The c93 demuxer depends on vocdec.o.
[ffmpeg.git] / libavformat / matroskadec.c
1 /*
2  * Matroska file demuxer (no muxer yet)
3  * Copyright (c) 2003-2004 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  * Specs available on the matroska project page:
28  * http://www.matroska.org/.
29  */
30
31 #include "avformat.h"
32 /* For codec_get_id(). */
33 #include "riff.h"
34 #include "intfloat_readwrite.h"
35 #include "matroska.h"
36
37 typedef struct Track {
38     MatroskaTrackType type;
39
40     /* Unique track number and track ID. stream_index is the index that
41      * the calling app uses for this track. */
42     uint32_t num;
43     uint32_t uid;
44     int stream_index;
45
46     char *name;
47     char language[4];
48
49     char *codec_id;
50     char *codec_name;
51
52     unsigned char *codec_priv;
53     int codec_priv_size;
54
55     uint64_t default_duration;
56     MatroskaTrackFlags flags;
57 } MatroskaTrack;
58
59 typedef struct MatroskaVideoTrack {
60     MatroskaTrack track;
61
62     int pixel_width;
63     int pixel_height;
64     int display_width;
65     int display_height;
66
67     uint32_t fourcc;
68
69     MatroskaAspectRatioMode ar_mode;
70     MatroskaEyeMode eye_mode;
71
72     //..
73 } MatroskaVideoTrack;
74
75 typedef struct MatroskaAudioTrack {
76     MatroskaTrack track;
77
78     int channels;
79     int bitdepth;
80     int internal_samplerate;
81     int samplerate;
82     int block_align;
83
84     /* real audio header */
85     int coded_framesize;
86     int sub_packet_h;
87     int frame_size;
88     int sub_packet_size;
89     int sub_packet_cnt;
90     int pkt_cnt;
91     uint8_t *buf;
92     //..
93 } MatroskaAudioTrack;
94
95 typedef struct MatroskaSubtitleTrack {
96     MatroskaTrack track;
97
98     int ass;
99     //..
100 } MatroskaSubtitleTrack;
101
102 #define MAX_TRACK_SIZE (FFMAX(FFMAX(sizeof(MatroskaVideoTrack), \
103                                     sizeof(MatroskaAudioTrack)), \
104                                     sizeof(MatroskaSubtitleTrack)))
105
106 typedef struct MatroskaLevel {
107     uint64_t start;
108     uint64_t length;
109 } MatroskaLevel;
110
111 typedef struct MatroskaDemuxIndex {
112   uint64_t        pos;   /* of the corresponding *cluster*! */
113   uint16_t        track; /* reference to 'num' */
114   uint64_t        time;  /* in nanoseconds */
115 } MatroskaDemuxIndex;
116
117 typedef struct MatroskaDemuxContext {
118     AVFormatContext *ctx;
119
120     /* ebml stuff */
121     int num_levels;
122     MatroskaLevel levels[EBML_MAX_DEPTH];
123     int level_up;
124
125     /* matroska stuff */
126     char *writing_app;
127     char *muxing_app;
128     int64_t created;
129
130     /* timescale in the file */
131     int64_t time_scale;
132
133     /* num_streams is the number of streams that av_new_stream() was called
134      * for ( = that are available to the calling program). */
135     int num_tracks;
136     int num_streams;
137     MatroskaTrack *tracks[MAX_STREAMS];
138
139     /* cache for ID peeking */
140     uint32_t peek_id;
141
142     /* byte position of the segment inside the stream */
143     offset_t segment_start;
144
145     /* The packet queue. */
146     AVPacket **packets;
147     int num_packets;
148
149     /* have we already parse metadata/cues/clusters? */
150     int metadata_parsed;
151     int index_parsed;
152     int done;
153
154     /* The index for seeking. */
155     int num_indexes;
156     MatroskaDemuxIndex *index;
157
158     /* What to skip before effectively reading a packet. */
159     int skip_to_keyframe;
160     AVStream *skip_to_stream;
161 } MatroskaDemuxContext;
162
163 /*
164  * The first few functions handle EBML file parsing. The rest
165  * is the document interpretation. Matroska really just is a
166  * EBML file.
167  */
168
169 /*
170  * Return: the amount of levels in the hierarchy that the
171  * current element lies higher than the previous one.
172  * The opposite isn't done - that's auto-done using master
173  * element reading.
174  */
175
176 static int
177 ebml_read_element_level_up (MatroskaDemuxContext *matroska)
178 {
179     ByteIOContext *pb = &matroska->ctx->pb;
180     offset_t pos = url_ftell(pb);
181     int num = 0;
182
183     while (matroska->num_levels > 0) {
184         MatroskaLevel *level = &matroska->levels[matroska->num_levels - 1];
185
186         if (pos >= level->start + level->length) {
187             matroska->num_levels--;
188             num++;
189         } else {
190             break;
191         }
192     }
193
194     return num;
195 }
196
197 /*
198  * Read: an "EBML number", which is defined as a variable-length
199  * array of bytes. The first byte indicates the length by giving a
200  * number of 0-bits followed by a one. The position of the first
201  * "one" bit inside the first byte indicates the length of this
202  * number.
203  * Returns: num. of bytes read. < 0 on error.
204  */
205
206 static int
207 ebml_read_num (MatroskaDemuxContext *matroska,
208                int                   max_size,
209                uint64_t             *number)
210 {
211     ByteIOContext *pb = &matroska->ctx->pb;
212     int len_mask = 0x80, read = 1, n = 1;
213     int64_t total = 0;
214
215     /* the first byte tells us the length in bytes - get_byte() can normally
216      * return 0, but since that's not a valid first ebmlID byte, we can
217      * use it safely here to catch EOS. */
218     if (!(total = get_byte(pb))) {
219         /* we might encounter EOS here */
220         if (!url_feof(pb)) {
221             offset_t pos = url_ftell(pb);
222             av_log(matroska->ctx, AV_LOG_ERROR,
223                    "Read error at pos. %"PRIu64" (0x%"PRIx64")\n",
224                    pos, pos);
225         }
226         return AVERROR_IO; /* EOS or actual I/O error */
227     }
228
229     /* get the length of the EBML number */
230     while (read <= max_size && !(total & len_mask)) {
231         read++;
232         len_mask >>= 1;
233     }
234     if (read > max_size) {
235         offset_t pos = url_ftell(pb) - 1;
236         av_log(matroska->ctx, AV_LOG_ERROR,
237                "Invalid EBML number size tag 0x%02x at pos %"PRIu64" (0x%"PRIx64")\n",
238                (uint8_t) total, pos, pos);
239         return AVERROR_INVALIDDATA;
240     }
241
242     /* read out length */
243     total &= ~len_mask;
244     while (n++ < read)
245         total = (total << 8) | get_byte(pb);
246
247     *number = total;
248
249     return read;
250 }
251
252 /*
253  * Read: the element content data ID.
254  * Return: the number of bytes read or < 0 on error.
255  */
256
257 static int
258 ebml_read_element_id (MatroskaDemuxContext *matroska,
259                       uint32_t             *id,
260                       int                  *level_up)
261 {
262     int read;
263     uint64_t total;
264
265     /* if we re-call this, use our cached ID */
266     if (matroska->peek_id != 0) {
267         if (level_up)
268             *level_up = 0;
269         *id = matroska->peek_id;
270         return 0;
271     }
272
273     /* read out the "EBML number", include tag in ID */
274     if ((read = ebml_read_num(matroska, 4, &total)) < 0)
275         return read;
276     *id = matroska->peek_id  = total | (1 << (read * 7));
277
278     /* level tracking */
279     if (level_up)
280         *level_up = ebml_read_element_level_up(matroska);
281
282     return read;
283 }
284
285 /*
286  * Read: element content length.
287  * Return: the number of bytes read or < 0 on error.
288  */
289
290 static int
291 ebml_read_element_length (MatroskaDemuxContext *matroska,
292                           uint64_t             *length)
293 {
294     /* clear cache since we're now beyond that data point */
295     matroska->peek_id = 0;
296
297     /* read out the "EBML number", include tag in ID */
298     return ebml_read_num(matroska, 8, length);
299 }
300
301 /*
302  * Return: the ID of the next element, or 0 on error.
303  * Level_up contains the amount of levels that this
304  * next element lies higher than the previous one.
305  */
306
307 static uint32_t
308 ebml_peek_id (MatroskaDemuxContext *matroska,
309               int                  *level_up)
310 {
311     uint32_t id;
312
313     if (ebml_read_element_id(matroska, &id, level_up) < 0)
314         return 0;
315
316     return id;
317 }
318
319 /*
320  * Seek to a given offset.
321  * 0 is success, -1 is failure.
322  */
323
324 static int
325 ebml_read_seek (MatroskaDemuxContext *matroska,
326                 offset_t              offset)
327 {
328     ByteIOContext *pb = &matroska->ctx->pb;
329
330     /* clear ID cache, if any */
331     matroska->peek_id = 0;
332
333     return (url_fseek(pb, offset, SEEK_SET) == offset) ? 0 : -1;
334 }
335
336 /*
337  * Skip the next element.
338  * 0 is success, -1 is failure.
339  */
340
341 static int
342 ebml_read_skip (MatroskaDemuxContext *matroska)
343 {
344     ByteIOContext *pb = &matroska->ctx->pb;
345     uint32_t id;
346     uint64_t length;
347     int res;
348
349     if ((res = ebml_read_element_id(matroska, &id, NULL)) < 0 ||
350         (res = ebml_read_element_length(matroska, &length)) < 0)
351         return res;
352
353     url_fskip(pb, length);
354
355     return 0;
356 }
357
358 /*
359  * Read the next element as an unsigned int.
360  * 0 is success, < 0 is failure.
361  */
362
363 static int
364 ebml_read_uint (MatroskaDemuxContext *matroska,
365                 uint32_t             *id,
366                 uint64_t             *num)
367 {
368     ByteIOContext *pb = &matroska->ctx->pb;
369     int n = 0, size, res;
370     uint64_t rlength;
371
372     if ((res = ebml_read_element_id(matroska, id, NULL)) < 0 ||
373         (res = ebml_read_element_length(matroska, &rlength)) < 0)
374         return res;
375     size = rlength;
376     if (size < 1 || size > 8) {
377         offset_t pos = url_ftell(pb);
378         av_log(matroska->ctx, AV_LOG_ERROR,
379                "Invalid uint element size %d at position %"PRId64" (0x%"PRIx64")\n",
380                 size, pos, pos);
381         return AVERROR_INVALIDDATA;
382     }
383
384     /* big-endian ordening; build up number */
385     *num = 0;
386     while (n++ < size)
387         *num = (*num << 8) | get_byte(pb);
388
389     return 0;
390 }
391
392 /*
393  * Read the next element as a signed int.
394  * 0 is success, < 0 is failure.
