mkv: report average framerate as minimal as well
[ffmpeg.git] / libavformat / matroskadec.c
1 /*
2  * Matroska file demuxer
3  * Copyright (c) 2003-2008 The Libav Project
4  *
5  * This file is part of Libav.
6  *
7  * Libav 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  * Libav 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 Libav; 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  * @author Ronald Bultje <rbultje@ronald.bitfreak.net>
26  * @author with a little help from Moritz Bunkus <moritz@bunkus.org>
27  * @author totally reworked by Aurelien Jacobs <aurel@gnuage.org>
28  * @see specs available on the Matroska project page: http://www.matroska.org/
29  */
30
31 #include <stdio.h>
32 #include "avformat.h"
33 #include "internal.h"
34 #include "avio_internal.h"
35 /* For ff_codec_get_id(). */
36 #include "riff.h"
37 #include "isom.h"
38 #include "rm.h"
39 #include "matroska.h"
40 #include "libavcodec/mpeg4audio.h"
41 #include "libavutil/intfloat.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/avstring.h"
44 #include "libavutil/lzo.h"
45 #include "libavutil/dict.h"
46 #if CONFIG_ZLIB
47 #include <zlib.h>
48 #endif
49 #if CONFIG_BZLIB
50 #include <bzlib.h>
51 #endif
52
53 typedef enum {
54     EBML_NONE,
55     EBML_UINT,
56     EBML_FLOAT,
57     EBML_STR,
58     EBML_UTF8,
59     EBML_BIN,
60     EBML_NEST,
61     EBML_PASS,
62     EBML_STOP,
63     EBML_TYPE_COUNT
64 } EbmlType;
65
66 typedef const struct EbmlSyntax {
67     uint32_t id;
68     EbmlType type;
69     int list_elem_size;
70     int data_offset;
71     union {
72         uint64_t    u;
73         double      f;
74         const char *s;
75         const struct EbmlSyntax *n;
76     } def;
77 } EbmlSyntax;
78
79 typedef struct {
80     int nb_elem;
81     void *elem;
82 } EbmlList;
83
84 typedef struct {
85     int      size;
86     uint8_t *data;
87     int64_t  pos;
88 } EbmlBin;
89
90 typedef struct {
91     uint64_t version;
92     uint64_t max_size;
93     uint64_t id_length;
94     char    *doctype;
95     uint64_t doctype_version;
96 } Ebml;
97
98 typedef struct {
99     uint64_t algo;
100     EbmlBin  settings;
101 } MatroskaTrackCompression;
102
103 typedef struct {
104     uint64_t scope;
105     uint64_t type;
106     MatroskaTrackCompression compression;
107 } MatroskaTrackEncoding;
108
109 typedef struct {
110     double   frame_rate;
111     uint64_t display_width;
112     uint64_t display_height;
113     uint64_t pixel_width;
114     uint64_t pixel_height;
115     uint64_t fourcc;
116 } MatroskaTrackVideo;
117
118 typedef struct {
119     double   samplerate;
120     double   out_samplerate;
121     uint64_t bitdepth;
122     uint64_t channels;
123
124     /* real audio header (extracted from extradata) */
125     int      coded_framesize;
126     int      sub_packet_h;
127     int      frame_size;
128     int      sub_packet_size;
129     int      sub_packet_cnt;
130     int      pkt_cnt;
131     uint64_t buf_timecode;
132     uint8_t *buf;
133 } MatroskaTrackAudio;
134
135 typedef struct {
136     uint64_t num;
137     uint64_t uid;
138     uint64_t type;
139     char    *name;
140     char    *codec_id;
141     EbmlBin  codec_priv;
142     char    *language;
143     double time_scale;
144     uint64_t default_duration;
145     uint64_t flag_default;
146     uint64_t flag_forced;
147     MatroskaTrackVideo video;
148     MatroskaTrackAudio audio;
149     EbmlList encodings;
150
151     AVStream *stream;
152     int64_t end_timecode;
153     int ms_compat;
154 } MatroskaTrack;
155
156 typedef struct {
157     uint64_t uid;
158     char *filename;
159     char *mime;
160     EbmlBin bin;
161
162     AVStream *stream;
163 } MatroskaAttachement;
164
165 typedef struct {
166     uint64_t start;
167     uint64_t end;
168     uint64_t uid;
169     char    *title;
170
171     AVChapter *chapter;
172 } MatroskaChapter;
173
174 typedef struct {
175     uint64_t track;
176     uint64_t pos;
177 } MatroskaIndexPos;
178
179 typedef struct {
180     uint64_t time;
181     EbmlList pos;
182 } MatroskaIndex;
183
184 typedef struct {
185     char *name;
186     char *string;
187     char *lang;
188     uint64_t def;
189     EbmlList sub;
190 } MatroskaTag;
191
192 typedef struct {
193     char    *type;
194     uint64_t typevalue;
195     uint64_t trackuid;
196     uint64_t chapteruid;
197     uint64_t attachuid;
198 } MatroskaTagTarget;
199
200 typedef struct {
201     MatroskaTagTarget target;
202     EbmlList tag;
203 } MatroskaTags;
204
205 typedef struct {
206     uint64_t id;
207     uint64_t pos;
208 } MatroskaSeekhead;
209
210 typedef struct {
211     uint64_t start;
212     uint64_t length;
213 } MatroskaLevel;
214
215 typedef struct {
216     AVFormatContext *ctx;
217
218     /* EBML stuff */
219     int num_levels;
220     MatroskaLevel levels[EBML_MAX_DEPTH];
221     int level_up;
222     uint32_t current_id;
223
224     uint64_t time_scale;
225     double   duration;
226     char    *title;
227     EbmlList tracks;
228     EbmlList attachments;
229     EbmlList chapters;
230     EbmlList index;
231     EbmlList tags;
232     EbmlList seekhead;
233
234     /* byte position of the segment inside the stream */
235     int64_t segment_start;
236
237     /* the packet queue */
238     AVPacket **packets;
239     int num_packets;
240     AVPacket *prev_pkt;
241
242     int done;
243
244     /* What to skip before effectively reading a packet. */
245     int skip_to_keyframe;
246     uint64_t skip_to_timecode;
247
248     /* File has a CUES element, but we defer parsing until it is needed. */
249     int cues_parsing_deferred;
250 } MatroskaDemuxContext;
251
252 typedef struct {
253     uint64_t duration;
254     int64_t  reference;
255     uint64_t non_simple;
256     EbmlBin  bin;
257 } MatroskaBlock;
258
259 typedef struct {
260     uint64_t timecode;
261     EbmlList blocks;
262 } MatroskaCluster;
263
264 static EbmlSyntax ebml_header[] = {
265     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
266     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
267     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
268     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
269     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
270     { EBML_ID_EBMLVERSION,            EBML_NONE },
271     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
272     { 0 }
273 };
274
275 static EbmlSyntax ebml_syntax[] = {
276     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
277     { 0 }
278 };
279
280 static EbmlSyntax matroska_info[] = {
281     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
282     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
283     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
284     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
285     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
286     { MATROSKA_ID_DATEUTC,            EBML_NONE },
287     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
288     { 0 }
289 };
290
291 static EbmlSyntax matroska_track_video[] = {
292     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
293     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
294     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
295     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
296     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
297     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,fourcc) },
298     { MATROSKA_ID_VIDEOPIXELCROPB,    EBML_NONE },
299     { MATROSKA_ID_VIDEOPIXELCROPT,    EBML_NONE },
300     { MATROSKA_ID_VIDEOPIXELCROPL,    EBML_NONE },
301     { MATROSKA_ID_VIDEOPIXELCROPR,    EBML_NONE },
302     { MATROSKA_ID_VIDEODISPLAYUNIT,   EBML_NONE },
303     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
304     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_NONE },
305     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
306     { 0 }
307 };
308
309 static EbmlSyntax matroska_track_audio[] = {
310     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
311     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
312     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
313     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
314     { 0 }
315 };
316
317 static EbmlSyntax matroska_track_encoding_compression[] = {
318     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
319     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
320     { 0 }
321 };
322
323 static EbmlSyntax matroska_track_encoding[] = {
324     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
325     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
326     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
327     { MATROSKA_ID_ENCODINGORDER,      EBML_NONE },
328     { 0 }
329 };
330
331 static EbmlSyntax matroska_track_encodings[] = {
332     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
333     { 0 }
334 };
335
336 static EbmlSyntax matroska_track[] = {
337     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
338     { MATROSKA_ID_TRACKNAME,            EBML_UTF8, 0, offsetof(MatroskaTrack,name) },
339     { MATROSKA_ID_TRACKUID,             EBML_UINT, 0, offsetof(MatroskaTrack,uid) },
340     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
341     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
342     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
343     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
344     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
345     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
346     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
347     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_UINT, 0, offsetof(MatroskaTrack,flag_forced), {.u=0} },
348     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
349     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
350     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
351     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
352     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
353     { MATROSKA_ID_CODECNAME,            EBML_NONE },
