matroskadec: simplify matroska_ebmlnum_sint()
[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 matroskadec.c
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 "avformat.h"
32 /* For codec_get_id(). */
33 #include "riff.h"
34 #include "isom.h"
35 #include "matroska.h"
36 #include "libavcodec/mpeg4audio.h"
37 #include "libavutil/intfloat_readwrite.h"
38 #include "libavutil/avstring.h"
39 #include "libavutil/lzo.h"
40 #ifdef CONFIG_ZLIB
41 #include <zlib.h>
42 #endif
43 #ifdef CONFIG_BZLIB
44 #include <bzlib.h>
45 #endif
46
47 typedef enum {
48     EBML_NONE,
49     EBML_UINT,
50     EBML_FLOAT,
51     EBML_STR,
52     EBML_UTF8,
53     EBML_BIN,
54     EBML_NEST,
55     EBML_PASS,
56     EBML_STOP,
57 } EbmlType;
58
59 typedef const struct EbmlSyntax {
60     uint32_t id;
61     EbmlType type;
62     int list_elem_size;
63     int data_offset;
64     union {
65         uint64_t    u;
66         double      f;
67         const char *s;
68         const struct EbmlSyntax *n;
69     } def;
70 } EbmlSyntax;
71
72 typedef struct {
73     int nb_elem;
74     void *elem;
75 } EbmlList;
76
77 typedef struct {
78     int      size;
79     uint8_t *data;
80     int64_t  pos;
81 } EbmlBin;
82
83 typedef struct {
84     uint64_t version;
85     uint64_t max_size;
86     uint64_t id_length;
87     char    *doctype;
88     uint64_t doctype_version;
89 } Ebml;
90
91 typedef struct {
92     uint64_t algo;
93     EbmlBin  settings;
94 } MatroskaTrackCompression;
95
96 typedef struct {
97     uint64_t scope;
98     uint64_t type;
99     MatroskaTrackCompression compression;
100 } MatroskaTrackEncoding;
101
102 typedef struct {
103     double   frame_rate;
104     uint64_t display_width;
105     uint64_t display_height;
106     uint64_t pixel_width;
107     uint64_t pixel_height;
108     uint64_t fourcc;
109 } MatroskaTrackVideo;
110
111 typedef struct {
112     double   samplerate;
113     double   out_samplerate;
114     uint64_t bitdepth;
115     uint64_t channels;
116
117     /* real audio header (extracted from extradata) */
118     int      coded_framesize;
119     int      sub_packet_h;
120     int      frame_size;
121     int      sub_packet_size;
122     int      sub_packet_cnt;
123     int      pkt_cnt;
124     uint8_t *buf;
125 } MatroskaTrackAudio;
126
127 typedef struct {
128     uint64_t num;
129     uint64_t type;
130     char    *codec_id;
131     EbmlBin  codec_priv;
132     char    *language;
133     double time_scale;
134     uint64_t default_duration;
135     uint64_t flag_default;
136     MatroskaTrackVideo video;
137     MatroskaTrackAudio audio;
138     EbmlList encodings;
139
140     AVStream *stream;
141 } MatroskaTrack;
142
143 typedef struct {
144     char *filename;
145     char *mime;
146     EbmlBin bin;
147 } MatroskaAttachement;
148
149 typedef struct {
150     uint64_t start;
151     uint64_t end;
152     uint64_t uid;
153     char    *title;
154 } MatroskaChapter;
155
156 typedef struct {
157     uint64_t track;
158     uint64_t pos;
159 } MatroskaIndexPos;
160
161 typedef struct {
162     uint64_t time;
163     EbmlList pos;
164 } MatroskaIndex;
165
166 typedef struct {
167     uint64_t id;
168     uint64_t pos;
169 } MatroskaSeekhead;
170
171 typedef struct {
172     uint64_t start;
173     uint64_t length;
174 } MatroskaLevel;
175
176 typedef struct {
177     AVFormatContext *ctx;
178
179     /* ebml stuff */
180     int num_levels;
181     MatroskaLevel levels[EBML_MAX_DEPTH];
182     int level_up;
183
184     uint64_t time_scale;
185     double   duration;
186     char    *title;
187     EbmlList tracks;
188     EbmlList attachments;
189     EbmlList chapters;
190     EbmlList index;
191     EbmlList seekhead;
192
193     /* num_streams is the number of streams that av_new_stream() was called
194      * for ( = that are available to the calling program). */
195     int num_streams;
196
197     /* byte position of the segment inside the stream */
198     offset_t segment_start;
199
200     /* The packet queue. */
201     AVPacket **packets;
202     int num_packets;
203
204     int done;
205     int has_cluster_id;
206
207     /* What to skip before effectively reading a packet. */
208     int skip_to_keyframe;
209     AVStream *skip_to_stream;
210 } MatroskaDemuxContext;
211
212 typedef struct {
213     uint64_t duration;
214     int64_t  reference;
215     EbmlBin  bin;
216 } MatroskaBlock;
217
218 typedef struct {
219     uint64_t timecode;
220     EbmlList blocks;
221 } MatroskaCluster;
222
223 #define ARRAY_SIZE(x)  (sizeof(x)/sizeof(*x))
224
225 static EbmlSyntax ebml_header[] = {
226     { EBML_ID_EBMLREADVERSION,        EBML_UINT, 0, offsetof(Ebml,version), {.u=EBML_VERSION} },
227     { EBML_ID_EBMLMAXSIZELENGTH,      EBML_UINT, 0, offsetof(Ebml,max_size), {.u=8} },
228     { EBML_ID_EBMLMAXIDLENGTH,        EBML_UINT, 0, offsetof(Ebml,id_length), {.u=4} },
229     { EBML_ID_DOCTYPE,                EBML_STR,  0, offsetof(Ebml,doctype), {.s="(none)"} },
230     { EBML_ID_DOCTYPEREADVERSION,     EBML_UINT, 0, offsetof(Ebml,doctype_version), {.u=1} },
231     { EBML_ID_EBMLVERSION,            EBML_NONE },
232     { EBML_ID_DOCTYPEVERSION,         EBML_NONE },
233     { EBML_ID_VOID,                   EBML_NONE },
234     { 0 }
235 };
236
237 static EbmlSyntax ebml_syntax[] = {
238     { EBML_ID_HEADER,                 EBML_NEST, 0, 0, {.n=ebml_header} },
239     { 0 }
240 };
241
242 static EbmlSyntax matroska_info[] = {
243     { MATROSKA_ID_TIMECODESCALE,      EBML_UINT,  0, offsetof(MatroskaDemuxContext,time_scale), {.u=1000000} },
244     { MATROSKA_ID_DURATION,           EBML_FLOAT, 0, offsetof(MatroskaDemuxContext,duration) },
245     { MATROSKA_ID_TITLE,              EBML_UTF8,  0, offsetof(MatroskaDemuxContext,title) },
246     { MATROSKA_ID_WRITINGAPP,         EBML_NONE },
247     { MATROSKA_ID_MUXINGAPP,          EBML_NONE },
248     { MATROSKA_ID_DATEUTC,            EBML_NONE },
249     { MATROSKA_ID_SEGMENTUID,         EBML_NONE },
250     { EBML_ID_VOID,                   EBML_NONE },
251     { 0 }
252 };
253
254 static EbmlSyntax matroska_track_video[] = {
255     { MATROSKA_ID_VIDEOFRAMERATE,     EBML_FLOAT,0, offsetof(MatroskaTrackVideo,frame_rate) },
256     { MATROSKA_ID_VIDEODISPLAYWIDTH,  EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_width) },
257     { MATROSKA_ID_VIDEODISPLAYHEIGHT, EBML_UINT, 0, offsetof(MatroskaTrackVideo,display_height) },
258     { MATROSKA_ID_VIDEOPIXELWIDTH,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_width) },
259     { MATROSKA_ID_VIDEOPIXELHEIGHT,   EBML_UINT, 0, offsetof(MatroskaTrackVideo,pixel_height) },
260     { MATROSKA_ID_VIDEOCOLORSPACE,    EBML_UINT, 0, offsetof(MatroskaTrackVideo,fourcc) },
261     { MATROSKA_ID_VIDEOFLAGINTERLACED,EBML_NONE },
262     { MATROSKA_ID_VIDEOSTEREOMODE,    EBML_NONE },
