]> git.sesse.net Git - ffmpeg/blob - libavformat/oggenc.c
Merge commit 'bd255f9feb4deea4c990e582f0ba3b90d7b64b4c'
[ffmpeg] / libavformat / oggenc.c
1 /*
2  * Ogg muxer
3  * Copyright (c) 2007 Baptiste Coudurier <baptiste dot coudurier at free dot fr>
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 #include "libavutil/crc.h"
23 #include "libavutil/opt.h"
24 #include "libavutil/mathematics.h"
25 #include "libavutil/opt.h"
26 #include "libavutil/random_seed.h"
27 #include "libavcodec/xiph.h"
28 #include "libavcodec/bytestream.h"
29 #include "libavcodec/flac.h"
30 #include "avformat.h"
31 #include "avio_internal.h"
32 #include "internal.h"
33 #include "vorbiscomment.h"
34
35 #define MAX_PAGE_SIZE 65025
36
37 typedef struct {
38     int64_t granule;
39     int stream_index;
40     uint8_t flags;
41     uint8_t segments_count;
42     uint8_t segments[255];
43     uint8_t data[MAX_PAGE_SIZE];
44     uint16_t size;
45 } OGGPage;
46
47 typedef struct {
48     unsigned page_counter;
49     uint8_t *header[3];
50     int header_len[3];
51     /** for theora granule */
52     int kfgshift;
53     int64_t last_kf_pts;
54     int vrev;
55     int eos;
56     unsigned page_count; ///< number of page buffered
57     OGGPage page; ///< current page
58     unsigned serial_num; ///< serial number
59     int64_t last_granule; ///< last packet granule
60 } OGGStreamContext;
61
62 typedef struct OGGPageList {
63     OGGPage page;
64     struct OGGPageList *next;
65 } OGGPageList;
66
67 typedef struct {
68     const AVClass *class;
69     OGGPageList *page_list;
70     int pref_size; ///< preferred page size (0 => fill all segments)
71 } OGGContext;
72
73 #define OFFSET(x) offsetof(OGGContext, x)
74 #define PARAM AV_OPT_FLAG_ENCODING_PARAM
75
76 static const AVOption options[] = {
77     { "oggpagesize", "Set preferred Ogg page size.",
78       offsetof(OGGContext, pref_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, MAX_PAGE_SIZE, AV_OPT_FLAG_ENCODING_PARAM},
79     { "pagesize", "preferred page size in bytes",
80         OFFSET(pref_size), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, MAX_PAGE_SIZE, PARAM },
81     { NULL },
82 };
83
84 static const AVClass ogg_muxer_class = {
85     .class_name = "Ogg muxer",
86     .item_name  = av_default_item_name,
87     .option     = options,
88     .version    = LIBAVUTIL_VERSION_INT,
89 };
90
91
92 static void ogg_update_checksum(AVFormatContext *s, AVIOContext *pb, int64_t crc_offset)
93 {
94     int64_t pos = avio_tell(pb);
95     uint32_t checksum = ffio_get_checksum(pb);
96     avio_seek(pb, crc_offset, SEEK_SET);
97     avio_wb32(pb, checksum);
98     avio_seek(pb, pos, SEEK_SET);
99 }
100
101 static int ogg_write_page(AVFormatContext *s, OGGPage *page, int extra_flags)
102 {
103     OGGStreamContext *oggstream = s->streams[page->stream_index]->priv_data;
104     AVIOContext *pb;
105     int64_t crc_offset;
106     int ret, size;
107     uint8_t *buf;
108
109     ret = avio_open_dyn_buf(&pb);
110     if (ret < 0)
111         return ret;
112     ffio_init_checksum(pb, ff_crc04C11DB7_update, 0);
113     ffio_wfourcc(pb, "OggS");
114     avio_w8(pb, 0);
115     avio_w8(pb, page->flags | extra_flags);
116     avio_wl64(pb, page->granule);
117     avio_wl32(pb, oggstream->serial_num);
118     avio_wl32(pb, oggstream->page_counter++);
