ffmpeg
对于音视频开发的同学想必都不陌生,它作为被普遍认可的编解码开源库,很多播放器都采用它来完成编解码的功能。作为一个Android开发人员,可以通过ffmepg自带的播放器ffplay
学习ffmpeg是如何拉取多媒体数据、解码、av同步的机制。
本文主要从URLProtocol的视角探究ffplay在demux数据之前是如何拉取数据的。
stream_open : 在ffplay::main函数中,负责根据filename决定拉取数据的协议,以及生成URLProtocol等数据
demux : 解码. 例如h264数据转换成yuv的过程,或者是aac音频数据转换成pcm的过程
解封装 : 在解码之前,用于将封装好的多媒体数据解成一个个视频轨、音频轨、字幕轨,对于不同的媒体资源格式有不同的解封装过程
在阅读源码分析之前,请确保清晰以下概念(大神请无视):
在上述的流程调用中,从ffurl_alloc函数开始进行URLProtocol和URLContext的初始化。如下所示:
int ffurl_alloc(URLContext **puc, const char *filename, int flags,
const AVIOInterruptCB *int_cb)
{
const URLProtocol *p = NULL;
// p -> ff_http_protocol
p = url_find_protocol(filename);
//初始化URLContext,并将protocol赋值进去
if (p)
return url_alloc_for_protocol(puc, p, filename, flags, int_cb);
//代码执行到此,代表URLContext初始化失败
*puc = NULL;
if (av_strstart(filename, "https:", NULL))
av_log(NULL, AV_LOG_WARNING, "https protocol not found, recompile FFmpeg with "
"openssl, gnutls "
"or securetransport enabled.\n");
return AVERROR_PROTOCOL_NOT_FOUND;
}
const URLProtocol ff_http_protocol = {
.name = “http”,
.url_open2 = http_open,
.url_read = http_read,
.url_write = http_write,
.priv_data_size = sizeof(HTTPContext),
};
url_find_protocol函数主要通过filename进行探查.如果匹配,则返回对应的URLProtocol,ffmpeg将所有支持的protocol以URLProtocol常量保存在protocol_list.c中。
static const struct URLProtocol *url_find_protocol(const char *filename)
{
const URLProtocol **protocols;
char proto_str[128], proto_nested[128], *ptr;
size_t proto_len = strspn(filename, URL_SCHEME_CHARS);
int i;
//filename == http://xxxx, proto_len = 4
if (filename[proto_len] != ':' &&
(strncmp(filename, "subfile,", 8) || !strchr(filename + proto_len + 1, ':')) ||
is_dos_path(filename))
strcpy(proto_str, "file");
else
av_strlcpy(proto_str, filename,
FFMIN(proto_len + 1, sizeof(proto_str))); //sizeof(proto_str) == 128, so proto_str == 'http'
if ((ptr = strchr(proto_str, ',')))
*ptr = '\0';
//proto_nested -> 'http'
av_strlcpy(proto_nested, proto_str, sizeof(proto_nested));
if ((ptr = strchr(proto_nested, '+')))
*ptr = '\0';
//将常量url_protocols装配到数组中,url_protocals -> [ff_http_protocol,....]
protocols = ffurl_get_protocols(NULL, NULL);
//分配失败,直接返回
if (!protocols)
return NULL;
for (i = 0; protocols[i]; i++) {
const URLProtocol *up = protocols[i];
//按照name属性匹配URLProtocal
if (!strcmp(proto_str, up->name)) {
av_freep(&protocols);
return up;
}
if (up->flags & URL_PROTOCOL_FLAG_NESTED_SCHEME &&
!strcmp(proto_nested, up->name)) {
av_freep(&protocols);
return up;
}
}
av_freep(&protocols);
return NULL;
}
由http_connect函数进入,http_open_cnx_internal函数主要完成底层protocol的生成,以及与服务器进行握手(通过底层protocol发送报文)
static int http_open_cnx_internal(URLContext *h, AVDictionary **options)
{
//默认底层协议为tcp
const char *path, *proxy_path, *lower_proto = "tcp", *local_path;
char hostname[1024], hoststr[1024], proto[10];
char auth[1024], proxyauth[1024] = "";
char path1[MAX_URL_SIZE];
char buf[1024], urlbuf[MAX_URL_SIZE];
int port, use_proxy, err, location_changed = 0;
HTTPContext *s = h->priv_data;
//av_url_split ->
av_url_split(proto, sizeof(proto), auth, sizeof(auth),
hostname, sizeof(hostname), &port,
path1, sizeof(path1), s->location);
ff_url_join(hoststr, sizeof(hoststr), NULL, NULL, hostname, port, NULL);
if (!strcmp(proto, "https")) {
lower_proto = "tls";
use_proxy = 0;
if (port < 0)
port = 443;
}
if (port < 0)
port = 80;
if (path1[0] == '\0')
path = "/";
else
path = path1;
local_path = path;
....
