Merge remote-tracking branch 'qatar/master'
[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 #include "avio_internal.h"
35 /* For ff_codec_get_id(). */
36 #include "riff.h"
37 #include "isom.h"
38 #include "rm.h"
39 #include "matroska.h"
40 #include "libavcodec/mpeg4audio.h"
41 #include "libavutil/intfloat_readwrite.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/avstring.h"
44 #include "libavutil/lzo.h"
45 #include "libavutil/dict.h"
46 #if CONFIG_ZLIB
47 #include <zlib.h>
48 #endif
49 #if CONFIG_BZLIB
50 #include <bzlib.h>
51 #endif
52
53 typedef enum {
54     EBML_NONE,
55     EBML_UINT,
56     EBML_FLOAT,
57     EBML_STR,
58     EBML_UTF8,
59     EBML_BIN,
60     EBML_NEST,
61     EBML_PASS,
62     EBML_STOP,
63     EBML_TYPE_COUNT
64 } EbmlType;
65
66 typedef const struct EbmlSyntax {
67     uint32_t id;
68     EbmlType type;
69     int list_elem_size;
70     int data_offset;
71     union {
72         uint64_t    u;
73         double      f;
74         const char *s;
75         const struct EbmlSyntax *n;
76     } def;
77 } EbmlSyntax;
78
79 typedef struct {
80     int nb_elem;
81     void *elem;
82 } EbmlList;
83
84 typedef struct {
85     int      size;
86     uint8_t *data;
87     int64_t  pos;
88 } EbmlBin;
89
90 typedef struct {
91     uint64_t version;
92     uint64_t max_size;
93     uint64_t id_length;
94     char    *doctype;
95     uint64_t doctype_version;
96 } Ebml;
97
98 typedef struct {
99     uint64_t algo;
100     EbmlBin  settings;
101 } MatroskaTrackCompression;
102
103 typedef struct {
104     uint64_t scope;
105     uint64_t type;
106     MatroskaTrackCompression compression;
107 } MatroskaTrackEncoding;
108
109 typedef struct {
110     double   frame_rate;
111     uint64_t display_width;
112     uint64_t display_height;
113     uint64_t pixel_width;
114     uint64_t pixel_height;
115     EbmlBin color_space;
116     uint64_t stereo_mode;
117 } MatroskaTrackVideo;
118
119 typedef struct {
120     double   samplerate;
121     double   out_samplerate;
122     uint64_t bitdepth;
123     uint64_t channels;
124
125     /* real audio header (extracted from extradata) */
126     int      coded_framesize;
127     int      sub_packet_h;
128     int      frame_size;
129     int      sub_packet_size;
130     int      sub_packet_cnt;
131     int      pkt_cnt;
132     uint64_t buf_timecode;
133     uint8_t *buf;
134 } MatroskaTrackAudio;
135
136 typedef struct {
137     uint64_t uid;
138     uint64_t type;
139 } MatroskaTrackPlane;
140
141 typedef struct {
142     EbmlList combine_planes;
143 } MatroskaTrackOperation;
144
145 typedef struct {
146     uint64_t num;
147     uint64_t uid;
148     uint64_t type;
149     char    *name;
150     char    *codec_id;
151     EbmlBin  codec_priv;
152     char    *language;
153     double time_scale;
154     uint64_t default_duration;
155     uint64_t flag_default;
156     uint64_t flag_forced;
157     MatroskaTrackVideo video;
158     MatroskaTrackAudio audio;
159     MatroskaTrackOperation operation;
160     EbmlList encodings;
161
162     AVStream *stream;
163     int64_t end_timecode;
164     int ms_compat;
165 } MatroskaTrack;
166
167 typedef struct {
168     uint64_t uid;
169     char *filename;
170     char *mime;
171     EbmlBin bin;
172
173     AVStream *stream;
174 } MatroskaAttachement;
175
176 typedef struct {
177     uint64_t start;
178     uint64_t end;
179     uint64_t uid;
180     char    *title;
181
182     AVChapter *chapter;
183 } MatroskaChapter;
184
185 typedef struct {
186     uint64_t track;
187     uint64_t pos;
188 } MatroskaIndexPos;
189
190 typedef struct {
191     uint64_t time;
192     EbmlList pos;
193 } MatroskaIndex;
194
195 typedef struct {
196     char *name;
197     char *string;
198     char *lang;
199     uint64_t def;
200     EbmlList sub;
201 } MatroskaTag;
202
203 typedef struct {
204     char    *type;
205     uint64_t typevalue;
206     uint64_t trackuid;
207     uint64_t chapteruid;
208     uint64_t attachuid;
209 } MatroskaTagTarget;
210
211 typedef struct {
212     MatroskaTagTarget target;
213     EbmlList tag;
214 } MatroskaTags;
215
216 typedef struct {
217     uint64_t id;
218     uint64_t pos;
219 } MatroskaSeekhead;
220
221 typedef struct {
222     uint64_t start;
223     uint64_t length;
224 } MatroskaLevel;
225
226 typedef struct {
227     AVFormatContext *ctx;
228
229     /* EBML stuff */
230     int num_levels;
231     MatroskaLevel levels[EBML_MAX_DEPTH];
232     int level_up;
233     uint32_t current_id;
234
235     uint64_t time_scale;
236     double   duration;
237     char    *title;
238     EbmlList tracks;
239     EbmlList attachments;
240     EbmlList chapters;
241     EbmlList index;
242     EbmlList tags;
243     EbmlList seekhead;
244
245     /* byte position of the segment inside the stream */
246     int64_t segment_start;
247
248     /* the packet queue */
249     AVPacket **packets;
250     int num_packets;
251     AVPacket *prev_pkt;
252
253     int done;
254
255     /* What to skip before effectively reading a packet. */
256     int skip_to_keyframe;
257     uint64_t skip_to_timecode;
258
259     /* File has a CUES element, but we defer parsing until it is needed. */
260     int cues_parsing_deferred;
261 } MatroskaDemuxContext;
262
263 typedef struct {
264     uint64_t duration;
265     int64_t  reference;
266     uint64_t non_simple;
267     EbmlBin  bin;
268 } MatroskaBlock;
269
270 typedef struct {
271     uint64_t timecode;
272     EbmlList blocks;
273 } MatroskaCluster;
274
275 static EbmlSyntax ebml_header[] = {
276     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
277     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
278     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
279     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
280     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
281     { EBML_ID_EBMLVERSION,            EBML_NONE },
282     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
283     { 0 }
284 };
285
286 static EbmlSyntax ebml_syntax[] = {
287     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
288     { 0 }
289 };
290
291 static EbmlSyntax matroska_info[] = {
292     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
293     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
294     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
295     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
296     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
297     { MATROSKA_ID_DATEUTC,            EBML_NONE },
298     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
299     { 0 }
300 };
301
302 static EbmlSyntax matroska_track_video[] = {
303     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
304     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
305     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
306     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
307     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
308     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_BIN,  0, offsetof(MatroskaTrackVideo,color_space) },
309     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,stereo_mode) },
310     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
311     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
312     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
313     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
314     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
315     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
316     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
317     { 0 }
318 };
319
320 static EbmlSyntax matroska_track_audio[] = {
321     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
322     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
323     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
324     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
325     { 0 }
326 };
327
328 static EbmlSyntax matroska_track_encoding_compression[] = {
329     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
330     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
331     { 0 }
332 };
333
334 static EbmlSyntax matroska_track_encoding[] = {
335     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
336     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
337     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
338     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
339     { 0 }
340 };
341
342 static EbmlSyntax matroska_track_encodings[] = {
343     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
344     { 0 }
345 };
346
347 static EbmlSyntax matroska_track_plane[] = {
348     { MATROSKA_ID_TRACKPLANEUID,  EBML_UINT, 0, offsetof(MatroskaTrackPlane,uid) },
349     { MATROSKA_ID_TRACKPLANETYPE, EBML_UINT, 0, offsetof(MatroskaTrackPlane,type) },
350     { 0 }
351 };
352
353 static EbmlSyntax matroska_track_combine_planes[] = {
354     { MATROSKA_ID_TRACKPLANE, EBML_NEST, sizeof(MatroskaTrackPlane), offsetof(MatroskaTrackOperation,combine_planes), {.n=matroska_track_plane} },
355     { 0 }
356 };
357
358 static EbmlSyntax matroska_track_operation[] = {
359     { MATROSKA_ID_TRACKCOMBINEPLANES, EBML_NEST, 0, 0, {.n=matroska_track_combine_planes} },
360     { 0 }
361 };
362
363 static EbmlSyntax matroska_track[] = {
364     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
365     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
366     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
367     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
368     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
369     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
370     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
371     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
372     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
373     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
374     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_UINT, 0, offsetof(MatroskaTrack,flag_forced), {.u=0} },
375     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
376     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
377     { MATROSKA_ID_TRACKOPERATION,       EBML_NEST, 0, offsetof(MatroskaTrack,operation), {.n=matroska_track_operation} },
