19dcdfac49aadf716bf1ed4333abd04a55d689c9
[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(EIO); /* 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(ENOMEM);
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(EIO);
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(ENOSYS);
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(ENOMEM);
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(EIO);
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(EIO);
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(EIO);
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(EIO);
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(EIO);
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;
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                             if (!track->default_duration)
1103                                 track->default_duration = 1000000000/num;
1104                             break;
1105                         }
1106
1107                         /* width of the size to display the video at */
1108                         case MATROSKA_ID_VIDEODISPLAYWIDTH: {
1109                             uint64_t num;
1110                             if ((res = ebml_read_uint(matroska, &id,
1111                                                       &num)) < 0)
1112                                 break;
1113                             videotrack->display_width = num;
1114                             break;
1115                         }
1116
1117                         /* height of the size to display the video at */
1118                         case MATROSKA_ID_VIDEODISPLAYHEIGHT: {
1119                             uint64_t num;
1120                             if ((res = ebml_read_uint(matroska, &id,
1121                                                       &num)) < 0)
1122                                 break;
1123                             videotrack->display_height = num;
1124                             break;
1125                         }
1126
1127                         /* width of the video in the file */
1128                         case MATROSKA_ID_VIDEOPIXELWIDTH: {
1129                             uint64_t num;
1130                             if ((res = ebml_read_uint(matroska, &id,
1131                                                       &num)) < 0)
1132                                 break;
1133                             videotrack->pixel_width = num;
1134                             break;
1135                         }
1136
1137                         /* height of the video in the file */
1138                         case MATROSKA_ID_VIDEOPIXELHEIGHT: {
1139                             uint64_t num;
1140                             if ((res = ebml_read_uint(matroska, &id,
1141                                                       &num)) < 0)
1142                                 break;
1143                             videotrack->pixel_height = num;
1144                             break;
1145                         }
1146
1147                         /* whether the video is interlaced */
1148                         case MATROSKA_ID_VIDEOFLAGINTERLACED: {
1149                             uint64_t num;
1150                             if ((res = ebml_read_uint(matroska, &id,
1151                                                       &num)) < 0)
1152                                 break;
1153                             if (num)
1154                                 track->flags |=
1155                                     MATROSKA_VIDEOTRACK_INTERLACED;
1156                             else
1157                                 track->flags &=
1158                                     ~MATROSKA_VIDEOTRACK_INTERLACED;
1159                             break;
1160                         }
1161
1162                         /* stereo mode (whether the video has two streams,
1163                          * where one is for the left eye and the other for
1164                          * the right eye, which creates a 3D-like
1165                          * effect) */
1166                         case MATROSKA_ID_VIDEOSTEREOMODE: {
1167                             uint64_t num;
1168                             if ((res = ebml_read_uint(matroska, &id,
1169                                                       &num)) < 0)
1170                                 break;
1171                             if (num != MATROSKA_EYE_MODE_MONO &&
1172                                 num != MATROSKA_EYE_MODE_LEFT &&
1173                                 num != MATROSKA_EYE_MODE_RIGHT &&
1174                                 num != MATROSKA_EYE_MODE_BOTH) {
1175                                 av_log(matroska->ctx, AV_LOG_INFO,
1176                                        "Ignoring unknown eye mode 0x%x\n",
1177                                        (uint32_t) num);
1178                                 break;
1179                             }
1180                             videotrack->eye_mode = num;
1181                             break;
1182                         }
1183
1184                         /* aspect ratio behaviour */
1185                         case MATROSKA_ID_VIDEOASPECTRATIO: {
1186                             uint64_t num;
1187                             if ((res = ebml_read_uint(matroska, &id,
1188                                                       &num)) < 0)
1189                                 break;
1190                             if (num != MATROSKA_ASPECT_RATIO_MODE_FREE &&
1191                                 num != MATROSKA_ASPECT_RATIO_MODE_KEEP &&
1192                                 num != MATROSKA_ASPECT_RATIO_MODE_FIXED) {
1193                                 av_log(matroska->ctx, AV_LOG_INFO,
1194                                        "Ignoring unknown aspect ratio 0x%x\n",
1195                                        (uint32_t) num);
1196                                 break;
1197                             }
1198                             videotrack->ar_mode = num;
1199                             break;
1200                         }
1201
1202                         /* colorspace (only matters for raw video)
1203                          * fourcc */
1204                         case MATROSKA_ID_VIDEOCOLOURSPACE: {
1205                             uint64_t num;
1206                             if ((res = ebml_read_uint(matroska, &id,
1207                                                       &num)) < 0)
1208                                 break;
1209                             videotrack->fourcc = num;
1210                             break;
1211                         }
1212
1213                         default:
1214                             av_log(matroska->ctx, AV_LOG_INFO,
1215                                    "Unknown video track header entry "
1216                                    "0x%x - ignoring\n", id);
1217                             /* pass-through */
1218
1219                         case EBML_ID_VOID:
1220                             res = ebml_read_skip(matroska);
1221                             break;
1222                     }
1223
1224                     if (matroska->level_up) {
1225                         matroska->level_up--;
1226                         break;
1227                     }
1228                 }
1229                 break;
1230             }
1231
1232             /* tracktype specific stuff for audio */
1233             case MATROSKA_ID_TRACKAUDIO: {
1234                 MatroskaAudioTrack *audiotrack;
1235                 if (!track->type)
1236                     track->type = MATROSKA_TRACK_TYPE_AUDIO;
1237                 if (track->type != MATROSKA_TRACK_TYPE_AUDIO) {
1238                     av_log(matroska->ctx, AV_LOG_INFO,
1239                            "audio data in non-audio track - ignoring\n");
1240                     res = AVERROR_INVALIDDATA;
1241                     break;
1242                 } else if ((res = ebml_read_master(matroska, &id)) < 0)
1243                     break;
1244                 audiotrack = (MatroskaAudioTrack *)track;
1245                 audiotrack->channels = 1;
1246                 audiotrack->samplerate = 8000;
1247
1248                 while (res == 0) {
1249                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1250                         res = AVERROR(EIO);
1251                         break;
1252                     } else if (matroska->level_up > 0) {
1253                         matroska->level_up--;
1254                         break;
1255                     }
1256
1257                     switch (id) {
1258                         /* samplerate */
1259                         case MATROSKA_ID_AUDIOSAMPLINGFREQ: {
1260                             double num;
