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 (preferrably 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;
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 && 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, duration = 0, is_segment = 0, is_variant = 0, bandwidth = 0;
206 enum KeyType key_type = KEY_NONE;
209 char key[MAX_URL_SIZE] = "";
216 if ((ret = avio_open2(&in, url, AVIO_FLAG_READ,
217 c->interrupt_callback, NULL)) < 0)
221 read_chomp_line(in, line, sizeof(line));
222 if (strcmp(line, "#EXTM3U")) {
223 ret = AVERROR_INVALIDDATA;
228 free_segment_list(var);
231 while (!url_feof(in)) {
232 read_chomp_line(in, line, sizeof(line));
233 if (av_strstart(line, "#EXT-X-STREAM-INF:", &ptr)) {
234 struct variant_info info = {{0}};
236 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_variant_args,
238 bandwidth = atoi(info.bandwidth);
239 } else if (av_strstart(line, "#EXT-X-KEY:", &ptr)) {
240 struct key_info info = {{0}};
241 ff_parse_key_value(ptr, (ff_parse_key_val_cb) handle_key_args,
245 if (!strcmp(info.method, "AES-128"))
246 key_type = KEY_AES_128;
247 if (!strncmp(info.iv, "0x", 2) || !strncmp(info.iv, "0X", 2)) {
248 ff_hex_to_data(iv, info.iv + 2);
251 av_strlcpy(key, info.uri, sizeof(key));
252 } else if (av_strstart(line, "#EXT-X-TARGETDURATION:", &ptr)) {
254 var = new_variant(c, 0, url, NULL);
256 ret = AVERROR(ENOMEM);
260 var->target_duration = atoi(ptr);
261 } else if (av_strstart(line, "#EXT-X-MEDIA-SEQUENCE:", &ptr)) {
263 var = new_variant(c, 0, url, NULL);
265 ret = AVERROR(ENOMEM);
269 var->start_seq_no = atoi(ptr);
270 } else if (av_strstart(line, "#EXT-X-ENDLIST", &ptr)) {
273 } else if (av_strstart(line, "#EXTINF:", &ptr)) {
275 duration = atoi(ptr);
276 } else if (av_strstart(line, "#", NULL)) {
278 } else if (line[0]) {
280 if (!new_variant(c, bandwidth, line, url)) {
281 ret = AVERROR(ENOMEM);
290 var = new_variant(c, 0, url, NULL);
292 ret = AVERROR(ENOMEM);
296 seg = av_malloc(sizeof(struct segment));
298 ret = AVERROR(ENOMEM);
301 seg->duration = duration;
302 seg->key_type = key_type;
304 memcpy(seg->iv, iv, sizeof(iv));
306 int seq = var->start_seq_no + var->n_segments;
307 memset(seg->iv, 0, sizeof(seg->iv));
308 AV_WB32(seg->iv + 12, seq);
310 ff_make_absolute_url(seg->key, sizeof(seg->key), url, key);
311 ff_make_absolute_url(seg->url, sizeof(seg->url), url, line);
312 dynarray_add(&var->segments, &var->n_segments, seg);
318 var->last_load_time = av_gettime();
326 static int open_input(struct variant *var)
328 struct segment *seg = var->segments[var->cur_seq_no - var->start_seq_no];
329 if (seg->key_type == KEY_NONE) {
330 return ffurl_open(&var->input, seg->url, AVIO_FLAG_READ,
331 &var->parent->interrupt_callback, NULL);
332 } else if (seg->key_type == KEY_AES_128) {
333 char iv[33], key[33], url[MAX_URL_SIZE];
335 if (strcmp(seg->key, var->key_url)) {
337 if (ffurl_open(&uc, seg->key, AVIO_FLAG_READ,
338 &var->parent->interrupt_callback, NULL) == 0) {
339 if (ffurl_read_complete(uc, var->key, sizeof(var->key))
340 != sizeof(var->key)) {
