2 * Apple HTTP Live Streaming demuxer
3 * Copyright (c) 2010 Martin Storsjo
5 * This file is part of FFmpeg.
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.
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.
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
24 * Apple HTTP Live Streaming demuxer
25 * http://tools.ietf.org/html/draft-pantos-http-live-streaming
28 #include "libavutil/avstring.h"
29 #include "libavutil/intreadwrite.h"
30 #include "libavutil/mathematics.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/dict.h"
33 #include "libavutil/time.h"
36 #include "avio_internal.h"
39 #define INITIAL_BUFFER_SIZE 32768
42 * An apple http stream consists of a playlist with media segment files,
43 * played sequentially. There may be several playlists with the same
44 * video content, in different bandwidth variants, that are played in
45 * parallel (preferably only one bandwidth variant at a time). In this case,
46 * the user supplied the url to a main playlist that only lists the variant
49 * If the main playlist doesn't point at any variants, we still create
50 * one anonymous toplevel variant for this, to maintain the structure.
60 char url[MAX_URL_SIZE];
61 char key[MAX_URL_SIZE];
62 enum KeyType key_type;
67 * Each variant has its own demuxer. If it currently is active,
68 * it has an open AVIOContext too, and potentially an AVPacket
69 * containing the next packet from this stream.
73 char url[MAX_URL_SIZE];
77 AVFormatContext *parent;
87 struct segment **segments;
88 int needed, cur_needed;
90 int64_t last_load_time;
92 char key_url[MAX_URL_SIZE];
96 typedef struct HLSContext {
98 struct variant **variants;
102 int64_t first_timestamp;
103 int64_t seek_timestamp;
105 AVIOInterruptCB *interrupt_callback;
106 char *user_agent; ///< holds HTTP user agent set as an AVOption to the HTTP protocol context
107 char *cookies; ///< holds HTTP cookie values set in either the initial response or as an AVOption to the HTTP protocol context
110 static int read_chomp_line(AVIOContext *s, char *buf, int maxlen)
112 int len = ff_get_line(s, buf, maxlen);
113 while (len > 0 && isspace(buf[len - 1]))
118 static void free_segment_list(struct variant *var)
121 for (i = 0; i < var->n_segments; i++)
122 av_free(var->segments[i]);
123 av_freep(&var->segments);
127 static void free_variant_list(HLSContext *c)
130 for (i = 0; i < c->n_variants; i++) {
131 struct variant *var = c->variants[i];
132 free_segment_list(var);
133 av_free_packet(&var->pkt);
134 av_free(var->pb.buffer);
136 ffurl_close(var->input);
139 avformat_close_input(&var->ctx);
143 av_freep(&c->variants);
144 av_freep(&c->cookies);
145 av_freep(&c->user_agent);
150 * Used to reset a statically allocated AVPacket to a clean slate,
151 * containing no data.
