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;
84 int64_t target_duration;
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 && av_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, is_segment = 0, is_variant = 0, bandwidth = 0;
210 int64_t duration = 0;
211 enum KeyType key_type = KEY_NONE;
214 char key[MAX_URL_SIZE] = "";
215 char line[MAX_URL_SIZE];
220 AVDictionary *opts = NULL;
222 /* Some HLS servers don't like being sent the range header */
223 av_dict_set(&opts, "seekable", "0", 0);
225 // broker prior HTTP options that should be consistent across requests
226 av_dict_set(&opts, "user-agent", c->user_agent, 0);
227 av_dict_set(&opts, "cookies", c->cookies, 0);
229 ret = avio_open2(&in, url, AVIO_FLAG_READ,
230 c->interrupt_callback, &opts);
236 read_chomp_line(in, line, sizeof(line));
237 if (strcmp(line, "#EXTM3U")) {
238 ret = AVERROR_INVALIDDATA;
243 free_segment_list(var);
246 while (!url_feof(in)) {
247 read_chomp_line(in, line, sizeof(line));
248 if (av_strstart(line, "#EXT-X-STREAM-INF:", &ptr)) {
249 struct variant_info info = {{0}};
251 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_variant_args,
253 bandwidth = atoi(info.bandwidth);
254 } else if (av_strstart(line, "#EXT-X-KEY:", &ptr)) {
255 struct key_info info = {{0}};
256 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_key_args,
260 if (!strcmp(info.method, "AES-128"))
261 key_type = KEY_AES_128;
262 if (!strncmp(info.iv, "0x", 2) || !strncmp(info.iv, "0X", 2)) {
263 ff_hex_to_data(iv, info.iv + 2);
266 av_strlcpy(key, info.uri, sizeof(key));
267 } else if (av_strstart(line, "#EXT-X-TARGETDURATION:", &ptr)) {
269 var = new_variant(c, 0, url, NULL);
271 ret = AVERROR(ENOMEM);
275 var->target_duration = atoi(ptr) * AV_TIME_BASE;
276 } else if (av_strstart(line, "#EXT-X-MEDIA-SEQUENCE:", &ptr)) {
278 var = new_variant(c, 0, url, NULL);
280 ret = AVERROR(ENOMEM);
284 var->start_seq_no = atoi(ptr);
285 } else if (av_strstart(line, "#EXT-X-ENDLIST", &ptr)) {
288 } else if (av_strstart(line, "#EXTINF:", &ptr)) {
290 duration = atof(ptr) * AV_TIME_BASE;
291 } else if (av_strstart(line, "#", NULL)) {
293 } else if (line[0]) {
295 if (!new_variant(c, bandwidth, line, url)) {
296 ret = AVERROR(ENOMEM);
305 var = new_variant(c, 0, url, NULL);
307 ret = AVERROR(ENOMEM);
311 seg = av_malloc(sizeof(struct segment));
313 ret = AVERROR(ENOMEM);
316 seg->duration = duration;
317 seg->key_type = key_type;
319 memcpy(seg->iv, iv, sizeof(iv));
321 int seq = var->start_seq_no + var->n_segments;
322 memset(seg->iv, 0, sizeof(seg->iv));
323 AV_WB32(seg->iv + 12, seq);
325 ff_make_absolute_url(seg->key, sizeof(seg->key), url, key);
326 ff_make_absolute_url(seg->url, sizeof(seg->url), url, line);
327 dynarray_add(&var->segments, &var->n_segments, seg);
333 var->last_load_time = av_gettime();
341 static int open_input(HLSContext *c, struct variant *var)
343 AVDictionary *opts = NULL;
345 struct segment *seg = var->segments[var->cur_seq_no - var->start_seq_no];
347 // broker prior HTTP options that should be consistent across requests
348 av_dict_set(&opts, "user-agent", c->user_agent, 0);
349 av_dict_set(&opts, "cookies", c->cookies, 0);
350 av_dict_set(&opts, "seekable", "0", 0);
352 if (seg->key_type == KEY_NONE) {
353 ret = ffurl_open(&var->input, seg->url, AVIO_FLAG_READ,
354 &var->parent->interrupt_callback, &opts);
356 } else if (seg->key_type == KEY_AES_128) {
357 char iv[33], key[33], url[MAX_URL_SIZE];
358 if (strcmp(seg->key, var->key_url)) {
360 if (ffurl_open(&uc, seg->key, AVIO_FLAG_READ,
361 &var->parent->interrupt_callback, &opts) == 0) {
362 if (ffurl_read_complete(uc, var->key, sizeof(var->key))
