2 * Apple HTTP Live Streaming demuxer
3 * Copyright (c) 2010 Martin Storsjo
5 * This file is part of Libav.
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.
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.
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
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;
108 static int read_chomp_line(AVIOContext *s, char *buf, int maxlen)
110 int len = ff_get_line(s, buf, maxlen);
111 while (len > 0 && av_isspace(buf[len - 1]))
116 static void free_segment_list(struct variant *var)
119 for (i = 0; i < var->n_segments; i++)
120 av_free(var->segments[i]);
121 av_freep(&var->segments);
125 static void free_variant_list(HLSContext *c)
128 for (i = 0; i < c->n_variants; i++) {
129 struct variant *var = c->variants[i];
130 free_segment_list(var);
131 av_free_packet(&var->pkt);
132 av_free(var->pb.buffer);
134 ffurl_close(var->input);
137 avformat_close_input(&var->ctx);
141 av_freep(&c->variants);
146 * Used to reset a statically allocated AVPacket to a clean slate,
147 * containing no data.
149 static void reset_packet(AVPacket *pkt)
155 static struct variant *new_variant(HLSContext *c, int bandwidth,
156 const char *url, const char *base)
158 struct variant *var = av_mallocz(sizeof(struct variant));
161 reset_packet(&var->pkt);
162 var->bandwidth = bandwidth;
163 ff_make_absolute_url(var->url, sizeof(var->url), base, url);
164 dynarray_add(&c->variants, &c->n_variants, var);
168 struct variant_info {
172 static void handle_variant_args(struct variant_info *info, const char *key,
173 int key_len, char **dest, int *dest_len)
175 if (!strncmp(key, "BANDWIDTH=", key_len)) {
176 *dest = info->bandwidth;
177 *dest_len = sizeof(info->bandwidth);
182 char uri[MAX_URL_SIZE];
187 static void handle_key_args(struct key_info *info, const char *key,
188 int key_len, char **dest, int *dest_len)
190 if (!strncmp(key, "METHOD=", key_len)) {
191 *dest = info->method;
192 *dest_len = sizeof(info->method);
193 } else if (!strncmp(key, "URI=", key_len)) {
195 *dest_len = sizeof(info->uri);
196 } else if (!strncmp(key, "IV=", key_len)) {
198 *dest_len = sizeof(info->iv);
202 static int parse_playlist(HLSContext *c, const char *url,
203 struct variant *var, AVIOContext *in)
205 int ret = 0, is_segment = 0, is_variant = 0, bandwidth = 0;
206 int64_t duration = 0;
207 enum KeyType key_type = KEY_NONE;
210 char key[MAX_URL_SIZE] = "";
217 if ((ret = avio_open2(&in, url, AVIO_FLAG_READ,
218 c->interrupt_callback, NULL)) < 0)
222 read_chomp_line(in, line, sizeof(line));
223 if (strcmp(line, "#EXTM3U")) {
224 ret = AVERROR_INVALIDDATA;
229 free_segment_list(var);
232 while (!in->eof_reached) {
233 read_chomp_line(in, line, sizeof(line));
234 if (av_strstart(line, "#EXT-X-STREAM-INF:", &ptr)) {
235 struct variant_info info = {{0}};
237 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_variant_args,
239 bandwidth = atoi(info.bandwidth);
240 } else if (av_strstart(line, "#EXT-X-KEY:", &ptr)) {
241 struct key_info info = {{0}};
242 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_key_args,
246 if (!strcmp(info.method, "AES-128"))