263     { MATROSKA_ID_VIDEOASPECTRATIO,   EBML_NONE },
264     { EBML_ID_VOID,                   EBML_NONE },
265     { 0 }
266 };
267
268 static EbmlSyntax matroska_track_audio[] = {
269     { MATROSKA_ID_AUDIOSAMPLINGFREQ,  EBML_FLOAT,0, offsetof(MatroskaTrackAudio,samplerate), {.f=8000.0} },
270     { MATROSKA_ID_AUDIOOUTSAMPLINGFREQ,EBML_FLOAT,0,offsetof(MatroskaTrackAudio,out_samplerate) },
271     { MATROSKA_ID_AUDIOBITDEPTH,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,bitdepth) },
272     { MATROSKA_ID_AUDIOCHANNELS,      EBML_UINT, 0, offsetof(MatroskaTrackAudio,channels), {.u=1} },
273     { EBML_ID_VOID,                   EBML_NONE },
274     { 0 }
275 };
276
277 static EbmlSyntax matroska_track_encoding_compression[] = {
278     { MATROSKA_ID_ENCODINGCOMPALGO,   EBML_UINT, 0, offsetof(MatroskaTrackCompression,algo), {.u=0} },
279     { MATROSKA_ID_ENCODINGCOMPSETTINGS,EBML_BIN, 0, offsetof(MatroskaTrackCompression,settings) },
280     { EBML_ID_VOID,                   EBML_NONE },
281     { 0 }
282 };
283
284 static EbmlSyntax matroska_track_encoding[] = {
285     { MATROSKA_ID_ENCODINGSCOPE,      EBML_UINT, 0, offsetof(MatroskaTrackEncoding,scope), {.u=1} },
286     { MATROSKA_ID_ENCODINGTYPE,       EBML_UINT, 0, offsetof(MatroskaTrackEncoding,type), {.u=0} },
287     { MATROSKA_ID_ENCODINGCOMPRESSION,EBML_NEST, 0, offsetof(MatroskaTrackEncoding,compression), {.n=matroska_track_encoding_compression} },
288     { EBML_ID_VOID,                   EBML_NONE },
289     { 0 }
290 };
291
292 static EbmlSyntax matroska_track_encodings[] = {
293     { MATROSKA_ID_TRACKCONTENTENCODING, EBML_NEST, sizeof(MatroskaTrackEncoding), offsetof(MatroskaTrack,encodings), {.n=matroska_track_encoding} },
294     { EBML_ID_VOID,                   EBML_NONE },
295     { 0 }
296 };
297
298 static EbmlSyntax matroska_track[] = {
299     { MATROSKA_ID_TRACKNUMBER,          EBML_UINT, 0, offsetof(MatroskaTrack,num) },
300     { MATROSKA_ID_TRACKTYPE,            EBML_UINT, 0, offsetof(MatroskaTrack,type) },
301     { MATROSKA_ID_CODECID,              EBML_STR,  0, offsetof(MatroskaTrack,codec_id) },
302     { MATROSKA_ID_CODECPRIVATE,         EBML_BIN,  0, offsetof(MatroskaTrack,codec_priv) },
303     { MATROSKA_ID_TRACKLANGUAGE,        EBML_UTF8, 0, offsetof(MatroskaTrack,language), {.s="eng"} },
304     { MATROSKA_ID_TRACKDEFAULTDURATION, EBML_UINT, 0, offsetof(MatroskaTrack,default_duration) },
305     { MATROSKA_ID_TRACKTIMECODESCALE,   EBML_FLOAT,0, offsetof(MatroskaTrack,time_scale), {.f=1.0} },
306     { MATROSKA_ID_TRACKFLAGDEFAULT,     EBML_UINT, 0, offsetof(MatroskaTrack,flag_default), {.u=1} },
307     { MATROSKA_ID_TRACKVIDEO,           EBML_NEST, 0, offsetof(MatroskaTrack,video), {.n=matroska_track_video} },
308     { MATROSKA_ID_TRACKAUDIO,           EBML_NEST, 0, offsetof(MatroskaTrack,audio), {.n=matroska_track_audio} },
309     { MATROSKA_ID_TRACKCONTENTENCODINGS,EBML_NEST, 0, 0, {.n=matroska_track_encodings} },
310     { MATROSKA_ID_TRACKUID,             EBML_NONE },
311     { MATROSKA_ID_TRACKNAME,            EBML_NONE },
312     { MATROSKA_ID_TRACKFLAGENABLED,     EBML_NONE },
313     { MATROSKA_ID_TRACKFLAGFORCED,      EBML_NONE },
314     { MATROSKA_ID_TRACKFLAGLACING,      EBML_NONE },
315     { MATROSKA_ID_CODECNAME,            EBML_NONE },
316     { MATROSKA_ID_CODECDECODEALL,       EBML_NONE },
317     { MATROSKA_ID_CODECINFOURL,         EBML_NONE },
318     { MATROSKA_ID_CODECDOWNLOADURL,     EBML_NONE },
319     { MATROSKA_ID_TRACKMINCACHE,        EBML_NONE },
320     { MATROSKA_ID_TRACKMAXCACHE,        EBML_NONE },
321     { EBML_ID_VOID,                     EBML_NONE },
322     { 0 }
323 };
324
325 static EbmlSyntax matroska_tracks[] = {
326     { MATROSKA_ID_TRACKENTRY,         EBML_NEST, sizeof(MatroskaTrack), offsetof(MatroskaDemuxContext,tracks), {.n=matroska_track} },
327     { EBML_ID_VOID,                   EBML_NONE },
328     { 0 }
329 };
330
331 static EbmlSyntax matroska_attachment[] = {
332     { MATROSKA_ID_FILENAME,           EBML_UTF8, 0, offsetof(MatroskaAttachement,filename) },
333     { MATROSKA_ID_FILEMIMETYPE,       EBML_STR,  0, offsetof(MatroskaAttachement,mime) },
334     { MATROSKA_ID_FILEDATA,           EBML_BIN,  0, offsetof(MatroskaAttachement,bin) },
335     { MATROSKA_ID_FILEUID,            EBML_NONE },
336     { EBML_ID_VOID,                   EBML_NONE },
337     { 0 }
338 };
339
340 static EbmlSyntax matroska_attachments[] = {
341     { MATROSKA_ID_ATTACHEDFILE,       EBML_NEST, sizeof(MatroskaAttachement), offsetof(MatroskaDemuxContext,attachments), {.n=matroska_attachment} },
342     { EBML_ID_VOID,                   EBML_NONE },
343     { 0 }
344 };
345
346 static EbmlSyntax matroska_chapter_display[] = {
347     { MATROSKA_ID_CHAPSTRING,         EBML_UTF8, 0, offsetof(MatroskaChapter,title) },
348     { EBML_ID_VOID,                   EBML_NONE },
349     { 0 }
350 };
351
352 static EbmlSyntax matroska_chapter_entry[] = {
353     { MATROSKA_ID_CHAPTERTIMESTART,   EBML_UINT, 0, offsetof(MatroskaChapter,start), {.u=AV_NOPTS_VALUE} },
354     { MATROSKA_ID_CHAPTERTIMEEND,     EBML_UINT, 0, offsetof(MatroskaChapter,end), {.u=AV_NOPTS_VALUE} },
355     { MATROSKA_ID_CHAPTERUID,         EBML_UINT, 0, offsetof(MatroskaChapter,uid) },
356     { MATROSKA_ID_CHAPTERDISPLAY,     EBML_NEST, 0, 0, {.n=matroska_chapter_display} },
357     { MATROSKA_ID_CHAPTERFLAGHIDDEN,  EBML_NONE },
358     { EBML_ID_VOID,                   EBML_NONE },
359     { 0 }
360 };
361
362 static EbmlSyntax matroska_chapter[] = {
363     { MATROSKA_ID_CHAPTERATOM,        EBML_NEST, sizeof(MatroskaChapter), offsetof(MatroskaDemuxContext,chapters), {.n=matroska_chapter_entry} },
364     { MATROSKA_ID_EDITIONUID,         EBML_NONE },
365     { MATROSKA_ID_EDITIONFLAGHIDDEN,  EBML_NONE },
366     { MATROSKA_ID_EDITIONFLAGDEFAULT, EBML_NONE },
367     { EBML_ID_VOID,                   EBML_NONE },
368     { 0 }
369 };
370
371 static EbmlSyntax matroska_chapters[] = {
372     { MATROSKA_ID_EDITIONENTRY,       EBML_NEST, 0, 0, {.n=matroska_chapter} },
373     { EBML_ID_VOID,                   EBML_NONE },
374     { 0 }
375 };
376
377 static EbmlSyntax matroska_index_pos[] = {
378     { MATROSKA_ID_CUETRACK,           EBML_UINT, 0, offsetof(MatroskaIndexPos,track) },
379     { MATROSKA_ID_CUECLUSTERPOSITION, EBML_UINT, 0, offsetof(MatroskaIndexPos,pos)   },
380     { EBML_ID_VOID,                   EBML_NONE },
381     { 0 }
382 };
383
384 static EbmlSyntax matroska_index_entry[] = {
385     { MATROSKA_ID_CUETIME,            EBML_UINT, 0, offsetof(MatroskaIndex,time) },
386     { MATROSKA_ID_CUETRACKPOSITION,   EBML_NEST, sizeof(MatroskaIndexPos), offsetof(MatroskaIndex,pos), {.n=matroska_index_pos} },
387     { EBML_ID_VOID,                   EBML_NONE },
388     { 0 }
