add ff_ prefix to metadata_conv()
[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     ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1052 }
1053
1054 static void matroska_convert_tags(AVFormatContext *s)
1055 {
1056     MatroskaDemuxContext *matroska = s->priv_data;
1057     MatroskaTags *tags = matroska->tags.elem;
1058     int i, j;
1059
1060     for (i=0; i < matroska->tags.nb_elem; i++) {
1061         if (tags[i].target.attachuid) {
1062             MatroskaAttachement *attachment = matroska->attachments.elem;
1063             for (j=0; j<matroska->attachments.nb_elem; j++)
1064                 if (attachment[j].uid == tags[i].target.attachuid)
1065                     matroska_convert_tag(s, &tags[i].tag,
1066                                          &attachment[j].stream->metadata, NULL);
1067         } else if (tags[i].target.chapteruid) {
1068             MatroskaChapter *chapter = matroska->chapters.elem;
1069             for (j=0; j<matroska->chapters.nb_elem; j++)
1070                 if (chapter[j].uid == tags[i].target.chapteruid)
1071                     matroska_convert_tag(s, &tags[i].tag,
1072                                          &chapter[j].chapter->metadata, NULL);
1073         } else if (tags[i].target.trackuid) {
1074             MatroskaTrack *track = matroska->tracks.elem;
1075             for (j=0; j<matroska->tracks.nb_elem; j++)
1076                 if (track[j].uid == tags[i].target.trackuid)
1077                     matroska_convert_tag(s, &tags[i].tag,
1078                                          &track[j].stream->metadata, NULL);
1079         } else {
1080             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1081                                  tags[i].target.type);
1082         }
1083     }
1084 }
1085
1086 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1087 {
1088     EbmlList *seekhead_list = &matroska->seekhead;
1089     MatroskaSeekhead *seekhead = seekhead_list->elem;
1090     uint32_t level_up = matroska->level_up;
1091     int64_t before_pos = url_ftell(matroska->ctx->pb);
1092     uint32_t saved_id = matroska->current_id;
1093     MatroskaLevel level;
1094     int i;
1095
1096     // we should not do any seeking in the streaming case
1097     if (url_is_streamed(matroska->ctx->pb) ||
1098         (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1099         return;
1100
1101     for (i=0; i<seekhead_list->nb_elem; i++) {
1102         int64_t offset = seekhead[i].pos + matroska->segment_start;
1103
1104         if (seekhead[i].pos <= before_pos
1105             || seekhead[i].id == MATROSKA_ID_SEEKHEAD
1106             || seekhead[i].id == MATROSKA_ID_CLUSTER)
1107             continue;
1108
1109         /* seek */
1110         if (url_fseek(matroska->ctx->pb, offset, SEEK_SET) != offset)
1111             continue;
1112
1113         /* We don't want to lose our seekhead level, so we add
1114          * a dummy. This is a crude hack. */
1115         if (matroska->num_levels == EBML_MAX_DEPTH) {
1116             av_log(matroska->ctx, AV_LOG_INFO,
1117                    "Max EBML element depth (%d) reached, "
1118                    "cannot parse further.\n", EBML_MAX_DEPTH);
1119             break;
1120         }
1121
1122         level.start = 0;
1123         level.length = (uint64_t)-1;
1124         matroska->levels[matroska->num_levels] = level;
1125         matroska->num_levels++;
1126         matroska->current_id = 0;
1127
1128         ebml_parse(matroska, matroska_segment, matroska);
1129
1130         /* remove dummy level */
1131         while (matroska->num_levels) {
1132             uint64_t length = matroska->levels[--matroska->num_levels].length;
1133             if (length == (uint64_t)-1)
1134                 break;
1135         }
1136     }
1137
1138     /* seek back */
1139     url_fseek(matroska->ctx->pb, before_pos, SEEK_SET);
1140     matroska->level_up = level_up;
1141     matroska->current_id = saved_id;
1142 }
1143
1144 static int matroska_aac_profile(char *codec_id)
1145 {
1146     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1147     int profile;
1148
1149     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1150         if (strstr(codec_id, aac_profiles[profile]))
1151             break;
1152     return profile + 1;
1153 }
1154
1155 static int matroska_aac_sri(int samplerate)
1156 {
1157     int sri;
1158
1159     for (sri=0; sri<FF_ARRAY_ELEMS(ff_mpeg4audio_sample_rates); sri++)
1160         if (ff_mpeg4audio_sample_rates[sri] == samplerate)
1161             break;
1162     return sri;
1163 }
1164
1165 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
1166 {
1167     MatroskaDemuxContext *matroska = s->priv_data;