395  */
396
397 static int
398 ebml_read_sint (MatroskaDemuxContext *matroska,
399                 uint32_t             *id,
400                 int64_t              *num)
401 {
402     ByteIOContext *pb = &matroska->ctx->pb;
403     int size, n = 1, negative = 0, res;
404     uint64_t rlength;
405
406     if ((res = ebml_read_element_id(matroska, id, NULL)) < 0 ||
407         (res = ebml_read_element_length(matroska, &rlength)) < 0)
408         return res;
409     size = rlength;
410     if (size < 1 || size > 8) {
411         offset_t pos = url_ftell(pb);
412         av_log(matroska->ctx, AV_LOG_ERROR,
413                "Invalid sint element size %d at position %"PRId64" (0x%"PRIx64")\n",
414                 size, pos, pos);
415         return AVERROR_INVALIDDATA;
416     }
417     if ((*num = get_byte(pb)) & 0x80) {
418         negative = 1;
419         *num &= ~0x80;
420     }
421     while (n++ < size)
422         *num = (*num << 8) | get_byte(pb);
423
424     /* make signed */
425     if (negative)
426         *num = *num - (1LL << ((8 * size) - 1));
427
428     return 0;
429 }
430
431 /*
432  * Read the next element as a float.
433  * 0 is success, < 0 is failure.
434  */
435
436 static int
437 ebml_read_float (MatroskaDemuxContext *matroska,
438                  uint32_t             *id,
439                  double               *num)
440 {
441     ByteIOContext *pb = &matroska->ctx->pb;
442     int size, res;
443     uint64_t rlength;
444
445     if ((res = ebml_read_element_id(matroska, id, NULL)) < 0 ||
446         (res = ebml_read_element_length(matroska, &rlength)) < 0)
447         return res;
448     size = rlength;
449
450     if (size == 4) {
451         *num= av_int2flt(get_be32(pb));
452     } else if(size==8){
453         *num= av_int2dbl(get_be64(pb));
454     } else{
455         offset_t pos = url_ftell(pb);
456         av_log(matroska->ctx, AV_LOG_ERROR,
457                "Invalid float element size %d at position %"PRIu64" (0x%"PRIx64")\n",
458                size, pos, pos);
459         return AVERROR_INVALIDDATA;
460     }
461
462     return 0;
463 }
464
465 /*
466  * Read the next element as an ASCII string.
467  * 0 is success, < 0 is failure.
468  */
469
470 static int
471 ebml_read_ascii (MatroskaDemuxContext *matroska,
472                  uint32_t             *id,
473                  char                **str)
474 {
475     ByteIOContext *pb = &matroska->ctx->pb;
476     int size, res;
477     uint64_t rlength;
478
479     if ((res = ebml_read_element_id(matroska, id, NULL)) < 0 ||
480         (res = ebml_read_element_length(matroska, &rlength)) < 0)
481         return res;
482     size = rlength;
483
484     /* ebml strings are usually not 0-terminated, so we allocate one
485      * byte more, read the string and NULL-terminate it ourselves. */
486     if (size < 0 || !(*str = av_malloc(size + 1))) {
487         av_log(matroska->ctx, AV_LOG_ERROR, "Memory allocation failed\n");
488         return AVERROR_NOMEM;
489     }
490     if (get_buffer(pb, (uint8_t *) *str, size) != size) {
491         offset_t pos = url_ftell(pb);
492         av_log(matroska->ctx, AV_LOG_ERROR,
493                "Read error at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
494         return AVERROR_IO;
495     }
496     (*str)[size] = '\0';
497
498     return 0;
499 }
500
501 /*
502  * Read the next element as a UTF-8 string.
503  * 0 is success, < 0 is failure.
504  */
505
506 static int
507 ebml_read_utf8 (MatroskaDemuxContext *matroska,
508                 uint32_t             *id,
509                 char                **str)
510 {
511   return ebml_read_ascii(matroska, id, str);
512 }
513
514 /*
515  * Read the next element as a date (nanoseconds since 1/1/2000).
516  * 0 is success, < 0 is failure.
517  */
518
519 static int
520 ebml_read_date (MatroskaDemuxContext *matroska,
521                 uint32_t             *id,
522                 int64_t              *date)
523 {
524   return ebml_read_sint(matroska, id, date);
525 }
526
527 /*
528  * Read the next element, but only the header. The contents
529  * are supposed to be sub-elements which can be read separately.
530  * 0 is success, < 0 is failure.
531  */
532
533 static int
534 ebml_read_master (MatroskaDemuxContext *matroska,
535                   uint32_t             *id)
536 {
537     ByteIOContext *pb = &matroska->ctx->pb;
538     uint64_t length;
539     MatroskaLevel *level;
540     int res;
541
542     if ((res = ebml_read_element_id(matroska, id, NULL)) < 0 ||
543         (res = ebml_read_element_length(matroska, &length)) < 0)
544         return res;
545
546     /* protect... (Heaven forbids that the '>' is true) */
547     if (matroska->num_levels >= EBML_MAX_DEPTH) {
548         av_log(matroska->ctx, AV_LOG_ERROR,
549                "File moves beyond max. allowed depth (%d)\n", EBML_MAX_DEPTH);
550         return AVERROR_NOTSUPP;
551     }
552
553     /* remember level */
554     level = &matroska->levels[matroska->num_levels++];
555     level->start = url_ftell(pb);
556     level->length = length;
557
558     return 0;
559 }
560
561 /*
562  * Read the next element as binary data.
563  * 0 is success, < 0 is failure.
564  */
565
566 static int
567 ebml_read_binary (MatroskaDemuxContext *matroska,
568                   uint32_t             *id,
569                   uint8_t             **binary,
570                   int                  *size)
571 {
572     ByteIOContext *pb = &matroska->ctx->pb;
573     uint64_t rlength;
574     int res;
575
576     if ((res = ebml_read_element_id(matroska, id, NULL)) < 0 ||
577         (res = ebml_read_element_length(matroska, &rlength)) < 0)
578         return res;
579     *size = rlength;
580
581     if (!(*binary = av_malloc(*size))) {
582         av_log(matroska->ctx, AV_LOG_ERROR,
583                "Memory allocation error\n");
584         return AVERROR_NOMEM;
585     }
586
587     if (get_buffer(pb, *binary, *size) != *size) {
588         offset_t pos = url_ftell(pb);
589         av_log(matroska->ctx, AV_LOG_ERROR,
590                "Read error at pos. %"PRIu64" (0x%"PRIx64")\n", pos, pos);
591         return AVERROR_IO;
592     }
593
594     return 0;
595 }
596
597 /*
598  * Read signed/unsigned "EBML" numbers.
599  * Return: number of bytes processed, < 0 on error.
600  * XXX: use ebml_read_num().
601  */
602
603 static int
604 matroska_ebmlnum_uint (uint8_t  *data,
605                        uint32_t  size,
606                        uint64_t *num)
607 {
608     int len_mask = 0x80, read = 1, n = 1, num_ffs = 0;
609     uint64_t total;
610
611     if (size <= 0)
612         return AVERROR_INVALIDDATA;
613
614     total = data[0];
615     while (read <= 8 && !(total & len_mask)) {
616         read++;
617         len_mask >>= 1;
618     }
619     if (read > 8)
620         return AVERROR_INVALIDDATA;
621
622     if ((total &= (len_mask - 1)) == len_mask - 1)
623         num_ffs++;
624     if (size < read)
625         return AVERROR_INVALIDDATA;
626     while (n < read) {
627         if (data[n] == 0xff)
628             num_ffs++;
629         total = (total << 8) | data[n];
630         n++;
631     }
632
633     if (read == num_ffs)
634         *num = (uint64_t)-1;
635     else
636         *num = total;
637
638     return read;
639 }
640
641 /*
642  * Same as above, but signed.
643  */
644
645 static int
646 matroska_ebmlnum_sint (uint8_t  *data,
647                        uint32_t  size,
648                        int64_t  *num)
649 {
650     uint64_t unum;
651     int res;
652
653     /* read as unsigned number first */
654     if ((res = matroska_ebmlnum_uint(data, size, &unum)) < 0)
655         return res;
656
657     /* make signed (weird way) */
658     if (unum == (uint64_t)-1)
659         *num = INT64_MAX;
660     else
661         *num = unum - ((1LL << ((7 * res) - 1)) - 1);
662
663     return res;
664 }
665
666 /*
667  * Read an EBML header.
668  * 0 is success, < 0 is failure.
669  */
670
671 static int
672 ebml_read_header (MatroskaDemuxContext *matroska,
673                   char                **doctype,
674                   int                  *version)
675 {
676     uint32_t id;
677     int level_up, res = 0;
678
679     /* default init */
680     if (doctype)
681         *doctype = NULL;
682     if (version)
683         *version = 1;
684
685     if (!(id = ebml_peek_id(matroska, &level_up)) ||
686         level_up != 0 || id != EBML_ID_HEADER) {
687         av_log(matroska->ctx, AV_LOG_ERROR,
688                "This is not an EBML file (id=0x%x/0x%x)\n", id, EBML_ID_HEADER);
689         return AVERROR_INVALIDDATA;
690     }
691     if ((res = ebml_read_master(matroska, &id)) < 0)
692         return res;
693
694     while (res == 0) {
695         if (!(id = ebml_peek_id(matroska, &level_up)))
696             return AVERROR_IO;
697
698         /* end-of-header */
699         if (level_up)
700             break;
701
702         switch (id) {
703             /* is our read version uptodate? */
704             case EBML_ID_EBMLREADVERSION: {
705                 uint64_t num;
706
707                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
708                     return res;
709                 if (num > EBML_VERSION) {
710                     av_log(matroska->ctx, AV_LOG_ERROR,
711                            "EBML version %"PRIu64" (> %d) is not supported\n",
712                            num, EBML_VERSION);
713                     return AVERROR_INVALIDDATA;
714                 }
715                 break;
716             }
717
718             /* we only handle 8 byte lengths at max */
719             case EBML_ID_EBMLMAXSIZELENGTH: {
720                 uint64_t num;
721
722                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
723                     return res;
724                 if (num > sizeof(uint64_t)) {
725                     av_log(matroska->ctx, AV_LOG_ERROR,
726                            "Integers of size %"PRIu64" (> %zd) not supported\n",
727                            num, sizeof(uint64_t));
728                     return AVERROR_INVALIDDATA;
729                 }
730                 break;
731             }
732
733             /* we handle 4 byte IDs at max */
734             case EBML_ID_EBMLMAXIDLENGTH: {
735                 uint64_t num;
736
737                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
738                     return res;
739                 if (num > sizeof(uint32_t)) {
740                     av_log(matroska->ctx, AV_LOG_ERROR,
741                            "IDs of size %"PRIu64" (> %zu) not supported\n",
742                             num, sizeof(uint32_t));
743                     return AVERROR_INVALIDDATA;
744                 }
745                 break;
746             }
747
748             case EBML_ID_DOCTYPE: {
749                 char *text;
750
751                 if ((res = ebml_read_ascii(matroska, &id, &text)) < 0)
752                     return res;
753                 if (doctype) {
754                     if (*doctype)
755                         av_free(*doctype);
756                     *doctype = text;
757                 } else
758                     av_free(text);
759                 break;
760             }
761
762             case EBML_ID_DOCTYPEREADVERSION: {
763                 uint64_t num;
764
765                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
766                     return res;
767                 if (version)
768                     *version = num;
769                 break;
770             }
771
772             default:
773                 av_log(matroska->ctx, AV_LOG_INFO,
774                        "Unknown data type 0x%x in EBML header", id);
775                 /* pass-through */
776
777             case EBML_ID_VOID:
778             /* we ignore these two, as they don't tell us anything we
779              * care about */
780             case EBML_ID_EBMLVERSION:
781             case EBML_ID_DOCTYPEVERSION:
782                 res = ebml_read_skip (matroska);
783                 break;
784         }
785     }
786
787     return 0;
788 }
789
790
791 static int
792 matroska_find_track_by_num (MatroskaDemuxContext *matroska,
793                             int                   num)
794 {
795     int i;
796
797     for (i = 0; i < matroska->num_tracks; i++)
798         if (matroska->tracks[i]->num == num)
799             return i;
800
801     return -1;
802 }
803
804
805 /*
806  * Put one packet in an application-supplied AVPacket struct.