153 static void reset_packet(AVPacket *pkt)
159 static struct variant *new_variant(HLSContext *c, int bandwidth,
160 const char *url, const char *base)
162 struct variant *var = av_mallocz(sizeof(struct variant));
165 reset_packet(&var->pkt);
166 var->bandwidth = bandwidth;
167 ff_make_absolute_url(var->url, sizeof(var->url), base, url);
168 dynarray_add(&c->variants, &c->n_variants, var);
172 struct variant_info {
176 static void handle_variant_args(struct variant_info *info, const char *key,
177 int key_len, char **dest, int *dest_len)
179 if (!strncmp(key, "BANDWIDTH=", key_len)) {
180 *dest = info->bandwidth;
181 *dest_len = sizeof(info->bandwidth);
186 char uri[MAX_URL_SIZE];
191 static void handle_key_args(struct key_info *info, const char *key,
192 int key_len, char **dest, int *dest_len)
194 if (!strncmp(key, "METHOD=", key_len)) {
195 *dest = info->method;
196 *dest_len = sizeof(info->method);
197 } else if (!strncmp(key, "URI=", key_len)) {
199 *dest_len = sizeof(info->uri);
200 } else if (!strncmp(key, "IV=", key_len)) {
202 *dest_len = sizeof(info->iv);
206 static int parse_playlist(HLSContext *c, const char *url,
207 struct variant *var, AVIOContext *in)
209 int ret = 0, duration = 0, is_segment = 0, is_variant = 0, bandwidth = 0;
210 enum KeyType key_type = KEY_NONE;
213 char key[MAX_URL_SIZE] = "";
219 AVDictionary *opts = NULL;
221 /* Some HLS servers dont like being sent the range header */
222 av_dict_set(&opts, "seekable", "0", 0);
224 // broker prior HTTP options that should be consistent across requests
225 av_dict_set(&opts, "user-agent", c->user_agent, 0);
226 av_dict_set(&opts, "cookies", c->cookies, 0);
228 ret = avio_open2(&in, url, AVIO_FLAG_READ,
229 c->interrupt_callback, &opts);
235 read_chomp_line(in, line, sizeof(line));
236 if (strcmp(line, "#EXTM3U")) {
237 ret = AVERROR_INVALIDDATA;
242 free_segment_list(var);
245 while (!url_feof(in)) {
246 read_chomp_line(in, line, sizeof(line));
247 if (av_strstart(line, "#EXT-X-STREAM-INF:", &ptr)) {
248 struct variant_info info = {{0}};
250 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_variant_args,
252 bandwidth = atoi(info.bandwidth);
253 } else if (av_strstart(line, "#EXT-X-KEY:", &ptr)) {
254 struct key_info info = {{0}};
255 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_key_args,
259 if (!strcmp(info.method, "AES-128"))
260 key_type = KEY_AES_128;
261 if (!strncmp(info.iv, "0x", 2) || !strncmp(info.iv, "0X", 2)) {
262 ff_hex_to_data(iv, info.iv + 2);
265 av_strlcpy(key, info.uri, sizeof(key));
266 } else if (av_strstart(line, "#EXT-X-TARGETDURATION:", &ptr)) {
268 var = new_variant(c, 0, url, NULL);
270 ret = AVERROR(ENOMEM);
274 var->target_duration = atoi(ptr);
275 } else if (av_strstart(line, "#EXT-X-MEDIA-SEQUENCE:", &ptr)) {
277 var = new_variant(c, 0, url, NULL);
279 ret = AVERROR(ENOMEM);
283 var->start_seq_no = atoi(ptr);
284 } else if (av_strstart(line, "#EXT-X-ENDLIST", &ptr)) {
287 } else if (av_strstart(line, "#EXTINF:", &ptr)) {
289 duration = atoi(ptr);
290 } else if (av_strstart(line, "#", NULL)) {
292 } else if (line[0]) {
294 if (!new_variant(c, bandwidth, line, url)) {
295 ret = AVERROR(ENOMEM);
304 var = new_variant(c, 0, url, NULL);
