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 {
 200     OGGStreamContext *oggstream = st->priv_data;
 201     OGGContext *ogg = s->priv_data;
 202     int total_segments = size / 255 + 1;
 203     uint8_t *p = data;
 204     int i, segments, len, flush = 0;
 205
 206     // Handles VFR by flushing page because this frame needs to have a timestamp
 207     // For theora, keyframes also need to have a timestamp to correctly mark
 208     // them as such, otherwise seeking will not work correctly at the very
 209     // least with old libogg versions.
 210     // Do not try to flush empty packets though, that will create broken files.
 211     if (st->codec->codec_id == CODEC_ID_THEORA &&
 212         oggstream->page.size &&
 213         (ogg_granule_to_timestamp(oggstream, granule) >
 214          ogg_granule_to_timestamp(oggstream, oggstream->last_granule) + 1 ||
 215          ogg_key_granule(oggstream, granule))) {
 216         if (oggstream->page.granule != -1)
 217             ogg_buffer_page(s, oggstream);
 218         flush = 1;
 219     }
 220
 221     for (i = 0; i < total_segments; ) {
 222         OGGPage *page = &oggstream->page;
 223
 224         segments = FFMIN(total_segments - i, 255 - page->segments_count);
 225
 226         if (i && !page->segments_count)
 227             page->flags |= 1; // continued packet
 228
 229         memset(page->segments+page->segments_count, 255, segments - 1);
 230         page->segments_count += segments - 1;
 231
 232         len = FFMIN(size, segments*255);
 233         page->segments[page->segments_count++] = len - (segments-1)*255;
 234         memcpy(page->data+page->size, p, len);
 235         p += len;
 236         size -= len;
 237         i += segments;
 238         page->size += len;
 239
 240         if (i == total_segments)
 241             page->granule = granule;
 242
 243         if(page->segments_count == 255 ||
 244            (ogg->pref_size > 0 && page->size >= ogg->pref_size)) {
 245            ogg_buffer_page(s, oggstream);
 246         }
 247     }
 248
 249     if (flush && oggstream->page.granule != -1)
 250         ogg_buffer_page(s, oggstream);
 251
 252     return 0;
 253 }
 254
 255 static uint8_t *ogg_write_vorbiscomment(int offset, int bitexact,
 256                                         int *header_len, AVDictionary **m, int framing_bit)
 257 {
 258     const char *vendor = bitexact ? "ffmpeg" : LIBAVFORMAT_IDENT;
 259     int size;
 260     uint8_t *p, *p0;
 261     unsigned int count;
 262
 263     ff_metadata_conv(m, ff_vorbiscomment_metadata_conv, NULL);
 264
 265     size = offset + ff_vorbiscomment_length(*m, vendor, &count) + framing_bit;
 266     p = av_mallocz(size);
 267     if (!p)
 268         return NULL;
 269     p0 = p;
 270
 271     p += offset;
 272     ff_vorbiscomment_write(&p, m, vendor, count);
 273     if (framing_bit)
 274         bytestream_put_byte(&p, 1);
 275
 276     *header_len = size;
 277     return p0;
 278 }
 279
 280 static int ogg_build_flac_headers(AVCodecContext *avctx,
 281                                   OGGStreamContext *oggstream, int bitexact,
 282                                   AVDictionary **m)
 283 {
 284     enum FLACExtradataFormat format;
 285     uint8_t *streaminfo;
 286     uint8_t *p;
 287
 288     if (!avpriv_flac_is_extradata_valid(avctx, &format, &streaminfo))
 289         return -1;
 290
 291     // first packet: STREAMINFO
 292     oggstream->header_len[0] = 51;
 293     oggstream->header[0] = av_mallocz(51); // per ogg flac specs