119     crc_offset = avio_tell(pb);
120     avio_wl32(pb, 0); // crc
121     avio_w8(pb, page->segments_count);
122     avio_write(pb, page->segments, page->segments_count);
123     avio_write(pb, page->data, page->size);
124
125     ogg_update_checksum(s, pb, crc_offset);
126     avio_flush(pb);
127
128     size = avio_close_dyn_buf(pb, &buf);
129     if (size < 0)
130         return size;
131
132     avio_write(s->pb, buf, size);
133     avio_flush(s->pb);
134     av_free(buf);
135     oggstream->page_count--;
136     return 0;
137 }
138
139 static int ogg_key_granule(OGGStreamContext *oggstream, int64_t granule)
140 {
141     return oggstream->kfgshift && !(granule & ((1<<oggstream->kfgshift)-1));
142 }
143
144 static int64_t ogg_granule_to_timestamp(OGGStreamContext *oggstream, int64_t granule)
145 {
146     if (oggstream->kfgshift)
147         return (granule>>oggstream->kfgshift) +
148             (granule & ((1<<oggstream->kfgshift)-1));
149     else
150         return granule;
151 }
152
153 static int ogg_compare_granule(AVFormatContext *s, OGGPage *next, OGGPage *page)
154 {
155     AVStream *st2 = s->streams[next->stream_index];
156     AVStream *st  = s->streams[page->stream_index];
157     int64_t next_granule, cur_granule;
158
159     if (next->granule == -1 || page->granule == -1)
160         return 0;
161
162     next_granule = av_rescale_q(ogg_granule_to_timestamp(st2->priv_data, next->granule),
163                                 st2->time_base, AV_TIME_BASE_Q);
164     cur_granule  = av_rescale_q(ogg_granule_to_timestamp(st->priv_data, page->granule),
165                                 st ->time_base, AV_TIME_BASE_Q);
166     return next_granule > cur_granule;
167 }
168
169 static int ogg_reset_cur_page(OGGStreamContext *oggstream)
170 {
171     oggstream->page.granule = -1;
172     oggstream->page.flags = 0;
173     oggstream->page.segments_count = 0;
174     oggstream->page.size = 0;
175     return 0;
176 }
177
178 static int ogg_buffer_page(AVFormatContext *s, OGGStreamContext *oggstream)
179 {
180     OGGContext *ogg = s->priv_data;
181     OGGPageList **p = &ogg->page_list;
182     OGGPageList *l = av_mallocz(sizeof(*l));
183
184     if (!l)
185         return AVERROR(ENOMEM);
186     l->page = oggstream->page;
187
188     oggstream->page_count++;
189     ogg_reset_cur_page(oggstream);
190
191     while (*p) {
192         if (ogg_compare_granule(s, &(*p)->page, &l->page))
193             break;
194         p = &(*p)->next;
195     }
196     l->next = *p;
197     *p = l;
198
199     return 0;
200 }
201
202 static int ogg_buffer_data(AVFormatContext *s, AVStream *st,
203                            uint8_t *data, unsigned size, int64_t granule,
204                            int header)
205 {
206     OGGStreamContext *oggstream = st->priv_data;
207     OGGContext *ogg = s->priv_data;
208     int total_segments = size / 255 + 1;
209     uint8_t *p = data;
210     int i, segments, len, flush = 0;
211
212     // Handles VFR by flushing page because this frame needs to have a timestamp
213     // For theora, keyframes also need to have a timestamp to correctly mark
214     // them as such, otherwise seeking will not work correctly at the very
215     // least with old libogg versions.
216     // Do not try to flush header packets though, that will create broken files.
