matroskadec: add generic element length validation.
[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     EBML_TYPE_COUNT
62 } EbmlType;
63
64 typedef const struct EbmlSyntax {
65     uint32_t id;
66     EbmlType type;
67     int list_elem_size;
68     int data_offset;
69     union {
70         uint64_t    u;
71         double      f;
72         const char *s;
73         const struct EbmlSyntax *n;
74     } def;
75 } EbmlSyntax;
76
77 typedef struct {
78     int nb_elem;
79     void *elem;
80 } EbmlList;
81
82 typedef struct {
83     int      size;
84     uint8_t *data;
85     int64_t  pos;
86 } EbmlBin;
87
88 typedef struct {
89     uint64_t version;
90     uint64_t max_size;
91     uint64_t id_length;
92     char    *doctype;
93     uint64_t doctype_version;
94 } Ebml;
95
96 typedef struct {
97     uint64_t algo;
98     EbmlBin  settings;
99 } MatroskaTrackCompression;
100
101 typedef struct {
102     uint64_t scope;
103     uint64_t type;
104     MatroskaTrackCompression compression;
105 } MatroskaTrackEncoding;
106
107 typedef struct {
108     double   frame_rate;
109     uint64_t display_width;
110     uint64_t display_height;
111     uint64_t pixel_width;
112     uint64_t pixel_height;
113     uint64_t fourcc;
114 } MatroskaTrackVideo;
115
116 typedef struct {
117     double   samplerate;
118     double   out_samplerate;
119     uint64_t bitdepth;
120     uint64_t channels;
121
122     /* real audio header (extracted from extradata) */
123     int      coded_framesize;
124     int      sub_packet_h;
125     int      frame_size;
126     int      sub_packet_size;
127     int      sub_packet_cnt;
128     int      pkt_cnt;
129     uint8_t *buf;
130 } MatroskaTrackAudio;
131
132 typedef struct {
133     uint64_t num;
134     uint64_t uid;
135     uint64_t type;
136     char    *name;
137     char    *codec_id;
138     EbmlBin  codec_priv;
139     char    *language;
140     double time_scale;
141     uint64_t default_duration;
142     uint64_t flag_default;
143     uint64_t flag_forced;
144     MatroskaTrackVideo video;
145     MatroskaTrackAudio audio;
146     EbmlList encodings;
147
148     AVStream *stream;
149     int64_t end_timecode;
150     int ms_compat;
151 } MatroskaTrack;
152
153 typedef struct {
154     uint64_t uid;
155     char *filename;
156     char *mime;
157     EbmlBin bin;
158
159     AVStream *stream;
160 } MatroskaAttachement;
161
162 typedef struct {
163     uint64_t start;
164     uint64_t end;
165     uint64_t uid;
166     char    *title;
167
168     AVChapter *chapter;
169 } MatroskaChapter;
170
171 typedef struct {
172     uint64_t track;
173     uint64_t pos;
174 } MatroskaIndexPos;
175
176 typedef struct {
177     uint64_t time;
178     EbmlList pos;
179 } MatroskaIndex;
180
181 typedef struct {
182     char *name;
183     char *string;
184     char *lang;
185     uint64_t def;
186     EbmlList sub;
187 } MatroskaTag;
188
189 typedef struct {
190     char    *type;
191     uint64_t typevalue;
192     uint64_t trackuid;
193     uint64_t chapteruid;
194     uint64_t attachuid;
195 } MatroskaTagTarget;
196
197 typedef struct {
198     MatroskaTagTarget target;
199     EbmlList tag;
200 } MatroskaTags;
201
202 typedef struct {
203     uint64_t id;
204     uint64_t pos;
205 } MatroskaSeekhead;
206
207 typedef struct {
208     uint64_t start;
209     uint64_t length;
210 } MatroskaLevel;
211
212 typedef struct {
213     AVFormatContext *ctx;
214
215     /* EBML stuff */
216     int num_levels;
217     MatroskaLevel levels[EBML_MAX_DEPTH];
218     int level_up;
219     uint32_t current_id;
220
221     uint64_t time_scale;
222     double   duration;
223     char    *title;
224     EbmlList tracks;
225     EbmlList attachments;
226     EbmlList chapters;
227     EbmlList index;
228     EbmlList tags;
229     EbmlList seekhead;
230
231     /* byte position of the segment inside the stream */
232     int64_t segment_start;
233
234     /* the packet queue */
235     AVPacket **packets;
236     int num_packets;
237     AVPacket *prev_pkt;
238
239     int done;
240
241     /* What to skip before effectively reading a packet. */
242     int skip_to_keyframe;
243     uint64_t skip_to_timecode;
244 } MatroskaDemuxContext;
245
246 typedef struct {
247     uint64_t duration;
248     int64_t  reference;
249     uint64_t non_simple;
250     EbmlBin  bin;
251 } MatroskaBlock;
252
253 typedef struct {
254     uint64_t timecode;
255     EbmlList blocks;
256 } MatroskaCluster;
257
258 static EbmlSyntax ebml_header[] = {
259     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
260     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
261     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
262     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
263     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
264     { EBML_ID_EBMLVERSION,            EBML_NONE },
265     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
266     { 0 }
267 };
268
269 static EbmlSyntax ebml_syntax[] = {
270     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
271     { 0 }
272 };
273
274 static EbmlSyntax matroska_info[] = {
275     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
276     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
277     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
278     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
279     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
280     { MATROSKA_ID_DATEUTC,            EBML_NONE },
281     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
282     { 0 }
283 };
284
285 static EbmlSyntax matroska_track_video[] = {
286     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
287     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
288     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
289     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
290     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
291     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,fourcc) },
292     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
293     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
294     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
295     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
296     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
297     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
298     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_NONE },
299     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
300     { 0 }
301 };
302
303 static EbmlSyntax matroska_track_audio[] = {
304     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
305     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
306     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
307     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
308     { 0 }
309 };
310
311 static EbmlSyntax matroska_track_encoding_compression[] = {
312     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
313     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
314     { 0 }
315 };
316
317 static EbmlSyntax matroska_track_encoding[] = {
318     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
319     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
320     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
321     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
322     { 0 }
323 };
324
325 static EbmlSyntax matroska_track_encodings[] = {
326     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
327     { 0 }
328 };
329
330 static EbmlSyntax matroska_track[] = {
331     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
332     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
333     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
334     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
335     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
336     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
337     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
338     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
339     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
340     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
341     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_UINT, 0, offsetof(MatroskaTrack,flag_forced), {.u=0} },
342     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
343     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
344     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
345     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
346     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
347     { MATROSKA_ID_CODECNAME,            EBML_NONE },
348     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
349     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
350     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
351     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
352     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
353     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
354     { 0 }
355 };
356
357 static EbmlSyntax matroska_tracks[] = {
358     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
359     { 0 }
360 };
361
362 static EbmlSyntax matroska_attachment[] = {
363     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
364     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
365     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
366     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
367     { MATROSKA_ID_FILEDESC,           EBML_NONE },
368     { 0 }
369 };
370
371 static EbmlSyntax matroska_attachments[] = {
372     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
373     { 0 }
374 };
375