1261                             if ((res = ebml_read_float(matroska, &id,
1262                                                        &num)) < 0)
1263                                 break;
1264                             audiotrack->internal_samplerate =
1265                             audiotrack->samplerate = num;
1266                             break;
1267                         }
1268
1269                         case MATROSKA_ID_AUDIOOUTSAMPLINGFREQ: {
1270                             double num;
1271                             if ((res = ebml_read_float(matroska, &id,
1272                                                        &num)) < 0)
1273                                 break;
1274                             audiotrack->samplerate = num;
1275                             break;
1276                         }
1277
1278                             /* bitdepth */
1279                         case MATROSKA_ID_AUDIOBITDEPTH: {
1280                             uint64_t num;
1281                             if ((res = ebml_read_uint(matroska, &id,
1282                                                       &num)) < 0)
1283                                 break;
1284                             audiotrack->bitdepth = num;
1285                             break;
1286                         }
1287
1288                             /* channels */
1289                         case MATROSKA_ID_AUDIOCHANNELS: {
1290                             uint64_t num;
1291                             if ((res = ebml_read_uint(matroska, &id,
1292                                                       &num)) < 0)
1293                                 break;
1294                             audiotrack->channels = num;
1295                             break;
1296                         }
1297
1298                         default:
1299                             av_log(matroska->ctx, AV_LOG_INFO,
1300                                    "Unknown audio track header entry "
1301                                    "0x%x - ignoring\n", id);
1302                             /* pass-through */
1303
1304                         case EBML_ID_VOID:
1305                             res = ebml_read_skip(matroska);
1306                             break;
1307                     }
1308
1309                     if (matroska->level_up) {
1310                         matroska->level_up--;
1311                         break;
1312                     }
1313                 }
1314                 break;
1315             }
1316
1317                 /* codec identifier */
1318             case MATROSKA_ID_CODECID: {
1319                 char *text;
1320                 if ((res = ebml_read_ascii(matroska, &id, &text)) < 0)
1321                     break;
1322                 track->codec_id = text;
1323                 break;
1324             }
1325
1326                 /* codec private data */
1327             case MATROSKA_ID_CODECPRIVATE: {
1328                 uint8_t *data;
1329                 int size;
1330                 if ((res = ebml_read_binary(matroska, &id, &data, &size) < 0))
1331                     break;
1332                 track->codec_priv = data;
1333                 track->codec_priv_size = size;
1334                 break;
1335             }
1336
1337                 /* name of the codec */
1338             case MATROSKA_ID_CODECNAME: {
1339                 char *text;
1340                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
1341                     break;
1342                 track->codec_name = text;
1343                 break;
1344             }
1345
1346                 /* name of this track */
1347             case MATROSKA_ID_TRACKNAME: {
1348                 char *text;
1349                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
1350                     break;
1351                 track->name = text;
1352                 break;
1353             }
1354
1355                 /* language (matters for audio/subtitles, mostly) */
1356             case MATROSKA_ID_TRACKLANGUAGE: {
1357                 char *text, *end;
1358                 if ((res = ebml_read_utf8(matroska, &id, &text)) < 0)
1359                     break;
1360                 if ((end = strchr(text, '-')))
1361                     *end = '\0';
1362                 if (strlen(text) == 3)
1363                     strcpy(track->language, text);
1364                 av_free(text);
1365                 break;
1366             }
1367
1368                 /* whether this is actually used */
1369             case MATROSKA_ID_TRACKFLAGENABLED: {
1370                 uint64_t num;
1371                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1372                     break;
1373                 if (num)
1374                     track->flags |= MATROSKA_TRACK_ENABLED;
1375                 else
1376                     track->flags &= ~MATROSKA_TRACK_ENABLED;
1377                 break;
1378             }
1379
1380                 /* whether it's the default for this track type */
1381             case MATROSKA_ID_TRACKFLAGDEFAULT: {
1382                 uint64_t num;
1383                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1384                     break;
1385                 if (num)
1386                     track->flags |= MATROSKA_TRACK_DEFAULT;
1387                 else
1388                     track->flags &= ~MATROSKA_TRACK_DEFAULT;
1389                 break;
1390             }
1391
1392                 /* lacing (like MPEG, where blocks don't end/start on frame
1393                  * boundaries) */
1394             case MATROSKA_ID_TRACKFLAGLACING: {
1395                 uint64_t num;
1396                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1397                     break;
1398                 if (num)
1399                     track->flags |= MATROSKA_TRACK_LACING;
1400                 else
1401                     track->flags &= ~MATROSKA_TRACK_LACING;
1402                 break;
1403             }
1404
1405                 /* default length (in time) of one data block in this track */
1406             case MATROSKA_ID_TRACKDEFAULTDURATION: {
1407                 uint64_t num;
1408                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
1409                     break;
1410                 track->default_duration = num;
1411                 break;
1412             }
1413
1414             default:
1415                 av_log(matroska->ctx, AV_LOG_INFO,
1416                        "Unknown track header entry 0x%x - ignoring\n", id);
1417                 /* pass-through */
1418
1419             case EBML_ID_VOID:
1420             /* we ignore these because they're nothing useful. */
1421             case MATROSKA_ID_CODECINFOURL:
1422             case MATROSKA_ID_CODECDOWNLOADURL:
1423             case MATROSKA_ID_TRACKMINCACHE:
1424             case MATROSKA_ID_TRACKMAXCACHE:
1425                 res = ebml_read_skip(matroska);
1426                 break;
1427         }
1428
1429         if (matroska->level_up) {
1430             matroska->level_up--;
1431             break;
1432         }
1433     }
1434
1435     return res;
1436 }
1437
1438 static int
1439 matroska_parse_tracks (MatroskaDemuxContext *matroska)
1440 {
1441     int res = 0;
1442     uint32_t id;
1443
1444     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing tracks...\n");
1445
1446     while (res == 0) {
1447         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1448             res = AVERROR(EIO);
1449             break;
1450         } else if (matroska->level_up) {
1451             matroska->level_up--;
1452             break;
1453         }
1454
1455         switch (id) {
1456             /* one track within the "all-tracks" header */
1457             case MATROSKA_ID_TRACKENTRY:
1458                 res = matroska_add_stream(matroska);
1459                 break;
1460
1461             default:
1462                 av_log(matroska->ctx, AV_LOG_INFO,
1463                        "Unknown entry 0x%x in track header\n", id);
1464                 /* fall-through */
1465
1466             case EBML_ID_VOID:
1467                 res = ebml_read_skip(matroska);
1468                 break;
1469         }
1470
1471         if (matroska->level_up) {
1472             matroska->level_up--;
1473             break;
1474         }
1475     }
1476
1477     return res;
1478 }
1479
1480 static int
1481 matroska_parse_index (MatroskaDemuxContext *matroska)
1482 {
1483     int res = 0;
1484     uint32_t id;
1485     MatroskaDemuxIndex idx;
1486
1487     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing index...\n");
1488
1489     while (res == 0) {
1490         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1491             res = AVERROR(EIO);
1492             break;
