SRS(simple-rtmp-server)流媒体服务器源码分析--RTMP信息Publish

前言

       上一节中已经介绍了rtmp流监听，一旦有rtmp请求之后处理，单独为该链接开启一个线程等等，这一节我们来分析rtmp请求到来之后，除了创建线程之外，还要做哪些工作。同样，只是代码框架分析。

首先对Publish做的工作大概描述一下：

      1、Rtmp协议几乎是从推送端过来的唯一协议（仅限于srs2.0及以下版本），因为Rtmp协议时效性。

      2、建立Rtmp协议链接：握手、建立网络连接、建立流通道、Publish/Play。

      3、接受Rtmp信息：Rtmp协议解析。

      4、成功接受到Rtmp信息处理：缓存、抖动矫正、分发到消费者堆栈（Play消费者，Forward消费者）、进行HLS切片等等。

这一讲内容只讲述RTMP协议Publish过程，SRS源码相关其他总结：

      SRS(simple-rtmp-server)流媒体服务器源码分析--系统启动
      SRS(simple-rtmp-server)流媒体服务器源码分析--RTMP消息play

      SRS(simple-rtmp-server)流媒体服务器源码分析--RTMP信息Publish

      SRS(simple-rtmp-server)流媒体服务器源码分析--HLS切片

Rtmp信息接受框架如下：

 

 一、进入循环

 

       接着上一节rtmp链接分析       

int SrsConnection::cycle()
{
    int ret = ERROR_SUCCESS;
    
    _srs_context->generate_id();
    id = _srs_context->get_id();
    
    ip = srs_get_peer_ip(st_netfd_fileno(stfd));
	srs_trace("ip:%s", ip);
    
    ret = do_cycle();
    
    // if socket io error, set to closed.
    if (srs_is_client_gracefully_close(ret)) {
        ret = ERROR_SOCKET_CLOSED;
    }
    
    // success.
    if (ret == ERROR_SUCCESS) {
        srs_trace("client finished.");
    }
    
    // client close peer.
    if (ret == ERROR_SOCKET_CLOSED) {
        srs_warn("client disconnect peer. ret=%d", ret);
    }

    return ERROR_SUCCESS;
}

进来之后获取客户端ip，进入do_cycle()函数，这里也要同样注意，在理解纯虚函数上是我的缺点。

 

二、rtmp协议链接

 

    RTMP协议是应用层协议，是要靠底层可靠的传输层协议（通常是TCP）来保证信息传输的可靠性的。在基于传输层协议的链接建立完成后，RTMP协议也要客户端和服务器通过“握手”来建立基于传输层链接之上的RTMP Connection链接，在Connection链接上会传输一些控制信息，如SetChunkSize,SetACKWindowSize。其中CreateStream命令会创建一个Stream链接，用于传输具体的音视频数据和控制这些信息传输的命令信息。RTMP协议传输时会对数据做自己的格式化，这种格式的消息我们称之为RTMP Message，而实际传输的时候为了更好地实现多路复用、分包和信息的公平性，发送端会把Message划分为带有Message ID的Chunk，每个Chunk可能是一个单独的Message，也可能是Message的一部分，在接受端会根据chunk中包含的data的长度，messageid和message的长度把chunk还原成完整的Message，从而实现信息的收发。

 

Rtmp协议规定：

（1）、 握手：固定步骤，RTMP连接都是以握手作为开始。

（2）、 建立连接：用于建立客户端和服务端之间的“网络连接”

（3）、 建立流：用于建立客户端和服务端之间的“网络流”，即多媒体通道。

（4）、 Publish/play：推流或者播放

 

1、握手 建立网络连接

 

   有关rtmp协议握手内容，可以参考rtmp协议博客：http://blog.csdn.net/ManagerUser/article/category/6996313

   直接来看建立网络连接：

// TODO: return detail message when error for client.
int SrsRtmpConn::do_cycle()
{
    int ret = ERROR_SUCCESS;
    
    srs_trace("RTMP client ip=%s", ip.c_str());

    rtmp->set_recv_timeout(SRS_CONSTS_RTMP_RECV_TIMEOUT_US);
    rtmp->set_send_timeout(SRS_CONSTS_RTMP_SEND_TIMEOUT_US);
    //1、rtmp握手
    if ((ret = rtmp->handshake()) != ERROR_SUCCESS) {
        srs_error("rtmp handshake failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("rtmp handshake success");
    //2、rtmp建立网络连接
    if ((ret = rtmp->connect_app(req)) != ERROR_SUCCESS) {
        srs_error("rtmp connect vhost/app failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("rtmp connect app success");
    
    // set client ip to request.
    req->ip = ip;
    
    // discovery vhost, resolve the vhost from config
    SrsConfDirective* parsed_vhost = _srs_config->get_vhost(req->vhost);
    if (parsed_vhost) {
        req->vhost = parsed_vhost->arg0();
    }
    
    srs_info("discovery app success. schema=%s, vhost=%s, port=%s, app=%s",
        req->schema.c_str(), req->vhost.c_str(), req->port.c_str(), req->app.c_str());
    
    if (req->schema.empty() || req->vhost.empty() || req->port.empty() || req->app.empty()) {
        ret = ERROR_RTMP_REQ_TCURL;
        srs_error("discovery tcUrl failed. "
            "tcUrl=%s, schema=%s, vhost=%s, port=%s, app=%s, ret=%d",
            req->tcUrl.c_str(), req->schema.c_str(), req->vhost.c_str(), req->port.c_str(), req->app.c_str(), ret);
        return ret;
    }
    
