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