总概
ff_ffplay.c
read_thread
---.>>>avformat_open_input
utils.c
avformat_open_input
--->>>s->iformat->read_header(s)
mov.c
.read_header = mov_read_header,
mov_read_header
-->>>>mov_read_default
1、分析
if (ffp->iformat_name)
is->iformat = av_find_input_format(ffp->iformat_name);
err = avformat_open_input(&ic, is->filename, is->iformat, &ffp->format_opts);
当ffp->iformat_name没有指定时,即ffp->iformat_name,会直接走
avformat_open_input
然后is->iformat什么时候解析呢?同时s->iformat->read_header(s)为什么就会走到mov下的read_header?带着这些问题,分析了avformat_open_input流程
avformat_open_input
--->>init_input
1.是否设置了pb,如果设置了就直接av_probe_input_buffer2.一般情况下pb均为空,有特殊情况,下文有做说明
2.是否设置了iformat,如果设置了直接返回;
如果没指定iformat,但是可以从文件名中猜出iformat,也成功.
3.hls一般会走到这里io_open,如果从文件名中也猜不出媒体格式,则只能打开这个文件进行探测了,先打开文件
--->>s->io_open
--->>>io_open_default
--->>>ffio_open_whitelist
--->>>ffurl_open_whitelist
-->>>ffurl_alloc
//探测是HTTP协议 URLProtocol ff_http_protocol
url_find_protocol(filename);
ffurl_connect //发送HTTP报文头,下载http.xxx.m3u8文件
--->>>ffurl_connect
--->>>url_open2或者url_open
---->>>av_probe_input_buffer2
--->>>av_probe_input_format2/* Guess file format. */会通过根据格式read_probe来选择最高分的解码器
--->>s->iformat->read_header(s)
其中io_open定义在avformat.h
int (*io_open)(struct AVFormatContext *s, AVIOContext **pb, const char *url,
int flags, AVDictionary **options);
然而io_open调用哪个文件的io_open呢?跟踪发现avformat_alloc_context
opotions.c
AVFormatContext *avformat_alloc_context(void)
{
AVFormatContext *ic;
ic = av_malloc(sizeof(AVFormatContext));
if (!ic) return ic;
avformat_get_context_defaults(ic);
return ic;
}
这个时候会调用avformat_get_context_defaults
static void avformat_get_context_defaults(AVFormatContext *s)
{
s->io_open = io_open_default;
s->io_close = io_close_default;
av_opt_set_defaults(s);
}
定义了io_open和io_close的函数指针,io_open_default和io_close_default
在 io_open_default又做了哪些操作?主要调用了ffio_open_whitelist
int ffio_open_whitelist(AVIOContext **s, const char *filename, int flags,
const AVIOInterruptCB *int_cb, AVDictionary **options,
const char *whitelist, const char *blacklist
)
{
err = ffurl_open_whitelist(&h, filename, flags, int_cb, options, whitelist, blacklist, NULL);
err = ffio_fdopen(s, h);
}
int ffurl_open_whitelist(URLContext **puc, const char *filename, int flags,
const AVIOInterruptCB *int_cb, AVDictionary **options,
const char *whitelist, const char* blacklist,
URLContext *parent)
{
AVDictionary *tmp_opts = NULL;
AVDictionaryEntry *e;
int ret = ffurl_alloc(puc, filename, flags, int_cb);
ret = ffurl_connect(*puc, options);
}
ffurl_open_whitelist主要执行ffurl_alloc和ffurl_connect
url_open根据协议调用配对的open2,比如http协议,即调用http.c的http_open
const URLProtocol ff_https_protocol = {
.name = "https",
.url_open2 = http_open,
.url_read = http_read,
.url_write = http_write,
.url_seek = http_seek,
.url_close = http_close,
.url_get_file_handle = http_get_file_handle,
.url_get_short_seek = http_get_short_seek,
.url_shutdown = http_shutdown,
.priv_data_size = sizeof(HTTPContext),
.priv_data_class = &https_context_class,