363 != sizeof(var->key)) {
364 av_log(NULL, AV_LOG_ERROR, "Unable to read key file %s\n",
369 av_log(NULL, AV_LOG_ERROR, "Unable to open key file %s\n",
372 av_strlcpy(var->key_url, seg->key, sizeof(var->key_url));
374 ff_data_to_hex(iv, seg->iv, sizeof(seg->iv), 0);
375 ff_data_to_hex(key, var->key, sizeof(var->key), 0);
376 iv[32] = key[32] = '\0';
377 if (strstr(seg->url, "://"))
378 snprintf(url, sizeof(url), "crypto+%s", seg->url);
380 snprintf(url, sizeof(url), "crypto:%s", seg->url);
381 if ((ret = ffurl_alloc(&var->input, url, AVIO_FLAG_READ,
382 &var->parent->interrupt_callback)) < 0)
384 av_opt_set(var->input->priv_data, "key", key, 0);
385 av_opt_set(var->input->priv_data, "iv", iv, 0);
386 /* Need to repopulate options */
388 av_dict_set(&opts, "seekable", "0", 0);
389 if ((ret = ffurl_connect(var->input, &opts)) < 0) {
390 ffurl_close(var->input);
397 ret = AVERROR(ENOSYS);
404 static int read_data(void *opaque, uint8_t *buf, int buf_size)
406 struct variant *v = opaque;
407 HLSContext *c = v->parent->priv_data;
412 /* If this is a live stream and the reload interval has elapsed since
413 * the last playlist reload, reload the variant playlists now. */
414 int64_t reload_interval = v->n_segments > 0 ?
415 v->segments[v->n_segments - 1]->duration :
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 / 2;
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 &&
461 v->parent->nb_streams >= v->stream_offset + v->ctx->nb_streams) {
463 for (i = v->stream_offset; i < v->stream_offset + v->ctx->nb_streams;
465 if (v->parent->streams[i]->discard < AVDISCARD_ALL)
470 av_log(v->parent, AV_LOG_INFO, "No longer receiving variant %d\n",
477 static int hls_read_header(AVFormatContext *s)
479 URLContext *u = (s->flags & AVFMT_FLAG_CUSTOM_IO) ? NULL : s->pb->opaque;
480 HLSContext *c = s->priv_data;
481 int ret = 0, i, j, stream_offset = 0;
483 c->interrupt_callback = &s->interrupt_callback;
485 // if the URL context is good, read important options we must broker later
486 if (u && u->prot->priv_data_class) {
487 // get the previous user agent & set back to null if string size is zero
488 av_freep(&c->user_agent);
489 av_opt_get(u->priv_data, "user-agent", 0, (uint8_t**)&(c->user_agent));
490 if (c->user_agent && !strlen(c->user_agent))
491 av_freep(&c->user_agent);
493 // get the previous cookies & set back to null if string size is zero
494 av_freep(&c->cookies);
495 av_opt_get(u->priv_data, "cookies", 0, (uint8_t**)&(c->cookies));
496 if (c->cookies && !strlen(c->cookies))
497 av_freep(&c->cookies);
500 if ((ret = parse_playlist(c, s->filename, NULL, s->pb)) < 0)
503 if (c->n_variants == 0) {
504 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
508 /* If the playlist only contained variants, parse each individual
509 * variant playlist. */
510 if (c->n_variants > 1 || c->variants[0]->n_segments == 0) {
511 for (i = 0; i < c->n_variants; i++) {
512 struct variant *v = c->variants[i];
513 if ((ret = parse_playlist(c, v->url, v, NULL)) < 0)
518 if (c->variants[0]->n_segments == 0) {
519 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
524 /* If this isn't a live stream, calculate the total duration of the
526 if (c->variants[0]->finished) {
527 int64_t duration = 0;
528 for (i = 0; i < c->variants[0]->n_segments; i++)
529 duration += c->variants[0]->segments[i]->duration;
530 s->duration = duration;
533 /* Open the demuxer for each variant */
534 for (i = 0; i < c->n_variants; i++) {
535 struct variant *v = c->variants[i];
536 AVInputFormat *in_fmt = NULL;
537 char bitrate_str[20];
540 if (v->n_segments == 0)
543 if (!(v->ctx = avformat_alloc_context())) {