217     if (st->codec->codec_id == AV_CODEC_ID_THEORA && !header &&
218         (ogg_granule_to_timestamp(oggstream, granule) >
219          ogg_granule_to_timestamp(oggstream, oggstream->last_granule) + 1 ||
220          ogg_key_granule(oggstream, granule))) {
221         if (oggstream->page.granule != -1)
222             ogg_buffer_page(s, oggstream);
223         flush = 1;
224     }
225
226     for (i = 0; i < total_segments; ) {
227         OGGPage *page = &oggstream->page;
228
229         segments = FFMIN(total_segments - i, 255 - page->segments_count);
230
231         if (i && !page->segments_count)
232             page->flags |= 1; // continued packet
233
234         memset(page->segments+page->segments_count, 255, segments - 1);
235         page->segments_count += segments - 1;
236
237         len = FFMIN(size, segments*255);
238         page->segments[page->segments_count++] = len - (segments-1)*255;
239         memcpy(page->data+page->size, p, len);
240         p += len;
241         size -= len;
242         i += segments;
243         page->size += len;
244
245         if (i == total_segments)
246             page->granule = granule;
247
248         if (!header && (page->segments_count == 255 ||
249             (ogg->pref_size > 0 && page->size >= ogg->pref_size))) {
250            ogg_buffer_page(s, oggstream);
251         }
252     }
253
254     if (flush && oggstream->page.granule != -1)
255         ogg_buffer_page(s, oggstream);
256
257     return 0;
258 }
259
260 static uint8_t *ogg_write_vorbiscomment(int offset, int bitexact,
261                                         int *header_len, AVDictionary **m, int framing_bit)
262 {
263     const char *vendor = bitexact ? "ffmpeg" : LIBAVFORMAT_IDENT;
264     int size;
265     uint8_t *p, *p0;
266     unsigned int count;
267
268     ff_metadata_conv(m, ff_vorbiscomment_metadata_conv, NULL);
269
270     size = offset + ff_vorbiscomment_length(*m, vendor, &count) + framing_bit;
271     p = av_mallocz(size);
272     if (!p)
273         return NULL;
274     p0 = p;
275
276     p += offset;
277     ff_vorbiscomment_write(&p, m, vendor, count);
278     if (framing_bit)
279         bytestream_put_byte(&p, 1);
280
281     *header_len = size;
282     return p0;
283 }
284
285 static int ogg_build_flac_headers(AVCodecContext *avctx,
286                                   OGGStreamContext *oggstream, int bitexact,
287                                   AVDictionary **m)
288 {
289     enum FLACExtradataFormat format;
290     uint8_t *streaminfo;
291     uint8_t *p;
292
293     if (!avpriv_flac_is_extradata_valid(avctx, &format, &streaminfo))
294         return -1;
295
296     // first packet: STREAMINFO
297     oggstream->header_len[0] = 51;
298     oggstream->header[0] = av_mallocz(51); // per ogg flac specs
299     p = oggstream->header[0];
300     if (!p)
301         return AVERROR(ENOMEM);
302     bytestream_put_byte(&p, 0x7F);
303     bytestream_put_buffer(&p, "FLAC", 4);
304     bytestream_put_byte(&p, 1); // major version
305     bytestream_put_byte(&p, 0); // minor version
306     bytestream_put_be16(&p, 1); // headers packets without this one
307     bytestream_put_buffer(&p, "fLaC", 4);
308     bytestream_put_byte(&p, 0x00); // streaminfo
309     bytestream_put_be24(&p, 34);
310     bytestream_put_buffer(&p, streaminfo, FLAC_STREAMINFO_SIZE);
311
312     // second packet: VorbisComment
313     p = ogg_write_vorbiscomment(4, bitexact, &oggstream->header_len[1], m, 0);
314     if (!p)
315         return AVERROR(ENOMEM);
316     oggstream->header[1] = p;
317     bytestream_put_byte(&p, 0x84); // last metadata block and vorbis comment
318     bytestream_put_be24(&p, oggstream->header_len[1] - 4);
319
320     return 0;
321 }
322
323 #define SPEEX_HEADER_SIZE 80
