Libav
rtpenc.c
Go to the documentation of this file.
1 /*
2  * RTP output format
3  * Copyright (c) 2002 Fabrice Bellard
4  *
5  * This file is part of Libav.
6  *
7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "avformat.h"
23 #include "mpegts.h"
24 #include "internal.h"
25 #include "libavutil/mathematics.h"
26 #include "libavutil/random_seed.h"
27 #include "libavutil/opt.h"
28 
29 #include "rtpenc.h"
30 
31 static const AVOption options[] = {
33  { "payload_type", "Specify RTP payload type", offsetof(RTPMuxContext, payload_type), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, 127, AV_OPT_FLAG_ENCODING_PARAM },
34  { "ssrc", "Stream identifier", offsetof(RTPMuxContext, ssrc), AV_OPT_TYPE_INT, { .i64 = 0 }, INT_MIN, INT_MAX, AV_OPT_FLAG_ENCODING_PARAM },
35  { "cname", "CNAME to include in RTCP SR packets", offsetof(RTPMuxContext, cname), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, AV_OPT_FLAG_ENCODING_PARAM },
36  { "seq", "Starting sequence number", offsetof(RTPMuxContext, seq), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 65535, AV_OPT_FLAG_ENCODING_PARAM },
37  { NULL },
38 };
39 
40 static const AVClass rtp_muxer_class = {
41  .class_name = "RTP muxer",
42  .item_name = av_default_item_name,
43  .option = options,
44  .version = LIBAVUTIL_VERSION_INT,
45 };
46 
47 #define RTCP_SR_SIZE 28
48 
49 static int is_supported(enum AVCodecID id)
50 {
51  switch(id) {
52  case AV_CODEC_ID_H263:
53  case AV_CODEC_ID_H263P:
54  case AV_CODEC_ID_H264:
57  case AV_CODEC_ID_MPEG4:
58  case AV_CODEC_ID_AAC:
59  case AV_CODEC_ID_MP2:
60  case AV_CODEC_ID_MP3:
63  case AV_CODEC_ID_PCM_S8:
68  case AV_CODEC_ID_PCM_U8:
70  case AV_CODEC_ID_AMR_NB:
71  case AV_CODEC_ID_AMR_WB:
72  case AV_CODEC_ID_VORBIS:
73  case AV_CODEC_ID_THEORA:
74  case AV_CODEC_ID_VP8:
77  case AV_CODEC_ID_ILBC:
78  case AV_CODEC_ID_MJPEG:
79  case AV_CODEC_ID_SPEEX:
80  case AV_CODEC_ID_OPUS:
81  return 1;
82  default:
83  return 0;
84  }
85 }
86 
88 {
89  RTPMuxContext *s = s1->priv_data;
90  int n;
91  AVStream *st;
92 
93  if (s1->nb_streams != 1) {
94  av_log(s1, AV_LOG_ERROR, "Only one stream supported in the RTP muxer\n");
95  return AVERROR(EINVAL);
96  }
97  st = s1->streams[0];
98  if (!is_supported(st->codec->codec_id)) {
99  av_log(s1, AV_LOG_ERROR, "Unsupported codec %x\n", st->codec->codec_id);
100 
101  return -1;
102  }
103 
104  if (s->payload_type < 0) {
105  /* Re-validate non-dynamic payload types */
106  if (st->id < RTP_PT_PRIVATE)
107  st->id = ff_rtp_get_payload_type(s1, st->codec, -1);
108 
109  s->payload_type = st->id;
110  } else {
111  /* private option takes priority */
112  st->id = s->payload_type;
113  }
114 
116  s->timestamp = s->base_timestamp;
117  s->cur_timestamp = 0;
118  if (!s->ssrc)
119  s->ssrc = av_get_random_seed();
120  s->first_packet = 1;
122  if (s1->start_time_realtime)
123  /* Round the NTP time to whole milliseconds. */
124  s->first_rtcp_ntp_time = (s1->start_time_realtime / 1000) * 1000 +
126  // Pick a random sequence start number, but in the lower end of the
127  // available range, so that any wraparound doesn't happen immediately.
