前面我们分析到WifiDisplaySource会调用ANetworkSession的接口去创建一个socket,并在这个socket上监听是否有客户端的连接请求。先来看看Wifi Display规范的一些流程图:
从之前的一篇文章中,当ANetworkSession创建好RTSP的listen socket后,就会把它加入到selelct中等待对方的连接,那我们首先来看ANetworkSession的threadLoop方法:
void ANetworkSession::threadLoop() { int res = select(maxFd + 1, &rs, &ws, NULL, NULL /* tv */); { Mutex::Autolock autoLock(mLock); List<sp<Session> > sessionsToAdd; for (size_t i = mSessions.size(); res > 0 && i-- > 0;) { const sp<Session> &session = mSessions.valueAt(i); int s = session->socket(); if (s < 0) { continue; } if (FD_ISSET(s, &rs) || FD_ISSET(s, &ws)) { --res; } if (FD_ISSET(s, &rs)) { if (session->isRTSPServer() || session->isTCPDatagramServer()) { struct sockaddr_in remoteAddr; socklen_t remoteAddrLen = sizeof(remoteAddr); int clientSocket = accept( s, (struct sockaddr *)&remoteAddr, &remoteAddrLen); if (clientSocket >= 0) { status_t err = MakeSocketNonBlocking(clientSocket); if (err != OK) { } else { in_addr_t addr = ntohl(remoteAddr.sin_addr.s_addr); ALOGI("incoming connection from %d.%d.%d.%d:%d " "(socket %d)", (addr >> 24), (addr >> 16) & 0xff, (addr >> 8) & 0xff, addr & 0xff, ntohs(remoteAddr.sin_port), clientSocket); sp<Session> clientSession = new Session( mNextSessionID++, Session::CONNECTED, clientSocket, session->getNotificationMessage()); clientSession->setMode( session->isRTSPServer() ? Session::MODE_RTSP : Session::MODE_DATAGRAM); sessionsToAdd.push_back(clientSession); } } else { ALOGE("accept returned error %d (%s)", errno, strerror(errno)); } } } while (!sessionsToAdd.empty()) { sp<Session> session = *sessionsToAdd.begin(); sessionsToAdd.erase(sessionsToAdd.begin()); mSessions.add(session->sessionID(), session); ALOGI("added clientSession %d", session->sessionID()); } }
上面在selelct循环中,首先只有刚创建的RTSP的listen socket,接着如果有客户端的连接请求,就会跳出select语句,然后调用accept去接收对方的连接。接着会去创建一个新的Session会话,我们去看它的构造函数:
ANetworkSession::Session::Session( int32_t sessionID, State state, int s, const sp<AMessage> ?ify) : mSessionID(sessionID), mState(state), mMode(MODE_DATAGRAM), mSocket(s), mNotify(notify), mSawReceiveFailure(false), mSawSendFailure(false), mUDPRetries(kMaxUDPRetries), mLastStallReportUs(-1ll) { if (mState == CONNECTED) { struct sockaddr_in localAddr; socklen_t localAddrLen = sizeof(localAddr); int res = getsockname( mSocket, (struct sockaddr *)&localAddr, &localAddrLen); CHECK_GE(res, 0); struct sockaddr_in remoteAddr; socklen_t remoteAddrLen = sizeof(remoteAddr); res = getpeername( mSocket, (struct sockaddr *)&remoteAddr, &remoteAddrLen); CHECK_GE(res, 0); sp<AMessage> msg = mNotify->dup(); msg->setInt32("sessionID", mSessionID); msg->setInt32("reason", kWhatClientConnected); msg->setString("server-ip", localAddrString.c_str()); msg->setInt32("server-port", ntohs(localAddr.sin_port)); msg->setString("client-ip", remoteAddrString.c_str()); msg->setInt32("client-port", ntohs(remoteAddr.sin_port)); msg->post(); } }
这里的状态mState为CONNECTED,所以会构建一个AMessage并post出去,由前一章的知识,我们知道这里的mNotify是一个kWhatRTSPNotify消息,会在WifiDisplaySource调用createRTSPServer时传进来的一个参数,所以这里创建的AMessage最终还是会被WifiDisplaySource去处理,跳过中间AMessage、ALooperRoster、ALooper的调用关系,我们直接到WifiDisplaySource的onMessageReceived去看如何处理:
case kWhatRTSPNotify: { int32_t reason; CHECK(msg->findInt32("reason", &reason)); switch (reason) { case ANetworkSession::kWhatError: { break; } case ANetworkSession::kWhatClientConnected: { int32_t sessionID; CHECK(msg->findInt32("sessionID", &sessionID)); if (mClientSessionID > 0) { ALOGW("A client tried to connect, but we already " "have one."); mNetSession->destroySession(sessionID); break; } CHECK_EQ(mState, AWAITING_CLIENT_CONNECTION); CHECK(msg->findString("client-ip", &mClientInfo.mRemoteIP)); CHECK(msg->findString("server-ip", &mClientInfo.mLocalIP)); if (mClientInfo.mRemoteIP == mClientInfo.mLocalIP) { // Disallow connections from the local interface // for security reasons. mNetSession->destroySession(sessionID); break; } CHECK(msg->findInt32( "server-port", &mClientInfo.mLocalPort)); mClientInfo.mPlaybackSessionID = -1; mClientSessionID = sessionID; ALOGI("We now have a client (%d) connected.", sessionID); mState = AWAITING_CLIENT_SETUP; status_t err = sendM1(sessionID); CHECK_EQ(err, (status_t)OK); break; } case ANetworkSession::kWhatData: { break; } case ANetworkSession::kWhatNetworkStall: { break; } default: TRESPASS(); } break; }
这里的reason是kWhatClientConnected,跳过前面不必要的case语句。如果先前已经连上其它的Sink device,这里就先断开之前的连接;如果没有,将新的SessionID赋予给mClientSessionID,并更改状态为AWAITING_CLIENT_SETUP,接着去看sendM1消息,这时候就要开始WifiDisplay M1~M7消息的发送了。
下面列举M1~M7消息的格式,有兴趣的可以去对照代码分析,我们后面着重分析M6(SetUp)和M7(Play)两个消息。
M1 reqeust:
OPTIONS * RTSP/1.0
Date: Tue, 29 Fri 2014 02:41:24 +0000
Server: stagefright/1.2 (Linux;Android 4.4)
CSeq: 1
Require: org.wfa.wfd1.0
M1 respose:
RTSP/1.0 200 OK
CSeq: 1
Date: Fri, Jan 01 2014 09:02:37 GMT
Public: org.wfa.wfd1.0, GET_PARAMETER, SET_PARAMETER
M2 request:
OPTIONS * RTSP/1.0
CSeq: 2
Require: org.wfa.wfd1.0
M2 response:
RTSP/1.0 200 OK
Date: Tue, 29 Fri 2014 02:41:25 +0000
Server: stagefright/1.2 (Linux;Android 4.3)
CSeq: 2
Public: org.wfa.wfd1.0, SETUP, TEARDOWN, PLAY, PAUSE, GET_PARAMETER, SET_PARAMETER
M3 request:
GET_PARAMETER rtsp://localhost/wfd1.0 RTSP/1.0
Date: Tue, 29 Fri 2014 02:41:25 +0000
Server: stagefright/1.2 (Linux;Android 4.3)
CSeq: 2
Content-Type: text/parameters
Content-Length: 83
wfd_content_protection
wfd_video_formats
wfd_audio_codecs
wfd_client_rtp_ports
M3 response:
RTSP/1.0 200 OK
CSeq: 2
Content-Length: 124
Content-Type: text/parameters
wfd_audio_codecs: LPCM 00000003 00, AAC 00000007 00
wfd_video_formats: 00 00 02 02 0000FFFF 0FFFFFFF 00000FFF 00 0000 0000 01 none none
wfd_content_protection: none
wfd_client_rtp_ports: RTP/AVP/UDP;unicast 19990 0 mode=play
M4 request:
SET_PARAMETER rtsp://localhost/wfd1.0 RTSP/1.0
Date: Tue, 29 Fri 2014 02:41:25 +0000
Server: stagefright/1.2 (Linux;Android 4.3)
CSeq: 3
Content-Type: text/parameters
Content-Length: 247
wfd_video_formats: 00 00 02 02 00000020 00000000 00000000 00 0000 0000 00 none none
wfd_audio_codecs: AAC 00000001 00
wfd_presentation_URL: rtsp://192.168.5.200/wfd1.0/streamid=0 none
wfd_client_rtp_ports: RTP/AVP/UDP;unicast 19990 0 mode=play
M4 response:
RTSP/1.0 200 OK
CSeq: 3
M5 request:
SET_PARAMETER rtsp://localhost/wfd1.0 RTSP/1.0