376 static EbmlSyntax matroska_chapter_display[] = {
377     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
378     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
379     { 0 }
380 };
381
382 static EbmlSyntax matroska_chapter_entry[] = {
383     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
384     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
385     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
386     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
387     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
388     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
389     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
390     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
391     { 0 }
392 };
393
394 static EbmlSyntax matroska_chapter[] = {
395     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
396     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
397     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
398     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
399     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
400     { 0 }
401 };
402
403 static EbmlSyntax matroska_chapters[] = {
404     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
405     { 0 }
406 };
407
408 static EbmlSyntax matroska_index_pos[] = {
409     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
410     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
411     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
412     { 0 }
413 };
414
415 static EbmlSyntax matroska_index_entry[] = {
416     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
417     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
418     { 0 }
419 };
420
421 static EbmlSyntax matroska_index[] = {
422     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
423     { 0 }
424 };
425
426 static EbmlSyntax matroska_simpletag[] = {
427     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
428     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
429     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
430     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
431     { MATROSKA_ID_TAGDEFAULT_BUG,     EBML_UINT, 0, offsetof(MatroskaTag,def) },
432     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
433     { 0 }
434 };
435
436 static EbmlSyntax matroska_tagtargets[] = {
437     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
438     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
439     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
440     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
441     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
442     { 0 }
443 };
444
445 static EbmlSyntax matroska_tag[] = {
446     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
447     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
448     { 0 }
449 };
450
451 static EbmlSyntax matroska_tags[] = {
452     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
453     { 0 }
454 };
455
456 static EbmlSyntax matroska_seekhead_entry[] = {
457     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
458     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
459     { 0 }
460 };
461
462 static EbmlSyntax matroska_seekhead[] = {
463     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
464     { 0 }
465 };
466
467 static EbmlSyntax matroska_segment[] = {
468     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
469     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
470     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
471     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
472     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
473     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
474     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
475     { MATROSKA_ID_CLUSTER,        EBML_STOP },
476     { 0 }
477 };
478
479 static EbmlSyntax matroska_segments[] = {
480     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
481     { 0 }
482 };
483
484 static EbmlSyntax matroska_blockgroup[] = {
485     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
486     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
487     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration), {.u=AV_NOPTS_VALUE} },
488     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
489     { 1,                          EBML_UINT, 0, offsetof(MatroskaBlock,non_simple), {.u=1} },
490     { 0 }
491 };
492
493 static EbmlSyntax matroska_cluster[] = {
494     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
495     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
496     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
497     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
498     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
499     { 0 }
500 };
501
502 static EbmlSyntax matroska_clusters[] = {
503     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
504     { MATROSKA_ID_INFO,           EBML_NONE },
505     { MATROSKA_ID_CUES,           EBML_NONE },
506     { MATROSKA_ID_TAGS,           EBML_NONE },
507     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
508     { 0 }
509 };
510
511 static const char *matroska_doctypes[] = { "matroska", "webm" };
512
513 /*
514  * Return: Whether we reached the end of a level in the hierarchy or not.
515  */
516 static int ebml_level_end(MatroskaDemuxContext *matroska)
517 {
518     ByteIOContext *pb = matroska->ctx->pb;
519     int64_t pos = url_ftell(pb);
520
521     if (matroska->num_levels > 0) {
522         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
523         if (pos - level->start >= level->length || matroska->current_id) {
524             matroska->num_levels--;
525             return 1;
526         }
527     }
528     return 0;
529 }
530
531 /*
532  * Read: an "EBML number", which is defined as a variable-length
533  * array of bytes. The first byte indicates the length by giving a
534  * number of 0-bits followed by a one. The position of the first
535  * "one" bit inside the first byte indicates the length of this
536  * number.
537  * Returns: number of bytes read, < 0 on error
538  */
539 static int ebml_read_num(MatroskaDemuxContext *matroska, ByteIOContext *pb,
540                          int max_size, uint64_t *number)
541 {
542     int read = 1, n = 1;
543     uint64_t total = 0;
544
545     /* The first byte tells us the length in bytes - get_byte() can normally
546      * return 0, but since that's not a valid first ebmlID byte, we can
547      * use it safely here to catch EOS. */
548     if (!(total = get_byte(pb))) {
549         /* we might encounter EOS here */
550         if (!url_feof(pb)) {
551             int64_t pos = url_ftell(pb);
552             av_log(matroska->ctx, AV_LOG_ERROR,
553                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
554                    pos, pos);
555         }
556         return AVERROR(EIO); /* EOS or actual I/O error */
557     }
558
559     /* get the length of the EBML number */
560     read = 8 - ff_log2_tab[total];
561     if (read > max_size) {
562         int64_t pos = url_ftell(pb) - 1;
563         av_log(matroska->ctx, AV_LOG_ERROR,
564                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
565                (uint8_t) total, pos, pos);
566         return AVERROR_INVALIDDATA;
567     }
568
569     /* read out length */
570     total ^= 1 << ff_log2_tab[total];
571     while (n++ < read)
572         total = (total << 8) | get_byte(pb);
573
574     *number = total;
575
576     return read;
577 }
578
579 /**
580  * Read a EBML length value.
581  * This needs special handling for the "unknown length" case which has multiple
582  * encodings.
583  */
584 static int ebml_read_length(MatroskaDemuxContext *matroska, ByteIOContext *pb,
585                             uint64_t *number)
586 {
587     int res = ebml_read_num(matroska, pb, 8, number);
588     if (res > 0 && *number + 1 == 1ULL << (7 * res))
589         *number = 0xffffffffffffffULL;
590     return res;
591 }
592
593 /*
594  * Read the next element as an unsigned int.
595  * 0 is success, < 0 is failure.
596  */
597 static int ebml_read_uint(ByteIOContext *pb, int size, uint64_t *num)
598 {
599     int n = 0;
600
601     if (size > 8)
602         return AVERROR_INVALIDDATA;
603
604     /* big-endian ordering; build up number */
605     *num = 0;
606     while (n++ < size)
607         *num = (*num << 8) | get_byte(pb);
608
609     return 0;
610 }
611
612 /*
613  * Read the next element as a float.
614  * 0 is success, < 0 is failure.
615  */
616 static int ebml_read_float(ByteIOContext *pb, int size, double *num)
617 {
618     if (size == 0) {
619         *num = 0;
620     } else if (size == 4) {
621         *num= av_int2flt(get_be32(pb));
622     } else if(size==8){
623         *num= av_int2dbl(get_be64(pb));
624     } else
625         return AVERROR_INVALIDDATA;
626
627     return 0;
628 }
629
630 /*
631  * Read the next element as an ASCII string.
632  * 0 is success, < 0 is failure.
633  */
634 static int ebml_read_ascii(ByteIOContext *pb, int size, char **str)
635 {
636     av_free(*str);
637     /* EBML strings are usually not 0-terminated, so we allocate one
638      * byte more, read the string and NULL-terminate it ourselves. */
639     if (!(*str = av_malloc(size + 1)))
640         return AVERROR(ENOMEM);
641     if (get_buffer(pb, (uint8_t *) *str, size) != size) {
642         av_freep(str);
643         return AVERROR(EIO);
644     }
645     (*str)[size] = '\0';
646
647     return 0;
648 }
649
650 /*
651  * Read the next element as binary data.
652  * 0 is success, < 0 is failure.