    // check vhost
    if ((ret = check_vhost()) != ERROR_SUCCESS) {
        srs_error("check vhost failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("check vhost success.");
    
    srs_trace("connect app, "
        "tcUrl=%s, pageUrl=%s, swfUrl=%s, schema=%s, vhost=%s, port=%s, app=%s, args=%s", 
        req->tcUrl.c_str(), req->pageUrl.c_str(), req->swfUrl.c_str(), 
        req->schema.c_str(), req->vhost.c_str(), req->port.c_str(),
        req->app.c_str(), (req->args? "(obj)":"null"));
    
    // show client identity
    if(req->args) {
        std::string srs_version;
        std::string srs_server_ip;
        int srs_pid = 0;
        int srs_id = 0;
        
        SrsAmf0Any* prop = NULL;
        if ((prop = req->args->ensure_property_string("srs_version")) != NULL) {
            srs_version = prop->to_str();
        }
        if ((prop = req->args->ensure_property_string("srs_server_ip")) != NULL) {
            srs_server_ip = prop->to_str();
        }
        if ((prop = req->args->ensure_property_number("srs_pid")) != NULL) {
            srs_pid = (int)prop->to_number();
        }
        if ((prop = req->args->ensure_property_number("srs_id")) != NULL) {
            srs_id = (int)prop->to_number();
        }
        
        srs_info("edge-srs ip=%s, version=%s, pid=%d, id=%d", 
            srs_server_ip.c_str(), srs_version.c_str(), srs_pid, srs_id);
        if (srs_pid > 0) {
            srs_trace("edge-srs ip=%s, version=%s, pid=%d, id=%d", 
                srs_server_ip.c_str(), srs_version.c_str(), srs_pid, srs_id);
        }
    }
    
    ret = service_cycle();
    
    http_hooks_on_close();

    return ret;
}

 在以上函数中，进行Rtmp握手，rtmp网络连接。建立网络连接之后要做一些确认工作。内容如下：

 进入service_cycle()

int SrsRtmpConn::service_cycle()
{    
    int ret = ERROR_SUCCESS;
    //1、服务器接收到连接命令消息后，发送确认窗口大小(Window Acknowledgement Size)协议消息到客户端，同时连接到连接命令中提到的应用程序。
    if ((ret = rtmp->set_window_ack_size((int)(2.5 * 1000 * 1000))) != ERROR_SUCCESS) {
        srs_error("set window acknowledgement size failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("set window acknowledgement size success");
    //2、服务器发送设置带宽()协议消息到客户端。
    if ((ret = rtmp->set_peer_bandwidth((int)(2.5 * 1000 * 1000), 2)) != ERROR_SUCCESS) {
        srs_error("set peer bandwidth failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("set peer bandwidth success");

    // get the ip which client connected.
    std::string local_ip = srs_get_local_ip(st_netfd_fileno(stfd));
    
    // do bandwidth test if connect to the vhost which is for bandwidth check.
    if (_srs_config->get_bw_check_enabled(req->vhost)) {
        return bandwidth->bandwidth_check(rtmp, skt, req, local_ip);
    }
    
    // do token traverse before serve it.
    // @see https://github.com/ossrs/srs/pull/239
    if (true) {
        bool vhost_is_edge = _srs_config->get_vhost_is_edge(req->vhost);
        bool edge_traverse = _srs_config->get_vhost_edge_token_traverse(req->vhost);
        if (vhost_is_edge && edge_traverse) {
            if ((ret = check_edge_token_traverse_auth()) != ERROR_SUCCESS) {
                srs_warn("token auth failed, ret=%d", ret);
                return ret;
            }
        }
    }
    
    // set chunk size to larger.
    // set the chunk size before any larger response greater than 128,
    // to make OBS happy, @see https://github.com/ossrs/srs/issues/454
    int chunk_size = _srs_config->get_chunk_size(req->vhost);
    if ((ret = rtmp->set_chunk_size(chunk_size)) != ERROR_SUCCESS) {
        srs_error("set chunk_size=%d failed. ret=%d", chunk_size, ret);
        return ret;
    }
    srs_info("set chunk_size=%d success", chunk_size);
    
    // response the client connect ok.
    if ((ret = rtmp->response_connect_app(req, local_ip.c_str())) != ERROR_SUCCESS) {
        srs_error("response connect app failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("response connect app success");
        
    if ((ret = rtmp->on_bw_done()) != ERROR_SUCCESS) {
        srs_error("on_bw_done failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("on_bw_done success");
    
    while (!disposed) {
        ret = stream_service_cycle();
        
        // stream service must terminated with error, never success.
        // when terminated with success, it's user required to stop.
        if (ret == ERROR_SUCCESS) {
            continue;
        }
        
        // when not system control error, fatal error, return.
        if (!srs_is_system_control_error(ret)) {
            if (ret != ERROR_SOCKET_TIMEOUT && !srs_is_client_gracefully_close(ret)) {
                srs_error("stream service cycle failed. ret=%d", ret);
            }
            return ret;
        }
        