247 key_type = KEY_AES_128;
248 if (!strncmp(info.iv, "0x", 2) || !strncmp(info.iv, "0X", 2)) {
249 ff_hex_to_data(iv, info.iv + 2);
252 av_strlcpy(key, info.uri, sizeof(key));
253 } else if (av_strstart(line, "#EXT-X-TARGETDURATION:", &ptr)) {
255 var = new_variant(c, 0, url, NULL);
257 ret = AVERROR(ENOMEM);
261 var->target_duration = atoi(ptr) * AV_TIME_BASE;
262 } else if (av_strstart(line, "#EXT-X-MEDIA-SEQUENCE:", &ptr)) {
264 var = new_variant(c, 0, url, NULL);
266 ret = AVERROR(ENOMEM);
270 var->start_seq_no = atoi(ptr);
271 } else if (av_strstart(line, "#EXT-X-ENDLIST", &ptr)) {
274 } else if (av_strstart(line, "#EXTINF:", &ptr)) {
276 duration = atof(ptr) * AV_TIME_BASE;
277 } else if (av_strstart(line, "#", NULL)) {
279 } else if (line[0]) {
281 if (!new_variant(c, bandwidth, line, url)) {
282 ret = AVERROR(ENOMEM);
291 var = new_variant(c, 0, url, NULL);
293 ret = AVERROR(ENOMEM);
297 seg = av_malloc(sizeof(struct segment));
299 ret = AVERROR(ENOMEM);
302 seg->duration = duration;
303 seg->key_type = key_type;
305 memcpy(seg->iv, iv, sizeof(iv));
307 int seq = var->start_seq_no + var->n_segments;
308 memset(seg->iv, 0, sizeof(seg->iv));
309 AV_WB32(seg->iv + 12, seq);
311 ff_make_absolute_url(seg->key, sizeof(seg->key), url, key);
312 ff_make_absolute_url(seg->url, sizeof(seg->url), url, line);
313 dynarray_add(&var->segments, &var->n_segments, seg);
319 var->last_load_time = av_gettime();
327 static int open_input(struct variant *var)
329 struct segment *seg = var->segments[var->cur_seq_no - var->start_seq_no];
330 if (seg->key_type == KEY_NONE) {
331 return ffurl_open(&var->input, seg->url, AVIO_FLAG_READ,
332 &var->parent->interrupt_callback, NULL);
333 } else if (seg->key_type == KEY_AES_128) {
334 char iv[33], key[33], url[MAX_URL_SIZE];
336 if (strcmp(seg->key, var->key_url)) {
338 if (ffurl_open(&uc, seg->key, AVIO_FLAG_READ,
339 &var->parent->interrupt_callback, NULL) == 0) {
340 if (ffurl_read_complete(uc, var->key, sizeof(var->key))
341 != sizeof(var->key)) {
342 av_log(NULL, AV_LOG_ERROR, "Unable to read key file %s\n",
347 av_log(NULL, AV_LOG_ERROR, "Unable to open key file %s\n",
350 av_strlcpy(var->key_url, seg->key, sizeof(var->key_url));
352 ff_data_to_hex(iv, seg->iv, sizeof(seg->iv), 0);
353 ff_data_to_hex(key, var->key, sizeof(var->key), 0);
354 iv[32] = key[32] = '\0';
355 if (strstr(seg->url, "://"))
356 snprintf(url, sizeof(url), "crypto+%s", seg->url);
358 snprintf(url, sizeof(url), "crypto:%s", seg->url);
359 if ((ret = ffurl_alloc(&var->input, url, AVIO_FLAG_READ,
360 &var->parent->interrupt_callback)) < 0)
362 av_opt_set(var->input->priv_data, "key", key, 0);
363 av_opt_set(var->input->priv_data, "iv", iv, 0);
364 if ((ret = ffurl_connect(var->input, NULL)) < 0) {
365 ffurl_close(var->input);
371 return AVERROR(ENOSYS);
374 static int read_data(void *opaque, uint8_t *buf, int buf_size)
376 struct variant *v = opaque;
377 HLSContext *c = v->parent->priv_data;
382 /* If this is a live stream and the reload interval has elapsed since
383 * the last playlist reload, reload the variant playlists now. */
384 int64_t reload_interval = v->n_segments > 0 ?