1168     EbmlList *attachements_list = &matroska->attachments;
1169     MatroskaAttachement *attachements;
1170     EbmlList *chapters_list = &matroska->chapters;
1171     MatroskaChapter *chapters;
1172     MatroskaTrack *tracks;
1173     EbmlList *index_list;
1174     MatroskaIndex *index;
1175     int index_scale = 1;
1176     uint64_t max_start = 0;
1177     Ebml ebml = { 0 };
1178     AVStream *st;
1179     int i, j, res;
1180
1181     matroska->ctx = s;
1182
1183     /* First read the EBML header. */
1184     if (ebml_parse(matroska, ebml_syntax, &ebml)
1185         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1186         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 2) {
1187         av_log(matroska->ctx, AV_LOG_ERROR,
1188                "EBML header using unsupported features\n"
1189                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1190                ebml.version, ebml.doctype, ebml.doctype_version);
1191         ebml_free(ebml_syntax, &ebml);
1192         return AVERROR_PATCHWELCOME;
1193     }
1194     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1195         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1196             break;
1197     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1198         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1199     }
1200     ebml_free(ebml_syntax, &ebml);
1201
1202     /* The next thing is a segment. */
1203     if ((res = ebml_parse(matroska, matroska_segments, matroska)) < 0)
1204         return res;
1205     matroska_execute_seekhead(matroska);
1206
1207     if (!matroska->time_scale)
1208         matroska->time_scale = 1000000;
1209     if (matroska->duration)
1210         matroska->ctx->duration = matroska->duration * matroska->time_scale
1211                                   * 1000 / AV_TIME_BASE;
1212     av_metadata_set2(&s->metadata, "title", matroska->title, 0);
1213
1214     tracks = matroska->tracks.elem;
1215     for (i=0; i < matroska->tracks.nb_elem; i++) {
1216         MatroskaTrack *track = &tracks[i];
1217         enum CodecID codec_id = CODEC_ID_NONE;
1218         EbmlList *encodings_list = &tracks->encodings;
1219         MatroskaTrackEncoding *encodings = encodings_list->elem;
1220         uint8_t *extradata = NULL;
1221         int extradata_size = 0;
1222         int extradata_offset = 0;
1223         ByteIOContext b;
1224
1225         /* Apply some sanity checks. */
1226         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1227             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1228             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1229             av_log(matroska->ctx, AV_LOG_INFO,
1230                    "Unknown or unsupported track type %"PRIu64"\n",
1231                    track->type);
1232             continue;
1233         }
1234         if (track->codec_id == NULL)
1235             continue;
1236
1237         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1238             if (!track->default_duration)
1239                 track->default_duration = 1000000000/track->video.frame_rate;
1240             if (!track->video.display_width)
1241                 track->video.display_width = track->video.pixel_width;
1242             if (!track->video.display_height)
1243                 track->video.display_height = track->video.pixel_height;
1244         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1245             if (!track->audio.out_samplerate)
1246                 track->audio.out_samplerate = track->audio.samplerate;
1247         }
1248         if (encodings_list->nb_elem > 1) {
1249             av_log(matroska->ctx, AV_LOG_ERROR,
1250                    "Multiple combined encodings no supported");
1251         } else if (encodings_list->nb_elem == 1) {
1252             if (encodings[0].type ||
1253                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1254 #if CONFIG_ZLIB
1255                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1256 #endif
1257 #if CONFIG_BZLIB
1258                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1259 #endif
1260                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1261                 encodings[0].scope = 0;
1262                 av_log(matroska->ctx, AV_LOG_ERROR,
1263                        "Unsupported encoding type");
1264             } else if (track->codec_priv.size && encodings[0].scope&2) {
1265                 uint8_t *codec_priv = track->codec_priv.data;
1266                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1267                                                     &track->codec_priv.size,
1268                                                     track);