653  */
654 static int ebml_read_binary(ByteIOContext *pb, int length, EbmlBin *bin)
655 {
656     av_free(bin->data);
657     if (!(bin->data = av_malloc(length)))
658         return AVERROR(ENOMEM);
659
660     bin->size = length;
661     bin->pos  = url_ftell(pb);
662     if (get_buffer(pb, bin->data, length) != length) {
663         av_freep(&bin->data);
664         return AVERROR(EIO);
665     }
666
667     return 0;
668 }
669
670 /*
671  * Read the next element, but only the header. The contents
672  * are supposed to be sub-elements which can be read separately.
673  * 0 is success, < 0 is failure.
674  */
675 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
676 {
677     ByteIOContext *pb = matroska->ctx->pb;
678     MatroskaLevel *level;
679
680     if (matroska->num_levels >= EBML_MAX_DEPTH) {
681         av_log(matroska->ctx, AV_LOG_ERROR,
682                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
683         return AVERROR(ENOSYS);
684     }
685
686     level = &matroska->levels[matroska->num_levels++];
687     level->start = url_ftell(pb);
688     level->length = length;
689
690     return 0;
691 }
692
693 /*
694  * Read signed/unsigned "EBML" numbers.
695  * Return: number of bytes processed, < 0 on error
696  */
697 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
698                                  uint8_t *data, uint32_t size, uint64_t *num)
699 {
700     ByteIOContext pb;
701     init_put_byte(&pb, data, size, 0, NULL, NULL, NULL, NULL);
702     return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
703 }
704
705 /*
706  * Same as above, but signed.
707  */
708 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
709                                  uint8_t *data, uint32_t size, int64_t *num)
710 {
711     uint64_t unum;
712     int res;
713
714     /* read as unsigned number first */
715     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
716         return res;
717
718     /* make signed (weird way) */
719     *num = unum - ((1LL << (7*res - 1)) - 1);
720
721     return res;
722 }
723
724 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
725                            EbmlSyntax *syntax, void *data);
726
727 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
728                          uint32_t id, void *data)
729 {
730     int i;
731     for (i=0; syntax[i].id; i++)
732         if (id == syntax[i].id)
733             break;
734     if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
735         matroska->num_levels > 0 &&
736         matroska->levels[matroska->num_levels-1].length == 0xffffffffffffff)
737         return 0;  // we reached the end of an unknown size cluster
738     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32)
739         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
740     return ebml_parse_elem(matroska, &syntax[i], data);
741 }
742
743 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
744                       void *data)
745 {
746     if (!matroska->current_id) {
747         uint64_t id;
748         int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
749         if (res < 0)
750             return res;
751         matroska->current_id = id | 1 << 7*res;
752     }
753     return ebml_parse_id(matroska, syntax, matroska->current_id, data);
754 }
755
756 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
757                            void *data)
758 {
759     int i, res = 0;
760
761     for (i=0; syntax[i].id; i++)
762         switch (syntax[i].type) {
763         case EBML_UINT:
764             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
765             break;
766         case EBML_FLOAT:
767             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
768             break;
769         case EBML_STR:
770         case EBML_UTF8:
771             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
772             break;
773         }
774
775     while (!res && !ebml_level_end(matroska))
776         res = ebml_parse(matroska, syntax, data);
777
778     return res;
779 }
780
781 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
782                            EbmlSyntax *syntax, void *data)
783 {
784     static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
785         [EBML_UINT]  = 8,
786         [EBML_FLOAT] = 8,
787         // max. 16 MB for strings
788         [EBML_STR]   = 0x1000000,
789         [EBML_UTF8]  = 0x1000000,
790         // max. 256 MB for binary data
791         [EBML_BIN]   = 0x10000000,
792         // no limits for anything else
793     };
794     ByteIOContext *pb = matroska->ctx->pb;
795     uint32_t id = syntax->id;
796     uint64_t length;
797     int res;
798
799     data = (char *)data + syntax->data_offset;
800     if (syntax->list_elem_size) {
801         EbmlList *list = data;
802         list->elem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
803         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
804         memset(data, 0, syntax->list_elem_size);
805         list->nb_elem++;
806     }
807
808     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
809         matroska->current_id = 0;
810         if ((res = ebml_read_length(matroska, pb, &length)) < 0)
811             return res;
812         if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
813             av_log(matroska->ctx, AV_LOG_ERROR,
814                    "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for syntax element %i\n",
815                    length, max_lengths[syntax->type], syntax->type);
816             return AVERROR_INVALIDDATA;
817         }
818     }
819
820     switch (syntax->type) {
821     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
822     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
823     case EBML_STR:
824     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
825     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
826     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
827                          return res;
828                      if (id == MATROSKA_ID_SEGMENT)
829                          matroska->segment_start = url_ftell(matroska->ctx->pb);
830                      return ebml_parse_nest(matroska, syntax->def.n, data);
831     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
832     case EBML_STOP:  return 1;
833     default:         return url_fseek(pb,length,SEEK_CUR)<0 ? AVERROR(EIO) : 0;
834     }
835     if (res == AVERROR_INVALIDDATA)
836         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
837     else if (res == AVERROR(EIO))
838         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
839     return res;
840 }
841
842 static void ebml_free(EbmlSyntax *syntax, void *data)
843 {
844     int i, j;
845     for (i=0; syntax[i].id; i++) {
846         void *data_off = (char *)data + syntax[i].data_offset;
847         switch (syntax[i].type) {
848         case EBML_STR:
849         case EBML_UTF8:  av_freep(data_off);                      break;
850         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
851         case EBML_NEST:
852             if (syntax[i].list_elem_size) {
853                 EbmlList *list = data_off;
854                 char *ptr = list->elem;
855                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
856                     ebml_free(syntax[i].def.n, ptr);
857                 av_free(list->elem);
858             } else
859                 ebml_free(syntax[i].def.n, data_off);
860         default:  break;
861         }
862     }
863 }
864
865
866 /*
867  * Autodetecting...
868  */
869 static int matroska_probe(AVProbeData *p)
870 {
871     uint64_t total = 0;
872     int len_mask = 0x80, size = 1, n = 1, i;
873
874     /* EBML header? */
875     if (AV_RB32(p->buf) != EBML_ID_HEADER)
876         return 0;
877
878     /* length of header */
879     total = p->buf[4];
880     while (size <= 8 && !(total & len_mask)) {
881         size++;
882         len_mask >>= 1;
883     }
884     if (size > 8)
885       return 0;
886     total &= (len_mask - 1);
887     while (n < size)
888         total = (total << 8) | p->buf[4 + n++];
889
890     /* Does the probe data contain the whole header? */
891     if (p->buf_size < 4 + size + total)
892       return 0;
893
894     /* The header should contain a known document type. For now,
895      * we don't parse the whole header but simply check for the
896      * availability of that array of characters inside the header.
