2 * MPEG-DASH ISO BMFF segmenter
3 * Copyright (c) 2014 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
27 #include "libavutil/avassert.h"
28 #include "libavutil/avstring.h"
29 #include "libavutil/intreadwrite.h"
30 #include "libavutil/mathematics.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/rational.h"
33 #include "libavutil/time_internal.h"
37 #include "avio_internal.h"
40 #include "os_support.h"
43 // See ISO/IEC 23009-1:2014 5.3.9.4.4
45 DASH_TMPL_ID_UNDEFINED = -1,
49 DASH_TMPL_ID_BANDWIDTH,
53 typedef struct Segment {
56 int range_length, index_length;
62 typedef struct OutputStream {
69 int64_t init_start_pos;
70 int init_range_length;
71 int nb_segments, segments_size, segment_index;
73 int64_t first_pts, start_pts, max_pts;
76 char bandwidth_str[64];
81 typedef struct DASHContext {
82 const AVClass *class; /* Class for private options. */
84 int extra_window_size;
90 OutputStream *streams;
91 int has_video, has_audio;
92 int64_t last_duration;
93 int64_t total_duration;
94 char availability_start_time[100];
96 const char *single_file_name;
97 const char *init_seg_name;
98 const char *media_seg_name;
99 AVRational min_frame_rate, max_frame_rate;
100 int ambiguous_frame_rate;
103 static int dash_write(void *opaque, uint8_t *buf, int buf_size)
105 OutputStream *os = opaque;
107 avio_write(os->out, buf, buf_size);
112 static void set_codec_str(AVFormatContext *s, AVCodecParameters *par,
115 const AVCodecTag *tags[2] = { NULL, NULL };
117 if (par->codec_type == AVMEDIA_TYPE_VIDEO)
118 tags[0] = ff_codec_movvideo_tags;
119 else if (par->codec_type == AVMEDIA_TYPE_AUDIO)
120 tags[0] = ff_codec_movaudio_tags;
124 tag = av_codec_get_tag(tags, par->codec_id);
132 if (!strcmp(str, "mp4a") || !strcmp(str, "mp4v")) {
134 tags[0] = ff_mp4_obj_type;
135 oti = av_codec_get_tag(tags, par->codec_id);
137 av_strlcatf(str, size, ".%02x", oti);
141 if (tag == MKTAG('m', 'p', '4', 'a')) {
142 if (par->extradata_size >= 2) {
143 int aot = par->extradata[0] >> 3;
145 aot = ((AV_RB16(par->extradata) >> 5) & 0x3f) + 32;
146 av_strlcatf(str, size, ".%d", aot);
148 } else if (tag == MKTAG('m', 'p', '4', 'v')) {
149 // Unimplemented, should output ProfileLevelIndication as a decimal number
150 av_log(s, AV_LOG_WARNING, "Incomplete RFC 6381 codec string for mp4v\n");
152 } else if (!strcmp(str, "avc1")) {
153 uint8_t *tmpbuf = NULL;
154 uint8_t *extradata = par->extradata;
155 int extradata_size = par->extradata_size;
158 if (extradata[0] != 1) {
160 if (avio_open_dyn_buf(&pb) < 0)
162 if (ff_isom_write_avcc(pb, extradata, extradata_size) < 0) {
163 ffio_free_dyn_buf(&pb);
166 extradata_size = avio_close_dyn_buf(pb, &extradata);
170 if (extradata_size >= 4)
171 av_strlcatf(str, size, ".%02x%02x%02x",
172 extradata[1], extradata[2], extradata[3]);
177 static void dash_free(AVFormatContext *s)
179 DASHContext *c = s->priv_data;
183 for (i = 0; i < s->nb_streams; i++) {
184 OutputStream *os = &c->streams[i];
185 if (os->ctx && os->ctx_inited)
186 av_write_trailer(os->ctx);
187 if (os->ctx && os->ctx->pb)
188 av_free(os->ctx->pb);
189 ff_format_io_close(s, &os->out);
191 avformat_free_context(os->ctx);
192 for (j = 0; j < os->nb_segments; j++)
193 av_free(os->segments[j]);
194 av_free(os->segments);
196 av_freep(&c->streams);
199 static void output_segment_list(OutputStream *os, AVIOContext *out, DASHContext *c)
201 int i, start_index = 0, start_number = 1;
202 if (c->window_size) {
203 start_index = FFMAX(os->nb_segments - c->window_size, 0);
204 start_number = FFMAX(os->segment_index - c->window_size, 1);
207 if (c->use_template) {
208 int timescale = c->use_timeline ? os->ctx->streams[0]->time_base.den : AV_TIME_BASE;
209 avio_printf(out, "\t\t\t\t<SegmentTemplate timescale=\"%d\" ", timescale);
210 if (!c->use_timeline)
211 avio_printf(out, "duration=\"%"PRId64"\" ", c->last_duration);