389 };
390
391 static EbmlSyntax matroska_index[] = {
392     { MATROSKA_ID_POINTENTRY,         EBML_NEST, sizeof(MatroskaIndex), offsetof(MatroskaDemuxContext,index), {.n=matroska_index_entry} },
393     { EBML_ID_VOID,                   EBML_NONE },
394     { 0 }
395 };
396
397 static EbmlSyntax matroska_tags[] = {
398     { EBML_ID_VOID,                   EBML_NONE },
399     { 0 }
400 };
401
402 static EbmlSyntax matroska_seekhead_entry[] = {
403     { MATROSKA_ID_SEEKID,             EBML_UINT, 0, offsetof(MatroskaSeekhead,id) },
404     { MATROSKA_ID_SEEKPOSITION,       EBML_UINT, 0, offsetof(MatroskaSeekhead,pos), {.u=-1} },
405     { EBML_ID_VOID,                   EBML_NONE },
406     { 0 }
407 };
408
409 static EbmlSyntax matroska_seekhead[] = {
410     { MATROSKA_ID_SEEKENTRY,          EBML_NEST, sizeof(MatroskaSeekhead), offsetof(MatroskaDemuxContext,seekhead), {.n=matroska_seekhead_entry} },
411     { EBML_ID_VOID,                   EBML_NONE },
412     { 0 }
413 };
414
415 static EbmlSyntax matroska_segment[] = {
416     { MATROSKA_ID_INFO,           EBML_NEST, 0, 0, {.n=matroska_info       } },
417     { MATROSKA_ID_TRACKS,         EBML_NEST, 0, 0, {.n=matroska_tracks     } },
418     { MATROSKA_ID_ATTACHMENTS,    EBML_NEST, 0, 0, {.n=matroska_attachments} },
419     { MATROSKA_ID_CHAPTERS,       EBML_NEST, 0, 0, {.n=matroska_chapters   } },
420     { MATROSKA_ID_CUES,           EBML_NEST, 0, 0, {.n=matroska_index      } },
421     { MATROSKA_ID_TAGS,           EBML_NEST, 0, 0, {.n=matroska_tags       } },
422     { MATROSKA_ID_SEEKHEAD,       EBML_NEST, 0, 0, {.n=matroska_seekhead   } },
423     { MATROSKA_ID_CLUSTER,        EBML_STOP, 0, offsetof(MatroskaDemuxContext,has_cluster_id) },
424     { EBML_ID_VOID,               EBML_NONE },
425     { 0 }
426 };
427
428 static EbmlSyntax matroska_segments[] = {
429     { MATROSKA_ID_SEGMENT,        EBML_NEST, 0, 0, {.n=matroska_segment    } },
430     { 0 }
431 };
432
433 static EbmlSyntax matroska_blockgroup[] = {
434     { MATROSKA_ID_BLOCK,          EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
435     { MATROSKA_ID_SIMPLEBLOCK,    EBML_BIN,  0, offsetof(MatroskaBlock,bin) },
436     { MATROSKA_ID_BLOCKDURATION,  EBML_UINT, 0, offsetof(MatroskaBlock,duration), {.u=AV_NOPTS_VALUE} },
437     { MATROSKA_ID_BLOCKREFERENCE, EBML_UINT, 0, offsetof(MatroskaBlock,reference) },
438     { EBML_ID_VOID,               EBML_NONE },
439     { 0 }
440 };
441
442 static EbmlSyntax matroska_cluster[] = {
443     { MATROSKA_ID_CLUSTERTIMECODE,EBML_UINT,0, offsetof(MatroskaCluster,timecode) },
444     { MATROSKA_ID_BLOCKGROUP,     EBML_NEST, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
445     { MATROSKA_ID_SIMPLEBLOCK,    EBML_PASS, sizeof(MatroskaBlock), offsetof(MatroskaCluster,blocks), {.n=matroska_blockgroup} },
446     { EBML_ID_VOID,               EBML_NONE },
447     { 0 }
448 };
449
450 static EbmlSyntax matroska_clusters[] = {
451     { MATROSKA_ID_CLUSTER,        EBML_NEST, 0, 0, {.n=matroska_cluster} },
452     { 0 }
453 };
454
455 /*
456  * Return: whether we reached the end of a level in the hierarchy or not
457  */
458 static int ebml_level_end(MatroskaDemuxContext *matroska)
459 {
460     ByteIOContext *pb = matroska->ctx->pb;
461     offset_t pos = url_ftell(pb);
462
463     if (matroska->num_levels > 0) {
464         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
465         if (pos - level->start >= level->length) {
466             matroska->num_levels--;
467             return 1;
468         }
469     }
470     return 0;
471 }
472
473 /*
474  * Read: an "EBML number", which is defined as a variable-length
475  * array of bytes. The first byte indicates the length by giving a
476  * number of 0-bits followed by a one. The position of the first
477  * "one" bit inside the first byte indicates the length of this
478  * number.
479  * Returns: num. of bytes read. < 0 on error.
480  */
481 static int ebml_read_num(MatroskaDemuxContext *matroska, ByteIOContext *pb,
482                          int max_size, uint64_t *number)
483 {
484     int len_mask = 0x80, read = 1, n = 1;
485     int64_t total = 0;
486
487     /* the first byte tells us the length in bytes - get_byte() can normally
488      * return 0, but since that's not a valid first ebmlID byte, we can
489      * use it safely here to catch EOS. */
490     if (!(total = get_byte(pb))) {
491         /* we might encounter EOS here */
492         if (!url_feof(pb)) {
493             offset_t pos = url_ftell(pb);
494             av_log(matroska->ctx, AV_LOG_ERROR,
495                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
496                    pos, pos);
497         }
498         return AVERROR(EIO); /* EOS or actual I/O error */
499     }
500
501     /* get the length of the EBML number */
502     while (read <= max_size && !(total & len_mask)) {
503         read++;
504         len_mask >>= 1;
505     }
506     if (read > max_size) {
507         offset_t pos = url_ftell(pb) - 1;
508         av_log(matroska->ctx, AV_LOG_ERROR,
509                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
510                (uint8_t) total, pos, pos);
511         return AVERROR_INVALIDDATA;
512     }
513
514     /* read out length */
515     total &= ~len_mask;
516     while (n++ < read)
517         total = (total << 8) | get_byte(pb);
518
519     *number = total;
520
521     return read;
522 }
523
524 /*
525  * Read: the element content data ID.
526  * 0 is success, < 0 is failure.
527  */
528 static int ebml_read_element_id(MatroskaDemuxContext *matroska, uint32_t *id)
529 {
530     int read;
531     uint64_t total;
532
533     /* read out the "EBML number", include tag in ID */
534     if ((read = ebml_read_num(matroska, matroska->ctx->pb, 4, &total)) < 0)
535         return read;
536     *id = total | (1 << (read * 7));
537
538     return 0;
539 }
540
541 /*
542  * Read the next element as an unsigned int.
543  * 0 is success, < 0 is failure.
544  */
545 static int ebml_read_uint(ByteIOContext *pb, int size, uint64_t *num)
546 {
547     int n = 0;
548
549     if (size < 1 || size > 8)
550         return AVERROR_INVALIDDATA;
551
552     /* big-endian ordening; build up number */
553     *num = 0;
554     while (n++ < size)
555         *num = (*num << 8) | get_byte(pb);
556
557     return 0;
558 }
559
560 /*
561  * Read the next element as a float.
562  * 0 is success, < 0 is failure.
563  */
564 static int ebml_read_float(ByteIOContext *pb, int size, double *num)
565 {
566     if (size == 4) {
567         *num= av_int2flt(get_be32(pb));
568     } else if(size==8){
569         *num= av_int2dbl(get_be64(pb));
570     } else
571         return AVERROR_INVALIDDATA;
572
573     return 0;
574 }
575
576 /*
577  * Read the next element as an ASCII string.