//拼凑lower protocol字符串(buf) -> tcp://{hostname}:{port}
ff_url_join(buf, sizeof(buf), lower_proto, NULL, hostname, port, NULL);
//s: HTTPContext, s->hd: URLContext
if (!s->hd) {
//匹配tcp protocal
err = ffurl_open_whitelist(&s->hd, buf, AVIO_FLAG_READ_WRITE,
&h->interrupt_callback, options,
h->protocol_whitelist, h->protocol_blacklist, h);
if (err < 0)
return err;
}
//进行http连接
err = http_connect(h, path, local_path, hoststr,
auth, proxyauth, &location_changed);
if (err < 0)
return err;
return location_changed;
}
该函数的主要作用是调用底层protocol发送报文,写入请求header并读取服务器返回的结果存入HTTPContext。例如,当filename为http://xxxx:port时,URLProtocol对应ff_http_protocol,同时private_data中嵌套了URLContext,该成员中还存放着对应ff_tcp_protocol的URLProtocol,这也是tcp在ffmpeg中视作lower protocol的来由。
static int http_connect(URLContext *h, const char *path, const char *local_path,
const char *hoststr, const char *auth,
const char *proxyauth, int *new_location)
{
HTTPContext *s = h->priv_data;
int post, err;
char headers[HTTP_HEADERS_SIZE] = "";
char *authstr = NULL, *proxyauthstr = NULL;
uint64_t off = s->off;
int len = 0;
const char *method;
int send_expect_100 = 0;
int ret;
/* send http header */
//第一次初始化AVFormatContext时,为FLAG_READ
post = h->flags & AVIO_FLAG_WRITE;
if (s->post_data) {
/* force POST method and disable chunked encoding when
* custom HTTP post data is set */
post = 1;
s->chunked_post = 0;
}
//method -> "GET"
if (s->method)
method = s->method;
else
method = post ? "POST" : "GET";
....
/* set default headers if needed */
if (!has_header(s->headers, "\r\nUser-Agent: "))
len += av_strlcatf(headers + len, sizeof(headers) - len,
"User-Agent: %s\r\n", s->user_agent);
if (!has_header(s->headers, "\r\nAccept: "))
len += av_strlcpy(headers + len, "Accept: */*\r\n",
sizeof(headers) - len);
// Note: we send this on purpose even when s->off is 0 when we're probing,
// since it allows us to detect more reliably if a (non-conforming)
// server supports seeking by analysing the reply headers.
if (!has_header(s->headers, "\r\nRange: ") && !post && (s->off > 0 || s->end_off || s->seekable == -1)) {
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Range: bytes=%"PRIu64"-", s->off);
if (s->end_off)
len += av_strlcatf(headers + len, sizeof(headers) - len,
"%"PRId64, s->end_off - 1);
len += av_strlcpy(headers + len, "\r\n",
sizeof(headers) - len);
}
if (send_expect_100 && !has_header(s->headers, "\r\nExpect: "))
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Expect: 100-continue\r\n");
if (!has_header(s->headers, "\r\nConnection: ")) {
if (s->multiple_requests)
len += av_strlcpy(headers + len, "Connection: keep-alive\r\n",
sizeof(headers) - len);
else
len += av_strlcpy(headers + len, "Connection: close\r\n",
sizeof(headers) - len);
}
if (!has_header(s->headers, "\r\nHost: "))
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Host: %s\r\n", hoststr);
if (!has_header(s->headers, "\r\nContent-Length: ") && s->post_data)
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Content-Length: %d\r\n", s->post_datalen);
if (!has_header(s->headers, "\r\nContent-Type: ") && s->content_type)
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Content-Type: %s\r\n", s->content_type);
if (!has_header(s->headers, "\r\nCookie: ") && s->cookies) {
char *cookies = NULL;
if (!get_cookies(s, &cookies, path, hoststr) && cookies) {