897      * Not fully fool-proof, but good enough. */
898     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
899         int probelen = strlen(matroska_doctypes[i]);
900         for (n = 4+size; n <= 4+size+total-probelen; n++)
901             if (!memcmp(p->buf+n, matroska_doctypes[i], probelen))
902                 return AVPROBE_SCORE_MAX;
903     }
904
905     // probably valid EBML header but no recognized doctype
906     return AVPROBE_SCORE_MAX/2;
907 }
908
909 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
910                                                  int num)
911 {
912     MatroskaTrack *tracks = matroska->tracks.elem;
913     int i;
914
915     for (i=0; i < matroska->tracks.nb_elem; i++)
916         if (tracks[i].num == num)
917             return &tracks[i];
918
919     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
920     return NULL;
921 }
922
923 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
924                                   MatroskaTrack *track)
925 {
926     MatroskaTrackEncoding *encodings = track->encodings.elem;
927     uint8_t* data = *buf;
928     int isize = *buf_size;
929     uint8_t* pkt_data = NULL;
930     int pkt_size = isize;
931     int result = 0;
932     int olen;
933
934     if (pkt_size >= 10000000)
935         return -1;
936
937     switch (encodings[0].compression.algo) {
938     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
939         return encodings[0].compression.settings.size;
940     case MATROSKA_TRACK_ENCODING_COMP_LZO:
941         do {
942             olen = pkt_size *= 3;
943             pkt_data = av_realloc(pkt_data, pkt_size+AV_LZO_OUTPUT_PADDING);
944             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
945         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
946         if (result)
947             goto failed;
948         pkt_size -= olen;
949         break;
950 #if CONFIG_ZLIB
951     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
952         z_stream zstream = {0};
953         if (inflateInit(&zstream) != Z_OK)
954             return -1;
955         zstream.next_in = data;
956         zstream.avail_in = isize;
957         do {
958             pkt_size *= 3;
959             pkt_data = av_realloc(pkt_data, pkt_size);
960             zstream.avail_out = pkt_size - zstream.total_out;
961             zstream.next_out = pkt_data + zstream.total_out;
962             result = inflate(&zstream, Z_NO_FLUSH);
963         } while (result==Z_OK && pkt_size<10000000);
964         pkt_size = zstream.total_out;
965         inflateEnd(&zstream);
966         if (result != Z_STREAM_END)
967             goto failed;
968         break;
969     }
970 #endif
971 #if CONFIG_BZLIB
972     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
973         bz_stream bzstream = {0};
974         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
975             return -1;
976         bzstream.next_in = data;
977         bzstream.avail_in = isize;
978         do {
979             pkt_size *= 3;
980             pkt_data = av_realloc(pkt_data, pkt_size);
981             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
982             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
983             result = BZ2_bzDecompress(&bzstream);
984         } while (result==BZ_OK && pkt_size<10000000);
985         pkt_size = bzstream.total_out_lo32;
986         BZ2_bzDecompressEnd(&bzstream);
987         if (result != BZ_STREAM_END)
988             goto failed;
989         break;
990     }
991 #endif
992     default:
993         return -1;
994     }
995
996     *buf = pkt_data;
997     *buf_size = pkt_size;
998     return 0;
999  failed:
1000     av_free(pkt_data);
1001     return -1;
1002 }
1003
1004 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
1005                                     AVPacket *pkt, uint64_t display_duration)
1006 {
1007     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
1008     for (; *ptr!=',' && ptr<end-1; ptr++);
1009     if (*ptr == ',')
1010         layer = ++ptr;
1011     for (; *ptr!=',' && ptr<end-1; ptr++);
1012     if (*ptr == ',') {
1013         int64_t end_pts = pkt->pts + display_duration;
1014         int sc = matroska->time_scale * pkt->pts / 10000000;
1015         int ec = matroska->time_scale * end_pts  / 10000000;
1016         int sh, sm, ss, eh, em, es, len;
1017         sh = sc/360000;  sc -= 360000*sh;
1018         sm = sc/  6000;  sc -=   6000*sm;
1019         ss = sc/   100;  sc -=    100*ss;
1020         eh = ec/360000;  ec -= 360000*eh;
1021         em = ec/  6000;  ec -=   6000*em;
1022         es = ec/   100;  ec -=    100*es;
1023         *ptr++ = '\0';
1024         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
1025         if (!(line = av_malloc(len)))
1026             return;
1027         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
1028                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
1029         av_free(pkt->data);
1030         pkt->data = line;
1031         pkt->size = strlen(line);
1032     }
1033 }
1034
1035 static void matroska_merge_packets(AVPacket *out, AVPacket *in)
1036 {
1037     out->data = av_realloc(out->data, out->size+in->size);
1038     memcpy(out->data+out->size, in->data, in->size);
1039     out->size += in->size;
1040     av_destruct_packet(in);
1041     av_free(in);
1042 }
1043
1044 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1045                                  AVMetadata **metadata, char *prefix)
1046 {
1047     MatroskaTag *tags = list->elem;
1048     char key[1024];
1049     int i;
1050
1051     for (i=0; i < list->nb_elem; i++) {
1052         const char *lang = strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1053
1054         if (!tags[i].name) {
1055             av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1056             continue;
1057         }
1058         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1059         else         av_strlcpy(key, tags[i].name, sizeof(key));
1060         if (tags[i].def || !lang) {
1061         av_metadata_set2(metadata, key, tags[i].string, 0);
1062         if (tags[i].sub.nb_elem)
1063             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1064         }
1065         if (lang) {
1066             av_strlcat(key, "-", sizeof(key));
1067             av_strlcat(key, lang, sizeof(key));
1068             av_metadata_set2(metadata, key, tags[i].string, 0);
1069             if (tags[i].sub.nb_elem)
1070                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1071         }
1072     }
1073     ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1074 }
1075
1076 static void matroska_convert_tags(AVFormatContext *s)
1077 {
1078     MatroskaDemuxContext *matroska = s->priv_data;
1079     MatroskaTags *tags = matroska->tags.elem;
1080     int i, j;
1081
1082     for (i=0; i < matroska->tags.nb_elem; i++) {
1083         if (tags[i].target.attachuid) {
1084             MatroskaAttachement *attachment = matroska->attachments.elem;
1085             for (j=0; j<matroska->attachments.nb_elem; j++)
1086                 if (attachment[j].uid == tags[i].target.attachuid)
1087                     matroska_convert_tag(s, &tags[i].tag,
1088                                          &attachment[j].stream->metadata, NULL);
1089         } else if (tags[i].target.chapteruid) {
1090             MatroskaChapter *chapter = matroska->chapters.elem;
1091             for (j=0; j<matroska->chapters.nb_elem; j++)
1092                 if (chapter[j].uid == tags[i].target.chapteruid)
1093                     matroska_convert_tag(s, &tags[i].tag,
1094                                          &chapter[j].chapter->metadata, NULL);
1095         } else if (tags[i].target.trackuid) {
1096             MatroskaTrack *track = matroska->tracks.elem;
1097             for (j=0; j<matroska->tracks.nb_elem; j++)
1098                 if (track[j].uid == tags[i].target.trackuid)
1099                     matroska_convert_tag(s, &tags[i].tag,
1100                                          &track[j].stream->metadata, NULL);
1101         } else {
1102             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1103                                  tags[i].target.type);
1104         }
1105     }
1106 }
1107
1108 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1109 {
1110     EbmlList *seekhead_list = &matroska->seekhead;
1111     MatroskaSeekhead *seekhead = seekhead_list->elem;
1112     uint32_t level_up = matroska->level_up;
1113     int64_t before_pos = url_ftell(matroska->ctx->pb);
1114     uint32_t saved_id = matroska->current_id;
1115     MatroskaLevel level;
1116     int i;
1117
1118     // we should not do any seeking in the streaming case
1119     if (url_is_streamed(matroska->ctx->pb) ||
1120         (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1121         return;
1122
1123     for (i=0; i<seekhead_list->nb_elem; i++) {
1124         int64_t offset = seekhead[i].pos + matroska->segment_start;
1125
1126         if (seekhead[i].pos <= before_pos
1127             || seekhead[i].id == MATROSKA_ID_SEEKHEAD
1128             || seekhead[i].id == MATROSKA_ID_CLUSTER)
1129             continue;
1130
1131         /* seek */
1132         if (url_fseek(matroska->ctx->pb, offset, SEEK_SET) != offset)
1133             continue;
1134
1135         /* We don't want to lose our seekhead level, so we add
1136          * a dummy. This is a crude hack. */
1137         if (matroska->num_levels == EBML_MAX_DEPTH) {
1138             av_log(matroska->ctx, AV_LOG_INFO,
1139                    "Max EBML element depth (%d) reached, "
1140                    "cannot parse further.\n", EBML_MAX_DEPTH);
