背景
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。首先先看一下函数调用的基本流程，如下图：
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。
avformat_open_input这个函数的作用是打开文件的链接，如果是网络连接，还会发起网络请求，并一直等待网络数据的返回，然后读取视频流的数据。
ffurl_open_whitelist函数的功能主要是打开文件链接，并填充一个URLContext *h结构体。该结构体的声明是在url.h文件里面，源码里面有
typedef struct URLContext {
const AVClass *av_class; /**< information for av_log(). Set by url_open(). */
const struct URLProtocol *prot;
void *priv_data;
char *filename; /**< specified URL */
int flags;
int max_packet_size; /**< if non zero, the stream is packetized with this max packet size */
int is_streamed; /**< true if streamed (no seek possible), default = false */
int is_connected;
AVIOInterruptCB interrupt_callback;
int64_t rw_timeout; /**< maximum time to wait for (network) read/write operation completion, in mcs */
const char *protocol_whitelist;
const char *protocol_blacklist;
} URLContext;这个结构体很重要，里面prot指向了具体URLProtocol结构体，该结构体里面包含有打开该协议的url的回调函数，如http，tcp，都有对应的open函数来处理。
ffurl_alloc
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;
}url_find_protocol: 根据输入的文件名或者url进行protocol探查，如filename以http开头的则探查返回的结果为ff_http_protocol
url_alloc_for_protocol: 初始化URLContext,并将上一个步骤返回的URLProtocol赋值其中
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
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;
}- 在protocol_list.c中罗列了所有ffmpeg支持的协议，例如http、rtmp、udp、file等。这些协议常量以URLProtocol结构的形式保存在url_protocols数组中。而url_find_protocol函数就是要通过filename匹配到特定的URLProtocol实例。
- 首先，url_find_protocol通过正则表达式判断出filename的协议长度,主要通过判断filename首个非字母非数字的下标来获得。例如http://www.baidu.com/xxx.mp4,则首个非字母非数字的字符为[:],其下标为4。所以proto_len = 4
- 随后，根据proto_len截取出filename中的协议字符串,例如http.再逐个跟ff_http_protocol等URLProtocol的name成员进行比较，如果相等，则代表匹配到URLProtocol。
ffurl_connect函数
该函数中唯一一个比较重要的函数就是
err = uc->prot->url_open2 ? uc->prot->url_open2(uc,
uc->filename,
uc->flags,
options) :
uc->prot->url_open(uc, uc->filename, uc->flags);首先判断是否prot->url_open2函数指针是否有赋值。我们从http.c中ff_http_protocol结构体的定义中可以发现
.url_open2 = http_open,所以在ffurl_connect函数中这里实际调用的是http_open函数。那么接下来就进入到我们的关键函数了。
http_open
http协议的基本实现都是在http.c文件中实现的。确定URLProtocol的实例是http之后，调用的就是http_open
static int http_open(URLContext *h, const char *uri, int flags,
AVDictionary **options)
{
av_log(NULL, AV_LOG_WARNING, "This is %s!\n",__FUNCTION__);
HTTPContext *s = h->priv_data;
int ret;
s->app_ctx = (AVApplicationContext *)(intptr_t)s->app_ctx_intptr;
if( s->seekable == 1 )
h->is_streamed = 0;
else
h->is_streamed = 1;
s->filesize = UINT64_MAX;
s->location = av_strdup(uri);
if (!s->location)
return AVERROR(ENOMEM);
if (options)
av_dict_copy(&s->chained_options, *options, 0);
if (s->headers) {
int len = strlen(s->headers);
if (len < 2 || strcmp("\r\n", s->headers + len - 2)) {
av_log(h, AV_LOG_WARNING,
"No trailing CRLF found in HTTP header.\n");
ret = av_reallocp(&s->headers, len + 3);
if (ret < 0)
return ret;
s->headers[len] = '\r';
s->headers[len + 1] = '\n';
s->headers[len + 2] = '\0';
}
}
if (s->listen) {
return http_listen(h, uri, flags, options);
}