.flags = URL_PROTOCOL_FLAG_NETWORK,
.default_whitelist = "http,https,tls,rtp,tcp,udp,crypto,httpproxy"
};
回到前面
avformat_open_input
--->>init_input
--->>s->io_open
--->>s->iformat->read_header(s)
//p是流的指针,其中buf就是原始数据
static int mov_probe(AVProbeData *p)
{
int64_t offset;
uint32_t tag;
int score = 0;
int moov_offset = -1;
/* check file header */
offset = 0;
for (;;) {
/* ignore invalid offset */
if ((offset + 8) > (unsigned int)p->buf_size)
break;
tag = AV_RL32(p->buf + offset + 4);//将p->buf+4即‘f’ ‘t’ 'y' 'p'组成一个int
switch(tag) {//判断这个int是不是ftyp
/* check for obvious tags */
case MKTAG('m','o','o','v'):
moov_offset = offset + 4;
case MKTAG('m','d','a','t'):
case MKTAG('p','n','o','t'): /* detect movs with preview pics like ew.mov and april.mov */
case MKTAG('u','d','t','a'): /* Packet Video PVAuthor adds this and a lot of more junk */
case MKTAG('f','t','y','p'):
if (AV_RB32(p->buf+offset) < 8 &&
(AV_RB32(p->buf+offset) != 1 ||
offset + 12 > (unsigned int)p->buf_size ||
AV_RB64(p->buf+offset + 8) == 0)) {
score = FFMAX(score, AVPROBE_SCORE_EXTENSION);
} else if (tag == MKTAG('f','t','y','p') &&//如果是的话score=100,说明就是mov格式的文件
( AV_RL32(p->buf + offset + 8) == MKTAG('j','p','2',' ')
|| AV_RL32(p->buf + offset + 8) == MKTAG('j','p','x',' ')
)) {
score = FFMAX(score, 5);
} else {
score = AVPROBE_SCORE_MAX;
}
offset = FFMAX(4, AV_RB32(p->buf+offset)) + offset;
break;
/* those are more common words, so rate then a bit less */
case MKTAG('e','d','i','w'): /* xdcam files have reverted first tags */
case MKTAG('w','i','d','e'):
case MKTAG('f','r','e','e'):
case MKTAG('j','u','n','k'):
case MKTAG('p','i','c','t'):
score = FFMAX(score, AVPROBE_SCORE_MAX - 5);
offset = FFMAX(4, AV_RB32(p->buf+offset)) + offset;
break;
case MKTAG(0x82,0x82,0x7f,0x7d):
case MKTAG('s','k','i','p'):
case MKTAG('u','u','i','d'):
case MKTAG('p','r','f','l'):
/* if we only find those cause probedata is too small at least rate them */
score = FFMAX(score, AVPROBE_SCORE_EXTENSION);
offset = FFMAX(4, AV_RB32(p->buf+offset)) + offset;
break;
default:
offset = FFMAX(4, AV_RB32(p->buf+offset)) + offset;
}
}
if(score > AVPROBE_SCORE_MAX - 50 && moov_offset != -1) {
/* moov atom in the header - we should make sure that this is not a
* MOV-packed MPEG-PS */
offset = moov_offset;
//p->buf_size = 2048
while(offset < (p->buf_size - 16)){ /* Sufficient space */
/* We found an actual hdlr atom */
if(AV_RL32(p->buf + offset ) == MKTAG('h','d','l','r') &&
AV_RL32(p->buf + offset + 8) == MKTAG('m','h','l','r') &&
AV_RL32(p->buf + offset + 12) == MKTAG('M','P','E','G')){
av_log(NULL, AV_LOG_WARNING, "Found media data tag MPEG indicating this is a MOV-packed MPEG-PS.\n");
//判断是否是MPEG-PS-in-MOV,如果是就返回scroe = 5
/* We found a media handler reference atom describing an
* MPEG-PS-in-MOV, return a
* low score to force expanding the probe window until
* mpegps_probe finds what it needs */
return 5;
}else
/* Keep looking */
offset+=2;
}
}