354     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
355     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
356     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
357     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
358     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
359     { MATROSKA_ID_TRACKMAXBLKADDID,     EBML_NONE },
360     { 0 }
361 };
362
363 static EbmlSyntax matroska_tracks[] = {
364     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
365     { 0 }
366 };
367
368 static EbmlSyntax matroska_attachment[] = {
369     { MATROSKA_ID_FILEUID,            EBML_UINT, 0, offsetof(MatroskaAttachement,uid) },
370     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
371     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
372     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
373     { MATROSKA_ID_FILEDESC,           EBML_NONE },
374     { 0 }
375 };
376
377 static EbmlSyntax matroska_attachments[] = {
378     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
379     { 0 }
380 };
381
382 static EbmlSyntax matroska_chapter_display[] = {
383     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
384     { MATROSKA_ID_CHAPLANG,           EBML_NONE },
385     { 0 }
386 };
387
388 static EbmlSyntax matroska_chapter_entry[] = {
389     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
390     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
391     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
392     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
393     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
394     { MATROSKA_ID_CHAPTERFLAGENABLED, EBML_NONE },
395     { MATROSKA_ID_CHAPTERPHYSEQUIV,   EBML_NONE },
396     { MATROSKA_ID_CHAPTERATOM,        EBML_NONE },
397     { 0 }
398 };
399
400 static EbmlSyntax matroska_chapter[] = {
401     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
402     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
403     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
404     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
405     { MATROSKA_ID_EDITIONFLAGORDERED, EBML_NONE },
406     { 0 }
407 };
408
409 static EbmlSyntax matroska_chapters[] = {
410     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
411     { 0 }
412 };
413
414 static EbmlSyntax matroska_index_pos[] = {
415     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
416     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
417     { MATROSKA_ID_CUEBLOCKNUMBER,     EBML_NONE },
418     { 0 }
419 };
420
421 static EbmlSyntax matroska_index_entry[] = {
422     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
423     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
424     { 0 }
425 };
426
427 static EbmlSyntax matroska_index[] = {
428     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
429     { 0 }
430 };
431
432 static EbmlSyntax matroska_simpletag[] = {
433     { MATROSKA_ID_TAGNAME,            EBML_UTF8, 0, offsetof(MatroskaTag,name) },
434     { MATROSKA_ID_TAGSTRING,          EBML_UTF8, 0, offsetof(MatroskaTag,string) },
435     { MATROSKA_ID_TAGLANG,            EBML_STR,  0, offsetof(MatroskaTag,lang), {.s="und"} },
436     { MATROSKA_ID_TAGDEFAULT,         EBML_UINT, 0, offsetof(MatroskaTag,def) },
437     { MATROSKA_ID_TAGDEFAULT_BUG,     EBML_UINT, 0, offsetof(MatroskaTag,def) },
438     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTag,sub), {.n=matroska_simpletag} },
439     { 0 }
440 };
441
442 static EbmlSyntax matroska_tagtargets[] = {
443     { MATROSKA_ID_TAGTARGETS_TYPE,      EBML_STR,  0, offsetof(MatroskaTagTarget,type) },
444     { MATROSKA_ID_TAGTARGETS_TYPEVALUE, EBML_UINT, 0, offsetof(MatroskaTagTarget,typevalue), {.u=50} },
445     { MATROSKA_ID_TAGTARGETS_TRACKUID,  EBML_UINT, 0, offsetof(MatroskaTagTarget,trackuid) },
446     { MATROSKA_ID_TAGTARGETS_CHAPTERUID,EBML_UINT, 0, offsetof(MatroskaTagTarget,chapteruid) },
447     { MATROSKA_ID_TAGTARGETS_ATTACHUID, EBML_UINT, 0, offsetof(MatroskaTagTarget,attachuid) },
448     { 0 }
449 };
450
451 static EbmlSyntax matroska_tag[] = {
452     { MATROSKA_ID_SIMPLETAG,          EBML_NEST, sizeof(MatroskaTag), offsetof(MatroskaTags,tag), {.n=matroska_simpletag} },
453     { MATROSKA_ID_TAGTARGETS,         EBML_NEST, 0, offsetof(MatroskaTags,target), {.n=matroska_tagtargets} },
454     { 0 }
455 };
456
457 static EbmlSyntax matroska_tags[] = {
458     { MATROSKA_ID_TAG,                EBML_NEST, sizeof(MatroskaTags), offsetof(MatroskaDemuxContext,tags), {.n=matroska_tag} },
459     { 0 }
460 };
461
462 static EbmlSyntax matroska_seekhead_entry[] = {
463     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
464     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
465     { 0 }
466 };
467
468 static EbmlSyntax matroska_seekhead[] = {
469     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
470     { 0 }
471 };
472
473 static EbmlSyntax matroska_segment[] = {
474     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
475     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
476     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
477     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
478     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
479     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
480     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
481     { MATROSKA_ID_CLUSTER,        EBML_STOP },
482     { 0 }
483 };
484
485 static EbmlSyntax matroska_segments[] = {
486     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
487     { 0 }
488 };
489
490 static EbmlSyntax matroska_blockgroup[] = {
491     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
492     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
493     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration), {.u=AV_NOPTS_VALUE} },
494     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
495     { 1,                          EBML_UINT, 0, offsetof(MatroskaBlock,non_simple), {.u=1} },
496     { 0 }
497 };
498
499 static EbmlSyntax matroska_cluster[] = {
500     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
501     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
502     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
503     { MATROSKA_ID_CLUSTERPOSITION,EBML_NONE },
504     { MATROSKA_ID_CLUSTERPREVSIZE,EBML_NONE },
505     { 0 }
506 };
507
508 static EbmlSyntax matroska_clusters[] = {
509     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
510     { MATROSKA_ID_INFO,           EBML_NONE },
511     { MATROSKA_ID_CUES,           EBML_NONE },
512     { MATROSKA_ID_TAGS,           EBML_NONE },
513     { MATROSKA_ID_SEEKHEAD,       EBML_NONE },
514     { 0 }
515 };
516
517 static const char *const matroska_doctypes[] = { "matroska", "webm" };
518
519 /*
520  * Return: Whether we reached the end of a level in the hierarchy or not.
521  */
522 static int ebml_level_end(MatroskaDemuxContext *matroska)
523 {
524     AVIOContext *pb = matroska->ctx->pb;
525     int64_t pos = avio_tell(pb);
526
527     if (matroska->num_levels > 0) {
528         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
529         if (pos - level->start >= level->length || matroska->current_id) {
530             matroska->num_levels--;
531             return 1;
532         }
533     }
534     return 0;
535 }
536
537 /*
538  * Read: an "EBML number", which is defined as a variable-length
539  * array of bytes. The first byte indicates the length by giving a
540  * number of 0-bits followed by a one. The position of the first
541  * "one" bit inside the first byte indicates the length of this
542  * number.
543  * Returns: number of bytes read, < 0 on error
544  */
545 static int ebml_read_num(MatroskaDemuxContext *matroska, AVIOContext *pb,
546                          int max_size, uint64_t *number)
547 {
548     int read = 1, n = 1;
549     uint64_t total = 0;
550
551     /* The first byte tells us the length in bytes - avio_r8() can normally
552      * return 0, but since that's not a valid first ebmlID byte, we can
553      * use it safely here to catch EOS. */
554     if (!(total = avio_r8(pb))) {
555         /* we might encounter EOS here */
556         if (!pb->eof_reached) {
557             int64_t pos = avio_tell(pb);
558             av_log(matroska->ctx, AV_LOG_ERROR,
559                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
560                    pos, pos);
561         }
562         return AVERROR(EIO); /* EOS or actual I/O error */
563     }
564
565     /* get the length of the EBML number */
566     read = 8 - ff_log2_tab[total];
567     if (read > max_size) {
568         int64_t pos = avio_tell(pb) - 1;
569         av_log(matroska->ctx, AV_LOG_ERROR,
570                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
571                (uint8_t) total, pos, pos);
572         return AVERROR_INVALIDDATA;
573     }
574
575     /* read out length */
576     total ^= 1 << ff_log2_tab[total];
577     while (n++ < read)
578         total = (total << 8) | avio_r8(pb);
579
580     *number = total;
581
582     return read;
583 }
584
585 /**
586  * Read a EBML length value.
587  * This needs special handling for the "unknown length" case which has multiple
588  * encodings.
589  */
590 static int ebml_read_length(MatroskaDemuxContext *matroska, AVIOContext *pb,
591                             uint64_t *number)
592 {
593     int res = ebml_read_num(matroska, pb, 8, number);
594     if (res > 0 && *number + 1 == 1ULL << (7 * res))
595         *number = 0xffffffffffffffULL;
596     return res;
597 }
598
599 /*
600  * Read the next element as an unsigned int.
601  * 0 is success, < 0 is failure.
602  */
603 static int ebml_read_uint(AVIOContext *pb, int size, uint64_t *num)
604 {
605     int n = 0;
606
607     if (size > 8)
608         return AVERROR_INVALIDDATA;
609
610     /* big-endian ordering; build up number */
611     *num = 0;
612     while (n++ < size)
613         *num = (*num << 8) | avio_r8(pb);
614
615     return 0;
616 }
617
618 /*
619  * Read the next element as a float.