1493         } else if (matroska->level_up) {
1494             matroska->level_up--;
1495             break;
1496         }
1497
1498         switch (id) {
1499             /* one single index entry ('point') */
1500             case MATROSKA_ID_POINTENTRY:
1501                 if ((res = ebml_read_master(matroska, &id)) < 0)
1502                     break;
1503
1504                 /* in the end, we hope to fill one entry with a
1505                  * timestamp, a file position and a tracknum */
1506                 idx.pos   = (uint64_t) -1;
1507                 idx.time  = (uint64_t) -1;
1508                 idx.track = (uint16_t) -1;
1509
1510                 while (res == 0) {
1511                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1512                         res = AVERROR(EIO);
1513                         break;
1514                     } else if (matroska->level_up) {
1515                         matroska->level_up--;
1516                         break;
1517                     }
1518
1519                     switch (id) {
1520                         /* one single index entry ('point') */
1521                         case MATROSKA_ID_CUETIME: {
1522                             uint64_t time;
1523                             if ((res = ebml_read_uint(matroska, &id,
1524                                                       &time)) < 0)
1525                                 break;
1526                             idx.time = time * matroska->time_scale;
1527                             break;
1528                         }
1529
1530                         /* position in the file + track to which it
1531                          * belongs */
1532                         case MATROSKA_ID_CUETRACKPOSITION:
1533                             if ((res = ebml_read_master(matroska, &id)) < 0)
1534                                 break;
1535
1536                             while (res == 0) {
1537                                 if (!(id = ebml_peek_id (matroska,
1538                                                     &matroska->level_up))) {
1539                                     res = AVERROR(EIO);
1540                                     break;
1541                                 } else if (matroska->level_up) {
1542                                     matroska->level_up--;
1543                                     break;
1544                                 }
1545
1546                                 switch (id) {
1547                                     /* track number */
1548                                     case MATROSKA_ID_CUETRACK: {
1549                                         uint64_t num;
1550                                         if ((res = ebml_read_uint(matroska,
1551                                                           &id, &num)) < 0)
1552                                             break;
1553                                         idx.track = num;
1554                                         break;
1555                                     }
1556
1557                                         /* position in file */
1558                                     case MATROSKA_ID_CUECLUSTERPOSITION: {
1559                                         uint64_t num;
1560                                         if ((res = ebml_read_uint(matroska,
1561                                                           &id, &num)) < 0)
1562                                             break;
1563                                         idx.pos = num+matroska->segment_start;
1564                                         break;
1565                                     }
1566
1567                                     default:
1568                                         av_log(matroska->ctx, AV_LOG_INFO,
1569                                                "Unknown entry 0x%x in "
1570                                                "CuesTrackPositions\n", id);
1571                                         /* fall-through */
1572
1573                                     case EBML_ID_VOID:
1574                                         res = ebml_read_skip(matroska);
1575                                         break;
1576                                 }
1577
1578                                 if (matroska->level_up) {
1579                                     matroska->level_up--;
1580                                     break;
1581                                 }
1582                             }
1583
1584                             break;
1585
1586                         default:
1587                             av_log(matroska->ctx, AV_LOG_INFO,
1588                                    "Unknown entry 0x%x in cuespoint "
1589                                    "index\n", id);
1590                             /* fall-through */
1591
1592                         case EBML_ID_VOID:
1593                             res = ebml_read_skip(matroska);
1594                             break;
1595                     }
1596
1597                     if (matroska->level_up) {
1598                         matroska->level_up--;
1599                         break;
1600                     }
1601                 }
1602
1603                 /* so let's see if we got what we wanted */
1604                 if (idx.pos   != (uint64_t) -1 &&
1605                     idx.time  != (uint64_t) -1 &&
1606                     idx.track != (uint16_t) -1) {
1607                     if (matroska->num_indexes % 32 == 0) {
1608                         /* re-allocate bigger index */
1609                         matroska->index =
1610                             av_realloc(matroska->index,
1611                                        (matroska->num_indexes + 32) *
1612                                        sizeof(MatroskaDemuxIndex));
1613                     }
1614                     matroska->index[matroska->num_indexes] = idx;
1615                     matroska->num_indexes++;
1616                 }
1617                 break;
1618
1619             default:
1620                 av_log(matroska->ctx, AV_LOG_INFO,
1621                        "Unknown entry 0x%x in cues header\n", id);
1622                 /* fall-through */
1623
1624             case EBML_ID_VOID:
1625                 res = ebml_read_skip(matroska);
1626                 break;
1627         }
1628
1629         if (matroska->level_up) {
1630             matroska->level_up--;
1631             break;
1632         }
1633     }
1634
1635     return res;
1636 }
1637
1638 static int
1639 matroska_parse_metadata (MatroskaDemuxContext *matroska)
1640 {
1641     int res = 0;
1642     uint32_t id;
1643
1644     while (res == 0) {
1645         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1646             res = AVERROR(EIO);
1647             break;
1648         } else if (matroska->level_up) {
1649             matroska->level_up--;
1650             break;
1651         }
1652
1653         switch (id) {
1654             /* Hm, this is unsupported... */
1655             default:
1656                 av_log(matroska->ctx, AV_LOG_INFO,
1657                        "Unknown entry 0x%x in metadata header\n", id);
1658                 /* fall-through */
1659
1660             case EBML_ID_VOID:
1661                 res = ebml_read_skip(matroska);
1662                 break;
1663         }
1664
1665         if (matroska->level_up) {
1666             matroska->level_up--;
1667             break;
1668         }
1669     }
1670
1671     return res;
1672 }
1673
1674 static int
1675 matroska_parse_seekhead (MatroskaDemuxContext *matroska)
1676 {
1677     int res = 0;
1678     uint32_t id;
1679
1680     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing seekhead...\n");
1681
1682     while (res == 0) {
1683         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1684             res = AVERROR(EIO);
1685             break;
1686         } else if (matroska->level_up) {
1687             matroska->level_up--;
1688             break;
1689         }
1690
1691         switch (id) {
1692             case MATROSKA_ID_SEEKENTRY: {
1693                 uint32_t seek_id = 0, peek_id_cache = 0;
1694                 uint64_t seek_pos = (uint64_t) -1, t;
1695
1696                 if ((res = ebml_read_master(matroska, &id)) < 0)
1697                     break;
1698
1699                 while (res == 0) {
1700                     if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1701                         res = AVERROR(EIO);
1702                         break;
1703                     } else if (matroska->level_up) {
1704                         matroska->level_up--;
1705                         break;
1706                     }
1707
1708                     switch (id) {
1709                         case MATROSKA_ID_SEEKID:
1710                             res = ebml_read_uint(matroska, &id, &t);
1711                             seek_id = t;
1712                             break;
1713
1714                         case MATROSKA_ID_SEEKPOSITION:
1715                             res = ebml_read_uint(matroska, &id, &seek_pos);
1716                             break;
1717
1718                         default:
1719                             av_log(matroska->ctx, AV_LOG_INFO,
1720                                    "Unknown seekhead ID 0x%x\n", id);
1721                             /* fall-through */
1722
1723                         case EBML_ID_VOID:
1724                             res = ebml_read_skip(matroska);
1725                             break;
1726                     }
1727
1728                     if (matroska->level_up) {
1729                         matroska->level_up--;
1730                         break;
1731                     }
1732                 }
1733
1734                 if (!seek_id || seek_pos == (uint64_t) -1) {
1735                     av_log(matroska->ctx, AV_LOG_INFO,
1736                            "Incomplete seekhead entry (0x%x/%"PRIu64")\n",
1737                            seek_id, seek_pos);
1738                     break;
1739                 }
1740
1741                 switch (seek_id) {
1742                     case MATROSKA_ID_CUES:
1743                     case MATROSKA_ID_TAGS: {
1744                         uint32_t level_up = matroska->level_up;
1745                         offset_t before_pos;
1746                         uint64_t length;
1747                         MatroskaLevel level;
1748
1749                         /* remember the peeked ID and the current position */
1750                         peek_id_cache = matroska->peek_id;
1751                         before_pos = url_ftell(matroska->ctx->pb);
1752
1753                         /* seek */
1754                         if ((res = ebml_read_seek(matroska, seek_pos +
1755                                                matroska->segment_start)) < 0)
1756                             return res;
1757
1758                         /* we don't want to lose our seekhead level, so we add
1759                          * a dummy. This is a crude hack. */
1760                         if (matroska->num_levels == EBML_MAX_DEPTH) {
1761                             av_log(matroska->ctx, AV_LOG_INFO,
1762                                    "Max EBML element depth (%d) reached, "
1763                                    "cannot parse further.\n", EBML_MAX_DEPTH);
1764                             return AVERROR_UNKNOWN;
1765                         }
1766
1767                         level.start = 0;
1768                         level.length = (uint64_t)-1;
1769                         matroska->levels[matroska->num_levels] = level;
1770                         matroska->num_levels++;
1771
1772                         /* check ID */
1773                         if (!(id = ebml_peek_id (matroska,
1774                                                  &matroska->level_up)))
1775                             goto finish;
1776                         if (id != seek_id) {
1777                             av_log(matroska->ctx, AV_LOG_INFO,
1778                                    "We looked for ID=0x%x but got "
1779                                    "ID=0x%x (pos=%"PRIu64")",
1780                                    seek_id, id, seek_pos +
1781                                    matroska->segment_start);
1782                             goto finish;
1783                         }
1784
1785                         /* read master + parse */
1786                         if ((res = ebml_read_master(matroska, &id)) < 0)
1787                             goto finish;
1788                         switch (id) {
1789                             case MATROSKA_ID_CUES:
1790                                 if (!(res = matroska_parse_index(matroska)) ||
1791                                     url_feof(matroska->ctx->pb)) {
1792                                     matroska->index_parsed = 1;
1793                                     res = 0;
1794                                 }
1795                                 break;
1796                             case MATROSKA_ID_TAGS:
1797                                 if (!(res = matroska_parse_metadata(matroska)) ||
1798                                    url_feof(matroska->ctx->pb)) {
1799                                     matroska->metadata_parsed = 1;
1800                                     res = 0;
1801                                 }
1802                                 break;
1803                         }
1804
1805                     finish:
1806                         /* remove dummy level */
1807                         while (matroska->num_levels) {
1808                             matroska->num_levels--;
1809                             length =
1810                                 matroska->levels[matroska->num_levels].length;
1811                             if (length == (uint64_t)-1)
1812                                 break;
1813                         }
1814
1815                         /* seek back */
1816                         if ((res = ebml_read_seek(matroska, before_pos)) < 0)
1817                             return res;
1818                         matroska->peek_id = peek_id_cache;
1819                         matroska->level_up = level_up;
1820                         break;
1821                     }
1822
1823                     default:
1824                         av_log(matroska->ctx, AV_LOG_INFO,
1825                                "Ignoring seekhead entry for ID=0x%x\n",
1826                                seek_id);
1827                         break;
1828                 }
1829
1830                 break;
1831             }
1832
1833             default:
1834                 av_log(matroska->ctx, AV_LOG_INFO,
1835                        "Unknown seekhead ID 0x%x\n", id);
1836                 /* fall-through */
1837
1838             case EBML_ID_VOID:
1839                 res = ebml_read_skip(matroska);
1840                 break;
1841         }
1842
1843         if (matroska->level_up) {
1844             matroska->level_up--;
1845             break;
1846         }
1847     }
1848
1849     return res;
1850 }
1851
1852 #define ARRAY_SIZE(x)  (sizeof(x)/sizeof(*x))
1853
1854 static int
1855 matroska_aac_profile (char *codec_id)
1856 {
1857     static const char *aac_profiles[] = {
1858         "MAIN", "LC", "SSR"
1859     };
1860     int profile;
1861
1862     for (profile=0; profile<ARRAY_SIZE(aac_profiles); profile++)
1863         if (strstr(codec_id, aac_profiles[profile]))
1864             break;
1865     return profile + 1;
1866 }
1867
1868 static int
1869 matroska_aac_sri (int samplerate)
1870 {
1871     static const int aac_sample_rates[] = {
1872         96000, 88200, 64000, 48000, 44100, 32000,
1873         24000, 22050, 16000, 12000, 11025,  8000,
1874     };
1875     int sri;
1876
1877     for (sri=0; sri<ARRAY_SIZE(aac_sample_rates); sri++)
1878         if (aac_sample_rates[sri] == samplerate)
1879             break;
1880     return sri;
1881 }
1882
1883 static int
1884 matroska_read_header (AVFormatContext    *s,
1885                       AVFormatParameters *ap)
1886 {
1887     MatroskaDemuxContext *matroska = s->priv_data;
1888     char *doctype;
1889     int version, last_level, res = 0;
1890     uint32_t id;
1891
1892     matroska->ctx = s;
1893
1894     /* First read the EBML header. */
1895     doctype = NULL;
1896     if ((res = ebml_read_header(matroska, &doctype, &version)) < 0)
1897         return res;
1898     if ((doctype == NULL) || strcmp(doctype, "matroska")) {
1899         av_log(matroska->ctx, AV_LOG_ERROR,
1900                "Wrong EBML doctype ('%s' != 'matroska').\n",
1901                doctype ? doctype : "(none)");
1902         if (doctype)
1903             av_free(doctype);
1904         return AVERROR_NOFMT;
1905     }
1906     av_free(doctype);
1907     if (version > 2) {
1908         av_log(matroska->ctx, AV_LOG_ERROR,
1909                "Matroska demuxer version 2 too old for file version %d\n",
1910                version);
1911         return AVERROR_NOFMT;
1912     }
1913
1914     /* The next thing is a segment. */
1915     while (1) {
1916         if (!(id = ebml_peek_id(matroska, &last_level)))
1917             return AVERROR(EIO);
1918         if (id == MATROSKA_ID_SEGMENT)
1919             break;
1920
1921         /* oi! */
1922         av_log(matroska->ctx, AV_LOG_INFO,
1923                "Expected a Segment ID (0x%x), but received 0x%x!\n",
1924                MATROSKA_ID_SEGMENT, id);
1925         if ((res = ebml_read_skip(matroska)) < 0)
1926             return res;
1927     }
1928
1929     /* We now have a Matroska segment.