        // for republish, continue service
        if (ret == ERROR_CONTROL_REPUBLISH) {
            // set timeout to a larger value, wait for encoder to republish.
            rtmp->set_send_timeout(SRS_REPUBLISH_RECV_TIMEOUT_US);
            rtmp->set_recv_timeout(SRS_REPUBLISH_SEND_TIMEOUT_US);
            
            srs_trace("control message(unpublish) accept, retry stream service.");
            continue;
        }
        
        // for "some" system control error, 
        // logical accept and retry stream service.
        if (ret == ERROR_CONTROL_RTMP_CLOSE) {
            // TODO: FIXME: use ping message to anti-death of socket.
            // @see: https://github.com/ossrs/srs/issues/39
            // set timeout to a larger value, for user paused.
            rtmp->set_recv_timeout(SRS_PAUSED_RECV_TIMEOUT_US);
            rtmp->set_send_timeout(SRS_PAUSED_SEND_TIMEOUT_US);
            
            srs_trace("control message(close) accept, retry stream service.");
            continue;
        }
        
        // for other system control message, fatal error.
        srs_error("control message(%d) reject as error. ret=%d", ret, ret);
        return ret;
    }
    
    return ret;
}

网络连接确认工作完成之后，进入客户身份识别代码中

 

2、client identify

 

进入stream_service_cycle()

int SrsRtmpConn::stream_service_cycle()
{
    int ret = ERROR_SUCCESS;
        
    SrsRtmpConnType type;
    if ((ret = rtmp->identify_client(res->stream_id, type, req->stream, req->duration)) != ERROR_SUCCESS) {
        if (!srs_is_client_gracefully_close(ret)) {
            srs_error("identify client failed. ret=%d", ret);
        }
        return ret;
    }
    req->strip();
    srs_trace("client identified, type=%s, stream_name=%s, duration=%.2f", 
        srs_client_type_string(type).c_str(), req->stream.c_str(), req->duration);
    
    // security check
    if ((ret = security->check(type, ip, req)) != ERROR_SUCCESS) {
        srs_error("security check failed. ret=%d", ret);
        return ret;
    }
    srs_info("security check ok");
    
    // Never allow the empty stream name, for HLS may write to a file with empty name.
    // @see https://github.com/ossrs/srs/issues/834
    if (req->stream.empty()) {
        ret = ERROR_RTMP_STREAM_NAME_EMPTY;
        srs_error("RTMP: Empty stream name not allowed, ret=%d", ret);
        return ret;
    }

    // client is identified, set the timeout to service timeout.
    rtmp->set_recv_timeout(SRS_CONSTS_RTMP_RECV_TIMEOUT_US);
    rtmp->set_send_timeout(SRS_CONSTS_RTMP_SEND_TIMEOUT_US);
    
    // find a source to serve.
    SrsSource* source = NULL;
    if ((ret = SrsSource::fetch_or_create(req, server, &source)) != ERROR_SUCCESS) {
        return ret;
    }
    srs_assert(source != NULL);
    
    // update the statistic when source disconveried.
    SrsStatistic* stat = SrsStatistic::instance();
    if ((ret = stat->on_client(_srs_context->get_id(), req, this, type)) != ERROR_SUCCESS) {
        srs_error("stat client failed. ret=%d", ret);
        return ret;
    }

    bool vhost_is_edge = _srs_config->get_vhost_is_edge(req->vhost);
    bool enabled_cache = _srs_config->get_gop_cache(req->vhost);
    srs_trace("source url=%s, ip=%s, cache=%d, is_edge=%d, source_id=%d[%d]",
        req->get_stream_url().c_str(), ip.c_str(), enabled_cache, vhost_is_edge, 
        source->source_id(), source->source_id());
    source->set_cache(enabled_cache);
    
    client_type = type;
    switch (type) {
        case SrsRtmpConnPlay: {
            srs_verbose("start to play stream %s.", req->stream.c_str());
            
            // response connection start play
            if ((ret = rtmp->start_play(res->stream_id)) != ERROR_SUCCESS) {
                srs_error("start to play stream failed. ret=%d", ret);
                return ret;
            }
            if ((ret = http_hooks_on_play()) != ERROR_SUCCESS) {
                srs_error("http hook on_play failed. ret=%d", ret);
                return ret;
            }
            
            srs_info("start to play stream %s success", req->stream.c_str());
            ret = playing(source);
            http_hooks_on_stop();
            
            return ret;
        }
        case SrsRtmpConnFMLEPublish: {
            srs_verbose("FMLE start to publish stream %s.", req->stream.c_str());
            
            if ((ret = rtmp->start_fmle_publish(res->stream_id)) != ERROR_SUCCESS) {
                srs_error("start to publish stream failed. ret=%d", ret);
                return ret;
            }
            
            return publishing(source);
        }
        case SrsRtmpConnHaivisionPublish: {
            srs_verbose("Haivision start to publish stream %s.", req->stream.c_str());
            
            if ((ret = rtmp->start_haivision_publish(res->stream_id)) != ERROR_SUCCESS) {
                srs_error("start to publish stream failed. ret=%d", ret);
                return ret;
            }
            
            return publishing(source);
        }
        case SrsRtmpConnFlashPublish: {
            srs_verbose("flash start to publish stream %s.", req->stream.c_str());
            
            if ((ret = rtmp->start_flash_publish(res->stream_id)) != ERROR_SUCCESS) {
                srs_error("flash start to publish stream failed. ret=%d", ret);
                return ret;
            }
            
            return publishing(source);
        }
        default: {
            ret = ERROR_SYSTEM_CLIENT_INVALID;
            srs_info("invalid client type=%d. ret=%d", type, ret);
            return ret;
        }
    }

    return ret;
}

 