av_log(NULL, AV_LOG_WARNING,"mdx http_listen %s!\n",__FUNCTION__);
av_application_will_http_open(s->app_ctx, (void*)h, uri);
ret = http_open_cnx(h, options);
av_application_did_http_open(s->app_ctx, (void*)h, uri, ret, s->http_code, s->filesize);
if (ret < 0)
av_dict_free(&s->chained_options);
return ret;
}在这个函数中，首先看一下av_application_will_http_open，这个函数主要是调用了av_application_on_http_event发送了AVAPP_EVENT_WILL_HTTP_OPEN的event，代表开始做http的open操作
#define AVAPP_EVENT_WILL_HTTP_OPEN 1 //AVAppHttpEvent真正的open操作是在后续的http_open_cnx。
另外，在http_open_cnx之后，会去调用 av_application_did_http_open发送了AVAPP_EVENT_DID_HTTP_OPEN的event，表明完成了connect，可以传输数据
#define AVAPP_EVENT_DID_HTTP_OPEN 2 //AVAppHttpEventhttp_open_cnx
static int http_open_cnx(URLContext *h, AVDictionary **options)
{
HTTPAuthType cur_auth_type, cur_proxy_auth_type;
HTTPContext *s = h->priv_data;
int location_changed, attempts = 0, redirects = 0;
redo:
av_dict_copy(options, s->chained_options, 0);
cur_auth_type = s->auth_state.auth_type;
cur_proxy_auth_type = s->auth_state.auth_type;
location_changed = http_open_cnx_internal(h, options);
if (location_changed < 0)
goto fail;
attempts++;
if (s->http_code == 401) {
if ((cur_auth_type == HTTP_AUTH_NONE || s->auth_state.stale) &&
s->auth_state.auth_type != HTTP_AUTH_NONE && attempts < 4) {
ffurl_closep(&s->hd);
goto redo;
} else
goto fail;
}
if (s->http_code == 407) {
if ((cur_proxy_auth_type == HTTP_AUTH_NONE || s->proxy_auth_state.stale) &&
s->proxy_auth_state.auth_type != HTTP_AUTH_NONE && attempts < 4) {
ffurl_closep(&s->hd);
goto redo;
} else
goto fail;
}
if ((s->http_code == 301 || s->http_code == 302 ||
s->http_code == 303 || s->http_code == 307) &&
location_changed == 1) {
/* url moved, get next */
ffurl_closep(&s->hd);
if (redirects++ >= MAX_REDIRECTS)
return AVERROR(EIO);
/* Restart the authentication process with the new target, which
* might use a different auth mechanism. */
memset(&s->auth_state, 0, sizeof(s->auth_state));
attempts = 0;
location_changed = 0;
goto redo;
}
return 0;
fail:
if (s->hd)
ffurl_closep(&s->hd);
if (location_changed < 0)
return location_changed;
return ff_http_averror(s->http_code, AVERROR(EIO));
}http_open_cnx_internal
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 ->
//根据传入的s->location其实就是视频的url，从url里面提取出hostname，port，以及path。
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")) {
av_dict_set_int(options, "fastopen", 0, 0);
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);
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;
}1、底层protocol默认采用的是tcp,如果是https协议的话会更改为tls.
2、调用ffurl_open_whitelist生成对应tcp的URLProtocol(ff_tcp_protocol)
3、调用底层protocol进行发送报文
关于建立TCP链接，这里多分析一下：
ff_url_join前面用到的时候是解析出hostname，但在这里，由于传入了lower_proto（它表示http 协议的下一层协议，一般都是tcp，所以该值初始化的时候就是tcp），所以buf的值是有lower_proto拼凑起来的tcp链接tcp://flv-meipai.8686c.com:80，就是tcp+域名。
接下来就是判断 s->hd 是否存在。默认情况下，是为NULL值，所以调用ffurl_open_whitelist开始打开tcp://flv-meipai.8686c.com:80，现在回到ffurl_open_whitelist函数了。这时通过url_find_protocol找到的就是tcp类型的URLProtocol，那么在ffurl_connect调用时调用的就是tcp.c里面的tcp_open函数，那么tcp的握手连接就在这个函数里面解析了。首先在tcp_open函数368行