306 ret = AVERROR(ENOMEM);
310 seg = av_malloc(sizeof(struct segment));
312 ret = AVERROR(ENOMEM);
315 seg->duration = duration;
316 seg->key_type = key_type;
318 memcpy(seg->iv, iv, sizeof(iv));
320 int seq = var->start_seq_no + var->n_segments;
321 memset(seg->iv, 0, sizeof(seg->iv));
322 AV_WB32(seg->iv + 12, seq);
324 ff_make_absolute_url(seg->key, sizeof(seg->key), url, key);
325 ff_make_absolute_url(seg->url, sizeof(seg->url), url, line);
326 dynarray_add(&var->segments, &var->n_segments, seg);
332 var->last_load_time = av_gettime();
340 static int open_input(HLSContext *c, struct variant *var)
342 AVDictionary *opts = NULL;
344 struct segment *seg = var->segments[var->cur_seq_no - var->start_seq_no];
346 // broker prior HTTP options that should be consistent across requests
347 av_dict_set(&opts, "user-agent", c->user_agent, 0);
348 av_dict_set(&opts, "cookies", c->cookies, 0);
349 av_dict_set(&opts, "seekable", "0", 0);
351 if (seg->key_type == KEY_NONE) {
352 ret = ffurl_open(&var->input, seg->url, AVIO_FLAG_READ,
353 &var->parent->interrupt_callback, &opts);
355 } else if (seg->key_type == KEY_AES_128) {
356 char iv[33], key[33], url[MAX_URL_SIZE];
357 if (strcmp(seg->key, var->key_url)) {
359 if (ffurl_open(&uc, seg->key, AVIO_FLAG_READ,
360 &var->parent->interrupt_callback, &opts) == 0) {
361 if (ffurl_read_complete(uc, var->key, sizeof(var->key))
362 != sizeof(var->key)) {
363 av_log(NULL, AV_LOG_ERROR, "Unable to read key file %s\n",
368 av_log(NULL, AV_LOG_ERROR, "Unable to open key file %s\n",
371 av_strlcpy(var->key_url, seg->key, sizeof(var->key_url));
373 ff_data_to_hex(iv, seg->iv, sizeof(seg->iv), 0);
374 ff_data_to_hex(key, var->key, sizeof(var->key), 0);
375 iv[32] = key[32] = '\0';
376 if (strstr(seg->url, "://"))
377 snprintf(url, sizeof(url), "crypto+%s", seg->url);
379 snprintf(url, sizeof(url), "crypto:%s", seg->url);
380 if ((ret = ffurl_alloc(&var->input, url, AVIO_FLAG_READ,
381 &var->parent->interrupt_callback)) < 0)
383 av_opt_set(var->input->priv_data, "key", key, 0);
384 av_opt_set(var->input->priv_data, "iv", iv, 0);
385 /* Need to repopulate options */
387 av_dict_set(&opts, "seekable", "0", 0);
388 if ((ret = ffurl_connect(var->input, &opts)) < 0) {
389 ffurl_close(var->input);
396 ret = AVERROR(ENOSYS);
403 static int read_data(void *opaque, uint8_t *buf, int buf_size)
405 struct variant *v = opaque;
406 HLSContext *c = v->parent->priv_data;
411 /* If this is a live stream and the reload interval has elapsed since
412 * the last playlist reload, reload the variant playlists now. */
413 int64_t reload_interval = v->n_segments > 0 ?
414 v->segments[v->n_segments - 1]->duration :
416 reload_interval *= 1000000;
420 av_gettime() - v->last_load_time >= reload_interval) {
421 if ((ret = parse_playlist(c, v->url, v, NULL)) < 0)
423 /* If we need to reload the playlist again below (if
424 * there's still no more segments), switch to a reload
425 * interval of half the target duration. */
426 reload_interval = v->target_duration * 500000LL;
428 if (v->cur_seq_no < v->start_seq_no) {
429 av_log(NULL, AV_LOG_WARNING,
430 "skipping %d segments ahead, expired from playlists\n",
431 v->start_seq_no - v->cur_seq_no);
432 v->cur_seq_no = v->start_seq_no;
