1 /*
2 * RTP input format
3 * Copyright (c) 2002 Fabrice Bellard.
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 #include "avformat.h"
22 #include "mpegts.h"
23 #include "bitstream.h"
24
25 #include <unistd.h>
26 #include "network.h"
27
28 #include "rtp_internal.h"
29 #include "rtp_h264.h"
30
31 //#define DEBUG
32
33 /* TODO: - add RTCP statistics reporting (should be optional).
34
35 - add support for h263/mpeg4 packetized output : IDEA: send a
36 buffer to 'rtp_write_packet' contains all the packets for ONE
37 frame. Each packet should have a four byte header containing
38 the length in big endian format (same trick as
39 'url_open_dyn_packet_buf')
40 */
41
42 /* statistics functions */
43 RTPDynamicProtocolHandler *RTPFirstDynamicPayloadHandler= NULL;
44
45 static RTPDynamicProtocolHandler mp4v_es_handler= {"MP4V-ES", CODEC_TYPE_VIDEO, CODEC_ID_MPEG4};
46 static RTPDynamicProtocolHandler mpeg4_generic_handler= {"mpeg4-generic", CODEC_TYPE_AUDIO, CODEC_ID_AAC};
47
register_dynamic_payload_handler(RTPDynamicProtocolHandler * handler)48 static void register_dynamic_payload_handler(RTPDynamicProtocolHandler *handler)
49 {
50 handler->next= RTPFirstDynamicPayloadHandler;
51 RTPFirstDynamicPayloadHandler= handler;
52 }
53
av_register_rtp_dynamic_payload_handlers(void)54 void av_register_rtp_dynamic_payload_handlers(void)
55 {
56 register_dynamic_payload_handler(&mp4v_es_handler);
57 register_dynamic_payload_handler(&mpeg4_generic_handler);
58 register_dynamic_payload_handler(&ff_h264_dynamic_handler);
59 }
60
rtcp_parse_packet(RTPDemuxContext * s,const unsigned char * buf,int len)61 static int rtcp_parse_packet(RTPDemuxContext *s, const unsigned char *buf, int len)
62 {
63 if (buf[1] != 200)
64 return -1;
65 s->last_rtcp_ntp_time = AV_RB64(buf + 8);
66 if (s->first_rtcp_ntp_time == AV_NOPTS_VALUE)
67 s->first_rtcp_ntp_time = s->last_rtcp_ntp_time;
68 s->last_rtcp_timestamp = AV_RB32(buf + 16);
69 return 0;
70 }
71
72 #define RTP_SEQ_MOD (1<<16)
73
74 /**
75 * called on parse open packet
76 */
rtp_init_statistics(RTPStatistics * s,uint16_t base_sequence)77 static void rtp_init_statistics(RTPStatistics *s, uint16_t base_sequence) // called on parse open packet.
78 {
79 memset(s, 0, sizeof(RTPStatistics));
80 s->max_seq= base_sequence;
81 s->probation= 1;
82 }
83
84 /**
85 * called whenever there is a large jump in sequence numbers, or when they get out of probation...
86 */
rtp_init_sequence(RTPStatistics * s,uint16_t seq)87 static void rtp_init_sequence(RTPStatistics *s, uint16_t seq)
88 {
89 s->max_seq= seq;
90 s->cycles= 0;
91 s->base_seq= seq -1;
92 s->bad_seq= RTP_SEQ_MOD + 1;
93 s->received= 0;
94 s->expected_prior= 0;
95 s->received_prior= 0;
96 s->jitter= 0;
97 s->transit= 0;
98 }
99
100 /**
101 * returns 1 if we should handle this packet.
102 */
rtp_valid_packet_in_sequence(RTPStatistics * s,uint16_t seq)103 static int rtp_valid_packet_in_sequence(RTPStatistics *s, uint16_t seq)
104 {
105 uint16_t udelta= seq - s->max_seq;
106 const int MAX_DROPOUT= 3000;
107 const int MAX_MISORDER = 100;
108 const int MIN_SEQUENTIAL = 2;
109
110 /* source not valid until MIN_SEQUENTIAL packets with sequence seq. numbers have been received */
111 if(s->probation)
112 {
113 if(seq==s->max_seq + 1) {
114 s->probation--;
115 s->max_seq= seq;
116 if(s->probation==0) {
117 rtp_init_sequence(s, seq);
118 s->received++;
119 return 1;
120 }
121 } else {
122 s->probation= MIN_SEQUENTIAL - 1;
123 s->max_seq = seq;
124 }
125 } else if (udelta < MAX_DROPOUT) {
126 // in order, with permissible gap
127 if(seq < s->max_seq) {
128 //sequence number wrapped; count antother 64k cycles
129 s->cycles += RTP_SEQ_MOD;
130 }
131 s->max_seq= seq;
132 } else if (udelta <= RTP_SEQ_MOD - MAX_MISORDER) {
133 // sequence made a large jump...