128  // (Immediate wraparound would be an issue for SRTP.)
129  if (s->seq < 0)
130  s->seq = av_get_random_seed() & 0x0fff;
131  else
132  s->seq &= 0xffff; // Use the given parameter, wrapped to the right interval
133 
134  if (s1->packet_size) {
135  if (s1->pb->max_packet_size)
136  s1->packet_size = FFMIN(s1->packet_size,
137  s1->pb->max_packet_size);
138  } else
139  s1->packet_size = s1->pb->max_packet_size;
140  if (s1->packet_size <= 12) {
141  av_log(s1, AV_LOG_ERROR, "Max packet size %d too low\n", s1->packet_size);
142  return AVERROR(EIO);
143  }
144  s->buf = av_malloc(s1->packet_size);
145  if (s->buf == NULL) {
146  return AVERROR(ENOMEM);
147  }
148  s->max_payload_size = s1->packet_size - 12;
149 
150  s->max_frames_per_packet = 0;
151  if (s1->max_delay > 0) {
152  if (st->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
154  if (!frame_size)
155  frame_size = st->codec->frame_size;
156  if (frame_size == 0) {
157  av_log(s1, AV_LOG_ERROR, "Cannot respect max delay: frame size = 0\n");
158  } else {
162  (AVRational){ frame_size, st->codec->sample_rate },
163  AV_ROUND_DOWN);
164  }
165  }
166  if (st->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
167  /* FIXME: We should round down here... */
168  s->max_frames_per_packet = av_rescale_q(s1->max_delay, (AVRational){1, 1000000}, st->codec->time_base);
169  }
170  }
171 
172  avpriv_set_pts_info(st, 32, 1, 90000);
173  switch(st->codec->codec_id) {
174  case AV_CODEC_ID_MP2:
175  case AV_CODEC_ID_MP3:
176  s->buf_ptr = s->buf + 4;
177  break;
180  break;
181  case AV_CODEC_ID_MPEG2TS:
182  n = s->max_payload_size / TS_PACKET_SIZE;
183  if (n < 1)
184  n = 1;
185  s->max_payload_size = n * TS_PACKET_SIZE;
186  s->buf_ptr = s->buf;
187  break;
188  case AV_CODEC_ID_H264:
189  /* check for H.264 MP4 syntax */
190  if (st->codec->extradata_size > 4 && st->codec->extradata[0] == 1) {
191  s->nal_length_size = (st->codec->extradata[4] & 0x03) + 1;
192  }
193  break;
194  case AV_CODEC_ID_VORBIS:
195  case AV_CODEC_ID_THEORA:
196  if (!s->max_frames_per_packet) s->max_frames_per_packet = 15;
197  s->max_frames_per_packet = av_clip(s->max_frames_per_packet, 1, 15);
198  s->max_payload_size -= 6; // ident+frag+tdt/vdt+pkt_num+pkt_length
199  s->num_frames = 0;
200  goto defaultcase;
202  /* Due to a historical error, the clock rate for G722 in RTP is
203  * 8000, even if the sample rate is 16000. See RFC 3551. */
204  avpriv_set_pts_info(st, 32, 1, 8000);
205  break;
206  case AV_CODEC_ID_OPUS:
207  if (st->codec->channels > 2) {
208  av_log(s1, AV_LOG_ERROR, "Multistream opus not supported in RTP\n");
209  goto fail;
210  }
211  /* The opus RTP RFC says that all opus streams should use 48000 Hz
212  * as clock rate, since all opus sample rates can be expressed in
213  * this clock rate, and sample rate changes on the fly are supported. */
