srs代码学习（1）--listen建立过程

srs的服务侦听的建立过程。

以rtmp服务为例 srs服务侦听的建立依靠从上到下的三个类。分别是

SrsServer  

SrsStreamListener  

SrsTcpListener

端口侦听过程为

1）main函数中调用全局变量_srs_server的  listen（）函数

if ((ret = _srs_server->listen()) != ERROR_SUCCESS) {
        return ret;
    }

在SrsServer的listen（）函数如下

<pre name="code" class="cpp">int SrsServer::listen()
{
    int ret = ERROR_SUCCESS;
    
    if ((ret = listen_rtmp()) != ERROR_SUCCESS) {
        return ret;
    }
    
    if ((ret = listen_http_api()) != ERROR_SUCCESS) {
        return ret;
    }
    
    if ((ret = listen_http_stream()) != ERROR_SUCCESS) {
        return ret;
    }
    
    if ((ret = listen_stream_caster()) != ERROR_SUCCESS) {
        return ret;
    }
    
    return ret;
}

listen_rtmp（）会建立一个rtmp的steamlistener

<pre name="code" class="cpp">int SrsServer::listen_rtmp()
{
    int ret = ERROR_SUCCESS;
    
    // stream service port.
    std::vector<std::string> ip_ports = _srs_config->get_listens();
    srs_assert((int)ip_ports.size() > 0);
    
    close_listeners(SrsListenerRtmpStream);
    
    for (int i = 0; i < (int)ip_ports.size(); i++) {
        SrsListener* listener = new SrsStreamListener(this, SrsListenerRtmpStream);
        listeners.push_back(listener);
        
        std::string ip;
        int port;
        srs_parse_endpoint(ip_ports[i], ip, port);
        
        if ((ret = listener->listen(ip, port)) != ERROR_SUCCESS) {
            srs_error("RTMP stream listen at %s:%d failed. ret=%d", ip.c_str(), port, ret);
            return ret;
        }
    }
    
    return ret;
}

这个streamlistener并不是真正的底层监听层。只是一个业务封装层。其类的继承顺序如下

在整个代码中，和其相识的类有SrsRtspListener  SrsHttpFlvListener两个类。这种类的主要作用是在底层链接建立有。给不同的上层放回链接信息。

代码走到这一步，侦听的socket还么米有建立起来，看SrsStreamListener的listen()函数

int SrsStreamListener::listen(string i, int p)
{
    int ret = ERROR_SUCCESS;
    
    ip = i;
    port = p;

    srs_freep(listener);
    listener = new SrsTcpListener(this, ip, port);

    if ((ret = listener->listen()) != ERROR_SUCCESS) {
        srs_error("tcp listen failed. ret=%d", ret);
        return ret;
    }
    
    srs_info("listen thread current_cid=%d, "
        "listen at port=%d, type=%d, fd=%d started success, ep=%s:%d",
        _srs_context->get_id(), p, type, listener->fd(), i.c_str(), p);

    srs_trace("%s listen at tcp://%s:%d, fd=%d", srs_listener_type2string(type).c_str(), ip.c_str(), port, listener->fd());

    return ret;
}

 listener = new SrsTcpListener(this, ip, port);

int SrsTcpListener::listen()
{
    int ret = ERROR_SUCCESS;
    
    if ((_fd = socket(AF_INET, SOCK_STREAM, 0)) == -1) {
        ret = ERROR_SOCKET_CREATE;
        srs_error("create linux socket error. port=%d, ret=%d", port, ret);
        return ret;
    }
    srs_verbose("create linux socket success. port=%d, fd=%d", port, _fd);
    
    int reuse_socket = 1;
    if (setsockopt(_fd, SOL_SOCKET, SO_REUSEADDR, &reuse_socket, sizeof(int)) == -1) {
        ret = ERROR_SOCKET_SETREUSE;
        srs_error("setsockopt reuse-addr error. port=%d, ret=%d", port, ret);
        return ret;
    }
    srs_verbose("setsockopt reuse-addr success. port=%d, fd=%d", port, _fd);
    