212 avio_printf(out, "initialization=\"%s\" media=\"%s\" startNumber=\"%d\">\n", c->init_seg_name, c->media_seg_name, c->use_timeline ? start_number : 1);
213 if (c->use_timeline) {
214 int64_t cur_time = 0;
215 avio_printf(out, "\t\t\t\t\t<SegmentTimeline>\n");
216 for (i = start_index; i < os->nb_segments; ) {
217 Segment *seg = os->segments[i];
219 avio_printf(out, "\t\t\t\t\t\t<S ");
220 if (i == start_index || seg->time != cur_time) {
221 cur_time = seg->time;
222 avio_printf(out, "t=\"%"PRId64"\" ", seg->time);
224 avio_printf(out, "d=\"%d\" ", seg->duration);
225 while (i + repeat + 1 < os->nb_segments &&
226 os->segments[i + repeat + 1]->duration == seg->duration &&
227 os->segments[i + repeat + 1]->time == os->segments[i + repeat]->time + os->segments[i + repeat]->duration)
230 avio_printf(out, "r=\"%d\" ", repeat);
231 avio_printf(out, "/>\n");
233 cur_time += (1 + repeat) * seg->duration;
235 avio_printf(out, "\t\t\t\t\t</SegmentTimeline>\n");
237 avio_printf(out, "\t\t\t\t</SegmentTemplate>\n");
238 } else if (c->single_file) {
239 avio_printf(out, "\t\t\t\t<BaseURL>%s</BaseURL>\n", os->initfile);
240 avio_printf(out, "\t\t\t\t<SegmentList timescale=\"%d\" duration=\"%"PRId64"\" startNumber=\"%d\">\n", AV_TIME_BASE, c->last_duration, start_number);
241 avio_printf(out, "\t\t\t\t\t<Initialization range=\"%"PRId64"-%"PRId64"\" />\n", os->init_start_pos, os->init_start_pos + os->init_range_length - 1);
242 for (i = start_index; i < os->nb_segments; i++) {
243 Segment *seg = os->segments[i];
244 avio_printf(out, "\t\t\t\t\t<SegmentURL mediaRange=\"%"PRId64"-%"PRId64"\" ", seg->start_pos, seg->start_pos + seg->range_length - 1);
245 if (seg->index_length)
246 avio_printf(out, "indexRange=\"%"PRId64"-%"PRId64"\" ", seg->start_pos, seg->start_pos + seg->index_length - 1);
247 avio_printf(out, "/>\n");
249 avio_printf(out, "\t\t\t\t</SegmentList>\n");
251 avio_printf(out, "\t\t\t\t<SegmentList timescale=\"%d\" duration=\"%"PRId64"\" startNumber=\"%d\">\n", AV_TIME_BASE, c->last_duration, start_number);
252 avio_printf(out, "\t\t\t\t\t<Initialization sourceURL=\"%s\" />\n", os->initfile);
253 for (i = start_index; i < os->nb_segments; i++) {
254 Segment *seg = os->segments[i];
255 avio_printf(out, "\t\t\t\t\t<SegmentURL media=\"%s\" />\n", seg->file);
257 avio_printf(out, "\t\t\t\t</SegmentList>\n");
261 static DASHTmplId dash_read_tmpl_id(const char *identifier, char *format_tag,
262 size_t format_tag_size, const char **ptr) {
263 const char *next_ptr;
264 DASHTmplId id_type = DASH_TMPL_ID_UNDEFINED;
266 if (av_strstart(identifier, "$$", &next_ptr)) {
267 id_type = DASH_TMPL_ID_ESCAPE;
269 } else if (av_strstart(identifier, "$RepresentationID$", &next_ptr)) {
270 id_type = DASH_TMPL_ID_REP_ID;
271 // default to basic format, as $RepresentationID$ identifiers
272 // are not allowed to have custom format-tags.
273 av_strlcpy(format_tag, "%d", format_tag_size);
275 } else { // the following identifiers may have an explicit format_tag
276 if (av_strstart(identifier, "$Number", &next_ptr))
277 id_type = DASH_TMPL_ID_NUMBER;
278 else if (av_strstart(identifier, "$Bandwidth", &next_ptr))
279 id_type = DASH_TMPL_ID_BANDWIDTH;
280 else if (av_strstart(identifier, "$Time", &next_ptr))
281 id_type = DASH_TMPL_ID_TIME;
283 id_type = DASH_TMPL_ID_UNDEFINED;
285 // next parse the dash format-tag and generate a c-string format tag
286 // (next_ptr now points at the first '%' at the beginning of the format-tag)
287 if (id_type != DASH_TMPL_ID_UNDEFINED) {
288 const char *number_format = (id_type == DASH_TMPL_ID_TIME) ? PRId64 : "d";
289 if (next_ptr[0] == '$') { // no dash format-tag
290 snprintf(format_tag, format_tag_size, "%%%s", number_format);
293 const char *width_ptr;
294 // only tolerate single-digit width-field (i.e. up to 9-digit width)
295 if (av_strstart(next_ptr, "%0", &width_ptr) &&
296 av_isdigit(width_ptr[0]) &&
297 av_strstart(&width_ptr[1], "d$", &next_ptr)) {
298 // yes, we're using a format tag to build format_tag.