1269                 if (offset < 0) {
1270                     track->codec_priv.data = NULL;
1271                     track->codec_priv.size = 0;
1272                     av_log(matroska->ctx, AV_LOG_ERROR,
1273                            "Failed to decode codec private data\n");
1274                 } else if (offset > 0) {
1275                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1276                     memcpy(track->codec_priv.data,
1277                            encodings[0].compression.settings.data, offset);
1278                     memcpy(track->codec_priv.data+offset, codec_priv,
1279                            track->codec_priv.size);
1280                     track->codec_priv.size += offset;
1281                 }
1282                 if (codec_priv != track->codec_priv.data)
1283                     av_free(codec_priv);
1284             }
1285         }
1286
1287         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1288             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1289                         strlen(ff_mkv_codec_tags[j].str))){
1290                 codec_id= ff_mkv_codec_tags[j].id;
1291                 break;
1292             }
1293         }
1294
1295         st = track->stream = av_new_stream(s, 0);
1296         if (st == NULL)
1297             return AVERROR(ENOMEM);
1298
1299         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1300             && track->codec_priv.size >= 40
1301             && track->codec_priv.data != NULL) {
1302             track->ms_compat = 1;
1303             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1304             codec_id = ff_codec_get_id(ff_codec_bmp_tags, track->video.fourcc);
1305             extradata_offset = 40;
1306         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1307                    && track->codec_priv.size >= 14
1308                    && track->codec_priv.data != NULL) {
1309             init_put_byte(&b, track->codec_priv.data, track->codec_priv.size,
1310                           URL_RDONLY, NULL, NULL, NULL, NULL);
1311             ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1312             codec_id = st->codec->codec_id;
1313             extradata_offset = FFMIN(track->codec_priv.size, 18);
1314         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1315                    && (track->codec_priv.size >= 86)
1316                    && (track->codec_priv.data != NULL)) {
1317             track->video.fourcc = AV_RL32(track->codec_priv.data);
1318             codec_id=ff_codec_get_id(codec_movvideo_tags, track->video.fourcc);
1319         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1320             switch (track->audio.bitdepth) {
1321             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1322             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1323             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1324             }
1325         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1326             switch (track->audio.bitdepth) {
1327             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1328             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1329             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1330             }
1331         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1332             codec_id = CODEC_ID_PCM_F64LE;
1333         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1334             int profile = matroska_aac_profile(track->codec_id);
1335             int sri = matroska_aac_sri(track->audio.samplerate);
1336             extradata = av_malloc(5);
1337             if (extradata == NULL)
1338                 return AVERROR(ENOMEM);
1339             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1340             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1341             if (strstr(track->codec_id, "SBR")) {
1342                 sri = matroska_aac_sri(track->audio.out_samplerate);
1343                 extradata[2] = 0x56;
1344                 extradata[3] = 0xE5;
1345                 extradata[4] = 0x80 | (sri<<3);
1346                 extradata_size = 5;
1347             } else
1348                 extradata_size = 2;
1349         } else if (codec_id == CODEC_ID_TTA) {
1350             extradata_size = 30;
1351             extradata = av_mallocz(extradata_size);
1352             if (extradata == NULL)
1353                 return AVERROR(ENOMEM);
1354             init_put_byte(&b, extradata, extradata_size, 1,
1355                           NULL, NULL, NULL, NULL);
1356             put_buffer(&b, "TTA1", 4);
1357             put_le16(&b, 1);
