Add a special function to mkv demxuer to parse length values that includes
[ffmpeg.git] / libavformat / matroskadec.c
1 /*
2  * Matroska file demuxer
3  * Copyright (c) 2003-2008 The FFmpeg Project
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23  * @file
24  * Matroska file demuxer
25  * by Ronald Bultje <rbultje@ronald.bitfreak.net>
26  * with a little help from Moritz Bunkus <moritz@bunkus.org>
27  * totally reworked by Aurelien Jacobs <aurel@gnuage.org>
28  * Specs available on the Matroska project page: http://www.matroska.org/.
29  */
30
31 #include <stdio.h>
32 #include "avformat.h"
33 #include "internal.h"
34 /* For ff_codec_get_id(). */
35 #include "riff.h"
36 #include "isom.h"
37 #include "rm.h"
38 #include "matroska.h"
39 #include "libavcodec/mpeg4audio.h"
40 #include "libavutil/intfloat_readwrite.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavutil/avstring.h"
43 #include "libavutil/lzo.h"
44 #if CONFIG_ZLIB
45 #include <zlib.h>
46 #endif
47 #if CONFIG_BZLIB
48 #include <bzlib.h>
49 #endif
50
51 typedef enum {
52     EBML_NONE,
53     EBML_UINT,
54     EBML_FLOAT,
55     EBML_STR,
56     EBML_UTF8,
57     EBML_BIN,
58     EBML_NEST,
59     EBML_PASS,
60     EBML_STOP,
61 } EbmlType;
62
63 typedef const struct EbmlSyntax {
64     uint32_t id;
65     EbmlType type;
66     int list_elem_size;
67     int data_offset;
68     union {
69         uint64_t    u;
70         double      f;
71         const char *s;
72         const struct EbmlSyntax *n;
73     } def;
74 } EbmlSyntax;
75
76 typedef struct {
77     int nb_elem;
78     void *elem;
79 } EbmlList;
80
81 typedef struct {
82     int      size;
83     uint8_t *data;
84     int64_t  pos;
85 } EbmlBin;
86
87 typedef struct {
88     uint64_t version;
89     uint64_t max_size;
90     uint64_t id_length;
91     char    *doctype;
92     uint64_t doctype_version;
93 } Ebml;
94
95 typedef struct {
96     uint64_t algo;
97     EbmlBin  settings;
98 } MatroskaTrackCompression;
99
100 typedef struct {
101     uint64_t scope;
102     uint64_t type;
103     MatroskaTrackCompression compression;
104 } MatroskaTrackEncoding;
105
106 typedef struct {
107     double   frame_rate;
108     uint64_t display_width;
109     uint64_t display_height;
110     uint64_t pixel_width;
111     uint64_t pixel_height;
112     uint64_t fourcc;
113 } MatroskaTrackVideo;
114
115 typedef struct {
116     double   samplerate;
117     double   out_samplerate;
118     uint64_t bitdepth;
119     uint64_t channels;
120
121     /* real audio header (extracted from extradata) */
122     int      coded_framesize;
123     int      sub_packet_h;
124     int      frame_size;
125     int      sub_packet_size;
126     int      sub_packet_cnt;
127     int      pkt_cnt;
128     uint8_t *buf;
129 } MatroskaTrackAudio;
130
131 typedef struct {
132     uint64_t num;
133     uint64_t uid;
134     uint64_t type;
135     char    *name;
136     char    *codec_id;
137     EbmlBin  codec_priv;
138     char    *language;
139     double time_scale;
140     uint64_t default_duration;
141     uint64_t flag_default;
142     uint64_t flag_forced;
143     MatroskaTrackVideo video;
144     MatroskaTrackAudio audio;
145     EbmlList encodings;
146
147     AVStream *stream;
148     int64_t end_timecode;
149     int ms_compat;
150 } MatroskaTrack;
151
152 typedef struct {
153     uint64_t uid;
154     char *filename;
155     char *mime;
156     EbmlBin bin;
157
158     AVStream *stream;
159 } MatroskaAttachement;
160
161 typedef struct {
162     uint64_t start;
163     uint64_t end;
164     uint64_t uid;
165     char    *title;
166
167     AVChapter *chapter;
168 } MatroskaChapter;
169
170 typedef struct {
171     uint64_t track;
172     uint64_t pos;
173 } MatroskaIndexPos;
174
175 typedef struct {
176     uint64_t time;
177     EbmlList pos;
178 } MatroskaIndex;
179
180 typedef struct {
181     char *name;
182     char *string;
183     char *lang;
184     uint64_t def;
185     EbmlList sub;
186 } MatroskaTag;
187
188 typedef struct {
189     char    *type;
190     uint64_t typevalue;
191     uint64_t trackuid;
192     uint64_t chapteruid;
193     uint64_t attachuid;
194 } MatroskaTagTarget;
195
196 typedef struct {
197     MatroskaTagTarget target;
198     EbmlList tag;
199 } MatroskaTags;
200
201 typedef struct {
202     uint64_t id;
203     uint64_t pos;
204 } MatroskaSeekhead;
205
206 typedef struct {
207     uint64_t start;
208     uint64_t length;
209 } MatroskaLevel;
210
211 typedef struct {
212     AVFormatContext *ctx;
213
214     /* EBML stuff */
215     int num_levels;
216     MatroskaLevel levels[EBML_MAX_DEPTH];
217     int level_up;
218     uint32_t current_id;
219
220     uint64_t time_scale;
221     double   duration;
222     char    *title;
223     EbmlList tracks;
224     EbmlList attachments;
225     EbmlList chapters;
226     EbmlList index;
227     EbmlList tags;
228     EbmlList seekhead;
229
230     /* byte position of the segment inside the stream */
231     int64_t segment_start;
232
233     /* the packet queue */
234     AVPacket **packets;
235     int num_packets;
236     AVPacket *prev_pkt;
237
238     int done;
239
240     /* What to skip before effectively reading a packet. */
241     int skip_to_keyframe;
242     uint64_t skip_to_timecode;
243 } MatroskaDemuxContext;
244
245 typedef struct {
246     uint64_t duration;
247     int64_t  reference;
248     uint64_t non_simple;
249     EbmlBin  bin;
250 } MatroskaBlock;
251
252 typedef struct {
253     uint64_t timecode;
254     EbmlList blocks;
255 } MatroskaCluster;
256
257 static EbmlSyntax ebml_header[] = {
258     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
259     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
260     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
261     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
262     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
263     { EBML_ID_EBMLVERSION,            EBML_NONE },
264     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
265     { 0 }
266 };
267
268 static EbmlSyntax ebml_syntax[] = {
269     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
270     { 0 }
271 };
272
273 static EbmlSyntax matroska_info[] = {
274     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
275     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
276     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
277     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
278     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
279     { MATROSKA_ID_DATEUTC,            EBML_NONE },
280     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
281     { 0 }
282 };
283
284 static EbmlSyntax matroska_track_video[] = {
285     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
286     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
287     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
288     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
289     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
290     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,fourcc) },
291     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
292     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
293     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
294     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
295     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
296     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
297     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_NONE },
298     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
299     { 0 }
300 };
301
302 static EbmlSyntax matroska_track_audio[] = {
303     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
304     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
305     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
306     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
307     { 0 }
308 };
309
310 static EbmlSyntax matroska_track_encoding_compression[] = {
311     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
312     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
313     { 0 }
314 };
315
316 static EbmlSyntax matroska_track_encoding[] = {
317     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
318     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
319     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
320     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
321     { 0 }
322 };
323
324 static EbmlSyntax matroska_track_encodings[] = {
325     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
326     { 0 }
327 };
328
329 static EbmlSyntax matroska_track[] = {
330     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
331     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
332     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
333     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
334     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
335     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
336     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
337     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
338     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
339     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
340     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_UINT, 0, offsetof(MatroskaTrack,flag_forced), {.u=0} },
341     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
342     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
343     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
344     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
345     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