1930      * Seeks are from the beginning of the segment,
1931      * after the segment ID/length. */
1932     if ((res = ebml_read_master(matroska, &id)) < 0)
1933         return res;
1934     matroska->segment_start = url_ftell(s->pb);
1935
1936     matroska->time_scale = 1000000;
1937     /* we've found our segment, start reading the different contents in here */
1938     while (res == 0) {
1939         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
1940             res = AVERROR(EIO);
1941             break;
1942         } else if (matroska->level_up) {
1943             matroska->level_up--;
1944             break;
1945         }
1946
1947         switch (id) {
1948             /* stream info */
1949             case MATROSKA_ID_INFO: {
1950                 if ((res = ebml_read_master(matroska, &id)) < 0)
1951                     break;
1952                 res = matroska_parse_info(matroska);
1953                 break;
1954             }
1955
1956             /* track info headers */
1957             case MATROSKA_ID_TRACKS: {
1958                 if ((res = ebml_read_master(matroska, &id)) < 0)
1959                     break;
1960                 res = matroska_parse_tracks(matroska);
1961                 break;
1962             }
1963
1964             /* stream index */
1965             case MATROSKA_ID_CUES: {
1966                 if (!matroska->index_parsed) {
1967                     if ((res = ebml_read_master(matroska, &id)) < 0)
1968                         break;
1969                     res = matroska_parse_index(matroska);
1970                 } else
1971                     res = ebml_read_skip(matroska);
1972                 break;
1973             }
1974
1975             /* metadata */
1976             case MATROSKA_ID_TAGS: {
1977                 if (!matroska->metadata_parsed) {
1978                     if ((res = ebml_read_master(matroska, &id)) < 0)
1979                         break;
1980                     res = matroska_parse_metadata(matroska);
1981                 } else
1982                     res = ebml_read_skip(matroska);
1983                 break;
1984             }
1985
1986             /* file index (if seekable, seek to Cues/Tags to parse it) */
1987             case MATROSKA_ID_SEEKHEAD: {
1988                 if ((res = ebml_read_master(matroska, &id)) < 0)
1989                     break;
1990                 res = matroska_parse_seekhead(matroska);
1991                 break;
1992             }
1993
1994             case MATROSKA_ID_CLUSTER: {
1995                 /* Do not read the master - this will be done in the next
1996                  * call to matroska_read_packet. */
1997                 res = 1;
1998                 break;
1999             }
2000
2001             default:
2002                 av_log(matroska->ctx, AV_LOG_INFO,
2003                        "Unknown matroska file header ID 0x%x\n", id);
2004             /* fall-through */
2005
2006             case EBML_ID_VOID:
2007                 res = ebml_read_skip(matroska);
2008                 break;
2009         }
2010
2011         if (matroska->level_up) {
2012             matroska->level_up--;
2013             break;
2014         }
2015     }
2016
2017     /* Have we found a cluster? */
2018     if (ebml_peek_id(matroska, NULL) == MATROSKA_ID_CLUSTER) {
2019         int i, j;
2020         MatroskaTrack *track;
2021         AVStream *st;
2022
2023         for (i = 0; i < matroska->num_tracks; i++) {
2024             enum CodecID codec_id = CODEC_ID_NONE;
2025             uint8_t *extradata = NULL;
2026             int extradata_size = 0;
2027             int extradata_offset = 0;
2028             track = matroska->tracks[i];
2029             track->stream_index = -1;
2030
2031             /* Apply some sanity checks. */
2032             if (track->codec_id == NULL)
2033                 continue;
2034
2035             for(j=0; ff_mkv_codec_tags[j].str; j++){
2036                 if(!strncmp(ff_mkv_codec_tags[j].str, track->codec_id,
2037                             strlen(ff_mkv_codec_tags[j].str))){
2038                     codec_id= ff_mkv_codec_tags[j].id;
2039                     break;
2040                 }
2041             }
2042
2043             /* Set the FourCC from the CodecID. */
2044             /* This is the MS compatibility mode which stores a
2045              * BITMAPINFOHEADER in the CodecPrivate. */
2046             if (!strcmp(track->codec_id,
2047                         MATROSKA_CODEC_ID_VIDEO_VFW_FOURCC) &&
2048                 (track->codec_priv_size >= 40) &&
2049                 (track->codec_priv != NULL)) {
2050                 MatroskaVideoTrack *vtrack = (MatroskaVideoTrack *) track;
2051
2052                 /* Offset of biCompression. Stored in LE. */
2053                 vtrack->fourcc = AV_RL32(track->codec_priv + 16);
2054                 codec_id = codec_get_id(codec_bmp_tags, vtrack->fourcc);
2055
2056             }
2057
2058             /* This is the MS compatibility mode which stores a
2059              * WAVEFORMATEX in the CodecPrivate. */
2060             else if (!strcmp(track->codec_id,
2061                              MATROSKA_CODEC_ID_AUDIO_ACM) &&
2062                 (track->codec_priv_size >= 18) &&
2063                 (track->codec_priv != NULL)) {
2064                 uint16_t tag;
2065
2066                 /* Offset of wFormatTag. Stored in LE. */
2067                 tag = AV_RL16(track->codec_priv);
2068                 codec_id = codec_get_id(codec_wav_tags, tag);
2069
2070             }
2071
2072             else if (codec_id == CODEC_ID_AAC && !track->codec_priv_size) {
2073                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *) track;
2074                 int profile = matroska_aac_profile(track->codec_id);
2075                 int sri = matroska_aac_sri(audiotrack->internal_samplerate);
2076                 extradata = av_malloc(5);
2077                 if (extradata == NULL)
2078                     return AVERROR(ENOMEM);
2079                 extradata[0] = (profile << 3) | ((sri&0x0E) >> 1);
2080                 extradata[1] = ((sri&0x01) << 7) | (audiotrack->channels<<3);
2081                 if (strstr(track->codec_id, "SBR")) {
2082                     sri = matroska_aac_sri(audiotrack->samplerate);
2083                     extradata[2] = 0x56;
2084                     extradata[3] = 0xE5;
2085                     extradata[4] = 0x80 | (sri<<3);
2086                     extradata_size = 5;
2087                 } else {
2088                     extradata_size = 2;
2089                 }
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(ENOMEM);
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             }
2114
2115             else if (codec_id == CODEC_ID_RA_144) {
2116                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2117                 audiotrack->samplerate = 8000;
2118                 audiotrack->channels = 1;
2119             }
2120
2121             else if (codec_id == CODEC_ID_RA_288 ||
2122                      codec_id == CODEC_ID_COOK ||
2123                      codec_id == CODEC_ID_ATRAC3) {
2124                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2125                 ByteIOContext b;
2126
2127                 init_put_byte(&b, track->codec_priv, track->codec_priv_size, 0,
2128                               NULL, NULL, NULL, NULL);
2129                 url_fskip(&b, 24);
2130                 audiotrack->coded_framesize = get_be32(&b);
2131                 url_fskip(&b, 12);