807  * Returns 0 on success or -1 on failure.
808  */
809
810 static int
811 matroska_deliver_packet (MatroskaDemuxContext *matroska,
812                          AVPacket             *pkt)
813 {
814     if (matroska->num_packets > 0) {
815         memcpy(pkt, matroska->packets[0], sizeof(AVPacket));
816         av_free(matroska->packets[0]);
817         if (matroska->num_packets > 1) {
818             memmove(&matroska->packets[0], &matroska->packets[1],
819                     (matroska->num_packets - 1) * sizeof(AVPacket *));
820             matroska->packets =
821                 av_realloc(matroska->packets, (matroska->num_packets - 1) *
822                            sizeof(AVPacket *));
823         } else {
824             av_freep(&matroska->packets);
825         }
826         matroska->num_packets--;
827         return 0;
828     }
829
830     return -1;
831 }
832
833 /*
834  * Put a packet into our internal queue. Will be delivered to the
835  * user/application during the next get_packet() call.
836  */
837
838 static void
839 matroska_queue_packet (MatroskaDemuxContext *matroska,
840                        AVPacket             *pkt)
841 {
842     matroska->packets =
843         av_realloc(matroska->packets, (matroska->num_packets + 1) *
844                    sizeof(AVPacket *));
845     matroska->packets[matroska->num_packets] = pkt;
846     matroska->num_packets++;
847 }
848
849
850 /*
851  * Autodetecting...
852  */
853
854 static int
855 matroska_probe (AVProbeData *p)
856 {
857     uint64_t total = 0;
858     int len_mask = 0x80, size = 1, n = 1;
859     uint8_t probe_data[] = { 'm', 'a', 't', 'r', 'o', 's', 'k', 'a' };
860
861     /* ebml header? */
862     if (AV_RB32(p->buf) != EBML_ID_HEADER)
863         return 0;
864
865     /* length of header */
866     total = p->buf[4];
867     while (size <= 8 && !(total & len_mask)) {
868         size++;
869         len_mask >>= 1;
870     }
871     if (size > 8)
872       return 0;
873     total &= (len_mask - 1);
874     while (n < size)
875         total = (total << 8) | p->buf[4 + n++];
876
877     /* does the probe data contain the whole header? */
878     if (p->buf_size < 4 + size + total)
879       return 0;
880
881     /* the header must contain the document type 'matroska'. For now,
882      * we don't parse the whole header but simply check for the
883      * availability of that array of characters inside the header.
884      * Not fully fool-proof, but good enough. */
885     for (n = 4 + size; n <= 4 + size + total - sizeof(probe_data); n++)
886         if (!memcmp (&p->buf[n], probe_data, sizeof(probe_data)))
887             return AVPROBE_SCORE_MAX;
888
889     return 0;
890 }
891
892 /*
893  * From here on, it's all XML-style DTD stuff... Needs no comments.
894  */
895
896 static int
897 matroska_parse_info (MatroskaDemuxContext *matroska)
898 {
899     int res = 0;
900     uint32_t id;
901
902     av_log(matroska->ctx, AV_LOG_DEBUG, "Parsing info...\n");
903
904     while (res == 0) {
905         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
906             res = AVERROR_IO;
907             break;
908         } else if (matroska->level_up) {
909             matroska->level_up--;
910             break;
911         }
912
913         switch (id) {
914             /* cluster timecode */
915             case MATROSKA_ID_TIMECODESCALE: {
916                 uint64_t num;
917                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
918                     break;
919                 matroska->time_scale = num;
920                 break;
921             }
922
923             case MATROSKA_ID_DURATION: {
924                 double num;
925                 if ((res = ebml_read_float(matroska, &id, &num)) < 0)
926                     break;
927                 matroska->ctx->duration = num * matroska->time_scale * 1000 / AV_TIME_BASE;
928                 break;
929             }
930
931             case MATROSKA_ID_TITLE: {
932                 char *text;
933                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
934                     break;
935                 strncpy(matroska->ctx->title, text,
936                         sizeof(matroska->ctx->title)-1);
937                 av_free(text);
938                 break;
939             }
940
941             case MATROSKA_ID_WRITINGAPP: {
942                 char *text;
943                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
944                     break;
945                 matroska->writing_app = text;
946                 break;
947             }
948
949             case MATROSKA_ID_MUXINGAPP: {
950                 char *text;
951                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
952                     break;
953                 matroska->muxing_app = text;
954                 break;
955             }
956
957             case MATROSKA_ID_DATEUTC: {
958                 int64_t time;
959                 if ((res = ebml_read_date(matroska, &id, &time)) < 0)
960                     break;
961                 matroska->created = time;
962                 break;
963             }
964
965             default:
966                 av_log(matroska->ctx, AV_LOG_INFO,
967                        "Unknown entry 0x%x in info header\n", id);
968                 /* fall-through */
969
970             case EBML_ID_VOID:
971                 res = ebml_read_skip(matroska);
972                 break;
973         }
974
975         if (matroska->level_up) {
976             matroska->level_up--;
977             break;
978         }
979     }
980
981     return res;
982 }
983
984 static int
985 matroska_add_stream (MatroskaDemuxContext *matroska)
986 {
987     int res = 0;
988     uint32_t id;
989     MatroskaTrack *track;
990
991     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing track, adding stream..,\n");
992
993     /* Allocate a generic track. As soon as we know its type we'll realloc. */
994     track = av_mallocz(MAX_TRACK_SIZE);
995     matroska->num_tracks++;
996     strcpy(track->language, "eng");
997
998     /* start with the master */
999     if ((res = ebml_read_master(matroska, &id)) < 0)
1000         return res;
1001
1002     /* try reading the trackentry headers */
1003     while (res == 0) {
1004         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1005             res = AVERROR_IO;
1006             break;
1007         } else if (matroska->level_up > 0) {
1008             matroska->level_up--;
1009             break;
1010         }
1011
1012         switch (id) {
1013             /* track number (unique stream ID) */
1014             case MATROSKA_ID_TRACKNUMBER: {
1015                 uint64_t num;
1016                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1017                     break;
1018                 track->num = num;
1019                 break;
1020             }
1021
1022             /* track UID (unique identifier) */
1023             case MATROSKA_ID_TRACKUID: {
1024                 uint64_t num;
1025                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1026                     break;
1027                 track->uid = num;
1028                 break;
1029             }
1030
1031             /* track type (video, audio, combined, subtitle, etc.) */
1032             case MATROSKA_ID_TRACKTYPE: {
1033                 uint64_t num;
1034                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1035                     break;
1036                 if (track->type && track->type != num) {
1037                     av_log(matroska->ctx, AV_LOG_INFO,
1038                            "More than one tracktype in an entry - skip\n");
1039                     break;
1040                 }
1041                 track->type = num;
1042
1043                 switch (track->type) {
1044                     case MATROSKA_TRACK_TYPE_VIDEO:
1045                     case MATROSKA_TRACK_TYPE_AUDIO:
1046                     case MATROSKA_TRACK_TYPE_SUBTITLE:
1047                         break;
1048                     case MATROSKA_TRACK_TYPE_COMPLEX:
1049                     case MATROSKA_TRACK_TYPE_LOGO:
1050                     case MATROSKA_TRACK_TYPE_CONTROL:
1051                     default:
1052                         av_log(matroska->ctx, AV_LOG_INFO,
1053                                "Unknown or unsupported track type 0x%x\n",
1054                                track->type);
1055                         track->type = 0;
1056                         break;
1057                 }
1058                 matroska->tracks[matroska->num_tracks - 1] = track;
1059                 break;
1060             }
1061
1062             /* tracktype specific stuff for video */
1063             case MATROSKA_ID_TRACKVIDEO: {
1064                 MatroskaVideoTrack *videotrack;
1065                 if (!track->type)
1066                     track->type = MATROSKA_TRACK_TYPE_VIDEO;
1067                 if (track->type != MATROSKA_TRACK_TYPE_VIDEO) {
1068                     av_log(matroska->ctx, AV_LOG_INFO,
1069                            "video data in non-video track - ignoring\n");
1070                     res = AVERROR_INVALIDDATA;
1071                     break;
1072                 } else if ((res = ebml_read_master(matroska, &id)) < 0)
1073                     break;
1074                 videotrack = (MatroskaVideoTrack *)track;
1075
1076                 while (res == 0) {
1077                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1078                         res = AVERROR_IO;
1079                         break;
1080                     } else if (matroska->level_up > 0) {
1081                         matroska->level_up--;
1082                         break;
1083                     }
1084
1085                     switch (id) {
1086                         /* fixme, this should be one-up, but I get it here */
1087                         case MATROSKA_ID_TRACKDEFAULTDURATION: {
1088                             uint64_t num;
1089                             if ((res = ebml_read_uint (matroska, &id,
1090                                                        &num)) < 0)
1091                                 break;
1092                             track->default_duration = num/matroska->time_scale;
1093                             break;
1094                         }
1095
1096                         /* video framerate */
1097                         case MATROSKA_ID_VIDEOFRAMERATE: {
1098                             double num;
1099                             if ((res = ebml_read_float(matroska, &id,
1100                                                        &num)) < 0)
1101                                 break;
1102                             track->default_duration = 1000000000/(matroska->time_scale*num);
1103                             break;
1104                         }
1105
1106                         /* width of the size to display the video at */
1107                         case MATROSKA_ID_VIDEODISPLAYWIDTH: {
1108                             uint64_t num;
1109                             if ((res = ebml_read_uint(matroska, &id,
1110                                                       &num)) < 0)
1111                                 break;
1112                             videotrack->display_width = num;
1113                             break;
1114                         }
1115
1116                         /* height of the size to display the video at */
1117                         case MATROSKA_ID_VIDEODISPLAYHEIGHT: {
1118                             uint64_t num;
1119                             if ((res = ebml_read_uint(matroska, &id,
1120                                                       &num)) < 0)
1121                                 break;
1122                             videotrack->display_height = num;
1123                             break;
1124                         }
1125
1126                         /* width of the video in the file */
1127                         case MATROSKA_ID_VIDEOPIXELWIDTH: {
1128                             uint64_t num;
1129                             if ((res = ebml_read_uint(matroska, &id,
1130                                                       &num)) < 0)
1131                                 break;
1132                             videotrack->pixel_width = num;
1133                             break;
1134                         }
1135
1136                         /* height of the video in the file */
1137                         case MATROSKA_ID_VIDEOPIXELHEIGHT: {
1138                             uint64_t num;
1139                             if ((res = ebml_read_uint(matroska, &id,
1140                                                       &num)) < 0)
1141                                 break;
1142                             videotrack->pixel_height = num;
1143                             break;
1144                         }
1145
1146                         /* whether the video is interlaced */