1141             break;
1142         }
1143
1144         level.start = 0;
1145         level.length = (uint64_t)-1;
1146         matroska->levels[matroska->num_levels] = level;
1147         matroska->num_levels++;
1148         matroska->current_id = 0;
1149
1150         ebml_parse(matroska, matroska_segment, matroska);
1151
1152         /* remove dummy level */
1153         while (matroska->num_levels) {
1154             uint64_t length = matroska->levels[--matroska->num_levels].length;
1155             if (length == (uint64_t)-1)
1156                 break;
1157         }
1158     }
1159
1160     /* seek back */
1161     url_fseek(matroska->ctx->pb, before_pos, SEEK_SET);
1162     matroska->level_up = level_up;
1163     matroska->current_id = saved_id;
1164 }
1165
1166 static int matroska_aac_profile(char *codec_id)
1167 {
1168     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1169     int profile;
1170
1171     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1172         if (strstr(codec_id, aac_profiles[profile]))
1173             break;
1174     return profile + 1;
1175 }
1176
1177 static int matroska_aac_sri(int samplerate)
1178 {
1179     int sri;
1180
1181     for (sri=0; sri<FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
1182         if (ff_mpeg4audio_sample_rates[sri] == samplerate)
1183             break;
1184     return sri;
1185 }
1186
1187 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
1188 {
1189     MatroskaDemuxContext *matroska = s->priv_data;
1190     EbmlList *attachements_list = &matroska->attachments;
1191     MatroskaAttachement *attachements;
1192     EbmlList *chapters_list = &matroska->chapters;
1193     MatroskaChapter *chapters;
1194     MatroskaTrack *tracks;
1195     EbmlList *index_list;
1196     MatroskaIndex *index;
1197     int index_scale = 1;
1198     uint64_t max_start = 0;
1199     Ebml ebml = { 0 };
1200     AVStream *st;
1201     int i, j, res;
1202
1203     matroska->ctx = s;
1204
1205     /* First read the EBML header. */
1206     if (ebml_parse(matroska, ebml_syntax, &ebml)
1207         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1208         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 2) {
1209         av_log(matroska->ctx, AV_LOG_ERROR,
1210                "EBML header using unsupported features\n"
1211                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1212                ebml.version, ebml.doctype, ebml.doctype_version);
1213         ebml_free(ebml_syntax, &ebml);
1214         return AVERROR_PATCHWELCOME;
1215     }
1216     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1217         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1218             break;
1219     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1220         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1221     }
1222     ebml_free(ebml_syntax, &ebml);
1223
1224     /* The next thing is a segment. */
1225     if ((res = ebml_parse(matroska, matroska_segments, matroska)) < 0)
1226         return res;
1227     matroska_execute_seekhead(matroska);
1228
1229     if (!matroska->time_scale)
1230         matroska->time_scale = 1000000;
1231     if (matroska->duration)
1232         matroska->ctx->duration = matroska->duration * matroska->time_scale
1233                                   * 1000 / AV_TIME_BASE;
1234     av_metadata_set2(&s->metadata, "title", matroska->title, 0);
1235
1236     tracks = matroska->tracks.elem;
1237     for (i=0; i < matroska->tracks.nb_elem; i++) {
1238         MatroskaTrack *track = &tracks[i];
1239         enum CodecID codec_id = CODEC_ID_NONE;
1240         EbmlList *encodings_list = &tracks->encodings;
1241         MatroskaTrackEncoding *encodings = encodings_list->elem;
1242         uint8_t *extradata = NULL;
1243         int extradata_size = 0;
1244         int extradata_offset = 0;
1245         ByteIOContext b;
1246
1247         /* Apply some sanity checks. */
1248         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1249             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1250             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1251             av_log(matroska->ctx, AV_LOG_INFO,
1252                    "Unknown or unsupported track type %"PRIu64"\n",
1253                    track->type);
1254             continue;
1255         }
1256         if (track->codec_id == NULL)
1257             continue;
1258
1259         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1260             if (!track->default_duration)
1261                 track->default_duration = 1000000000/track->video.frame_rate;
1262             if (!track->video.display_width)
1263                 track->video.display_width = track->video.pixel_width;
1264             if (!track->video.display_height)
1265                 track->video.display_height = track->video.pixel_height;
1266         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1267             if (!track->audio.out_samplerate)
1268                 track->audio.out_samplerate = track->audio.samplerate;
1269         }
1270         if (encodings_list->nb_elem > 1) {
1271             av_log(matroska->ctx, AV_LOG_ERROR,
1272                    "Multiple combined encodings no supported");
1273         } else if (encodings_list->nb_elem == 1) {
1274             if (encodings[0].type ||
1275                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1276 #if CONFIG_ZLIB
1277                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1278 #endif
1279 #if CONFIG_BZLIB
1280                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1281 #endif
1282                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1283                 encodings[0].scope = 0;
1284                 av_log(matroska->ctx, AV_LOG_ERROR,
1285                        "Unsupported encoding type");
1286             } else if (track->codec_priv.size && encodings[0].scope&2) {
1287                 uint8_t *codec_priv = track->codec_priv.data;
1288                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1289                                                     &track->codec_priv.size,
1290                                                     track);
1291                 if (offset < 0) {
1292                     track->codec_priv.data = NULL;
1293                     track->codec_priv.size = 0;
1294                     av_log(matroska->ctx, AV_LOG_ERROR,
1295                            "Failed to decode codec private data\n");
1296                 } else if (offset > 0) {
1297                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1298                     memcpy(track->codec_priv.data,
1299                            encodings[0].compression.settings.data, offset);
1300                     memcpy(track->codec_priv.data+offset, codec_priv,
1301                            track->codec_priv.size);
1302                     track->codec_priv.size += offset;
1303                 }
1304                 if (codec_priv != track->codec_priv.data)
1305                     av_free(codec_priv);
1306             }
1307         }
1308
1309         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1310             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1311                         strlen(ff_mkv_codec_tags[j].str))){
1312                 codec_id= ff_mkv_codec_tags[j].id;
1313                 break;
1314             }
1315         }
1316
1317         st = track->stream = av_new_stream(s, 0);
1318         if (st == NULL)
1319             return AVERROR(ENOMEM);
1320
1321         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1322             && track->codec_priv.size >= 40
1323             && track->codec_priv.data != NULL) {
1324             track->ms_compat = 1;
1325             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1326             codec_id = ff_codec_get_id(ff_codec_bmp_tags, track->video.fourcc);
1327             extradata_offset = 40;
1328         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1329                    && track->codec_priv.size >= 14
1330                    && track->codec_priv.data != NULL) {
1331             init_put_byte(&b, track->codec_priv.data, track->codec_priv.size,
1332                           URL_RDONLY, NULL, NULL, NULL, NULL);
1333             ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1334             codec_id = st->codec->codec_id;
1335             extradata_offset = FFMIN(track->codec_priv.size, 18);
1336         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1337                    && (track->codec_priv.size >= 86)
1338                    && (track->codec_priv.data != NULL)) {
1339             track->video.fourcc = AV_RL32(track->codec_priv.data);
1340             codec_id=ff_codec_get_id(codec_movvideo_tags, track->video.fourcc);
1341         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1342             switch (track->audio.bitdepth) {
1343             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1344             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1345             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1346             }
1347         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1348             switch (track->audio.bitdepth) {
1349             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1350             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1351             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1352             }
1353         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1354             codec_id = CODEC_ID_PCM_F64LE;