544 ret = AVERROR(ENOMEM);
552 /* If this is a live stream with more than 3 segments, start at the
553 * third last segment. */
554 v->cur_seq_no = v->start_seq_no;
555 if (!v->finished && v->n_segments > 3)
556 v->cur_seq_no = v->start_seq_no + v->n_segments - 3;
558 v->read_buffer = av_malloc(INITIAL_BUFFER_SIZE);
559 ffio_init_context(&v->pb, v->read_buffer, INITIAL_BUFFER_SIZE, 0, v,
560 read_data, NULL, NULL);
562 ret = av_probe_input_buffer(&v->pb, &in_fmt, v->segments[0]->url,
565 /* Free the ctx - it isn't initialized properly at this point,
566 * so avformat_close_input shouldn't be called. If
567 * avformat_open_input fails below, it frees and zeros the
568 * context, so it doesn't need any special treatment like this. */
569 av_log(s, AV_LOG_ERROR, "Error when loading first segment '%s'\n", v->segments[0]->url);
570 avformat_free_context(v->ctx);
575 v->stream_offset = stream_offset;
576 ret = avformat_open_input(&v->ctx, v->segments[0]->url, in_fmt, NULL);
580 v->ctx->ctx_flags &= ~AVFMTCTX_NOHEADER;
581 ret = avformat_find_stream_info(v->ctx, NULL);
584 snprintf(bitrate_str, sizeof(bitrate_str), "%d", v->bandwidth);
586 program = av_new_program(s, i);
589 av_dict_set(&program->metadata, "variant_bitrate", bitrate_str, 0);
591 /* Create new AVStreams for each stream in this variant */
592 for (j = 0; j < v->ctx->nb_streams; j++) {
593 AVStream *st = avformat_new_stream(s, NULL);
594 AVStream *ist = v->ctx->streams[j];
596 ret = AVERROR(ENOMEM);
599 ff_program_add_stream_index(s, i, stream_offset + j);
601 avpriv_set_pts_info(st, ist->pts_wrap_bits, ist->time_base.num, ist->time_base.den);
602 avcodec_copy_context(st->codec, v->ctx->streams[j]->codec);
604 av_dict_set(&st->metadata, "variant_bitrate", bitrate_str,
607 stream_offset += v->ctx->nb_streams;
611 c->first_timestamp = AV_NOPTS_VALUE;
612 c->seek_timestamp = AV_NOPTS_VALUE;
616 free_variant_list(c);
620 static int recheck_discard_flags(AVFormatContext *s, int first)
622 HLSContext *c = s->priv_data;
625 /* Check if any new streams are needed */
626 for (i = 0; i < c->n_variants; i++)
627 c->variants[i]->cur_needed = 0;
629 for (i = 0; i < s->nb_streams; i++) {
630 AVStream *st = s->streams[i];
631 struct variant *var = c->variants[s->streams[i]->id];
632 if (st->discard < AVDISCARD_ALL)
635 for (i = 0; i < c->n_variants; i++) {
636 struct variant *v = c->variants[i];
637 if (v->cur_needed && !v->needed) {
640 v->cur_seq_no = c->cur_seq_no;
641 v->pb.eof_reached = 0;
642 av_log(s, AV_LOG_INFO, "Now receiving variant %d\n", i);
643 } else if (first && !v->cur_needed && v->needed) {
645 ffurl_close(v->input);
649 av_log(s, AV_LOG_INFO, "No longer receiving variant %d\n", i);
655 static int hls_read_packet(AVFormatContext *s, AVPacket *pkt)
657 HLSContext *c = s->priv_data;
658 int ret, i, minvariant = -1;
660 if (c->first_packet) {
661 recheck_discard_flags(s, 1);
666 c->end_of_segment = 0;
667 for (i = 0; i < c->n_variants; i++) {
668 struct variant *var = c->variants[i];
669 /* Make sure we've got one buffered packet from each open variant
671 if (var->needed && !var->pkt.data) {
675 ret = av_read_frame(var->ctx, &var->pkt);
677 if (!url_feof(&var->pb) && ret != AVERROR_EOF)
679 reset_packet(&var->pkt);
682 if (c->first_timestamp == AV_NOPTS_VALUE &&
683 var->pkt.dts != AV_NOPTS_VALUE)
684 c->first_timestamp = av_rescale_q(var->pkt.dts,
685 var->ctx->streams[var->pkt.stream_index]->time_base,
689 if (c->seek_timestamp == AV_NOPTS_VALUE)
692 if (var->pkt.dts == AV_NOPTS_VALUE) {
693 c->seek_timestamp = AV_NOPTS_VALUE;
697 st = var->ctx->streams[var->pkt.stream_index];
698 ts_diff = av_rescale_rnd(var->pkt.dts, AV_TIME_BASE,