134 if(seq==s->bad_seq) {
135 // two sequential packets-- assume that the other side restarted without telling us; just resync.
136 rtp_init_sequence(s, seq);
137 } else {
138 s->bad_seq= (seq + 1) & (RTP_SEQ_MOD-1);
139 return 0;
140 }
141 } else {
142 // duplicate or reordered packet...
143 }
144 s->received++;
145 return 1;
146 }
147
148 #if 0
149 /**
150 * This function is currently unused; without a valid local ntp time, I don't see how we could calculate the
151 * difference between the arrival and sent timestamp. As a result, the jitter and transit statistics values
152 * never change. I left this in in case someone else can see a way. (rdm)
153 */
154 static void rtcp_update_jitter(RTPStatistics *s, uint32_t sent_timestamp, uint32_t arrival_timestamp)
155 {
156 uint32_t transit= arrival_timestamp - sent_timestamp;
157 int d;
158 s->transit= transit;
159 d= FFABS(transit - s->transit);
160 s->jitter += d - ((s->jitter + 8)>>4);
161 }
162 #endif
163
rtp_check_and_send_back_rr(RTPDemuxContext * s,int count)164 int rtp_check_and_send_back_rr(RTPDemuxContext *s, int count)
165 {
166 ByteIOContext *pb;
167 uint8_t *buf;
168 int len;
169 int rtcp_bytes;
170 RTPStatistics *stats= &s->statistics;
171 uint32_t lost;
172 uint32_t extended_max;
173 uint32_t expected_interval;
174 uint32_t received_interval;
175 uint32_t lost_interval;
176 uint32_t expected;
177 uint32_t fraction;
178 uint64_t ntp_time= s->last_rtcp_ntp_time; // TODO: Get local ntp time?
179
180 if (!s->rtp_ctx || (count < 1))
181 return -1;
182
183 /* TODO: I think this is way too often; RFC 1889 has algorithm for this */
184 /* XXX: mpeg pts hardcoded. RTCP send every 0.5 seconds */
185 s->octet_count += count;
186 rtcp_bytes = ((s->octet_count - s->last_octet_count) * RTCP_TX_RATIO_NUM) /
187 RTCP_TX_RATIO_DEN;
188 rtcp_bytes /= 50; // mmu_man: that's enough for me... VLC sends much less btw !?
189 if (rtcp_bytes < 28)
190 return -1;
191 s->last_octet_count = s->octet_count;
192
193 if (url_open_dyn_buf(&pb) < 0)
194 return -1;
195
196 // Receiver Report
197 put_byte(pb, (RTP_VERSION << 6) + 1); /* 1 report block */
198 put_byte(pb, 201);
199 put_be16(pb, 7); /* length in words - 1 */
200 put_be32(pb, s->ssrc); // our own SSRC
201 put_be32(pb, s->ssrc); // XXX: should be the server's here!
202 // some placeholders we should really fill...
203 // RFC 1889/p64
204 extended_max= stats->cycles + stats->max_seq;
205 expected= extended_max - stats->base_seq + 1;
206 lost= expected - stats->received;
207 lost= FFMIN(lost, 0xffffff); // clamp it since it's only 24 bits...
208 expected_interval= expected - stats->expected_prior;
209 stats->expected_prior= expected;
210 received_interval= stats->received - stats->received_prior;
211 stats->received_prior= stats->received;
212 lost_interval= expected_interval - received_interval;
213 if (expected_interval==0 || lost_interval<=0) fraction= 0;
214 else fraction = (lost_interval<<8)/expected_interval;
215
216 fraction= (fraction<<24) | lost;
217
218 put_be32(pb, fraction); /* 8 bits of fraction, 24 bits of total packets lost */
219 put_be32(pb, extended_max); /* max sequence received */
220 put_be32(pb, stats->jitter>>4); /* jitter */
221
222 if(s->last_rtcp_ntp_time==AV_NOPTS_VALUE)
223 {
224 put_be32(pb, 0); /* last SR timestamp */
225 put_be32(pb, 0); /* delay since last SR */
226 } else {
227 uint32_t middle_32_bits= s->last_rtcp_ntp_time>>16; // this is valid, right? do we need to handle 64 bit values special?