       通过客户身份识别identify_client()函数后，Srs就知道是publish还是play，进入相应的分支。首先这里分支有点多，SrsRtmpConnPlay播放rtmp流，SrsRtmpConnFMLEPublish，fmle发布rtmp流，我们接受rtmp流，就是从SrsRtmpConnFMLEPublish分支里面进去。

注意：使用vlc播放器播放从SrsRtmpConnPlay分支进入。

 

3、进入publish分支

 

进入publishing();

int SrsRtmpConn::publishing(SrsSource* source)
{
    int ret = ERROR_SUCCESS;

    if ((ret = refer->check(req->pageUrl, _srs_config->get_refer_publish(req->vhost))) != ERROR_SUCCESS) {
        srs_error("check publish_refer failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("check publish_refer success.");
    // http回调，通知vhost SRS发生了publish事件。
    if ((ret = http_hooks_on_publish()) != ERROR_SUCCESS) {
        srs_error("http hook on_publish failed. ret=%d", ret);
        return ret;
    }

    bool vhost_is_edge = _srs_config->get_vhost_is_edge(req->vhost);
    if ((ret = acquire_publish(source, vhost_is_edge)) == ERROR_SUCCESS) {
        // use isolate thread to recv,
        // @see: https://github.com/ossrs/srs/issues/237
        // 创建了一个接受线程对象，这个在后面用，
        SrsPublishRecvThread trd(rtmp, req, 
            st_netfd_fileno(stfd), 0, this, source,
            client_type != SrsRtmpConnFlashPublish,
            vhost_is_edge);

        srs_info("start to publish stream %s success", req->stream.c_str());
        ret = do_publishing(source, &trd);

        // stop isolate recv thread
        trd.stop();
    }
    
    // whatever the acquire publish, always release publish.
    // when the acquire error in the midlle-way, the publish state changed,
    // but failed, so we must cleanup it.
    // @see https://github.com/ossrs/srs/issues/474
    // @remark when stream is busy, should never release it.
    if (ret != ERROR_SYSTEM_STREAM_BUSY) {
        release_publish(source, vhost_is_edge);
    }

    http_hooks_on_unpublish();

    return ret;
}

 

三、单独创建rtmp接收线程

 

进入do_publishing()，等待接受rtmp信息。

int SrsRtmpConn::do_publishing(SrsSource* source, SrsPublishRecvThread* trd)
{
    int ret = ERROR_SUCCESS;
    
    SrsPithyPrint* pprint = SrsPithyPrint::create_rtmp_publish();
    SrsAutoFree(SrsPithyPrint, pprint);
    // 创建一个独立的接收线程，专门用于Rtmp协议的接受，因此，下一步进入rtmp接受线程至关重要。
    // start isolate recv thread.
    if ((ret = trd->start()) != ERROR_SUCCESS) {
        srs_error("start isolate recv thread failed. ret=%d", ret);
        return ret;
    }
    
    // change the isolate recv thread context id,
    // merge its log to current thread.
    int receive_thread_cid = trd->get_cid();
    trd->set_cid(_srs_context->get_id());
    
    // initialize the publish timeout.
    publish_1stpkt_timeout = _srs_config->get_publish_1stpkt_timeout(req->vhost);
    publish_normal_timeout = _srs_config->get_publish_normal_timeout(req->vhost);
    
    // set the sock options.
    set_sock_options();
    
    if (true) {
        bool mr = _srs_config->get_mr_enabled(req->vhost);
        int mr_sleep = _srs_config->get_mr_sleep_ms(req->vhost);
        srs_trace("start publish mr=%d/%d, p1stpt=%d, pnt=%d, tcp_nodelay=%d, rtcid=%d",
                  mr, mr_sleep, publish_1stpkt_timeout, publish_normal_timeout, tcp_nodelay, receive_thread_cid);
    }

    int64_t nb_msgs = 0;
    uint64_t nb_frames = 0;
    while (!disposed) {
	// 计算时间
        pprint->elapse();
        // 资源过期
        // when source is set to expired, disconnect it.
        if (expired) {
            ret = ERROR_USER_DISCONNECT;
            srs_trace("connection expired. ret=%d", ret);
            return ret;
        }
        // 等待
        // cond wait for timeout.
        if (nb_msgs == 0) {
            // when not got msgs, wait for a larger timeout.
            // @see https://github.com/ossrs/srs/issues/441
            trd->wait(publish_1stpkt_timeout);
        } else {
            trd->wait(publish_normal_timeout);
        }
		// 检测该线程出现问题
        // check the thread error code.
        if ((ret = trd->error_code()) != ERROR_SUCCESS) {
            if (!srs_is_system_control_error(ret) && !srs_is_client_gracefully_close(ret)) {
                srs_error("recv thread failed. ret=%d", ret);
            }
            return ret;
        }
        //上一次接受RTMP包数量如果等于现在接受的，说明本次没有接收到，publish超时
        // when not got any messages, timeout.
        if (trd->nb_msgs() <= nb_msgs) {
            ret = ERROR_SOCKET_TIMEOUT;
            srs_trace("publish timeout %dms, nb_msgs=%"PRId64", ret=%d",
                nb_msgs? publish_normal_timeout : publish_1stpkt_timeout, nb_msgs, ret);
            break;
        }
        nb_msgs = trd->nb_msgs();
        