return score;
}
mov_probe会得到score,其调用流程
init_input会得到iformat_name为mp4,从而read_header会调用mov.c文件里的mov_read_header
AVInputFormat ff_mov_demuxer = {
.name = "mov,mp4,m4a,3gp,3g2,mj2",
.long_name = NULL_IF_CONFIG_SMALL("QuickTime / MOV"),
.priv_class = &mov_class,
.priv_data_size = sizeof(MOVContext),
.extensions = "mov,mp4,m4a,3gp,3g2,mj2",
.read_probe = mov_probe,
.read_header = mov_read_header,
.read_packet = mov_read_packet,
.read_close = mov_read_close,
.read_seek = mov_read_seek,
.flags = AVFMT_NO_BYTE_SEEK,
};
下面开始分析mov_read_header
//在代码中atom,其实就是MP4协议中的box
typedef struct MOVAtom {
uint32_t type;//box类型
int64_t size; /* box整体大小(size+type+body三部分) total size (excluding the size and type fields) */
} MOVAtom;
static int mov_read_header(AVFormatContext *s)
{
MOVContext *mov = s->priv_data; //AVClass类型 memset 为0
AVIOContext *pb = s->pb;
int j, err;
MOVAtom atom = { AV_RL32("root") }; //创建父box,包含ftyp、moov、mdat三种类型的box并赋初值
int i;
...
mov->fc = s; //将AVFormatContext 赋值给MOVContext 方便书写,代码具有扩展性
mov->trak_index = -1;
/* .mov and .mp4 aren't streamable anyway (only progressive download if moov is before mdat) */
if (pb->seekable & AVIO_SEEKABLE_NORMAL)
atom.size = avio_size(pb); //源MP4文件整体大小
else
atom.size = INT64_MAX;
/* check MOV header */
//当读完moov的box时跳出,但这里只读一次,因为有mov_read_default接口会不断向下读取嵌套box,只要atom.size还有足够数据
do {
if (mov->moov_retry)
avio_seek(pb, 0, SEEK_SET);
//有嵌套BOX,继续往下读
if ((err = mov_read_default(mov, pb, atom)) < 0) {
av_log(s, AV_LOG_ERROR, "error reading header\n");
goto fail;
}
} while ((pb->seekable & AVIO_SEEKABLE_NORMAL) && !mov->found_moov && !mov->moov_retry++);
if (!mov->found_moov) { //moov是否读完标志
av_log(s, AV_LOG_ERROR, "moov atom not found\n");
err = AVERROR_INVALIDDATA;
goto fail;
}
av_log(mov->fc, AV_LOG_TRACE, "on_parse_exit_offset=%"PRId64"\n", avio_tell(pb));
if (pb->seekable & AVIO_SEEKABLE_NORMAL) {
......
}
/* copy timecode metadata from tmcd tracks to the related video streams */
for (i = 0; i < s->nb_streams; i++) {
...
}
export_orphan_timecode(s);
//s->streams是在读取trak box开的内存并赋值的
for (i = 0; i < s->nb_streams; i++) {
AVStream *st = s->streams[i];
MOVStreamContext *sc = st->priv_data;
fix_timescale(mov, sc); //mdhd中的时间缩放比例,一般为1
if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO &&
st->codecpar->codec_id == AV_CODEC_ID_AAC) {
st->internal->skip_samples = sc->start_pad;
}
if (st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO && sc->nb_frames_for_fps > 0 && sc->duration_for_fps > 0)
av_reduce(&st->avg_frame_rate.num, &st->avg_frame_rate.den,
sc->time_scale*(int64_t)sc->nb_frames_for_fps, sc->duration_for_fps, INT_MAX);
if (st->codecpar->codec_type == AVMEDIA_TYPE_SUBTITLE) {
if (st->codecpar->width <= 0 || st->codecpar->height <= 0) {
st->codecpar->width = sc->width;
st->codecpar->height = sc->height;
}
if (st->codecpar->codec_id == AV_CODEC_ID_DVD_SUBTITLE) {
if ((err = mov_rewrite_dvd_sub_extradata(st)) < 0)
goto fail;
}
}
...
}
if (mov->trex_data) {
...