341 av_log(NULL, AV_LOG_ERROR, "Unable to read key file %s\n",
346 av_log(NULL, AV_LOG_ERROR, "Unable to open key file %s\n",
349 av_strlcpy(var->key_url, seg->key, sizeof(var->key_url));
351 ff_data_to_hex(iv, seg->iv, sizeof(seg->iv), 0);
352 ff_data_to_hex(key, var->key, sizeof(var->key), 0);
353 iv[32] = key[32] = '\0';
354 if (strstr(seg->url, "://"))
355 snprintf(url, sizeof(url), "crypto+%s", seg->url);
357 snprintf(url, sizeof(url), "crypto:%s", seg->url);
358 if ((ret = ffurl_alloc(&var->input, url, AVIO_FLAG_READ,
359 &var->parent->interrupt_callback)) < 0)
361 av_opt_set(var->input->priv_data, "key", key, 0);
362 av_opt_set(var->input->priv_data, "iv", iv, 0);
363 if ((ret = ffurl_connect(var->input, NULL)) < 0) {
364 ffurl_close(var->input);
370 return AVERROR(ENOSYS);
373 static int read_data(void *opaque, uint8_t *buf, int buf_size)
375 struct variant *v = opaque;
376 HLSContext *c = v->parent->priv_data;
381 /* If this is a live stream and the reload interval has elapsed since
382 * the last playlist reload, reload the variant playlists now. */
383 int64_t reload_interval = v->n_segments > 0 ?
384 v->segments[v->n_segments - 1]->duration :
386 reload_interval *= 1000000;
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 * 500000;
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 if (ret < 0 && ret != AVERROR_EOF)
425 ffurl_close(v->input);
429 c->end_of_segment = 1;
430 c->cur_seq_no = v->cur_seq_no;
432 if (v->ctx && v->ctx->nb_streams) {
434 for (i = v->stream_offset; i < v->stream_offset + v->ctx->nb_streams;
436 if (v->parent->streams[i]->discard < AVDISCARD_ALL)
441 av_log(v->parent, AV_LOG_INFO, "No longer receiving variant %d\n",
448 static int hls_read_header(AVFormatContext *s)
450 HLSContext *c = s->priv_data;
451 int ret = 0, i, j, stream_offset = 0;
453 c->interrupt_callback = &s->interrupt_callback;
455 if ((ret = parse_playlist(c, s->filename, NULL, s->pb)) < 0)
458 if (c->n_variants == 0) {
459 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
463 /* If the playlist only contained variants, parse each individual
464 * variant playlist. */
465 if (c->n_variants > 1 || c->variants[0]->n_segments == 0) {
466 for (i = 0; i < c->n_variants; i++) {
467 struct variant *v = c->variants[i];
468 if ((ret = parse_playlist(c, v->url, v, NULL)) < 0)
473 if (c->variants[0]->n_segments == 0) {
474 av_log(NULL, AV_LOG_WARNING, "Empty playlist\n");
479 /* If this isn't a live stream, calculate the total duration of the
481 if (c->variants[0]->finished) {
482 int64_t duration = 0;
483 for (i = 0; i < c->variants[0]->n_segments; i++)
484 duration += c->variants[0]->segments[i]->duration;
485 s->duration = duration * AV_TIME_BASE;
488 /* Open the demuxer for each variant */
489 for (i = 0; i < c->n_variants; i++) {
490 struct variant *v = c->variants[i];
491 AVInputFormat *in_fmt = NULL;
492 char bitrate_str[20];
493 if (v->n_segments == 0)
496 if (!(v->ctx = avformat_alloc_context())) {
497 ret = AVERROR(ENOMEM);
505 /* If this is a live stream with more than 3 segments, start at the