2132                 audiotrack->sub_packet_h    = get_be16(&b);
2133                 audiotrack->frame_size      = get_be16(&b);
2134                 audiotrack->sub_packet_size = get_be16(&b);
2135                 audiotrack->buf = av_malloc(audiotrack->frame_size * audiotrack->sub_packet_h);
2136                 if (codec_id == CODEC_ID_RA_288) {
2137                     audiotrack->block_align = audiotrack->coded_framesize;
2138                     track->codec_priv_size = 0;
2139                 } else {
2140                     audiotrack->block_align = audiotrack->sub_packet_size;
2141                     extradata_offset = 78;
2142                     track->codec_priv_size -= extradata_offset;
2143                 }
2144             }
2145
2146             else if (codec_id == CODEC_ID_TEXT) {
2147                 MatroskaSubtitleTrack *subtrack=(MatroskaSubtitleTrack *)track;
2148                 if (!strcmp(track->codec_id, "S_TEXT/ASS") ||
2149                     !strcmp(track->codec_id, "S_TEXT/SSA") ||
2150                     !strcmp(track->codec_id, "S_ASS") ||
2151                     !strcmp(track->codec_id, "S_SSA"))
2152                     subtrack->ass = 1;
2153             }
2154
2155             if (codec_id == CODEC_ID_NONE) {
2156                 av_log(matroska->ctx, AV_LOG_INFO,
2157                        "Unknown/unsupported CodecID %s.\n",
2158                        track->codec_id);
2159             }
2160
2161             track->stream_index = matroska->num_streams;
2162
2163             matroska->num_streams++;
2164             st = av_new_stream(s, track->stream_index);
2165             if (st == NULL)
2166                 return AVERROR(ENOMEM);
2167             av_set_pts_info(st, 64, matroska->time_scale, 1000*1000*1000); /* 64 bit pts in ns */
2168
2169             st->codec->codec_id = codec_id;
2170             st->start_time = 0;
2171             if (strcmp(track->language, "und"))
2172                 strcpy(st->language, track->language);
2173
2174             if (track->default_duration)
2175                 av_reduce(&st->codec->time_base.num, &st->codec->time_base.den,
2176                           track->default_duration, 1000000000, 30000);
2177
2178             if(extradata){
2179                 st->codec->extradata = extradata;
2180                 st->codec->extradata_size = extradata_size;
2181             } else if(track->codec_priv && track->codec_priv_size > 0){
2182                 st->codec->extradata = av_malloc(track->codec_priv_size);
2183                 if(st->codec->extradata == NULL)
2184                     return AVERROR(ENOMEM);
2185                 st->codec->extradata_size = track->codec_priv_size;
2186                 memcpy(st->codec->extradata,track->codec_priv+extradata_offset,
2187                        track->codec_priv_size);
2188             }
2189
2190             if (track->type == MATROSKA_TRACK_TYPE_VIDEO) {
2191                 MatroskaVideoTrack *videotrack = (MatroskaVideoTrack *)track;
2192
2193                 st->codec->codec_type = CODEC_TYPE_VIDEO;
2194                 st->codec->codec_tag = videotrack->fourcc;
2195                 st->codec->width = videotrack->pixel_width;
2196                 st->codec->height = videotrack->pixel_height;
2197                 if (videotrack->display_width == 0)
2198                     videotrack->display_width= videotrack->pixel_width;
2199                 if (videotrack->display_height == 0)
2200                     videotrack->display_height= videotrack->pixel_height;
2201                 av_reduce(&st->codec->sample_aspect_ratio.num,
2202                           &st->codec->sample_aspect_ratio.den,
2203                           st->codec->height * videotrack->display_width,
2204                           st->codec-> width * videotrack->display_height,
2205                           255);
2206                 st->need_parsing = AVSTREAM_PARSE_HEADERS;
2207             } else if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2208                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2209
2210                 st->codec->codec_type = CODEC_TYPE_AUDIO;
2211                 st->codec->sample_rate = audiotrack->samplerate;
2212                 st->codec->channels = audiotrack->channels;
2213                 st->codec->block_align = audiotrack->block_align;
2214             } else if (track->type == MATROSKA_TRACK_TYPE_SUBTITLE) {
2215                 st->codec->codec_type = CODEC_TYPE_SUBTITLE;
2216             }
2217
2218             /* What do we do with private data? E.g. for Vorbis. */
2219         }
2220         res = 0;
2221     }
2222
2223     if (matroska->index_parsed) {
2224         int i, track, stream;
2225         for (i=0; i<matroska->num_indexes; i++) {
2226             MatroskaDemuxIndex *idx = &matroska->index[i];
2227             track = matroska_find_track_by_num(matroska, idx->track);
2228             stream = matroska->tracks[track]->stream_index;
2229             if (stream >= 0)
2230                 av_add_index_entry(matroska->ctx->streams[stream],
2231                                    idx->pos, idx->time/matroska->time_scale,
2232                                    0, 0, AVINDEX_KEYFRAME);
2233         }
2234     }
2235
2236     return res;
2237 }
2238
2239 static int
2240 matroska_parse_block(MatroskaDemuxContext *matroska, uint8_t *data, int size,
2241                      int64_t pos, uint64_t cluster_time, uint64_t duration,
2242                      int is_keyframe, int is_bframe)
2243 {
2244     int res = 0;
2245     int track;
2246     AVStream *st;
2247     AVPacket *pkt;
2248     uint8_t *origdata = data;
2249     int16_t block_time;
2250     uint32_t *lace_size = NULL;
2251     int n, flags, laces = 0;
2252     uint64_t num;
2253
2254     /* first byte(s): tracknum */
2255     if ((n = matroska_ebmlnum_uint(data, size, &num)) < 0) {
2256         av_log(matroska->ctx, AV_LOG_ERROR, "EBML block data error\n");
2257         av_free(origdata);
2258         return res;
2259     }
2260     data += n;
2261     size -= n;
2262
2263     /* fetch track from num */
2264     track = matroska_find_track_by_num(matroska, num);
2265     if (size <= 3 || track < 0 || track >= matroska->num_tracks) {
2266         av_log(matroska->ctx, AV_LOG_INFO,
2267                "Invalid stream %d or size %u\n", track, size);
2268         av_free(origdata);
2269         return res;
2270     }
2271     if (matroska->tracks[track]->stream_index < 0)
2272         return res;
2273     st = matroska->ctx->streams[matroska->tracks[track]->stream_index];
2274     if (st->discard >= AVDISCARD_ALL) {
2275         av_free(origdata);
2276         return res;
2277     }
2278     if (duration == AV_NOPTS_VALUE)
2279         duration = matroska->tracks[track]->default_duration / matroska->time_scale;
2280
2281     /* block_time (relative to cluster time) */
2282     block_time = AV_RB16(data);
2283     data += 2;
2284     flags = *data++;
2285     size -= 3;
2286     if (is_keyframe == -1)
2287         is_keyframe = flags & 0x80 ? PKT_FLAG_KEY : 0;
2288
2289     if (matroska->skip_to_keyframe) {
2290         if (!is_keyframe || st != matroska->skip_to_stream)
2291             return res;
2292         matroska->skip_to_keyframe = 0;
2293     }
2294
2295     switch ((flags & 0x06) >> 1) {
2296         case 0x0: /* no lacing */
2297             laces = 1;
2298             lace_size = av_mallocz(sizeof(int));
2299             lace_size[0] = size;
2300             break;
2301
2302         case 0x1: /* xiph lacing */