1147                         case MATROSKA_ID_VIDEOFLAGINTERLACED: {
1148                             uint64_t num;
1149                             if ((res = ebml_read_uint(matroska, &id,
1150                                                       &num)) < 0)
1151                                 break;
1152                             if (num)
1153                                 track->flags |=
1154                                     MATROSKA_VIDEOTRACK_INTERLACED;
1155                             else
1156                                 track->flags &=
1157                                     ~MATROSKA_VIDEOTRACK_INTERLACED;
1158                             break;
1159                         }
1160
1161                         /* stereo mode (whether the video has two streams,
1162                          * where one is for the left eye and the other for
1163                          * the right eye, which creates a 3D-like
1164                          * effect) */
1165                         case MATROSKA_ID_VIDEOSTEREOMODE: {
1166                             uint64_t num;
1167                             if ((res = ebml_read_uint(matroska, &id,
1168                                                       &num)) < 0)
1169                                 break;
1170                             if (num != MATROSKA_EYE_MODE_MONO &&
1171                                 num != MATROSKA_EYE_MODE_LEFT &&
1172                                 num != MATROSKA_EYE_MODE_RIGHT &&
1173                                 num != MATROSKA_EYE_MODE_BOTH) {
1174                                 av_log(matroska->ctx, AV_LOG_INFO,
1175                                        "Ignoring unknown eye mode 0x%x\n",
1176                                        (uint32_t) num);
1177                                 break;
1178                             }
1179                             videotrack->eye_mode = num;
1180                             break;
1181                         }
1182
1183                         /* aspect ratio behaviour */
1184                         case MATROSKA_ID_VIDEOASPECTRATIO: {
1185                             uint64_t num;
1186                             if ((res = ebml_read_uint(matroska, &id,
1187                                                       &num)) < 0)
1188                                 break;
1189                             if (num != MATROSKA_ASPECT_RATIO_MODE_FREE &&
1190                                 num != MATROSKA_ASPECT_RATIO_MODE_KEEP &&
1191                                 num != MATROSKA_ASPECT_RATIO_MODE_FIXED) {
1192                                 av_log(matroska->ctx, AV_LOG_INFO,
1193                                        "Ignoring unknown aspect ratio 0x%x\n",
1194                                        (uint32_t) num);
1195                                 break;
1196                             }
1197                             videotrack->ar_mode = num;
1198                             break;
1199                         }
1200
1201                         /* colourspace (only matters for raw video)
1202                          * fourcc */
1203                         case MATROSKA_ID_VIDEOCOLOURSPACE: {
1204                             uint64_t num;
1205                             if ((res = ebml_read_uint(matroska, &id,
1206                                                       &num)) < 0)
1207                                 break;
1208                             videotrack->fourcc = num;
1209                             break;
1210                         }
1211
1212                         default:
1213                             av_log(matroska->ctx, AV_LOG_INFO,
1214                                    "Unknown video track header entry "
1215                                    "0x%x - ignoring\n", id);
1216                             /* pass-through */
1217
1218                         case EBML_ID_VOID:
1219                             res = ebml_read_skip(matroska);
1220                             break;
1221                     }
1222
1223                     if (matroska->level_up) {
1224                         matroska->level_up--;
1225                         break;
1226                     }
1227                 }
1228                 break;
1229             }
1230
1231             /* tracktype specific stuff for audio */
1232             case MATROSKA_ID_TRACKAUDIO: {
1233                 MatroskaAudioTrack *audiotrack;
1234                 if (!track->type)
1235                     track->type = MATROSKA_TRACK_TYPE_AUDIO;
1236                 if (track->type != MATROSKA_TRACK_TYPE_AUDIO) {
1237                     av_log(matroska->ctx, AV_LOG_INFO,
1238                            "audio data in non-audio track - ignoring\n");
1239                     res = AVERROR_INVALIDDATA;
1240                     break;
1241                 } else if ((res = ebml_read_master(matroska, &id)) < 0)
1242                     break;
1243                 audiotrack = (MatroskaAudioTrack *)track;
1244                 audiotrack->channels = 1;
1245                 audiotrack->samplerate = 8000;
1246
1247                 while (res == 0) {
1248                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1249                         res = AVERROR_IO;
1250                         break;
1251                     } else if (matroska->level_up > 0) {
1252                         matroska->level_up--;
1253                         break;
1254                     }
1255
1256                     switch (id) {
1257                         /* samplerate */
1258                         case MATROSKA_ID_AUDIOSAMPLINGFREQ: {
1259                             double num;
1260                             if ((res = ebml_read_float(matroska, &id,
1261                                                        &num)) < 0)
1262                                 break;
1263                             audiotrack->internal_samplerate =
1264                             audiotrack->samplerate = num;
1265                             break;
1266                         }
1267
1268                         case MATROSKA_ID_AUDIOOUTSAMPLINGFREQ: {
1269                             double num;
1270                             if ((res = ebml_read_float(matroska, &id,
1271                                                        &num)) < 0)
1272                                 break;
1273                             audiotrack->samplerate = num;
1274                             break;
1275                         }
1276
1277                             /* bitdepth */
1278                         case MATROSKA_ID_AUDIOBITDEPTH: {
1279                             uint64_t num;
1280                             if ((res = ebml_read_uint(matroska, &id,
1281                                                       &num)) < 0)
1282                                 break;
1283                             audiotrack->bitdepth = num;
1284                             break;
1285                         }
1286
1287                             /* channels */
1288                         case MATROSKA_ID_AUDIOCHANNELS: {
1289                             uint64_t num;
1290                             if ((res = ebml_read_uint(matroska, &id,
1291                                                       &num)) < 0)
1292                                 break;
1293                             audiotrack->channels = num;
1294                             break;
1295                         }
1296
1297                         default:
1298                             av_log(matroska->ctx, AV_LOG_INFO,
1299                                    "Unknown audio track header entry "
1300                                    "0x%x - ignoring\n", id);
1301                             /* pass-through */
1302
1303                         case EBML_ID_VOID:
1304                             res = ebml_read_skip(matroska);
1305                             break;
1306                     }
1307
1308                     if (matroska->level_up) {
1309                         matroska->level_up--;
1310                         break;
1311                     }
1312                 }
1313                 break;
1314             }
1315
1316                 /* codec identifier */
1317             case MATROSKA_ID_CODECID: {
1318                 char *text;
1319                 if ((res = ebml_read_ascii(matroska, &id, &text)) < 0)
1320                     break;
1321                 track->codec_id = text;
1322                 break;
1323             }
1324
1325                 /* codec private data */
1326             case MATROSKA_ID_CODECPRIVATE: {
1327                 uint8_t *data;
1328                 int size;
1329                 if ((res = ebml_read_binary(matroska, &id, &data, &size) < 0))
1330                     break;
1331                 track->codec_priv = data;
1332                 track->codec_priv_size = size;
1333                 break;
1334             }
1335
1336                 /* name of the codec */
1337             case MATROSKA_ID_CODECNAME: {
1338                 char *text;
1339                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
1340                     break;
1341                 track->codec_name = text;
1342                 break;
1343             }
1344
1345                 /* name of this track */
1346             case MATROSKA_ID_TRACKNAME: {
1347                 char *text;
1348                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
1349                     break;
1350                 track->name = text;
1351                 break;
1352             }
1353
1354                 /* language (matters for audio/subtitles, mostly) */
1355             case MATROSKA_ID_TRACKLANGUAGE: {
1356                 char *text, *end;
1357                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
1358                     break;
1359                 if ((end = strchr(text, '-')))
1360                     *end = '\0';
1361                 if (strlen(text) == 3)
1362                     strcpy(track->language, text);
1363                 av_free(text);
1364                 break;
1365             }
1366
1367                 /* whether this is actually used */
1368             case MATROSKA_ID_TRACKFLAGENABLED: {
1369                 uint64_t num;
1370                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1371                     break;
1372                 if (num)
1373                     track->flags |= MATROSKA_TRACK_ENABLED;
1374                 else
1375                     track->flags &= ~MATROSKA_TRACK_ENABLED;
1376                 break;
1377             }
1378
1379                 /* whether it's the default for this track type */
1380             case MATROSKA_ID_TRACKFLAGDEFAULT: {
1381                 uint64_t num;
1382                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1383                     break;
1384                 if (num)
1385                     track->flags |= MATROSKA_TRACK_DEFAULT;
1386                 else
1387                     track->flags &= ~MATROSKA_TRACK_DEFAULT;
1388                 break;
1389             }
1390
1391                 /* lacing (like MPEG, where blocks don't end/start on frame
1392                  * boundaries) */
1393             case MATROSKA_ID_TRACKFLAGLACING: {
1394                 uint64_t num;
1395                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1396                     break;
1397                 if (num)
1398                     track->flags |= MATROSKA_TRACK_LACING;
1399                 else
1400                     track->flags &= ~MATROSKA_TRACK_LACING;
1401                 break;
1402             }
1403
1404                 /* default length (in time) of one data block in this track */
1405             case MATROSKA_ID_TRACKDEFAULTDURATION: {
1406                 uint64_t num;
1407                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1408                     break;
1409                 track->default_duration = num / matroska->time_scale;
1410                 break;
1411             }
1412
1413             default:
1414                 av_log(matroska->ctx, AV_LOG_INFO,
1415                        "Unknown track header entry 0x%x - ignoring\n", id);
1416                 /* pass-through */
1417
1418             case EBML_ID_VOID:
1419             /* we ignore these because they're nothing useful. */
1420             case MATROSKA_ID_CODECINFOURL:
1421             case MATROSKA_ID_CODECDOWNLOADURL:
1422             case MATROSKA_ID_TRACKMINCACHE:
1423             case MATROSKA_ID_TRACKMAXCACHE:
1424                 res = ebml_read_skip(matroska);
1425                 break;
1426         }
1427
1428         if (matroska->level_up) {
1429             matroska->level_up--;
1430             break;
1431         }
1432     }
1433
1434     return res;
1435 }
1436
1437 static int