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Cookie: %s\r\n", cookies);
av_free(cookies);
}
}
if (!has_header(s->headers, "\r\nIcy-MetaData: ") && s->icy)
len += av_strlcatf(headers + len, sizeof(headers) - len,
"Icy-MetaData: %d\r\n", 1);
/* now add in custom headers */
if (s->headers)
av_strlcpy(headers + len, s->headers, sizeof(headers) - len);
ret = snprintf(s->buffer, sizeof(s->buffer),
"%s %s HTTP/1.1\r\n"
"%s"
"%s"
"%s"
"%s%s"
"\r\n",
method,
path,
post && s->chunked_post ? "Transfer-Encoding: chunked\r\n" : "",
headers,
authstr ? authstr : "",
proxyauthstr ? "Proxy-" : "", proxyauthstr ? proxyauthstr : "");
av_log(h, AV_LOG_DEBUG, "request: %s\n", s->buffer);
if (strlen(headers) + 1 == sizeof(headers) ||
ret >= sizeof(s->buffer)) {
av_log(h, AV_LOG_ERROR, "overlong headers\n");
err = AVERROR(EINVAL);
goto done;
}
//写入请求的header
if ((err = ffurl_write(s->hd, s->buffer, strlen(s->buffer))) < 0)
goto done;
if (s->post_data)
if ((err = ffurl_write(s->hd, s->post_data, s->post_datalen)) < 0)
goto done;
/* init input buffer */
s->buf_ptr = s->buffer;
s->buf_end = s->buffer;
s->line_count = 0;
s->off = 0;
s->icy_data_read = 0;
s->filesize = UINT64_MAX;
s->willclose = 0;
s->end_chunked_post = 0;
s->end_header = 0;
if (post && !s->post_data && !send_expect_100) {
/* Pretend that it did work. We didn't read any header yet, since
* we've still to send the POST data, but the code calling this
* function will check http_code after we return. */
s->http_code = 200;
err = 0;
goto done;
}
/* wait for header */
//读取header
err = http_read_header(h, new_location);
if (err < 0)
goto done;
if (*new_location)
s->off = off;
err = (off == s->off) ? 0 : -1;
done:
av_freep(&authstr);
av_freep(&proxyauthstr);
return err;
}
在根据url生成URLProtocol以及根据读取服务器返回的header初始化HTTPContext后,需要进一步的判断媒体资源属于哪种格式,适用于哪种解码器进行demux.
/* Open input file and probe the format if necessary. */
static int init_input(AVFormatContext *s, const char *filename,
AVDictionary **options)
{
int ret;
AVProbeData pd = { filename, NULL, 0 };
int score = AVPROBE_SCORE_RETRY;
//usually no
if (s->pb) {
s->flags |= AVFMT_FLAG_CUSTOM_IO;
if (!s->iformat)
return av_probe_input_buffer2(s->pb, &s->iformat, filename,
s, 0, s->format_probesize);
else if (s->iformat->flags & AVFMT_NOFILE)
av_log(s, AV_LOG_WARNING, "Custom AVIOContext makes no sense and "
"will be ignored with AVFMT_NOFILE format.\n");
return 0;
}
//对应iformat已经初始化的情况
if ((s->iformat && s->iformat->flags & AVFMT_NOFILE) ||
(!s->iformat && (s->iformat = av_probe_input_format2(&pd, 0, &score))))
return score;
//call avio_open2, s->pb : AVIOContext
if ((ret = s->io_open(s, &s->pb, filename, AVIO_FLAG_READ | s->avio_flags, options)) < 0)
return ret;
return av_probe_input_buffer2(s->pb, &s->iformat, filename,
s, 0, s->format_probesize);
}
该函数的主要行为是通过读取数据匹配到合适的解码器,比如针对aac音频,那匹配的就是ff_aac_demuxer解码器。
int av_probe_input_buffer2(AVIOContext *pb, AVInputFormat **fmt,
const char *filename, void *logctx,
unsigned int offset, unsigned int max_probe_size)
{
AVProbeData pd = { filename ? filename : "" };
uint8_t *buf = NULL;
int ret = 0, probe_size, buf_offset = 0;
int score = 0;
int ret2;
//s->format_probesize = 0, PROBE_BUF_MAX == 1<<20
if (!max_probe_size)
max_probe_size = PROBE_BUF_MAX;
//URLContext
if (pb->av_class) {
uint8_t *mime_type_opt = NULL;
char *semi;
//在读取http协议头时获得,获取
av_opt_get(pb, "mime_type", AV_OPT_SEARCH_CHILDREN, &mime_type_opt);