 294     p = oggstream->header[0];
 295     if (!p)
 296         return AVERROR(ENOMEM);
 297     bytestream_put_byte(&p, 0x7F);
 298     bytestream_put_buffer(&p, "FLAC", 4);
 299     bytestream_put_byte(&p, 1); // major version
 300     bytestream_put_byte(&p, 0); // minor version
 301     bytestream_put_be16(&p, 1); // headers packets without this one
 302     bytestream_put_buffer(&p, "fLaC", 4);
 303     bytestream_put_byte(&p, 0x00); // streaminfo
 304     bytestream_put_be24(&p, 34);
 305     bytestream_put_buffer(&p, streaminfo, FLAC_STREAMINFO_SIZE);
 306
 307     // second packet: VorbisComment
 308     p = ogg_write_vorbiscomment(4, bitexact, &oggstream->header_len[1], m, 0);
 309     if (!p)
 310         return AVERROR(ENOMEM);
 311     oggstream->header[1] = p;
 312     bytestream_put_byte(&p, 0x84); // last metadata block and vorbis comment
 313     bytestream_put_be24(&p, oggstream->header_len[1] - 4);
 314
 315     return 0;
 316 }
 317
 318 #define SPEEX_HEADER_SIZE 80
 319
 320 static int ogg_build_speex_headers(AVCodecContext *avctx,
 321                                    OGGStreamContext *oggstream, int bitexact,
 322                                    AVDictionary **m)
 323 {
 324     uint8_t *p;
 325
 326     if (avctx->extradata_size < SPEEX_HEADER_SIZE)
 327         return -1;
 328
 329     // first packet: Speex header
 330     p = av_mallocz(SPEEX_HEADER_SIZE);
 331     if (!p)
 332         return AVERROR(ENOMEM);
 333     oggstream->header[0] = p;
 334     oggstream->header_len[0] = SPEEX_HEADER_SIZE;
 335     bytestream_put_buffer(&p, avctx->extradata, SPEEX_HEADER_SIZE);
 336     AV_WL32(&oggstream->header[0][68], 0);  // set extra_headers to 0
 337
 338     // second packet: VorbisComment
 339     p = ogg_write_vorbiscomment(0, bitexact, &oggstream->header_len[1], m, 0);
 340     if (!p)
 341         return AVERROR(ENOMEM);
 342     oggstream->header[1] = p;
 343
 344     return 0;
 345 }
 346
 347 static int ogg_write_header(AVFormatContext *s)
 348 {
 349     OGGStreamContext *oggstream;
 350     int i, j;
 351
 352     for (i = 0; i < s->nb_streams; i++) {
 353         AVStream *st = s->streams[i];
 354         unsigned serial_num = i;
 355
 356         if (st->codec->codec_type == AVMEDIA_TYPE_AUDIO)
 357             avpriv_set_pts_info(st, 64, 1, st->codec->sample_rate);
 358         else if (st->codec->codec_type == AVMEDIA_TYPE_VIDEO)
 359             avpriv_set_pts_info(st, 64, st->codec->time_base.num, st->codec->time_base.den);
 360         if (st->codec->codec_id != CODEC_ID_VORBIS &&
 361             st->codec->codec_id != CODEC_ID_THEORA &&
 362             st->codec->codec_id != CODEC_ID_SPEEX  &&
 363             st->codec->codec_id != CODEC_ID_FLAC) {
 364             av_log(s, AV_LOG_ERROR, "Unsupported codec id in stream %d\n", i);
 365             return -1;
 366         }
 367
 368         if (!st->codec->extradata || !st->codec->extradata_size) {
 369             av_log(s, AV_LOG_ERROR, "No extradata present\n");
 370             return -1;
 371         }
 372         oggstream = av_mallocz(sizeof(*oggstream));
 373         oggstream->page.stream_index = i;
 374
 375         if (!(st->codec->flags & CODEC_FLAG_BITEXACT))
 376             do {
 377                 serial_num = av_get_random_seed();
 378                 for (j = 0; j < i; j++) {
 379                     OGGStreamContext *sc = s->streams[j]->priv_data;
 380                     if (serial_num == sc->serial_num)
 381                         break;
