在前几篇文章说到的EventThread事件,我们可以知道EventThread这个线程负责端口的监听,监听的事件主要有两类,第一类是有新的RTSP连接请求事 件,第二类就是在原来已经连接的RTSP连接请求上响应消息事件,比如PALY,Setup,Describe等等。文章参考了http://blog.youkuaiyun.com/longlong530/article/details/43451717的文章在这里标示感谢!
那么对于监听到的事件,应该怎么样处理了,就在这个RTSPSession的这个方法中,我们详细的分析下这个方法。
主要看这个RTSPSession::Run()方法 在Run的switch这个语句中有一些状态的判断。
1. 状态机 kReadingFirstRequest
通过fInputStream.ReadRequest()调用Session绑定的socket,获取数据,该函数有两个返回码:
a) QTSS_NoErr,意味着已经收到该socket的完整数据,但是没有收集到本次请求的全部数据,这时需要接着请求监听读事件,获取更多数据组成完整的RTSP请求消息。
b) QTSS_RequestArrived,意味着此时已经收到完整请求的报文,可以正常进入下一状态机了。
c) E2BIG,表示缓存区已经溢出,默认缓存区的大小为( kRequestBufferSizeInBytes = 4096),进入kHaveNonTunnelMessage状态机,然后在改状态机下响应错误
case kReadingFirstRequest:
{
if ((err = fInputStream.ReadRequest()) == QTSS_NoErr)
{
// If the RequestStream returns QTSS_NoErr, it means
// that we've read all outstanding data off the socket,
// and still don't have a full request. Wait for more data.
//+rt use the socket that reads the data, may be different now.
fInputSocketP->RequestEvent(EV_RE);
return 0;
}
if ((err != QTSS_RequestArrived) && (err != E2BIG))
{
// Any other error implies that the client has gone away. At this point,
// we can't have 2 sockets, so we don't need to do the "half closed" check
// we do below
Assert(err > 0);
Assert(!this->IsLiveSession());
break;
}
if (err == QTSS_RequestArrived)
fState = kHTTPFilteringRequest;
// If we get an E2BIG, it means our buffer was overfilled.
// In that case, we can just jump into the following state, and
// the code their does a check for this error and returns an error.
if (err == E2BIG)
fState = kHaveNonTunnelMessage;
}
3. 状态机 kHaveNonTunnelMessage
进入此状态,说明请求报文格式是正确的,请求已进入受理状态,具体操作步骤如下:
a) 创建RTSPRequest对象,用于解析RTSP消息;
b) 此状态中对fReadMutex,fSessionMutex进行加锁,禁止在处理报文的过程中接收以RTP Interleaved接收RTP数据或者发出RTSP响应报文
c) 对错误码E2BIG、QTSS_BadArgument进行处理,响应qtssClientBadRequest;
d) 将状态机跳转到kFilteringRequest下;
case kHaveNonTunnelMessage:
{
// should only get here when fInputStream has a full message built.
Assert( fInputStream.GetRequestBuffer() );
Assert(fRequest == NULL);
fRequest = NEW RTSPRequest(this);
fRoleParams.rtspRequestParams.inRTSPRequest = fRequest;
fRoleParams.rtspRequestParams.inRTSPHeaders = fRequest->GetHeaderDictionary();
// We have an RTSP request and are about to begin processing. We need to
// make sure that anyone sending interleaved data on this session won't
// be allowed to do so until we are done sending our response
// We also make sure that a POST session can't snarf in while we're
// processing the request.
fReadMutex.Lock();
fSessionMutex.Lock();
// The fOutputStream's fBytesWritten counter is used to
// count the # of bytes for this RTSP response. So, at
// this point, reset it to 0 (we can then just let it increment
// until the next request comes in)
fOutputStream.ResetBytesWritten();
// Check for an overfilled buffer, and return an error.