299 snprintf(format_tag, format_tag_size, "%s%c%s", "%0", width_ptr[0], number_format);
302 av_log(NULL, AV_LOG_WARNING, "Failed to parse format-tag beginning with %s. Expected either a "
303 "closing '$' character or a format-string like '%%0[width]d', "
304 "where width must be a single digit\n", next_ptr);
305 id_type = DASH_TMPL_ID_UNDEFINED;
313 static void dash_fill_tmpl_params(char *dst, size_t buffer_size,
314 const char *template, int rep_id,
315 int number, int bit_rate,
318 const char *t_cur = template;
319 while (dst_pos < buffer_size - 1 && *t_cur) {
320 char format_tag[7]; // May be "%d", "%0Xd", or "%0Xlld" (for $Time$), where X is in [0-9]
323 const char *t_next = strchr(t_cur, '$'); // copy over everything up to the first '$' character
325 int num_copy_bytes = FFMIN(t_next - t_cur, buffer_size - dst_pos - 1);
326 av_strlcpy(&dst[dst_pos], t_cur, num_copy_bytes + 1);
328 dst_pos += num_copy_bytes;
330 } else { // no more DASH identifiers to substitute - just copy the rest over and break
331 av_strlcpy(&dst[dst_pos], t_cur, buffer_size - dst_pos);
335 if (dst_pos >= buffer_size - 1 || !*t_cur)
338 // t_cur is now pointing to a '$' character
339 id_type = dash_read_tmpl_id(t_cur, format_tag, sizeof(format_tag), &t_next);
341 case DASH_TMPL_ID_ESCAPE:
342 av_strlcpy(&dst[dst_pos], "$", 2);
345 case DASH_TMPL_ID_REP_ID:
346 n = snprintf(&dst[dst_pos], buffer_size - dst_pos, format_tag, rep_id);
348 case DASH_TMPL_ID_NUMBER:
349 n = snprintf(&dst[dst_pos], buffer_size - dst_pos, format_tag, number);
351 case DASH_TMPL_ID_BANDWIDTH:
352 n = snprintf(&dst[dst_pos], buffer_size - dst_pos, format_tag, bit_rate);
354 case DASH_TMPL_ID_TIME:
355 n = snprintf(&dst[dst_pos], buffer_size - dst_pos, format_tag, time);
357 case DASH_TMPL_ID_UNDEFINED:
358 // copy over one byte and advance
359 av_strlcpy(&dst[dst_pos], t_cur, 2);
364 // t_next points just past the processed identifier
365 // n is the number of bytes that were attempted to be written to dst
366 // (may have failed to write all because buffer_size).
369 dst_pos += FFMIN(n, buffer_size - dst_pos - 1);
374 static char *xmlescape(const char *str) {
375 int outlen = strlen(str)*3/2 + 6;
376 char *out = av_realloc(NULL, outlen + 1);
380 for (; *str; str++) {
381 if (pos + 6 > outlen) {
383 outlen = 2 * outlen + 6;
384 tmp = av_realloc(out, outlen + 1);
392 memcpy(&out[pos], "&", 5);
394 } else if (*str == '<') {
395 memcpy(&out[pos], "<", 4);
397 } else if (*str == '>') {
398 memcpy(&out[pos], ">", 4);
400 } else if (*str == '\'') {
401 memcpy(&out[pos], "'", 6);
403 } else if (*str == '\"') {
404 memcpy(&out[pos], """, 6);
414 static void write_time(AVIOContext *out, int64_t time)
416 int seconds = time / AV_TIME_BASE;
417 int fractions = time % AV_TIME_BASE;
418 int minutes = seconds / 60;
419 int hours = minutes / 60;
422 avio_printf(out, "PT");
424 avio_printf(out, "%dH", hours);
425 if (hours || minutes)
426 avio_printf(out, "%dM", minutes);
427 avio_printf(out, "%d.%dS", seconds, fractions / (AV_TIME_BASE / 10));
430 static void format_date_now(char *buf, int size)
432 time_t t = time(NULL);
433 struct tm *ptm, tmbuf;
434 ptm = gmtime_r(&t, &tmbuf);
436 if (!strftime(buf, size, "%Y-%m-%dT%H:%M:%S", ptm))
441 static int write_manifest(AVFormatContext *s, int final)
443 DASHContext *c = s->priv_data;
445 char temp_filename[1024];
447 AVDictionaryEntry *title = av_dict_get(s->metadata, "title", NULL, 0);
449 snprintf(temp_filename, sizeof(temp_filename), "%s.tmp", s->filename);
450 ret = s->io_open(s, &out, temp_filename, AVIO_FLAG_WRITE, NULL);
452 av_log(s, AV_LOG_ERROR, "Unable to open %s for writing\n", temp_filename);