228 uint32_t delay_since_last= ntp_time - s->last_rtcp_ntp_time;
229
230 put_be32(pb, middle_32_bits); /* last SR timestamp */
231 put_be32(pb, delay_since_last); /* delay since last SR */
232 }
233
234 // CNAME
235 put_byte(pb, (RTP_VERSION << 6) + 1); /* 1 report block */
236 put_byte(pb, 202);
237 len = strlen(s->hostname);
238 put_be16(pb, (6 + len + 3) / 4); /* length in words - 1 */
239 put_be32(pb, s->ssrc);
240 put_byte(pb, 0x01);
241 put_byte(pb, len);
242 put_buffer(pb, s->hostname, len);
243 // padding
244 for (len = (6 + len) % 4; len % 4; len++) {
245 put_byte(pb, 0);
246 }
247
248 put_flush_packet(pb);
249 len = url_close_dyn_buf(pb, &buf);
250 if ((len > 0) && buf) {
251 int result;
252 #if defined(DEBUG)
253 printf("sending %d bytes of RR\n", len);
254 #endif
255 result= url_write(s->rtp_ctx, buf, len);
256 #if defined(DEBUG)
257 printf("result from url_write: %d\n", result);
258 #endif
259 av_free(buf);
260 }
261 return 0;
262 }
263
264 /**
265 * open a new RTP parse context for stream 'st'. 'st' can be NULL for
266 * MPEG2TS streams to indicate that they should be demuxed inside the
267 * rtp demux (otherwise CODEC_ID_MPEG2TS packets are returned)
268 * TODO: change this to not take rtp_payload data, and use the new dynamic payload system.
269 */
rtp_parse_open(AVFormatContext * s1,AVStream * st,URLContext * rtpc,int payload_type,rtp_payload_data_t * rtp_payload_data)270 RTPDemuxContext *rtp_parse_open(AVFormatContext *s1, AVStream *st, URLContext *rtpc, int payload_type, rtp_payload_data_t *rtp_payload_data)
271 {
272 RTPDemuxContext *s;
273
274 s = av_mallocz(sizeof(RTPDemuxContext));
275 if (!s)
276 return NULL;
277 s->payload_type = payload_type;
278 s->last_rtcp_ntp_time = AV_NOPTS_VALUE;
279 s->first_rtcp_ntp_time = AV_NOPTS_VALUE;
280 s->ic = s1;
281 s->st = st;
282 s->rtp_payload_data = rtp_payload_data;
283 rtp_init_statistics(&s->statistics, 0); // do we know the initial sequence from sdp?
284 if (!strcmp(ff_rtp_enc_name(payload_type), "MP2T")) {
285 s->ts = mpegts_parse_open(s->ic);
286 if (s->ts == NULL) {
287 av_free(s);
288 return NULL;
289 }
290 } else {
291 switch(st->codec->codec_id) {
292 case CODEC_ID_MPEG1VIDEO:
293 case CODEC_ID_MPEG2VIDEO:
294 case CODEC_ID_MP2:
295 case CODEC_ID_MP3:
296 case CODEC_ID_MPEG4:
297 case CODEC_ID_H264:
298 st->need_parsing = AVSTREAM_PARSE_FULL;
299 break;
300 default:
301 break;
302 }
303 }
304 // needed to send back RTCP RR in RTSP sessions
305 s->rtp_ctx = rtpc;
306 gethostname(s->hostname, sizeof(s->hostname));
307 return s;
308 }
309
rtp_parse_mp4_au(RTPDemuxContext * s,const uint8_t * buf)310 static int rtp_parse_mp4_au(RTPDemuxContext *s, const uint8_t *buf)
311 {
312 int au_headers_length, au_header_size, i;
313 GetBitContext getbitcontext;
314 rtp_payload_data_t *infos;
315
316 infos = s->rtp_payload_data;
317
318 if (infos == NULL)
319 return -1;
320
321 /* decode the first 2 bytes where the AUHeader sections are stored
322 length in bits */
323 au_headers_length = AV_RB16(buf);
324
325 if (au_headers_length > RTP_MAX_PACKET_LENGTH)
326 return -1;
327
328 infos->au_headers_length_bytes = (au_headers_length + 7) / 8;
329