if (err == E2BIG)
{
(void)QTSSModuleUtils::SendErrorResponse(fRequest, qtssClientBadRequest,
qtssMsgRequestTooLong);
fState = kPostProcessingRequest;
break;
}
// Check for a corrupt base64 error, return an error
if (err == QTSS_BadArgument)
{
(void)QTSSModuleUtils::SendErrorResponse(fRequest, qtssClientBadRequest,
qtssMsgBadBase64);
fState = kPostProcessingRequest;
break;
}
Assert(err == QTSS_RequestArrived);
fState = kFilteringRequest;
// Note that there is no break here. We'd like to continue onto the next
// state at this point. This goes for every case in this case statement
}
4. 状态机kFilteringRequest
a) 刷新超时任务fTimeoutTask,运转RTSP会话的超时机制。
b) 通过fIsDataPacket属性进行判断当前数据是否是一个数据包,而不是一个信令消息。该属性判断方法在RTSPRequestStream::ReadRequest()中。RTP包格式以$字符开头,后面紧跟着一个字节是信道标示符,后面两个字节是数字长度,Darwin就用这个字符区分是否为数据包。
c) 这时第一次开始调用module了,角色为kRTSPFilterRole。注册了该角色模块只有一个QTSSRefMovieModule,
d) SetupRequest(),解析RTSP消息,同时创建一个客户会话(RTPSession),和产生当前请求的客户端连接相关联,这个会话会一直保持,直到客户端的流播放结束。
注意:服务器根据被调用的模块是否对请求做了应答来决定后面的调用(方法HasResponseBeenSent()),如果注册了RTSP Filter Role的某一个模块在被调用的时候对请求作出了应答,服务器将立即调用注册了RTSPPostprocessorRole的模块,不再调用其他尚未调用的注册了RTSP Filter Role的模块,否则服务器调用其它注册了RTSP Filter Role的模块。
case kFilteringRequest:
{
// We received something so auto refresh
// The need to auto refresh is because the api doesn't allow a module to refresh at this point
//
fTimeoutTask.RefreshTimeout();
//
// Before we even do this, check to see if this is a *data* packet,
// in which case this isn't an RTSP request, so we don't need to go
// through any of the remaining steps
if (fInputStream.IsDataPacket()) // can this interfere with MP3?
{
this->HandleIncomingDataPacket();
fState = kCleaningUp;
break;
}
//
// In case a module wants to replace the request
char* theReplacedRequest = NULL;
char* oldReplacedRequest = NULL;
// Setup the filter param block
QTSS_RoleParams theFilterParams;
theFilterParams.rtspFilterParams.inRTSPSession = this;
theFilterParams.rtspFilterParams.inRTSPRequest = fRequest;
theFilterParams.rtspFilterParams.outNewRequest = &theReplacedRequest;
// Invoke filter modules
numModules = QTSServerInterface::GetNumModulesInRole(QTSSModule::kRTSPFilterRole);
for (; (fCurrentModule < numModules) && ((!fRequest->HasResponseBeenSent()) || fModuleState.eventRequested); fCurrentModule++)
{
fModuleState.eventRequested = false;
fModuleState.idleTime = 0;
if (fModuleState.globalLockRequested )
{ fModuleState.globalLockRequested = false;
fModuleState.isGlobalLocked = true;
}
theModule = QTSServerInterface::GetModule(QTSSModule::kRTSPFilterRole, fCurrentModule);
(void)theModule->CallDispatch(QTSS_RTSPFilter_Role, &theFilterParams);
fModuleState.isGlobalLocked = false;
// If this module has requested an event, return and wait for the event to transpire
if (fModuleState.globalLockRequested) // call this request back locked
return this->CallLocked();
if (fModuleState.eventRequested)
{
this->ForceSameThread(); // We are holding mutexes, so we need to force
// the same thread to be used for next Run()
return fModuleState.idleTime; // If the module has requested idle time...