214  avpriv_set_pts_info(st, 32, 1, 48000);
215  break;
216  case AV_CODEC_ID_ILBC:
217  if (st->codec->block_align != 38 && st->codec->block_align != 50) {
218  av_log(s1, AV_LOG_ERROR, "Incorrect iLBC block size specified\n");
219  goto fail;
220  }
221  if (!s->max_frames_per_packet)
222  s->max_frames_per_packet = 1;
223  s->max_frames_per_packet = FFMIN(s->max_frames_per_packet,
224  s->max_payload_size / st->codec->block_align);
225  goto defaultcase;
226  case AV_CODEC_ID_AMR_NB:
227  case AV_CODEC_ID_AMR_WB:
228  if (!s->max_frames_per_packet)
229  s->max_frames_per_packet = 12;
230  if (st->codec->codec_id == AV_CODEC_ID_AMR_NB)
231  n = 31;
232  else
233  n = 61;
234  /* max_header_toc_size + the largest AMR payload must fit */
235  if (1 + s->max_frames_per_packet + n > s->max_payload_size) {
236  av_log(s1, AV_LOG_ERROR, "RTP max payload size too small for AMR\n");
237  goto fail;
238  }
239  if (st->codec->channels != 1) {
240  av_log(s1, AV_LOG_ERROR, "Only mono is supported\n");
241  goto fail;
242  }
243  case AV_CODEC_ID_AAC:
244  s->num_frames = 0;
245  default:
246 defaultcase:
247  if (st->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
248  avpriv_set_pts_info(st, 32, 1, st->codec->sample_rate);
249  }
250  s->buf_ptr = s->buf;
251  break;
252  }
253 
254  return 0;
255 
256 fail:
257  av_freep(&s->buf);
258  return AVERROR(EINVAL);
259 }
260 
261 /* send an rtcp sender report packet */
262 static void rtcp_send_sr(AVFormatContext *s1, int64_t ntp_time, int bye)
263 {
264  RTPMuxContext *s = s1->priv_data;
265  uint32_t rtp_ts;
266 
267  av_dlog(s1, "RTCP: %02x %"PRIx64" %x\n", s->payload_type, ntp_time, s->timestamp);
268 
269  s->last_rtcp_ntp_time = ntp_time;
270  rtp_ts = av_rescale_q(ntp_time - s->first_rtcp_ntp_time, (AVRational){1, 1000000},
271  s1->streams[0]->time_base) + s->base_timestamp;
272  avio_w8(s1->pb, RTP_VERSION << 6);
273  avio_w8(s1->pb, RTCP_SR);
274  avio_wb16(s1->pb, 6); /* length in words - 1 */
275  avio_wb32(s1->pb, s->ssrc);
276  avio_wb64(s1->pb, NTP_TO_RTP_FORMAT(ntp_time));
277  avio_wb32(s1->pb, rtp_ts);
278  avio_wb32(s1->pb, s->packet_count);
279  avio_wb32(s1->pb, s->octet_count);
280 
281  if (s->cname) {
282  int len = FFMIN(strlen(s->cname), 255);
283  avio_w8(s1->pb, (RTP_VERSION << 6) + 1);
284  avio_w8(s1->pb, RTCP_SDES);
285  avio_wb16(s1->pb, (7 + len + 3) / 4); /* length in words - 1 */
286 
287  avio_wb32(s1->pb, s->ssrc);
288  avio_w8(s1->pb, 0x01); /* CNAME */
289  avio_w8(s1->pb, len);
290  avio_write(s1->pb, s->cname, len);
291  avio_w8(s1->pb, 0); /* END */
292  for (len = (7 + len) % 4; len % 4; len++)
293  avio_w8(s1->pb, 0);
294  }
295 
296  if (bye) {
297  avio_w8(s1->pb, (RTP_VERSION << 6) | 1);
298  avio_w8(s1->pb, RTCP_BYE);
299  avio_wb16(s1->pb, 1); /* length in words - 1 */
300  avio_wb32(s1->pb, s->ssrc);
301  }
302 