av_url_split(proto, sizeof(proto), NULL, 0, hostname, sizeof(hostname),
&port, path, sizeof(path), uri);先根据uri解析出协议名以及hostname，然后调用以下的
ret = ijk_tcp_getaddrinfo_nonblock(hostname, portstr, &hints, &ai, s->addrinfo_timeout, &h->interrupt_callback, s->addrinfo_one_by_one);做DNS解析。这个函数是ijkplayer作者加上去的，标准的ffmpeg 里面并没有。它的功能是利用多线程来解析DNS。但实际上从代码上并没有看到有什么优势，其实还是阻塞等结果解析出来了才返回的，这个地方不是很懂为什么要这么改。
接下来就是创建socket了
fd = ff_socket(cur_ai->ai_family, cur_ai->ai_socktype, cur_ai->ai_protocol);调用ff_listen_connect函数进行tcp握手。
之后668行 会有一个调用av_application_on_tcp_will_open(s->app_ctx)，主要是发送了AVAPP_CTRL_WILL_TCP_OPEN（#define AVAPP_CTRL_WILL_TCP_OPEN 0x20001 //AVAppTcpIOControl）消息，表示要开始tcp连接；在此之后会调用av_application_on_tcp_did_open，发送一个AVAPP_CTRL_DID_TCP_OPEN（#define AVAPP_CTRL_DID_TCP_OPEN 0x20002 //AVAppTcpIOControl）消息，表示tcp连接成功。
至此，调用tcp协议的ffurl_open_whitelist函数就调用完成了，tcp握手连接也建立成功。再回到http_open_cnx_internal函数。继续调用
err = http_connect(h, path, local_path, hoststr, auth, proxyauth, &location_changed);http_connect
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;
}
1、ffurl_write: 通过lower protocol(例如tcp)写入协议头
从ffurl_write代码中可以看出，它实际调用的是url_write方法，而该prot的write方法，是http_write，它又是调用的ffurl_write(s->hd, buf, size);就是指http协议下一层的协议tcp的tcp_write方法。tcp_write方法最终调用就是ret = send(s->fd, buf, size, MSG_NOSIGNAL);系统的send方法。所以最终都是调用系统实现的Socket接口。至此，http_connect方法的发送request的请求就完毕了。剩下就是等待响应了。
2、http_read_header: 读取服务器返回的header,并保存在HTTPContext
http_read_header就是不断的读取网络返回的数据，并解析出来。
至此http_open_cnx_internal函数也调用完了。回到http_open_cnx函数。这时如果能正常获取数据，那么s->http_code的值应该是200，至此，http_open_cnx，http_open函数也返回了，流程可以直接返回到ffio_open_whitelist函数中，ffio_fdopen函数只是对AVIOContext结构体根据http request获取的数据进行一些赋值。那就可以直接返回到init_input函数了。接下来是
if (s->iformat)
return 0;
return av_probe_input_buffer2(s->pb, &s->iformat, filename,
判断如果s->iformat没有值，就根据filename解析出s->iformat。这也是在前面开头提到的，如果没有加av_find_input_format("flv")这个代码，那就要重新根据filename来解析数据了，这个函数比较耗时，需要读取到一定数据后才能解析出来。
probe(探查)
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);
}初始化时，AVInputFormat是无效的。因此会调用av_probe_input_buffer2进行媒体资源的格式探查
av_probe_input_buffer2
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;
}- 初始化AVFormatContext时并没有确定probe data的大小,所以这里会统一设置为1<<20
- 创建AVProbeData结构，并把之前URLProtocol读取到的mime_type赋值到其中
- 循环多次进行probe行为，每次循环探查的数据大小逐渐增加，当匹配到特定解码器后跳出循环。
- probe行为：a).通过tcp读取数据，读取数据的上限为probe_max(1<<20)。b).读取数据的过程中，如果遇到end of file之外的错误则探查行为失败，进入失败流程。c).读取数据之后，调用av_probe_input_format2(实质上调用的是av_probe_input_format3)进行format guess,并打分。
avio_read
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;
}
1、首次读取通过调用fill_buffer进行,如果是http协议的媒体资源,会通过底层的tcp去读取数据,并存放在AVIOContext->buffer中
2、进入第二次循环之后，会将AVIOContext->buffer拷贝到目标buf中
av_probe_input_format3
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;
}1、首先在AVProbeData数据中，查看是否有id3信息。id3一般在mp3音频中，该信息用于存放比如专辑名称等一些数据。
2、调用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。
参考链接：ffplay如何通过URLProtocol拉取http数据【源码篇】_哲学家♂的博客-CSDN博客
参考链接：Avformat_open_input函数的分析之--HTTP篇_baohonglai的博客-CSDN博客_avformat_open_input参数