620  * 0 is success, < 0 is failure.
621  */
622 static int ebml_read_float(AVIOContext *pb, int size, double *num)
623 {
624     if (size == 0) {
625         *num = 0;
626     } else if (size == 4) {
627         *num = av_int2float(avio_rb32(pb));
628     } else if (size == 8){
629         *num = av_int2double(avio_rb64(pb));
630     } else
631         return AVERROR_INVALIDDATA;
632
633     return 0;
634 }
635
636 /*
637  * Read the next element as an ASCII string.
638  * 0 is success, < 0 is failure.
639  */
640 static int ebml_read_ascii(AVIOContext *pb, int size, char **str)
641 {
642     char *res;
643
644     /* EBML strings are usually not 0-terminated, so we allocate one
645      * byte more, read the string and NULL-terminate it ourselves. */
646     if (!(res = av_malloc(size + 1)))
647         return AVERROR(ENOMEM);
648     if (avio_read(pb, (uint8_t *) res, size) != size) {
649         av_free(res);
650         return AVERROR(EIO);
651     }
652     (res)[size] = '\0';
653     av_free(*str);
654     *str = res;
655
656     return 0;
657 }
658
659 /*
660  * Read the next element as binary data.
661  * 0 is success, < 0 is failure.
662  */
663 static int ebml_read_binary(AVIOContext *pb, int length, EbmlBin *bin)
664 {
665     av_free(bin->data);
666     if (!(bin->data = av_malloc(length)))
667         return AVERROR(ENOMEM);
668
669     bin->size = length;
670     bin->pos  = avio_tell(pb);
671     if (avio_read(pb, bin->data, length) != length) {
672         av_freep(&bin->data);
673         return AVERROR(EIO);
674     }
675
676     return 0;
677 }
678
679 /*
680  * Read the next element, but only the header. The contents
681  * are supposed to be sub-elements which can be read separately.
682  * 0 is success, < 0 is failure.
683  */
684 static int ebml_read_master(MatroskaDemuxContext *matroska, uint64_t length)
685 {
686     AVIOContext *pb = matroska->ctx->pb;
687     MatroskaLevel *level;
688
689     if (matroska->num_levels >= EBML_MAX_DEPTH) {
690         av_log(matroska->ctx, AV_LOG_ERROR,
691                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
692         return AVERROR(ENOSYS);
693     }
694
695     level = &matroska->levels[matroska->num_levels++];
696     level->start = avio_tell(pb);
697     level->length = length;
698
699     return 0;
700 }
701
702 /*
703  * Read signed/unsigned "EBML" numbers.
704  * Return: number of bytes processed, < 0 on error
705  */
706 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
707                                  uint8_t *data, uint32_t size, uint64_t *num)
708 {
709     AVIOContext pb;
710     ffio_init_context(&pb, data, size, 0, NULL, NULL, NULL, NULL);
711     return ebml_read_num(matroska, &pb, FFMIN(size, 8), num);
712 }
713
714 /*
715  * Same as above, but signed.
716  */
717 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
718                                  uint8_t *data, uint32_t size, int64_t *num)
719 {
720     uint64_t unum;
721     int res;
722
723     /* read as unsigned number first */
724     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
725         return res;
726
727     /* make signed (weird way) */
728     *num = unum - ((1LL << (7*res - 1)) - 1);
729
730     return res;
731 }
732
733 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
734                            EbmlSyntax *syntax, void *data);
735
736 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
737                          uint32_t id, void *data)
738 {
739     int i;
740     for (i=0; syntax[i].id; i++)
741         if (id == syntax[i].id)
742             break;
743     if (!syntax[i].id && id == MATROSKA_ID_CLUSTER &&
744         matroska->num_levels > 0 &&
745         matroska->levels[matroska->num_levels-1].length == 0xffffffffffffff)
746         return 0;  // we reached the end of an unknown size cluster
747     if (!syntax[i].id && id != EBML_ID_VOID && id != EBML_ID_CRC32)
748         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
749     return ebml_parse_elem(matroska, &syntax[i], data);
750 }
751
752 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
753                       void *data)
754 {
755     if (!matroska->current_id) {
756         uint64_t id;
757         int res = ebml_read_num(matroska, matroska->ctx->pb, 4, &id);
758         if (res < 0)
759             return res;
760         matroska->current_id = id | 1 << 7*res;
761     }
762     return ebml_parse_id(matroska, syntax, matroska->current_id, data);
763 }
764
765 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
766                            void *data)
767 {
768     int i, res = 0;
769
770     for (i=0; syntax[i].id; i++)
771         switch (syntax[i].type) {
772         case EBML_UINT:
773             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
774             break;
775         case EBML_FLOAT:
776             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
777             break;
778         case EBML_STR:
779         case EBML_UTF8:
780             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
781             break;
782         }
783
784     while (!res && !ebml_level_end(matroska))
785         res = ebml_parse(matroska, syntax, data);
786
787     return res;
788 }
789
790 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
791                            EbmlSyntax *syntax, void *data)
792 {
793     static const uint64_t max_lengths[EBML_TYPE_COUNT] = {
794         [EBML_UINT]  = 8,
795         [EBML_FLOAT] = 8,
796         // max. 16 MB for strings
797         [EBML_STR]   = 0x1000000,
798         [EBML_UTF8]  = 0x1000000,
799         // max. 256 MB for binary data
800         [EBML_BIN]   = 0x10000000,
801         // no limits for anything else
802     };
803     AVIOContext *pb = matroska->ctx->pb;
804     uint32_t id = syntax->id;
805     uint64_t length;
806     int res;
807     void *newelem;
808
809     data = (char *)data + syntax->data_offset;
810     if (syntax->list_elem_size) {
811         EbmlList *list = data;
812         newelem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
813         if (!newelem)
814             return AVERROR(ENOMEM);
815         list->elem = newelem;
816         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
817         memset(data, 0, syntax->list_elem_size);
818         list->nb_elem++;
819     }
820
821     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP) {
822         matroska->current_id = 0;
823         if ((res = ebml_read_length(matroska, pb, &length)) < 0)
824             return res;
825         if (max_lengths[syntax->type] && length > max_lengths[syntax->type]) {
826             av_log(matroska->ctx, AV_LOG_ERROR,
827                    "Invalid length 0x%"PRIx64" > 0x%"PRIx64" for syntax element %i\n",
828                    length, max_lengths[syntax->type], syntax->type);
829             return AVERROR_INVALIDDATA;
830         }
831     }
832
833     switch (syntax->type) {
834     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
835     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
836     case EBML_STR:
837     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
838     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
839     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
840                          return res;
841                      if (id == MATROSKA_ID_SEGMENT)
842                          matroska->segment_start = avio_tell(matroska->ctx->pb);
843                      return ebml_parse_nest(matroska, syntax->def.n, data);
844     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
845     case EBML_STOP:  return 1;
846     default:         return avio_skip(pb,length)<0 ? AVERROR(EIO) : 0;
847     }
848     if (res == AVERROR_INVALIDDATA)
849         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
850     else if (res == AVERROR(EIO))
851         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
852     return res;
853 }
854
855 static void ebml_free(EbmlSyntax *syntax, void *data)
856 {
857     int i, j;
858     for (i=0; syntax[i].id; i++) {
859         void *data_off = (char *)data + syntax[i].data_offset;
860         switch (syntax[i].type) {
861         case EBML_STR:
862         case EBML_UTF8:  av_freep(data_off);                      break;
863         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
864         case EBML_NEST:
865             if (syntax[i].list_elem_size) {
866                 EbmlList *list = data_off;
867                 char *ptr = list->elem;
868                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
869                     ebml_free(syntax[i].def.n, ptr);
870                 av_free(list->elem);
871             } else
872                 ebml_free(syntax[i].def.n, data_off);
873         default:  break;
874         }
875     }
876 }
877
878
879 /*
880  * Autodetecting...
881  */
882 static int matroska_probe(AVProbeData *p)
883 {
884     uint64_t total = 0;
885     int len_mask = 0x80, size = 1, n = 1, i;
886
887     /* EBML header? */
888     if (AV_RB32(p->buf) != EBML_ID_HEADER)
889         return 0;
890
891     /* length of header */
892     total = p->buf[4];
893     while (size <= 8 && !(total & len_mask)) {
894         size++;
895         len_mask >>= 1;
896     }
897     if (size > 8)
898       return 0;
899     total &= (len_mask - 1);
900     while (n < size)
901         total = (total << 8) | p->buf[4 + n++];
902
903     /* Does the probe data contain the whole header? */
904     if (p->buf_size < 4 + size + total)
905       return 0;
906
907     /* The header should contain a known document type. For now,
908      * we don't parse the whole header but simply check for the
909      * availability of that array of characters inside the header.