1355         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1356             int profile = matroska_aac_profile(track->codec_id);
1357             int sri = matroska_aac_sri(track->audio.samplerate);
1358             extradata = av_malloc(5);
1359             if (extradata == NULL)
1360                 return AVERROR(ENOMEM);
1361             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1362             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1363             if (strstr(track->codec_id, "SBR")) {
1364                 sri = matroska_aac_sri(track->audio.out_samplerate);
1365                 extradata[2] = 0x56;
1366                 extradata[3] = 0xE5;
1367                 extradata[4] = 0x80 | (sri<<3);
1368                 extradata_size = 5;
1369             } else
1370                 extradata_size = 2;
1371         } else if (codec_id == CODEC_ID_TTA) {
1372             extradata_size = 30;
1373             extradata = av_mallocz(extradata_size);
1374             if (extradata == NULL)
1375                 return AVERROR(ENOMEM);
1376             init_put_byte(&b, extradata, extradata_size, 1,
1377                           NULL, NULL, NULL, NULL);
1378             put_buffer(&b, "TTA1", 4);
1379             put_le16(&b, 1);
1380             put_le16(&b, track->audio.channels);
1381             put_le16(&b, track->audio.bitdepth);
1382             put_le32(&b, track->audio.out_samplerate);
1383             put_le32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1384         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1385                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1386             extradata_offset = 26;
1387         } else if (codec_id == CODEC_ID_RA_144) {
1388             track->audio.out_samplerate = 8000;
1389             track->audio.channels = 1;
1390         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1391                    codec_id == CODEC_ID_ATRAC3 || codec_id == CODEC_ID_SIPR) {
1392             int flavor;
1393             init_put_byte(&b, track->codec_priv.data,track->codec_priv.size,
1394                           0, NULL, NULL, NULL, NULL);
1395             url_fskip(&b, 22);
1396             flavor                       = get_be16(&b);
1397             track->audio.coded_framesize = get_be32(&b);
1398             url_fskip(&b, 12);
1399             track->audio.sub_packet_h    = get_be16(&b);
1400             track->audio.frame_size      = get_be16(&b);
1401             track->audio.sub_packet_size = get_be16(&b);
1402             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1403             if (codec_id == CODEC_ID_RA_288) {
1404                 st->codec->block_align = track->audio.coded_framesize;
1405                 track->codec_priv.size = 0;
1406             } else {
1407                 if (codec_id == CODEC_ID_SIPR && flavor < 4) {
1408                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1409                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1410                     st->codec->bit_rate = sipr_bit_rate[flavor];
1411                 }
1412                 st->codec->block_align = track->audio.sub_packet_size;
1413                 extradata_offset = 78;
1414             }
1415         }
1416         track->codec_priv.size -= extradata_offset;
1417
1418         if (codec_id == CODEC_ID_NONE)
1419             av_log(matroska->ctx, AV_LOG_INFO,
1420                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1421
1422         if (track->time_scale < 0.01)
1423             track->time_scale = 1.0;
1424         av_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1425
1426         st->codec->codec_id = codec_id;
1427         st->start_time = 0;
1428         if (strcmp(track->language, "und"))
1429             av_metadata_set2(&st->metadata, "language", track->language, 0);
1430         av_metadata_set2(&st->metadata, "title", track->name, 0);
1431
1432         if (track->flag_default)
1433             st->disposition |= AV_DISPOSITION_DEFAULT;
1434         if (track->flag_forced)
1435             st->disposition |= AV_DISPOSITION_FORCED;
1436
1437         if (track->default_duration)
1438             av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
1439                       track->default_duration, 1000000000, 30000);
1440
1441         if (!st->codec->extradata) {
1442             if(extradata){
1443                 st->codec->extradata = extradata;
1444                 st->codec->extradata_size = extradata_size;
1445             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1446                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1447                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1448                 if(st->codec->extradata == NULL)
1449                     return AVERROR(ENOMEM);
1450                 st->codec->extradata_size = track->codec_priv.size;
1451                 memcpy(st->codec->extradata,
1452                        track->codec_priv.data + extradata_offset,
1453                        track->codec_priv.size);
1454             }
1455         }
1456
1457         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1458             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1459             st->codec->codec_tag  = track->video.fourcc;
1460             st->codec->width  = track->video.pixel_width;
1461             st->codec->height = track->video.pixel_height;
1462             av_reduce(&st->sample_aspect_ratio.num,
1463                       &st->sample_aspect_ratio.den,
1464                       st->codec->height * track->video.display_width,
1465                       st->codec-> width * track->video.display_height,
1466                       255);
1467             if (st->codec->codec_id != CODEC_ID_H264)
1468             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1469             if (track->default_duration)
1470                 st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
1471         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1472             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1473             st->codec->sample_rate = track->audio.out_samplerate;
1474             st->codec->channels = track->audio.channels;
1475             if (st->codec->codec_id != CODEC_ID_AAC)
1476             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1477         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1478             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1479         }
1480     }
1481
1482     attachements = attachements_list->elem;
1483     for (j=0; j<attachements_list->nb_elem; j++) {
1484         if (!(attachements[j].filename && attachements[j].mime &&
1485               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1486             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1487         } else {
1488             AVStream *st = av_new_stream(s, 0);
1489             if (st == NULL)
1490                 break;
1491             av_metadata_set2(&st->metadata, "filename",attachements[j].filename, 0);
1492             st->codec->codec_id = CODEC_ID_NONE;
1493             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1494             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1495             if(st->codec->extradata == NULL)
1496                 break;
1497             st->codec->extradata_size = attachements[j].bin.size;
1498             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1499
1500             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1501                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1502                              strlen(ff_mkv_mime_tags[i].str))) {
1503                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1504                     break;
1505                 }
1506             }
1507             attachements[j].stream = st;
1508         }
1509     }
1510
1511     chapters = chapters_list->elem;
1512     for (i=0; i<chapters_list->nb_elem; i++)
1513         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1514             && (max_start==0 || chapters[i].start > max_start)) {
1515             chapters[i].chapter =
1516             ff_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1517                            chapters[i].start, chapters[i].end,
1518                            chapters[i].title);
1519             av_metadata_set2(&chapters[i].chapter->metadata,
1520                              "title", chapters[i].title, 0);
1521             max_start = chapters[i].start;
1522         }
1523
1524     index_list = &matroska->index;
1525     index = index_list->elem;
1526     if (index_list->nb_elem
1527         && index[0].time > 100000000000000/matroska->time_scale) {
1528         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1529         index_scale = matroska->time_scale;
1530     }
1531     for (i=0; i<index_list->nb_elem; i++) {
1532         EbmlList *pos_list = &index[i].pos;
1533         MatroskaIndexPos *pos = pos_list->elem;
1534         for (j=0; j<pos_list->nb_elem; j++) {
1535             MatroskaTrack *track = matroska_find_track_by_num(matroska,
1536                                                               pos[j].track);
1537             if (track && track->stream)
1538                 av_add_index_entry(track->stream,
1539                                    pos[j].pos + matroska->segment_start,
1540                                    index[i].time/index_scale, 0, 0,
1541                                    AVINDEX_KEYFRAME);
1542         }
1543     }
1544
1545     matroska_convert_tags(s);
1546
1547     return 0;
1548 }
1549
1550 /*
1551  * Put one packet in an application-supplied AVPacket struct.