303  avio_flush(s1->pb);
304 }
305 
306 /* send an rtp packet. sequence number is incremented, but the caller
307  must update the timestamp itself */
308 void ff_rtp_send_data(AVFormatContext *s1, const uint8_t *buf1, int len, int m)
309 {
310  RTPMuxContext *s = s1->priv_data;
311 
312  av_dlog(s1, "rtp_send_data size=%d\n", len);
313 
314  /* build the RTP header */
315  avio_w8(s1->pb, RTP_VERSION << 6);
316  avio_w8(s1->pb, (s->payload_type & 0x7f) | ((m & 0x01) << 7));
317  avio_wb16(s1->pb, s->seq);
318  avio_wb32(s1->pb, s->timestamp);
319  avio_wb32(s1->pb, s->ssrc);
320 
321  avio_write(s1->pb, buf1, len);
322  avio_flush(s1->pb);
323 
324  s->seq = (s->seq + 1) & 0xffff;
325  s->octet_count += len;
326  s->packet_count++;
327 }
328 
329 /* send an integer number of samples and compute time stamp and fill
330  the rtp send buffer before sending. */
332  const uint8_t *buf1, int size, int sample_size_bits)
333 {
334  RTPMuxContext *s = s1->priv_data;
335  int len, max_packet_size, n;
336  /* Calculate the number of bytes to get samples aligned on a byte border */
337  int aligned_samples_size = sample_size_bits/av_gcd(sample_size_bits, 8);
338 
339  max_packet_size = (s->max_payload_size / aligned_samples_size) * aligned_samples_size;
340  /* Not needed, but who knows. Don't check if samples aren't an even number of bytes. */
341  if ((sample_size_bits % 8) == 0 && ((8 * size) % sample_size_bits) != 0)
342  return AVERROR(EINVAL);
343  n = 0;
344  while (size > 0) {
345  s->buf_ptr = s->buf;
346  len = FFMIN(max_packet_size, size);
347 
348  /* copy data */
349  memcpy(s->buf_ptr, buf1, len);
350  s->buf_ptr += len;
351  buf1 += len;
352  size -= len;
353  s->timestamp = s->cur_timestamp + n * 8 / sample_size_bits;
354  ff_rtp_send_data(s1, s->buf, s->buf_ptr - s->buf, 0);
355  n += (s->buf_ptr - s->buf);
356  }
357  return 0;
358 }
359 
361  const uint8_t *buf1, int size)
362 {
363  RTPMuxContext *s = s1->priv_data;
364  int len, count, max_packet_size;
365 
366  max_packet_size = s->max_payload_size;
367 
368  /* test if we must flush because not enough space */
369  len = (s->buf_ptr - s->buf);
370  if ((len + size) > max_packet_size) {
371  if (len > 4) {
372  ff_rtp_send_data(s1, s->buf, s->buf_ptr - s->buf, 0);
373  s->buf_ptr = s->buf + 4;
374  }
375  }
376  if (s->buf_ptr == s->buf + 4) {
377  s->timestamp = s->cur_timestamp;
378  }
379 
380  /* add the packet */
381  if (size > max_packet_size) {
382  /* big packet: fragment */
383  count = 0;
384  while (size > 0) {
385  len = max_packet_size - 4;
386  if (len > size)
387  len = size;
388  /* build fragmented packet */
389  s->buf[0] = 0;
390  s->buf[1] = 0;
391  s->buf[2] = count >> 8;
392  s->buf[3] = count;
393  memcpy(s->buf + 4, buf1, len);
394  ff_rtp_send_data(s1, s->buf, len + 4, 0);
395  size -= len;
396  buf1 += len;
397  count += len;