910      * Not fully fool-proof, but good enough. */
911     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++) {
912         int probelen = strlen(matroska_doctypes[i]);
913         if (total < probelen)
914             continue;
915         for (n = 4+size; n <= 4+size+total-probelen; n++)
916             if (!memcmp(p->buf+n, matroska_doctypes[i], probelen))
917                 return AVPROBE_SCORE_MAX;
918     }
919
920     // probably valid EBML header but no recognized doctype
921     return AVPROBE_SCORE_MAX/2;
922 }
923
924 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
925                                                  int num)
926 {
927     MatroskaTrack *tracks = matroska->tracks.elem;
928     int i;
929
930     for (i=0; i < matroska->tracks.nb_elem; i++)
931         if (tracks[i].num == num)
932             return &tracks[i];
933
934     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
935     return NULL;
936 }
937
938 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
939                                   MatroskaTrack *track)
940 {
941     MatroskaTrackEncoding *encodings = track->encodings.elem;
942     uint8_t* data = *buf;
943     int isize = *buf_size;
944     uint8_t* pkt_data = NULL;
945     uint8_t av_unused *newpktdata;
946     int pkt_size = isize;
947     int result = 0;
948     int olen;
949
950     if (pkt_size >= 10000000)
951         return -1;
952
953     switch (encodings[0].compression.algo) {
954     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
955         return encodings[0].compression.settings.size;
956     case MATROSKA_TRACK_ENCODING_COMP_LZO:
957         do {
958             olen = pkt_size *= 3;
959             pkt_data = av_realloc(pkt_data, pkt_size+AV_LZO_OUTPUT_PADDING);
960             result = av_lzo1x_decode(pkt_data, &olen, data, &isize);
961         } while (result==AV_LZO_OUTPUT_FULL && pkt_size<10000000);
962         if (result)
963             goto failed;
964         pkt_size -= olen;
965         break;
966 #if CONFIG_ZLIB
967     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
968         z_stream zstream = {0};
969         if (inflateInit(&zstream) != Z_OK)
970             return -1;
971         zstream.next_in = data;
972         zstream.avail_in = isize;
973         do {
974             pkt_size *= 3;
975             newpktdata = av_realloc(pkt_data, pkt_size);
976             if (!newpktdata) {
977                 inflateEnd(&zstream);
978                 goto failed;
979             }
980             pkt_data = newpktdata;
981             zstream.avail_out = pkt_size - zstream.total_out;
982             zstream.next_out = pkt_data + zstream.total_out;
983             result = inflate(&zstream, Z_NO_FLUSH);
984         } while (result==Z_OK && pkt_size<10000000);
985         pkt_size = zstream.total_out;
986         inflateEnd(&zstream);
987         if (result != Z_STREAM_END)
988             goto failed;
989         break;
990     }
991 #endif
992 #if CONFIG_BZLIB
993     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
994         bz_stream bzstream = {0};
995         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
996             return -1;
997         bzstream.next_in = data;
998         bzstream.avail_in = isize;
999         do {
1000             pkt_size *= 3;
1001             newpktdata = av_realloc(pkt_data, pkt_size);
1002             if (!newpktdata) {
1003                 BZ2_bzDecompressEnd(&bzstream);
1004                 goto failed;
1005             }
1006             pkt_data = newpktdata;
1007             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
1008             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
1009             result = BZ2_bzDecompress(&bzstream);
1010         } while (result==BZ_OK && pkt_size<10000000);
1011         pkt_size = bzstream.total_out_lo32;
1012         BZ2_bzDecompressEnd(&bzstream);
1013         if (result != BZ_STREAM_END)
1014             goto failed;
1015         break;
1016     }
1017 #endif
1018     default:
1019         return -1;
1020     }
1021
1022     *buf = pkt_data;
1023     *buf_size = pkt_size;
1024     return 0;
1025  failed:
1026     av_free(pkt_data);
1027     return -1;
1028 }
1029
1030 static void matroska_fix_ass_packet(MatroskaDemuxContext *matroska,
1031                                     AVPacket *pkt, uint64_t display_duration)
1032 {
1033     char *line, *layer, *ptr = pkt->data, *end = ptr+pkt->size;
1034     for (; *ptr!=',' && ptr<end-1; ptr++);
1035     if (*ptr == ',')
1036         layer = ++ptr;
1037     for (; *ptr!=',' && ptr<end-1; ptr++);
1038     if (*ptr == ',') {
1039         int64_t end_pts = pkt->pts + display_duration;
1040         int sc = matroska->time_scale * pkt->pts / 10000000;
1041         int ec = matroska->time_scale * end_pts  / 10000000;
1042         int sh, sm, ss, eh, em, es, len;
1043         sh = sc/360000;  sc -= 360000*sh;
1044         sm = sc/  6000;  sc -=   6000*sm;
1045         ss = sc/   100;  sc -=    100*ss;
1046         eh = ec/360000;  ec -= 360000*eh;
1047         em = ec/  6000;  ec -=   6000*em;
1048         es = ec/   100;  ec -=    100*es;
1049         *ptr++ = '\0';
1050         len = 50 + end-ptr + FF_INPUT_BUFFER_PADDING_SIZE;
1051         if (!(line = av_malloc(len)))
1052             return;
1053         snprintf(line,len,"Dialogue: %s,%d:%02d:%02d.%02d,%d:%02d:%02d.%02d,%s\r\n",
1054                  layer, sh, sm, ss, sc, eh, em, es, ec, ptr);
1055         av_free(pkt->data);
1056         pkt->data = line;
1057         pkt->size = strlen(line);
1058     }
1059 }
1060
1061 static int matroska_merge_packets(AVPacket *out, AVPacket *in)
1062 {
1063     void *newdata = av_realloc(out->data, out->size+in->size);
1064     if (!newdata)
1065         return AVERROR(ENOMEM);
1066     out->data = newdata;
1067     memcpy(out->data+out->size, in->data, in->size);
1068     out->size += in->size;
1069     av_destruct_packet(in);
1070     av_free(in);
1071     return 0;
1072 }
1073
1074 static void matroska_convert_tag(AVFormatContext *s, EbmlList *list,
1075                                  AVDictionary **metadata, char *prefix)
1076 {
1077     MatroskaTag *tags = list->elem;
1078     char key[1024];
1079     int i;
1080
1081     for (i=0; i < list->nb_elem; i++) {
1082         const char *lang = strcmp(tags[i].lang, "und") ? tags[i].lang : NULL;
1083
1084         if (!tags[i].name) {
1085             av_log(s, AV_LOG_WARNING, "Skipping invalid tag with no TagName.\n");
1086             continue;
1087         }
1088         if (prefix)  snprintf(key, sizeof(key), "%s/%s", prefix, tags[i].name);
1089         else         av_strlcpy(key, tags[i].name, sizeof(key));
1090         if (tags[i].def || !lang) {
1091         av_dict_set(metadata, key, tags[i].string, 0);
1092         if (tags[i].sub.nb_elem)
1093             matroska_convert_tag(s, &tags[i].sub, metadata, key);
1094         }
1095         if (lang) {
1096             av_strlcat(key, "-", sizeof(key));
1097             av_strlcat(key, lang, sizeof(key));
1098             av_dict_set(metadata, key, tags[i].string, 0);
1099             if (tags[i].sub.nb_elem)
1100                 matroska_convert_tag(s, &tags[i].sub, metadata, key);
1101         }
1102     }
1103     ff_metadata_conv(metadata, NULL, ff_mkv_metadata_conv);
1104 }
1105
1106 static void matroska_convert_tags(AVFormatContext *s)
1107 {
1108     MatroskaDemuxContext *matroska = s->priv_data;
1109     MatroskaTags *tags = matroska->tags.elem;
1110     int i, j;
1111
1112     for (i=0; i < matroska->tags.nb_elem; i++) {
1113         if (tags[i].target.attachuid) {
1114             MatroskaAttachement *attachment = matroska->attachments.elem;
1115             for (j=0; j<matroska->attachments.nb_elem; j++)
1116                 if (attachment[j].uid == tags[i].target.attachuid
1117                     && attachment[j].stream)
1118                     matroska_convert_tag(s, &tags[i].tag,
1119                                          &attachment[j].stream->metadata, NULL);
1120         } else if (tags[i].target.chapteruid) {
1121             MatroskaChapter *chapter = matroska->chapters.elem;
1122             for (j=0; j<matroska->chapters.nb_elem; j++)
1123                 if (chapter[j].uid == tags[i].target.chapteruid
1124                     && chapter[j].chapter)
1125                     matroska_convert_tag(s, &tags[i].tag,
1126                                          &chapter[j].chapter->metadata, NULL);
1127         } else if (tags[i].target.trackuid) {
1128             MatroskaTrack *track = matroska->tracks.elem;
1129             for (j=0; j<matroska->tracks.nb_elem; j++)
1130                 if (track[j].uid == tags[i].target.trackuid && track[j].stream)
1131                     matroska_convert_tag(s, &tags[i].tag,
1132                                          &track[j].stream->metadata, NULL);
1133         } else {
1134             matroska_convert_tag(s, &tags[i].tag, &s->metadata,
1135                                  tags[i].target.type);
1136         }
1137     }
1138 }
1139
1140 static int matroska_parse_seekhead_entry(MatroskaDemuxContext *matroska, int idx)
1141 {
1142     EbmlList *seekhead_list = &matroska->seekhead;
1143     MatroskaSeekhead *seekhead = seekhead_list->elem;
1144     uint32_t level_up = matroska->level_up;
1145     int64_t before_pos = avio_tell(matroska->ctx->pb);
1146     uint32_t saved_id = matroska->current_id;
1147     MatroskaLevel level;
1148     int64_t offset;
1149     int ret = 0;
1150
1151     if (idx >= seekhead_list->nb_elem
1152             || seekhead[idx].id == MATROSKA_ID_SEEKHEAD
1153             || seekhead[idx].id == MATROSKA_ID_CLUSTER)
1154         return 0;
1155
1156     /* seek */
1157     offset = seekhead[idx].pos + matroska->segment_start;
1158     if (avio_seek(matroska->ctx->pb, offset, SEEK_SET) == offset) {
1159         /* We don't want to lose our seekhead level, so we add
1160          * a dummy. This is a crude hack. */
1161         if (matroska->num_levels == EBML_MAX_DEPTH) {
1162             av_log(matroska->ctx, AV_LOG_INFO,
1163                    "Max EBML element depth (%d) reached, "
1164                    "cannot parse further.\n", EBML_MAX_DEPTH);
1165             ret = AVERROR_INVALIDDATA;
1166         } else {
1167             level.start = 0;
1168             level.length = (uint64_t)-1;
1169             matroska->levels[matroska->num_levels] = level;
1170             matroska->num_levels++;
1171             matroska->current_id = 0;
1172
1173             ret = ebml_parse(matroska, matroska_segment, matroska);
1174
1175             /* remove dummy level */
1176             while (matroska->num_levels) {
1177                 uint64_t length = matroska->levels[--matroska->num_levels].length;
1178                 if (length == (uint64_t)-1)
1179                     break;
1180             }
1181         }
1182     }
1183     /* seek back */
1184     avio_seek(matroska->ctx->pb, before_pos, SEEK_SET);
1185     matroska->level_up = level_up;
1186     matroska->current_id = saved_id;
1187
1188     return ret;
1189 }
1190
1191 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
1192 {
1193     EbmlList *seekhead_list = &matroska->seekhead;
1194     int64_t before_pos = avio_tell(matroska->ctx->pb);
1195     int i;
1196
1197     // we should not do any seeking in the streaming case
1198     if (!matroska->ctx->pb->seekable ||
1199         (matroska->ctx->flags & AVFMT_FLAG_IGNIDX))
1200         return;
1201
1202     for (i = 0; i < seekhead_list->nb_elem; i++) {
1203         MatroskaSeekhead *seekhead = seekhead_list->elem;
1204         if (seekhead[i].pos <= before_pos)
1205             continue;
1206
1207         // defer cues parsing until we actually need cue data.