 382                 }
 383             } while (j < i);
 384         oggstream->serial_num = serial_num;
 385
 386         st->priv_data = oggstream;
 387         if (st->codec->codec_id == CODEC_ID_FLAC) {
 388             int err = ogg_build_flac_headers(st->codec, oggstream,
 389                                              st->codec->flags & CODEC_FLAG_BITEXACT,
 390                                              &s->metadata);
 391             if (err) {
 392                 av_log(s, AV_LOG_ERROR, "Error writing FLAC headers\n");
 393                 av_freep(&st->priv_data);
 394                 return err;
 395             }
 396         } else if (st->codec->codec_id == CODEC_ID_SPEEX) {
 397             int err = ogg_build_speex_headers(st->codec, oggstream,
 398                                               st->codec->flags & CODEC_FLAG_BITEXACT,
 399                                               &s->metadata);
 400             if (err) {
 401                 av_log(s, AV_LOG_ERROR, "Error writing Speex headers\n");
 402                 av_freep(&st->priv_data);
 403                 return err;
 404             }
 405         } else {
 406             uint8_t *p;
 407             const char *cstr = st->codec->codec_id == CODEC_ID_VORBIS ? "vorbis" : "theora";
 408             int header_type = st->codec->codec_id == CODEC_ID_VORBIS ? 3 : 0x81;
 409             int framing_bit = st->codec->codec_id == CODEC_ID_VORBIS ? 1 : 0;
 410
 411             if (avpriv_split_xiph_headers(st->codec->extradata, st->codec->extradata_size,
 412                                       st->codec->codec_id == CODEC_ID_VORBIS ? 30 : 42,
 413                                       oggstream->header, oggstream->header_len) < 0) {
 414                 av_log(s, AV_LOG_ERROR, "Extradata corrupted\n");
 415                 av_freep(&st->priv_data);
 416                 return -1;
 417             }
 418
 419             p = ogg_write_vorbiscomment(7, st->codec->flags & CODEC_FLAG_BITEXACT,
 420                                         &oggstream->header_len[1], &s->metadata,
 421                                         framing_bit);
 422             if (!p)
 423                 return AVERROR(ENOMEM);
 424
 425             oggstream->header[1] = p;
 426             bytestream_put_byte(&p, header_type);
 427             bytestream_put_buffer(&p, cstr, 6);
 428
 429             if (st->codec->codec_id == CODEC_ID_THEORA) {
 430                 /** KFGSHIFT is the width of the less significant section of the granule position
 431                     The less significant section is the frame count since the last keyframe */
 432                 oggstream->kfgshift = ((oggstream->header[0][40]&3)<<3)|(oggstream->header[0][41]>>5);
 433                 oggstream->vrev = oggstream->header[0][9];
 434                 av_log(s, AV_LOG_DEBUG, "theora kfgshift %d, vrev %d\n",
 435                        oggstream->kfgshift, oggstream->vrev);
 436             }
 437         }
 438     }
 439
 440     for (j = 0; j < s->nb_streams; j++) {
 441         OGGStreamContext *oggstream = s->streams[j]->priv_data;
 442         ogg_buffer_data(s, s->streams[j], oggstream->header[0],
 443                         oggstream->header_len[0], 0);
 444         oggstream->page.flags |= 2; // bos
 445         ogg_buffer_page(s, oggstream);
 446     }
 447     for (j = 0; j < s->nb_streams; j++) {
 448         AVStream *st = s->streams[j];
 449         OGGStreamContext *oggstream = st->priv_data;
 450         for (i = 1; i < 3; i++) {