1358             put_le16(&b, track->audio.channels);
1359             put_le16(&b, track->audio.bitdepth);
1360             put_le32(&b, track->audio.out_samplerate);
1361             put_le32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1362         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1363                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1364             extradata_offset = 26;
1365         } else if (codec_id == CODEC_ID_RA_144) {
1366             track->audio.out_samplerate = 8000;
1367             track->audio.channels = 1;
1368         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1369                    codec_id == CODEC_ID_ATRAC3 || codec_id == CODEC_ID_SIPR) {
1370             int flavor;
1371             init_put_byte(&b, track->codec_priv.data,track->codec_priv.size,
1372                           0, NULL, NULL, NULL, NULL);
1373             url_fskip(&b, 22);
1374             flavor                       = get_be16(&b);
1375             track->audio.coded_framesize = get_be32(&b);
1376             url_fskip(&b, 12);
1377             track->audio.sub_packet_h    = get_be16(&b);
1378             track->audio.frame_size      = get_be16(&b);
1379             track->audio.sub_packet_size = get_be16(&b);
1380             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1381             if (codec_id == CODEC_ID_RA_288) {
1382                 st->codec->block_align = track->audio.coded_framesize;
1383                 track->codec_priv.size = 0;
1384             } else {
1385                 if (codec_id == CODEC_ID_SIPR && flavor < 4) {
1386                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1387                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1388                     st->codec->bit_rate = sipr_bit_rate[flavor];
1389                 }
1390                 st->codec->block_align = track->audio.sub_packet_size;
1391                 extradata_offset = 78;
1392             }
1393         }
1394         track->codec_priv.size -= extradata_offset;
1395
1396         if (codec_id == CODEC_ID_NONE)
1397             av_log(matroska->ctx, AV_LOG_INFO,
1398                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1399
1400         if (track->time_scale < 0.01)
1401             track->time_scale = 1.0;
1402         av_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1403
1404         st->codec->codec_id = codec_id;
1405         st->start_time = 0;
1406         if (strcmp(track->language, "und"))
1407             av_metadata_set2(&st->metadata, "language", track->language, 0);
1408         av_metadata_set2(&st->metadata, "title", track->name, 0);
1409
1410         if (track->flag_default)
1411             st->disposition |= AV_DISPOSITION_DEFAULT;
1412         if (track->flag_forced)
1413             st->disposition |= AV_DISPOSITION_FORCED;
1414
1415         if (track->default_duration)
1416             av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
1417                       track->default_duration, 1000000000, 30000);
1418
1419         if (!st->codec->extradata) {
1420             if(extradata){
1421                 st->codec->extradata = extradata;
1422                 st->codec->extradata_size = extradata_size;
1423             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1424                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1425                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1426                 if(st->codec->extradata == NULL)
1427                     return AVERROR(ENOMEM);
1428                 st->codec->extradata_size = track->codec_priv.size;
1429                 memcpy(st->codec->extradata,
1430                        track->codec_priv.data + extradata_offset,
1431                        track->codec_priv.size);
1432             }
1433         }
1434
1435         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1436             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1437             st->codec->codec_tag  = track->video.fourcc;
1438             st->codec->width  = track->video.pixel_width;
1439             st->codec->height = track->video.pixel_height;
1440             av_reduce(&st->sample_aspect_ratio.num,
1441                       &st->sample_aspect_ratio.den,
1442                       st->codec->height * track->video.display_width,
1443                       st->codec-> width * track->video.display_height,
1444                       255);
1445             if (st->codec->codec_id != CODEC_ID_H264)
1446             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1447             if (track->default_duration)
1448                 st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