}
//
// Check to see if this module has replaced the request. If so, check
// to see if there is an old replacement that we should delete
if (theReplacedRequest != NULL)
{
if (oldReplacedRequest != NULL)
delete [] oldReplacedRequest;
fRequest->SetVal(qtssRTSPReqFullRequest, theReplacedRequest, ::strlen(theReplacedRequest));
oldReplacedRequest = theReplacedRequest;
theReplacedRequest = NULL;
}
}
fCurrentModule = 0;
if (fRequest->HasResponseBeenSent())
{
fState = kPostProcessingRequest;
break;
}
if (fSentOptionsRequest && this->ParseOptionsResponse())
{
fRoundTripTime = (SInt32) (OS::Milliseconds() - fOptionsRequestSendTime);
//qtss_printf("RTSPSession::Run RTT time = %"_S32BITARG_" msec\n", fRoundTripTime);
fState = kSendingResponse;
break;
}
else
// Otherwise, this is a normal request, so parse it and get the RTPSession.
this->SetupRequest();
// This might happen if there is some syntax or other error,
// or if it is an OPTIONS request
if (fRequest->HasResponseBeenSent())
{
fState = kPostProcessingRequest;
break;
}
fState = kRoutingRequest;
}
5. 状态机kPreprocessingRequest
遍历调用所有注册了QTSS_RTSPPreProcessor_Role角色的模块。在这个角色模式下,分别处理了每种RTSP消息,比如本次的点播请求的Describe、Setup、Play指令,模块中针对各种消息都有对应的单独函数处理。 处理完每次RTSP请求后即进入下一状态kPostProcessingRequest,待下轮循环进入本状态机再处理下一个RTSP消息。
case kPreprocessingRequest:
{
// Invoke preprocessor modules
numModules = QTSServerInterface::GetNumModulesInRole(QTSSModule::kRTSPPreProcessorRole);
{
// Manipulation of the RTPSession from the point of view of
// a module is guarenteed to be atomic by the API.
Assert(fRTPSession != NULL);
OSMutexLocker locker(fRTPSession->GetSessionMutex());
for (; (fCurrentModule < numModules) && ((!fRequest->HasResponseBeenSent()) || fModuleState.eventRequested); fCurrentModule++)
{
fModuleState.eventRequested = false;
fModuleState.idleTime = 0;
if (fModuleState.globalLockRequested )
{ fModuleState.globalLockRequested = false;
fModuleState.isGlobalLocked = true;
}
theModule = QTSServerInterface::GetModule(QTSSModule::kRTSPPreProcessorRole, fCurrentModule);
(void)theModule->CallDispatch(QTSS_RTSPPreProcessor_Role, &fRoleParams);
fModuleState.isGlobalLocked = false;
// The way the API is set up currently, the first module that adds a stream
// to the session is responsible for sending RTP packets for the session.
if (fRTPSession->HasAnRTPStream() && (fRTPSession->GetPacketSendingModule() == NULL))
fRTPSession->SetPacketSendingModule(theModule);
if (fModuleState.globalLockRequested) // call this request back locked
return this->CallLocked();
// If this module has requested an event, return and wait for the event to transpire
if (fModuleState.eventRequested)
{
this->ForceSameThread(); // We are holding mutexes, so we need to force
// the same thread to be used for next Run()
return fModuleState.idleTime; // If the module has requested idle time...