578  * 0 is success, < 0 is failure.
579  */
580 static int ebml_read_ascii(ByteIOContext *pb, int size, char **str)
581 {
582     av_free(*str);
583     /* ebml strings are usually not 0-terminated, so we allocate one
584      * byte more, read the string and NULL-terminate it ourselves. */
585     if (!(*str = av_malloc(size + 1)))
586         return AVERROR(ENOMEM);
587     if (get_buffer(pb, (uint8_t *) *str, size) != size) {
588         av_free(*str);
589         return AVERROR(EIO);
590     }
591     (*str)[size] = '\0';
592
593     return 0;
594 }
595
596 /*
597  * Read the next element as binary data.
598  * 0 is success, < 0 is failure.
599  */
600 static int ebml_read_binary(ByteIOContext *pb, int length, EbmlBin *bin)
601 {
602     av_free(bin->data);
603     if (!(bin->data = av_malloc(length)))
604         return AVERROR(ENOMEM);
605
606     bin->size = length;
607     bin->pos  = url_ftell(pb);
608     if (get_buffer(pb, bin->data, length) != length)
609         return AVERROR(EIO);
610
611     return 0;
612 }
613
614 /*
615  * Read the next element, but only the header. The contents
616  * are supposed to be sub-elements which can be read separately.
617  * 0 is success, < 0 is failure.
618  */
619 static int ebml_read_master(MatroskaDemuxContext *matroska, int length)
620 {
621     ByteIOContext *pb = matroska->ctx->pb;
622     MatroskaLevel *level;
623
624     if (matroska->num_levels >= EBML_MAX_DEPTH) {
625         av_log(matroska->ctx, AV_LOG_ERROR,
626                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
627         return AVERROR(ENOSYS);
628     }
629
630     level = &matroska->levels[matroska->num_levels++];
631     level->start = url_ftell(pb);
632     level->length = length;
633
634     return 0;
635 }
636
637 /*
638  * Read signed/unsigned "EBML" numbers.
639  * Return: number of bytes processed, < 0 on error.
640  */
641 static int matroska_ebmlnum_uint(MatroskaDemuxContext *matroska,
642                                  uint8_t *data, uint32_t size, uint64_t *num)
643 {
644     ByteIOContext pb;
645     init_put_byte(&pb, data, size, 0, NULL, NULL, NULL, NULL);
646     return ebml_read_num(matroska, &pb, 8, num);
647 }
648
649 /*
650  * Same as above, but signed.
651  */
652 static int matroska_ebmlnum_sint(MatroskaDemuxContext *matroska,
653                                  uint8_t *data, uint32_t size, int64_t *num)
654 {
655     uint64_t unum;
656     int res;
657
658     /* read as unsigned number first */
659     if ((res = matroska_ebmlnum_uint(matroska, data, size, &unum)) < 0)
660         return res;
661
662     /* make signed (weird way) */
663     *num = unum - ((1LL << (7*res - 1)) - 1);
664
665     return res;
666 }
667
668 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
669                            EbmlSyntax *syntax, void *data);
670
671 static int ebml_parse_id(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
672                          uint32_t id, void *data)
673 {
674     int i;
675     for (i=0; syntax[i].id; i++)
676         if (id == syntax[i].id)
677             break;
678     if (!syntax[i].id)
679         av_log(matroska->ctx, AV_LOG_INFO, "Unknown entry 0x%X\n", id);
680     return ebml_parse_elem(matroska, &syntax[i], data);
681 }
682
683 static int ebml_parse(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
684                       void *data)
685 {
686     uint32_t id;
687     int res = ebml_read_element_id(matroska, &id);
688     return res < 0 ? res : ebml_parse_id(matroska, syntax, id, data);
689 }
690
691 static int ebml_parse_nest(MatroskaDemuxContext *matroska, EbmlSyntax *syntax,
692                            void *data)
693 {
694     int i, res = 0;
695
696     for (i=0; syntax[i].id; i++)
697         switch (syntax[i].type) {
698         case EBML_UINT:
699             *(uint64_t *)((char *)data+syntax[i].data_offset) = syntax[i].def.u;
700             break;
701         case EBML_FLOAT:
702             *(double   *)((char *)data+syntax[i].data_offset) = syntax[i].def.f;
703             break;
704         case EBML_STR:
705         case EBML_UTF8:
706             *(char    **)((char *)data+syntax[i].data_offset) = av_strdup(syntax[i].def.s);
707             break;
708         }
709
710     while (!res && !ebml_level_end(matroska))
711         res = ebml_parse(matroska, syntax, data);
712
713     return res;
714 }
715
716 static int ebml_parse_elem(MatroskaDemuxContext *matroska,
717                            EbmlSyntax *syntax, void *data)
718 {
719     ByteIOContext *pb = matroska->ctx->pb;
720     uint32_t id = syntax->id;
721     uint64_t length;
722     int res;
723
724     data = (char *)data + syntax->data_offset;
725     if (syntax->list_elem_size) {
726         EbmlList *list = data;
727         list->elem = av_realloc(list->elem, (list->nb_elem+1)*syntax->list_elem_size);
728         data = (char*)list->elem + list->nb_elem*syntax->list_elem_size;
729         memset(data, 0, syntax->list_elem_size);
730         list->nb_elem++;
731     }
732
733     if (syntax->type != EBML_PASS && syntax->type != EBML_STOP)
734         if ((res = ebml_read_num(matroska, pb, 8, &length)) < 0)
735             return res;
736
737     switch (syntax->type) {
738     case EBML_UINT:  res = ebml_read_uint  (pb, length, data);  break;
739     case EBML_FLOAT: res = ebml_read_float (pb, length, data);  break;
740     case EBML_STR:
741     case EBML_UTF8:  res = ebml_read_ascii (pb, length, data);  break;
742     case EBML_BIN:   res = ebml_read_binary(pb, length, data);  break;
743     case EBML_NEST:  if ((res=ebml_read_master(matroska, length)) < 0)
744                          return res;
745                      if (id == MATROSKA_ID_SEGMENT)
746                          matroska->segment_start = url_ftell(matroska->ctx->pb);
747                      return ebml_parse_nest(matroska, syntax->def.n, data);
748     case EBML_PASS:  return ebml_parse_id(matroska, syntax->def.n, id, data);
749     case EBML_STOP:  *(int *)data = 1;      return 1;
750     default:         url_fskip(pb, length); return 0;
751     }
752     if (res == AVERROR_INVALIDDATA)
753         av_log(matroska->ctx, AV_LOG_ERROR, "Invalid element\n");
754     else if (res == AVERROR(EIO))
755         av_log(matroska->ctx, AV_LOG_ERROR, "Read error\n");
756     return res;
757 }
758
759 static void ebml_free(EbmlSyntax *syntax, void *data)
760 {
761     int i, j;
762     for (i=0; syntax[i].id; i++) {
763         void *data_off = (char *)data + syntax[i].data_offset;
764         switch (syntax[i].type) {
765         case EBML_STR:
766         case EBML_UTF8:  av_freep(data_off);                      break;
767         case EBML_BIN:   av_freep(&((EbmlBin *)data_off)->data);  break;
768         case EBML_NEST:
769             if (syntax[i].list_elem_size) {
770                 EbmlList *list = data_off;
771                 char *ptr = list->elem;
772                 for (j=0; j<list->nb_elem; j++, ptr+=syntax[i].list_elem_size)
773                     ebml_free(syntax[i].def.n, ptr);
774                 av_free(list->elem);
775             } else
776                 ebml_free(syntax[i].def.n, data_off);
777         default:  break;
778         }
779     }
780 }
781
782
783 /*
784  * Autodetecting...
785  */
786 static int matroska_probe(AVProbeData *p)
787 {
788     uint64_t total = 0;
789     int len_mask = 0x80, size = 1, n = 1;
790     char probe_data[] = "matroska";
791
792     /* ebml header? */
793     if (AV_RB32(p->buf) != EBML_ID_HEADER)
794         return 0;
795
796     /* length of header */
797     total = p->buf[4];
798     while (size <= 8 && !(total & len_mask)) {
799         size++;
800         len_mask >>= 1;
801     }
802     if (size > 8)
803       return 0;
804     total &= (len_mask - 1);
805     while (n < size)
806         total = (total << 8) | p->buf[4 + n++];
807
808     /* does the probe data contain the whole header? */
809     if (p->buf_size < 4 + size + total)
810       return 0;
811
812     /* the header must contain the document type 'matroska'. For now,
813      * we don't parse the whole header but simply check for the
814      * availability of that array of characters inside the header.