1438 matroska_parse_tracks (MatroskaDemuxContext *matroska)
1439 {
1440     int res = 0;
1441     uint32_t id;
1442
1443     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing tracks...\n");
1444
1445     while (res == 0) {
1446         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1447             res = AVERROR_IO;
1448             break;
1449         } else if (matroska->level_up) {
1450             matroska->level_up--;
1451             break;
1452         }
1453
1454         switch (id) {
1455             /* one track within the "all-tracks" header */
1456             case MATROSKA_ID_TRACKENTRY:
1457                 res = matroska_add_stream(matroska);
1458                 break;
1459
1460             default:
1461                 av_log(matroska->ctx, AV_LOG_INFO,
1462                        "Unknown entry 0x%x in track header\n", id);
1463                 /* fall-through */
1464
1465             case EBML_ID_VOID:
1466                 res = ebml_read_skip(matroska);
1467                 break;
1468         }
1469
1470         if (matroska->level_up) {
1471             matroska->level_up--;
1472             break;
1473         }
1474     }
1475
1476     return res;
1477 }
1478
1479 static int
1480 matroska_parse_index (MatroskaDemuxContext *matroska)
1481 {
1482     int res = 0;
1483     uint32_t id;
1484     MatroskaDemuxIndex idx;
1485
1486     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing index...\n");
1487
1488     while (res == 0) {
1489         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1490             res = AVERROR_IO;
1491             break;
1492         } else if (matroska->level_up) {
1493             matroska->level_up--;
1494             break;
1495         }
1496
1497         switch (id) {
1498             /* one single index entry ('point') */
1499             case MATROSKA_ID_POINTENTRY:
1500                 if ((res = ebml_read_master(matroska, &id)) < 0)
1501                     break;
1502
1503                 /* in the end, we hope to fill one entry with a
1504                  * timestamp, a file position and a tracknum */
1505                 idx.pos   = (uint64_t) -1;
1506                 idx.time  = (uint64_t) -1;
1507                 idx.track = (uint16_t) -1;
1508
1509                 while (res == 0) {
1510                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1511                         res = AVERROR_IO;
1512                         break;
1513                     } else if (matroska->level_up) {
1514                         matroska->level_up--;
1515                         break;
1516                     }
1517
1518                     switch (id) {
1519                         /* one single index entry ('point') */
1520                         case MATROSKA_ID_CUETIME: {
1521                             uint64_t time;
1522                             if ((res = ebml_read_uint(matroska, &id,
1523                                                       &time)) < 0)
1524                                 break;
1525                             idx.time = time * matroska->time_scale;
1526                             break;
1527                         }
1528
1529                         /* position in the file + track to which it
1530                          * belongs */
1531                         case MATROSKA_ID_CUETRACKPOSITION:
1532                             if ((res = ebml_read_master(matroska, &id)) < 0)
1533                                 break;
1534
1535                             while (res == 0) {
1536                                 if (!(id = ebml_peek_id (matroska,
1537                                                     &matroska->level_up))) {
1538                                     res = AVERROR_IO;
1539                                     break;
1540                                 } else if (matroska->level_up) {
1541                                     matroska->level_up--;
1542                                     break;
1543                                 }
1544
1545                                 switch (id) {
1546                                     /* track number */
1547                                     case MATROSKA_ID_CUETRACK: {
1548                                         uint64_t num;
1549                                         if ((res = ebml_read_uint(matroska,
1550                                                           &id, &num)) < 0)
1551                                             break;
1552                                         idx.track = num;
1553                                         break;
1554                                     }
1555
1556                                         /* position in file */
1557                                     case MATROSKA_ID_CUECLUSTERPOSITION: {
1558                                         uint64_t num;
1559                                         if ((res = ebml_read_uint(matroska,
1560                                                           &id, &num)) < 0)
1561                                             break;
1562                                         idx.pos = num+matroska->segment_start;
1563                                         break;
1564                                     }
1565
1566                                     default:
1567                                         av_log(matroska->ctx, AV_LOG_INFO,
1568                                                "Unknown entry 0x%x in "
1569                                                "CuesTrackPositions\n", id);
1570                                         /* fall-through */
1571
1572                                     case EBML_ID_VOID:
1573                                         res = ebml_read_skip(matroska);
1574                                         break;
1575                                 }
1576
1577                                 if (matroska->level_up) {
1578                                     matroska->level_up--;
1579                                     break;
1580                                 }
1581                             }
1582
1583                             break;
1584
1585                         default:
1586                             av_log(matroska->ctx, AV_LOG_INFO,
1587                                    "Unknown entry 0x%x in cuespoint "
1588                                    "index\n", id);
1589                             /* fall-through */
1590
1591                         case EBML_ID_VOID:
1592                             res = ebml_read_skip(matroska);
1593                             break;
1594                     }
1595
1596                     if (matroska->level_up) {
1597                         matroska->level_up--;
1598                         break;
1599                     }
1600                 }
1601
1602                 /* so let's see if we got what we wanted */
1603                 if (idx.pos   != (uint64_t) -1 &&
1604                     idx.time  != (uint64_t) -1 &&
1605                     idx.track != (uint16_t) -1) {
1606                     if (matroska->num_indexes % 32 == 0) {
1607                         /* re-allocate bigger index */
1608                         matroska->index =
1609                             av_realloc(matroska->index,
1610                                        (matroska->num_indexes + 32) *
1611                                        sizeof(MatroskaDemuxIndex));
1612                     }
1613                     matroska->index[matroska->num_indexes] = idx;
1614                     matroska->num_indexes++;
1615                 }
1616                 break;
1617
1618             default:
1619                 av_log(matroska->ctx, AV_LOG_INFO,
1620                        "Unknown entry 0x%x in cues header\n", id);
1621                 /* fall-through */
1622
1623             case EBML_ID_VOID:
1624                 res = ebml_read_skip(matroska);
1625                 break;
1626         }
1627
1628         if (matroska->level_up) {
1629             matroska->level_up--;
1630             break;
1631         }
1632     }
1633
1634     return res;
1635 }
1636
1637 static int
1638 matroska_parse_metadata (MatroskaDemuxContext *matroska)
1639 {
1640     int res = 0;
1641     uint32_t id;
1642
1643     while (res == 0) {
1644         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1645             res = AVERROR_IO;
1646             break;
1647         } else if (matroska->level_up) {
1648             matroska->level_up--;
1649             break;
1650         }
1651
1652         switch (id) {
1653             /* Hm, this is unsupported... */
1654             default:
1655                 av_log(matroska->ctx, AV_LOG_INFO,
1656                        "Unknown entry 0x%x in metadata header\n", id);
1657                 /* fall-through */
1658
1659             case EBML_ID_VOID:
1660                 res = ebml_read_skip(matroska);
1661                 break;
1662         }
1663
1664         if (matroska->level_up) {
1665             matroska->level_up--;
1666             break;
1667         }
1668     }
1669
1670     return res;
1671 }
1672
1673 static int
1674 matroska_parse_seekhead (MatroskaDemuxContext *matroska)
1675 {
1676     int res = 0;
1677     uint32_t id;
1678
1679     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing seekhead...\n");
1680
1681     while (res == 0) {
1682         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1683             res = AVERROR_IO;
1684             break;
1685         } else if (matroska->level_up) {
1686             matroska->level_up--;
1687             break;
1688         }
1689
1690         switch (id) {
1691             case MATROSKA_ID_SEEKENTRY: {
1692                 uint32_t seek_id = 0, peek_id_cache = 0;
1693                 uint64_t seek_pos = (uint64_t) -1, t;
1694
1695                 if ((res = ebml_read_master(matroska, &id)) < 0)
1696                     break;
1697
1698                 while (res == 0) {
1699                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1700                         res = AVERROR_IO;
1701                         break;
1702                     } else if (matroska->level_up) {
1703                         matroska->level_up--;
1704                         break;
1705                     }
1706
1707                     switch (id) {
1708                         case MATROSKA_ID_SEEKID:
1709                             res = ebml_read_uint(matroska, &id, &t);
1710                             seek_id = t;
1711                             break;
1712
1713                         case MATROSKA_ID_SEEKPOSITION:
1714                             res = ebml_read_uint(matroska, &id, &seek_pos);
1715                             break;
1716
1717                         default:
1718                             av_log(matroska->ctx, AV_LOG_INFO,
1719                                    "Unknown seekhead ID 0x%x\n", id);
1720                             /* fall-through */
1721
1722                         case EBML_ID_VOID:
1723                             res = ebml_read_skip(matroska);
1724                             break;
1725                     }
1726
1727                     if (matroska->level_up) {
1728                         matroska->level_up--;
1729                         break;
1730                     }
1731                 }
1732
1733                 if (!seek_id || seek_pos == (uint64_t) -1) {
1734                     av_log(matroska->ctx, AV_LOG_INFO,
1735                            "Incomplete seekhead entry (0x%x/%"PRIu64")\n",
1736                            seek_id, seek_pos);
1737                     break;
1738                 }
1739
1740                 switch (seek_id) {
1741                     case MATROSKA_ID_CUES:
1742                     case MATROSKA_ID_TAGS: {
1743                         uint32_t level_up = matroska->level_up;
1744                         offset_t before_pos;
1745                         uint64_t length;
1746                         MatroskaLevel level;
1747
1748                         /* remember the peeked ID and the current position */
1749                         peek_id_cache = matroska->peek_id;
1750                         before_pos = url_ftell(&matroska->ctx->pb);
1751
1752                         /* seek */
1753                         if ((res = ebml_read_seek(matroska, seek_pos +
1754                                                matroska->segment_start)) < 0)
1755                             return res;
1756
1757                         /* we don't want to lose our seekhead level, so we add
1758                          * a dummy. This is a crude hack. */
1759                         if (matroska->num_levels == EBML_MAX_DEPTH) {
1760                             av_log(matroska->ctx, AV_LOG_INFO,
1761                                    "Max EBML element depth (%d) reached, "
1762                                    "cannot parse further.\n", EBML_MAX_DEPTH);
1763                             return AVERROR_UNKNOWN;