434 if (v->cur_seq_no >= v->start_seq_no + v->n_segments) {
437 while (av_gettime() - v->last_load_time < reload_interval) {
438 if (ff_check_interrupt(c->interrupt_callback))
442 /* Enough time has elapsed since the last reload */
446 ret = open_input(c, v);
450 ret = ffurl_read(v->input, buf, buf_size);
453 ffurl_close(v->input);
457 c->end_of_segment = 1;
458 c->cur_seq_no = v->cur_seq_no;
460 if (v->ctx && v->ctx->nb_streams && v->parent->nb_streams >= v->stream_offset + v->ctx->nb_streams) {
462 for (i = v->stream_offset; i < v->stream_offset + v->ctx->nb_streams;
464 if (v->parent->streams[i]->discard < AVDISCARD_ALL)
469 av_log(v->parent, AV_LOG_INFO, "No longer receiving variant %d\n",
476 static int hls_read_header(AVFormatContext *s)
478 URLContext *u = s->pb->opaque;
479 HLSContext *c = s->priv_data;
480 int ret = 0, i, j, stream_offset = 0;
482 c->interrupt_callback = &s->interrupt_callback;
484 // if the URL context is good, read important options we must broker later
485 if (u && u->prot->priv_data_class) {
486 // get the previous user agent & set back to null if string size is zero
487 av_freep(&c->user_agent);
488 av_opt_get(u->priv_data, "user-agent", 0, (uint8_t**)&(c->user_agent));
489 if (c->user_agent && !strlen(c->user_agent))
490 av_freep(&c->user_agent);
492 // get the previous cookies & set back to null if string size is zero
493 av_freep(&c->cookies);
494 av_opt_get(u->priv_data, "cookies", 0, (uint8_t**)&(c->cookies));
495 if (c->cookies && !strlen(c->cookies))
496 av_freep(&c->cookies);
499 if ((ret = parse_playlist(c, s->filename, NULL, s->pb)) < 0)
502 if (c->n_variants == 0) {
503 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
507 /* If the playlist only contained variants, parse each individual
508 * variant playlist. */
509 if (c->n_variants > 1 || c->variants[0]->n_segments == 0) {
510 for (i = 0; i < c->n_variants; i++) {
511 struct variant *v = c->variants[i];
512 if ((ret = parse_playlist(c, v->url, v, NULL)) < 0)
517 if (c->variants[0]->n_segments == 0) {
518 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
523 /* If this isn't a live stream, calculate the total duration of the
525 if (c->variants[0]->finished) {
526 int64_t duration = 0;
527 for (i = 0; i < c->variants[0]->n_segments; i++)
528 duration += c->variants[0]->segments[i]->duration;
529 s->duration = duration * AV_TIME_BASE;
532 /* Open the demuxer for each variant */
533 for (i = 0; i < c->n_variants; i++) {
534 struct variant *v = c->variants[i];
535 AVInputFormat *in_fmt = NULL;
536 char bitrate_str[20];
537 AVProgram *program = NULL;
538 if (v->n_segments == 0)
541 if (!(v->ctx = avformat_alloc_context())) {
542 ret = AVERROR(ENOMEM);
550 /* If this is a live stream with more than 3 segments, start at the
551 * third last segment. */
552 v->cur_seq_no = v->start_seq_no;
553 if (!v->finished && v->n_segments > 3)
554 v->cur_seq_no = v->start_seq_no + v->n_segments - 3;
556 v->read_buffer = av_malloc(INITIAL_BUFFER_SIZE);
557 ffio_init_context(&v->pb, v->read_buffer, INITIAL_BUFFER_SIZE, 0, v,
558 read_data, NULL, NULL);
560 ret = av_probe_input_buffer(&v->pb, &in_fmt, v->segments[0]->url,