    sockaddr_in addr;
    addr.sin_family = AF_INET;
    addr.sin_port = htons(port);
    addr.sin_addr.s_addr = inet_addr(ip.c_str());
    if (bind(_fd, (const sockaddr*)&addr, sizeof(sockaddr_in)) == -1) {
        ret = ERROR_SOCKET_BIND;
        srs_error("bind socket error. ep=%s:%d, ret=%d", ip.c_str(), port, ret);
        return ret;
    }
    srs_verbose("bind socket success. ep=%s:%d, fd=%d", ip.c_str(), port, _fd);
    
    if (::listen(_fd, SERVER_LISTEN_BACKLOG) == -1) {
        ret = ERROR_SOCKET_LISTEN;
        srs_error("listen socket error. ep=%s:%d, ret=%d", ip.c_str(), port, ret);
        return ret;
    }
    srs_verbose("listen socket success. ep=%s:%d, fd=%d", ip.c_str(), port, _fd);
    
    if ((_stfd = st_netfd_open_socket(_fd)) == NULL){
        ret = ERROR_ST_OPEN_SOCKET;
        srs_error("st_netfd_open_socket open socket failed. ep=%s:%d, ret=%d", ip.c_str(), port, ret);
        return ret;
    }
    srs_verbose("st open socket success. ep=%s:%d, fd=%d", ip.c_str(), port, _fd);
    
    if ((ret = pthread->start()) != ERROR_SUCCESS) {
        srs_error("st_thread_create listen thread error. ep=%s:%d, ret=%d", ip.c_str(), port, ret);
        return ret;
    }
    srs_verbose("create st listen thread success, ep=%s:%d", ip.c_str(), port);
    
    return ret;
}

在仔细看SrsTcpListener类。继承关系如下

这个类中很有趣的有了个线程类SrsReusableThread* pthread，这个线程是什么时候建立的？答案是在构造函数中

SrsTcpListener::SrsTcpListener(ISrsTcpHandler* h, string i, int p)
{
    handler = h;
    ip = i;
    port = p;

    _fd = -1;
    _stfd = NULL;

    pthread = new SrsReusableThread("tcp", this);
}

观察线程类的结构

很奇怪，SrsReusableThread竟然也继承了一个handler类。这就表示。这个类还是一个封装类，并不是底层的真正的线程类。果然我们发现了其一个变量 internal::SrsThread* pthread，这个应该靠近点底层了吧。先放下不讲。我们是来追这个线程类也我的业务类啥关系的。这时另外一个有意思的变量ISrsReusableThreadHandler* handler，还记得
SrsTcpListener 类么，其有一个ISrsReusableThreadHandler的接口。通过这个接口，把线程和业务链接起来。看看其cycle()函数

int SrsReusableThread::cycle()
{
    return handler->cycle();
}

int SrsTcpListener::cycle()
{
    int ret = ERROR_SUCCESS;
    
    st_netfd_t client_stfd = st_accept(_stfd, NULL, NULL, ST_UTIME_NO_TIMEOUT);
    
    if(client_stfd == NULL){
        // ignore error.
        if (errno != EINTR) {
            srs_error("ignore accept thread stoppped for accept client error");
        }
        return ret;
    }
    srs_verbose("get a client. fd=%d", st_netfd_fileno(client_stfd));
    
    if ((ret = handler->on_tcp_client(client_stfd)) != ERROR_SUCCESS) {
        srs_warn("accept client error. ret=%d", ret);
        return ret;
    }
    
    return ret;
}

int SrsStreamListener::on_tcp_client(st_netfd_t stfd)
{
    int ret = ERROR_SUCCESS;
    
    if ((ret = server->accept_client(type, stfd)) != ERROR_SUCCESS) {
        srs_warn("accept client error. ret=%d", ret);
        return ret;
    }

    return ret;
}

到现在还有两个问题要搞明白：

1）srs的线程模型

2）如何管理各个链接。

这个下一次在摸索