1764                         }
1765
1766                         level.start = 0;
1767                         level.length = (uint64_t)-1;
1768                         matroska->levels[matroska->num_levels] = level;
1769                         matroska->num_levels++;
1770
1771                         /* check ID */
1772                         if (!(id = ebml_peek_id (matroska,
1773                                                  &matroska->level_up)))
1774                             goto finish;
1775                         if (id != seek_id) {
1776                             av_log(matroska->ctx, AV_LOG_INFO,
1777                                    "We looked for ID=0x%x but got "
1778                                    "ID=0x%x (pos=%"PRIu64")",
1779                                    seek_id, id, seek_pos +
1780                                    matroska->segment_start);
1781                             goto finish;
1782                         }
1783
1784                         /* read master + parse */
1785                         if ((res = ebml_read_master(matroska, &id)) < 0)
1786                             goto finish;
1787                         switch (id) {
1788                             case MATROSKA_ID_CUES:
1789                                 if (!(res = matroska_parse_index(matroska)) ||
1790                                     url_feof(&matroska->ctx->pb)) {
1791                                     matroska->index_parsed = 1;
1792                                     res = 0;
1793                                 }
1794                                 break;
1795                             case MATROSKA_ID_TAGS:
1796                                 if (!(res = matroska_parse_metadata(matroska)) ||
1797                                    url_feof(&matroska->ctx->pb)) {
1798                                     matroska->metadata_parsed = 1;
1799                                     res = 0;
1800                                 }
1801                                 break;
1802                         }
1803
1804                     finish:
1805                         /* remove dummy level */
1806                         while (matroska->num_levels) {
1807                             matroska->num_levels--;
1808                             length =
1809                                 matroska->levels[matroska->num_levels].length;
1810                             if (length == (uint64_t)-1)
1811                                 break;
1812                         }
1813
1814                         /* seek back */
1815                         if ((res = ebml_read_seek(matroska, before_pos)) < 0)
1816                             return res;
1817                         matroska->peek_id = peek_id_cache;
1818                         matroska->level_up = level_up;
1819                         break;
1820                     }
1821
1822                     default:
1823                         av_log(matroska->ctx, AV_LOG_INFO,
1824                                "Ignoring seekhead entry for ID=0x%x\n",
1825                                seek_id);
1826                         break;
1827                 }
1828
1829                 break;
1830             }
1831
1832             default:
1833                 av_log(matroska->ctx, AV_LOG_INFO,
1834                        "Unknown seekhead ID 0x%x\n", id);
1835                 /* fall-through */
1836
1837             case EBML_ID_VOID:
1838                 res = ebml_read_skip(matroska);
1839                 break;
1840         }
1841
1842         if (matroska->level_up) {
1843             matroska->level_up--;
1844             break;
1845         }
1846     }
1847
1848     return res;
1849 }
1850
1851 #define ARRAY_SIZE(x)  (sizeof(x)/sizeof(*x))
1852
1853 static int
1854 matroska_aac_profile (char *codec_id)
1855 {
1856     static const char *aac_profiles[] = {
1857         "MAIN", "LC", "SSR"
1858     };
1859     int profile;
1860
1861     for (profile=0; profile<ARRAY_SIZE(aac_profiles); profile++)
1862         if (strstr(codec_id, aac_profiles[profile]))
1863             break;
1864     return profile + 1;
1865 }
1866
1867 static int
1868 matroska_aac_sri (int samplerate)
1869 {
1870     static const int aac_sample_rates[] = {
1871         96000, 88200, 64000, 48000, 44100, 32000,
1872         24000, 22050, 16000, 12000, 11025,  8000,
1873     };
1874     int sri;
1875
1876     for (sri=0; sri<ARRAY_SIZE(aac_sample_rates); sri++)
1877         if (aac_sample_rates[sri] == samplerate)
1878             break;
1879     return sri;
1880 }
1881
1882 static int
1883 matroska_read_header (AVFormatContext    *s,
1884                       AVFormatParameters *ap)
1885 {
1886     MatroskaDemuxContext *matroska = s->priv_data;
1887     char *doctype;
1888     int version, last_level, res = 0;
1889     uint32_t id;
1890
1891     matroska->ctx = s;
1892
1893     /* First read the EBML header. */
1894     doctype = NULL;
1895     if ((res = ebml_read_header(matroska, &doctype, &version)) < 0)
1896         return res;
1897     if ((doctype == NULL) || strcmp(doctype, "matroska")) {
1898         av_log(matroska->ctx, AV_LOG_ERROR,
1899                "Wrong EBML doctype ('%s' != 'matroska').\n",
1900                doctype ? doctype : "(none)");
1901         if (doctype)
1902             av_free(doctype);
1903         return AVERROR_NOFMT;
1904     }
1905     av_free(doctype);
1906     if (version > 2) {
1907         av_log(matroska->ctx, AV_LOG_ERROR,
1908                "Matroska demuxer version 2 too old for file version %d\n",
1909                version);
1910         return AVERROR_NOFMT;
1911     }
1912
1913     /* The next thing is a segment. */
1914     while (1) {
1915         if (!(id = ebml_peek_id(matroska, &last_level)))
1916             return AVERROR_IO;
1917         if (id == MATROSKA_ID_SEGMENT)
1918             break;
1919
1920         /* oi! */
1921         av_log(matroska->ctx, AV_LOG_INFO,
1922                "Expected a Segment ID (0x%x), but received 0x%x!\n",
1923                MATROSKA_ID_SEGMENT, id);
1924         if ((res = ebml_read_skip(matroska)) < 0)
1925             return res;
1926     }
1927
1928     /* We now have a Matroska segment.
1929      * Seeks are from the beginning of the segment,
1930      * after the segment ID/length. */
1931     if ((res = ebml_read_master(matroska, &id)) < 0)
1932         return res;
1933     matroska->segment_start = url_ftell(&s->pb);
1934
1935     matroska->time_scale = 1000000;
1936     /* we've found our segment, start reading the different contents in here */
1937     while (res == 0) {
1938         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1939             res = AVERROR_IO;
1940             break;
1941         } else if (matroska->level_up) {
1942             matroska->level_up--;
1943             break;
1944         }
1945
1946         switch (id) {
1947             /* stream info */
1948             case MATROSKA_ID_INFO: {
1949                 if ((res = ebml_read_master(matroska, &id)) < 0)
1950                     break;
1951                 res = matroska_parse_info(matroska);
1952                 break;
1953             }
1954
1955             /* track info headers */
1956             case MATROSKA_ID_TRACKS: {
1957                 if ((res = ebml_read_master(matroska, &id)) < 0)
1958                     break;
1959                 res = matroska_parse_tracks(matroska);
1960                 break;
1961             }
1962
1963             /* stream index */
1964             case MATROSKA_ID_CUES: {
1965                 if (!matroska->index_parsed) {
1966                     if ((res = ebml_read_master(matroska, &id)) < 0)
1967                         break;
1968                     res = matroska_parse_index(matroska);
1969                 } else
1970                     res = ebml_read_skip(matroska);
1971                 break;
1972             }
1973
1974             /* metadata */
1975             case MATROSKA_ID_TAGS: {
1976                 if (!matroska->metadata_parsed) {
1977                     if ((res = ebml_read_master(matroska, &id)) < 0)
1978                         break;
1979                     res = matroska_parse_metadata(matroska);
1980                 } else
1981                     res = ebml_read_skip(matroska);
1982                 break;
1983             }
1984
1985             /* file index (if seekable, seek to Cues/Tags to parse it) */
1986             case MATROSKA_ID_SEEKHEAD: {
1987                 if ((res = ebml_read_master(matroska, &id)) < 0)
1988                     break;
1989                 res = matroska_parse_seekhead(matroska);
1990                 break;
1991             }
1992
1993             case MATROSKA_ID_CLUSTER: {
1994                 /* Do not read the master - this will be done in the next
1995                  * call to matroska_read_packet. */
1996                 res = 1;
1997                 break;
1998             }
1999
2000             default:
2001                 av_log(matroska->ctx, AV_LOG_INFO,
2002                        "Unknown matroska file header ID 0x%x\n", id);
2003             /* fall-through */
2004
2005             case EBML_ID_VOID:
2006                 res = ebml_read_skip(matroska);
2007                 break;
2008         }
2009
2010         if (matroska->level_up) {
2011             matroska->level_up--;
2012             break;
2013         }
2014     }
2015
2016     /* Have we found a cluster? */
2017     if (ebml_peek_id(matroska, NULL) == MATROSKA_ID_CLUSTER) {
2018         int i, j;
2019         MatroskaTrack *track;
2020         AVStream *st;
2021
2022         for (i = 0; i < matroska->num_tracks; i++) {
2023             enum CodecID codec_id = CODEC_ID_NONE;
2024             uint8_t *extradata = NULL;
2025             int extradata_size = 0;
2026             int extradata_offset = 0;
2027             track = matroska->tracks[i];
2028             track->stream_index = -1;
2029
2030             /* Apply some sanity checks. */
2031             if (track->codec_id == NULL)
2032                 continue;
2033
2034             for(j=0; ff_mkv_codec_tags[j].str; j++){
2035                 if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
2036                             strlen(ff_mkv_codec_tags[j].str))){
2037                     codec_id= ff_mkv_codec_tags[j].id;
2038                     break;
2039                 }
2040             }
2041
2042             /* Set the FourCC from the CodecID. */
2043             /* This is the MS compatibility mode which stores a
2044              * BITMAPINFOHEADER in the CodecPrivate. */
2045             if (!strcmp(track->codec_id,
2046                         MATROSKA_CODEC_ID_VIDEO_VFW_FOURCC) &&
2047                 (track->codec_priv_size >= 40) &&
2048                 (track->codec_priv != NULL)) {
2049                 MatroskaVideoTrack *vtrack = (MatroskaVideoTrack *) track;
2050
2051                 /* Offset of biCompression. Stored in LE. */
2052                 vtrack->fourcc = AV_RL32(track->codec_priv + 16);
2053                 codec_id = codec_get_id(codec_bmp_tags, vtrack->fourcc);
2054
2055             }
2056
2057             /* This is the MS compatibility mode which stores a
2058              * WAVEFORMATEX in the CodecPrivate. */
2059             else if (!strcmp(track->codec_id,
2060                              MATROSKA_CODEC_ID_AUDIO_ACM) &&
2061                 (track->codec_priv_size >= 18) &&
2062                 (track->codec_priv != NULL)) {
2063                 uint16_t tag;
2064
2065                 /* Offset of wFormatTag. Stored in LE. */
2066                 tag = AV_RL16(track->codec_priv);
2067                 codec_id = codec_get_id(codec_wav_tags, tag);
2068
2069             }
2070
2071             else if (codec_id == CODEC_ID_AAC && !track->codec_priv_size) {
2072                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *) track;
2073                 int profile = matroska_aac_profile(track->codec_id);
2074                 int sri = matroska_aac_sri(audiotrack->internal_samplerate);
2075                 extradata = av_malloc(5);
2076                 if (extradata == NULL)
2077                     return AVERROR_NOMEM;
2078                 extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
2079                 extradata[1] = ((sri&0x01) << 7) | (audiotrack->channels<<3);
2080                 if (strstr(track->codec_id, "SBR")) {
2081                     sri = matroska_aac_sri(audiotrack->samplerate);
2082                     extradata[2] = 0x56;
2083                     extradata[3] = 0xE5;
2084                     extradata[4] = 0x80 | (sri<<3);
2085                     extradata_size = 5;
2086                 } else {
2087                     extradata_size = 2;
2088                 }
2089                 track->default_duration = 1024*1000 / audiotrack->internal_samplerate;
2090             }
2091
2092             else if (codec_id == CODEC_ID_TTA) {