Date: Tue, 29 Fri 2014 02:41:25 +0000
Server: stagefright/1.2 (Linux;Android 4.3)
CSeq: 4
Content-Type: text/parameters
Content-Length: 27
wfd_trigger_method: SETUP
M5 response:
RTSP/1.0 200 OK
CSeq: 4
M6 request:
SETUP rtsp://192.168.5.200/wfd1.0/streamid=0 RTSP/1.0
CSeq: 3
Transport: RTP/AVP/UDP;unicast;client_port=19990
M6 response:
RTSP/1.0 200 OK
Date: Tue, 29 Fri 2014 02:41:25 +0000
Server: stagefright/1.2 (Linux;Android 4.3)
CSeq: 3
Session: 988982966;timeout=30
Transport: RTP/AVP/UDP;unicast;client_port=19990;server_port=22220
我们先来看处理M6 request的方法,代码在onSetupRequest中:
status_t WifiDisplaySource::onSetupRequest( int32_t sessionID, int32_t cseq, const sp<ParsedMessage> &data) { CHECK_EQ(sessionID, mClientSessionID); if (mClientInfo.mPlaybackSessionID != -1) { sendErrorResponse(sessionID, "400 Bad Request", cseq); return ERROR_MALFORMED; } int32_t playbackSessionID = makeUniquePlaybackSessionID(); sp<AMessage> notify = new AMessage(kWhatPlaybackSessionNotify, id()); notify->setInt32("playbackSessionID", playbackSessionID); notify->setInt32("sessionID", sessionID); sp<PlaybackSession> playbackSession = new PlaybackSession( mNetSession, notify, mInterfaceAddr, mHDCP, mMediaPath.c_str()); looper()->registerHandler(playbackSession); AString uri; data->getRequestField(1, &uri); if (strncasecmp("rtsp://", uri.c_str(), 7)) { sendErrorResponse(sessionID, "400 Bad Request", cseq); return ERROR_MALFORMED; } if (!(uri.startsWith("rtsp://") && uri.endsWith("/wfd1.0/streamid=0"))) { sendErrorResponse(sessionID, "404 Not found", cseq); return ERROR_MALFORMED; } RTPSender::TransportMode rtcpMode = RTPSender::TRANSPORT_UDP; if (clientRtcp < 0) { rtcpMode = RTPSender::TRANSPORT_NONE; } status_t err = playbackSession->init( mClientInfo.mRemoteIP.c_str(), clientRtp, rtpMode, clientRtcp, rtcpMode, mSinkSupportsAudio, mUsingPCMAudio, mSinkSupportsVideo, mChosenVideoResolutionType, mChosenVideoResolutionIndex, mChosenVideoProfile, mChosenVideoLevel); if (err != OK) { looper()->unregisterHandler(playbackSession->id()); playbackSession.clear(); } mClientInfo.mPlaybackSessionID = playbackSessionID; mClientInfo.mPlaybackSession = playbackSession; mState = AWAITING_CLIENT_PLAY; scheduleReaper(); scheduleKeepAlive(sessionID); return OK; }
跳过前面的关于RTSP回复消息的组织,这里还会创建一个PlaybackSession对象,并调用它的init方法做初始化。
根据前面的背景知识介绍,设备之间的交互将由Session来管理。在代码中,Session的概念由WifiDisplaySource的内部类PlaybackSession来表示。先来看和其相关的类图结构,如下图所示:
由上图可知:
- PlaybackSession及其内部类Track都从AHandler派生。故它们的工作也依赖于消息循环和处理。Track代表视频流或音频流。
- Track内部通过mMediaPull变量指向一个MediaPull对象。而MediaPull对象则保存了一个MediaSource对象。在PlaybackSession中,此MediaSource的真正类型为SurfaceMediaSource。它表明该Media的源来自Surface。
- BufferQueue从ISurfaceTexure中派生,根据前面对SurfaceFlinger的介绍,它就是SurfaceFlinger代码示例中代表虚拟设备的State的surface变量。
- 左图中,MediaPull通过kWhatPull消息不断调用MediaSource的read函数。
- 右图中,SurfaceMediaSource的read函数由通过mBufferQueue来读取数据。
-
那么mBufferQueue的数据来自什么地方呢?对,正是来自SurfaceFlinger。
当然,PlaybackSession拿到这些数据后还需要做编码,然后才能发送给远端设备。由于篇幅关系,本文就不再讨论这些问题了。
当双方设备准备就绪后,MediaPull会通过kWhatPull消息处理不断调用MediaSource的read函数。在SurfaceMediaSource实现的read函数中,来自SurfaceFlinger的混屏后的数据经由BufferQueue传递到MediaPull中。