324
325 static int ogg_build_speex_headers(AVCodecContext *avctx,
326                                    OGGStreamContext *oggstream, int bitexact,
327                                    AVDictionary **m)
328 {
329     uint8_t *p;
330
331     if (avctx->extradata_size < SPEEX_HEADER_SIZE)
332         return -1;
333
334     // first packet: Speex header
335     p = av_mallocz(SPEEX_HEADER_SIZE);
336     if (!p)
337         return AVERROR(ENOMEM);
338     oggstream->header[0] = p;
339     oggstream->header_len[0] = SPEEX_HEADER_SIZE;
340     bytestream_put_buffer(&p, avctx->extradata, SPEEX_HEADER_SIZE);
341     AV_WL32(&oggstream->header[0][68], 0);  // set extra_headers to 0
342
343     // second packet: VorbisComment
344     p = ogg_write_vorbiscomment(0, bitexact, &oggstream->header_len[1], m, 0);
345     if (!p)
346         return AVERROR(ENOMEM);
347     oggstream->header[1] = p;
348
349     return 0;
350 }
351
352 #define OPUS_HEADER_SIZE 19
353
354 static int ogg_build_opus_headers(AVCodecContext *avctx,
355                                   OGGStreamContext *oggstream, int bitexact,
356                                   AVDictionary **m)
357 {
358     uint8_t *p;
359
360     if (avctx->extradata_size < OPUS_HEADER_SIZE)
361         return -1;
362
363     /* first packet: Opus header */
364     p = av_mallocz(avctx->extradata_size);
365     if (!p)
366         return AVERROR(ENOMEM);
367     oggstream->header[0] = p;
368     oggstream->header_len[0] = avctx->extradata_size;
369     bytestream_put_buffer(&p, avctx->extradata, avctx->extradata_size);
370
371     /* second packet: VorbisComment */
372     p = ogg_write_vorbiscomment(8, bitexact, &oggstream->header_len[1], m, 0);
373     if (!p)
374         return AVERROR(ENOMEM);
375     oggstream->header[1] = p;
376     bytestream_put_buffer(&p, "OpusTags", 8);
377
378     return 0;
379 }
380
381 static int ogg_write_header(AVFormatContext *s)
382 {
383     OGGStreamContext *oggstream;
384     int i, j;
385
386     for (i = 0; i < s->nb_streams; i++) {
387         AVStream *st = s->streams[i];
388         unsigned serial_num = i;
389
390         if (st->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
391             if (st->codec->codec_id == AV_CODEC_ID_OPUS)
392                 /* Opus requires a fixed 48kHz clock */
393                 avpriv_set_pts_info(st, 64, 1, 48000);
394             else
395                 avpriv_set_pts_info(st, 64, 1, st->codec->sample_rate);
396         } else if (st->codec->codec_type == AVMEDIA_TYPE_VIDEO)
397             avpriv_set_pts_info(st, 64, st->codec->time_base.num, st->codec->time_base.den);
398         if (st->codec->codec_id != AV_CODEC_ID_VORBIS &&
399             st->codec->codec_id != AV_CODEC_ID_THEORA &&
400             st->codec->codec_id != AV_CODEC_ID_SPEEX  &&
401             st->codec->codec_id != AV_CODEC_ID_FLAC   &&
402             st->codec->codec_id != AV_CODEC_ID_OPUS) {
403             av_log(s, AV_LOG_ERROR, "Unsupported codec id in stream %d\n", i);
404             return -1;
405         }
406
407         if (!st->codec->extradata || !st->codec->extradata_size) {
408             av_log(s, AV_LOG_ERROR, "No extradata present\n");
409             return -1;
410         }
411         oggstream = av_mallocz(sizeof(*oggstream));
412         oggstream->page.stream_index = i;
413
414         if (!(st->codec->flags & CODEC_FLAG_BITEXACT))
415             do {
416                 serial_num = av_get_random_seed();
417                 for (j = 0; j < i; j++) {
418                     OGGStreamContext *sc = s->streams[j]->priv_data;
419                     if (serial_num == sc->serial_num)
420                         break;
421                 }
422             } while (j < i);