1552  * Returns 0 on success or -1 on failure.
1553  */
1554 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1555                                    AVPacket *pkt)
1556 {
1557     if (matroska->num_packets > 0) {
1558         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1559         av_free(matroska->packets[0]);
1560         if (matroska->num_packets > 1) {
1561             memmove(&matroska->packets[0], &matroska->packets[1],
1562                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1563             matroska->packets =
1564                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
1565                            sizeof(AVPacket *));
1566         } else {
1567             av_freep(&matroska->packets);
1568         }
1569         matroska->num_packets--;
1570         return 0;
1571     }
1572
1573     return -1;
1574 }
1575
1576 /*
1577  * Free all packets in our internal queue.
1578  */
1579 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1580 {
1581     if (matroska->packets) {
1582         int n;
1583         for (n = 0; n < matroska->num_packets; n++) {
1584             av_free_packet(matroska->packets[n]);
1585             av_free(matroska->packets[n]);
1586         }
1587         av_freep(&matroska->packets);
1588         matroska->num_packets = 0;
1589     }
1590 }
1591
1592 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1593                                 int size, int64_t pos, uint64_t cluster_time,
1594                                 uint64_t duration, int is_keyframe,
1595                                 int64_t cluster_pos)
1596 {
1597     uint64_t timecode = AV_NOPTS_VALUE;
1598     MatroskaTrack *track;
1599     int res = 0;
1600     AVStream *st;
1601     AVPacket *pkt;
1602     int16_t block_time;
1603     uint32_t *lace_size = NULL;
1604     int n, flags, laces = 0;
1605     uint64_t num;
1606
1607     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1608         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1609         return res;
1610     }
1611     data += n;
1612     size -= n;
1613
1614     track = matroska_find_track_by_num(matroska, num);
1615     if (size <= 3 || !track || !track->stream) {
1616         av_log(matroska->ctx, AV_LOG_INFO,
1617                "Invalid stream %"PRIu64" or size %u\n", num, size);
1618         return res;
1619     }
1620     st = track->stream;
1621     if (st->discard >= AVDISCARD_ALL)
1622         return res;
1623     if (duration == AV_NOPTS_VALUE)
1624         duration = track->default_duration / matroska->time_scale;
1625
1626     block_time = AV_RB16(data);
1627     data += 2;
1628     flags = *data++;
1629     size -= 3;
1630     if (is_keyframe == -1)
1631         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1632
1633     if (cluster_time != (uint64_t)-1
1634         && (block_time >= 0 || cluster_time >= -block_time)) {
1635         timecode = cluster_time + block_time;
1636         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1637             && timecode < track->end_timecode)
1638             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1639         if (is_keyframe)
1640             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1641         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1642     }
1643
1644     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1645         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1646             return res;
1647         matroska->skip_to_keyframe = 0;
1648     }
1649
1650     switch ((flags & 0x06) >> 1) {
1651         case 0x0: /* no lacing */
1652             laces = 1;
1653             lace_size = av_mallocz(sizeof(int));
1654             lace_size[0] = size;
1655             break;
1656
1657         case 0x1: /* Xiph lacing */
1658         case 0x2: /* fixed-size lacing */
1659         case 0x3: /* EBML lacing */
1660             assert(size>0); // size <=3 is checked before size-=3 above
1661             laces = (*data) + 1;
1662             data += 1;
1663             size -= 1;
1664             lace_size = av_mallocz(laces * sizeof(int));
1665
1666             switch ((flags & 0x06) >> 1) {
1667                 case 0x1: /* Xiph lacing */ {
1668                     uint8_t temp;
1669                     uint32_t total = 0;
1670                     for (n = 0; res == 0 && n < laces - 1; n++) {
1671                         while (1) {
1672                             if (size == 0) {
1673                                 res = -1;
1674                                 break;
1675                             }
1676                             temp = *data;
1677                             lace_size[n] += temp;
1678                             data += 1;
1679                             size -= 1;
1680                             if (temp != 0xff)
1681                                 break;
1682                         }
1683                         total += lace_size[n];
1684                     }
1685                     lace_size[n] = size - total;
1686                     break;
1687                 }
1688
1689                 case 0x2: /* fixed-size lacing */
1690                     for (n = 0; n < laces; n++)
1691                         lace_size[n] = size / laces;
1692                     break;
1693
1694                 case 0x3: /* EBML lacing */ {
1695                     uint32_t total;
1696                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1697                     if (n < 0) {
1698                         av_log(matroska->ctx, AV_LOG_INFO,
1699                                "EBML block data error\n");
1700                         break;
1701                     }
1702                     data += n;
1703                     size -= n;
1704                     total = lace_size[0] = num;
1705                     for (n = 1; res == 0 && n < laces - 1; n++) {
1706                         int64_t snum;
1707                         int r;
1708                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1709                         if (r < 0) {
1710                             av_log(matroska->ctx, AV_LOG_INFO,
1711                                    "EBML block data error\n");
1712                             break;
1713                         }
1714                         data += r;
1715                         size -= r;
1716                         lace_size[n] = lace_size[n - 1] + snum;
1717                         total += lace_size[n];
1718                     }
1719                     lace_size[n] = size - total;
1720                     break;
1721                 }
1722             }
1723             break;
1724     }
1725
1726     if (res == 0) {
1727         for (n = 0; n < laces; n++) {
1728             if ((st->codec->codec_id == CODEC_ID_RA_288 ||
1729                  st->codec->codec_id == CODEC_ID_COOK ||
1730                  st->codec->codec_id == CODEC_ID_SIPR ||
1731                  st->codec->codec_id == CODEC_ID_ATRAC3) &&
1732                  st->codec->block_align && track->audio.sub_packet_size) {
1733                 int a = st->codec->block_align;
1734                 int sps = track->audio.sub_packet_size;
1735                 int cfs = track->audio.coded_framesize;
1736                 int h = track->audio.sub_packet_h;
1737                 int y = track->audio.sub_packet_cnt;
1738                 int w = track->audio.frame_size;
1739                 int x;
1740
1741                 if (!track->audio.pkt_cnt) {
1742                     if (st->codec->codec_id == CODEC_ID_RA_288)
1743                         for (x=0; x<h/2; x++)
1744                             memcpy(track->audio.buf+x*2*w+y*cfs,
1745                                    data+x*cfs, cfs);
1746                     else if (st->codec->codec_id == CODEC_ID_SIPR)
1747                         memcpy(track->audio.buf + y*w, data, w);
1748                     else
1749                         for (x=0; x<w/sps; x++)