455 avio_printf(out, "<?xml version=\"1.0\" encoding=\"utf-8\"?>\n");
456 avio_printf(out, "<MPD xmlns:xsi=\"http://www.w3.org/2001/XMLSchema-instance\"\n"
457 "\txmlns=\"urn:mpeg:dash:schema:mpd:2011\"\n"
458 "\txmlns:xlink=\"http://www.w3.org/1999/xlink\"\n"
459 "\txsi:schemaLocation=\"urn:mpeg:DASH:schema:MPD:2011 http://standards.iso.org/ittf/PubliclyAvailableStandards/MPEG-DASH_schema_files/DASH-MPD.xsd\"\n"
460 "\tprofiles=\"urn:mpeg:dash:profile:isoff-live:2011\"\n"
461 "\ttype=\"%s\"\n", final ? "static" : "dynamic");
463 avio_printf(out, "\tmediaPresentationDuration=\"");
464 write_time(out, c->total_duration);
465 avio_printf(out, "\"\n");
467 int64_t update_period = c->last_duration / AV_TIME_BASE;
469 if (c->use_template && !c->use_timeline)
471 avio_printf(out, "\tminimumUpdatePeriod=\"PT%"PRId64"S\"\n", update_period);
472 avio_printf(out, "\tsuggestedPresentationDelay=\"PT%"PRId64"S\"\n", c->last_duration / AV_TIME_BASE);
473 if (!c->availability_start_time[0] && s->nb_streams > 0 && c->streams[0].nb_segments > 0) {
474 format_date_now(c->availability_start_time, sizeof(c->availability_start_time));
476 if (c->availability_start_time[0])
477 avio_printf(out, "\tavailabilityStartTime=\"%s\"\n", c->availability_start_time);
478 format_date_now(now_str, sizeof(now_str));
480 avio_printf(out, "\tpublishTime=\"%s\"\n", now_str);
481 if (c->window_size && c->use_template) {
482 avio_printf(out, "\ttimeShiftBufferDepth=\"");
483 write_time(out, c->last_duration * c->window_size);
484 avio_printf(out, "\"\n");
487 avio_printf(out, "\tminBufferTime=\"");
488 write_time(out, c->last_duration);
489 avio_printf(out, "\">\n");
490 avio_printf(out, "\t<ProgramInformation>\n");
492 char *escaped = xmlescape(title->value);
493 avio_printf(out, "\t\t<Title>%s</Title>\n", escaped);
496 avio_printf(out, "\t</ProgramInformation>\n");
497 if (c->window_size && s->nb_streams > 0 && c->streams[0].nb_segments > 0 && !c->use_template) {
498 OutputStream *os = &c->streams[0];
499 int start_index = FFMAX(os->nb_segments - c->window_size, 0);
500 int64_t start_time = av_rescale_q(os->segments[start_index]->time, s->streams[0]->time_base, AV_TIME_BASE_Q);
501 avio_printf(out, "\t<Period start=\"");
502 write_time(out, start_time);
503 avio_printf(out, "\">\n");
505 avio_printf(out, "\t<Period start=\"PT0.0S\">\n");
509 avio_printf(out, "\t\t<AdaptationSet contentType=\"video\" segmentAlignment=\"true\" bitstreamSwitching=\"true\"");
510 if (c->max_frame_rate.num && !c->ambiguous_frame_rate)
511 avio_printf(out, " %s=\"%d/%d\"", (av_cmp_q(c->min_frame_rate, c->max_frame_rate) < 0) ? "maxFrameRate" : "frameRate", c->max_frame_rate.num, c->max_frame_rate.den);
512 avio_printf(out, ">\n");
514 for (i = 0; i < s->nb_streams; i++) {
515 AVStream *st = s->streams[i];
516 OutputStream *os = &c->streams[i];
518 if (st->codecpar->codec_type != AVMEDIA_TYPE_VIDEO)
521 avio_printf(out, "\t\t\t<Representation id=\"%d\" mimeType=\"video/mp4\" codecs=\"%s\"%s width=\"%d\" height=\"%d\"", i, os->codec_str, os->bandwidth_str, st->codecpar->width, st->codecpar->height);
522 if (st->avg_frame_rate.num)
523 avio_printf(out, " frameRate=\"%d/%d\"", st->avg_frame_rate.num, st->avg_frame_rate.den);
524 avio_printf(out, ">\n");
526 output_segment_list(&c->streams[i], out, c);
527 avio_printf(out, "\t\t\t</Representation>\n");
529 avio_printf(out, "\t\t</AdaptationSet>\n");
532 avio_printf(out, "\t\t<AdaptationSet contentType=\"audio\" segmentAlignment=\"true\" bitstreamSwitching=\"true\">\n");
533 for (i = 0; i < s->nb_streams; i++) {
534 AVStream *st = s->streams[i];