330 /* skip AU headers length section (2 bytes) */
331 buf += 2;
332
333 init_get_bits(&getbitcontext, buf, infos->au_headers_length_bytes * 8);
334
335 /* XXX: Wrong if optionnal additional sections are present (cts, dts etc...) */
336 au_header_size = infos->sizelength + infos->indexlength;
337 if (au_header_size <= 0 || (au_headers_length % au_header_size != 0))
338 return -1;
339
340 infos->nb_au_headers = au_headers_length / au_header_size;
341 infos->au_headers = av_malloc(sizeof(struct AUHeaders) * infos->nb_au_headers);
342
343 /* XXX: We handle multiple AU Section as only one (need to fix this for interleaving)
344 In my test, the FAAD decoder does not behave correctly when sending each AU one by one
345 but does when sending the whole as one big packet... */
346 infos->au_headers[0].size = 0;
347 infos->au_headers[0].index = 0;
348 for (i = 0; i < infos->nb_au_headers; ++i) {
349 infos->au_headers[0].size += get_bits_long(&getbitcontext, infos->sizelength);
350 infos->au_headers[0].index = get_bits_long(&getbitcontext, infos->indexlength);
351 }
352
353 infos->nb_au_headers = 1;
354
355 return 0;
356 }
357
358 /**
359 * This was the second switch in rtp_parse packet. Normalizes time, if required, sets stream_index, etc.
360 */
finalize_packet(RTPDemuxContext * s,AVPacket * pkt,uint32_t timestamp)361 static void finalize_packet(RTPDemuxContext *s, AVPacket *pkt, uint32_t timestamp)
362 {
363 switch(s->st->codec->codec_id) {
364 case CODEC_ID_MP2:
365 case CODEC_ID_MPEG1VIDEO:
366 case CODEC_ID_MPEG2VIDEO:
367 if (s->last_rtcp_ntp_time != AV_NOPTS_VALUE) {
368 int64_t addend;
369
370 int delta_timestamp;
371 /* XXX: is it really necessary to unify the timestamp base ? */
372 /* compute pts from timestamp with received ntp_time */
373 delta_timestamp = timestamp - s->last_rtcp_timestamp;
374 /* convert to 90 kHz without overflow */
375 addend = (s->last_rtcp_ntp_time - s->first_rtcp_ntp_time) >> 14;
376 addend = (addend * 5625) >> 14;
377 pkt->pts = addend + delta_timestamp;
378 }
379 break;
380 case CODEC_ID_AAC:
381 case CODEC_ID_H264:
382 case CODEC_ID_MPEG4:
383 pkt->pts = timestamp;
384 break;
385 default:
386 /* no timestamp info yet */
387 break;
388 }
389 pkt->stream_index = s->st->index;
390 }
391
392 /**
393 * Parse an RTP or RTCP packet directly sent as a buffer.
394 * @param s RTP parse context.
395 * @param pkt returned packet
396 * @param buf input buffer or NULL to read the next packets
397 * @param len buffer len
398 * @return 0 if a packet is returned, 1 if a packet is returned and more can follow
399 * (use buf as NULL to read the next). -1 if no packet (error or no more packet).
400 */
rtp_parse_packet(RTPDemuxContext * s,AVPacket * pkt,const uint8_t * buf,int len)401 int rtp_parse_packet(RTPDemuxContext *s, AVPacket *pkt,
402 const uint8_t *buf, int len)
403 {
404 unsigned int ssrc, h;
405 int payload_type, seq, ret, flags = 0;
406 AVStream *st;
407 uint32_t timestamp;
408 int rv= 0;
409
410 if (!buf) {
411 /* return the next packets, if any */
412 if(s->st && s->parse_packet) {
413 timestamp= 0; ///< Should not be used if buf is NULL, but should be set to the timestamp of the packet returned....