385 v->segments[v->n_segments - 1]->duration :
390 av_gettime() - v->last_load_time >= reload_interval) {
391 if ((ret = parse_playlist(c, v->url, v, NULL)) < 0)
393 /* If we need to reload the playlist again below (if
394 * there's still no more segments), switch to a reload
395 * interval of half the target duration. */
396 reload_interval = v->target_duration / 2;
398 if (v->cur_seq_no < v->start_seq_no) {
399 av_log(NULL, AV_LOG_WARNING,
400 "skipping %d segments ahead, expired from playlists\n",
401 v->start_seq_no - v->cur_seq_no);
402 v->cur_seq_no = v->start_seq_no;
404 if (v->cur_seq_no >= v->start_seq_no + v->n_segments) {
407 while (av_gettime() - v->last_load_time < reload_interval) {
408 if (ff_check_interrupt(c->interrupt_callback))
412 /* Enough time has elapsed since the last reload */
420 ret = ffurl_read(v->input, buf, buf_size);
423 ffurl_close(v->input);
427 c->end_of_segment = 1;
428 c->cur_seq_no = v->cur_seq_no;
430 if (v->ctx && v->ctx->nb_streams &&
431 v->parent->nb_streams >= v->stream_offset + v->ctx->nb_streams) {
433 for (i = v->stream_offset; i < v->stream_offset + v->ctx->nb_streams;
435 if (v->parent->streams[i]->discard < AVDISCARD_ALL)
440 av_log(v->parent, AV_LOG_INFO, "No longer receiving variant %d\n",
447 static int hls_read_header(AVFormatContext *s)
449 HLSContext *c = s->priv_data;
450 int ret = 0, i, j, stream_offset = 0;
452 c->interrupt_callback = &s->interrupt_callback;
454 if ((ret = parse_playlist(c, s->filename, NULL, s->pb)) < 0)
457 if (c->n_variants == 0) {
458 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
462 /* If the playlist only contained variants, parse each individual
463 * variant playlist. */
464 if (c->n_variants > 1 || c->variants[0]->n_segments == 0) {
465 for (i = 0; i < c->n_variants; i++) {
466 struct variant *v = c->variants[i];
467 if ((ret = parse_playlist(c, v->url, v, NULL)) < 0)
472 if (c->variants[0]->n_segments == 0) {
473 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
478 /* If this isn't a live stream, calculate the total duration of the
480 if (c->variants[0]->finished) {
481 int64_t duration = 0;
482 for (i = 0; i < c->variants[0]->n_segments; i++)
483 duration += c->variants[0]->segments[i]->duration;
484 s->duration = duration;
487 /* Open the demuxer for each variant */
488 for (i = 0; i < c->n_variants; i++) {
489 struct variant *v = c->variants[i];
490 AVInputFormat *in_fmt = NULL;
491 char bitrate_str[20];
494 if (v->n_segments == 0)
497 if (!(v->ctx = avformat_alloc_context())) {
498 ret = AVERROR(ENOMEM);
506 /* If this is a live stream with more than 3 segments, start at the
507 * third last segment. */
508 v->cur_seq_no = v->start_seq_no;
509 if (!v->finished && v->n_segments > 3)
510 v->cur_seq_no = v->start_seq_no + v->n_segments - 3;
512 v->read_buffer = av_malloc(INITIAL_BUFFER_SIZE);
513 ffio_init_context(&v->pb, v->read_buffer, INITIAL_BUFFER_SIZE, 0, v,
514 read_data, NULL, NULL);
516 ret = av_probe_input_buffer(&v->pb, &in_fmt, v->segments[0]->url,
519 /* Free the ctx - it isn't initialized properly at this point,
520 * so avformat_close_input shouldn't be called. If
521 * avformat_open_input fails below, it frees and zeros the
522 * context, so it doesn't need any special treatment like this. */