535 OutputStream *os = &c->streams[i];
537 if (st->codecpar->codec_type != AVMEDIA_TYPE_AUDIO)
540 avio_printf(out, "\t\t\t<Representation id=\"%d\" mimeType=\"audio/mp4\" codecs=\"%s\"%s audioSamplingRate=\"%d\">\n", i, os->codec_str, os->bandwidth_str, st->codecpar->sample_rate);
541 avio_printf(out, "\t\t\t\t<AudioChannelConfiguration schemeIdUri=\"urn:mpeg:dash:23003:3:audio_channel_configuration:2011\" value=\"%d\" />\n", st->codecpar->channels);
542 output_segment_list(&c->streams[i], out, c);
543 avio_printf(out, "\t\t\t</Representation>\n");
545 avio_printf(out, "\t\t</AdaptationSet>\n");
547 avio_printf(out, "\t</Period>\n");
548 avio_printf(out, "</MPD>\n");
550 ff_format_io_close(s, &out);
551 return avpriv_io_move(temp_filename, s->filename);
554 static int dash_init(AVFormatContext *s)
556 DASHContext *c = s->priv_data;
558 AVOutputFormat *oformat;
562 if (c->single_file_name)
566 c->ambiguous_frame_rate = 0;
568 av_strlcpy(c->dirname, s->filename, sizeof(c->dirname));
569 ptr = strrchr(c->dirname, '/');
571 av_strlcpy(basename, &ptr[1], sizeof(basename));
574 c->dirname[0] = '\0';
575 av_strlcpy(basename, s->filename, sizeof(basename));
578 ptr = strrchr(basename, '.');
582 oformat = av_guess_format("mp4", NULL, NULL);
584 return AVERROR_MUXER_NOT_FOUND;
586 c->streams = av_mallocz(sizeof(*c->streams) * s->nb_streams);
588 return AVERROR(ENOMEM);
590 for (i = 0; i < s->nb_streams; i++) {
591 OutputStream *os = &c->streams[i];
592 AVFormatContext *ctx;
594 AVDictionary *opts = NULL;
597 os->bit_rate = s->streams[i]->codecpar->bit_rate;
599 snprintf(os->bandwidth_str, sizeof(os->bandwidth_str),
600 " bandwidth=\"%d\"", os->bit_rate);
602 int level = s->strict_std_compliance >= FF_COMPLIANCE_STRICT ?
603 AV_LOG_ERROR : AV_LOG_WARNING;
604 av_log(s, level, "No bit rate set for stream %d\n", i);
605 if (s->strict_std_compliance >= FF_COMPLIANCE_STRICT)
606 return AVERROR(EINVAL);
609 ctx = avformat_alloc_context();
611 return AVERROR(ENOMEM);
613 ctx->oformat = oformat;
614 ctx->interrupt_callback = s->interrupt_callback;
615 ctx->opaque = s->opaque;
616 ctx->io_close = s->io_close;
617 ctx->io_open = s->io_open;
619 if (!(st = avformat_new_stream(ctx, NULL)))
620 return AVERROR(ENOMEM);
621 avcodec_parameters_copy(st->codecpar, s->streams[i]->codecpar);
622 st->sample_aspect_ratio = s->streams[i]->sample_aspect_ratio;
623 st->time_base = s->streams[i]->time_base;
624 ctx->avoid_negative_ts = s->avoid_negative_ts;
625 ctx->flags = s->flags;
627 ctx->pb = avio_alloc_context(os->iobuf, sizeof(os->iobuf), AVIO_FLAG_WRITE, os, NULL, dash_write, NULL);
629 return AVERROR(ENOMEM);
631 if (c->single_file) {
632 if (c->single_file_name)
633 dash_fill_tmpl_params(os->initfile, sizeof(os->initfile), c->single_file_name, i, 0, os->bit_rate, 0);
635 snprintf(os->initfile, sizeof(os->initfile), "%s-stream%d.m4s", basename, i);
637 dash_fill_tmpl_params(os->initfile, sizeof(os->initfile), c->init_seg_name, i, 0, os->bit_rate, 0);
639 snprintf(filename, sizeof(filename), "%s%s", c->dirname, os->initfile);
640 ret = s->io_open(s, &os->out, filename, AVIO_FLAG_WRITE, NULL);
643 os->init_start_pos = 0;
645 av_dict_set(&opts, "movflags", "frag_custom+dash+delay_moov", 0);
646 if ((ret = avformat_init_output(ctx, &opts)) < 0)
652 av_log(s, AV_LOG_VERBOSE, "Representation %d init segment will be written to: %s\n", i, filename);
654 s->streams[i]->time_base = st->time_base;
655 // If the muxer wants to shift timestamps, request to have them shifted
656 // already before being handed to this muxer, so we don't have mismatches
657 // between the MPD and the actual segments.
