcpy(track->audio.buf + y*w, data, w);
1688                     else
1689                         for (x=0; x<w/sps; x++)
1690                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1691
1692                     if (++track->audio.sub_packet_cnt >= h) {
1693                         if (st->codec->codec_id == CODEC_ID_SIPR)
1694                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
1695                         track->audio.sub_packet_cnt = 0;
1696                         track->audio.pkt_cnt = h*w / a;
1697                     }
1698                 }
1699                 while (track->audio.pkt_cnt) {
1700                     pkt = av_mallocz(sizeof(AVPacket));
1701                     av_new_packet(pkt, a);
1702                     memcpy(pkt->data, track->audio.buf
1703                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1704                     pkt->pos = pos;
1705                     pkt->stream_index = st->index;
1706                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1707                 }
1708             } else {
1709                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1710                 int offset = 0, pkt_size = lace_size[n];
1711                 uint8_t *pkt_data = data;
1712
1713                 if (lace_size[n] > size) {
1714                     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
1715                     break;
1716                 }
1717
1718                 if (encodings && encodings->scope & 1) {
1719                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1720                     if (offset < 0)
1721                         continue;
1722                 }
1723
1724                 pkt = av_mallocz(sizeof(AVPacket));
1725                 /* XXX: prevent data copy... */
1726                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1727                     av_free(pkt);
1728                     res = AVERROR(ENOMEM);
1729                     break;
1730                 }
1731                 if (offset)
1732                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1733                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1734
1735                 if (pkt_data != data)
1736                     av_free(pkt_data);
1737
1738                 if (n == 0)
1739                     pkt->flags = is_keyframe;
1740                 pkt->stream_index = st->index;
1741
1742                 if (track->ms_compat)
1743                     pkt->dts = timecode;
1744                 else
1745                     pkt->pts = timecode;
1746                 pkt->pos = pos;
1747                 if (st->codec->codec_id == CODEC_ID_TEXT)
1748                     pkt->convergence_duration = duration;
1749                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1750                     pkt->duration = duration;
1751
1752                 if (st->codec->codec_id == CODEC_ID_SSA)
1753                     matroska_fix_ass_packet(matroska, pkt, duration);
1754
1755                 if (matroska->prev_pkt &&
1756                     timecode != AV_NOPTS_VALUE &&
1757                     matroska->prev_pkt->pts == timecode &&
1758                     matroska->prev_pkt->stream_index == st->index)
1759                     matroska_merge_packets(matroska->prev_pkt, pkt);
1760                 else {
1761                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1762                     matroska->prev_pkt = pkt;
1763                 }
1764             }
1765
1766             if (timecode != AV_NOPTS_VALUE)
1767                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1768             data += lace_size[n];
1769             size -= lace_size[n];
1770         }
1771     }
1772
1773     av_free(lace_size);
1774     return res;
1775 }
1776
1777 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1778 {
1779     MatroskaCluster cluster = { 0 };
1780     EbmlList *blocks_list;
1781     MatroskaBlock *blocks;
1782     int i, res;
1783     int64_t pos = url_ftell(matroska->ctx->pb);
1784     matroska->prev_pkt = NULL;
1785     if (matroska->has_cluster_id){
1786         /* For the first cluster we parse, its ID was already read as
1787            part of matroska_read_header(), so don't read it again */
1788         res = ebml_parse_id(matroska, matroska_clusters,
1789                             MATROSKA_ID_CLUSTER, &cluster);
1790         pos -= 4;  /* sizeof the ID which was already read */
1791         matroska->has_cluster_id = 0;
1792     } else
1793         res = ebml_parse(matroska, matroska_clusters, &cluster);
1794     blocks_list = &cluster.blocks;
1795     blocks = blocks_list->elem;
1796     for (i=0; i<blocks_list->nb_elem; i++)
1797         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
1798             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
1799             res=matroska_parse_block(matroska,
1800                                      blocks[i].bin.data, blocks[i].bin.size,
1801                                      blocks[i].bin.pos,  cluster.timecode,
1802                                      blocks[i].duration, is_keyframe,
1803                                      pos);
1804         }
1805     ebml_free(matroska_cluster, &cluster);
1806     if (res < 0)  matroska->done = 1;
1807     return res;
1808 }
1809
1810 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1811 {
1812     MatroskaDemuxContext *matroska = s->priv_data;
1813
1814     while (matroska_deliver_packet(matroska, pkt)) {
1815         if (matroska->done)
1816             return AVERROR_EOF;
1817         matroska_parse_cluster(matroska);
1818     }
1819
1820     return 0;
1821 }
1822
1823 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1824                               int64_t timestamp, int flags)
1825 {
1826     MatroskaDemuxContext *matroska = s->priv_data;
1827     MatroskaTrack *tracks = matroska->tracks.elem;
1828     AVStream *st = s->streams[stream_index];
1829     int i, index, index_sub, index_min;
1830
1831     if (!st->nb_index_entries)
1832         return 0;
1833     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
1834
1835     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1836         url_fseek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
1837         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1838             matroska_clear_queue(matroska);
1839             if (matroska_parse_cluster(matroska) < 0)
1840                 break;
1841         }
1842     }
1843
1844     matroska_clear_queue(matroska);
1845     if (index < 0)
1846         return 0;
1847
1848     index_min = index;
1849     for (i=0; i < matroska->tracks.nb_elem; i++) {
1850         tracks[i].end_timecode = 0;
1851         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
1852             && !tracks[i].stream->discard != AVDISCARD_ALL) {
1853             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
1854             if (index_sub >= 0
1855                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
1856                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
1857                 index_min = index_sub;
1858         }
1859     }
1860
1861     url_fseek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
1862     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
1863     matroska->skip_to_timecode = st->index_entries[index].timestamp;
1864     matroska->done = 0;
1865     av_update_cur_dts(s, st, st->index_entries[index].timestamp);
1866     return 0;
1867 }
1868
1869 static int matroska_read_close(AVFormatContext *s)
1870 {
1871     MatroskaDemuxContext *matroska = s->priv_data;
1872     MatroskaTrack *tracks = matroska->tracks.elem;
1873     int n;
1874
1875     matroska_clear_queue(matroska);
1876
1877     for (n=0; n < matroska->tracks.nb_elem; n++)
1878         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
1879             av_free(tracks[n].audio.buf);
1880     ebml_free(matroska_segment, matroska);
1881
1882     return 0;
1883 }
1884
1885 AVInputFormat matroska_demuxer = {
1886     "matroska",
1887     NULL_IF_CONFIG_SMALL("Matroska file format"),
1888     sizeof(MatroskaDemuxContext),
1889     matroska_probe,
1890     matroska_read_header,
1891     matroska_read_packet,
1892     matroska_read_close,
1893     matroska_read_seek,
1894     .metadata_conv = ff_mkv_metadata_conv,
1895 };