815      * Not fully fool-proof, but good enough. */
816     for (n = 4+size; n <= 4+size+total-(sizeof(probe_data)-1); n++)
817         if (!memcmp(p->buf+n, probe_data, sizeof(probe_data)-1))
818             return AVPROBE_SCORE_MAX;
819
820     return 0;
821 }
822
823 static MatroskaTrack *matroska_find_track_by_num(MatroskaDemuxContext *matroska,
824                                                  int num)
825 {
826     MatroskaTrack *tracks = matroska->tracks.elem;
827     int i;
828
829     for (i=0; i < matroska->tracks.nb_elem; i++)
830         if (tracks[i].num == num)
831             return &tracks[i];
832
833     av_log(matroska->ctx, AV_LOG_ERROR, "Invalid track number %d\n", num);
834     return NULL;
835 }
836
837 static int matroska_decode_buffer(uint8_t** buf, int* buf_size,
838                                   MatroskaTrack *track)
839 {
840     MatroskaTrackEncoding *encodings = track->encodings.elem;
841     uint8_t* data = *buf;
842     int isize = *buf_size;
843     uint8_t* pkt_data = NULL;
844     int pkt_size = isize;
845     int result = 0;
846     int olen;
847
848     switch (encodings[0].compression.algo) {
849     case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP:
850         return encodings[0].compression.settings.size;
851     case MATROSKA_TRACK_ENCODING_COMP_LZO:
852         do {
853             olen = pkt_size *= 3;
854             pkt_data = av_realloc(pkt_data,
855                                   pkt_size+LZO_OUTPUT_PADDING);
856             result = lzo1x_decode(pkt_data, &olen, data, &isize);
857         } while (result==LZO_OUTPUT_FULL && pkt_size<10000000);
858         if (result)
859             goto failed;
860         pkt_size -= olen;
861         break;
862 #ifdef CONFIG_ZLIB
863     case MATROSKA_TRACK_ENCODING_COMP_ZLIB: {
864         z_stream zstream = {0};
865         if (inflateInit(&zstream) != Z_OK)
866             return -1;
867         zstream.next_in = data;
868         zstream.avail_in = isize;
869         do {
870             pkt_size *= 3;
871             pkt_data = av_realloc(pkt_data, pkt_size);
872             zstream.avail_out = pkt_size - zstream.total_out;
873             zstream.next_out = pkt_data + zstream.total_out;
874             result = inflate(&zstream, Z_NO_FLUSH);
875         } while (result==Z_OK && pkt_size<10000000);
876         pkt_size = zstream.total_out;
877         inflateEnd(&zstream);
878         if (result != Z_STREAM_END)
879             goto failed;
880         break;
881     }
882 #endif
883 #ifdef CONFIG_BZLIB
884     case MATROSKA_TRACK_ENCODING_COMP_BZLIB: {
885         bz_stream bzstream = {0};
886         if (BZ2_bzDecompressInit(&bzstream, 0, 0) != BZ_OK)
887             return -1;
888         bzstream.next_in = data;
889         bzstream.avail_in = isize;
890         do {
891             pkt_size *= 3;
892             pkt_data = av_realloc(pkt_data, pkt_size);
893             bzstream.avail_out = pkt_size - bzstream.total_out_lo32;
894             bzstream.next_out = pkt_data + bzstream.total_out_lo32;
895             result = BZ2_bzDecompress(&bzstream);
896         } while (result==BZ_OK && pkt_size<10000000);
897         pkt_size = bzstream.total_out_lo32;
898         BZ2_bzDecompressEnd(&bzstream);
899         if (result != BZ_STREAM_END)
900             goto failed;
901         break;
902     }
903 #endif
904     }
905
906     *buf = pkt_data;
907     *buf_size = pkt_size;
908     return 0;
909  failed:
910     av_free(pkt_data);
911     return -1;
912 }
913
914 static void matroska_execute_seekhead(MatroskaDemuxContext *matroska)
915 {
916     EbmlList *seekhead_list = &matroska->seekhead;
917     MatroskaSeekhead *seekhead = seekhead_list->elem;
918     uint32_t level_up = matroska->level_up;
919     offset_t before_pos = url_ftell(matroska->ctx->pb);
920     MatroskaLevel level;
921     int i;
922
923     for (i=0; i<seekhead_list->nb_elem; i++) {
924         offset_t offset = seekhead[i].pos + matroska->segment_start;
925
926         if (seekhead[i].pos <= before_pos
927             || seekhead[i].id == MATROSKA_ID_SEEKHEAD
928             || seekhead[i].id == MATROSKA_ID_CLUSTER)
929             continue;
930
931         /* seek */
932         if (url_fseek(matroska->ctx->pb, offset, SEEK_SET) != offset)
933             continue;
934
935         /* we don't want to lose our seekhead level, so we add
936          * a dummy. This is a crude hack. */
937         if (matroska->num_levels == EBML_MAX_DEPTH) {
938             av_log(matroska->ctx, AV_LOG_INFO,
939                    "Max EBML element depth (%d) reached, "
940                    "cannot parse further.\n", EBML_MAX_DEPTH);
941             break;
942         }
943
944         level.start = 0;
945         level.length = (uint64_t)-1;
946         matroska->levels[matroska->num_levels] = level;
947         matroska->num_levels++;
948
949         ebml_parse(matroska, matroska_segment, matroska);
950
951         /* remove dummy level */
952         while (matroska->num_levels) {
953             uint64_t length = matroska->levels[--matroska->num_levels].length;
954             if (length == (uint64_t)-1)
955                 break;
956         }
957     }
958
959     /* seek back */
960     url_fseek(matroska->ctx->pb, before_pos, SEEK_SET);
961     matroska->level_up = level_up;
962 }
963
964 static int matroska_aac_profile(char *codec_id)
965 {
966     static const char *aac_profiles[] = { "MAIN", "LC", "SSR" };
967     int profile;
968
969     for (profile=0; profile<ARRAY_SIZE(aac_profiles); profile++)
970         if (strstr(codec_id, aac_profiles[profile]))
971             break;
972     return profile + 1;
973 }
974
975 static int matroska_aac_sri(int samplerate)
976 {
977     int sri;
978
979     for (sri=0; sri<ARRAY_SIZE(ff_mpeg4audio_sample_rates); sri++)
980         if (ff_mpeg4audio_sample_rates[sri] == samplerate)
981             break;
982     return sri;
983 }
984
985 static int matroska_read_header(AVFormatContext *s, AVFormatParameters *ap)
986 {
987     MatroskaDemuxContext *matroska = s->priv_data;
988     EbmlList *attachements_list = &matroska->attachments;
989     MatroskaAttachement *attachements;
990     EbmlList *chapters_list = &matroska->chapters;
991     MatroskaChapter *chapters;
992     MatroskaTrack *tracks;
993     EbmlList *index_list;
994     MatroskaIndex *index;
995     Ebml ebml = { 0 };
996     AVStream *st;
997     int i, j;
998
999     matroska->ctx = s;
1000
1001     /* First read the EBML header. */
1002     if (ebml_parse(matroska, ebml_syntax, &ebml)
1003         || ebml.version > EBML_VERSION       || ebml.max_size > sizeof(uint64_t)
1004         || ebml.id_length > sizeof(uint32_t) || strcmp(ebml.doctype, "matroska")
1005         || ebml.doctype_version > 2) {
1006         av_log(matroska->ctx, AV_LOG_ERROR,
1007                "EBML header using unsupported features\n"
1008                "(EBML version %"PRIu64", doctype %s, doc version %"PRIu64")\n",
1009                ebml.version, ebml.doctype, ebml.doctype_version);
1010         return AVERROR_NOFMT;
1011     }
1012     ebml_free(ebml_syntax, &ebml);
1013
1014     /* The next thing is a segment. */
1015     if (ebml_parse(matroska, matroska_segments, matroska) < 0)
1016         return -1;
1017     matroska_execute_seekhead(matroska);
1018
1019     if (matroska->duration)
1020         matroska->ctx->duration = matroska->duration * matroska->time_scale
1021                                   * 1000 / AV_TIME_BASE;
1022     if (matroska->title)
1023         strncpy(matroska->ctx->title, matroska->title,
1024                 sizeof(matroska->ctx->title)-1);
1025
1026     tracks = matroska->tracks.elem;
1027     for (i=0; i < matroska->tracks.nb_elem; i++) {
1028         MatroskaTrack *track = &tracks[i];
1029         enum CodecID codec_id = CODEC_ID_NONE;
1030         EbmlList *encodings_list = &tracks->encodings;
1031         MatroskaTrackEncoding *encodings = encodings_list->elem;
1032         uint8_t *extradata = NULL;
1033         int extradata_size = 0;
1034         int extradata_offset = 0;
1035
1036         /* Apply some sanity checks. */
1037         if (track->type != MATROSKA_TRACK_TYPE_VIDEO &&
1038             track->type != MATROSKA_TRACK_TYPE_AUDIO &&
1039             track->type != MATROSKA_TRACK_TYPE_SUBTITLE) {
1040             av_log(matroska->ctx, AV_LOG_INFO,
1041                    "Unknown or unsupported track type %"PRIu64"\n",
1042                    track->type);
1043             continue;
1044         }
1045         if (track->codec_id == NULL)
1046             continue;
1047
1048         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1049             if (!track->default_duration)