346     { MATROSKA_ID_CODECNAME,            EBML_NONE },
347     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
348     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
349     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
350     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
351     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
352     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
353     { 0 }
354 };
355
356 static EbmlSyntax matroska_tracks[] = {
357     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
358     { 0 }
359 };
360
361 static EbmlSyntax matroska_attachment[] = {
362     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
363     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
364     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
365     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
366     { MATROSKA_ID_FILEDESC,           EBML_NONE },
367     { 0 }
368 };
369
370 static EbmlSyntax matroska_attachments[] = {
371     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
372     { 0 }
373 };
374
375 static EbmlSyntax matroska_chapter_display[] = {
376     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
377     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
378     { 0 }
379 };
380
381 static EbmlSyntax matroska_chapter_entry[] = {
382     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
383     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
384     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
385     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
386     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
387     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
388     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
389     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
390     { 0 }
391 };
392
393 static EbmlSyntax matroska_chapter[] = {
394     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
395     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
396     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
397     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
398     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
399     { 0 }
400 };
401
402 static EbmlSyntax matroska_chapters[] = {
403     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
404     { 0 }
405 };
406
407 static EbmlSyntax matroska_index_pos[] = {
408     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
409     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
410     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
411     { 0 }
412 };
413
414 static EbmlSyntax matroska_index_entry[] = {
415     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
416     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
417     { 0 }
418 };
419
420 static EbmlSyntax matroska_index[] = {
421     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
422     { 0 }
423 };
424
425 static EbmlSyntax matroska_simpletag[] = {
426     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
427     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
428     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
429     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
430     { MATROSKA_ID_TAGDEFAULT_BUG,     EBML_UINT, 0, offsetof(MatroskaTag,def) },
431     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
432     { 0 }
433 };
434
435 static EbmlSyntax matroska_tagtargets[] = {
436     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
437     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
438     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
439     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
440     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
441     { 0 }
442 };
443
444 static EbmlSyntax matroska_tag[] = {
445     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
446     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
447     { 0 }
448 };
449
450 static EbmlSyntax matroska_tags[] = {
451     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
452     { 0 }
453 };
454
455 static EbmlSyntax matroska_seekhead_entry[] = {
456     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
457     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
458     { 0 }
459 };
460
461 static EbmlSyntax matroska_seekhead[] = {
462     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
463     { 0 }
464 };
465
466 static EbmlSyntax matroska_segment[] = {
467     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
468     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
469     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
470     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
471     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
472     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
473     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
474     { MATROSKA_ID_CLUSTER,        EBML_STOP },
475     { 0 }
476 };
477
478 static EbmlSyntax matroska_segments[] = {
479     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
480     { 0 }
481 };
482
483 static EbmlSyntax matroska_blockgroup[] = {
484     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
485     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
486     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration), {.u=AV_NOPTS_VALUE} },
487     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
488     { 1,                          EBML_UINT, 0, offsetof(MatroskaBlock,non_simple), {.u=1} },
489     { 0 }
490 };
491
492 static EbmlSyntax matroska_cluster[] = {
493     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
494     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
495     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
496     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
497     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
498     { 0 }
499 };
500
501 static EbmlSyntax matroska_clusters[] = {
502     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
503     { MATROSKA_ID_INFO,           EBML_NONE },
504     { MATROSKA_ID_CUES,           EBML_NONE },
505     { MATROSKA_ID_TAGS,           EBML_NONE },
506     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
507     { 0 }
508 };
509
510 static const char *matroska_doctypes[] = { "matroska", "webm" };
511
512 /*
513  * Return: Whether we reached the end of a level in the hierarchy or not.
514  */
515 static int ebml_level_end(MatroskaDemuxContext *matroska)
516 {
517     ByteIOContext *pb = matroska->ctx->pb;
518     int64_t pos = url_ftell(pb);
519
520     if (matroska->num_levels > 0) {
521         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
522         if (pos - level->start >= level->length || matroska->current_id) {
523             matroska->num_levels--;
524             return 1;
525         }
526     }
527     return 0;
528 }
529
530 /*
531  * Read: an "EBML number", which is defined as a variable-length
532  * array of bytes. The first byte indicates the length by giving a
533  * number of 0-bits followed by a one. The position of the first
534  * "one" bit inside the first byte indicates the length of this
535  * number.
536  * Returns: number of bytes read, < 0 on error
537  */
538 static int ebml_read_num(MatroskaDemuxContext *matroska, ByteIOContext *pb,
539                          int max_size, uint64_t *number)
540 {
541     int read = 1, n = 1;
542     uint64_t total = 0;
543
544     /* The first byte tells us the length in bytes - get_byte() can normally
545      * return 0, but since that's not a valid first ebmlID byte, we can
546      * use it safely here to catch EOS. */
547     if (!(total = get_byte(pb))) {
548         /* we might encounter EOS here */
549         if (!url_feof(pb)) {
550             int64_t pos = url_ftell(pb);
551             av_log(matroska->ctx, AV_LOG_ERROR,
552                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
553                    pos, pos);
554         }
555         return AVERROR(EIO); /* EOS or actual I/O error */
556     }
557
558     /* get the length of the EBML number */
559     read = 8 - ff_log2_tab[total];
560     if (read > max_size) {
561         int64_t pos = url_ftell(pb) - 1;
562         av_log(matroska->ctx, AV_LOG_ERROR,
563                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
564                (uint8_t) total, pos, pos);
565         return AVERROR_INVALIDDATA;
566     }
567
568     /* read out length */
569     total ^= 1 << ff_log2_tab[total];
570     while (n++ < read)
571         total = (total << 8) | get_byte(pb);
572
573     *number = total;
574
575     return read;
576 }
577
578 /**
579  * Read a EBML length value.
580  * This needs special handling for the "unknown length" case which has multiple
581  * encodings.
582  */
583 static int ebml_read_length(MatroskaDemuxContext *matroska, ByteIOContext *pb,
584                             uint64_t *number)
585 {
586     int res = ebml_read_num(matroska, pb, 8, number);
587     if (res > 0 && *number + 1 == 1ULL << (7 * res))
588         *number = 0xffffffffffffffULL;
589     return res;
590 }
591
592 /*
593  * Read the next element as an unsigned int.
594  * 0 is success, < 0 is failure.
595  */
596 static int ebml_read_uint(ByteIOContext *pb, int size, uint64_t *num)
597 {
598     int n = 0;
599
600     if (size > 8)
601         return AVERROR_INVALIDDATA;
602
603     /* big-endian ordering; build up number */
604     *num = 0;
605     while (n++ < size)
606         *num = (*num << 8) | get_byte(pb);
607
608     return 0;
609 }
610
611 /*
612  * Read the next element as a float.