398  }
399  } else {
400  if (s->buf_ptr == s->buf + 4) {
401  /* no fragmentation possible */
402  s->buf[0] = 0;
403  s->buf[1] = 0;
404  s->buf[2] = 0;
405  s->buf[3] = 0;
406  }
407  memcpy(s->buf_ptr, buf1, size);
408  s->buf_ptr += size;
409  }
410 }
411 
413  const uint8_t *buf1, int size)
414 {
415  RTPMuxContext *s = s1->priv_data;
416  int len, max_packet_size;
417 
418  max_packet_size = s->max_payload_size;
419 
420  while (size > 0) {
421  len = max_packet_size;
422  if (len > size)
423  len = size;
424 
425  s->timestamp = s->cur_timestamp;
426  ff_rtp_send_data(s1, buf1, len, (len == size));
427 
428  buf1 += len;
429  size -= len;
430  }
431 }
432 
433 /* NOTE: size is assumed to be an integer multiple of TS_PACKET_SIZE */
435  const uint8_t *buf1, int size)
436 {
437  RTPMuxContext *s = s1->priv_data;
438  int len, out_len;
439 
440  while (size >= TS_PACKET_SIZE) {
441  len = s->max_payload_size - (s->buf_ptr - s->buf);
442  if (len > size)
443  len = size;
444  memcpy(s->buf_ptr, buf1, len);
445  buf1 += len;
446  size -= len;
447  s->buf_ptr += len;
448 
449  out_len = s->buf_ptr - s->buf;
450  if (out_len >= s->max_payload_size) {
451  ff_rtp_send_data(s1, s->buf, out_len, 0);
452  s->buf_ptr = s->buf;
453  }
454  }
455 }
456 
457 static int rtp_send_ilbc(AVFormatContext *s1, const uint8_t *buf, int size)
458 {
459  RTPMuxContext *s = s1->priv_data;
460  AVStream *st = s1->streams[0];
461  int frame_duration = av_get_audio_frame_duration(st->codec, 0);
462  int frame_size = st->codec->block_align;
463  int frames = size / frame_size;
464 
465  while (frames > 0) {
466  int n = FFMIN(s->max_frames_per_packet - s->num_frames, frames);
467 
468  if (!s->num_frames) {
469  s->buf_ptr = s->buf;
470  s->timestamp = s->cur_timestamp;
471  }
472  memcpy(s->buf_ptr, buf, n * frame_size);
473  frames -= n;
474  s->num_frames += n;
475  s->buf_ptr += n * frame_size;
476  buf += n * frame_size;
477  s->cur_timestamp += n * frame_duration;
478 
479  if (s->num_frames == s->max_frames_per_packet) {
480  ff_rtp_send_data(s1, s->buf, s->buf_ptr - s->buf, 1);
481  s->num_frames = 0;
482  }
483  }
484  return 0;
485 }
486 
488 {
489  RTPMuxContext *s = s1->priv_data;
490  AVStream *st = s1->streams[0];
491  int rtcp_bytes;
492  int size= pkt->size;
493 
494  av_dlog(s1, "%d: write len=%d\n", pkt->stream_index, size);
495 
496  rtcp_bytes = ((s->octet_count - s->last_octet_count) * RTCP_TX_RATIO_NUM) /
498  if ((s->first_packet || ((rtcp_bytes >= RTCP_SR_SIZE) &&
499  (ff_ntp_time() - s->last_rtcp_ntp_time > 5000000))) &&
500  !(s->flags & FF_RTP_FLAG_SKIP_RTCP)) {
501  rtcp_send_sr(s1, ff_ntp_time(), 0);
503  s->first_packet = 0;
504  }
505  s->cur_timestamp = s->base_timestamp + pkt->pts;
506 
507  switch(st->codec->codec_id) {
510  case AV_CODEC_ID_PCM_U8:
511  case AV_CODEC_ID_PCM_S8:
512  return rtp_send_samples(s1, pkt->data, size, 8 * st->codec->channels);
517  return rtp_send_samples(s1, pkt->data, size, 16 * st->codec->channels);
519  /* The actual sample size is half a byte per sample, but since the
520  * stream clock rate is 8000 Hz while the sample rate is 16000 Hz,
521  * the correct parameter for send_samples_bits is 8 bits per stream
522  * clock. */
523  return rtp_send_samples(s1, pkt->data, size, 8 * st->codec->channels);
525  return rtp_send_samples(s1, pkt->data, size,
527  case AV_CODEC_ID_MP2:
528  case AV_CODEC_ID_MP3:
529  rtp_send_mpegaudio(s1, pkt->data, size);
530  break;
533  ff_rtp_send_mpegvideo(s1, pkt->data, size);
534  break;
535  case AV_CODEC_ID_AAC:
536  if (s->flags & FF_RTP_FLAG_MP4A_LATM)
537  ff_rtp_send_latm(s1, pkt->data, size);
538  else
539  ff_rtp_send_aac(s1, pkt->data, size);
540  break;
541  case AV_CODEC_ID_AMR_NB:
542  case AV_CODEC_ID_AMR_WB:
543  ff_rtp_send_amr(s1, pkt->data, size);
544  break;
545  case AV_CODEC_ID_MPEG2TS:
546  rtp_send_mpegts_raw(s1, pkt->data, size);
547  break;
548  case AV_CODEC_ID_H264:
549  ff_rtp_send_h264(s1, pkt->data, size);
550  break;
551  case AV_CODEC_ID_H263:
552  if (s->flags & FF_RTP_FLAG_RFC2190) {
553  int mb_info_size = 0;
554  const uint8_t *mb_info =
556  &mb_info_size);
557  ff_rtp_send_h263_rfc2190(s1, pkt->data, size, mb_info, mb_info_size);
558  break;
559  }
560  /* Fallthrough */
561  case AV_CODEC_ID_H263P:
562  ff_rtp_send_h263(s1, pkt->data, size);
563  break;
564  case AV_CODEC_ID_VORBIS:
565  case AV_CODEC_ID_THEORA:
566  ff_rtp_send_xiph(s1, pkt->data, size);
567  break;
568  case AV_CODEC_ID_VP8:
569  ff_rtp_send_vp8(s1, pkt->data, size);
570  break;
571  case AV_CODEC_ID_ILBC:
572  rtp_send_ilbc(s1, pkt->data, size);
573  break;
574  case AV_CODEC_ID_MJPEG:
575  ff_rtp_send_jpeg(s1, pkt->data, size);
576  break;
577  case AV_CODEC_ID_OPUS:
578  if (size > s->max_payload_size) {
579  av_log(s1, AV_LOG_ERROR,
580  "Packet size %d too large for max RTP payload size %d\n",
581  size, s->max_payload_size);
582  return AVERROR(EINVAL);
583  }
584  /* Intentional fallthrough */
585  default:
586  /* better than nothing : send the codec raw data */
587  rtp_send_raw(s1, pkt->data, size);
588  break;
589  }
590  return 0;
591 }
592 
594 {
595  RTPMuxContext *s = s1->priv_data;
596 
597  /* If the caller closes and recreates ->pb, this might actually
598  * be NULL here even if it was successfully allocated at the start. */
599  if (s1->pb && (s->flags & FF_RTP_FLAG_SEND_BYE))
600  rtcp_send_sr(s1, ff_ntp_time(), 1);
601  av_freep(&s->buf);
602 
603  return 0;
604 }
605 
607  .name = "rtp",
608  .long_name = NULL_IF_CONFIG_SMALL("RTP output"),
609  .priv_data_size = sizeof(RTPMuxContext),
610  .audio_codec = AV_CODEC_ID_PCM_MULAW,
611  .video_codec = AV_CODEC_ID_MPEG4,
615  .priv_class = &rtp_muxer_class,
616 };