1449         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1450             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1451             st->codec->sample_rate = track->audio.out_samplerate;
1452             st->codec->channels = track->audio.channels;
1453             if (st->codec->codec_id != CODEC_ID_AAC)
1454             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1455         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1456             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1457         }
1458     }
1459
1460     attachements = attachements_list->elem;
1461     for (j=0; j<attachements_list->nb_elem; j++) {
1462         if (!(attachements[j].filename && attachements[j].mime &&
1463               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1464             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1465         } else {
1466             AVStream *st = av_new_stream(s, 0);
1467             if (st == NULL)
1468                 break;
1469             av_metadata_set2(&st->metadata, "filename",attachements[j].filename, 0);
1470             st->codec->codec_id = CODEC_ID_NONE;
1471             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1472             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1473             if(st->codec->extradata == NULL)
1474                 break;
1475             st->codec->extradata_size = attachements[j].bin.size;
1476             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1477
1478             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1479                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1480                              strlen(ff_mkv_mime_tags[i].str))) {
1481                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1482                     break;
1483                 }
1484             }
1485             attachements[j].stream = st;
1486         }
1487     }
1488
1489     chapters = chapters_list->elem;
1490     for (i=0; i<chapters_list->nb_elem; i++)
1491         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1492             && (max_start==0 || chapters[i].start > max_start)) {
1493             chapters[i].chapter =
1494             ff_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1495                            chapters[i].start, chapters[i].end,
1496                            chapters[i].title);
1497             av_metadata_set2(&chapters[i].chapter->metadata,
1498                              "title", chapters[i].title, 0);
1499             max_start = chapters[i].start;
1500         }
1501
1502     index_list = &matroska->index;
1503     index = index_list->elem;
1504     if (index_list->nb_elem
1505         && index[0].time > 100000000000000/matroska->time_scale) {
1506         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1507         index_scale = matroska->time_scale;
1508     }
1509     for (i=0; i<index_list->nb_elem; i++) {
1510         EbmlList *pos_list = &index[i].pos;
1511         MatroskaIndexPos *pos = pos_list->elem;
1512         for (j=0; j<pos_list->nb_elem; j++) {
1513             MatroskaTrack *track = matroska_find_track_by_num(matroska,
1514                                                               pos[j].track);
1515             if (track && track->stream)
1516                 av_add_index_entry(track->stream,
1517                                    pos[j].pos + matroska->segment_start,
1518                                    index[i].time/index_scale, 0, 0,
1519                                    AVINDEX_KEYFRAME);
1520         }
1521     }
1522
1523     matroska_convert_tags(s);
1524
1525     return 0;
1526 }
1527
1528 /*
1529  * Put one packet in an application-supplied AVPacket struct.
1530  * Returns 0 on success or -1 on failure.
1531  */
1532 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1533                                    AVPacket *pkt)
1534 {
1535     if (matroska->num_packets > 0) {
1536         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1537         av_free(matroska->packets[0]);
1538         if (matroska->num_packets > 1) {
1539             memmove(&matroska->packets[0], &matroska->packets[1],
1540                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1541             matroska->packets =
1542                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
1543                            sizeof(AVPacket *));
1544         } else {
1545             av_freep(&matroska->packets);
1546         }
1547         matroska->num_packets--;
1548         return 0;
1549     }
1550
1551     return -1;
1552 }
1553
1554 /*
1555  * Free all packets in our internal queue.