563 /* Free the ctx - it isn't initialized properly at this point,
564 * so avformat_close_input shouldn't be called. If
565 * avformat_open_input fails below, it frees and zeros the
566 * context, so it doesn't need any special treatment like this. */
567 av_log(s, AV_LOG_ERROR, "Error when loading first segment '%s'\n", v->segments[0]->url);
568 avformat_free_context(v->ctx);
573 ret = avformat_open_input(&v->ctx, v->segments[0]->url, in_fmt, NULL);
577 v->stream_offset = stream_offset;
578 v->ctx->ctx_flags &= ~AVFMTCTX_NOHEADER;
579 ret = avformat_find_stream_info(v->ctx, NULL);
582 snprintf(bitrate_str, sizeof(bitrate_str), "%d", v->bandwidth);
584 /* Create new AVprogram for variant i */
585 program = av_new_program(s, i);
588 av_dict_set(&program->metadata, "variant_bitrate", bitrate_str, 0);
590 /* Create new AVStreams for each stream in this variant */
591 for (j = 0; j < v->ctx->nb_streams; j++) {
592 AVStream *st = avformat_new_stream(s, NULL);
594 ret = AVERROR(ENOMEM);
597 ff_program_add_stream_index(s, i, stream_offset + j);
599 avcodec_copy_context(st->codec, v->ctx->streams[j]->codec);
601 av_dict_set(&st->metadata, "variant_bitrate", bitrate_str,
604 stream_offset += v->ctx->nb_streams;
608 c->first_timestamp = AV_NOPTS_VALUE;
609 c->seek_timestamp = AV_NOPTS_VALUE;
613 free_variant_list(c);
617 static int recheck_discard_flags(AVFormatContext *s, int first)
619 HLSContext *c = s->priv_data;
622 /* Check if any new streams are needed */
623 for (i = 0; i < c->n_variants; i++)
624 c->variants[i]->cur_needed = 0;
626 for (i = 0; i < s->nb_streams; i++) {
627 AVStream *st = s->streams[i];
628 struct variant *var = c->variants[s->streams[i]->id];
629 if (st->discard < AVDISCARD_ALL)
632 for (i = 0; i < c->n_variants; i++) {
633 struct variant *v = c->variants[i];
634 if (v->cur_needed && !v->needed) {
637 v->cur_seq_no = c->cur_seq_no;
638 v->pb.eof_reached = 0;
639 av_log(s, AV_LOG_INFO, "Now receiving variant %d\n", i);
640 } else if (first && !v->cur_needed && v->needed) {
642 ffurl_close(v->input);
646 av_log(s, AV_LOG_INFO, "No longer receiving variant %d\n", i);
652 static int hls_read_packet(AVFormatContext *s, AVPacket *pkt)
654 HLSContext *c = s->priv_data;
655 int ret, i, minvariant = -1;
657 if (c->first_packet) {
658 recheck_discard_flags(s, 1);
663 c->end_of_segment = 0;
664 for (i = 0; i < c->n_variants; i++) {
665 struct variant *var = c->variants[i];
666 /* Make sure we've got one buffered packet from each open variant
668 if (var->needed && !var->pkt.data) {
672 ret = av_read_frame(var->ctx, &var->pkt);
674 if (!url_feof(&var->pb) && ret != AVERROR_EOF)
676 reset_packet(&var->pkt);
679 if (c->first_timestamp == AV_NOPTS_VALUE)
680 c->first_timestamp = var->pkt.dts;
683 if (c->seek_timestamp == AV_NOPTS_VALUE)
686 if (var->pkt.dts == AV_NOPTS_VALUE) {
687 c->seek_timestamp = AV_NOPTS_VALUE;
691 st = var->ctx->streams[var->pkt.stream_index];
692 ts_diff = av_rescale_rnd(var->pkt.dts, AV_TIME_BASE,
693 st->time_base.den, AV_ROUND_DOWN) -
695 if (ts_diff >= 0 && (c->seek_flags & AVSEEK_FLAG_ANY ||
696 var->pkt.flags & AV_PKT_FLAG_KEY)) {
697 c->seek_timestamp = AV_NOPTS_VALUE;
702 /* Check if this stream has the packet with the lowest dts */