423         oggstream->serial_num = serial_num;
424
425         st->priv_data = oggstream;
426         if (st->codec->codec_id == AV_CODEC_ID_FLAC) {
427             int err = ogg_build_flac_headers(st->codec, oggstream,
428                                              st->codec->flags & CODEC_FLAG_BITEXACT,
429                                              &s->metadata);
430             if (err) {
431                 av_log(s, AV_LOG_ERROR, "Error writing FLAC headers\n");
432                 av_freep(&st->priv_data);
433                 return err;
434             }
435         } else if (st->codec->codec_id == AV_CODEC_ID_SPEEX) {
436             int err = ogg_build_speex_headers(st->codec, oggstream,
437                                               st->codec->flags & CODEC_FLAG_BITEXACT,
438                                               &s->metadata);
439             if (err) {
440                 av_log(s, AV_LOG_ERROR, "Error writing Speex headers\n");
441                 av_freep(&st->priv_data);
442                 return err;
443             }
444         } else if (st->codec->codec_id == AV_CODEC_ID_OPUS) {
445             int err = ogg_build_opus_headers(st->codec, oggstream,
446                                              st->codec->flags & CODEC_FLAG_BITEXACT,
447                                              &s->metadata);
448             if (err) {
449                 av_log(s, AV_LOG_ERROR, "Error writing Opus headers\n");
450                 av_freep(&st->priv_data);
451                 return err;
452             }
453         } else {
454             uint8_t *p;
455             const char *cstr = st->codec->codec_id == AV_CODEC_ID_VORBIS ? "vorbis" : "theora";
456             int header_type = st->codec->codec_id == AV_CODEC_ID_VORBIS ? 3 : 0x81;
457             int framing_bit = st->codec->codec_id == AV_CODEC_ID_VORBIS ? 1 : 0;
458
459             if (avpriv_split_xiph_headers(st->codec->extradata, st->codec->extradata_size,
460                                       st->codec->codec_id == AV_CODEC_ID_VORBIS ? 30 : 42,
461                                       oggstream->header, oggstream->header_len) < 0) {
462                 av_log(s, AV_LOG_ERROR, "Extradata corrupted\n");
463                 av_freep(&st->priv_data);
464                 return -1;
465             }
466
467             p = ogg_write_vorbiscomment(7, st->codec->flags & CODEC_FLAG_BITEXACT,
468                                         &oggstream->header_len[1], &s->metadata,
469                                         framing_bit);
470             oggstream->header[1] = p;
471             if (!p)
472                 return AVERROR(ENOMEM);
473
474             bytestream_put_byte(&p, header_type);
475             bytestream_put_buffer(&p, cstr, 6);
476
477             if (st->codec->codec_id == AV_CODEC_ID_THEORA) {
478                 /** KFGSHIFT is the width of the less significant section of the granule position
479                     The less significant section is the frame count since the last keyframe */
480                 oggstream->kfgshift = ((oggstream->header[0][40]&3)<<3)|(oggstream->header[0][41]>>5);
481                 oggstream->vrev = oggstream->header[0][9];
482                 av_log(s, AV_LOG_DEBUG, "theora kfgshift %d, vrev %d\n",
483                        oggstream->kfgshift, oggstream->vrev);
484             }
485         }
486     }
487
488     for (j = 0; j < s->nb_streams; j++) {
489         OGGStreamContext *oggstream = s->streams[j]->priv_data;
490         ogg_buffer_data(s, s->streams[j], oggstream->header[0],
491                         oggstream->header_len[0], 0, 1);
492         oggstream->page.flags |= 2; // bos
493         ogg_buffer_page(s, oggstream);
494     }
495     for (j = 0; j < s->nb_streams; j++) {