1556  */
1557 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1558 {
1559     if (matroska->packets) {
1560         int n;
1561         for (n = 0; n < matroska->num_packets; n++) {
1562             av_free_packet(matroska->packets[n]);
1563             av_free(matroska->packets[n]);
1564         }
1565         av_freep(&matroska->packets);
1566         matroska->num_packets = 0;
1567     }
1568 }
1569
1570 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1571                                 int size, int64_t pos, uint64_t cluster_time,
1572                                 uint64_t duration, int is_keyframe,
1573                                 int64_t cluster_pos)
1574 {
1575     uint64_t timecode = AV_NOPTS_VALUE;
1576     MatroskaTrack *track;
1577     int res = 0;
1578     AVStream *st;
1579     AVPacket *pkt;
1580     int16_t block_time;
1581     uint32_t *lace_size = NULL;
1582     int n, flags, laces = 0;
1583     uint64_t num;
1584
1585     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1586         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1587         return res;
1588     }
1589     data += n;
1590     size -= n;
1591
1592     track = matroska_find_track_by_num(matroska, num);
1593     if (size <= 3 || !track || !track->stream) {
1594         av_log(matroska->ctx, AV_LOG_INFO,
1595                "Invalid stream %"PRIu64" or size %u\n", num, size);
1596         return res;
1597     }
1598     st = track->stream;
1599     if (st->discard >= AVDISCARD_ALL)
1600         return res;
1601     if (duration == AV_NOPTS_VALUE)
1602         duration = track->default_duration / matroska->time_scale;
1603
1604     block_time = AV_RB16(data);
1605     data += 2;
1606     flags = *data++;
1607     size -= 3;
1608     if (is_keyframe == -1)
1609         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1610
1611     if (cluster_time != (uint64_t)-1
1612         && (block_time >= 0 || cluster_time >= -block_time)) {
1613         timecode = cluster_time + block_time;
1614         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1615             && timecode < track->end_timecode)
1616             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1617         if (is_keyframe)
1618             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1619         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1620     }
1621
1622     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1623         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1624             return res;
1625         matroska->skip_to_keyframe = 0;
1626     }
1627
1628     switch ((flags & 0x06) >> 1) {
1629         case 0x0: /* no lacing */
1630             laces = 1;
1631             lace_size = av_mallocz(sizeof(int));
1632             lace_size[0] = size;
1633             break;
1634
1635         case 0x1: /* Xiph lacing */
1636         case 0x2: /* fixed-size lacing */
1637         case 0x3: /* EBML lacing */
1638             assert(size>0); // size <=3 is checked before size-=3 above
1639             laces = (*data) + 1;
1640             data += 1;
1641             size -= 1;
1642             lace_size = av_mallocz(laces * sizeof(int));
1643
1644             switch ((flags & 0x06) >> 1) {
1645                 case 0x1: /* Xiph lacing */ {
1646                     uint8_t temp;
1647                     uint32_t total = 0;
1648                     for (n = 0; res == 0 && n < laces - 1; n++) {
1649                         while (1) {
1650                             if (size == 0) {
1651                                 res = -1;
1652                                 break;
1653                             }
1654                             temp = *data;
1655                             lace_size[n] += temp;
1656                             data += 1;
1657                             size -= 1;
1658                             if (temp != 0xff)
1659                                 break;
1660                         }
1661                         total += lace_size[n];
1662                     }
1663                     lace_size[n] = size - total;
1664                     break;
1665                 }
1666
1667                 case 0x2: /* fixed-size lacing */
1668                     for (n = 0; n < laces; n++)
1669                         lace_size[n] = size / laces;
1670                     break;
1671
1672                 case 0x3: /* EBML lacing */ {
1673                     uint32_t total;
1674                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1675                     if (n < 0) {
1676                         av_log(matroska->ctx, AV_LOG_INFO,
1677                                "EBML block data error\n");
1678                         break;
1679                     }
1680                     data += n;
1681                     size -= n;
1682                     total = lace_size[0] = num;
1683                     for (n = 1; res == 0 && n < laces - 1; n++) {
1684                         int64_t snum;