658 s->avoid_negative_ts = ctx->avoid_negative_ts;
659 if (st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO) {
660 AVRational avg_frame_rate = s->streams[i]->avg_frame_rate;
661 if (avg_frame_rate.num > 0) {
662 if (av_cmp_q(avg_frame_rate, c->min_frame_rate) < 0)
663 c->min_frame_rate = avg_frame_rate;
664 if (av_cmp_q(c->max_frame_rate, avg_frame_rate) < 0)
665 c->max_frame_rate = avg_frame_rate;
667 c->ambiguous_frame_rate = 1;
670 } else if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO) {
674 set_codec_str(s, st->codecpar, os->codec_str, sizeof(os->codec_str));
675 os->first_pts = AV_NOPTS_VALUE;
676 os->max_pts = AV_NOPTS_VALUE;
677 os->last_dts = AV_NOPTS_VALUE;
678 os->segment_index = 1;
681 if (!c->has_video && c->min_seg_duration <= 0) {
682 av_log(s, AV_LOG_WARNING, "no video stream and no min seg duration set\n");
683 return AVERROR(EINVAL);
688 static int dash_write_header(AVFormatContext *s)
690 DASHContext *c = s->priv_data;
692 for (i = 0; i < s->nb_streams; i++) {
693 OutputStream *os = &c->streams[i];
694 if ((ret = avformat_write_header(os->ctx, NULL)) < 0) {
699 ret = write_manifest(s, 0);
701 av_log(s, AV_LOG_VERBOSE, "Manifest written to: %s\n", s->filename);
705 static int add_segment(OutputStream *os, const char *file,
706 int64_t time, int duration,
707 int64_t start_pos, int64_t range_length,
708 int64_t index_length)
712 if (os->nb_segments >= os->segments_size) {
713 os->segments_size = (os->segments_size + 1) * 2;
714 if ((err = av_reallocp(&os->segments, sizeof(*os->segments) *
715 os->segments_size)) < 0) {
716 os->segments_size = 0;
721 seg = av_mallocz(sizeof(*seg));
723 return AVERROR(ENOMEM);
724 av_strlcpy(seg->file, file, sizeof(seg->file));
726 seg->duration = duration;
727 if (seg->time < 0) { // If pts<0, it is expected to be cut away with an edit list
728 seg->duration += seg->time;
731 seg->start_pos = start_pos;
732 seg->range_length = range_length;
733 seg->index_length = index_length;
734 os->segments[os->nb_segments++] = seg;
739 static void write_styp(AVIOContext *pb)
742 ffio_wfourcc(pb, "styp");
743 ffio_wfourcc(pb, "msdh");
744 avio_wb32(pb, 0); /* minor */
745 ffio_wfourcc(pb, "msdh");
746 ffio_wfourcc(pb, "msix");
749 static void find_index_range(AVFormatContext *s, const char *full_path,
750 int64_t pos, int *index_length)
756 ret = s->io_open(s, &pb, full_path, AVIO_FLAG_READ, NULL);
759 if (avio_seek(pb, pos, SEEK_SET) != pos) {
760 ff_format_io_close(s, &pb);
763 ret = avio_read(pb, buf, 8);
764 ff_format_io_close(s, &pb);
767 if (AV_RL32(&buf[4]) != MKTAG('s', 'i', 'd', 'x'))
769 *index_length = AV_RB32(&buf[0]);
772 static int update_stream_extradata(AVFormatContext *s, OutputStream *os,
773 AVCodecParameters *par)
777 if (os->ctx->streams[0]->codecpar->extradata_size || !par->extradata_size)
780 extradata = av_malloc(par->extradata_size);
783 return AVERROR(ENOMEM);
785 memcpy(extradata, par->extradata, par->extradata_size);
787 os->ctx->streams[0]->codecpar->extradata = extradata;
788 os->ctx->streams[0]->codecpar->extradata_size = par->extradata_size;
790 set_codec_str(s, par, os->codec_str, sizeof(os->codec_str));
795 static int dash_flush(AVFormatContext *s, int final, int stream)
797 DASHContext *c = s->priv_data;
799 int cur_flush_segment_index = 0;
801 cur_flush_segment_index = c->streams[stream].segment_index;
803 for (i = 0; i < s->nb_streams; i++) {
804 OutputStream *os = &c->streams[i];
805 char filename[1024] = "", full_path[1024], temp_path[1024];
807 int range_length, index_length = 0;
809 if (!os->packets_written)
812 // Flush the single stream that got a keyframe right now.
813 // Flush all audio streams as well, in sync with video keyframes,
814 // but not the other video streams.
815 if (stream >= 0 && i != stream) {
816 if (s->streams[i]->codecpar->codec_type != AVMEDIA_TYPE_AUDIO)
818 // Make sure we don't flush audio streams multiple times, when
819 // all video streams are flushed one at a time.