        // Update the stat for video fps.
        // @remark https://github.com/ossrs/srs/issues/851
        SrsStatistic* stat = SrsStatistic::instance();
        if ((ret = stat->on_video_frames(req, (int)(trd->nb_video_frames() - nb_frames))) != ERROR_SUCCESS) {
            return ret;
        }
	// 获取视频帧数量
        nb_frames = trd->nb_video_frames();

        // reportable
        if (pprint->can_print()) {
            kbps->sample();
            bool mr = _srs_config->get_mr_enabled(req->vhost);
            int mr_sleep = _srs_config->get_mr_sleep_ms(req->vhost);
            srs_trace("<- "SRS_CONSTS_LOG_CLIENT_PUBLISH
                " time=%"PRId64", okbps=%d,%d,%d, ikbps=%d,%d,%d, mr=%d/%d, p1stpt=%d, pnt=%d", pprint->age(),
                kbps->get_send_kbps(), kbps->get_send_kbps_30s(), kbps->get_send_kbps_5m(),
                kbps->get_recv_kbps(), kbps->get_recv_kbps_30s(), kbps->get_recv_kbps_5m(),
                mr, mr_sleep, publish_1stpkt_timeout, publish_normal_timeout
            );
        }
    }

    return ret;
}

该while(!disposed)是一个检测rtmp连接循环，不是真正Rtmp接受线程。

 

四、rtmp接受

 

int SrsRecvThread::cycle()
{
    int ret = ERROR_SUCCESS;

    while (!trd->interrupted()) {
        if (!handler->can_handle()) {
            st_usleep(timeout * 1000);
            continue;
        }
    	
        SrsCommonMessage* msg = NULL;
		srs_trace("come in isloate recv thread\n");
        // recv and handle message
        ret = rtmp->recv_message(&msg);
        if (ret == ERROR_SUCCESS) {
            ret = handler->handle(msg);
        }
    
        if (ret != ERROR_SUCCESS) {
            if (!srs_is_client_gracefully_close(ret) && !srs_is_system_control_error(ret)) {
                srs_error("thread process message failed. ret=%d", ret);
            }
    
            // we use no timeout to recv, should never got any error.
            trd->interrupt();
            
            // notice the handler got a recv error.
            handler->on_recv_error(ret);
    
            return ret;
        }
        srs_verbose("thread loop recv message. ret=%d", ret);
    }

    return ret;

ok，进入真正Rtmp接受线程，当然第一步是rtmp接受了，第二步就是rtmp信息处理了。我们来看rtmp接受。

 

1、rtmp接受

 

rtmp接受还是挺复杂的，绕了这么大的弯子，终于来到rtmp接受了，具体代码框架流程请看框架图。

int SrsProtocol::recv_message(SrsCommonMessage** pmsg)
{
    *pmsg = NULL;
    
    int ret = ERROR_SUCCESS;
    
    while (true) {
        SrsCommonMessage* msg = NULL;

	//接受RTMP块流
        if ((ret = recv_interlaced_message(&msg)) != ERROR_SUCCESS) {
            if (ret != ERROR_SOCKET_TIMEOUT && !srs_is_client_gracefully_close(ret)) {
                srs_error("recv interlaced message failed. ret=%d", ret);
            }
            srs_freep(msg);
            return ret;
        }
        srs_trace("entire msg received");
        
        if (!msg) {
            srs_info("got empty message without error.");
            continue;
        }
        
        if (msg->size <= 0 || msg->header.payload_length <= 0) {
            srs_trace("ignore empty message(type=%d, size=%d, time=%"PRId64", sid=%d).",
                msg->header.message_type, msg->header.payload_length,
                msg->header.timestamp, msg->header.stream_id);
            srs_freep(msg);
            continue;
        }
        
        if ((ret = on_recv_message(msg)) != ERROR_SUCCESS) {
            srs_error("hook the received msg failed. ret=%d", ret);
            srs_freep(msg);
            return ret;
        }
        
        srs_trace("got a msg, cid=%d, type=%d, size=%d, time=%"PRId64, 
            msg->header.perfer_cid, msg->header.message_type, msg->header.payload_length, 
            msg->header.timestamp);
        *pmsg = msg;
        break;
    }
    
    return ret;
}

进入接受recv_interlaced_message（）函数，

 

int SrsProtocol::recv_interlaced_message(SrsCommonMessage** pmsg)
{
    int ret = ERROR_SUCCESS;
    
    // chunk stream basic header.
    char fmt = 0;
    int cid = 0;
    // 检查基本信息   1、CSID  2、Chunk type:0,1,2,3
    if ((ret = read_basic_header(fmt, cid)) != ERROR_SUCCESS) {
        if (ret != ERROR_SOCKET_TIMEOUT && !srs_is_client_gracefully_close(ret)) {
            srs_error("read basic header failed. ret=%d", ret);
        }
        return ret;
    }
    srs_verbose("read basic header success. fmt=%d, cid=%d", fmt, cid);
    
    // the cid must not negative.
    srs_assert(cid >= 0);
    
    // get the cached chunk stream.
    SrsChunkStream* chunk = NULL;
    