2093                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *) track;
2094                 ByteIOContext b;
2095                 extradata_size = 30;
2096                 extradata = av_mallocz(extradata_size);
2097                 if (extradata == NULL)
2098                     return AVERROR_NOMEM;
2099                 init_put_byte(&b, extradata, extradata_size, 1,
2100                               NULL, NULL, NULL, NULL);
2101                 put_buffer(&b, (uint8_t *) "TTA1", 4);
2102                 put_le16(&b, 1);
2103                 put_le16(&b, audiotrack->channels);
2104                 put_le16(&b, audiotrack->bitdepth);
2105                 put_le32(&b, audiotrack->samplerate);
2106                 put_le32(&b, matroska->ctx->duration * audiotrack->samplerate);
2107             }
2108
2109             else if (codec_id == CODEC_ID_RV10 || codec_id == CODEC_ID_RV20 ||
2110                      codec_id == CODEC_ID_RV30 || codec_id == CODEC_ID_RV40) {
2111                 extradata_offset = 26;
2112                 track->codec_priv_size -= extradata_offset;
2113                 track->flags |= MATROSKA_TRACK_REAL_V;
2114             }
2115
2116             else if (codec_id == CODEC_ID_RA_144) {
2117                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2118                 audiotrack->samplerate = 8000;
2119                 audiotrack->channels = 1;
2120             }
2121
2122             else if (codec_id == CODEC_ID_RA_288 ||
2123                      codec_id == CODEC_ID_COOK ||
2124                      codec_id == CODEC_ID_ATRAC3) {
2125                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2126                 ByteIOContext b;
2127
2128                 init_put_byte(&b, track->codec_priv, track->codec_priv_size, 0,
2129                               NULL, NULL, NULL, NULL);
2130                 url_fskip(&b, 24);
2131                 audiotrack->coded_framesize = get_be32(&b);
2132                 url_fskip(&b, 12);
2133                 audiotrack->sub_packet_h    = get_be16(&b);
2134                 audiotrack->frame_size      = get_be16(&b);
2135                 audiotrack->sub_packet_size = get_be16(&b);
2136                 audiotrack->buf = av_malloc(audiotrack->frame_size * audiotrack->sub_packet_h);
2137                 if (codec_id == CODEC_ID_RA_288) {
2138                     audiotrack->block_align = audiotrack->coded_framesize;
2139                     track->codec_priv_size = 0;
2140                 } else {
2141                     audiotrack->block_align = audiotrack->sub_packet_size;
2142                     extradata_offset = 78;
2143                     track->codec_priv_size -= extradata_offset;
2144                 }
2145             }
2146
2147             else if (codec_id == CODEC_ID_TEXT) {
2148                 MatroskaSubtitleTrack *subtrack=(MatroskaSubtitleTrack *)track;
2149                 if (!strcmp(track->codec_id, "S_TEXT/ASS") ||
2150                     !strcmp(track->codec_id, "S_TEXT/SSA"))
2151                     subtrack->ass = 1;
2152             }
2153
2154             if (codec_id == CODEC_ID_NONE) {
2155                 av_log(matroska->ctx, AV_LOG_INFO,
2156                        "Unknown/unsupported CodecID %s.\n",
2157                        track->codec_id);
2158             }
2159
2160             track->stream_index = matroska->num_streams;
2161
2162             matroska->num_streams++;
2163             st = av_new_stream(s, track->stream_index);
2164             if (st == NULL)
2165                 return AVERROR_NOMEM;
2166             av_set_pts_info(st, 64, matroska->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
2167
2168             st->codec->codec_id = codec_id;
2169             st->start_time = 0;
2170             if (strcmp(track->language, "und"))
2171                 strcpy(st->language, track->language);
2172
2173             if (track->default_duration)
2174                 av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
2175                           track->default_duration, 1000, 30000);
2176
2177             if(extradata){
2178                 st->codec->extradata = extradata;
2179                 st->codec->extradata_size = extradata_size;
2180             } else if(track->codec_priv && track->codec_priv_size > 0){
2181                 st->codec->extradata = av_malloc(track->codec_priv_size);
2182                 if(st->codec->extradata == NULL)
2183                     return AVERROR_NOMEM;
2184                 st->codec->extradata_size = track->codec_priv_size;
2185                 memcpy(st->codec->extradata,track->codec_priv+extradata_offset,
2186                        track->codec_priv_size);
2187             }
2188
2189             if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2190                 MatroskaVideoTrack *videotrack = (MatroskaVideoTrack *)track;
2191
2192                 st->codec->codec_type = CODEC_TYPE_VIDEO;
2193                 st->codec->codec_tag = videotrack->fourcc;
2194                 st->codec->width = videotrack->pixel_width;
2195                 st->codec->height = videotrack->pixel_height;
2196                 if (videotrack->display_width == 0)
2197                     videotrack->display_width= videotrack->pixel_width;
2198                 if (videotrack->display_height == 0)
2199                     videotrack->display_height= videotrack->pixel_height;
2200                 av_reduce(&st->codec->sample_aspect_ratio.num,
2201                           &st->codec->sample_aspect_ratio.den,
2202                           st->codec->height * videotrack->display_width,
2203                           st->codec-> width * videotrack->display_height,
2204                           255);
2205                 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2206             } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2207                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2208
2209                 st->codec->codec_type = CODEC_TYPE_AUDIO;
2210                 st->codec->sample_rate = audiotrack->samplerate;
2211                 st->codec->channels = audiotrack->channels;
2212                 st->codec->block_align = audiotrack->block_align;
2213             } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2214                 st->codec->codec_type = CODEC_TYPE_SUBTITLE;
2215             }
2216
2217             /* What do we do with private data? E.g. for Vorbis. */
2218         }
2219         res = 0;
2220     }
2221
2222     if (matroska->index_parsed) {
2223         int i, track, stream;
2224         for (i=0; i<matroska->num_indexes; i++) {
2225             MatroskaDemuxIndex *idx = &matroska->index[i];
2226             track = matroska_find_track_by_num(matroska, idx->track);
2227             stream = matroska->tracks[track]->stream_index;
2228             if (stream >= 0)
2229                 av_add_index_entry(matroska->ctx->streams[stream],
2230                                    idx->pos, idx->time/matroska->time_scale,
2231                                    0, 0, AVINDEX_KEYFRAME);
2232         }
2233     }
2234
2235     return res;
2236 }
2237
2238 static inline int
2239 rv_offset(uint8_t *data, int slice, int slices)
2240 {
2241     return AV_RL32(data+8*slice+4) + 8*slices;
2242 }
2243
2244 static int
2245 matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data, int size,
2246                      int64_t pos, uint64_t cluster_time, uint64_t duration,
2247                      int is_keyframe, int is_bframe)
2248 {
2249     int res = 0;
2250     int track;
2251     AVStream *st;
2252     AVPacket *pkt;
2253     uint8_t *origdata = data;
2254     int16_t block_time;
2255     uint32_t *lace_size = NULL;
2256     int n, flags, laces = 0;
2257     uint64_t num;
2258
2259     /* first byte(s): tracknum */
2260     if ((n = matroska_ebmlnum_uint(data, size, &num)) < 0) {
2261         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
2262         av_free(origdata);
2263         return res;
2264     }
2265     data += n;
2266     size -= n;
2267
2268     /* fetch track from num */
2269     track = matroska_find_track_by_num(matroska, num);
2270     if (size <= 3 || track < 0 || track >= matroska->num_tracks) {
2271         av_log(matroska->ctx, AV_LOG_INFO,
2272                "Invalid stream %d or size %u\n", track, size);
2273         av_free(origdata);
2274         return res;
2275     }
2276     if (matroska->tracks[track]->stream_index < 0)
2277         return res;
2278     st = matroska->ctx->streams[matroska->tracks[track]->stream_index];
2279     if (st->discard >= AVDISCARD_ALL) {
2280         av_free(origdata);
2281         return res;
2282     }
2283     if (duration == AV_NOPTS_VALUE)
2284         duration = matroska->tracks[track]->default_duration;
2285
2286     /* block_time (relative to cluster time) */
2287     block_time = AV_RB16(data);
2288     data += 2;
2289     flags = *data++;
2290     size -= 3;
2291     if (is_keyframe == -1)
2292         is_keyframe = flags & 1 ? PKT_FLAG_KEY : 0;
2293
2294     if (matroska->skip_to_keyframe) {
2295         if (!is_keyframe || st != matroska->skip_to_stream)
2296             return res;
2297         matroska->skip_to_keyframe = 0;
2298     }
2299
2300     switch ((flags & 0x06) >> 1) {
2301         case 0x0: /* no lacing */
2302             laces = 1;
2303             lace_size = av_mallocz(sizeof(int));
2304             lace_size[0] = size;
2305             break;
2306
2307         case 0x1: /* xiph lacing */
2308         case 0x2: /* fixed-size lacing */
2309         case 0x3: /* EBML lacing */
2310             if (size == 0) {
2311                 res = -1;
2312                 break;
2313             }
2314             laces = (*data) + 1;
2315             data += 1;
2316             size -= 1;
2317             lace_size = av_mallocz(laces * sizeof(int));
2318
2319             switch ((flags & 0x06) >> 1) {
2320                 case 0x1: /* xiph lacing */ {
2321                     uint8_t temp;
2322                     uint32_t total = 0;
2323                     for (n = 0; res == 0 && n < laces - 1; n++) {
2324                         while (1) {
2325                             if (size == 0) {
2326                                 res = -1;
2327                                 break;
2328                             }
2329                             temp = *data;
2330                             lace_size[n] += temp;
2331                             data += 1;
2332                             size -= 1;
2333                             if (temp != 0xff)
2334                                 break;
2335                         }
2336                         total += lace_size[n];
2337                     }
2338                     lace_size[n] = size - total;
2339                     break;
2340                 }
2341
2342                 case 0x2: /* fixed-size lacing */
2343                     for (n = 0; n < laces; n++)
2344                         lace_size[n] = size / laces;
2345                     break;
2346
2347                 case 0x3: /* EBML lacing */ {
2348                     uint32_t total;
2349                     n = matroska_ebmlnum_uint(data, size, &num);
2350                     if (n < 0) {
2351                         av_log(matroska->ctx, AV_LOG_INFO,
2352                                "EBML block data error\n");
2353                         break;
2354                     }
2355                     data += n;
2356                     size -= n;
2357                     total = lace_size[0] = num;
2358                     for (n = 1; res == 0 && n < laces - 1; n++) {
2359                         int64_t snum;
2360                         int r;
2361                         r = matroska_ebmlnum_sint (data, size, &snum);
2362                         if (r < 0) {
2363                             av_log(matroska->ctx, AV_LOG_INFO,
2364                                    "EBML block data error\n");
2365                             break;
2366                         }
2367                         data += r;
2368                         size -= r;
2369                         lace_size[n] = lace_size[n - 1] + snum;
2370                         total += lace_size[n];
2371                     }
2372                     lace_size[n] = size - total;
2373                     break;
2374                 }
2375             }
2376             break;
2377     }
2378
2379     if (res == 0) {
2380         int real_v = matroska->tracks[track]->flags & MATROSKA_TRACK_REAL_V;
2381         uint64_t timecode = AV_NOPTS_VALUE;
2382
2383         if (cluster_time != (uint64_t)-1 && cluster_time + block_time >= 0)
2384             timecode = cluster_time + block_time;
2385
2386         for (n = 0; n < laces; n++) {
2387             int slice, slices = 1;
2388
2389             if (real_v) {
2390                 slices = *data++ + 1;
2391                 lace_size[n]--;