523 avformat_free_context(v->ctx);
528 v->stream_offset = stream_offset;
529 ret = avformat_open_input(&v->ctx, v->segments[0]->url, in_fmt, NULL);
533 v->ctx->ctx_flags &= ~AVFMTCTX_NOHEADER;
534 ret = avformat_find_stream_info(v->ctx, NULL);
537 snprintf(bitrate_str, sizeof(bitrate_str), "%d", v->bandwidth);
539 program = av_new_program(s, i);
542 av_dict_set(&program->metadata, "variant_bitrate", bitrate_str, 0);
544 /* Create new AVStreams for each stream in this variant */
545 for (j = 0; j < v->ctx->nb_streams; j++) {
546 AVStream *st = avformat_new_stream(s, NULL);
547 AVStream *ist = v->ctx->streams[j];
549 ret = AVERROR(ENOMEM);
552 ff_program_add_stream_index(s, i, stream_offset + j);
554 avpriv_set_pts_info(st, ist->pts_wrap_bits, ist->time_base.num, ist->time_base.den);
555 avcodec_copy_context(st->codec, v->ctx->streams[j]->codec);
557 av_dict_set(&st->metadata, "variant_bitrate", bitrate_str,
560 stream_offset += v->ctx->nb_streams;
564 c->first_timestamp = AV_NOPTS_VALUE;
565 c->seek_timestamp = AV_NOPTS_VALUE;
569 free_variant_list(c);
573 static int recheck_discard_flags(AVFormatContext *s, int first)
575 HLSContext *c = s->priv_data;
578 /* Check if any new streams are needed */
579 for (i = 0; i < c->n_variants; i++)
580 c->variants[i]->cur_needed = 0;;
582 for (i = 0; i < s->nb_streams; i++) {
583 AVStream *st = s->streams[i];
584 struct variant *var = c->variants[s->streams[i]->id];
585 if (st->discard < AVDISCARD_ALL)
588 for (i = 0; i < c->n_variants; i++) {
589 struct variant *v = c->variants[i];
590 if (v->cur_needed && !v->needed) {
593 v->cur_seq_no = c->cur_seq_no;
594 v->pb.eof_reached = 0;
595 av_log(s, AV_LOG_INFO, "Now receiving variant %d\n", i);
596 } else if (first && !v->cur_needed && v->needed) {
598 ffurl_close(v->input);
602 av_log(s, AV_LOG_INFO, "No longer receiving variant %d\n", i);
608 static int hls_read_packet(AVFormatContext *s, AVPacket *pkt)
610 HLSContext *c = s->priv_data;
611 int ret, i, minvariant = -1;
613 if (c->first_packet) {
614 recheck_discard_flags(s, 1);
619 c->end_of_segment = 0;
620 for (i = 0; i < c->n_variants; i++) {
621 struct variant *var = c->variants[i];
622 /* Make sure we've got one buffered packet from each open variant
624 if (var->needed && !var->pkt.data) {
628 ret = av_read_frame(var->ctx, &var->pkt);
630 if (!var->pb.eof_reached)
632 reset_packet(&var->pkt);
635 if (c->first_timestamp == AV_NOPTS_VALUE &&
636 var->pkt.dts != AV_NOPTS_VALUE)
637 c->first_timestamp = av_rescale_q(var->pkt.dts,
638 var->ctx->streams[var->pkt.stream_index]->time_base,
642 if (c->seek_timestamp == AV_NOPTS_VALUE)
645 if (var->pkt.dts == AV_NOPTS_VALUE) {
646 c->seek_timestamp = AV_NOPTS_VALUE;
650 st = var->ctx->streams[var->pkt.stream_index];
651 ts_diff = av_rescale_rnd(var->pkt.dts, AV_TIME_BASE,
652 st->time_base.den, AV_ROUND_DOWN) -
654 if (ts_diff >= 0 && (c->seek_flags & AVSEEK_FLAG_ANY ||
655 var->pkt.flags & AV_PKT_FLAG_KEY)) {
656 c->seek_timestamp = AV_NOPTS_VALUE;
659 av_free_packet(&var->pkt);
660 reset_packet(&var->pkt);
663 /* Check if this stream still is on an earlier segment number, or
664 * has the packet with the lowest dts */
666 struct variant *minvar = c->variants[minvariant];