    // use chunk stream cache to get the chunk info.
    // @see https://github.com/ossrs/srs/issues/249
    if (cid < SRS_PERF_CHUNK_STREAM_CACHE) {
        // chunk stream cache hit.
        srs_verbose("cs-cache hit, cid=%d", cid);
        // already init, use it direclty
        chunk = cs_cache[cid];
        srs_verbose("cached chunk stream: fmt=%d, cid=%d, size=%d, message(type=%d, size=%d, time=%"PRId64", sid=%d)",
            chunk->fmt, chunk->cid, (chunk->msg? chunk->msg->size : 0), chunk->header.message_type, chunk->header.payload_length,
            chunk->header.timestamp, chunk->header.stream_id);
    } else {
        // chunk stream cache miss, use map.
        if (chunk_streams.find(cid) == chunk_streams.end()) {
            chunk = chunk_streams[cid] = new SrsChunkStream(cid);
            // set the perfer cid of chunk,
            // which will copy to the message received.
            chunk->header.perfer_cid = cid;
            srs_verbose("cache new chunk stream: fmt=%d, cid=%d", fmt, cid);
        } else {
            chunk = chunk_streams[cid];
            srs_verbose("cached chunk stream: fmt=%d, cid=%d, size=%d, message(type=%d, size=%d, time=%"PRId64", sid=%d)",
                chunk->fmt, chunk->cid, (chunk->msg? chunk->msg->size : 0), chunk->header.message_type, chunk->header.payload_length,
                chunk->header.timestamp, chunk->header.stream_id);
        }
    }
    // 检查信息头信息
    // chunk stream message header
    if ((ret = read_message_header(chunk, fmt)) != ERROR_SUCCESS) {
        if (ret != ERROR_SOCKET_TIMEOUT && !srs_is_client_gracefully_close(ret)) {
            srs_error("read message header failed. ret=%d", ret);
        }
        return ret;
    }
    srs_verbose("read message header success. "
            "fmt=%d, ext_time=%d, size=%d, message(type=%d, size=%d, time=%"PRId64", sid=%d)", 
            fmt, chunk->extended_timestamp, (chunk->msg? chunk->msg->size : 0), chunk->header.message_type, 
            chunk->header.payload_length, chunk->header.timestamp, chunk->header.stream_id);

	//获取rtmp负载信息
    // read msg payload from chunk stream.
    SrsCommonMessage* msg = NULL;
    if ((ret = read_message_payload(chunk, &msg)) != ERROR_SUCCESS) {
        if (ret != ERROR_SOCKET_TIMEOUT && !srs_is_client_gracefully_close(ret)) {
            srs_error("read message payload failed. ret=%d", ret);
        }
        return ret;
    }
    // 没有获得一个完整的RTMP信息，试着获取下一个块流
    // not got an entire RTMP message, try next chunk.
    if (!msg) {
        srs_verbose("get partial message success. size=%d, message(type=%d, size=%d, time=%"PRId64", sid=%d)",
                (msg? msg->size : (chunk->msg? chunk->msg->size : 0)), chunk->header.message_type, chunk->header.payload_length,
                chunk->header.timestamp, chunk->header.stream_id);
        return ret;
    }
    // 获取一个完整的RTMP信息
    *pmsg = msg;
    srs_info("get entire message success. size=%d, message(type=%d, size=%d, time=%"PRId64", sid=%d)",
            (msg? msg->size : (chunk->msg? chunk->msg->size : 0)), chunk->header.message_type, chunk->header.payload_length,
            chunk->header.timestamp, chunk->header.stream_id);
            
    return ret;
}

这里面就有一些相关Rtmp协议的东西，检车Basic Header、Message Header获取payload信息，ok，rtmp信息接收到了。

 

2、Rtmp信息处理

 

 

int SrsPublishRecvThread::handle(SrsCommonMessage* msg)
{
    int ret = ERROR_SUCCESS;
    
    // when cid changed, change it.
    if (ncid != cid) {
        _srs_context->set_id(ncid);
        cid = ncid;
    }
	
    _nb_msgs++;
	
    if (msg->header.is_video()) {
	// 接受到的视频帧数量
        video_frames++;
	srs_trace("message:%d, video:%d\n", _nb_msgs, video_frames);
    }
    
    // log to show the time of recv thread.
    srs_trace("recv thread now=%"PRId64"us, got msg time=%"PRId64"ms, size=%d",
        srs_update_system_time_ms(), msg->header.timestamp, msg->size);

    // the rtmp connection will handle this message
    ret = _conn->handle_publish_message(_source, msg, _is_fmle, _is_edge);

    // must always free it,
    // the source will copy it if need to use.
    srs_freep(msg);
    
    return ret;
}

首先做了一个统计，统计所有接受到的数据块和视频数据块。

 

进入process_publish_message（）函数

 

int SrsRtmpConn::process_publish_message(SrsSource* source, SrsCommonMessage* msg, bool vhost_is_edge)
{
    int ret = ERROR_SUCCESS;
    //发布到源站服务器
    // for edge, directly proxy message to origin.
    if (vhost_is_edge) {
        if ((ret = source->on_edge_proxy_publish(msg)) != ERROR_SUCCESS) {
            srs_error("edge publish proxy msg failed. ret=%d", ret);
            return ret;
        }
        return ret;
    }
    