613  * 0 is success, < 0 is failure.
614  */
615 static int ebml_read_float(ByteIOContext *pb, int size, double *num)
616 {
617     if (size == 0) {
618         *num = 0;
619     } else if (size == 4) {
620         *num= av_int2flt(get_be32(pb));
621     } else if(size==8){
622         *num= av_int2dbl(get_be64(pb));
623     } else
624         return AVERROR_INVALIDDATA;
625
626     return 0;
627 }
628
629 /*
630  * Read the next element as an ASCII string.
631  * 0 is success, < 0 is failure.
632  */
633 static int ebml_read_ascii(ByteIOContext *pb, int size, char **str)
634 {
635     av_free(*str);
636     /* EBML strings are usually not 0-terminated, so we allocate one
637      * byte more, read the string and NULL-terminate it ourselves. */
638     if (!(*str = av_malloc(size + 1)))
639         return AVERROR(ENOMEM);
640     if (get_buffer(pb, (uint8_t *) *str, size) != size) {
641         av_freep(str);
642         return AVERROR(EIO);
643     }
644     (*str)[size] = '\0';
645
646     return 0;
647 }
648
649 /*
650  * Read the next element as binary data.
651  * 0 is success, < 0 is failure.
652  */
653 static int ebml_read_binary(ByteIOContext *pb, int length, EbmlBin *bin)
654 {
655     av_free(bin->data);
656     if (!(bin->data = av_malloc(length)))
657         return AVERROR(ENOMEM);
658
659     bin->size = length;
660     bin->pos  = url_ftell(pb);
661     if (get_buffer(pb, bin->data, length) != length) {
662         av_freep(&bin->data);
663         return AVERROR(EIO);
664     }
665
666     return 0;
667 }
668
669 /*
670  * Read the next element, but only the header. The contents
671  * are supposed to be sub-elements which can be read separately.
672  * 0 is success, < 0 is failure.
673  */
674 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
675 {
676     ByteIOContext *pb = matroska->ctx->pb;
677     MatroskaLevel *level;
678
679     if (matroska->num_levels >= EBML_MAX_DEPTH) {
680         av_log(matroska->ctx, AV_LOG_ERROR,
681                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
682         return AVERROR(ENOSYS);
683     }
684
685     level = &matroska->levels[matroska->num_levels++];
686     level->start = url_ftell(pb);
687     level->length = length;
688
689     return 0;
690 }
691
692 /*
693  * Read signed/unsigned "EBML" numbers.
694  * Return: number of bytes processed, < 0 on error
695  */
696 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
697                                  uint8_t *data, uint32_t size, uint64_t *num)
698 {
699     ByteIOContext pb;
700     init_put_byte(&pb, data, size, 0, NULL, NULL, NULL, NULL);
701     return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
702 }
703
704 /*
705  * Same as above, but signed.
706  */
707 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
708                                  uint8_t *data, uint32_t size, int64_t *num)
709 {
710     uint64_t unum;
711     int res;
712
713     /* read as unsigned number first */
714     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
715         return res;
716
717     /* make signed (weird way) */
718     *num = unum - ((1LL << (7*res - 1)) - 1);
719
720     return res;
721 }
722
723 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
724                            EbmlSyntax *syntax, void *data);
725
726 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
727                          uint32_t id, void *data)
728 {
729     int i;
730     for (i=0; syntax[i].id; i++)
731         if (id == syntax[i].id)
732             break;
733     if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
734         matroska->num_levels > 0 &&
735         matroska->levels[matroska->num_levels-1].length == 0xffffffffffffff)
736         return 0;  // we reached the end of an unknown size cluster
737     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32)
738         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
739     return ebml_parse_elem(matroska, &syntax[i], data);
740 }
741
742 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
743                       void *data)
744 {
745     if (!matroska->current_id) {
746         uint64_t id;
747         int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
748         if (res < 0)
749             return res;
750         matroska->current_id = id | 1 << 7*res;
751     }
752     return ebml_parse_id(matroska, syntax, matroska->current_id, data);
753 }
754
755 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
756                            void *data)
757 {
758     int i, res = 0;
759
760     for (i=0; syntax[i].id; i++)
761         switch (syntax[i].type) {
762         case EBML_UINT:
763             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
764             break;
765         case EBML_FLOAT:
766             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
767             break;
768         case EBML_STR:
769         case EBML_UTF8:
770             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
771             break;
772         }
773
774     while (!res && !ebml_level_end(matroska))
775         res = ebml_parse(matroska, syntax, data);
776
777     return res;
778 }
779
780 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
781                            EbmlSyntax *syntax, void *data)
782 {
783     ByteIOContext *pb = matroska->ctx->pb;
784     uint32_t id = syntax->id;
785     uint64_t length;
786     int res;
787
788     data = (char *)data + syntax->data_offset;
789     if (syntax->list_elem_size) {
790         EbmlList *list = data;
791         list->elem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
792         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
793         memset(data, 0, syntax->list_elem_size);
794         list->nb_elem++;
795     }
796
797     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
798         matroska->current_id = 0;
799         if ((res = ebml_read_length(matroska, pb, &length)) < 0)
800             return res;
801     }
802
803     switch (syntax->type) {
804     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
805     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
806     case EBML_STR:
807     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
808     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
809     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
810                          return res;
811                      if (id == MATROSKA_ID_SEGMENT)
812                          matroska->segment_start = url_ftell(matroska->ctx->pb);
813                      return ebml_parse_nest(matroska, syntax->def.n, data);
814     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
815     case EBML_STOP:  return 1;
816     default:         return url_fseek(pb,length,SEEK_CUR)<0 ? AVERROR(EIO) : 0;
817     }
818     if (res == AVERROR_INVALIDDATA)
819         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
820     else if (res == AVERROR(EIO))
821         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
822     return res;
823 }
824
825 static void ebml_free(EbmlSyntax *syntax, void *data)
826 {
827     int i, j;
828     for (i=0; syntax[i].id; i++) {
829         void *data_off = (char *)data + syntax[i].data_offset;
830         switch (syntax[i].type) {
831         case EBML_STR:
832         case EBML_UTF8:  av_freep(data_off);                      break;
833         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
834         case EBML_NEST:
835             if (syntax[i].list_elem_size) {
836                 EbmlList *list = data_off;
837                 char *ptr = list->elem;
838                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
839                     ebml_free(syntax[i].def.n, ptr);
840                 av_free(list->elem);
841             } else
842                 ebml_free(syntax[i].def.n, data_off);
843         default:  break;
844         }
845     }
846 }
847
848
849 /*
850  * Autodetecting...
851  */
852 static int matroska_probe(AVProbeData *p)
853 {
854     uint64_t total = 0;
855     int len_mask = 0x80, size = 1, n = 1, i;
856
857     /* EBML header? */
858     if (AV_RB32(p->buf) != EBML_ID_HEADER)
859         return 0;
860
861     /* length of header */
862     total = p->buf[4];
863     while (size <= 8 && !(total & len_mask)) {
864         size++;
865         len_mask >>= 1;
866     }
867     if (size > 8)
868       return 0;
869     total &= (len_mask - 1);
870     while (n < size)
871         total = (total << 8) | p->buf[4 + n++];
872
873     /* Does the probe data contain the whole header? */
874     if (p->buf_size < 4 + size + total)
875       return 0;
876
877     /* The header should contain a known document type. For now,
878      * we don't parse the whole header but simply check for the
879      * availability of that array of characters inside the header.