667 if (minvariant < 0 || var->cur_seq_no < minvar->cur_seq_no) {
669 } else if (var->cur_seq_no == minvar->cur_seq_no) {
670 int64_t dts = var->pkt.dts;
671 int64_t mindts = minvar->pkt.dts;
672 AVStream *st = var->ctx->streams[var->pkt.stream_index];
673 AVStream *minst = minvar->ctx->streams[minvar->pkt.stream_index];
675 if (dts == AV_NOPTS_VALUE) {
677 } else if (mindts != AV_NOPTS_VALUE) {
678 if (st->start_time != AV_NOPTS_VALUE)
679 dts -= st->start_time;
680 if (minst->start_time != AV_NOPTS_VALUE)
681 mindts -= minst->start_time;
683 if (av_compare_ts(dts, st->time_base,
684 mindts, minst->time_base) < 0)
690 if (c->end_of_segment) {
691 if (recheck_discard_flags(s, 0))
694 /* If we got a packet, return it */
695 if (minvariant >= 0) {
696 *pkt = c->variants[minvariant]->pkt;
697 pkt->stream_index += c->variants[minvariant]->stream_offset;
698 reset_packet(&c->variants[minvariant]->pkt);
704 static int hls_close(AVFormatContext *s)
706 HLSContext *c = s->priv_data;
708 free_variant_list(c);
712 static int hls_read_seek(AVFormatContext *s, int stream_index,
713 int64_t timestamp, int flags)
715 HLSContext *c = s->priv_data;
718 if ((flags & AVSEEK_FLAG_BYTE) || !c->variants[0]->finished)
719 return AVERROR(ENOSYS);
721 c->seek_flags = flags;
722 c->seek_timestamp = stream_index < 0 ? timestamp :
723 av_rescale_rnd(timestamp, AV_TIME_BASE,
724 s->streams[stream_index]->time_base.den,
725 flags & AVSEEK_FLAG_BACKWARD ?
726 AV_ROUND_DOWN : AV_ROUND_UP);
727 timestamp = av_rescale_rnd(timestamp, AV_TIME_BASE, stream_index >= 0 ?
728 s->streams[stream_index]->time_base.den :
729 AV_TIME_BASE, flags & AVSEEK_FLAG_BACKWARD ?
730 AV_ROUND_DOWN : AV_ROUND_UP);
731 if (s->duration < c->seek_timestamp) {
732 c->seek_timestamp = AV_NOPTS_VALUE;
737 for (i = 0; i < c->n_variants; i++) {
739 struct variant *var = c->variants[i];
740 int64_t pos = c->first_timestamp == AV_NOPTS_VALUE ?
741 0 : c->first_timestamp;
743 ffurl_close(var->input);
746 av_free_packet(&var->pkt);
747 reset_packet(&var->pkt);
748 var->pb.eof_reached = 0;
749 /* Clear any buffered data */
750 var->pb.buf_end = var->pb.buf_ptr = var->pb.buffer;
751 /* Reset the pos, to let the mpegts demuxer know we've seeked. */
754 /* Locate the segment that contains the target timestamp */
755 for (j = 0; j < var->n_segments; j++) {
756 if (timestamp >= pos &&
757 timestamp < pos + var->segments[j]->duration) {
758 var->cur_seq_no = var->start_seq_no + j;
762 pos += var->segments[j]->duration;
765 c->seek_timestamp = AV_NOPTS_VALUE;
770 static int hls_probe(AVProbeData *p)
772 /* Require #EXTM3U at the start, and either one of the ones below
773 * somewhere for a proper match. */
774 if (strncmp(p->buf, "#EXTM3U", 7))
776 if (strstr(p->buf, "#EXT-X-STREAM-INF:") ||
777 strstr(p->buf, "#EXT-X-TARGETDURATION:") ||
778 strstr(p->buf, "#EXT-X-MEDIA-SEQUENCE:"))
779 return AVPROBE_SCORE_MAX;
783 AVInputFormat ff_hls_demuxer = {
784 .name = "hls,applehttp",
785 .long_name = NULL_IF_CONFIG_SMALL("Apple HTTP Live Streaming"),
786 .priv_data_size = sizeof(HLSContext),
787 .read_probe = hls_probe,
788 .read_header = hls_read_header,
789 .read_packet = hls_read_packet,
790 .read_close = hls_close,
791 .read_seek = hls_read_seek,