    // process audio packet
    if (msg->header.is_audio()) {
        if ((ret = source->on_audio(msg)) != ERROR_SUCCESS) {
            srs_error("source process audio message failed. ret=%d", ret);
            return ret;
        }
        return ret;
    }
    // process video packet
    if (msg->header.is_video()) {
        if ((ret = source->on_video(msg)) != ERROR_SUCCESS) {
            srs_error("source process video message failed. ret=%d", ret);
            return ret;
        }
        return ret;
    }
    
    // process aggregate packet
    if (msg->header.is_aggregate()) {
        if ((ret = source->on_aggregate(msg)) != ERROR_SUCCESS) {
            srs_error("source process aggregate message failed. ret=%d", ret);
            return ret;
        }
        return ret;
    }
    
    // process onMetaData
    if (msg->header.is_amf0_data() || msg->header.is_amf3_data()) {
        SrsPacket* pkt = NULL;
        if ((ret = rtmp->decode_message(msg, &pkt)) != ERROR_SUCCESS) {
            srs_error("decode onMetaData message failed. ret=%d", ret);
            return ret;
        }
        SrsAutoFree(SrsPacket, pkt);
        if (dynamic_cast(pkt)) {
            SrsOnMetaDataPacket* metadata = dynamic_cast(pkt);
            if ((ret = source->on_meta_data(msg, metadata)) != ERROR_SUCCESS) {
                srs_error("source process onMetaData message failed. ret=%d", ret);
                return ret;
            }
            srs_info("process onMetaData message success.");
            return ret;
        }
        
        srs_info("ignore AMF0/AMF3 data message.");
        return ret;
    }
    
    return ret;
}

该函数做了一下几点处理：

 

（1）发布到源站服务器（相关CDN知识请参考：http://blog.csdn.net/ManagerUser/article/details/73909997）

（2）处理音频包

（3）处理视频包

注意：这里对音视频数据包的处理差不多，所有只看一下视频数据包的处理即可。

进入on_video（）函数

 

int SrsSource::on_video(SrsCommonMessage* shared_video)
{
    int ret = ERROR_SUCCESS;
    
    // monotically increase detect.
    if (!mix_correct && is_monotonically_increase) {
        if (last_packet_time > 0 && shared_video->header.timestamp < last_packet_time) {
            is_monotonically_increase = false;
            srs_warn("VIDEO: stream not monotonically increase, please open mix_correct.");
        }
    }
    last_packet_time = shared_video->header.timestamp;
    
    // drop any unknown header video.
    // @see https://github.com/ossrs/srs/issues/421
    if (!SrsFlvCodec::video_is_acceptable(shared_video->payload, shared_video->size)) {
        char b0 = 0x00;
        if (shared_video->size > 0) {
            b0 = shared_video->payload[0];
        }
        
        srs_warn("drop unknown header video, size=%d, bytes[0]=%#x", shared_video->size, b0);
        return ret;
    }
    
    // convert shared_video to msg, user should not use shared_video again.
    // the payload is transfer to msg, and set to NULL in shared_video.
    //将SrsCommonMessage类的所有数据完全拷贝给SrsSharedPtrMessage
    SrsSharedPtrMessage msg;
    if ((ret = msg.create(shared_video)) != ERROR_SUCCESS) {
        srs_error("initialize the video failed. ret=%d", ret);
        return ret;
    }
    srs_info("Video dts=%"PRId64", size=%d", msg.timestamp, msg.size);
    
    // directly process the audio message.
    if (!mix_correct) {
        return on_video_imp(&msg);
    }
    // msg.copy()返回msg自己指针
    // insert msg to the queue.
    mix_queue->push(msg.copy());
    
    // fetch someone from mix queue.
    SrsSharedPtrMessage* m = mix_queue->pop();
    if (!m) {
        return ret;
    }
    //消费单调递增消息
    // consume the monotonically increase message.
    if (m->is_audio()) {
        ret = on_audio_imp(m);
    } else {
        ret = on_video_imp(m);
    }
    srs_freep(m);
    
    return ret;
}

该函数做了rtmp块数据的拷贝，不用太在意，只关系接受到的rtmp信息在SrsCommonMessage类中即可。

 

进入on_video_imp()

 

int SrsSource::on_video_imp(SrsSharedPtrMessage* msg)
{
    int ret = ERROR_SUCCESS;
    
    srs_info("Video dts=%"PRId64", size=%d", msg->timestamp, msg->size);
    
    bool is_sequence_header = SrsFlvCodec::video_is_sequence_header(msg->payload, msg->size);
    
    // whether consumer should drop for the duplicated sequence header.
    bool drop_for_reduce = false;
    if (is_sequence_header && cache_sh_video && _srs_config->get_reduce_sequence_header(_req->vhost)) {
        if (cache_sh_video->size == msg->size) {
            drop_for_reduce = srs_bytes_equals(cache_sh_video->payload, msg->payload, msg->size);
            srs_warn("drop for reduce sh video, size=%d", msg->size);
        }
    }
    // cache the sequence header if h264
    // donot cache the sequence header to gop_cache, return here.
    if (is_sequence_header) {
        srs_freep(cache_sh_video);
	// 缓存到shareMessage
        cache_sh_video = msg->copy();
        
        // parse detail audio codec
        SrsAvcAacCodec codec;
        
        // user can disable the sps parse to workaround when parse sps failed.
        // @see https://github.com/ossrs/srs/issues/474
        // 是否解析SPS
        codec.avc_parse_sps = _srs_config->get_parse_sps(_req->vhost);