707 struct variant *minvar = c->variants[minvariant];
708 int64_t dts = var->pkt.dts;
709 int64_t mindts = minvar->pkt.dts;
710 AVStream *st = var->ctx->streams[ var->pkt.stream_index];
711 AVStream *minst= minvar->ctx->streams[minvar->pkt.stream_index];
713 if( st->start_time != AV_NOPTS_VALUE) dts -= st->start_time;
714 if(minst->start_time != AV_NOPTS_VALUE) mindts -= minst->start_time;
716 if (av_compare_ts(dts, st->time_base, mindts, minst->time_base) < 0)
721 if (c->end_of_segment) {
722 if (recheck_discard_flags(s, 0))
725 /* If we got a packet, return it */
726 if (minvariant >= 0) {
727 *pkt = c->variants[minvariant]->pkt;
728 pkt->stream_index += c->variants[minvariant]->stream_offset;
729 reset_packet(&c->variants[minvariant]->pkt);
735 static int hls_close(AVFormatContext *s)
737 HLSContext *c = s->priv_data;
739 free_variant_list(c);
743 static int hls_read_seek(AVFormatContext *s, int stream_index,
744 int64_t timestamp, int flags)
746 HLSContext *c = s->priv_data;
749 if ((flags & AVSEEK_FLAG_BYTE) || !c->variants[0]->finished)
750 return AVERROR(ENOSYS);
752 c->seek_flags = flags;
753 c->seek_timestamp = stream_index < 0 ? timestamp :
754 av_rescale_rnd(timestamp, AV_TIME_BASE,
755 s->streams[stream_index]->time_base.den,
756 flags & AVSEEK_FLAG_BACKWARD ?
757 AV_ROUND_DOWN : AV_ROUND_UP);
758 timestamp = av_rescale_rnd(timestamp, 1, stream_index >= 0 ?
759 s->streams[stream_index]->time_base.den :
760 AV_TIME_BASE, flags & AVSEEK_FLAG_BACKWARD ?
761 AV_ROUND_DOWN : AV_ROUND_UP);
762 if (s->duration < c->seek_timestamp) {
763 c->seek_timestamp = AV_NOPTS_VALUE;
768 for (i = 0; i < c->n_variants; i++) {
770 struct variant *var = c->variants[i];
771 int64_t pos = c->first_timestamp == AV_NOPTS_VALUE ? 0 :
772 av_rescale_rnd(c->first_timestamp, 1, stream_index >= 0 ?
773 s->streams[stream_index]->time_base.den :
774 AV_TIME_BASE, flags & AVSEEK_FLAG_BACKWARD ?
775 AV_ROUND_DOWN : AV_ROUND_UP);
777 ffurl_close(var->input);
780 av_free_packet(&var->pkt);
781 reset_packet(&var->pkt);
782 var->pb.eof_reached = 0;
783 /* Clear any buffered data */
784 var->pb.buf_end = var->pb.buf_ptr = var->pb.buffer;
785 /* Reset the pos, to let the mpegts demuxer know we've seeked. */
788 /* Locate the segment that contains the target timestamp */
789 for (j = 0; j < var->n_segments; j++) {
790 if (timestamp >= pos &&
791 timestamp < pos + var->segments[j]->duration) {
792 var->cur_seq_no = var->start_seq_no + j;
796 pos += var->segments[j]->duration;
799 c->seek_timestamp = AV_NOPTS_VALUE;
804 static int hls_probe(AVProbeData *p)
806 /* Require #EXTM3U at the start, and either one of the ones below
807 * somewhere for a proper match. */
808 if (strncmp(p->buf, "#EXTM3U", 7))
810 if (strstr(p->buf, "#EXT-X-STREAM-INF:") ||
811 strstr(p->buf, "#EXT-X-TARGETDURATION:") ||
812 strstr(p->buf, "#EXT-X-MEDIA-SEQUENCE:"))
813 return AVPROBE_SCORE_MAX;
817 AVInputFormat ff_hls_demuxer = {
818 .name = "hls,applehttp",
819 .long_name = NULL_IF_CONFIG_SMALL("Apple HTTP Live Streaming"),
820 .priv_data_size = sizeof(HLSContext),
821 .read_probe = hls_probe,
822 .read_header = hls_read_header,
823 .read_packet = hls_read_packet,
824 .read_close = hls_close,
825 .read_seek = hls_read_seek,