1685                         int r;
1686                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1687                         if (r < 0) {
1688                             av_log(matroska->ctx, AV_LOG_INFO,
1689                                    "EBML block data error\n");
1690                             break;
1691                         }
1692                         data += r;
1693                         size -= r;
1694                         lace_size[n] = lace_size[n - 1] + snum;
1695                         total += lace_size[n];
1696                     }
1697                     lace_size[n] = size - total;
1698                     break;
1699                 }
1700             }
1701             break;
1702     }
1703
1704     if (res == 0) {
1705         for (n = 0; n < laces; n++) {
1706             if ((st->codec->codec_id == CODEC_ID_RA_288 ||
1707                  st->codec->codec_id == CODEC_ID_COOK ||
1708                  st->codec->codec_id == CODEC_ID_SIPR ||
1709                  st->codec->codec_id == CODEC_ID_ATRAC3) &&
1710                  st->codec->block_align && track->audio.sub_packet_size) {
1711                 int a = st->codec->block_align;
1712                 int sps = track->audio.sub_packet_size;
1713                 int cfs = track->audio.coded_framesize;
1714                 int h = track->audio.sub_packet_h;
1715                 int y = track->audio.sub_packet_cnt;
1716                 int w = track->audio.frame_size;
1717                 int x;
1718
1719                 if (!track->audio.pkt_cnt) {
1720                     if (st->codec->codec_id == CODEC_ID_RA_288)
1721                         for (x=0; x<h/2; x++)
1722                             memcpy(track->audio.buf+x*2*w+y*cfs,
1723                                    data+x*cfs, cfs);
1724                     else if (st->codec->codec_id == CODEC_ID_SIPR)
1725                         memcpy(track->audio.buf + y*w, data, w);
1726                     else
1727                         for (x=0; x<w/sps; x++)
1728                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1729
1730                     if (++track->audio.sub_packet_cnt >= h) {
1731                         if (st->codec->codec_id == CODEC_ID_SIPR)
1732                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
1733                         track->audio.sub_packet_cnt = 0;
1734                         track->audio.pkt_cnt = h*w / a;
1735                     }
1736                 }
1737                 while (track->audio.pkt_cnt) {
1738                     pkt = av_mallocz(sizeof(AVPacket));
1739                     av_new_packet(pkt, a);
1740                     memcpy(pkt->data, track->audio.buf
1741                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1742                     pkt->pos = pos;
1743                     pkt->stream_index = st->index;
1744                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1745                 }
1746             } else {
1747                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1748                 int offset = 0, pkt_size = lace_size[n];
1749                 uint8_t *pkt_data = data;
1750
1751                 if (pkt_size > size) {
1752                     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
1753                     break;
1754                 }
1755
1756                 if (encodings && encodings->scope & 1) {
1757                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1758                     if (offset < 0)
1759                         continue;
1760                 }
1761
1762                 pkt = av_mallocz(sizeof(AVPacket));
1763                 /* XXX: prevent data copy... */
1764                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1765                     av_free(pkt);
1766                     res = AVERROR(ENOMEM);
1767                     break;
1768                 }
1769                 if (offset)
1770                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1771                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1772
1773                 if (pkt_data != data)
1774                     av_free(pkt_data);
1775
1776                 if (n == 0)
1777                     pkt->flags = is_keyframe;
1778                 pkt->stream_index = st->index;
1779
1780                 if (track->ms_compat)
1781                     pkt->dts = timecode;
1782                 else
1783                     pkt->pts = timecode;
1784                 pkt->pos = pos;
1785                 if (st->codec->codec_id == CODEC_ID_TEXT)
1786                     pkt->convergence_duration = duration;
1787                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1788                     pkt->duration = duration;
1789
1790                 if (st->codec->codec_id == CODEC_ID_SSA)
1791                     matroska_fix_ass_packet(matroska, pkt, duration);
1792
1793                 if (matroska->prev_pkt &&