880      * Not fully fool-proof, but good enough. */
881     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
882         int probelen = strlen(matroska_doctypes[i]);
883         for (n = 4+size; n <= 4+size+total-probelen; n++)
884             if (!memcmp(p->buf+n, matroska_doctypes[i], probelen))
885                 return AVPROBE_SCORE_MAX;
886     }
887
888     // probably valid EBML header but no recognized doctype
889     return AVPROBE_SCORE_MAX/2;
890 }
891
892 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
893                                                  int num)
894 {
895     MatroskaTrack *tracks = matroska->tracks.elem;
896     int i;
897
898     for (i=0; i < matroska->tracks.nb_elem; i++)
899         if (tracks[i].num == num)
900             return &tracks[i];
901
902     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
903     return NULL;
904 }
905
906 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
907                                   MatroskaTrack *track)
908 {
909     MatroskaTrackEncoding *encodings = track->encodings.elem;
910     uint8_t* data = *buf;
911     int isize = *buf_size;
912     uint8_t* pkt_data = NULL;
913     int pkt_size = isize;
914     int result = 0;
915     int olen;
916
917     if (pkt_size >= 10000000)
918         return -1;
919
920     switch (encodings[0].compression.algo) {
921     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
922         return encodings[0].compression.settings.size;
923     case MATROSKA_TRACK_ENCODING_COMP_LZO:
924         do {
925             olen = pkt_size *= 3;
926             pkt_data = av_realloc(pkt_data, pkt_size+AV_LZO_OUTPUT_PADDING);
927             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
928         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
929         if (result)
930             goto failed;
931         pkt_size -= olen;
932         break;
933 #if CONFIG_ZLIB
934     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
935         z_stream zstream = {0};
936         if (inflateInit(&zstream) != Z_OK)
937             return -1;
938         zstream.next_in = data;
939         zstream.avail_in = isize;
940         do {
941             pkt_size *= 3;
942             pkt_data = av_realloc(pkt_data, pkt_size);
943             zstream.avail_out = pkt_size - zstream.total_out;
944             zstream.next_out = pkt_data + zstream.total_out;
945             result = inflate(&zstream, Z_NO_FLUSH);
946         } while (result==Z_OK && pkt_size<10000000);
947         pkt_size = zstream.total_out;
948         inflateEnd(&zstream);
949         if (result != Z_STREAM_END)
950             goto failed;
951         break;
952     }
953 #endif
954 #if CONFIG_BZLIB
955     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
956         bz_stream bzstream = {0};
957         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
958             return -1;
959         bzstream.next_in = data;
960         bzstream.avail_in = isize;
961         do {
962             pkt_size *= 3;
963             pkt_data = av_realloc(pkt_data, pkt_size);
964             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
965             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
966             result = BZ2_bzDecompress(&bzstream);
967         } while (result==BZ_OK && pkt_size<10000000);
968         pkt_size = bzstream.total_out_lo32;
969         BZ2_bzDecompressEnd(&bzstream);
970         if (result != BZ_STREAM_END)
971             goto failed;
972         break;
973     }
974 #endif
975     default:
976         return -1;
977     }
978
979     *buf = pkt_data;
980     *buf_size = pkt_size;
981     return 0;
982  failed:
983     av_free(pkt_data);
984     return -1;
985 }
986
987 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
988                                     AVPacket *pkt, uint64_t display_duration)
989 {
990     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
991     for (; *ptr!=',' && ptr<end-1; ptr++);
992     if (*ptr == ',')
993         layer = ++ptr;
994     for (; *ptr!=',' && ptr<end-1; ptr++);
995     if (*ptr == ',') {
996         int64_t end_pts = pkt->pts + display_duration;
997         int sc = matroska->time_scale * pkt->pts / 10000000;
998         int ec = matroska->time_scale * end_pts  / 10000000;
999         int sh, sm, ss, eh, em, es, len;
1000         sh = sc/360000;  sc -= 360000*sh;
1001         sm = sc/  6000;  sc -=   6000*sm;
1002         ss = sc/   100;  sc -=    100*ss;
1003         eh = ec/360000;  ec -= 360000*eh;
1004         em = ec/  6000;  ec -=   6000*em;
1005         es = ec/   100;  ec -=    100*es;
1006         *ptr++ = '\0';
1007         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
1008         if (!(line = av_malloc(len)))
1009             return;
1010         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
1011                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
1012         av_free(pkt->data);
1013         pkt->data = line;
1014         pkt->size = strlen(line);
1015     }
1016 }
1017
1018 static void matroska_merge_packets(AVPacket *out, AVPacket *in)
1019 {
1020     out->data = av_realloc(out->data, out->size+in->size);
1021     memcpy(out->data+out->size, in->data, in->size);
1022     out->size += in->size;
1023     av_destruct_packet(in);
1024     av_free(in);
1025 }
1026
1027 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1028                                  AVMetadata **metadata, char *prefix)
1029 {
1030     MatroskaTag *tags = list->elem;
1031     char key[1024];
1032     int i;
1033
1034     for (i=0; i < list->nb_elem; i++) {
1035         const char *lang = strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1036         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1037         else         av_strlcpy(key, tags[i].name, sizeof(key));
1038         if (tags[i].def || !lang) {
1039         av_metadata_set2(metadata, key, tags[i].string, 0);
1040         if (tags[i].sub.nb_elem)
1041             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1042         }
1043         if (lang) {
1044             av_strlcat(key, "-", sizeof(key));
1045             av_strlcat(key, lang, sizeof(key));
1046             av_metadata_set2(metadata, key, tags[i].string, 0);
1047             if (tags[i].sub.nb_elem)
1048                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1049         }
1050     }
1051 }
1052
1053 static void matroska_convert_tags(AVFormatContext *s)
1054 {
1055     MatroskaDemuxContext *matroska = s->priv_data;
1056     MatroskaTags *tags = matroska->tags.elem;
1057     int i, j;
1058
1059     for (i=0; i < matroska->tags.nb_elem; i++) {
1060         if (tags[i].target.attachuid) {
1061             MatroskaAttachement *attachment = matroska->attachments.elem;
1062             for (j=0; j<matroska->attachments.nb_elem; j++)
1063                 if (attachment[j].uid == tags[i].target.attachuid)
1064                     matroska_convert_tag(s, &tags[i].tag,
1065                                          &attachment[j].stream->metadata, NULL);
1066         } else if (tags[i].target.chapteruid) {
1067             MatroskaChapter *chapter = matroska->chapters.elem;
1068             for (j=0; j<matroska->chapters.nb_elem; j++)
1069                 if (chapter[j].uid == tags[i].target.chapteruid)
1070                     matroska_convert_tag(s, &tags[i].tag,
1071                                          &chapter[j].chapter->metadata, NULL);
1072         } else if (tags[i].target.trackuid) {
1073             MatroskaTrack *track = matroska->tracks.elem;
1074             for (j=0; j<matroska->tracks.nb_elem; j++)
1075                 if (track[j].uid == tags[i].target.trackuid)
1076                     matroska_convert_tag(s, &tags[i].tag,
1077                                          &track[j].stream->metadata, NULL);
1078         } else {
1079             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1080                                  tags[i].target.type);
1081         }
1082     }
1083 }
1084
1085 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1086 {
1087     EbmlList *seekhead_list = &matroska->seekhead;
1088     MatroskaSeekhead *seekhead = seekhead_list->elem;
1089     uint32_t level_up = matroska->level_up;
1090     int64_t before_pos = url_ftell(matroska->ctx->pb);
1091     uint32_t saved_id = matroska->current_id;
1092     MatroskaLevel level;
1093     int i;
1094
1095     // we should not do any seeking in the streaming case
1096     if (url_is_streamed(matroska->ctx->pb) ||
1097         (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1098         return;
1099
1100     for (i=0; i<seekhead_list->nb_elem; i++) {
1101         int64_t offset = seekhead[i].pos + matroska->segment_start;
1102
1103         if (seekhead[i].pos <= before_pos
1104             || seekhead[i].id == MATROSKA_ID_SEEKHEAD
1105             || seekhead[i].id == MATROSKA_ID_CLUSTER)
1106             continue;
1107
1108         /* seek */
1109         if (url_fseek(matroska->ctx->pb, offset, SEEK_SET) != offset)
1110             continue;
1111
1112         /* We don't want to lose our seekhead level, so we add
1113          * a dummy. This is a crude hack. */
1114         if (matroska->num_levels == EBML_MAX_DEPTH) {
1115             av_log(matroska->ctx, AV_LOG_INFO,
1116                    "Max EBML element depth (%d) reached, "
1117                    "cannot parse further.\n", EBML_MAX_DEPTH);
1118             break;
1119         }
1120
1121         level.start = 0;
1122         level.length = (uint64_t)-1;
1123         matroska->levels[matroska->num_levels] = level;