496         AVStream *st = s->streams[j];
497         OGGStreamContext *oggstream = st->priv_data;
498         for (i = 1; i < 3; i++) {
499             if (oggstream->header_len[i])
500                 ogg_buffer_data(s, st, oggstream->header[i],
501                                 oggstream->header_len[i], 0, 1);
502         }
503         ogg_buffer_page(s, oggstream);
504     }
505     return 0;
506 }
507
508 static void ogg_write_pages(AVFormatContext *s, int flush)
509 {
510     OGGContext *ogg = s->priv_data;
511     OGGPageList *next, *p;
512
513     if (!ogg->page_list)
514         return;
515
516     for (p = ogg->page_list; p; ) {
517         OGGStreamContext *oggstream =
518             s->streams[p->page.stream_index]->priv_data;
519         if (oggstream->page_count < 2 && !flush)
520             break;
521         ogg_write_page(s, &p->page,
522                        flush && oggstream->page_count == 1 ? 4 : 0); // eos
523         next = p->next;
524         av_freep(&p);
525         p = next;
526     }
527     ogg->page_list = p;
528 }
529
530 static int ogg_write_packet(AVFormatContext *s, AVPacket *pkt)
531 {
532     AVStream *st = s->streams[pkt->stream_index];
533     OGGStreamContext *oggstream = st->priv_data;
534     int ret;
535     int64_t granule;
536
537     if (st->codec->codec_id == AV_CODEC_ID_THEORA) {
538         int64_t pts = oggstream->vrev < 1 ? pkt->pts : pkt->pts + pkt->duration;
539         int pframe_count;
540         if (pkt->flags & AV_PKT_FLAG_KEY)
541             oggstream->last_kf_pts = pts;
542         pframe_count = pts - oggstream->last_kf_pts;
543         // prevent frame count from overflow if key frame flag is not set
544         if (pframe_count >= (1<<oggstream->kfgshift)) {
545             oggstream->last_kf_pts += pframe_count;
546             pframe_count = 0;
547         }
548         granule = (oggstream->last_kf_pts<<oggstream->kfgshift) | pframe_count;
549     } else if (st->codec->codec_id == AV_CODEC_ID_OPUS)
550         granule = pkt->pts + pkt->duration + av_rescale_q(st->codec->delay, (AVRational){ 1, st->codec->sample_rate }, st->time_base);
551     else
552         granule = pkt->pts + pkt->duration;
553
554     ret = ogg_buffer_data(s, st, pkt->data, pkt->size, granule, 0);
555     if (ret < 0)
556         return ret;
557
558     ogg_write_pages(s, 0);
559
560     oggstream->last_granule = granule;
561
562     return 0;
563 }
564
565 static int ogg_write_trailer(AVFormatContext *s)
566 {
567     int i;
568
569     /* flush current page */
570     for (i = 0; i < s->nb_streams; i++)
571         ogg_buffer_page(s, s->streams[i]->priv_data);
572
573     ogg_write_pages(s, 1);
574
575     for (i = 0; i < s->nb_streams; i++) {
576         AVStream *st = s->streams[i];
577         OGGStreamContext *oggstream = st->priv_data;
578         if (st->codec->codec_id == AV_CODEC_ID_FLAC ||
579             st->codec->codec_id == AV_CODEC_ID_SPEEX ||
580             st->codec->codec_id == AV_CODEC_ID_OPUS) {
581             av_freep(&oggstream->header[0]);
582         }
583         av_freep(&oggstream->header[1]);
584         av_freep(&st->priv_data);
585     }
586     return 0;
587 }
588
589 AVOutputFormat ff_ogg_muxer = {
590     .name              = "ogg",
591     .long_name         = NULL_IF_CONFIG_SMALL("Ogg"),
592     .mime_type         = "application/ogg",
593     .extensions        = "ogg,ogv,spx,opus",
594     .priv_data_size    = sizeof(OGGContext),
595     .audio_codec       = AV_CODEC_ID_FLAC,
596     .video_codec       = AV_CODEC_ID_THEORA,
597     .write_header      = ogg_write_header,
598     .write_packet      = ogg_write_packet,
599     .write_trailer     = ogg_write_trailer,
600     .priv_class        = &ogg_muxer_class,
601 };