1050                 track->default_duration = 1000000000/track->video.frame_rate;
1051             if (!track->video.display_width)
1052                 track->video.display_width = track->video.pixel_width;
1053             if (!track->video.display_height)
1054                 track->video.display_height = track->video.pixel_height;
1055         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1056             if (!track->audio.out_samplerate)
1057                 track->audio.out_samplerate = track->audio.samplerate;
1058         }
1059         if (encodings_list->nb_elem > 1) {
1060             av_log(matroska->ctx, AV_LOG_ERROR,
1061                    "Multiple combined encodings no supported");
1062         } else if (encodings_list->nb_elem == 1) {
1063             if (encodings[0].type ||
1064                 (encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP &&
1065 #ifdef CONFIG_ZLIB
1066                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_ZLIB &&
1067 #endif
1068 #ifdef CONFIG_BZLIB
1069                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_BZLIB &&
1070 #endif
1071                  encodings[0].compression.algo != MATROSKA_TRACK_ENCODING_COMP_LZO)) {
1072                 encodings[0].scope = 0;
1073                 av_log(matroska->ctx, AV_LOG_ERROR,
1074                        "Unsupported encoding type");
1075             } else if (track->codec_priv.size && encodings[0].scope&2) {
1076                 uint8_t *codec_priv = track->codec_priv.data;
1077                 int offset = matroska_decode_buffer(&track->codec_priv.data,
1078                                                     &track->codec_priv.size,
1079                                                     track);
1080                 if (offset < 0) {
1081                     track->codec_priv.data = NULL;
1082                     track->codec_priv.size = 0;
1083                     av_log(matroska->ctx, AV_LOG_ERROR,
1084                            "Failed to decode codec private data\n");
1085                 } else if (offset > 0) {
1086                     track->codec_priv.data = av_malloc(track->codec_priv.size + offset);
1087                     memcpy(track->codec_priv.data,
1088                            encodings[0].compression.settings.data, offset);
1089                     memcpy(track->codec_priv.data+offset, codec_priv,
1090                            track->codec_priv.size);
1091                     track->codec_priv.size += offset;
1092                 }
1093                 if (codec_priv != track->codec_priv.data)
1094                     av_free(codec_priv);
1095             }
1096         }
1097
1098         for(j=0; ff_mkv_codec_tags[j].id != CODEC_ID_NONE; j++){
1099             if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
1100                         strlen(ff_mkv_codec_tags[j].str))){
1101                 codec_id= ff_mkv_codec_tags[j].id;
1102                 break;
1103             }
1104         }
1105
1106         st = track->stream = av_new_stream(s, matroska->num_streams++);
1107         if (st == NULL)
1108             return AVERROR(ENOMEM);
1109
1110         if (!strcmp(track->codec_id, "V_MS/VFW/FOURCC")
1111             && track->codec_priv.size >= 40
1112             && track->codec_priv.data != NULL) {
1113             track->video.fourcc = AV_RL32(track->codec_priv.data + 16);
1114             codec_id = codec_get_id(codec_bmp_tags, track->video.fourcc);
1115         } else if (!strcmp(track->codec_id, "A_MS/ACM")
1116                    && track->codec_priv.size >= 18
1117                    && track->codec_priv.data != NULL) {
1118             uint16_t tag = AV_RL16(track->codec_priv.data);
1119             codec_id = codec_get_id(codec_wav_tags, tag);
1120         } else if (!strcmp(track->codec_id, "V_QUICKTIME")
1121                    && (track->codec_priv.size >= 86)
1122                    && (track->codec_priv.data != NULL)) {
1123             track->video.fourcc = AV_RL32(track->codec_priv.data);
1124             codec_id=codec_get_id(codec_movvideo_tags, track->video.fourcc);
1125         } else if (codec_id == CODEC_ID_AAC && !track->codec_priv.size) {
1126             int profile = matroska_aac_profile(track->codec_id);
1127             int sri = matroska_aac_sri(track->audio.samplerate);
1128             extradata = av_malloc(5);
1129             if (extradata == NULL)
1130                 return AVERROR(ENOMEM);
1131             extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
1132             extradata[1] = ((sri&0x01) << 7) | (track->audio.channels<<3);
1133             if (strstr(track->codec_id, "SBR")) {
1134                 sri = matroska_aac_sri(track->audio.out_samplerate);
1135                 extradata[2] = 0x56;
1136                 extradata[3] = 0xE5;
1137                 extradata[4] = 0x80 | (sri<<3);
1138                 extradata_size = 5;
1139             } else
1140                 extradata_size = 2;
1141         } else if (codec_id == CODEC_ID_TTA) {
1142             ByteIOContext b;
1143             extradata_size = 30;
1144             extradata = av_mallocz(extradata_size);
1145             if (extradata == NULL)
1146                 return AVERROR(ENOMEM);
1147             init_put_byte(&b, extradata, extradata_size, 1,
1148                           NULL, NULL, NULL, NULL);
1149             put_buffer(&b, "TTA1", 4);
1150             put_le16(&b, 1);
1151             put_le16(&b, track->audio.channels);
1152             put_le16(&b, track->audio.bitdepth);
1153             put_le32(&b, track->audio.out_samplerate);
1154             put_le32(&b, matroska->ctx->duration * track->audio.out_samplerate);
1155         } else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
1156                    codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
1157             extradata_offset = 26;
1158             track->codec_priv.size -= extradata_offset;
1159         } else if (codec_id == CODEC_ID_RA_144) {
1160             track->audio.out_samplerate = 8000;
1161             track->audio.channels = 1;
1162         } else if (codec_id == CODEC_ID_RA_288 || codec_id == CODEC_ID_COOK ||
1163                    codec_id == CODEC_ID_ATRAC3) {
1164             ByteIOContext b;
1165
1166             init_put_byte(&b, track->codec_priv.data,track->codec_priv.size,
1167                           0, NULL, NULL, NULL, NULL);
1168             url_fskip(&b, 24);
1169             track->audio.coded_framesize = get_be32(&b);
1170             url_fskip(&b, 12);
1171             track->audio.sub_packet_h    = get_be16(&b);
1172             track->audio.frame_size      = get_be16(&b);
1173             track->audio.sub_packet_size = get_be16(&b);
1174             track->audio.buf = av_malloc(track->audio.frame_size * track->audio.sub_packet_h);
1175             if (codec_id == CODEC_ID_RA_288) {
1176                 st->codec->block_align = track->audio.coded_framesize;
1177                 track->codec_priv.size = 0;
1178             } else {
1179                 st->codec->block_align = track->audio.sub_packet_size;
1180                 extradata_offset = 78;
1181                 track->codec_priv.size -= extradata_offset;
1182             }
1183         }
1184
1185         if (codec_id == CODEC_ID_NONE)
1186             av_log(matroska->ctx, AV_LOG_INFO,
1187                    "Unknown/unsupported CodecID %s.\n", track->codec_id);
1188
1189         av_set_pts_info(st, 64, matroska->time_scale*track->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
1190
1191         st->codec->codec_id = codec_id;
1192         st->start_time = 0;
1193         if (strcmp(track->language, "und"))
1194             av_strlcpy(st->language, track->language, 4);
1195
1196         if (track->flag_default)
1197             st->disposition |= AV_DISPOSITION_DEFAULT;
1198
1199         if (track->default_duration)
1200             av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
1201                       track->default_duration, 1000000000, 30000);
1202
1203         if(extradata){
1204             st->codec->extradata = extradata;
1205             st->codec->extradata_size = extradata_size;
1206         } else if(track->codec_priv.data && track->codec_priv.size > 0){
1207             st->codec->extradata = av_malloc(track->codec_priv.size);
1208             if(st->codec->extradata == NULL)
1209                 return AVERROR(ENOMEM);
1210             st->codec->extradata_size = track->codec_priv.size;
1211             memcpy(st->codec->extradata,
1212                    track->codec_priv.data + extradata_offset,
1213                    track->codec_priv.size);