1124         matroska->num_levels++;
1125         matroska->current_id = 0;
1126
1127         ebml_parse(matroska, matroska_segment, matroska);
1128
1129         /* remove dummy level */
1130         while (matroska->num_levels) {
1131             uint64_t length = matroska->levels[--matroska->num_levels].length;
1132             if (length == (uint64_t)-1)
1133                 break;
1134         }
1135     }
1136
1137     /* seek back */
1138     url_fseek(matroska->ctx->pb, before_pos, SEEK_SET);
1139     matroska->level_up = level_up;
1140     matroska->current_id = saved_id;
1141 }
1142
1143 static int matroska_aac_profile(char *codec_id)
1144 {
1145     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1146     int profile;
1147
1148     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1149         if (strstr(codec_id, aac_profiles[profile]))
1150             break;
1151     return profile + 1;
1152 }
1153
1154 static int matroska_aac_sri(int samplerate)
1155 {
1156     int sri;
1157
1158     for (sri=0; sri<FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
1159         if (ff_mpeg4audio_sample_rates[sri] == samplerate)
1160             break;
1161     return sri;
1162 }
1163
1164 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
1165 {
1166     MatroskaDemuxContext *matroska = s->priv_data;
1167     EbmlList *attachements_list = &matroska->attachments;
1168     MatroskaAttachement *attachements;
1169     EbmlList *chapters_list = &matroska->chapters;
1170     MatroskaChapter *chapters;
1171     MatroskaTrack *tracks;
1172     EbmlList *index_list;
1173     MatroskaIndex *index;
1174     int index_scale = 1;
1175     uint64_t max_start = 0;
1176     Ebml ebml = { 0 };
1177     AVStream *st;
1178     int i, j, res;
1179
1180     matroska->ctx = s;
1181
1182     /* First read the EBML header. */
1183     if (ebml_parse(matroska, ebml_syntax, &ebml)
1184         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1185         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 2) {
1186         av_log(matroska->ctx, AV_LOG_ERROR,
1187                "EBML header using unsupported features\n"
1188                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1189                ebml.version, ebml.doctype, ebml.doctype_version);
1190         ebml_free(ebml_syntax, &ebml);
1191         return AVERROR_PATCHWELCOME;
1192     }
1193     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1194         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1195             break;
1196     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1197         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1198     }
1199     ebml_free(ebml_syntax, &ebml);
1200
1201     /* The next thing is a segment. */
1202     if ((res = ebml_parse(matroska, matroska_segments, matroska)) < 0)
1203         return res;
1204     matroska_execute_seekhead(matroska);
1205
1206     if (!matroska->time_scale)
1207         matroska->time_scale = 1000000;
1208     if (matroska->duration)
1209         matroska->ctx->duration = matroska->duration * matroska->time_scale
1210                                   * 1000 / AV_TIME_BASE;
1211     av_metadata_set2(&s->metadata, "title", matroska->title, 0);
1212
1213     tracks = matroska->tracks.elem;
1214     for (i=0; i < matroska->tracks.nb_elem; i++) {
1215         MatroskaTrack *track = &tracks[i];
1216         enum CodecID codec_id = CODEC_ID_NONE;
1217         EbmlList *encodings_list = &tracks->encodings;
1218         MatroskaTrackEncoding *encodings = encodings_list->elem;
1219         uint8_t *extradata = NULL;
1220         int extradata_size = 0;
1221         int extradata_offset = 0;
1222         ByteIOContext b;
1223
1224         /* Apply some sanity checks. */
1225         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1226             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1227             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1228             av_log(matroska->ctx, AV_LOG_INFO,
1229                    "Unknown or unsupported track type %"PRIu64"\n",
1230                    track->type);
1231             continue;
1232         }
1233         if (track->codec_id == NULL)
1234             continue;
1235
1236         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1237             if (!track->default_duration)
1238                 track->default_duration = 1000000000/track->video.frame_rate;
1239             if (!track->video.display_width)
1240                 track->video.display_width = track->video.pixel_width;
1241             if (!track->video.display_height)
1242                 track->video.display_height = track->video.pixel_height;
1243         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1244             if (!track->audio.out_samplerate)
1245                 track->audio.out_samplerate = track->audio.samplerate;
1246         }
1247         if (encodings_list->nb_elem > 1) {
1248             av_log(matroska->ctx, AV_LOG_ERROR,
1249                    "Multiple combined encodings no supported");
1250         } else if (encodings_list->nb_elem == 1) {
1251             if (encodings[0].type ||
1252                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1253 #if CONFIG_ZLIB
1254                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1255 #endif
1256 #if CONFIG_BZLIB
1257                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1258 #endif
1259                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1260                 encodings[0].scope = 0;
1261                 av_log(matroska->ctx, AV_LOG_ERROR,
1262                        "Unsupported encoding type");
1263             } else if (track->codec_priv.size && encodings[0].scope&2) {
1264                 uint8_t *codec_priv = track->codec_priv.data;
1265                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1266                                                     &track->codec_priv.size,
1267                                                     track);
1268                 if (offset < 0) {
1269                     track->codec_priv.data = NULL;
1270                     track->codec_priv.size = 0;
1271                     av_log(matroska->ctx, AV_LOG_ERROR,
1272                            "Failed to decode codec private data\n");
1273                 } else if (offset > 0) {
1274                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1275                     memcpy(track->codec_priv.data,
1276                            encodings[0].compression.settings.data, offset);
1277                     memcpy(track->codec_priv.data+offset, codec_priv,
1278                            track->codec_priv.size);
1279                     track->codec_priv.size += offset;
1280                 }
1281                 if (codec_priv != track->codec_priv.data)
1282                     av_free(codec_priv);
1283             }
1284         }
1285
1286         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1287             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1288                         strlen(ff_mkv_codec_tags[j].str))){
1289                 codec_id= ff_mkv_codec_tags[j].id;
1290                 break;
1291             }
1292         }
1293
1294         st = track->stream = av_new_stream(s, 0);
1295         if (st == NULL)
1296             return AVERROR(ENOMEM);
1297
1298         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1299             && track->codec_priv.size >= 40
1300             && track->codec_priv.data != NULL) {
1301             track->ms_compat = 1;
1302             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1303             codec_id = ff_codec_get_id(ff_codec_bmp_tags, track->video.fourcc);
1304             extradata_offset = 40;
1305         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1306                    && track->codec_priv.size >= 14
1307                    && track->codec_priv.data != NULL) {
1308             init_put_byte(&b, track->codec_priv.data, track->codec_priv.size,
1309                           URL_RDONLY, NULL, NULL, NULL, NULL);
1310             ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1311             codec_id = st->codec->codec_id;
1312             extradata_offset = FFMIN(track->codec_priv.size, 18);
1313         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1314                    && (track->codec_priv.size >= 86)
1315                    && (track->codec_priv.data != NULL)) {
1316             track->video.fourcc = AV_RL32(track->codec_priv.data);
1317             codec_id=ff_codec_get_id(codec_movvideo_tags, track->video.fourcc);
1318         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1319             switch (track->audio.bitdepth) {
1320             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1321             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1322             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1323             }
1324         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1325             switch (track->audio.bitdepth) {
1326             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1327             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1328             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1329             }
1330         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1331             codec_id = CODEC_ID_PCM_F64LE;
1332         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1333             int profile = matroska_aac_profile(track->codec_id);