378     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
379     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
380     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
381     { MATROSKA_ID_CODECNAME,            EBML_NONE },
382     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
383     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
384     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
385     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
386     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
387     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
388     { 0 }
389 };
390
391 static EbmlSyntax matroska_tracks[] = {
392     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
393     { 0 }
394 };
395
396 static EbmlSyntax matroska_attachment[] = {
397     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
398     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
399     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
400     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
401     { MATROSKA_ID_FILEDESC,           EBML_NONE },
402     { 0 }
403 };
404
405 static EbmlSyntax matroska_attachments[] = {
406     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
407     { 0 }
408 };
409
410 static EbmlSyntax matroska_chapter_display[] = {
411     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
412     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
413     { 0 }
414 };
415
416 static EbmlSyntax matroska_chapter_entry[] = {
417     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
418     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
419     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
420     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
421     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
422     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
423     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
424     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
425     { 0 }
426 };
427
428 static EbmlSyntax matroska_chapter[] = {
429     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
430     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
431     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
432     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
433     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
434     { 0 }
435 };
436
437 static EbmlSyntax matroska_chapters[] = {
438     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
439     { 0 }
440 };
441
442 static EbmlSyntax matroska_index_pos[] = {
443     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
444     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
445     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
446     { 0 }
447 };
448
449 static EbmlSyntax matroska_index_entry[] = {
450     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
451     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
452     { 0 }
453 };
454
455 static EbmlSyntax matroska_index[] = {
456     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
457     { 0 }
458 };
459
460 static EbmlSyntax matroska_simpletag[] = {
461     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
462     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
463     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
464     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
465     { MATROSKA_ID_TAGDEFAULT_BUG,     EBML_UINT, 0, offsetof(MatroskaTag,def) },
466     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
467     { 0 }
468 };
469
470 static EbmlSyntax matroska_tagtargets[] = {
471     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
472     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
473     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
474     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
475     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
476     { 0 }
477 };
478
479 static EbmlSyntax matroska_tag[] = {
480     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
481     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
482     { 0 }
483 };
484
485 static EbmlSyntax matroska_tags[] = {
486     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
487     { 0 }
488 };
489
490 static EbmlSyntax matroska_seekhead_entry[] = {
491     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
492     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
493     { 0 }
494 };
495
496 static EbmlSyntax matroska_seekhead[] = {
497     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
498     { 0 }
499 };
500
501 static EbmlSyntax matroska_segment[] = {
502     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
503     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
504     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
505     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
506     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
507     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
508     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
509     { MATROSKA_ID_CLUSTER,        EBML_STOP },
510     { 0 }
511 };
512
513 static EbmlSyntax matroska_segments[] = {
514     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
515     { 0 }
516 };
517
518 static EbmlSyntax matroska_blockgroup[] = {
519     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
520     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
521     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration) },
522     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
523     { 1,                          EBML_UINT, 0, offsetof(MatroskaBlock,non_simple), {.u=1} },
524     { 0 }
525 };
526
527 static EbmlSyntax matroska_cluster[] = {
528     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
529     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
530     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
531     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
532     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
533     { 0 }
534 };
535
536 static EbmlSyntax matroska_clusters[] = {
537     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
538     { MATROSKA_ID_INFO,           EBML_NONE },
539     { MATROSKA_ID_CUES,           EBML_NONE },
540     { MATROSKA_ID_TAGS,           EBML_NONE },
541     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
542     { 0 }
543 };
544
545 static const char *matroska_doctypes[] = { "matroska", "webm" };
546
547 /*
548  * Return: Whether we reached the end of a level in the hierarchy or not.
549  */
550 static int ebml_level_end(MatroskaDemuxContext *matroska)
551 {
552     AVIOContext *pb = matroska->ctx->pb;
553     int64_t pos = avio_tell(pb);
554
555     if (matroska->num_levels > 0) {
556         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
557         if (pos - level->start >= level->length || matroska->current_id) {
558             matroska->num_levels--;
559             return 1;
560         }
561     }
562     return 0;
563 }
564
565 /*
566  * Read: an "EBML number", which is defined as a variable-length
567  * array of bytes. The first byte indicates the length by giving a
568  * number of 0-bits followed by a one. The position of the first
569  * "one" bit inside the first byte indicates the length of this
570  * number.
571  * Returns: number of bytes read, < 0 on error
572  */
573 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
574                          int max_size, uint64_t *number)
575 {
576     int read = 1, n = 1;
577     uint64_t total = 0;
578
579     /* The first byte tells us the length in bytes - avio_r8() can normally
580      * return 0, but since that's not a valid first ebmlID byte, we can
581      * use it safely here to catch EOS. */
582     if (!(total = avio_r8(pb))) {
583         /* we might encounter EOS here */
584         if (!url_feof(pb)) {
585             int64_t pos = avio_tell(pb);
586             av_log(matroska->ctx, AV_LOG_ERROR,
587                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
588                    pos, pos);
589         }
590         return AVERROR(EIO); /* EOS or actual I/O error */
591     }
592
593     /* get the length of the EBML number */
594     read = 8 - ff_log2_tab[total];
595     if (read > max_size) {
596         int64_t pos = avio_tell(pb) - 1;
597         av_log(matroska->ctx, AV_LOG_ERROR,
598                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
599                (uint8_t) total, pos, pos);
600         return AVERROR_INVALIDDATA;
601     }
602
603     /* read out length */
604     total ^= 1 << ff_log2_tab[total];
605     while (n++ < read)
606         total = (total << 8) | avio_r8(pb);
607
608     *number = total;
609
610     return read;
611 }
612
613 /**
614  * Read a EBML length value.
615  * This needs special handling for the "unknown length" case which has multiple
616  * encodings.
617  */
618 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
619                             uint64_t *number)
620 {
621     int res = ebml_read_num(matroska, pb, 8, number);
622     if (res > 0 && *number + 1 == 1ULL << (7 * res))
623         *number = 0xffffffffffffffULL;
624     return res;
625 }
626
627 /*
628  * Read the next element as an unsigned int.
629  * 0 is success, < 0 is failure.
630  */
631 static int ebml_read_uint(AVIOContext *pb, int size, uint64_t *num)
632 {
633     int n = 0;
634
635     if (size > 8)
636         return AVERROR_INVALIDDATA;
637
638     /* big-endian ordering; build up number */
639     *num = 0;
640     while (n++ < size)
641         *num = (*num << 8) | avio_r8(pb);
642
643     return 0;
644 }
645
646 /*
647  * Read the next element as a float.
648  * 0 is success, < 0 is failure.
649  */
650 static int ebml_read_float(AVIOContext *pb, int size, double *num)
651 {
652     if (size == 0) {
653         *num = 0;
654     } else if (size == 4) {
655         *num= av_int2flt(avio_rb32(pb));
656     } else if(size==8){
657         *num= av_int2dbl(avio_rb64(pb));
658     } else
659         return AVERROR_INVALIDDATA;
660
661     return 0;
662 }
663
664 /*
665  * Read the next element as an ASCII string.
666  * 0 is success, < 0 is failure.