pd.mime_type = (const char *)mime_type_opt;
semi = pd.mime_type ? strchr(pd.mime_type, ';') : NULL;
if (semi) {
*semi = '\0';
}
}
//PROBE_BUF_MIN -> 2048
//probe的大小从2048开始,随后以2倍大小增加
for (probe_size = PROBE_BUF_MIN; probe_size <= max_probe_size && !*fmt;
//每次循环都会增加1倍的probe_size
probe_size = FFMIN(probe_size << 1,
FFMAX(max_probe_size, probe_size + 1))) {
//AVPROBE_SCORE_RETRY = 25
score = probe_size < max_probe_size ? AVPROBE_SCORE_RETRY : 0;
/* Read probe data. */
//分配buf内存空间
if ((ret = av_reallocp(&buf, probe_size + AVPROBE_PADDING_SIZE)) < 0)
goto fail;
//读取多媒体数据
//1.读到数据.ret > 0
//2.没读到数据走if分支
if ((ret = avio_read(pb, buf + buf_offset,
probe_size - buf_offset)) < 0) {
/* Fail if error was not end of file, otherwise, lower score. */
if (ret != AVERROR_EOF)
goto fail;
score = 0;
ret = 0; /* error was end of file, nothing read */
}
//buf_offset初始化时为0
buf_offset += ret;
if (buf_offset < offset)
continue;
pd.buf_size = buf_offset - offset;
pd.buf = &buf[offset];
//设置probe data末尾的extra allocated bytes为0
memset(pd.buf + pd.buf_size, 0, AVPROBE_PADDING_SIZE);
/* Guess file format. */
*fmt = av_probe_input_format2(&pd, 1, &score);
if (*fmt) {
/* This can only be true in the last iteration. */
if (score <= AVPROBE_SCORE_RETRY) {
av_log(logctx, AV_LOG_WARNING,
"Format %s detected only with low score of %d, "
"misdetection possible!\n", (*fmt)->name, score);
} else
av_log(logctx, AV_LOG_DEBUG,
"Format %s probed with size=%d and score=%d\n",
(*fmt)->name, probe_size, score);
#if 0
FILE *f = fopen("probestat.tmp", "ab");
fprintf(f, "probe_size:%d format:%s score:%d filename:%s\n", probe_size, (*fmt)->name, score, filename);
fclose(f);
#endif
}
}
if (!*fmt)
ret = AVERROR_INVALIDDATA;
fail:
/* Rewind. Reuse probe buffer to avoid seeking. */
ret2 = ffio_rewind_with_probe_data(pb, &buf, buf_offset);
if (ret >= 0)
ret = ret2;
av_freep(&pd.mime_type);
return ret < 0 ? ret : score;
}
AVFormatContext
时并没有确定probe data的大小,所以这里会统一设置为1<<20AVProbeData
结构,并把之前URLProtocol读取到的mime_type赋值到其中av_probe_input_format3
)进行format guess,并打分。ffmpeg自带的数据读取函数,读取数据之后会存放在AVIOContext->buffer之中
/**
* size -> probe_size,第一次probe_size为2048
**/
int avio_read(AVIOContext *s, unsigned char *buf, int size)
{
int len, size1;
size1 = size;
while (size > 0) {
len = FFMIN(s->buf_end - s->buf_ptr, size);
//s-> write_flag = 0
if (len == 0 || s->write_flag) {
//s->update_checksum = null
//s->buffer_size = IO_BUFFER_SIZE //32768
if((s->direct || size > s->buffer_size) && !s->update_checksum) {
// bypass the buffer and read data directly into buf
if(s->read_packet)
len = s->read_packet(s->opaque, buf, size);
if (len <= 0) {
/* do not modify buffer if EOF reached so that a seek back can
be done without rereading data */
s->eof_reached = 1;
if(len<0)
s->error= len;
break;
} else {
s->pos += len;
s->bytes_read += len;
size -= len;
buf += len;
// reset the buffer
s->buf_ptr = s->buffer;
s->buf_end = s->buffer/* + len*/;
}
} else {
//goto this
//通过tcp读取一次数据,len!=0
fill_buffer(s);
len = s->buf_end - s->buf_ptr;
//如果数据读取完毕,终止循环
if (len == 0)
break;
}
} else {
//下一次循环,goto this
//将s->buf_ptr的内容复制到buf
memcpy(buf, s->buf_ptr, len);
buf += len;
s->buf_ptr += len;
//size减去已读数据的长度
size -= len;
}
}
....
return size1 - size;
}
AVIOContext
->buffer中AVIOContext
->buffer拷贝到目标buf中av_probe_input_format3
函数的主要行为是format guess.