506 * third last segment. */
507 v->cur_seq_no = v->start_seq_no;
508 if (!v->finished && v->n_segments > 3)
509 v->cur_seq_no = v->start_seq_no + v->n_segments - 3;
511 v->read_buffer = av_malloc(INITIAL_BUFFER_SIZE);
512 ffio_init_context(&v->pb, v->read_buffer, INITIAL_BUFFER_SIZE, 0, v,
513 read_data, NULL, NULL);
515 ret = av_probe_input_buffer(&v->pb, &in_fmt, v->segments[0]->url,
518 /* Free the ctx - it isn't initialized properly at this point,
519 * so avformat_close_input shouldn't be called. If
520 * avformat_open_input fails below, it frees and zeros the
521 * context, so it doesn't need any special treatment like this. */
522 av_log(s, AV_LOG_ERROR, "Error when loading first segment '%s'\n", v->segments[0]->url);
523 avformat_free_context(v->ctx);
528 ret = avformat_open_input(&v->ctx, v->segments[0]->url, in_fmt, NULL);
531 v->stream_offset = stream_offset;
532 snprintf(bitrate_str, sizeof(bitrate_str), "%d", v->bandwidth);
533 /* Create new AVStreams for each stream in this variant */
534 for (j = 0; j < v->ctx->nb_streams; j++) {
535 AVStream *st = avformat_new_stream(s, NULL);
537 ret = AVERROR(ENOMEM);
541 avcodec_copy_context(st->codec, v->ctx->streams[j]->codec);
543 av_dict_set(&st->metadata, "variant_bitrate", bitrate_str,
546 stream_offset += v->ctx->nb_streams;
550 c->first_timestamp = AV_NOPTS_VALUE;
551 c->seek_timestamp = AV_NOPTS_VALUE;
555 free_variant_list(c);
559 static int recheck_discard_flags(AVFormatContext *s, int first)
561 HLSContext *c = s->priv_data;
564 /* Check if any new streams are needed */
565 for (i = 0; i < c->n_variants; i++)
566 c->variants[i]->cur_needed = 0;
568 for (i = 0; i < s->nb_streams; i++) {
569 AVStream *st = s->streams[i];
570 struct variant *var = c->variants[s->streams[i]->id];
571 if (st->discard < AVDISCARD_ALL)
574 for (i = 0; i < c->n_variants; i++) {
575 struct variant *v = c->variants[i];
576 if (v->cur_needed && !v->needed) {
579 v->cur_seq_no = c->cur_seq_no;
580 v->pb.eof_reached = 0;
581 av_log(s, AV_LOG_INFO, "Now receiving variant %d\n", i);
582 } else if (first && !v->cur_needed && v->needed) {
584 ffurl_close(v->input);
588 av_log(s, AV_LOG_INFO, "No longer receiving variant %d\n", i);
594 static int hls_read_packet(AVFormatContext *s, AVPacket *pkt)
596 HLSContext *c = s->priv_data;
597 int ret, i, minvariant = -1;
599 if (c->first_packet) {
600 recheck_discard_flags(s, 1);
605 c->end_of_segment = 0;
606 for (i = 0; i < c->n_variants; i++) {
607 struct variant *var = c->variants[i];
608 /* Make sure we've got one buffered packet from each open variant
610 if (var->needed && !var->pkt.data) {
614 ret = av_read_frame(var->ctx, &var->pkt);
616 if (!url_feof(&var->pb))
618 reset_packet(&var->pkt);
621 if (c->first_timestamp == AV_NOPTS_VALUE)
622 c->first_timestamp = var->pkt.dts;
625 if (c->seek_timestamp == AV_NOPTS_VALUE)
628 if (var->pkt.dts == AV_NOPTS_VALUE) {
629 c->seek_timestamp = AV_NOPTS_VALUE;
633 st = var->ctx->streams[var->pkt.stream_index];
634 ts_diff = av_rescale_rnd(var->pkt.dts, AV_TIME_BASE,
635 st->time_base.den, AV_ROUND_DOWN) -