 451             if (oggstream && oggstream->header_len[i])
 452                 ogg_buffer_data(s, st, oggstream->header[i],
 453                                 oggstream->header_len[i], 0);
 454         }
 455         ogg_buffer_page(s, oggstream);
 456     }
 457     return 0;
 458 }
 459
 460 static void ogg_write_pages(AVFormatContext *s, int flush)
 461 {
 462     OGGContext *ogg = s->priv_data;
 463     OGGPageList *next, *p;
 464
 465     if (!ogg->page_list)
 466         return;
 467
 468     for (p = ogg->page_list; p; ) {
 469         OGGStreamContext *oggstream =
 470             s->streams[p->page.stream_index]->priv_data;
 471         if (oggstream->page_count < 2 && !flush)
 472             break;
 473         ogg_write_page(s, &p->page,
 474                        flush && oggstream->page_count == 1 ? 4 : 0); // eos
 475         next = p->next;
 476         av_freep(&p);
 477         p = next;
 478     }
 479     ogg->page_list = p;
 480 }
 481
 482 static int ogg_write_packet(AVFormatContext *s, AVPacket *pkt)
 483 {
 484     AVStream *st = s->streams[pkt->stream_index];
 485     OGGStreamContext *oggstream = st->priv_data;
 486     int ret;
 487     int64_t granule;
 488
 489     if (st->codec->codec_id == CODEC_ID_THEORA) {
 490         int64_t pts = oggstream->vrev < 1 ? pkt->pts : pkt->pts + pkt->duration;
 491         int pframe_count;
 492         if (pkt->flags & AV_PKT_FLAG_KEY)
 493             oggstream->last_kf_pts = pts;
 494         pframe_count = pts - oggstream->last_kf_pts;
 495         // prevent frame count from overflow if key frame flag is not set
 496         if (pframe_count >= (1<<oggstream->kfgshift)) {
 497             oggstream->last_kf_pts += pframe_count;
 498             pframe_count = 0;
 499         }
 500         granule = (oggstream->last_kf_pts<<oggstream->kfgshift) | pframe_count;
 501     } else
 502         granule = pkt->pts + pkt->duration;
 503
 504     ret = ogg_buffer_data(s, st, pkt->data, pkt->size, granule);
 505     if (ret < 0)
 506         return ret;
 507
 508     ogg_write_pages(s, 0);
 509
 510     oggstream->last_granule = granule;
 511
 512     return 0;
 513 }
 514
 515 static int ogg_write_trailer(AVFormatContext *s)
 516 {
 517     int i;
 518
 519     /* flush current page */
 520     for (i = 0; i < s->nb_streams; i++)
 521         ogg_buffer_page(s, s->streams[i]->priv_data);
 522
 523     ogg_write_pages(s, 1);
 524
 525     for (i = 0; i < s->nb_streams; i++) {
 526         AVStream *st = s->streams[i];
 527         OGGStreamContext *oggstream = st->priv_data;
 528         if (st->codec->codec_id == CODEC_ID_FLAC ||
 529             st->codec->codec_id == CODEC_ID_SPEEX) {
 530             av_freep(&oggstream->header[0]);
 531             av_freep(&oggstream->header[1]);
 532         }
 533         else
 534             av_freep(&oggstream->header[1]);
 535         av_freep(&st->priv_data);
 536     }
 537     return 0;
 538 }
 539
 540 AVOutputFormat ff_ogg_muxer = {
 541     .name              = "ogg",
 542     .long_name         = NULL_IF_CONFIG_SMALL("Ogg"),
 543     .mime_type         = "application/ogg",
 544     .extensions        = "ogg,ogv,spx",
 545     .priv_data_size    = sizeof(OGGContext),
 546     .audio_codec       = CODEC_ID_FLAC,
 547     .video_codec       = CODEC_ID_THEORA,
 548     .write_header      = ogg_write_header,
 549     .write_packet      = ogg_write_packet,
 550     .write_trailer     = ogg_write_trailer,
 551     .priv_class = &ogg_muxer_class,
 552 };