699 st->time_base.den, AV_ROUND_DOWN) -
701 if (ts_diff >= 0 && (c->seek_flags & AVSEEK_FLAG_ANY ||
702 var->pkt.flags & AV_PKT_FLAG_KEY)) {
703 c->seek_timestamp = AV_NOPTS_VALUE;
706 av_free_packet(&var->pkt);
707 reset_packet(&var->pkt);
710 /* Check if this stream still is on an earlier segment number, or
711 * has the packet with the lowest dts */
713 struct variant *minvar = c->variants[minvariant];
714 if (minvariant < 0 || var->cur_seq_no < minvar->cur_seq_no) {
716 } else if (var->cur_seq_no == minvar->cur_seq_no) {
717 int64_t dts = var->pkt.dts;
718 int64_t mindts = minvar->pkt.dts;
719 AVStream *st = var->ctx->streams[var->pkt.stream_index];
720 AVStream *minst = minvar->ctx->streams[minvar->pkt.stream_index];
722 if (dts == AV_NOPTS_VALUE) {
724 } else if (mindts != AV_NOPTS_VALUE) {
725 if (st->start_time != AV_NOPTS_VALUE)
726 dts -= st->start_time;
727 if (minst->start_time != AV_NOPTS_VALUE)
728 mindts -= minst->start_time;
730 if (av_compare_ts(dts, st->time_base,
731 mindts, minst->time_base) < 0)
737 if (c->end_of_segment) {
738 if (recheck_discard_flags(s, 0))
741 /* If we got a packet, return it */
742 if (minvariant >= 0) {
743 *pkt = c->variants[minvariant]->pkt;
744 pkt->stream_index += c->variants[minvariant]->stream_offset;
745 reset_packet(&c->variants[minvariant]->pkt);
751 static int hls_close(AVFormatContext *s)
753 HLSContext *c = s->priv_data;
755 free_variant_list(c);
759 static int hls_read_seek(AVFormatContext *s, int stream_index,
760 int64_t timestamp, int flags)
762 HLSContext *c = s->priv_data;
765 if ((flags & AVSEEK_FLAG_BYTE) || !c->variants[0]->finished)
766 return AVERROR(ENOSYS);
768 c->seek_flags = flags;
769 c->seek_timestamp = stream_index < 0 ? timestamp :
770 av_rescale_rnd(timestamp, AV_TIME_BASE,
771 s->streams[stream_index]->time_base.den,
772 flags & AVSEEK_FLAG_BACKWARD ?
773 AV_ROUND_DOWN : AV_ROUND_UP);
774 timestamp = av_rescale_rnd(timestamp, AV_TIME_BASE, stream_index >= 0 ?
775 s->streams[stream_index]->time_base.den :
776 AV_TIME_BASE, flags & AVSEEK_FLAG_BACKWARD ?
777 AV_ROUND_DOWN : AV_ROUND_UP);
778 if (s->duration < c->seek_timestamp) {
779 c->seek_timestamp = AV_NOPTS_VALUE;
784 for (i = 0; i < c->n_variants; i++) {
786 struct variant *var = c->variants[i];
787 int64_t pos = c->first_timestamp == AV_NOPTS_VALUE ?
788 0 : c->first_timestamp;
790 ffurl_close(var->input);
793 av_free_packet(&var->pkt);
794 reset_packet(&var->pkt);
795 var->pb.eof_reached = 0;
796 /* Clear any buffered data */
797 var->pb.buf_end = var->pb.buf_ptr = var->pb.buffer;
798 /* Reset the pos, to let the mpegts demuxer know we've seeked. */
801 /* Locate the segment that contains the target timestamp */
802 for (j = 0; j < var->n_segments; j++) {
803 if (timestamp >= pos &&
804 timestamp < pos + var->segments[j]->duration) {
805 var->cur_seq_no = var->start_seq_no + j;
809 pos += var->segments[j]->duration;
812 c->seek_timestamp = AV_NOPTS_VALUE;
817 static int hls_probe(AVProbeData *p)
819 /* Require #EXTM3U at the start, and either one of the ones below
820 * somewhere for a proper match. */
821 if (strncmp(p->buf, "#EXTM3U", 7))
823 if (strstr(p->buf, "#EXT-X-STREAM-INF:") ||
824 strstr(p->buf, "#EXT-X-TARGETDURATION:") ||
825 strstr(p->buf, "#EXT-X-MEDIA-SEQUENCE:"))
826 return AVPROBE_SCORE_MAX;
830 AVInputFormat ff_hls_demuxer = {
831 .name = "hls,applehttp",
832 .long_name = NULL_IF_CONFIG_SMALL("Apple HTTP Live Streaming"),
833 .priv_data_size = sizeof(HLSContext),
834 .read_probe = hls_probe,
835 .read_header = hls_read_header,
836 .read_packet = hls_read_packet,
837 .read_close = hls_close,
838 .read_seek = hls_read_seek,