1208         if (seekhead[i].id == MATROSKA_ID_CUES) {
1209             matroska->cues_parsing_deferred = 1;
1210             continue;
1211         }
1212
1213         if (matroska_parse_seekhead_entry(matroska, i) < 0)
1214             break;
1215     }
1216 }
1217
1218 static void matroska_parse_cues(MatroskaDemuxContext *matroska) {
1219     EbmlList *seekhead_list = &matroska->seekhead;
1220     MatroskaSeekhead *seekhead = seekhead_list->elem;
1221     EbmlList *index_list;
1222     MatroskaIndex *index;
1223     int index_scale = 1;
1224     int i, j;
1225
1226     for (i = 0; i < seekhead_list->nb_elem; i++)
1227         if (seekhead[i].id == MATROSKA_ID_CUES)
1228             break;
1229     assert(i <= seekhead_list->nb_elem);
1230
1231     matroska_parse_seekhead_entry(matroska, i);
1232
1233     index_list = &matroska->index;
1234     index = index_list->elem;
1235     if (index_list->nb_elem
1236         && index[0].time > 1E14/matroska->time_scale) {
1237         av_log(matroska->ctx, AV_LOG_WARNING, "Working around broken index.\n");
1238         index_scale = matroska->time_scale;
1239     }
1240     for (i = 0; i < index_list->nb_elem; i++) {
1241         EbmlList *pos_list = &index[i].pos;
1242         MatroskaIndexPos *pos = pos_list->elem;
1243         for (j = 0; j < pos_list->nb_elem; j++) {
1244             MatroskaTrack *track = matroska_find_track_by_num(matroska, pos[j].track);
1245             if (track && track->stream)
1246                 av_add_index_entry(track->stream,
1247                                    pos[j].pos + matroska->segment_start,
1248                                    index[i].time/index_scale, 0, 0,
1249                                    AVINDEX_KEYFRAME);
1250         }
1251     }
1252 }
1253
1254 static int matroska_aac_profile(char *codec_id)
1255 {
1256     static const char * const aac_profiles[] = { "MAIN", "LC", "SSR" };
1257     int profile;
1258
1259     for (profile=0; profile<FF_ARRAY_ELEMS(aac_profiles); profile++)
1260         if (strstr(codec_id, aac_profiles[profile]))
1261             break;
1262     return profile + 1;
1263 }
1264
1265 static int matroska_aac_sri(int samplerate)
1266 {
1267     int sri;
1268
1269     for (sri=0; sri<FF_ARRAY_ELEMS(avpriv_mpeg4audio_sample_rates); sri++)
1270         if (avpriv_mpeg4audio_sample_rates[sri] == samplerate)
1271             break;
1272     return sri;
1273 }
1274
1275 static int matroska_read_header(AVFormatContext *s)
1276 {
1277     MatroskaDemuxContext *matroska = s->priv_data;
1278     EbmlList *attachements_list = &matroska->attachments;
1279     MatroskaAttachement *attachements;
1280     EbmlList *chapters_list = &matroska->chapters;
1281     MatroskaChapter *chapters;
1282     MatroskaTrack *tracks;
1283     uint64_t max_start = 0;
1284     Ebml ebml = { 0 };
1285     AVStream *st;
1286     int i, j, res;
1287
1288     matroska->ctx = s;
1289
1290     /* First read the EBML header. */
1291     if (ebml_parse(matroska, ebml_syntax, &ebml)
1292         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1293         || ebml.id_length > sizeof(uint32_t) || ebml.doctype_version > 2) {
1294         av_log(matroska->ctx, AV_LOG_ERROR,
1295                "EBML header using unsupported features\n"
1296                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1297                ebml.version, ebml.doctype, ebml.doctype_version);
1298         ebml_free(ebml_syntax, &ebml);
1299         return AVERROR_PATCHWELCOME;
1300     }
1301     for (i = 0; i < FF_ARRAY_ELEMS(matroska_doctypes); i++)
1302         if (!strcmp(ebml.doctype, matroska_doctypes[i]))
1303             break;
1304     if (i >= FF_ARRAY_ELEMS(matroska_doctypes)) {
1305         av_log(s, AV_LOG_WARNING, "Unknown EBML doctype '%s'\n", ebml.doctype);
1306     }
1307     ebml_free(ebml_syntax, &ebml);
1308
1309     /* The next thing is a segment. */
1310     if ((res = ebml_parse(matroska, matroska_segments, matroska)) < 0)
1311         return res;
1312     matroska_execute_seekhead(matroska);
1313
1314     if (!matroska->time_scale)
1315         matroska->time_scale = 1000000;
1316     if (matroska->duration)
1317         matroska->ctx->duration = matroska->duration * matroska->time_scale
1318                                   * 1000 / AV_TIME_BASE;
1319     av_dict_set(&s->metadata, "title", matroska->title, 0);
1320
1321     tracks = matroska->tracks.elem;
1322     for (i=0; i < matroska->tracks.nb_elem; i++) {
1323         MatroskaTrack *track = &tracks[i];
1324         enum CodecID codec_id = CODEC_ID_NONE;
1325         EbmlList *encodings_list = &tracks->encodings;
1326         MatroskaTrackEncoding *encodings = encodings_list->elem;
1327         uint8_t *extradata = NULL;
1328         int extradata_size = 0;
1329         int extradata_offset = 0;
1330         AVIOContext b;
1331
1332         /* Apply some sanity checks. */
1333         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1334             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1335             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1336             av_log(matroska->ctx, AV_LOG_INFO,
1337                    "Unknown or unsupported track type %"PRIu64"\n",
1338                    track->type);
1339             continue;
1340         }
1341         if (track->codec_id == NULL)
1342             continue;
1343
1344         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1345             if (!track->default_duration)
1346                 track->default_duration = 1000000000/track->video.frame_rate;
1347             if (!track->video.display_width)
1348                 track->video.display_width = track->video.pixel_width;
1349             if (!track->video.display_height)
1350                 track->video.display_height = track->video.pixel_height;
1351         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1352             if (!track->audio.out_samplerate)
1353                 track->audio.out_samplerate = track->audio.samplerate;
1354         }
1355         if (encodings_list->nb_elem > 1) {
1356             av_log(matroska->ctx, AV_LOG_ERROR,
1357                    "Multiple combined encodings not supported");
1358         } else if (encodings_list->nb_elem == 1) {
1359             if (encodings[0].type ||
1360                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1361 #if CONFIG_ZLIB
1362                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1363 #endif
1364 #if CONFIG_BZLIB
1365                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1366 #endif
1367                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1368                 encodings[0].scope = 0;
1369                 av_log(matroska->ctx, AV_LOG_ERROR,
1370                        "Unsupported encoding type");
1371             } else if (track->codec_priv.size && encodings[0].scope&2) {
1372                 uint8_t *codec_priv = track->codec_priv.data;
1373                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1374                                                     &track->codec_priv.size,
1375                                                     track);
1376                 if (offset < 0) {
1377                     track->codec_priv.data = NULL;
1378                     track->codec_priv.size = 0;
1379                     av_log(matroska->ctx, AV_LOG_ERROR,
1380                            "Failed to decode codec private data\n");
1381                 } else if (offset > 0) {
1382                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1383                     memcpy(track->codec_priv.data,
1384                            encodings[0].compression.settings.data, offset);
1385                     memcpy(track->codec_priv.data+offset, codec_priv,
1386                            track->codec_priv.size);
1387                     track->codec_priv.size += offset;
1388                 }
1389                 if (codec_priv != track->codec_priv.data)
1390                     av_free(codec_priv);
1391             }
1392         }
1393
1394         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1395             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1396                         strlen(ff_mkv_codec_tags[j].str))){
1397                 codec_id= ff_mkv_codec_tags[j].id;
1398                 break;
1399             }
1400         }
1401
1402         st = track->stream = avformat_new_stream(s, NULL);
1403         if (st == NULL)
1404             return AVERROR(ENOMEM);
1405
1406         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1407             && track->codec_priv.size >= 40
1408             && track->codec_priv.data != NULL) {
1409             track->ms_compat = 1;
1410             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1411             codec_id = ff_codec_get_id(ff_codec_bmp_tags, track->video.fourcc);
1412             extradata_offset = 40;
1413         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1414                    && track->codec_priv.size >= 14
1415                    && track->codec_priv.data != NULL) {
1416             int ret;
1417             ffio_init_context(&b, track->codec_priv.data, track->codec_priv.size,
1418                           AVIO_FLAG_READ, NULL, NULL, NULL, NULL);
1419             ret = ff_get_wav_header(&b, st->codec, track->codec_priv.size);
1420             if (ret < 0)
1421                 return ret;
1422             codec_id = st->codec->codec_id;
1423             extradata_offset = FFMIN(track->codec_priv.size, 18);
1424         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1425                    && (track->codec_priv.size >= 86)
1426                    && (track->codec_priv.data != NULL)) {