667  */
668 static int ebml_read_ascii(AVIOContext *pb, int size, char **str)
669 {
670     av_free(*str);
671     /* EBML strings are usually not 0-terminated, so we allocate one
672      * byte more, read the string and NULL-terminate it ourselves. */
673     if (!(*str = av_malloc(size + 1)))
674         return AVERROR(ENOMEM);
675     if (avio_read(pb, (uint8_t *) *str, size) != size) {
676         av_freep(str);
677         return AVERROR(EIO);
678     }
679     (*str)[size] = '\0';
680
681     return 0;
682 }
683
684 /*
685  * Read the next element as binary data.
686  * 0 is success, < 0 is failure.
687  */
688 static int ebml_read_binary(AVIOContext *pb, int length, EbmlBin *bin)
689 {
690     av_free(bin->data);
691     if (!(bin->data = av_malloc(length)))
692         return AVERROR(ENOMEM);
693
694     bin->size = length;
695     bin->pos  = avio_tell(pb);
696     if (avio_read(pb, bin->data, length) != length) {
697         av_freep(&bin->data);
698         return AVERROR(EIO);
699     }
700
701     return 0;
702 }
703
704 /*
705  * Read the next element, but only the header. The contents
706  * are supposed to be sub-elements which can be read separately.
707  * 0 is success, < 0 is failure.
708  */
709 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
710 {
711     AVIOContext *pb = matroska->ctx->pb;
712     MatroskaLevel *level;
713
714     if (matroska->num_levels >= EBML_MAX_DEPTH) {
715         av_log(matroska->ctx, AV_LOG_ERROR,
716                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
717         return AVERROR(ENOSYS);
718     }
719
720     level = &matroska->levels[matroska->num_levels++];
721     level->start = avio_tell(pb);
722     level->length = length;
723
724     return 0;
725 }
726
727 /*
728  * Read signed/unsigned "EBML" numbers.
729  * Return: number of bytes processed, < 0 on error
730  */
731 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
732                                  uint8_t *data, uint32_t size, uint64_t *num)
733 {
734     AVIOContext pb;
735     ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
736     return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
737 }
738
739 /*
740  * Same as above, but signed.
741  */
742 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
743                                  uint8_t *data, uint32_t size, int64_t *num)
744 {
745     uint64_t unum;
746     int res;
747
748     /* read as unsigned number first */
749     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
750         return res;
751
752     /* make signed (weird way) */
753     *num = unum - ((1LL << (7*res - 1)) - 1);
754
755     return res;
756 }
757
758 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
759                            EbmlSyntax *syntax, void *data);
760
761 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
762                          uint32_t id, void *data)
763 {
764     int i;
765     for (i=0; syntax[i].id; i++)
766         if (id == syntax[i].id)
767             break;
768     if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
769         matroska->num_levels > 0 &&
770         matroska->levels[matroska->num_levels-1].length == 0xffffffffffffff)
771         return 0;  // we reached the end of an unknown size cluster
772     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32)
773         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
774     return ebml_parse_elem(matroska, &syntax[i], data);
775 }
776
777 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
778                       void *data)
779 {
780     if (!matroska->current_id) {
781         uint64_t id;
782         int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
783         if (res < 0)
784             return res;
785         matroska->current_id = id | 1 << 7*res;
786     }
787     return ebml_parse_id(matroska, syntax, matroska->current_id, data);
788 }
789
790 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
791                            void *data)
792 {
793     int i, res = 0;
794
795     for (i=0; syntax[i].id; i++)
796         switch (syntax[i].type) {
797         case EBML_UINT:
798             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
799             break;
800         case EBML_FLOAT:
801             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
802             break;
803         case EBML_STR:
804         case EBML_UTF8:
805             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
806             break;
807         }
808
809     while (!res && !ebml_level_end(matroska))
810         res = ebml_parse(matroska, syntax, data);
811
812     return res;
813 }
814
815 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
816                            EbmlSyntax *syntax, void *data)
817 {
818     static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
819         [EBML_UINT]  = 8,
820         [EBML_FLOAT] = 8,
821         // max. 16 MB for strings
822         [EBML_STR]   = 0x1000000,
823         [EBML_UTF8]  = 0x1000000,
824         // max. 256 MB for binary data
825         [EBML_BIN]   = 0x10000000,
826         // no limits for anything else
827     };
828     AVIOContext *pb = matroska->ctx->pb;
829     uint32_t id = syntax->id;
830     uint64_t length;
831     int res;
832     void *newelem;
833
834     data = (char *)data + syntax->data_offset;
835     if (syntax->list_elem_size) {
836         EbmlList *list = data;
837         newelem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
838         if (!newelem)
839             return AVERROR(ENOMEM);
840         list->elem = newelem;
841         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
842         memset(data, 0, syntax->list_elem_size);
843         list->nb_elem++;
844     }
845
846     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
847         matroska->current_id = 0;
848         if ((res = ebml_read_length(matroska, pb, &length)) < 0)
849             return res;
850         if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
851             av_log(matroska->ctx, AV_LOG_ERROR,
852                    "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for syntax element %i\n",
853                    length, max_lengths[syntax->type], syntax->type);
854             return AVERROR_INVALIDDATA;
855         }
856     }
857
858     switch (syntax->type) {
859     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
860     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
861     case EBML_STR:
862     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
863     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
864     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
865                          return res;
866                      if (id == MATROSKA_ID_SEGMENT)
867                          matroska->segment_start = avio_tell(matroska->ctx->pb);
868                      return ebml_parse_nest(matroska, syntax->def.n, data);
869     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
870     case EBML_STOP:  return 1;
871     default:         return avio_skip(pb,length)<0 ? AVERROR(EIO) : 0;
872     }
873     if (res == AVERROR_INVALIDDATA)
874         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
875     else if (res == AVERROR(EIO))
876         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
877     return res;
878 }
879
880 static void ebml_free(EbmlSyntax *syntax, void *data)
881 {
882     int i, j;
883     for (i=0; syntax[i].id; i++) {
884         void *data_off = (char *)data + syntax[i].data_offset;
885         switch (syntax[i].type) {
886         case EBML_STR:
887         case EBML_UTF8:  av_freep(data_off);                      break;
888         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
889         case EBML_NEST:
890             if (syntax[i].list_elem_size) {
891                 EbmlList *list = data_off;
892                 char *ptr = list->elem;
893                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
894                     ebml_free(syntax[i].def.n, ptr);
895                 av_free(list->elem);
896             } else
897                 ebml_free(syntax[i].def.n, data_off);
898         default:  break;
899         }
900     }
901 }
902
903
904 /*
905  * Autodetecting...
906  */
907 static int matroska_probe(AVProbeData *p)
908 {
909     uint64_t total = 0;
910     int len_mask = 0x80, size = 1, n = 1, i;
911
912     /* EBML header? */
913     if (AV_RB32(p->buf) != EBML_ID_HEADER)
914         return 0;
915
916     /* length of header */
917     total = p->buf[4];
918     while (size <= 8 && !(total & len_mask)) {
919         size++;
920         len_mask >>= 1;
921     }
922     if (size > 8)
923       return 0;
924     total &= (len_mask - 1);
925     while (n < size)
926         total = (total << 8) | p->buf[4 + n++];
927
928     /* Does the probe data contain the whole header? */
929     if (p->buf_size < 4 + size + total)
930       return 0;
931
932     /* The header should contain a known document type. For now,
933      * we don't parse the whole header but simply check for the
934      * availability of that array of characters inside the header.
