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