1794                     timecode != AV_NOPTS_VALUE &&
1795                     matroska->prev_pkt->pts == timecode &&
1796                     matroska->prev_pkt->stream_index == st->index &&
1797                     st->codec->codec_id == CODEC_ID_SSA)
1798                     matroska_merge_packets(matroska->prev_pkt, pkt);
1799                 else {
1800                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1801                     matroska->prev_pkt = pkt;
1802                 }
1803             }
1804
1805             if (timecode != AV_NOPTS_VALUE)
1806                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1807             data += lace_size[n];
1808             size -= lace_size[n];
1809         }
1810     }
1811
1812     av_free(lace_size);
1813     return res;
1814 }
1815
1816 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1817 {
1818     MatroskaCluster cluster = { 0 };
1819     EbmlList *blocks_list;
1820     MatroskaBlock *blocks;
1821     int i, res;
1822     int64_t pos = url_ftell(matroska->ctx->pb);
1823     matroska->prev_pkt = NULL;
1824     if (matroska->current_id)
1825         pos -= 4;  /* sizeof the ID which was already read */
1826     res = ebml_parse(matroska, matroska_clusters, &cluster);
1827     blocks_list = &cluster.blocks;
1828     blocks = blocks_list->elem;
1829     for (i=0; i<blocks_list->nb_elem; i++)
1830         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
1831             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
1832             res=matroska_parse_block(matroska,
1833                                      blocks[i].bin.data, blocks[i].bin.size,
1834                                      blocks[i].bin.pos,  cluster.timecode,
1835                                      blocks[i].duration, is_keyframe,
1836                                      pos);
1837         }
1838     ebml_free(matroska_cluster, &cluster);
1839     if (res < 0)  matroska->done = 1;
1840     return res;
1841 }
1842
1843 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1844 {
1845     MatroskaDemuxContext *matroska = s->priv_data;
1846
1847     while (matroska_deliver_packet(matroska, pkt)) {
1848         if (matroska->done)
1849             return AVERROR_EOF;
1850         matroska_parse_cluster(matroska);
1851     }
1852
1853     return 0;
1854 }
1855
1856 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1857                               int64_t timestamp, int flags)
1858 {
1859     MatroskaDemuxContext *matroska = s->priv_data;
1860     MatroskaTrack *tracks = matroska->tracks.elem;
1861     AVStream *st = s->streams[stream_index];
1862     int i, index, index_sub, index_min;
1863
1864     if (!st->nb_index_entries)
1865         return 0;
1866     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
1867
1868     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1869         url_fseek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
1870         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1871             matroska_clear_queue(matroska);
1872             if (matroska_parse_cluster(matroska) < 0)
1873                 break;
1874         }
1875     }
1876
1877     matroska_clear_queue(matroska);
1878     if (index < 0)
1879         return 0;
1880
1881     index_min = index;
1882     for (i=0; i < matroska->tracks.nb_elem; i++) {
1883         tracks[i].end_timecode = 0;
1884         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
1885             && !tracks[i].stream->discard != AVDISCARD_ALL) {
1886             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
1887             if (index_sub >= 0
1888                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
1889                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
1890                 index_min = index_sub;
1891         }
1892     }
1893
1894     url_fseek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
1895     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
1896     matroska->skip_to_timecode = st->index_entries[index].timestamp;
1897     matroska->done = 0;
1898     av_update_cur_dts(s, st, st->index_entries[index].timestamp);
1899     return 0;
1900 }
1901
1902 static int matroska_read_close(AVFormatContext *s)
1903 {
1904     MatroskaDemuxContext *matroska = s->priv_data;
1905     MatroskaTrack *tracks = matroska->tracks.elem;
1906     int n;
1907
1908     matroska_clear_queue(matroska);
1909
1910     for (n=0; n < matroska->tracks.nb_elem; n++)
1911         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
1912             av_free(tracks[n].audio.buf);
1913     ebml_free(matroska_segment, matroska);
1914
1915     return 0;
1916 }
1917
1918 AVInputFormat matroska_demuxer = {
1919     "matroska,webm",
1920     NULL_IF_CONFIG_SMALL("Matroska/WebM file format"),
1921     sizeof(MatroskaDemuxContext),
1922     matroska_probe,
1923     matroska_read_header,
1924     matroska_read_packet,
1925     matroska_read_close,
1926     matroska_read_seek,
1927 };