1334             int sri = matroska_aac_sri(track->audio.samplerate);
1335             extradata = av_malloc(5);
1336             if (extradata == NULL)
1337                 return AVERROR(ENOMEM);
1338             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1339             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1340             if (strstr(track->codec_id, "SBR")) {
1341                 sri = matroska_aac_sri(track->audio.out_samplerate);
1342                 extradata[2] = 0x56;
1343                 extradata[3] = 0xE5;
1344                 extradata[4] = 0x80 | (sri<<3);
1345                 extradata_size = 5;
1346             } else
1347                 extradata_size = 2;
1348         } else if (codec_id == CODEC_ID_TTA) {
1349             extradata_size = 30;
1350             extradata = av_mallocz(extradata_size);
1351             if (extradata == NULL)
1352                 return AVERROR(ENOMEM);
1353             init_put_byte(&b, extradata, extradata_size, 1,
1354                           NULL, NULL, NULL, NULL);
1355             put_buffer(&b, "TTA1", 4);
1356             put_le16(&b, 1);
1357             put_le16(&b, track->audio.channels);
1358             put_le16(&b, track->audio.bitdepth);
1359             put_le32(&b, track->audio.out_samplerate);
1360             put_le32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1361         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1362                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1363             extradata_offset = 26;
1364         } else if (codec_id == CODEC_ID_RA_144) {
1365             track->audio.out_samplerate = 8000;
1366             track->audio.channels = 1;
1367         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1368                    codec_id == CODEC_ID_ATRAC3 || codec_id == CODEC_ID_SIPR) {
1369             int flavor;
1370             init_put_byte(&b, track->codec_priv.data,track->codec_priv.size,
1371                           0, NULL, NULL, NULL, NULL);
1372             url_fskip(&b, 22);
1373             flavor                       = get_be16(&b);
1374             track->audio.coded_framesize = get_be32(&b);
1375             url_fskip(&b, 12);
1376             track->audio.sub_packet_h    = get_be16(&b);
1377             track->audio.frame_size      = get_be16(&b);
1378             track->audio.sub_packet_size = get_be16(&b);
1379             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1380             if (codec_id == CODEC_ID_RA_288) {
1381                 st->codec->block_align = track->audio.coded_framesize;
1382                 track->codec_priv.size = 0;
1383             } else {
1384                 if (codec_id == CODEC_ID_SIPR && flavor < 4) {
1385                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1386                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1387                     st->codec->bit_rate = sipr_bit_rate[flavor];
1388                 }
1389                 st->codec->block_align = track->audio.sub_packet_size;
1390                 extradata_offset = 78;
1391             }
1392         }
1393         track->codec_priv.size -= extradata_offset;
1394
1395         if (codec_id == CODEC_ID_NONE)
1396             av_log(matroska->ctx, AV_LOG_INFO,
1397                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1398
1399         if (track->time_scale < 0.01)
1400             track->time_scale = 1.0;
1401         av_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1402
1403         st->codec->codec_id = codec_id;
1404         st->start_time = 0;
1405         if (strcmp(track->language, "und"))
1406             av_metadata_set2(&st->metadata, "language", track->language, 0);
1407         av_metadata_set2(&st->metadata, "title", track->name, 0);
1408
1409         if (track->flag_default)
1410             st->disposition |= AV_DISPOSITION_DEFAULT;
1411         if (track->flag_forced)
1412             st->disposition |= AV_DISPOSITION_FORCED;
1413
1414         if (track->default_duration)
1415             av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
1416                       track->default_duration, 1000000000, 30000);
1417
1418         if (!st->codec->extradata) {
1419             if(extradata){
1420                 st->codec->extradata = extradata;
1421                 st->codec->extradata_size = extradata_size;
1422             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1423                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1424                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1425                 if(st->codec->extradata == NULL)
1426                     return AVERROR(ENOMEM);
1427                 st->codec->extradata_size = track->codec_priv.size;
1428                 memcpy(st->codec->extradata,
1429                        track->codec_priv.data + extradata_offset,
1430                        track->codec_priv.size);
1431             }
1432         }
1433
1434         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1435             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1436             st->codec->codec_tag  = track->video.fourcc;
1437             st->codec->width  = track->video.pixel_width;
1438             st->codec->height = track->video.pixel_height;
1439             av_reduce(&st->sample_aspect_ratio.num,
1440                       &st->sample_aspect_ratio.den,
1441                       st->codec->height * track->video.display_width,
1442                       st->codec-> width * track->video.display_height,
1443                       255);
1444             if (st->codec->codec_id != CODEC_ID_H264)
1445             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1446             if (track->default_duration)
1447                 st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
1448         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1449             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1450             st->codec->sample_rate = track->audio.out_samplerate;
1451             st->codec->channels = track->audio.channels;
1452             if (st->codec->codec_id != CODEC_ID_AAC)
1453             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1454         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1455             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1456         }
1457     }
1458
1459     attachements = attachements_list->elem;
1460     for (j=0; j<attachements_list->nb_elem; j++) {
1461         if (!(attachements[j].filename && attachements[j].mime &&
1462               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1463             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1464         } else {
1465             AVStream *st = av_new_stream(s, 0);
1466             if (st == NULL)
1467                 break;
1468             av_metadata_set2(&st->metadata, "filename",attachements[j].filename, 0);
1469             st->codec->codec_id = CODEC_ID_NONE;
1470             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1471             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1472             if(st->codec->extradata == NULL)
1473                 break;
1474             st->codec->extradata_size = attachements[j].bin.size;
1475             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1476
1477             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1478                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1479                              strlen(ff_mkv_mime_tags[i].str))) {
1480                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1481                     break;
1482                 }
1483             }
1484             attachements[j].stream = st;
1485         }
1486     }
1487
1488     chapters = chapters_list->elem;
1489     for (i=0; i<chapters_list->nb_elem; i++)
1490         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1491             && (max_start==0 || chapters[i].start > max_start)) {
1492             chapters[i].chapter =
1493             ff_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1494                            chapters[i].start, chapters[i].end,
1495                            chapters[i].title);
1496             av_metadata_set2(&chapters[i].chapter->metadata,
1497                              "title", chapters[i].title, 0);
1498             max_start = chapters[i].start;
1499         }
1500
1501     index_list = &matroska->index;
1502     index = index_list->elem;
1503     if (index_list->nb_elem
1504         && index[0].time > 100000000000000/matroska->time_scale) {
1505         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1506         index_scale = matroska->time_scale;
1507     }
1508     for (i=0; i<index_list->nb_elem; i++) {
1509         EbmlList *pos_list = &index[i].pos;
1510         MatroskaIndexPos *pos = pos_list->elem;
1511         for (j=0; j<pos_list->nb_elem; j++) {
1512             MatroskaTrack *track = matroska_find_track_by_num(matroska,
1513                                                               pos[j].track);
1514             if (track && track->stream)
1515                 av_add_index_entry(track->stream,
1516                                    pos[j].pos + matroska->segment_start,
1517                                    index[i].time/index_scale, 0, 0,
1518                                    AVINDEX_KEYFRAME);
1519         }
1520     }
1521
1522     matroska_convert_tags(s);
1523
1524     return 0;
1525 }
1526
1527 /*
1528  * Put one packet in an application-supplied AVPacket struct.