935      * Not fully fool-proof, but good enough. */
936     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
937         int probelen = strlen(matroska_doctypes[i]);
938         if (total < probelen)
939             continue;
940         for (n = 4+size; n <= 4+size+total-probelen; n++)
941             if (!memcmp(p->buf+n, matroska_doctypes[i], probelen))
942                 return AVPROBE_SCORE_MAX;
943     }
944
945     // probably valid EBML header but no recognized doctype
946     return AVPROBE_SCORE_MAX/2;
947 }
948
949 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
950                                                  int num)
951 {
952     MatroskaTrack *tracks = matroska->tracks.elem;
953     int i;
954
955     for (i=0; i < matroska->tracks.nb_elem; i++)
956         if (tracks[i].num == num)
957             return &tracks[i];
958
959     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
960     return NULL;
961 }
962
963 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
964                                   MatroskaTrack *track)
965 {
966     MatroskaTrackEncoding *encodings = track->encodings.elem;
967     uint8_t* data = *buf;
968     int isize = *buf_size;
969     uint8_t* pkt_data = NULL;
970     uint8_t* newpktdata;
971     int pkt_size = isize;
972     int result = 0;
973     int olen;
974
975     if (pkt_size >= 10000000)
976         return -1;
977
978     switch (encodings[0].compression.algo) {
979     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
980         return encodings[0].compression.settings.size;
981     case MATROSKA_TRACK_ENCODING_COMP_LZO:
982         do {
983             olen = pkt_size *= 3;
984             pkt_data = av_realloc(pkt_data, pkt_size+AV_LZO_OUTPUT_PADDING);
985             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
986         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
987         if (result)
988             goto failed;
989         pkt_size -= olen;
990         break;
991 #if CONFIG_ZLIB
992     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
993         z_stream zstream = {0};
994         if (inflateInit(&zstream) != Z_OK)
995             return -1;
996         zstream.next_in = data;
997         zstream.avail_in = isize;
998         do {
999             pkt_size *= 3;
1000             newpktdata = av_realloc(pkt_data, pkt_size);
1001             if (!newpktdata) {
1002                 inflateEnd(&zstream);
1003                 goto failed;
1004             }
1005             pkt_data = newpktdata;
1006             zstream.avail_out = pkt_size - zstream.total_out;
1007             zstream.next_out = pkt_data + zstream.total_out;
1008             if (pkt_data) {
1009                 result = inflate(&zstream, Z_NO_FLUSH);
1010             } else
1011                 result = Z_MEM_ERROR;
1012         } while (result==Z_OK && pkt_size<10000000);
1013         pkt_size = zstream.total_out;
1014         inflateEnd(&zstream);
1015         if (result != Z_STREAM_END)
1016             goto failed;
1017         break;
1018     }
1019 #endif
1020 #if CONFIG_BZLIB
1021     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
1022         bz_stream bzstream = {0};
1023         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
1024             return -1;
1025         bzstream.next_in = data;
1026         bzstream.avail_in = isize;
1027         do {
1028             pkt_size *= 3;
1029             newpktdata = av_realloc(pkt_data, pkt_size);
1030             if (!newpktdata) {
1031                 BZ2_bzDecompressEnd(&bzstream);
1032                 goto failed;
1033             }
1034             pkt_data = newpktdata;
1035             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1036             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1037             if (pkt_data) {
1038                 result = BZ2_bzDecompress(&bzstream);
1039             } else
1040                 result = BZ_MEM_ERROR;
1041         } while (result==BZ_OK && pkt_size<10000000);
1042         pkt_size = bzstream.total_out_lo32;
1043         BZ2_bzDecompressEnd(&bzstream);
1044         if (result != BZ_STREAM_END)
1045             goto failed;
1046         break;
1047     }
1048 #endif
1049     default:
1050         return -1;
1051     }
1052
1053     *buf = pkt_data;
1054     *buf_size = pkt_size;
1055     return 0;
1056  failed:
1057     av_free(pkt_data);
1058     return -1;
1059 }
1060
1061 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
1062                                     AVPacket *pkt, uint64_t display_duration)
1063 {
1064     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
1065     for (; *ptr!=',' && ptr<end-1; ptr++);
1066     if (*ptr == ',')
1067         ptr++;
1068     layer = ptr;
1069     for (; *ptr!=',' && ptr<end-1; ptr++);
1070     if (*ptr == ',') {
1071         int64_t end_pts = pkt->pts + display_duration;
1072         int sc = matroska->time_scale * pkt->pts / 10000000;
1073         int ec = matroska->time_scale * end_pts  / 10000000;
1074         int sh, sm, ss, eh, em, es, len;
1075         sh = sc/360000;  sc -= 360000*sh;
1076         sm = sc/  6000;  sc -=   6000*sm;
1077         ss = sc/   100;  sc -=    100*ss;
1078         eh = ec/360000;  ec -= 360000*eh;
1079         em = ec/  6000;  ec -=   6000*em;
1080         es = ec/   100;  ec -=    100*es;
1081         *ptr++ = '\0';
1082         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
1083         if (!(line = av_malloc(len)))
1084             return;
1085         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
1086                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
1087         av_free(pkt->data);
1088         pkt->data = line;
1089         pkt->size = strlen(line);
1090     }
1091 }
1092
1093 static int matroska_merge_packets(AVPacket *out, AVPacket *in)
1094 {
1095     void *newdata = av_realloc(out->data, out->size+in->size);
1096     if (!newdata)
1097         return AVERROR(ENOMEM);
1098     out->data = newdata;
1099     memcpy(out->data+out->size, in->data, in->size);
1100     out->size += in->size;
1101     av_destruct_packet(in);
1102     av_free(in);
1103     return 0;
1104 }
1105
1106 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1107                                  AVDictionary **metadata, char *prefix)
1108 {
1109     MatroskaTag *tags = list->elem;
1110     char key[1024];
1111     int i;
1112
1113     for (i=0; i < list->nb_elem; i++) {
1114         const char *lang = strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1115
1116         if (!tags[i].name) {
1117             av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1118             continue;
1119         }
1120         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1121         else         av_strlcpy(key, tags[i].name, sizeof(key));
1122         if (tags[i].def || !lang) {
1123         av_dict_set(metadata, key, tags[i].string, 0);
1124         if (tags[i].sub.nb_elem)
1125             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1126         }
1127         if (lang) {
1128             av_strlcat(key, "-", sizeof(key));
1129             av_strlcat(key, lang, sizeof(key));
1130             av_dict_set(metadata, key, tags[i].string, 0);
1131             if (tags[i].sub.nb_elem)
1132                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1133         }
1134     }
1135     ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1136 }
1137
1138 static void matroska_convert_tags(AVFormatContext *s)
1139 {
1140     MatroskaDemuxContext *matroska = s->priv_data;
1141     MatroskaTags *tags = matroska->tags.elem;
1142     int i, j;
1143
1144     for (i=0; i < matroska->tags.nb_elem; i++) {
1145         if (tags[i].target.attachuid) {
1146             MatroskaAttachement *attachment = matroska->attachments.elem;
1147             for (j=0; j<matroska->attachments.nb_elem; j++)
1148                 if (attachment[j].uid == tags[i].target.attachuid
1149                     && attachment[j].stream)
1150                     matroska_convert_tag(s, &tags[i].tag,
1151                                          &attachment[j].stream->metadata, NULL);
1152         } else if (tags[i].target.chapteruid) {
1153             MatroskaChapter *chapter = matroska->chapters.elem;
1154             for (j=0; j<matroska->chapters.nb_elem; j++)
1155                 if (chapter[j].uid == tags[i].target.chapteruid
1156                     && chapter[j].chapter)
1157                     matroska_convert_tag(s, &tags[i].tag,
1158                                          &chapter[j].chapter->metadata, NULL);
1159         } else if (tags[i].target.trackuid) {
1160             MatroskaTrack *track = matroska->tracks.elem;
1161             for (j=0; j<matroska->tracks.nb_elem; j++)
1162                 if (track[j].uid == tags[i].target.trackuid && track[j].stream)
1163                     matroska_convert_tag(s, &tags[i].tag,
1164                                          &track[j].stream->metadata, NULL);
1165         } else {
1166             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1167                                  tags[i].target.type);
1168         }
1169     }
1170 }
1171
1172 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska, int idx)
1173 {
1174     EbmlList *seekhead_list = &matroska->seekhead;
1175     MatroskaSeekhead *seekhead = seekhead_list->elem;
1176     uint32_t level_up = matroska->level_up;
1177     int64_t before_pos = avio_tell(matroska->ctx->pb);
1178     uint32_t saved_id = matroska->current_id;
1179     MatroskaLevel level;
1180     int64_t offset;
1181     int ret = 0;
1182
1183     if (idx >= seekhead_list->nb_elem
1184             || seekhead[idx].id == MATROSKA_ID_SEEKHEAD
1185             || seekhead[idx].id == MATROSKA_ID_CLUSTER)
1186         return 0;
1187
1188     /* seek */
1189     offset = seekhead[idx].pos + matroska->segment_start;
1190     if (avio_seek(matroska->ctx->pb, offset, SEEK_SET) == offset) {
1191         /* We don't want to lose our seekhead level, so we add
1192          * a dummy. This is a crude hack. */
1193         if (matroska->num_levels == EBML_MAX_DEPTH) {
1194             av_log(matroska->ctx, AV_LOG_INFO,
1195                    "Max EBML element depth (%d) reached, "
1196                    "cannot parse further.\n", EBML_MAX_DEPTH);
1197             ret = AVERROR_INVALIDDATA;
1198         } else {
1199             level.start = 0;
1200             level.length = (uint64_t)-1;
1201             matroska->levels[matroska->num_levels] = level;
1202             matroska->num_levels++;
1203             matroska->current_id = 0;
1204
1205             ret = ebml_parse(matroska, matroska_segment, matroska);
1206
1207             /* remove dummy level */
1208             while (matroska->num_levels) {
1209                 uint64_t length = matroska->levels[--matroska->num_levels].length;
1210                 if (length == (uint64_t)-1)
1211                     break;
1212             }
1213         }
1214     }
1215     /* seek back */
1216     avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
1217     matroska->level_up = level_up;
1218     matroska->current_id = saved_id;
1219
1220     return ret;
1221 }
1222
1223 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1224 {
1225     EbmlList *seekhead_list = &matroska->seekhead;
1226     MatroskaSeekhead *seekhead = seekhead_list->elem;
1227     int64_t before_pos = avio_tell(matroska->ctx->pb);
1228     int i;
1229
1230     // we should not do any seeking in the streaming case
1231     if (!matroska->ctx->pb->seekable ||
1232         (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1233         return;
1234
1235     for (i = 0; i < seekhead_list->nb_elem; i++) {
1236         if (seekhead[i].pos <= before_pos)
1237             continue;
1238
1239         // defer cues parsing until we actually need cue data.