1427             track->video.fourcc = AV_RL32(track->codec_priv.data);
1428             codec_id=ff_codec_get_id(ff_codec_movvideo_tags, track->video.fourcc);
1429         } else if (codec_id == CODEC_ID_PCM_S16BE) {
1430             switch (track->audio.bitdepth) {
1431             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1432             case 24:  codec_id = CODEC_ID_PCM_S24BE;  break;
1433             case 32:  codec_id = CODEC_ID_PCM_S32BE;  break;
1434             }
1435         } else if (codec_id == CODEC_ID_PCM_S16LE) {
1436             switch (track->audio.bitdepth) {
1437             case  8:  codec_id = CODEC_ID_PCM_U8;     break;
1438             case 24:  codec_id = CODEC_ID_PCM_S24LE;  break;
1439             case 32:  codec_id = CODEC_ID_PCM_S32LE;  break;
1440             }
1441         } else if (codec_id==CODEC_ID_PCM_F32LE && track->audio.bitdepth==64) {
1442             codec_id = CODEC_ID_PCM_F64LE;
1443         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1444             int profile = matroska_aac_profile(track->codec_id);
1445             int sri = matroska_aac_sri(track->audio.samplerate);
1446             extradata = av_mallocz(5 + FF_INPUT_BUFFER_PADDING_SIZE);
1447             if (extradata == NULL)
1448                 return AVERROR(ENOMEM);
1449             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1450             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1451             if (strstr(track->codec_id, "SBR")) {
1452                 sri = matroska_aac_sri(track->audio.out_samplerate);
1453                 extradata[2] = 0x56;
1454                 extradata[3] = 0xE5;
1455                 extradata[4] = 0x80 | (sri<<3);
1456                 extradata_size = 5;
1457             } else
1458                 extradata_size = 2;
1459         } else if (codec_id == CODEC_ID_TTA) {
1460             extradata_size = 30;
1461             extradata = av_mallocz(extradata_size);
1462             if (extradata == NULL)
1463                 return AVERROR(ENOMEM);
1464             ffio_init_context(&b, extradata, extradata_size, 1,
1465                           NULL, NULL, NULL, NULL);
1466             avio_write(&b, "TTA1", 4);
1467             avio_wl16(&b, 1);
1468             avio_wl16(&b, track->audio.channels);
1469             avio_wl16(&b, track->audio.bitdepth);
1470             avio_wl32(&b, track->audio.out_samplerate);
1471             avio_wl32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1472         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1473                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1474             extradata_offset = 26;
1475         } else if (codec_id == CODEC_ID_RA_144) {
1476             track->audio.out_samplerate = 8000;
1477             track->audio.channels = 1;
1478         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1479                    codec_id == CODEC_ID_ATRAC3 || codec_id == CODEC_ID_SIPR) {
1480             int flavor;
1481             ffio_init_context(&b, track->codec_priv.data,track->codec_priv.size,
1482                           0, NULL, NULL, NULL, NULL);
1483             avio_skip(&b, 22);
1484             flavor                       = avio_rb16(&b);
1485             track->audio.coded_framesize = avio_rb32(&b);
1486             avio_skip(&b, 12);
1487             track->audio.sub_packet_h    = avio_rb16(&b);
1488             track->audio.frame_size      = avio_rb16(&b);
1489             track->audio.sub_packet_size = avio_rb16(&b);
1490             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1491             if (codec_id == CODEC_ID_RA_288) {
1492                 st->codec->block_align = track->audio.coded_framesize;
1493                 track->codec_priv.size = 0;
1494             } else {
1495                 if (codec_id == CODEC_ID_SIPR && flavor < 4) {
1496                     const int sipr_bit_rate[4] = { 6504, 8496, 5000, 16000 };
1497                     track->audio.sub_packet_size = ff_sipr_subpk_size[flavor];
1498                     st->codec->bit_rate = sipr_bit_rate[flavor];
1499                 }
1500                 st->codec->block_align = track->audio.sub_packet_size;
1501                 extradata_offset = 78;
1502             }
1503         }
1504         track->codec_priv.size -= extradata_offset;
1505
1506         if (codec_id == CODEC_ID_NONE)
1507             av_log(matroska->ctx, AV_LOG_INFO,
1508                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1509
1510         if (track->time_scale < 0.01)
1511             track->time_scale = 1.0;
1512         avpriv_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1513
1514         st->codec->codec_id = codec_id;
1515         st->start_time = 0;
1516         if (strcmp(track->language, "und"))
1517             av_dict_set(&st->metadata, "language", track->language, 0);
1518         av_dict_set(&st->metadata, "title", track->name, 0);
1519
1520         if (track->flag_default)
1521             st->disposition |= AV_DISPOSITION_DEFAULT;
1522         if (track->flag_forced)
1523             st->disposition |= AV_DISPOSITION_FORCED;
1524
1525         if (!st->codec->extradata) {
1526             if(extradata){
1527                 st->codec->extradata = extradata;
1528                 st->codec->extradata_size = extradata_size;
1529             } else if(track->codec_priv.data && track->codec_priv.size > 0){
1530                 st->codec->extradata = av_mallocz(track->codec_priv.size +
1531                                                   FF_INPUT_BUFFER_PADDING_SIZE);
1532                 if(st->codec->extradata == NULL)
1533                     return AVERROR(ENOMEM);
1534                 st->codec->extradata_size = track->codec_priv.size;
1535                 memcpy(st->codec->extradata,
1536                        track->codec_priv.data + extradata_offset,
1537                        track->codec_priv.size);
1538             }
1539         }
1540
1541         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1542             st->codec->codec_type = AVMEDIA_TYPE_VIDEO;
1543             st->codec->codec_tag  = track->video.fourcc;
1544             st->codec->width  = track->video.pixel_width;
1545             st->codec->height = track->video.pixel_height;
1546             av_reduce(&st->sample_aspect_ratio.num,
1547                       &st->sample_aspect_ratio.den,
1548                       st->codec->height * track->video.display_width,
1549                       st->codec-> width * track->video.display_height,
1550                       255);
1551             if (st->codec->codec_id != CODEC_ID_H264)
1552             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1553             if (track->default_duration)
1554                 st->r_frame_rate   =
1555                 st->avg_frame_rate = av_d2q(1000000000.0/track->default_duration, INT_MAX);
1556         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1557             st->codec->codec_type = AVMEDIA_TYPE_AUDIO;
1558             st->codec->sample_rate = track->audio.out_samplerate;
1559             st->codec->channels = track->audio.channels;
1560             if (st->codec->codec_id != CODEC_ID_AAC)
1561             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1562         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1563             st->codec->codec_type = AVMEDIA_TYPE_SUBTITLE;
1564         }
1565     }
1566
1567     attachements = attachements_list->elem;
1568     for (j=0; j<attachements_list->nb_elem; j++) {
1569         if (!(attachements[j].filename && attachements[j].mime &&
1570               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1571             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1572         } else {
1573             AVStream *st = avformat_new_stream(s, NULL);
1574             if (st == NULL)
1575                 break;
1576             av_dict_set(&st->metadata, "filename",attachements[j].filename, 0);
1577             av_dict_set(&st->metadata, "mimetype", attachements[j].mime, 0);
1578             st->codec->codec_id = CODEC_ID_NONE;
1579             st->codec->codec_type = AVMEDIA_TYPE_ATTACHMENT;
1580             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1581             if(st->codec->extradata == NULL)
1582                 break;
1583             st->codec->extradata_size = attachements[j].bin.size;
1584             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1585
1586             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1587                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1588                              strlen(ff_mkv_mime_tags[i].str))) {
1589                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1590                     break;
1591                 }
1592             }
1593             attachements[j].stream = st;
1594         }
1595     }
1596
1597     chapters = chapters_list->elem;
1598     for (i=0; i<chapters_list->nb_elem; i++)
1599         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid
1600             && (max_start==0 || chapters[i].start > max_start)) {
1601             chapters[i].chapter =
1602             avpriv_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1603                            chapters[i].start, chapters[i].end,
1604                            chapters[i].title);
1605             av_dict_set(&chapters[i].chapter->metadata,
1606                              "title", chapters[i].title, 0);
1607             max_start = chapters[i].start;
1608         }
1609
1610     matroska_convert_tags(s);
1611
1612     return 0;
1613 }
1614
1615 /*
1616  * Put one packet in an application-supplied AVPacket struct.