}
if (mov->use_mfra_for > 0) {
...
}
for (i = 0; i < mov->bitrates_count && i < s->nb_streams; i++) {
if (mov->bitrates[i]) {
s->streams[i]->codecpar->bit_rate = mov->bitrates[i];//计算码率
}
}
ff_rfps_calculate(s);
for (i = 0; i < s->nb_streams; i++) {
AVStream *st = s->streams[i];
MOVStreamContext *sc = st->priv_data;
switch (st->codecpar->codec_type) {
case AVMEDIA_TYPE_AUDIO:
err = ff_replaygain_export(st, s->metadata);
if (err < 0)
goto fail;
break;
case AVMEDIA_TYPE_VIDEO:
...
}
ff_configure_buffers_for_index(s, AV_TIME_BASE);
for (i = 0; i < mov->frag_index.nb_items; i++)
if (mov->frag_index.item[i].moof_offset <= mov->fragment.moof_offset)
mov->frag_index.item[i].headers_read = 1;
return 0;
fail:
mov_read_close(s);
return err;
}
//将所有解析atom的接口都注册到一起
typedef struct MOVParseTableEntry {
uint32_t type;
int (*parse)(MOVContext *ctx, AVIOContext *pb, MOVAtom atom);
} MOVParseTableEntry;
static const MOVParseTableEntry mov_default_parse_table[] = {
{ MKTAG('f','t','y','p'), mov_read_ftyp },
{ MKTAG('t','r','a','k'), mov_read_trak },
{ MKTAG('t','r','a','f'), mov_read_default },
{ MKTAG('s','t','s','c'), mov_read_stsc },
{ MKTAG('s','t','s','d'), mov_read_stsd }, /* sample description */
{ MKTAG('s','t','s','s'), mov_read_stss }, /* sync sample */
{ MKTAG('s','t','s','z'), mov_read_stsz }, /* sample size */
{ MKTAG('s','t','t','s'), mov_read_stts },
{ MKTAG('s','t','z','2'), mov_read_stsz }, /* compact sample size */
.....
}
//这个接口是最重要的接口,主要是负责读取嵌套的box,里面有循环代码,只要atom.size数据足够多的
//比如 moov->trak->mdia->stbl->stsd、stts、stss、ctts、stco等
//所以上面mov_read_header接口中只用执行一次mov_read_default就行了,就是从创建的父box开始循环读取嵌套box
static int mov_read_default(MOVContext *c, AVIOContext *pb, MOVAtom atom)
{
int64_t total_size = 0;//读取一个atom(无论是否是嵌套box),已经读取的字节数
MOVAtom a;
int i;
if (c->atom_depth > 10) {
av_log(c->fc, AV_LOG_ERROR, "Atoms too deeply nested\n");
return AVERROR_INVALIDDATA;
}
c->atom_depth ++; //atom读取层数
if (atom.size < 0)
atom.size = INT64_MAX;
while (total_size <= atom.size - 8 && !avio_feof(pb)) {
int (*parse)(MOVContext*, AVIOContext*, MOVAtom) = NULL;
a.size = atom.size;
a.type=0;
if (atom.size >= 8) {
a.size = avio_rb32(pb);//该box大小(size+type+body大小)
a.type = avio_rl32(pb);//该box类型
//类型比对
if (((a.type == MKTAG('f','r','e','e') && c->moov_retry) ||
a.type == MKTAG('h','o','o','v')) &&
a.size >= 8 &&
c->fc->strict_std_compliance < FF_COMPLIANCE_STRICT) {
uint32_t type;
avio_skip(pb, 4);
type = avio_rl32(pb);
avio_seek(pb, -8, SEEK_CUR);
if (type == MKTAG('m','v','h','d') ||
type == MKTAG('c','m','o','v')) {
av_log(c->fc, AV_LOG_ERROR, "Detected moov in a free or hoov atom.\n");
a.type = MKTAG('m','o','o','v');
}
}
if (atom.type != MKTAG('r','o','o','t') &&
atom.type != MKTAG('m','o','o','v')) {