1214         }
1215
1216         if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
1217             st->codec->codec_type = CODEC_TYPE_VIDEO;
1218             st->codec->codec_tag  = track->video.fourcc;
1219             st->codec->width  = track->video.pixel_width;
1220             st->codec->height = track->video.pixel_height;
1221             av_reduce(&st->codec->sample_aspect_ratio.num,
1222                       &st->codec->sample_aspect_ratio.den,
1223                       st->codec->height * track->video.display_width,
1224                       st->codec-> width * track->video.display_height,
1225                       255);
1226             st->need_parsing = AVSTREAM_PARSE_HEADERS;
1227         } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
1228             st->codec->codec_type = CODEC_TYPE_AUDIO;
1229             st->codec->sample_rate = track->audio.out_samplerate;
1230             st->codec->channels = track->audio.channels;
1231         } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
1232             st->codec->codec_type = CODEC_TYPE_SUBTITLE;
1233         }
1234     }
1235
1236     attachements = attachements_list->elem;
1237     for (j=0; j<attachements_list->nb_elem; j++) {
1238         if (!(attachements[j].filename && attachements[j].mime &&
1239               attachements[j].bin.data && attachements[j].bin.size > 0)) {
1240             av_log(matroska->ctx, AV_LOG_ERROR, "incomplete attachment\n");
1241         } else {
1242             AVStream *st = av_new_stream(s, matroska->num_streams++);
1243             if (st == NULL)
1244                 break;
1245             st->filename          = av_strdup(attachements[j].filename);
1246             st->codec->codec_id = CODEC_ID_NONE;
1247             st->codec->codec_type = CODEC_TYPE_ATTACHMENT;
1248             st->codec->extradata  = av_malloc(attachements[j].bin.size);
1249             if(st->codec->extradata == NULL)
1250                 break;
1251             st->codec->extradata_size = attachements[j].bin.size;
1252             memcpy(st->codec->extradata, attachements[j].bin.data, attachements[j].bin.size);
1253
1254             for (i=0; ff_mkv_mime_tags[i].id != CODEC_ID_NONE; i++) {
1255                 if (!strncmp(ff_mkv_mime_tags[i].str, attachements[j].mime,
1256                              strlen(ff_mkv_mime_tags[i].str))) {
1257                     st->codec->codec_id = ff_mkv_mime_tags[i].id;
1258                     break;
1259                 }
1260             }
1261         }
1262     }
1263
1264     chapters = chapters_list->elem;
1265     for (i=0; i<chapters_list->nb_elem; i++)
1266         if (chapters[i].start != AV_NOPTS_VALUE && chapters[i].uid)
1267             ff_new_chapter(s, chapters[i].uid, (AVRational){1, 1000000000},
1268                            chapters[i].start, chapters[i].end,
1269                            chapters[i].title);
1270
1271     index_list = &matroska->index;
1272     index = index_list->elem;
1273     for (i=0; i<index_list->nb_elem; i++) {
1274         EbmlList *pos_list = &index[i].pos;
1275         MatroskaIndexPos *pos = pos_list->elem;
1276         for (j=0; j<pos_list->nb_elem; j++) {
1277             MatroskaTrack *track = matroska_find_track_by_num(matroska,
1278                                                               pos[j].track);
1279             if (track && track->stream)
1280                 av_add_index_entry(track->stream,
1281                                    pos[j].pos + matroska->segment_start,
1282                                    index[i].time*matroska->time_scale/AV_TIME_BASE,
1283                                    0, 0, AVINDEX_KEYFRAME);
1284         }
1285     }
1286
1287     return 0;
1288 }
1289
1290 /*
1291  * Put a packet into our internal queue. Will be delivered to the
1292  * user/application during the next get_packet() call.
1293  */
1294 static void matroska_queue_packet(MatroskaDemuxContext *matroska, AVPacket *pkt)
1295 {
1296     matroska->packets =
1297         av_realloc(matroska->packets, (matroska->num_packets + 1) *
1298                    sizeof(AVPacket *));
1299     matroska->packets[matroska->num_packets] = pkt;
1300     matroska->num_packets++;
1301 }
1302
1303 /*
1304  * Put one packet in an application-supplied AVPacket struct.
1305  * Returns 0 on success or -1 on failure.
1306  */
1307 static int matroska_deliver_packet(MatroskaDemuxContext *matroska,
1308                                    AVPacket *pkt)
1309 {
1310     if (matroska->num_packets > 0) {
1311         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
1312         av_free(matroska->packets[0]);
1313         if (matroska->num_packets > 1) {
1314             memmove(&matroska->packets[0], &matroska->packets[1],
1315                     (matroska->num_packets - 1) * sizeof(AVPacket *));
1316             matroska->packets =
1317                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
1318                            sizeof(AVPacket *));
1319         } else {
1320             av_freep(&matroska->packets);
1321         }
1322         matroska->num_packets--;
1323         return 0;
1324     }
1325
1326     return -1;
1327 }
1328
1329 /*
1330  * Free all packets in our internal queue.
1331  */
1332 static void matroska_clear_queue(MatroskaDemuxContext *matroska)
1333 {
1334     if (matroska->packets) {
1335         int n;
1336         for (n = 0; n < matroska->num_packets; n++) {
1337             av_free_packet(matroska->packets[n]);
1338             av_free(matroska->packets[n]);
1339         }
1340         av_free(matroska->packets);
1341         matroska->packets = NULL;
1342         matroska->num_packets = 0;
1343     }
1344 }
1345
1346 static int matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data,
1347                                 int size, int64_t pos, uint64_t cluster_time,
1348                                 uint64_t duration, int is_keyframe)
1349 {
1350     MatroskaTrack *track;
1351     int res = 0;
1352     AVStream *st;
1353     AVPacket *pkt;
1354     int16_t block_time;
1355     uint32_t *lace_size = NULL;
1356     int n, flags, laces = 0;
1357     uint64_t num;
1358
1359     if ((n = matroska_ebmlnum_uint(matroska, data, size, &num)) < 0) {
1360         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
1361         return res;
1362     }
1363     data += n;
1364     size -= n;
1365
1366     track = matroska_find_track_by_num(matroska, num);
1367     if (size <= 3 || !track || !track->stream) {
1368         av_log(matroska->ctx, AV_LOG_INFO,
1369                "Invalid stream %"PRIu64" or size %u\n", num, size);
1370         return res;
1371     }
1372     st = track->stream;
1373     if (st->discard >= AVDISCARD_ALL)
1374         return res;
1375     if (duration == AV_NOPTS_VALUE)
1376         duration = track->default_duration / matroska->time_scale;
1377
1378     block_time = AV_RB16(data);
1379     data += 2;
1380     flags = *data++;
1381     size -= 3;
1382     if (is_keyframe == -1)
1383         is_keyframe = flags & 0x80 ? PKT_FLAG_KEY : 0;
1384
1385     if (matroska->skip_to_keyframe) {
1386         if (!is_keyframe || st != matroska->skip_to_stream)
1387             return res;
1388         matroska->skip_to_keyframe = 0;
1389     }
1390
1391     switch ((flags & 0x06) >> 1) {
1392         case 0x0: /* no lacing */
1393             laces = 1;
1394             lace_size = av_mallocz(sizeof(int));
1395             lace_size[0] = size;
1396             break;
1397
1398         case 0x1: /* xiph lacing */
1399         case 0x2: /* fixed-size lacing */
1400         case 0x3: /* EBML lacing */
1401             assert(size>0); // size <=3 is checked before size-=3 above
1402             laces = (*data) + 1;
1403             data += 1;
1404             size -= 1;
1405             lace_size = av_mallocz(laces * sizeof(int));
1406
1407             switch ((flags & 0x06) >> 1) {
1408                 case 0x1: /* xiph lacing */ {
1409                     uint8_t temp;
1410                     uint32_t total = 0;
1411                     for (n = 0; res == 0 && n < laces - 1; n++) {
1412                         while (1) {
1413                             if (size == 0) {
1414                                 res = -1;
1415                                 break;
1416                             }
1417                             temp = *data;
1418                             lace_size[n] += temp;
1419                             data += 1;
1420                             size -= 1;
1421                             if (temp != 0xff)
1422                                 break;
1423                         }