414 rv= s->parse_packet(s, pkt, ×tamp, NULL, 0, flags);
415 finalize_packet(s, pkt, timestamp);
416 return rv;
417 } else {
418 // TODO: Move to a dynamic packet handler (like above)
419 if (s->read_buf_index >= s->read_buf_size)
420 return -1;
421 ret = mpegts_parse_packet(s->ts, pkt, s->buf + s->read_buf_index,
422 s->read_buf_size - s->read_buf_index);
423 if (ret < 0)
424 return -1;
425 s->read_buf_index += ret;
426 if (s->read_buf_index < s->read_buf_size)
427 return 1;
428 else
429 return 0;
430 }
431 }
432
433 if (len < 12)
434 return -1;
435
436 if ((buf[0] & 0xc0) != (RTP_VERSION << 6))
437 return -1;
438 if (buf[1] >= 200 && buf[1] <= 204) {
439 rtcp_parse_packet(s, buf, len);
440 return -1;
441 }
442 payload_type = buf[1] & 0x7f;
443 seq = AV_RB16(buf + 2);
444 timestamp = AV_RB32(buf + 4);
445 ssrc = AV_RB32(buf + 8);
446 /* store the ssrc in the RTPDemuxContext */
447 s->ssrc = ssrc;
448
449 /* NOTE: we can handle only one payload type */
450 if (s->payload_type != payload_type)
451 return -1;
452
453 st = s->st;
454 // only do something with this if all the rtp checks pass...
455 if(!rtp_valid_packet_in_sequence(&s->statistics, seq))
456 {
457 av_log(st?st->codec:NULL, AV_LOG_ERROR, "RTP: PT=%02x: bad cseq %04x expected=%04x\n",
458 payload_type, seq, ((s->seq + 1) & 0xffff));
459 return -1;
460 }
461
462 s->seq = seq;
463 len -= 12;
464 buf += 12;
465
466 if (!st) {
467 /* specific MPEG2TS demux support */
468 ret = mpegts_parse_packet(s->ts, pkt, buf, len);
469 if (ret < 0)
470 return -1;
471 if (ret < len) {
472 s->read_buf_size = len - ret;
473 memcpy(s->buf, buf + ret, s->read_buf_size);
474 s->read_buf_index = 0;
475 return 1;
476 }
477 } else if (s->parse_packet) {
478 rv = s->parse_packet(s, pkt, ×tamp, buf, len, flags);
479 } else {
480 // at this point, the RTP header has been stripped; This is ASSUMING that there is only 1 CSRC, which in't wise.
481 switch(st->codec->codec_id) {
482 case CODEC_ID_MP2:
483 /* better than nothing: skip mpeg audio RTP header */
484 if (len <= 4)
485 return -1;
486 h = AV_RB32(buf);
487 len -= 4;
488 buf += 4;
489 av_new_packet(pkt, len);
490 memcpy(pkt->data, buf, len);
491 break;
492 case CODEC_ID_MPEG1VIDEO:
493 case CODEC_ID_MPEG2VIDEO:
494 /* better than nothing: skip mpeg video RTP header */
495 if (len <= 4)
496 return -1;
497 h = AV_RB32(buf);
498 buf += 4;
499 len -= 4;
500 if (h & (1 << 26)) {
501 /* mpeg2 */
502 if (len <= 4)
503 return -1;
504 buf += 4;
505 len -= 4;
506 }
507 av_new_packet(pkt, len);
508 memcpy(pkt->data, buf, len);
509 break;
510 // moved from below, verbatim. this is because this section handles packets, and the lower switch handles
511 // timestamps.
512 // TODO: Put this into a dynamic packet handler...
513 case CODEC_ID_AAC:
514 if (rtp_parse_mp4_au(s, buf))
515 return -1;
516 {
517 rtp_payload_data_t *infos = s->rtp_payload_data;
518 if (infos == NULL)
519 return -1;
520 buf += infos->au_headers_length_bytes + 2;
521 len -= infos->au_headers_length_bytes + 2;
522
523 /* XXX: Fixme we only handle the case where rtp_parse_mp4_au define
524 one au_header */
525 av_new_packet(pkt, infos->au_headers[0].size);
526 memcpy(pkt->data, buf, infos->au_headers[0].size);
527 buf += infos->au_headers[0].size;
528 len -= infos->au_headers[0].size;
529 }
530 s->read_buf_size = len;
531 rv= 0;
532 break;
533 default:
534 av_new_packet(pkt, len);
535 memcpy(pkt->data, buf, len);
536 break;
537 }
538
539 // now perform timestamp things....
540 finalize_packet(s, pkt, timestamp);
541 }
542 return rv;
543 }
544
rtp_parse_close(RTPDemuxContext * s)545 void rtp_parse_close(RTPDemuxContext *s)
546 {
547 // TODO: fold this into the protocol specific data fields.
548 if (!strcmp(ff_rtp_enc_name(s->payload_type), "MP2T")) {
549 mpegts_parse_close(s->ts);
550 }
551 av_free(s);
552 }
553