2392             }
2393
2394             for (slice=0; slice<slices; slice++) {
2395                 int slice_size, slice_offset = 0;
2396                 if (real_v)
2397                     slice_offset = rv_offset(data, slice, slices);
2398                 if (slice+1 == slices)
2399                     slice_size = lace_size[n] - slice_offset;
2400                 else
2401                     slice_size = rv_offset(data, slice+1, slices) - slice_offset;
2402
2403                 if (st->codec->codec_id == CODEC_ID_RA_288 ||
2404                     st->codec->codec_id == CODEC_ID_COOK ||
2405                     st->codec->codec_id == CODEC_ID_ATRAC3) {
2406                     MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)matroska->tracks[track];
2407                     int a = st->codec->block_align;
2408                     int sps = audiotrack->sub_packet_size;
2409                     int cfs = audiotrack->coded_framesize;
2410                     int h = audiotrack->sub_packet_h;
2411                     int y = audiotrack->sub_packet_cnt;
2412                     int w = audiotrack->frame_size;
2413                     int x;
2414
2415                     if (!audiotrack->pkt_cnt) {
2416                         if (st->codec->codec_id == CODEC_ID_RA_288)
2417                             for (x=0; x<h/2; x++)
2418                                 memcpy(audiotrack->buf+x*2*w+y*cfs,
2419                                        data+x*cfs, cfs);
2420                         else
2421                             for (x=0; x<w/sps; x++)
2422                                 memcpy(audiotrack->buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
2423
2424                         if (++audiotrack->sub_packet_cnt >= h) {
2425                             audiotrack->sub_packet_cnt = 0;
2426                             audiotrack->pkt_cnt = h*w / a;
2427                         }
2428                     }
2429                     while (audiotrack->pkt_cnt) {
2430                         pkt = av_mallocz(sizeof(AVPacket));
2431                         av_new_packet(pkt, a);
2432                         memcpy(pkt->data, audiotrack->buf
2433                                + a * (h*w / a - audiotrack->pkt_cnt--), a);
2434                         pkt->pos = pos;
2435                         pkt->stream_index = matroska->tracks[track]->stream_index;
2436                         matroska_queue_packet(matroska, pkt);
2437                     }
2438                 } else {
2439                     int offset = 0;
2440
2441                     if (st->codec->codec_id == CODEC_ID_TEXT
2442                         && ((MatroskaSubtitleTrack *)(matroska->tracks[track]))->ass) {
2443                         int i;
2444                         for (i=0; i<8 && data[slice_offset+offset]; offset++)
2445                             if (data[slice_offset+offset] == ',')
2446                                 i++;
2447                     }
2448
2449                     pkt = av_mallocz(sizeof(AVPacket));
2450                     /* XXX: prevent data copy... */
2451                     if (av_new_packet(pkt, slice_size-offset) < 0) {
2452                         res = AVERROR_NOMEM;
2453                         n = laces-1;
2454                         break;
2455                     }
2456                     memcpy (pkt->data, data+slice_offset+offset, slice_size-offset);
2457
2458                     if (n == 0)
2459                         pkt->flags = is_keyframe;
2460                     pkt->stream_index = matroska->tracks[track]->stream_index;
2461
2462                     pkt->pts = timecode;
2463                     pkt->pos = pos;
2464                     pkt->duration = duration;
2465
2466                     matroska_queue_packet(matroska, pkt);
2467                 }
2468
2469                 if (timecode != AV_NOPTS_VALUE)
2470                     timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
2471             }
2472             data += lace_size[n];
2473         }
2474     }
2475
2476     av_free(lace_size);
2477     av_free(origdata);
2478     return res;
2479 }
2480
2481 static int
2482 matroska_parse_blockgroup (MatroskaDemuxContext *matroska,
2483                            uint64_t              cluster_time)
2484 {
2485     int res = 0;
2486     uint32_t id;
2487     int is_bframe = 0;
2488     int is_keyframe = PKT_FLAG_KEY, last_num_packets = matroska->num_packets;
2489     uint64_t duration = AV_NOPTS_VALUE;
2490     uint8_t *data;
2491     int size = 0;
2492     int64_t pos = 0;
2493
2494     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing blockgroup...\n");
2495
2496     while (res == 0) {
2497         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
2498             res = AVERROR_IO;
2499             break;
2500         } else if (matroska->level_up) {
2501             matroska->level_up--;
2502             break;
2503         }
2504
2505         switch (id) {
2506             /* one block inside the group. Note, block parsing is one
2507              * of the harder things, so this code is a bit complicated.
2508              * See http://www.matroska.org/ for documentation. */
2509             case MATROSKA_ID_BLOCK: {
2510                 pos = url_ftell(&matroska->ctx->pb);
2511                 res = ebml_read_binary(matroska, &id, &data, &size);
2512                 break;
2513             }
2514
2515             case MATROSKA_ID_BLOCKDURATION: {
2516                 if ((res = ebml_read_uint(matroska, &id, &duration)) < 0)
2517                     break;
2518                 break;
2519             }
2520
2521             case MATROSKA_ID_BLOCKREFERENCE: {
2522                 int64_t num;
2523                 /* We've found a reference, so not even the first frame in
2524                  * the lace is a key frame. */
2525                 is_keyframe = 0;
2526                 if (last_num_packets != matroska->num_packets)
2527                     matroska->packets[last_num_packets]->flags = 0;
2528                 if ((res = ebml_read_sint(matroska, &id, &num)) < 0)
2529                     break;
2530                 if (num > 0)
2531                     is_bframe = 1;
2532                 break;
2533             }
2534
2535             default:
2536                 av_log(matroska->ctx, AV_LOG_INFO,
2537                        "Unknown entry 0x%x in blockgroup data\n", id);
2538                 /* fall-through */
2539
2540             case EBML_ID_VOID:
2541                 res = ebml_read_skip(matroska);
2542                 break;
2543         }
2544
2545         if (matroska->level_up) {
2546             matroska->level_up--;
2547             break;
2548         }
2549     }
2550
2551     if (res)
2552         return res;
2553
2554     if (size > 0)
2555         res = matroska_parse_block(matroska, data, size, pos, cluster_time,
2556                                    duration, is_keyframe, is_bframe);
2557
2558     return res;
2559 }
2560
2561 static int
2562 matroska_parse_cluster (MatroskaDemuxContext *matroska)
2563 {
2564     int res = 0;
2565     uint32_t id;
2566     uint64_t cluster_time = 0;
2567     uint8_t *data;
2568     int64_t pos;
2569     int size;
2570
2571     av_log(matroska->ctx, AV_LOG_DEBUG,
2572            "parsing cluster at %"PRId64"\n", url_ftell(&matroska->ctx->pb));
2573
2574     while (res == 0) {
2575         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
2576             res = AVERROR_IO;
2577             break;
2578         } else if (matroska->level_up) {
2579             matroska->level_up--;
2580             break;
2581         }
2582
2583         switch (id) {
2584             /* cluster timecode */
2585             case MATROSKA_ID_CLUSTERTIMECODE: {
2586                 uint64_t num;
2587                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
2588                     break;
2589                 cluster_time = num;
2590                 break;
2591             }
2592
2593                 /* a group of blocks inside a cluster */
2594             case MATROSKA_ID_BLOCKGROUP:
2595                 if ((res = ebml_read_master(matroska, &id)) < 0)
2596                     break;
2597                 res = matroska_parse_blockgroup(matroska, cluster_time);
2598                 break;
2599
2600             case MATROSKA_ID_SIMPLEBLOCK:
2601                 pos = url_ftell(&matroska->ctx->pb);
2602                 res = ebml_read_binary(matroska, &id, &data, &size);
2603                 if (res == 0)
2604                     res = matroska_parse_block(matroska, data, size, pos,
2605                                                cluster_time, AV_NOPTS_VALUE,
2606                                                -1, 0);
2607                 break;
2608
2609             default:
2610                 av_log(matroska->ctx, AV_LOG_INFO,
2611                        "Unknown entry 0x%x in cluster data\n", id);
2612                 /* fall-through */
2613
2614             case EBML_ID_VOID:
2615                 res = ebml_read_skip(matroska);
2616                 break;
2617         }
2618
2619         if (matroska->level_up) {
2620             matroska->level_up--;
2621             break;
2622         }
2623     }
2624
2625     return res;
2626 }
2627
2628 static int
2629 matroska_read_packet (AVFormatContext *s,
2630                       AVPacket        *pkt)
2631 {
2632     MatroskaDemuxContext *matroska = s->priv_data;
2633     int res;
2634     uint32_t id;
2635
2636     /* Read stream until we have a packet queued. */
2637     while (matroska_deliver_packet(matroska, pkt)) {
2638
2639         /* Have we already reached the end? */
2640         if (matroska->done)
2641             return AVERROR_IO;
2642
2643         res = 0;
2644         while (res == 0) {
2645             if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
2646                 return AVERROR_IO;
2647             } else if (matroska->level_up) {
2648                 matroska->level_up--;
2649                 break;
2650             }
2651
2652             switch (id) {
2653                 case MATROSKA_ID_CLUSTER:
2654                     if ((res = ebml_read_master(matroska, &id)) < 0)
2655                         break;
2656                     if ((res = matroska_parse_cluster(matroska)) == 0)
2657                         res = 1; /* Parsed one cluster, let's get out. */
2658                     break;
2659
2660                 default:
2661                 case EBML_ID_VOID:
2662                     res = ebml_read_skip(matroska);
2663                     break;
2664             }
2665
2666             if (matroska->level_up) {
2667                 matroska->level_up--;
2668                 break;
2669             }
2670         }
2671
2672         if (res == -1)
2673             matroska->done = 1;
2674     }
2675
2676     return 0;
2677 }
2678
2679 static int
2680 matroska_read_seek (AVFormatContext *s, int stream_index, int64_t timestamp,
2681                     int flags)
2682 {
2683     MatroskaDemuxContext *matroska = s->priv_data;
2684     AVStream *st = s->streams[stream_index];
2685     int index;
2686
2687     /* find index entry */
2688     index = av_index_search_timestamp(st, timestamp, flags);
2689     if (index < 0)
2690         return 0;
2691
2692     /* do the seek */
2693     url_fseek(&s->pb, st->index_entries[index].pos, SEEK_SET);
2694     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
2695     matroska->skip_to_stream = st;
2696     matroska->num_packets = 0;
2697     matroska->peek_id = 0;
2698     return 0;
2699 }
2700
2701 static int
2702 matroska_read_close (AVFormatContext *s)
2703 {
2704     MatroskaDemuxContext *matroska = s->priv_data;
2705     int n = 0;
2706
2707     av_free(matroska->writing_app);
2708     av_free(matroska->muxing_app);
2709     av_free(matroska->index);
2710
2711     if (matroska->packets != NULL) {
2712         for (n = 0; n < matroska->num_packets; n++) {
2713             av_free_packet(matroska->packets[n]);
2714             av_free(matroska->packets[n]);
2715         }
2716         av_free(matroska->packets);
2717     }
2718
2719     for (n = 0; n < matroska->num_tracks; n++) {
2720         MatroskaTrack *track = matroska->tracks[n];
2721         av_free(track->codec_id);
2722         av_free(track->codec_name);
2723         av_free(track->codec_priv);
2724         av_free(track->name);
2725
2726         if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2727             MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2728             av_free(audiotrack->buf);
2729         }
2730
2731         av_free(track);
2732     }
2733
2734     return 0;
2735 }
2736
2737 AVInputFormat matroska_demuxer = {
2738     "matroska",
2739     "Matroska file format",
2740     sizeof(MatroskaDemuxContext),
2741     matroska_probe,
2742     matroska_read_header,
2743     matroska_read_packet,
2744     matroska_read_close,
2745     matroska_read_seek,
2746 };