2303         case 0x2: /* fixed-size lacing */
2304         case 0x3: /* EBML lacing */
2305             if (size == 0) {
2306                 res = -1;
2307                 break;
2308             }
2309             laces = (*data) + 1;
2310             data += 1;
2311             size -= 1;
2312             lace_size = av_mallocz(laces * sizeof(int));
2313
2314             switch ((flags & 0x06) >> 1) {
2315                 case 0x1: /* xiph lacing */ {
2316                     uint8_t temp;
2317                     uint32_t total = 0;
2318                     for (n = 0; res == 0 && n < laces - 1; n++) {
2319                         while (1) {
2320                             if (size == 0) {
2321                                 res = -1;
2322                                 break;
2323                             }
2324                             temp = *data;
2325                             lace_size[n] += temp;
2326                             data += 1;
2327                             size -= 1;
2328                             if (temp != 0xff)
2329                                 break;
2330                         }
2331                         total += lace_size[n];
2332                     }
2333                     lace_size[n] = size - total;
2334                     break;
2335                 }
2336
2337                 case 0x2: /* fixed-size lacing */
2338                     for (n = 0; n < laces; n++)
2339                         lace_size[n] = size / laces;
2340                     break;
2341
2342                 case 0x3: /* EBML lacing */ {
2343                     uint32_t total;
2344                     n = matroska_ebmlnum_uint(data, size, &num);
2345                     if (n < 0) {
2346                         av_log(matroska->ctx, AV_LOG_INFO,
2347                                "EBML block data error\n");
2348                         break;
2349                     }
2350                     data += n;
2351                     size -= n;
2352                     total = lace_size[0] = num;
2353                     for (n = 1; res == 0 && n < laces - 1; n++) {
2354                         int64_t snum;
2355                         int r;
2356                         r = matroska_ebmlnum_sint (data, size, &snum);
2357                         if (r < 0) {
2358                             av_log(matroska->ctx, AV_LOG_INFO,
2359                                    "EBML block data error\n");
2360                             break;
2361                         }
2362                         data += r;
2363                         size -= r;
2364                         lace_size[n] = lace_size[n - 1] + snum;
2365                         total += lace_size[n];
2366                     }
2367                     lace_size[n] = size - total;
2368                     break;
2369                 }
2370             }
2371             break;
2372     }
2373
2374     if (res == 0) {
2375         uint64_t timecode = AV_NOPTS_VALUE;
2376
2377         if (cluster_time != (uint64_t)-1
2378             && (block_time >= 0 || cluster_time >= -block_time))
2379             timecode = cluster_time + block_time;
2380
2381         for (n = 0; n < laces; n++) {
2382             if (st->codec->codec_id == CODEC_ID_RA_288 ||
2383                 st->codec->codec_id == CODEC_ID_COOK ||
2384                 st->codec->codec_id == CODEC_ID_ATRAC3) {
2385                 MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)matroska->tracks[track];
2386                 int a = st->codec->block_align;
2387                 int sps = audiotrack->sub_packet_size;
2388                 int cfs = audiotrack->coded_framesize;
2389                 int h = audiotrack->sub_packet_h;
2390                 int y = audiotrack->sub_packet_cnt;
2391                 int w = audiotrack->frame_size;
2392                 int x;
2393
2394                 if (!audiotrack->pkt_cnt) {
2395                     if (st->codec->codec_id == CODEC_ID_RA_288)
2396                         for (x=0; x<h/2; x++)
2397                             memcpy(audiotrack->buf+x*2*w+y*cfs,
2398                                    data+x*cfs, cfs);
2399                     else
2400                         for (x=0; x<w/sps; x++)
2401                             memcpy(audiotrack->buf+sps*(h*x+((h+1)/2)*(y&1)+(y>>1)), data+x*sps, sps);
2402
2403                     if (++audiotrack->sub_packet_cnt >= h) {
2404                         audiotrack->sub_packet_cnt = 0;
2405                         audiotrack->pkt_cnt = h*w / a;
2406                     }
2407                 }
2408                 while (audiotrack->pkt_cnt) {
2409                     pkt = av_mallocz(sizeof(AVPacket));
2410                     av_new_packet(pkt, a);
2411                     memcpy(pkt->data, audiotrack->buf
2412                            + a * (h*w / a - audiotrack->pkt_cnt--), a);
2413                     pkt->pos = pos;
2414                     pkt->stream_index = matroska->tracks[track]->stream_index;
2415                     matroska_queue_packet(matroska, pkt);
2416                 }
2417             } else {
2418                 int offset = 0;
2419
2420                 if (st->codec->codec_id == CODEC_ID_TEXT
2421                     && ((MatroskaSubtitleTrack *)(matroska->tracks[track]))->ass) {
2422                     int i;
2423                     for (i=0; i<8 && data[offset]; offset++)
2424                         if (data[offset] == ',')
2425                             i++;
2426                 }
2427
2428                 pkt = av_mallocz(sizeof(AVPacket));
2429                 /* XXX: prevent data copy... */
2430                 if (av_new_packet(pkt, lace_size[n]-offset) < 0) {
2431                     res = AVERROR(ENOMEM);
2432                     n = laces-1;
2433                     break;
2434                 }
2435                 memcpy (pkt->data, data+offset, lace_size[n]-offset);
2436
2437                 if (n == 0)
2438                     pkt->flags = is_keyframe;
2439                 pkt->stream_index = matroska->tracks[track]->stream_index;
2440
2441                 pkt->pts = timecode;
2442                 pkt->pos = pos;
2443                 pkt->duration = duration;
2444
2445                 matroska_queue_packet(matroska, pkt);
2446             }
2447
2448             if (timecode != AV_NOPTS_VALUE)
2449                 timecode = duration ? timecode + duration : AV_NOPTS_VALUE;
2450             data += lace_size[n];
2451         }
2452     }
2453
2454     av_free(lace_size);
2455     av_free(origdata);
2456     return res;
2457 }
2458
2459 static int
2460 matroska_parse_blockgroup (MatroskaDemuxContext *matroska,
2461                            uint64_t              cluster_time)
2462 {
2463     int res = 0;
2464     uint32_t id;
2465     int is_bframe = 0;
2466     int is_keyframe = PKT_FLAG_KEY, last_num_packets = matroska->num_packets;
2467     uint64_t duration = AV_NOPTS_VALUE;
2468     uint8_t *data;
2469     int size = 0;
2470     int64_t pos = 0;
2471
2472     av_log(matroska->ctx, AV_LOG_DEBUG, "parsing blockgroup...\n");
2473
2474     while (res == 0) {
2475         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
2476             res = AVERROR(EIO);
2477             break;
2478         } else if (matroska->level_up) {
2479             matroska->level_up--;
2480             break;
2481         }
2482
2483         switch (id) {
2484             /* one block inside the group. Note, block parsing is one
2485              * of the harder things, so this code is a bit complicated.