637 if (ts_diff >= 0 && (c->seek_flags & AVSEEK_FLAG_ANY ||
638 var->pkt.flags & AV_PKT_FLAG_KEY)) {
639 c->seek_timestamp = AV_NOPTS_VALUE;
644 /* Check if this stream has the packet with the lowest dts */
646 if (minvariant < 0 ||
647 var->pkt.dts < c->variants[minvariant]->pkt.dts)
651 if (c->end_of_segment) {
652 if (recheck_discard_flags(s, 0))
655 /* If we got a packet, return it */
656 if (minvariant >= 0) {
657 *pkt = c->variants[minvariant]->pkt;
658 pkt->stream_index += c->variants[minvariant]->stream_offset;
659 reset_packet(&c->variants[minvariant]->pkt);
665 static int hls_close(AVFormatContext *s)
667 HLSContext *c = s->priv_data;
669 free_variant_list(c);
673 static int hls_read_seek(AVFormatContext *s, int stream_index,
674 int64_t timestamp, int flags)
676 HLSContext *c = s->priv_data;
679 if ((flags & AVSEEK_FLAG_BYTE) || !c->variants[0]->finished)
680 return AVERROR(ENOSYS);
682 c->seek_flags = flags;
683 c->seek_timestamp = stream_index < 0 ? timestamp :
684 av_rescale_rnd(timestamp, AV_TIME_BASE,
685 s->streams[stream_index]->time_base.den,
686 flags & AVSEEK_FLAG_BACKWARD ?
687 AV_ROUND_DOWN : AV_ROUND_UP);
688 timestamp = av_rescale_rnd(timestamp, 1, stream_index >= 0 ?
689 s->streams[stream_index]->time_base.den :
690 AV_TIME_BASE, flags & AVSEEK_FLAG_BACKWARD ?
691 AV_ROUND_DOWN : AV_ROUND_UP);
692 if (s->duration < c->seek_timestamp) {
693 c->seek_timestamp = AV_NOPTS_VALUE;
698 for (i = 0; i < c->n_variants; i++) {
700 struct variant *var = c->variants[i];
701 int64_t pos = c->first_timestamp == AV_NOPTS_VALUE ? 0 :
702 av_rescale_rnd(c->first_timestamp, 1, stream_index >= 0 ?
703 s->streams[stream_index]->time_base.den :
704 AV_TIME_BASE, flags & AVSEEK_FLAG_BACKWARD ?
705 AV_ROUND_DOWN : AV_ROUND_UP);
707 ffurl_close(var->input);
710 av_free_packet(&var->pkt);
711 reset_packet(&var->pkt);
712 var->pb.eof_reached = 0;
713 /* Clear any buffered data */
714 var->pb.buf_end = var->pb.buf_ptr = var->pb.buffer;
715 /* Reset the pos, to let the mpegts demuxer know we've seeked. */
718 /* Locate the segment that contains the target timestamp */
719 for (j = 0; j < var->n_segments; j++) {
720 if (timestamp >= pos &&
721 timestamp < pos + var->segments[j]->duration) {
722 var->cur_seq_no = var->start_seq_no + j;
726 pos += var->segments[j]->duration;
729 c->seek_timestamp = AV_NOPTS_VALUE;
734 static int hls_probe(AVProbeData *p)
736 /* Require #EXTM3U at the start, and either one of the ones below
737 * somewhere for a proper match. */
738 if (strncmp(p->buf, "#EXTM3U", 7))
740 if (strstr(p->buf, "#EXT-X-STREAM-INF:") ||
741 strstr(p->buf, "#EXT-X-TARGETDURATION:") ||
742 strstr(p->buf, "#EXT-X-MEDIA-SEQUENCE:"))
743 return AVPROBE_SCORE_MAX;
747 AVInputFormat ff_hls_demuxer = {
748 .name = "hls,applehttp",
749 .long_name = NULL_IF_CONFIG_SMALL("Apple HTTP Live Streaming format"),
750 .priv_data_size = sizeof(HLSContext),
751 .read_probe = hls_probe,
752 .read_header = hls_read_header,
753 .read_packet = hls_read_packet,
754 .read_close = hls_close,
755 .read_seek = hls_read_seek,