AVInputFormat *av_probe_input_format3(AVProbeData *pd, int is_opened,
int *score_ret)
{
AVProbeData lpd = *pd;
AVInputFormat *fmt1 = NULL, *fmt;
int score, score_max = 0;
const static uint8_t zerobuffer[AVPROBE_PADDING_SIZE];
enum nodat {
NO_ID3,
ID3_ALMOST_GREATER_PROBE,
ID3_GREATER_PROBE,
ID3_GREATER_MAX_PROBE,
} nodat = NO_ID3;
if (!lpd.buf)
lpd.buf = (unsigned char *) zerobuffer;
//如果含id3信息(通常是mp3),则将buf移动到id3信息之后数据部分之前
if (lpd.buf_size > 10 && ff_id3v2_match(lpd.buf, ID3v2_DEFAULT_MAGIC)) {
int id3len = ff_id3v2_tag_len(lpd.buf);
if (lpd.buf_size > id3len + 16) {
if (lpd.buf_size < 2LL*id3len + 16)
nodat = ID3_ALMOST_GREATER_PROBE;
lpd.buf += id3len;
lpd.buf_size -= id3len;
} else if (id3len >= PROBE_BUF_MAX) {
nodat = ID3_GREATER_MAX_PROBE;
} else
nodat = ID3_GREATER_PROBE;
}
fmt = NULL;
//0.初始化时fmt1为null
//1.ffplay::main -> register_all
//2.REGISTER_DEMUXER (AAC,aac); ,注册解码器,加入新的AVInputFormat(ff_aac_demuxer: AVInputFormat)
//遍历AVInputFormat链表,如ff_mp3_demuxer等
while ((fmt1 = av_iformat_next(fmt1))) {
if (!is_opened == !(fmt1->flags & AVFMT_NOFILE) && strcmp(fmt1->name, "image2"))
continue;
score = 0;
if (fmt1->read_probe) {
//mp3 -> mp3_read_probe
score = fmt1->read_probe(&lpd);
if (score)
av_log(NULL, AV_LOG_TRACE, "Probing %s score:%d size:%d\n", fmt1->name, score, lpd.buf_size);
if (fmt1->extensions && av_match_ext(lpd.filename, fmt1->extensions)) {
switch (nodat) {
case NO_ID3:
score = FFMAX(score, 1);
break;
case ID3_GREATER_PROBE:
case ID3_ALMOST_GREATER_PROBE:
score = FFMAX(score, AVPROBE_SCORE_EXTENSION / 2 - 1);
break;
case ID3_GREATER_MAX_PROBE:
score = FFMAX(score, AVPROBE_SCORE_EXTENSION);
break;
}
}
} else if (fmt1->extensions) { //如果demuxer有登记自身的后缀,如ff_mp3_demuxer->extensions == 'mp2,mp3,m2a,mpa'
if (av_match_ext(lpd.filename, fmt1->extensions))
score = AVPROBE_SCORE_EXTENSION;
}
if (av_match_name(lpd.mime_type, fmt1->mime_type)) {
if (AVPROBE_SCORE_MIME > score) {
av_log(NULL, AV_LOG_DEBUG, "Probing %s score:%d increased to %d due to MIME type\n", fmt1->name, score, AVPROBE_SCORE_MIME);
score = AVPROBE_SCORE_MIME;
}
}
if (score > score_max) {
score_max = score;
fmt = fmt1;
} else if (score == score_max)
fmt = NULL;
}
if (nodat == ID3_GREATER_PROBE)
score_max = FFMIN(AVPROBE_SCORE_EXTENSION / 2 - 1, score_max);
*score_ret = score_max;
return fmt;
}
av_iformat_next
遍历解码器。在初始化AVFormatContext之后,会调用av_register_all注册解码器和编码器。在单次循环中,调用对应解码器的read_probe进行打分,比如mp3音频对应的就是mp3_read_probe
函数,随后根据最高分选出适合的解码器