820 if (c->has_video && os->segment_index > cur_flush_segment_index)
824 if (!os->init_range_length) {
825 av_write_frame(os->ctx, NULL);
826 os->init_range_length = avio_tell(os->ctx->pb);
828 ff_format_io_close(s, &os->out);
831 start_pos = avio_tell(os->ctx->pb);
833 if (!c->single_file) {
834 dash_fill_tmpl_params(filename, sizeof(filename), c->media_seg_name, i, os->segment_index, os->bit_rate, os->start_pts);
835 snprintf(full_path, sizeof(full_path), "%s%s", c->dirname, filename);
836 snprintf(temp_path, sizeof(temp_path), "%s.tmp", full_path);
837 ret = s->io_open(s, &os->out, temp_path, AVIO_FLAG_WRITE, NULL);
840 write_styp(os->ctx->pb);
842 snprintf(full_path, sizeof(full_path), "%s%s", c->dirname, os->initfile);
845 av_write_frame(os->ctx, NULL);
846 avio_flush(os->ctx->pb);
847 os->packets_written = 0;
849 range_length = avio_tell(os->ctx->pb) - start_pos;
850 if (c->single_file) {
851 find_index_range(s, full_path, start_pos, &index_length);
853 ff_format_io_close(s, &os->out);
854 ret = avpriv_io_move(temp_path, full_path);
858 add_segment(os, filename, os->start_pts, os->max_pts - os->start_pts, start_pos, range_length, index_length);
859 av_log(s, AV_LOG_VERBOSE, "Representation %d media segment %d written to: %s\n", i, os->segment_index, full_path);
862 if (c->window_size || (final && c->remove_at_exit)) {
863 for (i = 0; i < s->nb_streams; i++) {
864 OutputStream *os = &c->streams[i];
866 int remove = os->nb_segments - c->window_size - c->extra_window_size;
867 if (final && c->remove_at_exit)
868 remove = os->nb_segments;
870 for (j = 0; j < remove; j++) {
872 snprintf(filename, sizeof(filename), "%s%s", c->dirname, os->segments[j]->file);
874 av_free(os->segments[j]);
876 os->nb_segments -= remove;
877 memmove(os->segments, os->segments + remove, os->nb_segments * sizeof(*os->segments));
883 ret = write_manifest(s, final);
887 static int dash_write_packet(AVFormatContext *s, AVPacket *pkt)
889 DASHContext *c = s->priv_data;
890 AVStream *st = s->streams[pkt->stream_index];
891 OutputStream *os = &c->streams[pkt->stream_index];
892 int64_t seg_end_duration = (os->segment_index) * (int64_t) c->min_seg_duration;
895 ret = update_stream_extradata(s, os, st->codecpar);
899 // Fill in a heuristic guess of the packet duration, if none is available.
900 // The mp4 muxer will do something similar (for the last packet in a fragment)
901 // if nothing is set (setting it for the other packets doesn't hurt).
902 // By setting a nonzero duration here, we can be sure that the mp4 muxer won't
903 // invoke its heuristic (this doesn't have to be identical to that algorithm),
904 // so that we know the exact timestamps of fragments.
905 if (!pkt->duration && os->last_dts != AV_NOPTS_VALUE)
906 pkt->duration = pkt->dts - os->last_dts;
907 os->last_dts = pkt->dts;
909 // If forcing the stream to start at 0, the mp4 muxer will set the start
910 // timestamps to 0. Do the same here, to avoid mismatches in duration/timestamps.
911 if (os->first_pts == AV_NOPTS_VALUE &&
912 s->avoid_negative_ts == AVFMT_AVOID_NEG_TS_MAKE_ZERO) {
913 pkt->pts -= pkt->dts;
917 if (os->first_pts == AV_NOPTS_VALUE)
918 os->first_pts = pkt->pts;
920 if ((!c->has_video || st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO) &&
921 pkt->flags & AV_PKT_FLAG_KEY && os->packets_written &&
922 av_compare_ts(pkt->pts - os->first_pts, st->time_base,
923 seg_end_duration, AV_TIME_BASE_Q) >= 0) {
924 int64_t prev_duration = c->last_duration;
926 c->last_duration = av_rescale_q(pkt->pts - os->start_pts,
929 c->total_duration = av_rescale_q(pkt->pts - os->first_pts,
933 if ((!c->use_timeline || !c->use_template) && prev_duration) {
934 if (c->last_duration < prev_duration*9/10 ||
935 c->last_duration > prev_duration*11/10) {
936 av_log(s, AV_LOG_WARNING,
937 "Segment durations differ too much, enable use_timeline "
938 "and use_template, or keep a stricter keyframe interval\n");
942 if ((ret = dash_flush(s, 0, pkt->stream_index)) < 0)
946 if (!os->packets_written) {
947 // If we wrote a previous segment, adjust the start time of the segment
948 // to the end of the previous one (which is the same as the mp4 muxer
949 // does). This avoids gaps in the timeline.