if (a.type == MKTAG('t','r','a','k') ||
a.type == MKTAG('m','d','a','t')) {
av_log(c->fc, AV_LOG_ERROR, "Broken file, trak/mdat not at top-level\n");
avio_skip(pb, -8);
c->atom_depth --;
return 0;
}
}
total_size += 8; //已经读了8个字节
if (a.size == 1 && total_size + 8 <= atom.size) { /* 64 bit extended size */
a.size = avio_rb64(pb) - 8;
total_size += 8;
}
}
av_log(c->fc, AV_LOG_TRACE, "type:'%s' parent:'%s' sz: %"PRId64" %"PRId64" %"PRId64"\n",
av_fourcc2str(a.type), av_fourcc2str(atom.type), a.size, total_size, atom.size);
if (a.size == 0) {
a.size = atom.size - total_size + 8;
}
a.size -= 8;
if (a.size < 0)
break;
a.size = FFMIN(a.size, atom.size - total_size);
//通过比对将对应的解析box接口进行赋值
for (i = 0; mov_default_parse_table[i].type; i++)
if (mov_default_parse_table[i].type == a.type) {
parse = mov_default_parse_table[i].parse;
break;
}
// container is user data
if (!parse && (atom.type == MKTAG('u','d','t','a') ||
atom.type == MKTAG('i','l','s','t')))
parse = mov_read_udta_string;
// Supports parsing the QuickTime Metadata Keys.
// https://developer.apple.com/library/mac/documentation/QuickTime/QTFF/Metadata/Metadata.html
if (!parse && c->found_hdlr_mdta &&
atom.type == MKTAG('m','e','t','a') &&
a.type == MKTAG('k','e','y','s') &&
c->meta_keys_count == 0) {
parse = mov_read_keys;
}
if (!parse) { /* skip leaf atoms data */
avio_skip(pb, a.size);//起播的时候,会先seek到文件moov分区起始地址
} else {
int64_t start_pos = avio_tell(pb);//当前文件指针位置和文件首地址的偏移,为了验证该box或atom是否已经读完
int64_t left;
int err = parse(c, pb, a);
if (err < 0) {
c->atom_depth --;
return err;
}
if (c->found_moov && c->found_mdat &&
((!(pb->seekable & AVIO_SEEKABLE_NORMAL) || c->fc->flags & AVFMT_FLAG_IGNIDX || c->frag_index.complete) ||
start_pos + a.size == avio_size(pb))) {
if (!(pb->seekable & AVIO_SEEKABLE_NORMAL) || c->fc->flags & AVFMT_FLAG_IGNIDX || c->frag_index.complete)
c->next_root_atom = start_pos + a.size;
c->atom_depth --;
return 0;
}
left = a.size - avio_tell(pb) + start_pos;//验证 该box或atom是否已经读完
if (left > 0) /* skip garbage at atom end */
avio_skip(pb, left);//起播的时候,会先seek到文件结尾
else if (left < 0) {
av_log(c->fc, AV_LOG_WARNING,
"overread end of atom '%s' by %"PRId64" bytes\n",
av_fourcc2str(a.type), -left);
avio_seek(pb, left, SEEK_CUR);
}
}
total_size += a.size;//一个atom已经读完
}
if (total_size < atom.size && atom.size < 0x7ffff)
avio_skip(pb, atom.size - total_size);
c->atom_depth --;
return 0;
}
mov_read_default解析了如下分区
顺便说一下,mp4起播过程会seek两次,第一次是离文件结尾73结束的位置的free分区,第二次是seek到moov分区的起始地址
重要发现
起播过程,左图会seek两次,右图不会seek
同一个视频源,一个是带free分区,一个是没有free分区,带free分区的,起播会先seek到free分区,然后再seek到moov分区,也就是说起播会有两次seek,而不带free分区的视频源,起播不会seek
结论:
把free分区去掉或者mp4文件只流ftyp/moov/mdat三个分区,可以提高起播速度
经测试发现
如果文件长度设为free分区的起始地址
带有free分区的文件也不会有seek操作,如果带有free分区的视频,长度是本身文件的长度,就会有两次seek