	// AAC 高级音频编码
	// AVC 高级视频编码
        SrsCodecSample sample;
        if ((ret = codec.video_avc_demux(msg->payload, msg->size, &sample)) != ERROR_SUCCESS) {
            srs_error("source codec demux video failed. ret=%d", ret);
            return ret;
        }
        
        // when got video stream info.
        SrsStatistic* stat = SrsStatistic::instance();
        if ((ret = stat->on_video_info(_req, SrsCodecVideoAVC, codec.avc_profile, codec.avc_level)) != ERROR_SUCCESS) {
            return ret;
        }
        
        srs_trace("%dB video sh,  codec(%d, profile=%s, level=%s, %dx%d, %dkbps, %dfps, %ds)",
            msg->size, codec.video_codec_id,
            srs_codec_avc_profile2str(codec.avc_profile).c_str(),
            srs_codec_avc_level2str(codec.avc_level).c_str(), codec.width, codec.height,
            codec.video_data_rate / 1000, codec.frame_rate, codec.duration);
    }
    
#ifdef SRS_AUTO_HLS
    if ((ret = hls->on_video(msg, is_sequence_header)) != ERROR_SUCCESS) {
        // apply the error strategy for hls.
        // @see https://github.com/ossrs/srs/issues/264
        std::string hls_error_strategy = _srs_config->get_hls_on_error(_req->vhost);
        if (srs_config_hls_is_on_error_ignore(hls_error_strategy)) {
            srs_warn("hls process video message failed, ignore and disable hls. ret=%d", ret);
            
            // unpublish, ignore ret.
            hls->on_unpublish();
            
            // ignore.
            ret = ERROR_SUCCESS;
        } else if (srs_config_hls_is_on_error_continue(hls_error_strategy)) {
            if (srs_hls_can_continue(ret, cache_sh_video, msg)) {
                ret = ERROR_SUCCESS;
            } else {
                srs_warn("hls continue video failed. ret=%d", ret);
                return ret;
            }
        } else {
            srs_warn("hls disconnect publisher for video error. ret=%d", ret);
            return ret;
        }
    }
#endif
    
#ifdef SRS_AUTO_DVR
    if ((ret = dvr->on_video(msg)) != ERROR_SUCCESS) {
        srs_warn("dvr process video message failed, ignore and disable dvr. ret=%d", ret);
        
        // unpublish, ignore ret.
        dvr->on_unpublish();
        
        // ignore.
        ret = ERROR_SUCCESS;
    }
#endif

#ifdef SRS_AUTO_HDS
    if ((ret = hds->on_video(msg)) != ERROR_SUCCESS) {
        srs_warn("hds process video message failed, ignore and disable dvr. ret=%d", ret);
        
        // unpublish, ignore ret.
        hds->on_unpublish();
        // ignore.
        ret = ERROR_SUCCESS;
    }
#endif
    // 将信息入到消费者堆栈里面
    // copy to all consumer
    if (!drop_for_reduce) {;
        for (int i = 0; i < (int)consumers.size(); i++) {
            SrsConsumer* consumer = consumers.at(i);
            if ((ret = consumer->enqueue(msg, atc, jitter_algorithm)) != ERROR_SUCCESS) {
                srs_error("dispatch the video failed. ret=%d", ret);
                return ret;
            }
        }
        srs_info("dispatch video success.");
    }

    // copy to all forwarders.
    if (!forwarders.empty()) {
		printf("forward to other server");
        std::vector::iterator it;
        for (it = forwarders.begin(); it != forwarders.end(); ++it) {
            SrsForwarder* forwarder = *it;
            if ((ret = forwarder->on_video(msg)) != ERROR_SUCCESS) {
                srs_error("forwarder process video message failed. ret=%d", ret);
                return ret;
            }
        }
    }
    
    // when sequence header, donot push to gop cache and adjust the timestamp.
    if (is_sequence_header) {
        return ret;
    }

    // cache the last gop packets
    if ((ret = gop_cache->cache(msg)) != ERROR_SUCCESS) {
        srs_error("gop cache msg failed. ret=%d", ret);
        return ret;
    }
    srs_verbose("cache gop success.");
    
    // if atc, update the sequence header to abs time.
    if (atc) {
        if (cache_sh_video) {
            cache_sh_video->timestamp = msg->timestamp;
        }
        if (cache_metadata) {
            cache_metadata->timestamp = msg->timestamp;
        }
    }
    
    return ret;
}

该函数有点长，也非常重要，分为段来介绍；

前半段代码通过SrsAvcAacCodec类来检测视频编码，这里RTMP协议支持视频编码格式H264，音频AAC。

后半部分代码主要做了转发入栈操作：

（1）HLS转发（该部分内容在后面文章中介绍）

（2）将rtmp信息转发给消费者列表（用户列表，用户在play的时候直接到该列表中去数据并进行play）

（3）将rtmp信息forward到其他服务器（该部分内容在后面章节中介绍）

 

五、Publish总结

 

至此，rtmp信息publish过程就结束了，总结一下：

    （1）rtmp协议链接。包括握手，创建网络连接，控制信息交互。

    （2）创建单独的接受线程去处理Rtmp信息接受。

    （3）缓存数据到SrsCommonMessage类。

    （4）数据转发，转发给消费者，forward给其他服务器（边缘服务器或者源站服务器）