1240         if (seekhead[i].id == MATROSKA_ID_CUES) {
1241             matroska->cues_parsing_deferred = 1;
1242             continue;
1243         }
1244
1245         if (matroska_parse_seekhead_entry(matroska, i) < 0)
1246             break;
1247     }
1248 }
1249
1250 static void matroska_add_index_entries(MatroskaDemuxContext *matroska) {
1251     EbmlList *index_list;
1252     MatroskaIndex *index;
1253     int index_scale = 1;
1254     int i, j;
1255
1256     index_list = &matroska->index;
1257     index = index_list->elem;
1258     if (index_list->nb_elem
1259         && index[0].time > 1E14/matroska->time_scale) {
1260         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1261         index_scale = matroska->time_scale;
1262     }
1263     for (i = 0; i < index_list->nb_elem; i++) {
1264         EbmlList *pos_list = &index[i].pos;
1265         MatroskaIndexPos *pos = pos_list->elem;
1266         for (j = 0; j < pos_list->nb_elem; j++) {
1267             MatroskaTrack *track = matroska_find_track_by_num(matroska, pos[j].track);
1268             if (track && track->stream)
1269                 av_add_index_entry(track->stream,
1270                                    pos[j].pos + matroska->segment_start,
1271                                    index[i].time/index_scale, 0, 0,
1272                                    AVINDEX_KEYFRAME);
1273         }
1274     }
1275 }
1276
1277 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1278     EbmlList *seekhead_list = &matroska->seekhead;
1279     MatroskaSeekhead *seekhead = seekhead_list->elem;
1280     int i;
1281
1282     for (i = 0; i < seekhead_list->nb_elem; i++)
1283         if (seekhead[i].id == MATROSKA_ID_CUES)
1284             break;
1285     assert(i <= seekhead_list->nb_elem);
1286
1287     matroska_parse_seekhead_entry(matroska, i);
1288     matroska_add_index_entries(matroska);
1289 }
1290
1291 static int matroska_aac_profile(char *codec_id)
1292 {
1293     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1294     int profile;
1295
1296     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1297         if (strstr(codec_id, aac_profiles[profile]))
1298             break;
1299     return profile + 1;
1300 }
1301
1302 static int matroska_aac_sri(int samplerate)
1303 {
1304     int sri;
1305
1306     for (sri=0; sri<FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1307         if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
1308             break;
1309     return sri;
1310 }
1311
1312 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
1313 {
1314     MatroskaDemuxContext *matroska = s->priv_data;
1315     EbmlList *attachements_list = &matroska->attachments;
1316     MatroskaAttachement *attachements;
1317     EbmlList *chapters_list = &matroska->chapters;
1318     MatroskaChapter *chapters;
1319     MatroskaTrack *tracks;
1320     uint64_t max_start = 0;
1321     Ebml ebml = { 0 };
1322     AVStream *st;
1323     int i, j, k, res;
1324
1325     matroska->ctx = s;
1326
1327     /* First read the EBML header. */
1328     if (ebml_parse(matroska, ebml_syntax, &ebml)
1329         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1330         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 3) {
1331         av_log(matroska->ctx, AV_LOG_ERROR,
1332                "EBML header using unsupported features\n"
1333                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1334                ebml.version, ebml.doctype, ebml.doctype_version);
1335         ebml_free(ebml_syntax, &ebml);
1336         return AVERROR_PATCHWELCOME;
1337     } else if (ebml.doctype_version == 3) {
1338         av_log(matroska->ctx, AV_LOG_WARNING,
1339                "EBML header using unsupported features\n"
1340                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1341                ebml.version, ebml.doctype, ebml.doctype_version);
1342     }
1343     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1344         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1345             break;
1346     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1347         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1348     }
1349     ebml_free(ebml_syntax, &ebml);
1350
1351     /* The next thing is a segment. */
1352     if ((res = ebml_parse(matroska, matroska_segments, matroska)) < 0)
1353         return res;
1354     matroska_execute_seekhead(matroska);
1355
1356     if (!matroska->time_scale)
1357         matroska->time_scale = 1000000;
1358     if (matroska->duration)
1359         matroska->ctx->duration = matroska->duration * matroska->time_scale
1360                                   * 1000 / AV_TIME_BASE;
1361     av_dict_set(&s->metadata, "title", matroska->title, 0);
1362
1363     tracks = matroska->tracks.elem;
1364     for (i=0; i < matroska->tracks.nb_elem; i++) {
1365         MatroskaTrack *track = &tracks[i];
1366         enum CodecID codec_id = CODEC_ID_NONE;
1367         EbmlList *encodings_list = &track->encodings;
1368         MatroskaTrackEncoding *encodings = encodings_list->elem;
1369         uint8_t *extradata = NULL;
1370         int extradata_size = 0;
1371         int extradata_offset = 0;
1372         uint32_t fourcc = 0;
1373         AVIOContext b;
1374
1375         /* Apply some sanity checks. */
1376         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1377             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1378             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1379             av_log(matroska->ctx, AV_LOG_INFO,
1380                    "Unknown or unsupported track type %"PRIu64"\n",
1381                    track->type);
1382             continue;
1383         }
1384         if (track->codec_id == NULL)
1385             continue;
1386
1387         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1388             if (!track->default_duration)
1389                 track->default_duration = 1000000000/track->video.frame_rate;
1390             if (!track->video.display_width)
1391                 track->video.display_width = track->video.pixel_width;
1392             if (!track->video.display_height)
1393                 track->video.display_height = track->video.pixel_height;
1394             if (track->video.color_space.size == 4)
1395                 fourcc = AV_RL32(track->video.color_space.data);
1396         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1397             if (!track->audio.out_samplerate)
1398                 track->audio.out_samplerate = track->audio.samplerate;
1399         }
1400         if (encodings_list->nb_elem > 1) {
1401             av_log(matroska->ctx, AV_LOG_ERROR,
1402                    "Multiple combined encodings not supported");
1403         } else if (encodings_list->nb_elem == 1) {
1404             if (encodings[0].type ||
1405                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1406 #if CONFIG_ZLIB
1407                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1408 #endif
1409 #if CONFIG_BZLIB
1410                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1411 #endif
1412                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1413                 encodings[0].scope = 0;
1414                 av_log(matroska->ctx, AV_LOG_ERROR,
1415                        "Unsupported encoding type");
1416             } else if (track->codec_priv.size && encodings[0].scope&2) {
1417                 uint8_t *codec_priv = track->codec_priv.data;
1418                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1419                                                     &track->codec_priv.size,
1420                                                     track);
1421                 if (offset < 0) {
1422                     track->codec_priv.data = NULL;
1423                     track->codec_priv.size = 0;
1424                     av_log(matroska->ctx, AV_LOG_ERROR,
1425                            "Failed to decode codec private data\n");
1426                 } else if (offset > 0) {
1427                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1428                     memcpy(track->codec_priv.data,
1429                            encodings[0].compression.settings.data, offset);
1430                     memcpy(track->codec_priv.data+offset, codec_priv,
1431                            track->codec_priv.size);
1432                     track->codec_priv.size += offset;
1433                 }
1434                 if (codec_priv != track->codec_priv.data)
1435                     av_free(codec_priv);
1436             }
1437         }
1438
1439         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1440             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1441                         strlen(ff_mkv_codec_tags[j].str))){
1442                 codec_id= ff_mkv_codec_tags[j].id;
1443                 break;
1444             }
1445         }
1446
1447         st = track->stream = avformat_new_stream(s, NULL);
1448         if (st == NULL)
1449             return AVERROR(ENOMEM);
1450
1451         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1452             && track->codec_priv.size >= 40
1453             && track->codec_priv.data != NULL) {
1454             track->ms_compat = 1;
1455             fourcc = AV_RL32(track->codec_priv.data + 16);
1456             codec_id = ff_codec_get_id(ff_codec_bmp_tags, fourcc);
1457             extradata_offset = 40;
1458         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1459                    && track->codec_priv.size >= 14
1460                    && track->codec_priv.data != NULL) {
1461             int ret;
1462             ffio_init_context(&b, track->codec_priv.data, track->codec_priv.size,
1463                           AVIO_FLAG_READ, NULL, NULL, NULL, NULL);
1464             ret = ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1465             if (ret < 0)
1466                 return ret;
1467             codec_id = st->codec->codec_id;
1468             extradata_offset = FFMIN(track->codec_priv.size, 18);
1469         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1470                    && (track->codec_priv.size >= 86)
1471                    && (track->codec_priv.data != NULL)) {
1472             fourcc = AV_RL32(track->codec_priv.data);
1473             codec_id = ff_codec_get_id(codec_movvideo_tags, fourcc);
1474         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1475             switch (track->audio.bitdepth) {