1617  * Returns 0 on success or -1 on failure.
1618  */
1619 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1620                                    AVPacket *pkt)
1621 {
1622     if (matroska->num_packets > 0) {
1623         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1624         av_free(matroska->packets[0]);
1625         if (matroska->num_packets > 1) {
1626             void *newpackets;
1627             memmove(&matroska->packets[0], &matroska->packets[1],
1628                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1629             newpackets = av_realloc(matroska->packets,
1630                             (matroska->num_packets - 1) * sizeof(AVPacket *));
1631             if (newpackets)
1632                 matroska->packets = newpackets;
1633         } else {
1634             av_freep(&matroska->packets);
1635         }
1636         matroska->num_packets--;
1637         return 0;
1638     }
1639
1640     return -1;
1641 }
1642
1643 /*
1644  * Free all packets in our internal queue.
1645  */
1646 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1647 {
1648     if (matroska->packets) {
1649         int n;
1650         for (n = 0; n < matroska->num_packets; n++) {
1651             av_free_packet(matroska->packets[n]);
1652             av_free(matroska->packets[n]);
1653         }
1654         av_freep(&matroska->packets);
1655         matroska->num_packets = 0;
1656     }
1657 }
1658
1659 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1660                                 int size, int64_t pos, uint64_t cluster_time,
1661                                 uint64_t duration, int is_keyframe,
1662                                 int64_t cluster_pos)
1663 {
1664     uint64_t timecode = AV_NOPTS_VALUE;
1665     MatroskaTrack *track;
1666     int res = 0;
1667     AVStream *st;
1668     AVPacket *pkt;
1669     int16_t block_time;
1670     uint32_t *lace_size = NULL;
1671     int n, flags, laces = 0;
1672     uint64_t num;
1673
1674     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1675         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1676         return res;
1677     }
1678     data += n;
1679     size -= n;
1680
1681     track = matroska_find_track_by_num(matroska, num);
1682     if (!track || !track->stream) {
1683         av_log(matroska->ctx, AV_LOG_INFO,
1684                "Invalid stream %"PRIu64" or size %u\n", num, size);
1685         return AVERROR_INVALIDDATA;
1686     } else if (size <= 3)
1687         return 0;
1688     st = track->stream;
1689     if (st->discard >= AVDISCARD_ALL)
1690         return res;
1691     if (duration == AV_NOPTS_VALUE)
1692         duration = track->default_duration / matroska->time_scale;
1693
1694     block_time = AV_RB16(data);
1695     data += 2;
1696     flags = *data++;
1697     size -= 3;
1698     if (is_keyframe == -1)
1699         is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0;
1700
1701     if (cluster_time != (uint64_t)-1
1702         && (block_time >= 0 || cluster_time >= -block_time)) {
1703         timecode = cluster_time + block_time;
1704         if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE
1705             && timecode < track->end_timecode)
1706             is_keyframe = 0;  /* overlapping subtitles are not key frame */
1707         if (is_keyframe)
1708             av_add_index_entry(st, cluster_pos, timecode, 0,0,AVINDEX_KEYFRAME);
1709         track->end_timecode = FFMAX(track->end_timecode, timecode+duration);
1710     }
1711
1712     if (matroska->skip_to_keyframe && track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1713         if (!is_keyframe || timecode < matroska->skip_to_timecode)
1714             return res;
1715         matroska->skip_to_keyframe = 0;
1716     }
1717
1718     switch ((flags & 0x06) >> 1) {
1719         case 0x0: /* no lacing */
1720             laces = 1;
1721             lace_size = av_mallocz(sizeof(int));
1722             lace_size[0] = size;
1723             break;
1724
1725         case 0x1: /* Xiph lacing */
1726         case 0x2: /* fixed-size lacing */
1727         case 0x3: /* EBML lacing */
1728             assert(size>0); // size <=3 is checked before size-=3 above
1729             laces = (*data) + 1;
1730             data += 1;
1731             size -= 1;
1732             lace_size = av_mallocz(laces * sizeof(int));
1733
1734             switch ((flags & 0x06) >> 1) {
1735                 case 0x1: /* Xiph lacing */ {
1736                     uint8_t temp;
1737                     uint32_t total = 0;
1738                     for (n = 0; res == 0 && n < laces - 1; n++) {
1739                         while (1) {
1740                             if (size == 0) {
1741                                 res = -1;
1742                                 break;
1743                             }
1744                             temp = *data;
1745                             lace_size[n] += temp;
1746                             data += 1;
1747                             size -= 1;
1748                             if (temp != 0xff)
1749                                 break;
1750                         }
1751                         total += lace_size[n];
1752                     }
1753                     lace_size[n] = size - total;
1754                     break;
1755                 }
1756
1757                 case 0x2: /* fixed-size lacing */
1758                     for (n = 0; n < laces; n++)
1759                         lace_size[n] = size / laces;
1760                     break;
1761
1762                 case 0x3: /* EBML lacing */ {
1763                     uint32_t total;
1764                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1765                     if (n < 0) {
1766                         av_log(matroska->ctx, AV_LOG_INFO,
1767                                "EBML block data error\n");
1768                         break;
1769                     }
1770                     data += n;
1771                     size -= n;
1772                     total = lace_size[0] = num;
1773                     for (n = 1; res == 0 && n < laces - 1; n++) {
1774                         int64_t snum;
1775                         int r;
1776                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1777                         if (r < 0) {
1778                             av_log(matroska->ctx, AV_LOG_INFO,
1779                                    "EBML block data error\n");
1780                             break;
1781                         }
1782                         data += r;
1783                         size -= r;
1784                         lace_size[n] = lace_size[n - 1] + snum;
1785                         total += lace_size[n];
1786                     }
1787                     lace_size[laces - 1] = size - total;
1788                     break;
1789                 }
1790             }
1791             break;
1792     }
1793
1794     if (res == 0) {
1795         for (n = 0; n < laces; n++) {
1796             if ((st->codec->codec_id == CODEC_ID_RA_288 ||
1797                  st->codec->codec_id == CODEC_ID_COOK ||
1798                  st->codec->codec_id == CODEC_ID_SIPR ||
1799                  st->codec->codec_id == CODEC_ID_ATRAC3) &&
1800                  st->codec->block_align && track->audio.sub_packet_size) {
1801                 int a = st->codec->block_align;
1802                 int sps = track->audio.sub_packet_size;
1803                 int cfs = track->audio.coded_framesize;
1804                 int h = track->audio.sub_packet_h;
1805                 int y = track->audio.sub_packet_cnt;
1806                 int w = track->audio.frame_size;
1807                 int x;
1808
1809                 if (!track->audio.pkt_cnt) {
1810                     if (track->audio.sub_packet_cnt == 0)
1811                         track->audio.buf_timecode = timecode;
1812                     if (st->codec->codec_id == CODEC_ID_RA_288) {
1813                         if (size < cfs * h / 2) {
1814                             av_log(matroska->ctx, AV_LOG_ERROR,
1815                                    "Corrupt int4 RM-style audio packet size\n");
1816                             res = AVERROR_INVALIDDATA;
1817                             goto end;
1818                         }
1819                         for (x=0; x<h/2; x++)
1820                             memcpy(track->audio.buf+x*2*w+y*cfs,
1821                                    data+x*cfs, cfs);
1822                     } else if (st->codec->codec_id == CODEC_ID_SIPR) {
1823                         if (size < w) {
1824                             av_log(matroska->ctx, AV_LOG_ERROR,
1825                                    "Corrupt sipr RM-style audio packet size\n");
1826                             res = AVERROR_INVALIDDATA;
1827                             goto end;
1828                         }
1829                         memcpy(track->audio.buf + y*w, data, w);
1830                     } else {
1831                         if (size < sps * w / sps) {
1832                             av_log(matroska->ctx, AV_LOG_ERROR,
1833                                    "Corrupt generic RM-style audio packet size\n");
1834                             res = AVERROR_INVALIDDATA;