950 if (os->max_pts != AV_NOPTS_VALUE)
951 os->start_pts = os->max_pts;
953 os->start_pts = pkt->pts;
955 if (os->max_pts == AV_NOPTS_VALUE)
956 os->max_pts = pkt->pts + pkt->duration;
958 os->max_pts = FFMAX(os->max_pts, pkt->pts + pkt->duration);
959 os->packets_written++;
960 return ff_write_chained(os->ctx, 0, pkt, s, 0);
963 static int dash_write_trailer(AVFormatContext *s)
965 DASHContext *c = s->priv_data;
967 if (s->nb_streams > 0) {
968 OutputStream *os = &c->streams[0];
969 // If no segments have been written so far, try to do a crude
970 // guess of the segment duration
971 if (!c->last_duration)
972 c->last_duration = av_rescale_q(os->max_pts - os->start_pts,
973 s->streams[0]->time_base,
975 c->total_duration = av_rescale_q(os->max_pts - os->first_pts,
976 s->streams[0]->time_base,
979 dash_flush(s, 1, -1);
981 if (c->remove_at_exit) {
984 for (i = 0; i < s->nb_streams; i++) {
985 OutputStream *os = &c->streams[i];
986 snprintf(filename, sizeof(filename), "%s%s", c->dirname, os->initfile);
995 static int dash_check_bitstream(struct AVFormatContext *s, const AVPacket *avpkt)
997 DASHContext *c = s->priv_data;
998 OutputStream *os = &c->streams[avpkt->stream_index];
999 AVFormatContext *oc = os->ctx;
1000 if (oc->oformat->check_bitstream) {
1002 AVPacket pkt = *avpkt;
1003 pkt.stream_index = 0;
1004 ret = oc->oformat->check_bitstream(oc, &pkt);
1006 AVStream *st = s->streams[avpkt->stream_index];
1007 AVStream *ost = oc->streams[0];
1008 st->internal->bsfcs = ost->internal->bsfcs;
1009 st->internal->nb_bsfcs = ost->internal->nb_bsfcs;
1010 ost->internal->bsfcs = NULL;
1011 ost->internal->nb_bsfcs = 0;
1018 #define OFFSET(x) offsetof(DASHContext, x)
1019 #define E AV_OPT_FLAG_ENCODING_PARAM
1020 static const AVOption options[] = {
1021 { "window_size", "number of segments kept in the manifest", OFFSET(window_size), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, E },
1022 { "extra_window_size", "number of segments kept outside of the manifest before removing from disk", OFFSET(extra_window_size), AV_OPT_TYPE_INT, { .i64 = 5 }, 0, INT_MAX, E },
1023 { "min_seg_duration", "minimum segment duration (in microseconds)", OFFSET(min_seg_duration), AV_OPT_TYPE_INT64, { .i64 = 5000000 }, 0, INT_MAX, E },
1024 { "remove_at_exit", "remove all segments when finished", OFFSET(remove_at_exit), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, E },
1025 { "use_template", "Use SegmentTemplate instead of SegmentList", OFFSET(use_template), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, E },
1026 { "use_timeline", "Use SegmentTimeline in SegmentTemplate", OFFSET(use_timeline), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, E },
1027 { "single_file", "Store all segments in one file, accessed using byte ranges", OFFSET(single_file), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, E },
1028 { "single_file_name", "DASH-templated name to be used for baseURL. Implies storing all segments in one file, accessed using byte ranges", OFFSET(single_file_name), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, E },
1029 { "init_seg_name", "DASH-templated name to used for the initialization segment", OFFSET(init_seg_name), AV_OPT_TYPE_STRING, {.str = "init-stream$RepresentationID$.m4s"}, 0, 0, E },
1030 { "media_seg_name", "DASH-templated name to used for the media segments", OFFSET(media_seg_name), AV_OPT_TYPE_STRING, {.str = "chunk-stream$RepresentationID$-$Number%05d$.m4s"}, 0, 0, E },
1034 static const AVClass dash_class = {
1035 .class_name = "dash muxer",
1036 .item_name = av_default_item_name,
1038 .version = LIBAVUTIL_VERSION_INT,
1041 AVOutputFormat ff_dash_muxer = {
1043 .long_name = NULL_IF_CONFIG_SMALL("DASH Muxer"),
1044 .priv_data_size = sizeof(DASHContext),
1045 .audio_codec = AV_CODEC_ID_AAC,
1046 .video_codec = AV_CODEC_ID_H264,
1047 .flags = AVFMT_GLOBALHEADER | AVFMT_NOFILE | AVFMT_TS_NEGATIVE,
1049 .write_header = dash_write_header,
1050 .write_packet = dash_write_packet,
1051 .write_trailer = dash_write_trailer,
1052 .deinit = dash_free,
1053 .codec_tag = (const AVCodecTag* const []){ ff_mp4_obj_type, 0 },
1054 .check_bitstream = dash_check_bitstream,
1055 .priv_class = &dash_class,