1750                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1751
1752                     if (++track->audio.sub_packet_cnt >= h) {
1753                         if (st->codec->codec_id == CODEC_ID_SIPR)
1754                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
1755                         track->audio.sub_packet_cnt = 0;
1756                         track->audio.pkt_cnt = h*w / a;
1757                     }
1758                 }
1759                 while (track->audio.pkt_cnt) {
1760                     pkt = av_mallocz(sizeof(AVPacket));
1761                     av_new_packet(pkt, a);
1762                     memcpy(pkt->data, track->audio.buf
1763                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1764                     pkt->pos = pos;
1765                     pkt->stream_index = st->index;
1766                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1767                 }
1768             } else {
1769                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1770                 int offset = 0, pkt_size = lace_size[n];
1771                 uint8_t *pkt_data = data;
1772
1773                 if (pkt_size > size) {
1774                     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
1775                     break;
1776                 }
1777
1778                 if (encodings && encodings->scope & 1) {
1779                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1780                     if (offset < 0)
1781                         continue;
1782                 }
1783
1784                 pkt = av_mallocz(sizeof(AVPacket));
1785                 /* XXX: prevent data copy... */
1786                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1787                     av_free(pkt);
1788                     res = AVERROR(ENOMEM);
1789                     break;
1790                 }
1791                 if (offset)
1792                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1793                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1794
1795                 if (pkt_data != data)
1796                     av_free(pkt_data);
1797
1798                 if (n == 0)
1799                     pkt->flags = is_keyframe;
1800                 pkt->stream_index = st->index;
1801
1802                 if (track->ms_compat)
1803                     pkt->dts = timecode;
1804                 else
1805                     pkt->pts = timecode;
1806                 pkt->pos = pos;
1807                 if (st->codec->codec_id == CODEC_ID_TEXT)
1808                     pkt->convergence_duration = duration;
1809                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1810                     pkt->duration = duration;
1811
1812                 if (st->codec->codec_id == CODEC_ID_SSA)
1813                     matroska_fix_ass_packet(matroska, pkt, duration);
1814
1815                 if (matroska->prev_pkt &&
1816                     timecode != AV_NOPTS_VALUE &&
1817                     matroska->prev_pkt->pts == timecode &&
1818                     matroska->prev_pkt->stream_index == st->index &&
1819                     st->codec->codec_id == CODEC_ID_SSA)
1820                     matroska_merge_packets(matroska->prev_pkt, pkt);
1821                 else {
1822                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1823                     matroska->prev_pkt = pkt;
1824                 }
1825             }
1826
1827             if (timecode != AV_NOPTS_VALUE)
1828                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1829             data += lace_size[n];
1830             size -= lace_size[n];
1831         }
1832     }
1833
1834     av_free(lace_size);
1835     return res;
1836 }
1837
1838 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1839 {
1840     MatroskaCluster cluster = { 0 };
1841     EbmlList *blocks_list;
1842     MatroskaBlock *blocks;
1843     int i, res;
1844     int64_t pos = url_ftell(matroska->ctx->pb);
1845     matroska->prev_pkt = NULL;
1846     if (matroska->current_id)
1847         pos -= 4;  /* sizeof the ID which was already read */
1848     res = ebml_parse(matroska, matroska_clusters, &cluster);
1849     blocks_list = &cluster.blocks;
1850     blocks = blocks_list->elem;
1851     for (i=0; i<blocks_list->nb_elem; i++)
1852         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
1853             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
1854             res=matroska_parse_block(matroska,
1855                                      blocks[i].bin.data, blocks[i].bin.size,
1856                                      blocks[i].bin.pos,  cluster.timecode,
1857                                      blocks[i].duration, is_keyframe,
1858                                      pos);
1859         }
1860     ebml_free(matroska_cluster, &cluster);
1861     if (res < 0)  matroska->done = 1;
1862     return res;
1863 }
1864
1865 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1866 {
1867     MatroskaDemuxContext *matroska = s->priv_data;
1868
1869     while (matroska_deliver_packet(matroska, pkt)) {
1870         if (matroska->done)
1871             return AVERROR_EOF;
1872         matroska_parse_cluster(matroska);
1873     }
1874
1875     return 0;
1876 }
1877
1878 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1879                               int64_t timestamp, int flags)
1880 {
1881     MatroskaDemuxContext *matroska = s->priv_data;
1882     MatroskaTrack *tracks = matroska->tracks.elem;
1883     AVStream *st = s->streams[stream_index];
1884     int i, index, index_sub, index_min;
1885
1886     if (!st->nb_index_entries)
1887         return 0;
1888     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
1889
1890     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1891         url_fseek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
1892         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1893             matroska_clear_queue(matroska);
1894             if (matroska_parse_cluster(matroska) < 0)
1895                 break;
1896         }
1897     }
1898
1899     matroska_clear_queue(matroska);
1900     if (index < 0)
1901         return 0;
1902
1903     index_min = index;
1904     for (i=0; i < matroska->tracks.nb_elem; i++) {
1905         tracks[i].end_timecode = 0;
1906         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
1907             && !tracks[i].stream->discard != AVDISCARD_ALL) {
1908             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
1909             if (index_sub >= 0
1910                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
1911                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
1912                 index_min = index_sub;
1913         }
1914     }
1915
1916     url_fseek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
1917     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
1918     matroska->skip_to_timecode = st->index_entries[index].timestamp;
1919     matroska->done = 0;
1920     av_update_cur_dts(s, st, st->index_entries[index].timestamp);
1921     return 0;
1922 }
1923
1924 static int matroska_read_close(AVFormatContext *s)
1925 {
1926     MatroskaDemuxContext *matroska = s->priv_data;
1927     MatroskaTrack *tracks = matroska->tracks.elem;
1928     int n;
1929
1930     matroska_clear_queue(matroska);
1931
1932     for (n=0; n < matroska->tracks.nb_elem; n++)
1933         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
1934             av_free(tracks[n].audio.buf);
1935     ebml_free(matroska_segment, matroska);
1936
1937     return 0;
1938 }
1939
1940 AVInputFormat ff_matroska_demuxer = {
1941     "matroska,webm",
1942     NULL_IF_CONFIG_SMALL("Matroska/WebM file format"),
1943     sizeof(MatroskaDemuxContext),
1944     matroska_probe,
1945     matroska_read_header,
1946     matroska_read_packet,
1947     matroska_read_close,
1948     matroska_read_seek,
1949 };