1835                             goto end;
1836                         }
1837                         for (x=0; x<w/sps; x++)
1838                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1839                     }
1840
1841                     if (++track->audio.sub_packet_cnt >= h) {
1842                         if (st->codec->codec_id == CODEC_ID_SIPR)
1843                             ff_rm_reorder_sipr_data(track->audio.buf, h, w);
1844                         track->audio.sub_packet_cnt = 0;
1845                         track->audio.pkt_cnt = h*w / a;
1846                     }
1847                 }
1848                 while (track->audio.pkt_cnt) {
1849                     pkt = av_mallocz(sizeof(AVPacket));
1850                     av_new_packet(pkt, a);
1851                     memcpy(pkt->data, track->audio.buf
1852                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1853                     pkt->pts = track->audio.buf_timecode;
1854                     track->audio.buf_timecode = AV_NOPTS_VALUE;
1855                     pkt->pos = pos;
1856                     pkt->stream_index = st->index;
1857                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1858                 }
1859             } else {
1860                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1861                 int offset = 0, pkt_size = lace_size[n];
1862                 uint8_t *pkt_data = data;
1863
1864                 if (pkt_size > size) {
1865                     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid packet size\n");
1866                     break;
1867                 }
1868
1869                 if (encodings && encodings->scope & 1) {
1870                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1871                     if (offset < 0)
1872                         continue;
1873                 }
1874
1875                 pkt = av_mallocz(sizeof(AVPacket));
1876                 /* XXX: prevent data copy... */
1877                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1878                     av_free(pkt);
1879                     res = AVERROR(ENOMEM);
1880                     break;
1881                 }
1882                 if (offset)
1883                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1884                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1885
1886                 if (pkt_data != data)
1887                     av_free(pkt_data);
1888
1889                 if (n == 0)
1890                     pkt->flags = is_keyframe;
1891                 pkt->stream_index = st->index;
1892
1893                 if (track->ms_compat)
1894                     pkt->dts = timecode;
1895                 else
1896                     pkt->pts = timecode;
1897                 pkt->pos = pos;
1898                 if (st->codec->codec_id == CODEC_ID_TEXT)
1899                     pkt->convergence_duration = duration;
1900                 else if (track->type != MATROSKA_TRACK_TYPE_SUBTITLE)
1901                     pkt->duration = duration;
1902
1903                 if (st->codec->codec_id == CODEC_ID_SSA)
1904                     matroska_fix_ass_packet(matroska, pkt, duration);
1905
1906                 if (matroska->prev_pkt &&
1907                     timecode != AV_NOPTS_VALUE &&
1908                     matroska->prev_pkt->pts == timecode &&
1909                     matroska->prev_pkt->stream_index == st->index &&
1910                     st->codec->codec_id == CODEC_ID_SSA)
1911                     matroska_merge_packets(matroska->prev_pkt, pkt);
1912                 else {
1913                     dynarray_add(&matroska->packets,&matroska->num_packets,pkt);
1914                     matroska->prev_pkt = pkt;
1915                 }
1916             }
1917
1918             if (timecode != AV_NOPTS_VALUE)
1919                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1920             data += lace_size[n];
1921             size -= lace_size[n];
1922         }
1923     }
1924
1925 end:
1926     av_free(lace_size);
1927     return res;
1928 }
1929
1930 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1931 {
1932     MatroskaCluster cluster = { 0 };
1933     EbmlList *blocks_list;
1934     MatroskaBlock *blocks;
1935     int i, res;
1936     int64_t pos = avio_tell(matroska->ctx->pb);
1937     matroska->prev_pkt = NULL;
1938     if (matroska->current_id)
1939         pos -= 4;  /* sizeof the ID which was already read */
1940     res = ebml_parse(matroska, matroska_clusters, &cluster);
1941     blocks_list = &cluster.blocks;
1942     blocks = blocks_list->elem;
1943     for (i=0; i<blocks_list->nb_elem && !res; i++)
1944         if (blocks[i].bin.size > 0 && blocks[i].bin.data) {
1945             int is_keyframe = blocks[i].non_simple ? !blocks[i].reference : -1;
1946             if (!blocks[i].non_simple)
1947                 blocks[i].duration = AV_NOPTS_VALUE;
1948             res=matroska_parse_block(matroska,
1949                                      blocks[i].bin.data, blocks[i].bin.size,
1950                                      blocks[i].bin.pos,  cluster.timecode,
1951                                      blocks[i].duration, is_keyframe,
1952                                      pos);
1953         }
1954     ebml_free(matroska_cluster, &cluster);
1955     if (res < 0)  matroska->done = 1;
1956     return res;
1957 }
1958
1959 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1960 {
1961     MatroskaDemuxContext *matroska = s->priv_data;
1962     int ret = 0;
1963
1964     while (!ret && matroska_deliver_packet(matroska, pkt)) {
1965         if (matroska->done)
1966             return AVERROR_EOF;
1967         ret = matroska_parse_cluster(matroska);
1968     }
1969
1970     return ret;
1971 }
1972
1973 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1974                               int64_t timestamp, int flags)
1975 {
1976     MatroskaDemuxContext *matroska = s->priv_data;
1977     MatroskaTrack *tracks = matroska->tracks.elem;
1978     AVStream *st = s->streams[stream_index];
1979     int i, index, index_sub, index_min;
1980
1981     /* Parse the CUES now since we need the index data to seek. */
1982     if (matroska->cues_parsing_deferred) {
1983         matroska_parse_cues(matroska);
1984         matroska->cues_parsing_deferred = 0;
1985     }
1986
1987     if (!st->nb_index_entries)
1988         return 0;
1989     timestamp = FFMAX(timestamp, st->index_entries[0].timestamp);
1990
1991     if ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1992         avio_seek(s->pb, st->index_entries[st->nb_index_entries-1].pos, SEEK_SET);
1993         matroska->current_id = 0;
1994         while ((index = av_index_search_timestamp(st, timestamp, flags)) < 0) {
1995             matroska_clear_queue(matroska);
1996             if (matroska_parse_cluster(matroska) < 0)
1997                 break;
1998         }
1999     }
2000
2001     matroska_clear_queue(matroska);
2002     if (index < 0)
2003         return 0;
2004
2005     index_min = index;
2006     for (i=0; i < matroska->tracks.nb_elem; i++) {
2007         tracks[i].audio.pkt_cnt = 0;
2008         tracks[i].audio.sub_packet_cnt = 0;
2009         tracks[i].audio.buf_timecode = AV_NOPTS_VALUE;
2010         tracks[i].end_timecode = 0;
2011         if (tracks[i].type == MATROSKA_TRACK_TYPE_SUBTITLE
2012             && !tracks[i].stream->discard != AVDISCARD_ALL) {
2013             index_sub = av_index_search_timestamp(tracks[i].stream, st->index_entries[index].timestamp, AVSEEK_FLAG_BACKWARD);
2014             if (index_sub >= 0
2015                 && st->index_entries[index_sub].pos < st->index_entries[index_min].pos
2016                 && st->index_entries[index].timestamp - st->index_entries[index_sub].timestamp < 30000000000/matroska->time_scale)
2017                 index_min = index_sub;
2018         }
2019     }
2020
2021     avio_seek(s->pb, st->index_entries[index_min].pos, SEEK_SET);
2022     matroska->current_id = 0;
2023     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
2024     matroska->skip_to_timecode = st->index_entries[index].timestamp;
2025     matroska->done = 0;
2026     ff_update_cur_dts(s, st, st->index_entries[index].timestamp);
2027     return 0;
2028 }
2029
2030 static int matroska_read_close(AVFormatContext *s)
2031 {
2032     MatroskaDemuxContext *matroska = s->priv_data;
2033     MatroskaTrack *tracks = matroska->tracks.elem;
2034     int n;
2035
2036     matroska_clear_queue(matroska);
2037
2038     for (n=0; n < matroska->tracks.nb_elem; n++)
2039         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
2040             av_free(tracks[n].audio.buf);
2041     ebml_free(matroska_segment, matroska);
2042
2043     return 0;
2044 }
2045
2046 AVInputFormat ff_matroska_demuxer = {
2047     .name           = "matroska,webm",
2048     .long_name      = NULL_IF_CONFIG_SMALL("Matroska/WebM file format"),
2049     .priv_data_size = sizeof(MatroskaDemuxContext),
2050     .read_probe     = matroska_probe,
2051     .read_header    = matroska_read_header,
2052     .read_packet    = matroska_read_packet,
2053     .read_close     = matroska_read_close,
2054     .read_seek      = matroska_read_seek,
2055 };