}
}
}
fCurrentModule = 0;
if (fRequest->HasResponseBeenSent())
{
fState = kPostProcessingRequest;
break;
}
fState = kProcessingRequest;
}
今天详细的看了下RTSPSession的会话处理代码,该模块主要是Run方法内对会话的状态机进行管理,实现对RTSP会话的处理,我以点播Movie文件夹下的视频文件为例,深入的学习了(Describe、Setup、Play的点播流程),并按照状态机的处理顺序对他们进行一一解析,如下:
【转载请注明出处】:http://blog.youkuaiyun.com/longlong530
1. 状态机 kReadingFirstRequest
通过fInputStream.ReadRequest()调用Session绑定的socket,获取数据,该函数有两个返回码:
a) QTSS_NoErr,意味着已经收到该socket的完整数据,但是没有收集到本次请求的全部数据,这时需要接着请求监听读事件,获取更多数据组成完整的RTSP请求消息。
b) QTSS_RequestArrived,意味着此时已经收到完整请求的报文,可以正常进入下一状态机了。
c) E2BIG,表示缓存区已经溢出,默认缓存区的大小为( kRequestBufferSizeInBytes = 4096),进入kHaveNonTunnelMessage状态机,然后在改状态机下响应错误
2. 状态机 kHTTPFilteringRequest
该状态机下检测是否为RTSP-over-HTTP tunneling,并直接切换到kHaveNonTunnelMessage状态。由于与RTSP无关,该部分的函数PreFilterForHTTPProxyTunnel(),暂时就不深入研究了。
3. 状态机 kHaveNonTunnelMessage
进入此状态,说明请求报文格式是正确的,请求已进入受理状态,具体操作步骤如下:
a) 创建RTSPRequest对象,用于解析RTSP消息;
b) 此状态中对fReadMutex,fSessionMutex进行加锁,禁止在处理报文的过程中接收以RTP Interleaved接收RTP数据或者发出RTSP响应报文
c) 对错误码E2BIG、QTSS_BadArgument进行处理,响应qtssClientBadRequest;
d) 将状态机跳转到kFilteringRequest下;
4. 状态机kFilteringRequest
a) 刷新超时任务fTimeoutTask,运转RTSP会话的超时机制。
b) 通过fIsDataPacket属性进行判断当前数据是否是一个数据包,而不是一个信令消息。该属性判断方法在RTSPRequestStream::ReadRequest()中。RTP包格式以$字符开头,后面紧跟着一个字节是信道标示符,后面两个字节是数字长度,Darwin就用这个字符区分是否为数据包。
c) 这时第一次开始调用module了,角色为kRTSPFilterRole。注册了该角色模块只有一个QTSSRefMovieModule,
d) SetupRequest(),解析RTSP消息,同时创建一个客户会话(RTPSession),和产生当前请求的客户端连接相关联,这个会话会一直保持,直到客户端的流播放结束。
注意: 服务器根据被调用的模块是否对请求做了应答来决定后面的调用(方法HasResponseBeenSent()),如果注册了RTSP Filter Role的某一个模块在被调用的时候对请求作出了应答,服务器将立即调用注册了RTSPPostprocessorRole的模块,不再调用其他尚未调用的注册了RTSP Filter Role的模块,否则服务器调用其它注册了RTSP Filter Role的模块。
5. 状态机kRoutingRequest
这里只走到一个模块QTSSReflectorModule中,调用其RedirectBroadcast()方法,主要是增加两个字段到inParams->inRTSPRequest中。key值分别为qtssRTSPReqRootDir、qtssRTSPReqFilePath。 配置文件中如下两个字段并未设置,上面设置的路径逻辑就不会走到。
最后跳转到鉴权状态机kAuthenticatingRequest
6. 状态机kAuthenticatingRequest、kAuthorizingRequest。
这里对鉴权方面后续会单独深入研究,目前项目也不涉及。 鉴权完成后,跳转到核心的状态机kPreprocessingRequest中。
7. 状态机kPreprocessingRequest
遍历调用所有注册了QTSS_RTSPPreProcessor_Role角色的模块。在这个角色模式下,分别处理了每种RTSP消息,比如本次的点播请求的Describe、Setup、Play指令,模块中针对各种消息都有对应的单独函数处理。 处理完每次RTSP请求后即进入下一状态kPostProcessingRequest,待下轮循环进入本状态机再处理下一个RTSP消息。
(以下状态后续补充)
8. 状态机kPostProcessingRequest
9. 状态机kSendingResponse
10.状态机kCleaningUp
清理数据,同时释放kHaveNonTunnelMessage状态中获取的锁。
11.状态机kReadingRequest
【转载请注明出处】:http://blog.youkuaiyun.com/longlong530
简单的画了下RTSPSession的状态机流程,如下图:
本文详细解析RTSPSession中的状态机流程,包括事件监听、请求解析及响应处理等关键步骤,深入探讨Describe、Setup、Play指令的处理过程。
2704

被折叠的 条评论
为什么被折叠?