1476             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1477             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1478             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1479             }
1480         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1481             switch (track->audio.bitdepth) {
1482             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1483             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1484             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1485             }
1486         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1487             codec_id = CODEC_ID_PCM_F64LE;
1488         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1489             int profile = matroska_aac_profile(track->codec_id);
1490             int sri = matroska_aac_sri(track->audio.samplerate);
1491             extradata = av_malloc(5);
1492             if (extradata == NULL)
1493                 return AVERROR(ENOMEM);
1494             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1495             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1496             if (strstr(track->codec_id, "SBR")) {
1497                 sri = matroska_aac_sri(track->audio.out_samplerate);
1498                 extradata[2] = 0x56;
1499                 extradata[3] = 0xE5;
1500                 extradata[4] = 0x80 | (sri<<3);
1501                 extradata_size = 5;
1502             } else
1503                 extradata_size = 2;
1504         } else if (codec_id == CODEC_ID_TTA) {
1505             extradata_size = 30;
1506             extradata = av_mallocz(extradata_size);
1507             if (extradata == NULL)
1508                 return AVERROR(ENOMEM);
1509             ffio_init_context(&b, extradata, extradata_size, 1,
1510                           NULL, NULL, NULL, NULL);
1511             avio_write(&b, "TTA1", 4);
1512             avio_wl16(&b, 1);
1513             avio_wl16(&b, track->audio.channels);
1514             avio_wl16(&b, track->audio.bitdepth);
1515             avio_wl32(&b, track->audio.out_samplerate);
1516             avio_wl32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1517         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1518                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1519             extradata_offset = 26;
1520         } else if (codec_id == CODEC_ID_RA_144) {
1521             track->audio.out_samplerate = 8000;
1522             track->audio.channels = 1;
1523         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1524                    codec_id == CODEC_ID_ATRAC3 || codec_id == CODEC_ID_SIPR) {
1525             int flavor;
1526             ffio_init_context(&b, track->codec_priv.data,track->codec_priv.size,
1527                           0, NULL, NULL, NULL, NULL);
1528             avio_skip(&b, 22);
1529             flavor                       = avio_rb16(&b);
1530             track->audio.coded_framesize = avio_rb32(&b);
1531             avio_skip(&b, 12);
1532             track->audio.sub_packet_h    = avio_rb16(&b);
1533             track->audio.frame_size      = avio_rb16(&b);
1534             track->audio.sub_packet_size = avio_rb16(&b);
1535             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1536             if (codec_id == CODEC_ID_RA_288) {
1537                 st->codec->block_align = track->audio.coded_framesize;
1538                 track->codec_priv.size = 0;
1539             } else {
1540                 if (codec_id == CODEC_ID_SIPR && flavor < 4) {
1541                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1542                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1543                     st->codec->bit_rate = sipr_bit_rate[flavor];
1544                 }
1545                 st->codec->block_align = track->audio.sub_packet_size;
1546                 extradata_offset = 78;
1547             }
1548         }
1549         track->codec_priv.size -= extradata_offset;
1550
1551         if (codec_id == CODEC_ID_NONE)
1552             av_log(matroska->ctx, AV_LOG_INFO,
1553                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1554
1555         if (track->time_scale < 0.01)
1556             track->time_scale = 1.0;
1557         avpriv_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1558
1559         st->codec->codec_id = codec_id;
1560         st->start_time = 0;
1561         if (strcmp(track->language, "und"))
1562             av_dict_set(&st->metadata, "language", track->language, 0);
1563         av_dict_set(&st->metadata, "title", track->name, 0);
1564
1565         if (track->flag_default)
1566             st->disposition |= AV_DISPOSITION_DEFAULT;
1567         if (track->flag_forced)
1568             st->disposition |= AV_DISPOSITION_FORCED;
1569
1570         if (!st->codec->extradata) {
1571             if(extradata){
1572                 st->codec->extradata = extradata;
1573                 st->codec->extradata_size = extradata_size;
1574             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1575                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1576                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1577                 if(st->codec->extradata == NULL)
1578                     return AVERROR(ENOMEM);
1579                 st->codec->extradata_size = track->codec_priv.size;
1580                 memcpy(st->codec->extradata,
1581                        track->codec_priv.data + extradata_offset,
1582                        track->codec_priv.size);
1583             }
1584         }
1585
1586         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1587             MatroskaTrackPlane *planes = track->operation.combine_planes.elem;
1588
1589             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1590             st->codec->codec_tag  = fourcc;
1591             st->codec->width  = track->video.pixel_width;
1592             st->codec->height = track->video.pixel_height;
1593             av_reduce(&st->sample_aspect_ratio.num,
1594                       &st->sample_aspect_ratio.den,
1595                       st->codec->height * track->video.display_width,
1596                       st->codec-> width * track->video.display_height,
1597                       255);
1598             if (st->codec->codec_id != CODEC_ID_H264)
1599             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1600             if (track->default_duration)
1601                 st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
1602
1603             /* export stereo mode flag as metadata tag */
1604             if (track->video.stereo_mode && track->video.stereo_mode < MATROSKA_VIDEO_STEREO_MODE_COUNT)
1605                 av_dict_set(&st->metadata, "stereo_mode", matroska_video_stereo_mode[track->video.stereo_mode], 0);
1606
1607             /* if we have virtual track, mark the real tracks */
1608             for (j=0; j < track->operation.combine_planes.nb_elem; j++) {
1609                 char buf[32];
1610                 if (planes[j].type >= MATROSKA_VIDEO_STEREO_PLANE_COUNT)
1611                     continue;
1612                 snprintf(buf, sizeof(buf), "%s_%d",
1613                          matroska_video_stereo_plane[planes[j].type], i);
1614                 for (k=0; k < matroska->tracks.nb_elem; k++)
1615                     if (planes[j].uid == tracks[k].uid) {
1616                         av_dict_set(&s->streams[k]->metadata,
1617                                     "stereo_mode", buf, 0);
1618                         break;
1619                     }
1620             }
1621         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1622             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1623             st->codec->sample_rate = track->audio.out_samplerate;
1624             st->codec->channels = track->audio.channels;
1625             if (st->codec->codec_id != CODEC_ID_AAC)
1626             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1627         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1628             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1629         }
1630     }
1631
1632     attachements = attachements_list->elem;
1633     for (j=0; j<attachements_list->nb_elem; j++) {
1634         if (!(attachements[j].filename && attachements[j].mime &&
1635               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1636             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1637         } else {
1638             AVStream *st = avformat_new_stream(s, NULL);
1639             if (st == NULL)
1640                 break;
1641             av_dict_set(&st->metadata, "filename",attachements[j].filename, 0);
1642             av_dict_set(&st->metadata, "mimetype", attachements[j].mime, 0);
1643             st->codec->codec_id = CODEC_ID_NONE;
1644             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1645             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1646             if(st->codec->extradata == NULL)
1647                 break;
1648             st->codec->extradata_size = attachements[j].bin.size;
1649             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1650
1651             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1652                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1653                              strlen(ff_mkv_mime_tags[i].str))) {
1654                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1655                     break;
1656                 }
1657             }
1658             attachements[j].stream = st;
1659         }
1660     }
1661
1662     chapters = chapters_list->elem;
1663     for (i=0; i<chapters_list->nb_elem; i++)
1664         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1665             && (max_start==0 || chapters[i].start > max_start)) {
1666             chapters[i].chapter =
1667             avpriv_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1668                            chapters[i].start, chapters[i].end,
1669                            chapters[i].title);
1670             av_dict_set(&chapters[i].chapter->metadata,
1671                              "title", chapters[i].title, 0);
1672             max_start = chapters[i].start;
1673         }
1674
1675     matroska_add_index_entries(matroska);
1676
1677     matroska_convert_tags(s);
1678
1679     return 0;
1680 }
1681
1682 /*
1683  * Put one packet in an application-supplied AVPacket struct.