1529  * Returns 0 on success or -1 on failure.
1530  */
1531 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1532                                    AVPacket *pkt)
1533 {
1534     if (matroska->num_packets > 0) {
1535         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1536         av_free(matroska->packets[0]);
1537         if (matroska->num_packets > 1) {
1538             memmove(&matroska->packets[0], &matroska->packets[1],
1539                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1540             matroska->packets =
1541                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
1542                            sizeof(AVPacket *));
1543         } else {
1544             av_freep(&matroska->packets);
1545         }
1546         matroska->num_packets--;
1547         return 0;
1548     }
1549
1550     return -1;
1551 }
1552
1553 /*
1554  * Free all packets in our internal queue.
1555  */
1556 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1557 {
1558     if (matroska->packets) {
1559         int n;
1560         for (n = 0; n < matroska->num_packets; n++) {
1561             av_free_packet(matroska->packets[n]);
1562             av_free(matroska->packets[n]);
1563         }
1564         av_freep(&matroska->packets);
1565         matroska->num_packets = 0;
1566     }
1567 }
1568
1569 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1570                                 int size, int64_t pos, uint64_t cluster_time,
1571                                 uint64_t duration, int is_keyframe,
1572                                 int64_t cluster_pos)
1573 {
1574     uint64_t timecode = AV_NOPTS_VALUE;
1575     MatroskaTrack *track;
1576     int res = 0;
1577     AVStream *st;
1578     AVPacket *pkt;
1579     int16_t block_time;
1580     uint32_t *lace_size = NULL;
1581     int n, flags, laces = 0;
1582     uint64_t num;
1583
1584     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1585         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1586         return res;
1587     }
1588     data += n;
1589     size -= n;
1590
1591     track = matroska_find_track_by_num(matroska, num);
1592     if (size <= 3 || !track || !track->stream) {
1593         av_log(matroska->ctx, AV_LOG_INFO,
1594                "Invalid stream %"PRIu64" or size %u\n", num, size);
1595         return res;
1596     }
1597     st = track->stream;
1598     if (st->discard >= AVDISCARD_ALL)
1599         return res;
1600     if (duration == AV_NOPTS_VALUE)
1601         duration = track->default_duration / matroska->time_scale;
1602
1603     block_time = AV_RB16(data);
1604     data += 2;
1605     flags = *data++;
1606     size -= 3;
1607     if (is_keyframe == -1)
1608         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1609
1610     if (cluster_time != (uint64_t)-1
1611         && (block_time >= 0 || cluster_time >= -block_time)) {
1612         timecode = cluster_time + block_time;
1613         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1614             && timecode < track->end_timecode)
1615             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1616         if (is_keyframe)
1617             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1618         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1619     }
1620
1621     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1622         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1623             return res;
1624         matroska->skip_to_keyframe = 0;
1625     }
1626
1627     switch ((flags & 0x06) >> 1) {
1628         case 0x0: /* no lacing */
1629             laces = 1;
1630             lace_size = av_mallocz(sizeof(int));
1631             lace_size[0] = size;
1632             break;
1633
1634         case 0x1: /* Xiph lacing */
1635         case 0x2: /* fixed-size lacing */
1636         case 0x3: /* EBML lacing */
1637             assert(size>0); // size <=3 is checked before size-=3 above
1638             laces = (*data) + 1;
1639             data += 1;
1640             size -= 1;
1641             lace_size = av_mallocz(laces * sizeof(int));
1642
1643             switch ((flags & 0x06) >> 1) {
1644                 case 0x1: /* Xiph lacing */ {
1645                     uint8_t temp;
1646                     uint32_t total = 0;
1647                     for (n = 0; res == 0 && n < laces - 1; n++) {
1648                         while (1) {
1649                             if (size == 0) {
1650                                 res = -1;
1651                                 break;
1652                             }
1653                             temp = *data;
1654                             lace_size[n] += temp;
1655                             data += 1;
1656                             size -= 1;
1657                             if (temp != 0xff)
1658                                 break;
1659                         }
1660                         total += lace_size[n];
1661                     }
1662                     lace_size[n] = size - total;
1663                     break;
1664                 }
1665
1666                 case 0x2: /* fixed-size lacing */
1667                     for (n = 0; n < laces; n++)
1668                         lace_size[n] = size / laces;
1669                     break;
1670
1671                 case 0x3: /* EBML lacing */ {
1672                     uint32_t total;
1673                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1674                     if (n < 0) {
1675                         av_log(matroska->ctx, AV_LOG_INFO,
1676                                "EBML block data error\n");
1677                         break;
1678                     }
1679                     data += n;
1680                     size -= n;
1681                     total = lace_size[0] = num;
1682                     for (n = 1; res == 0 && n < laces - 1; n++) {
1683                         int64_t snum;
1684                         int r;
1685                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1686                         if (r < 0) {
1687                             av_log(matroska->ctx, AV_LOG_INFO,
1688                                    "EBML block data error\n");
1689                             break;
1690                         }
1691                         data += r;
1692                         size -= r;
1693                         lace_size[n] = lace_size[n - 1] + snum;
1694                         total += lace_size[n];
1695                     }
1696                     lace_size[n] = size - total;
1697                     break;
1698                 }
1699             }
1700             break;
1701     }
1702
1703     if (res == 0) {
1704         for (n = 0; n < laces; n++) {
1705             if ((st->codec->codec_id == CODEC_ID_RA_288 ||
1706                  st->codec->codec_id == CODEC_ID_COOK ||
1707                  st->codec->codec_id == CODEC_ID_SIPR ||
1708                  st->codec->codec_id == CODEC_ID_ATRAC3) &&
1709                  st->codec->block_align && track->audio.sub_packet_size) {
1710                 int a = st->codec->block_align;
1711                 int sps = track->audio.sub_packet_size;
1712                 int cfs = track->audio.coded_framesize;
1713                 int h = track->audio.sub_packet_h;
1714                 int y = track->audio.sub_packet_cnt;
1715                 int w = track->audio.frame_size;
1716                 int x;
1717
1718                 if (!track->audio.pkt_cnt) {
1719                     if (st->codec->codec_id == CODEC_ID_RA_288)
1720                         for (x=0; x<h/2; x++)
1721                             memcpy(track->audio.buf+x*2*w+y*cfs,
1722                                    data+x*cfs, cfs);
1723                     else if (st->codec->codec_id == CODEC_ID_SIPR)
1724                         memcpy(track->audio.buf + y*w, data, w);
1725                     else
1726                         for (x=0; x<w/sps; x++)