2486              * See http://www.matroska.org/ for documentation. */
2487             case MATROSKA_ID_BLOCK: {
2488                 pos = url_ftell(matroska->ctx->pb);
2489                 res = ebml_read_binary(matroska, &id, &data, &size);
2490                 break;
2491             }
2492
2493             case MATROSKA_ID_BLOCKDURATION: {
2494                 if ((res = ebml_read_uint(matroska, &id, &duration)) < 0)
2495                     break;
2496                 break;
2497             }
2498
2499             case MATROSKA_ID_BLOCKREFERENCE: {
2500                 int64_t num;
2501                 /* We've found a reference, so not even the first frame in
2502                  * the lace is a key frame. */
2503                 is_keyframe = 0;
2504                 if (last_num_packets != matroska->num_packets)
2505                     matroska->packets[last_num_packets]->flags = 0;
2506                 if ((res = ebml_read_sint(matroska, &id, &num)) < 0)
2507                     break;
2508                 if (num > 0)
2509                     is_bframe = 1;
2510                 break;
2511             }
2512
2513             default:
2514                 av_log(matroska->ctx, AV_LOG_INFO,
2515                        "Unknown entry 0x%x in blockgroup data\n", id);
2516                 /* fall-through */
2517
2518             case EBML_ID_VOID:
2519                 res = ebml_read_skip(matroska);
2520                 break;
2521         }
2522
2523         if (matroska->level_up) {
2524             matroska->level_up--;
2525             break;
2526         }
2527     }
2528
2529     if (res)
2530         return res;
2531
2532     if (size > 0)
2533         res = matroska_parse_block(matroska, data, size, pos, cluster_time,
2534                                    duration, is_keyframe, is_bframe);
2535
2536     return res;
2537 }
2538
2539 static int
2540 matroska_parse_cluster (MatroskaDemuxContext *matroska)
2541 {
2542     int res = 0;
2543     uint32_t id;
2544     uint64_t cluster_time = 0;
2545     uint8_t *data;
2546     int64_t pos;
2547     int size;
2548
2549     av_log(matroska->ctx, AV_LOG_DEBUG,
2550            "parsing cluster at %"PRId64"\n", url_ftell(matroska->ctx->pb));
2551
2552     while (res == 0) {
2553         if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
2554             res = AVERROR(EIO);
2555             break;
2556         } else if (matroska->level_up) {
2557             matroska->level_up--;
2558             break;
2559         }
2560
2561         switch (id) {
2562             /* cluster timecode */
2563             case MATROSKA_ID_CLUSTERTIMECODE: {
2564                 uint64_t num;
2565                 if ((res = ebml_read_uint(matroska, &id, &num)) < 0)
2566                     break;
2567                 cluster_time = num;
2568                 break;
2569             }
2570
2571                 /* a group of blocks inside a cluster */
2572             case MATROSKA_ID_BLOCKGROUP:
2573                 if ((res = ebml_read_master(matroska, &id)) < 0)
2574                     break;
2575                 res = matroska_parse_blockgroup(matroska, cluster_time);
2576                 break;
2577
2578             case MATROSKA_ID_SIMPLEBLOCK:
2579                 pos = url_ftell(matroska->ctx->pb);
2580                 res = ebml_read_binary(matroska, &id, &data, &size);
2581                 if (res == 0)
2582                     res = matroska_parse_block(matroska, data, size, pos,
2583                                                cluster_time, AV_NOPTS_VALUE,
2584                                                -1, 0);
2585                 break;
2586
2587             default:
2588                 av_log(matroska->ctx, AV_LOG_INFO,
2589                        "Unknown entry 0x%x in cluster data\n", id);
2590                 /* fall-through */
2591
2592             case EBML_ID_VOID:
2593                 res = ebml_read_skip(matroska);
2594                 break;
2595         }
2596
2597         if (matroska->level_up) {
2598             matroska->level_up--;
2599             break;
2600         }
2601     }
2602
2603     return res;
2604 }
2605
2606 static int
2607 matroska_read_packet (AVFormatContext *s,
2608                       AVPacket        *pkt)
2609 {
2610     MatroskaDemuxContext *matroska = s->priv_data;
2611     int res;
2612     uint32_t id;
2613
2614     /* Read stream until we have a packet queued. */
2615     while (matroska_deliver_packet(matroska, pkt)) {
2616
2617         /* Have we already reached the end? */
2618         if (matroska->done)
2619             return AVERROR(EIO);
2620
2621         res = 0;
2622         while (res == 0) {
2623             if (!(id = ebml_peek_id(matroska, &matroska->level_up))) {
2624                 return AVERROR(EIO);
2625             } else if (matroska->level_up) {
2626                 matroska->level_up--;
2627                 break;
2628             }
2629
2630             switch (id) {
2631                 case MATROSKA_ID_CLUSTER:
2632                     if ((res = ebml_read_master(matroska, &id)) < 0)
2633                         break;
2634                     if ((res = matroska_parse_cluster(matroska)) == 0)
2635                         res = 1; /* Parsed one cluster, let's get out. */
2636                     break;
2637
2638                 default:
2639                 case EBML_ID_VOID:
2640                     res = ebml_read_skip(matroska);
2641                     break;
2642             }
2643
2644             if (matroska->level_up) {
2645                 matroska->level_up--;
2646                 break;
2647             }
2648         }
2649
2650         if (res == -1)
2651             matroska->done = 1;
2652     }
2653
2654     return 0;
2655 }
2656
2657 static int
2658 matroska_read_seek (AVFormatContext *s, int stream_index, int64_t timestamp,
2659                     int flags)
2660 {
2661     MatroskaDemuxContext *matroska = s->priv_data;
2662     AVStream *st = s->streams[stream_index];
2663     int index;
2664
2665     /* find index entry */
2666     index = av_index_search_timestamp(st, timestamp, flags);
2667     if (index < 0)
2668         return 0;
2669
2670     /* do the seek */
2671     url_fseek(s->pb, st->index_entries[index].pos, SEEK_SET);
2672     matroska->skip_to_keyframe = !(flags & AVSEEK_FLAG_ANY);
2673     matroska->skip_to_stream = st;
2674     matroska->num_packets = 0;
2675     matroska->peek_id = 0;
2676     return 0;
2677 }
2678
2679 static int
2680 matroska_read_close (AVFormatContext *s)
2681 {
2682     MatroskaDemuxContext *matroska = s->priv_data;
2683     int n = 0;
2684
2685     av_free(matroska->writing_app);
2686     av_free(matroska->muxing_app);
2687     av_free(matroska->index);
2688
2689     if (matroska->packets != NULL) {
2690         for (n = 0; n < matroska->num_packets; n++) {
2691             av_free_packet(matroska->packets[n]);
2692             av_free(matroska->packets[n]);
2693         }
2694         av_free(matroska->packets);
2695     }
2696
2697     for (n = 0; n < matroska->num_tracks; n++) {
2698         MatroskaTrack *track = matroska->tracks[n];
2699         av_free(track->codec_id);
2700         av_free(track->codec_name);
2701         av_free(track->codec_priv);
2702         av_free(track->name);
2703
2704         if (track->type == MATROSKA_TRACK_TYPE_AUDIO) {
2705             MatroskaAudioTrack *audiotrack = (MatroskaAudioTrack *)track;
2706             av_free(audiotrack->buf);
2707         }
2708
2709         av_free(track);
2710     }
2711
2712     return 0;
2713 }
2714
2715 AVInputFormat matroska_demuxer = {
2716     "matroska",
2717     "Matroska file format",
2718     sizeof(MatroskaDemuxContext),
2719     matroska_probe,
2720     matroska_read_header,
2721     matroska_read_packet,
2722     matroska_read_close,
2723     matroska_read_seek,
2724 };