1684  * Returns 0 on success or -1 on failure.
1685  */
1686 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1687                                    AVPacket *pkt)
1688 {
1689     if (matroska->num_packets > 0) {
1690         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1691         av_free(matroska->packets[0]);
1692         if (matroska->num_packets > 1) {
1693             void *newpackets;
1694             memmove(&matroska->packets[0], &matroska->packets[1],
1695                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1696             newpackets = av_realloc(matroska->packets,
1697                             (matroska->num_packets - 1) * sizeof(AVPacket *));
1698             if (newpackets)
1699                 matroska->packets = newpackets;
1700         } else {
1701             av_freep(&matroska->packets);
1702         }
1703         matroska->num_packets--;
1704         return 0;
1705     }
1706
1707     return -1;
1708 }
1709
1710 /*
1711  * Free all packets in our internal queue.
1712  */
1713 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1714 {
1715     if (matroska->packets) {
1716         int n;
1717         for (n = 0; n < matroska->num_packets; n++) {
1718             av_free_packet(matroska->packets[n]);
1719             av_free(matroska->packets[n]);
1720         }
1721         av_freep(&matroska->packets);
1722         matroska->num_packets = 0;
1723     }
1724 }
1725
1726 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1727                                 int size, int64_t pos, uint64_t cluster_time,
1728                                 uint64_t duration, int is_keyframe,
1729                                 int64_t cluster_pos)
1730 {
1731     uint64_t timecode = AV_NOPTS_VALUE;
1732     MatroskaTrack *track;
1733     int res = 0;
1734     AVStream *st;
1735     AVPacket *pkt;
1736     int16_t block_time;
1737     uint32_t *lace_size = NULL;
1738     int n, flags, laces = 0;
1739     uint64_t num;
1740
1741     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1742         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1743         return res;
1744     }
1745     data += n;
1746     size -= n;
1747
1748     track = matroska_find_track_by_num(matroska, num);
1749     if (!track || !track->stream) {
1750         av_log(matroska->ctx, AV_LOG_INFO,
1751                "Invalid stream %"PRIu64" or size %u\n", num, size);
1752         return res;
1753     } else if (size <= 3)
1754         return 0;
1755     st = track->stream;
1756     if (st->discard >= AVDISCARD_ALL)
1757         return res;
1758     if (!duration)
1759         duration = track->default_duration / matroska->time_scale;
1760
1761     block_time = AV_RB16(data);
1762     data += 2;
1763     flags = *data++;
1764     size -= 3;
1765     if (is_keyframe == -1)
1766         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1767
1768     if (cluster_time != (uint64_t)-1
1769         && (block_time >= 0 || cluster_time >= -block_time)) {
1770         timecode = cluster_time + block_time;
1771         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1772             && timecode < track->end_timecode)
1773             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1774         if (is_keyframe)
1775             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1776         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1777     }
1778
1779     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1780         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1781             return res;
1782         matroska->skip_to_keyframe = 0;
1783     }
1784
1785     switch ((flags & 0x06) >> 1) {
1786         case 0x0: /* no lacing */
1787             laces = 1;
1788             lace_size = av_mallocz(sizeof(int));
1789             lace_size[0] = size;
1790             break;
1791
1792         case 0x1: /* Xiph lacing */
1793         case 0x2: /* fixed-size lacing */
1794         case 0x3: /* EBML lacing */
1795             assert(size>0); // size <=3 is checked before size-=3 above
1796             laces = (*data) + 1;
1797             data += 1;
1798             size -= 1;
1799             lace_size = av_mallocz(laces * sizeof(int));
1800
1801             switch ((flags & 0x06) >> 1) {
1802                 case 0x1: /* Xiph lacing */ {
1803                     uint8_t temp;
1804                     uint32_t total = 0;
1805                     for (n = 0; res == 0 && n < laces - 1; n++) {
1806                         while (1) {
1807                             if (size == 0) {
1808                                 res = -1;
1809                                 break;
1810                             }
1811                             temp = *data;
1812                             lace_size[n] += temp;
1813                             data += 1;
1814                             size -= 1;
1815                             if (temp != 0xff)
1816                                 break;
1817                         }
1818                         total += lace_size[n];
1819                     }
1820                     lace_size[n] = size - total;
1821                     break;
1822                 }
1823
1824                 case 0x2: /* fixed-size lacing */
1825                     for (n = 0; n < laces; n++)
1826                         lace_size[n] = size / laces;
1827                     break;
1828
1829                 case 0x3: /* EBML lacing */ {
1830                     uint32_t total;
1831                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1832                     if (n < 0) {
1833                         av_log(matroska->ctx, AV_LOG_INFO,
1834                                "EBML block data error\n");
1835                         break;
1836                     }
1837                     data += n;
1838                     size -= n;
1839                     total = lace_size[0] = num;
1840                     for (n = 1; res == 0 && n < laces - 1; n++) {
1841                         int64_t snum;
1842                         int r;
1843                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1844                         if (r < 0) {
1845                             av_log(matroska->ctx, AV_LOG_INFO,
1846                                    "EBML block data error\n");
1847                             break;
1848                         }
1849                         data += r;
1850                         size -= r;
1851                         lace_size[n] = lace_size[n - 1] + snum;
1852                         total += lace_size[n];
1853                     }
1854                     lace_size[laces - 1] = size - total;
1855                     break;
1856                 }
1857             }
1858             break;
1859     }
1860
1861     if (res == 0) {
1862         for (n = 0; n < laces; n++) {
1863             if ((st->codec->codec_id == CODEC_ID_RA_288 ||
1864                  st->codec->codec_id == CODEC_ID_COOK ||
1865                  st->codec->codec_id == CODEC_ID_SIPR ||
1866                  st->codec->codec_id == CODEC_ID_ATRAC3) &&
1867                  st->codec->block_align && track->audio.sub_packet_size) {
1868                 int a = st->codec->block_align;
1869                 int sps = track->audio.sub_packet_size;
1870                 int cfs = track->audio.coded_framesize;
1871                 int h = track->audio.sub_packet_h;
1872                 int y = track->audio.sub_packet_cnt;
1873                 int w = track->audio.frame_size;
1874                 int x;
1875
1876                 if (!track->audio.pkt_cnt) {
1877                     if (track->audio.sub_packet_cnt == 0)
1878                         track->audio.buf_timecode = timecode;
1879                     if (st->codec->codec_id == CODEC_ID_RA_288)
1880                         for (x=0; x<h/2; x++)
1881                             memcpy(track->audio.buf+x*2*w+y*cfs,
1882                                    data+x*cfs, cfs);
1883                     else if (st->codec->codec_id == CODEC_ID_SIPR)
1884                         memcpy(track->audio.buf + y*w, data, w);
1885                     else
1886                         for (x=0; x<w/sps; x++)
1887                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1888
1889                     if (++track->audio.sub_packet_cnt >= h) {
1890                         if (st->codec->codec_id == CODEC_ID_SIPR)