1727                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1728
1729                     if (++track->audio.sub_packet_cnt >= h) {
1730                         if (st->codec->codec_id == CODEC_ID_SIPR)
1731                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
1732                         track->audio.sub_packet_cnt = 0;
1733                         track->audio.pkt_cnt = h*w / a;
1734                     }
1735                 }
1736                 while (track->audio.pkt_cnt) {
1737                     pkt = av_mallocz(sizeof(AVPacket));
1738                     av_new_packet(pkt, a);
1739                     memcpy(pkt->data, track->audio.buf
1740                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1741                     pkt->pos = pos;
1742                     pkt->stream_index = st->index;
1743                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1744                 }
1745             } else {
1746                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1747                 int offset = 0, pkt_size = lace_size[n];
1748                 uint8_t *pkt_data = data;
1749
1750                 if (pkt_size > size) {
1751                     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
1752                     break;
1753                 }
1754
1755                 if (encodings && encodings->scope & 1) {
1756                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1757                     if (offset < 0)
1758                         continue;
1759                 }
1760
1761                 pkt = av_mallocz(sizeof(AVPacket));
1762                 /* XXX: prevent data copy... */
1763                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1764                     av_free(pkt);
1765                     res = AVERROR(ENOMEM);
1766                     break;
1767                 }
1768                 if (offset)
1769                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1770                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1771
1772                 if (pkt_data != data)
1773                     av_free(pkt_data);
1774
1775                 if (n == 0)
1776                     pkt->flags = is_keyframe;
1777                 pkt->stream_index = st->index;
1778
1779                 if (track->ms_compat)
1780                     pkt->dts = timecode;
1781                 else
1782                     pkt->pts = timecode;
1783                 pkt->pos = pos;
1784                 if (st->codec->codec_id == CODEC_ID_TEXT)
1785                     pkt->convergence_duration = duration;
1786                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1787                     pkt->duration = duration;
1788
1789                 if (st->codec->codec_id == CODEC_ID_SSA)
1790                     matroska_fix_ass_packet(matroska, pkt, duration);
1791
1792                 if (matroska->prev_pkt &&
1793                     timecode != AV_NOPTS_VALUE &&
1794                     matroska->prev_pkt->pts == timecode &&
1795                     matroska->prev_pkt->stream_index == st->index &&
1796                     st->codec->codec_id == CODEC_ID_SSA)
1797                     matroska_merge_packets(matroska->prev_pkt, pkt);
1798                 else {
1799                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1800                     matroska->prev_pkt = pkt;
1801                 }
1802             }
1803
1804             if (timecode != AV_NOPTS_VALUE)
1805                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1806             data += lace_size[n];
1807             size -= lace_size[n];
1808         }
1809     }
1810
1811     av_free(lace_size);
1812     return res;
1813 }
1814
1815 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1816 {
1817     MatroskaCluster cluster = { 0 };
1818     EbmlList *blocks_list;
1819     MatroskaBlock *blocks;
1820     int i, res;
1821     int64_t pos = url_ftell(matroska->ctx->pb);
1822     matroska->prev_pkt = NULL;
1823     if (matroska->current_id)
1824         pos -= 4;  /* sizeof the ID which was already read */
1825     res = ebml_parse(matroska, matroska_clusters, &cluster);
1826     blocks_list = &cluster.blocks;
1827     blocks = blocks_list->elem;
1828     for (i=0; i<blocks_list->nb_elem; i++)
1829         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
1830             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
1831             res=matroska_parse_block(matroska,
1832                                      blocks[i].bin.data, blocks[i].bin.size,
1833                                      blocks[i].bin.pos,  cluster.timecode,
1834                                      blocks[i].duration, is_keyframe,
1835                                      pos);
1836         }
1837     ebml_free(matroska_cluster, &cluster);
1838     if (res < 0)  matroska->done = 1;
1839     return res;
1840 }
1841
1842 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1843 {
1844     MatroskaDemuxContext *matroska = s->priv_data;
1845
1846     while (matroska_deliver_packet(matroska, pkt)) {
1847         if (matroska->done)
1848             return AVERROR_EOF;
1849         matroska_parse_cluster(matroska);
1850     }
1851
1852     return 0;
1853 }
1854
1855 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1856                               int64_t timestamp, int flags)
1857 {
1858     MatroskaDemuxContext *matroska = s->priv_data;
1859     MatroskaTrack *tracks = matroska->tracks.elem;
1860     AVStream *st = s->streams[stream_index];
1861     int i, index, index_sub, index_min;
1862
1863     if (!st->nb_index_entries)
1864         return 0;
1865     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
1866
1867     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1868         url_fseek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
1869         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1870             matroska_clear_queue(matroska);
1871             if (matroska_parse_cluster(matroska) < 0)
1872                 break;
1873         }
1874     }
1875
1876     matroska_clear_queue(matroska);
1877     if (index < 0)
1878         return 0;
1879
1880     index_min = index;
1881     for (i=0; i < matroska->tracks.nb_elem; i++) {
1882         tracks[i].end_timecode = 0;
1883         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
1884             && !tracks[i].stream->discard != AVDISCARD_ALL) {
1885             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
1886             if (index_sub >= 0
1887                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
1888                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
1889                 index_min = index_sub;
1890         }
1891     }
1892
1893     url_fseek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
1894     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
1895     matroska->skip_to_timecode = st->index_entries[index].timestamp;
1896     matroska->done = 0;
1897     av_update_cur_dts(s, st, st->index_entries[index].timestamp);
1898     return 0;
1899 }
1900
1901 static int matroska_read_close(AVFormatContext *s)
1902 {
1903     MatroskaDemuxContext *matroska = s->priv_data;
1904     MatroskaTrack *tracks = matroska->tracks.elem;
1905     int n;
1906
1907     matroska_clear_queue(matroska);
1908
1909     for (n=0; n < matroska->tracks.nb_elem; n++)
1910         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
1911             av_free(tracks[n].audio.buf);
1912     ebml_free(matroska_segment, matroska);
1913
1914     return 0;
1915 }
1916
1917 AVInputFormat matroska_demuxer = {
1918     "matroska,webm",
1919     NULL_IF_CONFIG_SMALL("Matroska/WebM file format"),
1920     sizeof(MatroskaDemuxContext),
1921     matroska_probe,
1922     matroska_read_header,
1923     matroska_read_packet,
1924     matroska_read_close,
1925     matroska_read_seek,
1926     .metadata_conv = ff_mkv_metadata_conv,
1927 };