1424                         total += lace_size[n];
1425                     }
1426                     lace_size[n] = size - total;
1427                     break;
1428                 }
1429
1430                 case 0x2: /* fixed-size lacing */
1431                     for (n = 0; n < laces; n++)
1432                         lace_size[n] = size / laces;
1433                     break;
1434
1435                 case 0x3: /* EBML lacing */ {
1436                     uint32_t total;
1437                     n = matroska_ebmlnum_uint(matroska, data, size, &num);
1438                     if (n < 0) {
1439                         av_log(matroska->ctx, AV_LOG_INFO,
1440                                "EBML block data error\n");
1441                         break;
1442                     }
1443                     data += n;
1444                     size -= n;
1445                     total = lace_size[0] = num;
1446                     for (n = 1; res == 0 && n < laces - 1; n++) {
1447                         int64_t snum;
1448                         int r;
1449                         r = matroska_ebmlnum_sint(matroska, data, size, &snum);
1450                         if (r < 0) {
1451                             av_log(matroska->ctx, AV_LOG_INFO,
1452                                    "EBML block data error\n");
1453                             break;
1454                         }
1455                         data += r;
1456                         size -= r;
1457                         lace_size[n] = lace_size[n - 1] + snum;
1458                         total += lace_size[n];
1459                     }
1460                     lace_size[n] = size - total;
1461                     break;
1462                 }
1463             }
1464             break;
1465     }
1466
1467     if (res == 0) {
1468         uint64_t timecode = AV_NOPTS_VALUE;
1469
1470         if (cluster_time != (uint64_t)-1
1471             && (block_time >= 0 || cluster_time >= -block_time))
1472             timecode = cluster_time + block_time;
1473
1474         for (n = 0; n < laces; n++) {
1475             if (st->codec->codec_id == CODEC_ID_RA_288 ||
1476                 st->codec->codec_id == CODEC_ID_COOK ||
1477                 st->codec->codec_id == CODEC_ID_ATRAC3) {
1478                 int a = st->codec->block_align;
1479                 int sps = track->audio.sub_packet_size;
1480                 int cfs = track->audio.coded_framesize;
1481                 int h = track->audio.sub_packet_h;
1482                 int y = track->audio.sub_packet_cnt;
1483                 int w = track->audio.frame_size;
1484                 int x;
1485
1486                 if (!track->audio.pkt_cnt) {
1487                     if (st->codec->codec_id == CODEC_ID_RA_288)
1488                         for (x=0; x<h/2; x++)
1489                             memcpy(track->audio.buf+x*2*w+y*cfs,
1490                                    data+x*cfs, cfs);
1491                     else
1492                         for (x=0; x<w/sps; x++)
1493                             memcpy(track->audio.buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
1494
1495                     if (++track->audio.sub_packet_cnt >= h) {
1496                         track->audio.sub_packet_cnt = 0;
1497                         track->audio.pkt_cnt = h*w / a;
1498                     }
1499                 }
1500                 while (track->audio.pkt_cnt) {
1501                     pkt = av_mallocz(sizeof(AVPacket));
1502                     av_new_packet(pkt, a);
1503                     memcpy(pkt->data, track->audio.buf
1504                            + a * (h*w / a - track->audio.pkt_cnt--), a);
1505                     pkt->pos = pos;
1506                     pkt->stream_index = st->index;
1507                     matroska_queue_packet(matroska, pkt);
1508                 }
1509             } else {
1510                 MatroskaTrackEncoding *encodings = track->encodings.elem;
1511                 int offset = 0, pkt_size = lace_size[n];
1512                 uint8_t *pkt_data = data;
1513
1514                 if (encodings && encodings->scope & 1) {
1515                     offset = matroska_decode_buffer(&pkt_data,&pkt_size, track);
1516                     if (offset < 0)
1517                         continue;
1518                 }
1519
1520                 pkt = av_mallocz(sizeof(AVPacket));
1521                 /* XXX: prevent data copy... */
1522                 if (av_new_packet(pkt, pkt_size+offset) < 0) {
1523                     av_free(pkt);
1524                     res = AVERROR(ENOMEM);
1525                     n = laces-1;
1526                     break;
1527                 }
1528                 if (offset)
1529                     memcpy (pkt->data, encodings->compression.settings.data, offset);
1530                 memcpy (pkt->data+offset, pkt_data, pkt_size);
1531
1532                 if (pkt_data != data)
1533                     av_free(pkt_data);
1534
1535                 if (n == 0)
1536                     pkt->flags = is_keyframe;
1537                 pkt->stream_index = st->index;
1538
1539                 pkt->pts = timecode;
1540                 pkt->pos = pos;
1541                 pkt->duration = duration;
1542
1543                 matroska_queue_packet(matroska, pkt);
1544             }
1545
1546             if (timecode != AV_NOPTS_VALUE)
1547                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
1548             data += lace_size[n];
1549         }
1550     }
1551
1552     av_free(lace_size);
1553     return res;
1554 }
1555
1556 static int matroska_parse_cluster(MatroskaDemuxContext *matroska)
1557 {
1558     MatroskaCluster cluster = { 0 };
1559     EbmlList *blocks_list;
1560     MatroskaBlock *blocks;
1561     int i, res;
1562     if (matroska->has_cluster_id){
1563         /* For the first cluster we parse, it's ID was already read as
1564            part of matroska_read_header(), so don't read it again */
1565         res = ebml_parse_id(matroska, matroska_clusters,
1566                             MATROSKA_ID_CLUSTER, &cluster);
1567         matroska->has_cluster_id = 0;
1568     } else
1569         res = ebml_parse(matroska, matroska_clusters, &cluster);
1570     blocks_list = &cluster.blocks;
1571     blocks = blocks_list->elem;
1572     for (i=0; !res && i<blocks_list->nb_elem; i++)
1573         if (blocks[i].bin.size > 0)
1574             res=matroska_parse_block(matroska,
1575                                      blocks[i].bin.data, blocks[i].bin.size,
1576                                      blocks[i].bin.pos,  cluster.timecode,
1577                                      blocks[i].duration, !blocks[i].reference);
1578     ebml_free(matroska_cluster, &cluster);
1579     return res;
1580 }
1581
1582 static int matroska_read_packet(AVFormatContext *s, AVPacket *pkt)
1583 {
1584     MatroskaDemuxContext *matroska = s->priv_data;
1585
1586     while (matroska_deliver_packet(matroska, pkt)) {
1587         if (matroska->done)
1588             return AVERROR(EIO);
1589         if (matroska_parse_cluster(matroska) < 0)
1590             matroska->done = 1;
1591     }
1592
1593     return 0;
1594 }
1595
1596 static int matroska_read_seek(AVFormatContext *s, int stream_index,
1597                               int64_t timestamp, int flags)
1598 {
1599     MatroskaDemuxContext *matroska = s->priv_data;
1600     AVStream *st = s->streams[stream_index];
1601     int index;
1602
1603     index = av_index_search_timestamp(st, timestamp, flags);
1604     if (index < 0)
1605         return 0;
1606
1607     matroska_clear_queue(matroska);
1608
1609     url_fseek(s->pb, st->index_entries[index].pos, SEEK_SET);
1610     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
1611     matroska->skip_to_stream = st;
1612     av_update_cur_dts(s, st, st->index_entries[index].timestamp);
1613     return 0;
1614 }
1615
1616 static int matroska_read_close(AVFormatContext *s)
1617 {
1618     MatroskaDemuxContext *matroska = s->priv_data;
1619     MatroskaTrack *tracks = matroska->tracks.elem;
1620     int n;
1621
1622     matroska_clear_queue(matroska);
1623
1624     for (n=0; n < matroska->tracks.nb_elem; n++)
1625         if (tracks[n].type == MATROSKA_TRACK_TYPE_AUDIO)
1626             av_free(tracks[n].audio.buf);
1627     ebml_free(matroska_segment, matroska);
1628
1629     return 0;
1630 }
1631
1632 AVInputFormat matroska_demuxer = {
1633     "matroska",
1634     NULL_IF_CONFIG_SMALL("Matroska file format"),
1635     sizeof(MatroskaDemuxContext),
1636     matroska_probe,
1637     matroska_read_header,
1638     matroska_read_packet,
1639     matroska_read_close,
1640     matroska_read_seek,
1641 };