1891                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
1892                         track->audio.sub_packet_cnt = 0;
1893                         track->audio.pkt_cnt = h*w / a;
1894                     }
1895                 }
1896                 while (track->audio.pkt_cnt) {
1897                     pkt = av_mallocz(sizeof(AVPacket));
1898                     av_new_packet(pkt, a);
1899                     memcpy(pkt->data, track->audio.buf
1900                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1901                     pkt->pts = track->audio.buf_timecode;
1902                     track->audio.buf_timecode = AV_NOPTS_VALUE;
1903                     pkt->pos = pos;
1904                     pkt->stream_index = st->index;
1905                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1906                 }
1907             } else {
1908                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1909                 int offset = 0, pkt_size = lace_size[n];
1910                 uint8_t *pkt_data = data;
1911
1912                 if (pkt_size > size) {
1913                     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
1914                     break;
1915                 }
1916
1917                 if (encodings && encodings->scope & 1) {
1918                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1919                     if (offset < 0)
1920                         continue;
1921                 }
1922
1923                 pkt = av_mallocz(sizeof(AVPacket));
1924                 /* XXX: prevent data copy... */
1925                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1926                     av_free(pkt);
1927                     res = AVERROR(ENOMEM);
1928                     break;
1929                 }
1930                 if (offset)
1931                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1932                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1933
1934                 if (pkt_data != data)
1935                     av_free(pkt_data);
1936
1937                 if (n == 0)
1938                     pkt->flags = is_keyframe;
1939                 pkt->stream_index = st->index;
1940
1941                 if (track->ms_compat)
1942                     pkt->dts = timecode;
1943                 else
1944                     pkt->pts = timecode;
1945                 pkt->pos = pos;
1946                 if (st->codec->codec_id == CODEC_ID_TEXT)
1947                     pkt->convergence_duration = duration;
1948                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1949                     pkt->duration = duration;
1950
1951                 if (st->codec->codec_id == CODEC_ID_SSA)
1952                     matroska_fix_ass_packet(matroska, pkt, duration);
1953
1954                 if (matroska->prev_pkt &&
1955                     timecode != AV_NOPTS_VALUE &&
1956                     matroska->prev_pkt->pts == timecode &&
1957                     matroska->prev_pkt->stream_index == st->index &&
1958                     st->codec->codec_id == CODEC_ID_SSA)
1959                     matroska_merge_packets(matroska->prev_pkt, pkt);
1960                 else {
1961                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1962                     matroska->prev_pkt = pkt;
1963                 }
1964             }
1965
1966             if (timecode != AV_NOPTS_VALUE)
1967                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1968             data += lace_size[n];
1969             size -= lace_size[n];
1970         }
1971     }
1972
1973     av_free(lace_size);
1974     return res;
1975 }
1976
1977 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1978 {
1979     MatroskaCluster cluster = { 0 };
1980     EbmlList *blocks_list;
1981     MatroskaBlock *blocks;
1982     int i, res;
1983     int64_t pos = avio_tell(matroska->ctx->pb);
1984     matroska->prev_pkt = NULL;
1985     if (matroska->current_id)
1986         pos -= 4;  /* sizeof the ID which was already read */
1987     res = ebml_parse(matroska, matroska_clusters, &cluster);
1988     blocks_list = &cluster.blocks;
1989     blocks = blocks_list->elem;
1990     for (i=0; i<blocks_list->nb_elem; i++)
1991         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
1992             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
1993             res=matroska_parse_block(matroska,
1994                                      blocks[i].bin.data, blocks[i].bin.size,
1995                                      blocks[i].bin.pos,  cluster.timecode,
1996                                      blocks[i].duration, is_keyframe,
1997                                      pos);
1998         }
1999     ebml_free(matroska_cluster, &cluster);
2000     if (res < 0)  matroska->done = 1;
2001     return res;
2002 }
2003
2004 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
2005 {
2006     MatroskaDemuxContext *matroska = s->priv_data;
2007
2008     while (matroska_deliver_packet(matroska, pkt)) {
2009         if (matroska->done)
2010             return AVERROR_EOF;
2011         matroska_parse_cluster(matroska);
2012     }
2013
2014     return 0;
2015 }
2016
2017 static int matroska_read_seek(AVFormatContext *s, int stream_index,
2018                               int64_t timestamp, int flags)
2019 {
2020     MatroskaDemuxContext *matroska = s->priv_data;
2021     MatroskaTrack *tracks = matroska->tracks.elem;
2022     AVStream *st = s->streams[stream_index];
2023     int i, index, index_sub, index_min;
2024
2025     /* Parse the CUES now since we need the index data to seek. */
2026     if (matroska->cues_parsing_deferred) {
2027         matroska_parse_cues(matroska);
2028         matroska->cues_parsing_deferred = 0;
2029     }
2030
2031     if (!st->nb_index_entries)
2032         return 0;
2033     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
2034
2035     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
2036         avio_seek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
2037         matroska->current_id = 0;
2038         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
2039             matroska_clear_queue(matroska);
2040             if (matroska_parse_cluster(matroska) < 0)
2041                 break;
2042         }
2043     }
2044
2045     matroska_clear_queue(matroska);
2046     if (index < 0)
2047         return 0;
2048
2049     index_min = index;
2050     for (i=0; i < matroska->tracks.nb_elem; i++) {
2051         tracks[i].audio.pkt_cnt = 0;
2052         tracks[i].audio.sub_packet_cnt = 0;
2053         tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
2054         tracks[i].end_timecode = 0;
2055         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
2056             && !tracks[i].stream->discard != AVDISCARD_ALL) {
2057             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
2058             if (index_sub >= 0
2059                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
2060                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
2061                 index_min = index_sub;
2062         }
2063     }
2064
2065     avio_seek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
2066     matroska->current_id = 0;
2067     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
2068     matroska->skip_to_timecode = st->index_entries[index].timestamp;
2069     matroska->done = 0;
2070     ff_update_cur_dts(s, st, st->index_entries[index].timestamp);
2071     return 0;
2072 }
2073
2074 static int matroska_read_close(AVFormatContext *s)
2075 {
2076     MatroskaDemuxContext *matroska = s->priv_data;
2077     MatroskaTrack *tracks = matroska->tracks.elem;
2078     int n;
2079
2080     matroska_clear_queue(matroska);
2081
2082     for (n=0; n < matroska->tracks.nb_elem; n++)
2083         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
2084             av_free(tracks[n].audio.buf);
2085     ebml_free(matroska_segment, matroska);
2086
2087     return 0;
2088 }
2089
2090 AVInputFormat ff_matroska_demuxer = {
2091     .name           = "matroska,webm",
2092     .long_name      = NULL_IF_CONFIG_SMALL("Matroska/WebM file format"),
2093     .priv_data_size = sizeof(MatroskaDemuxContext),
2094     .read_probe     = matroska_probe,
2095     .read_header    = matroska_read_header,
2096     .read_packet    = matroska_read_packet,
2097     .read_close     = matroska_read_close,
2098     .read_seek      = matroska_read_seek,
2099 };