Re: [h323plus] Twice H323Connection::Unlock() call issue

28 Dec 2015


      Hi Jan, 
I think, your solution will work, but in my opinion we have to fix  H323Connection::SendSignalSetup function, because it (only) has strange behavior. It has different internal state (locked and unlocked) for the same return reason (Not H323Connection::EndedByCallerAbort). 
Take a look to my proposition (a bit draft variant but in anyway) - in worst case (if not able to Lock the mutexes) we only lost the real reason (why does it fails) and return EndedByCallerAbort. 
H323Connection::SendSignalSetup 
{       CallEndReason reason = NumCallEndReasons;
//... 
      // Do the transport connect       connectionState = AwaitingTransportConnect; 
      // Release the mutex as can deadlock trying to clear call during connect.       Unlock(); 
      PBoolean connectFailed = false;       if (!signallingChannel->IsOpen()) {         signallingChannel->SetWriteTimeout(100);         connectFailed = !signallingChannel->Connect();       } 
      // See if transport connect failed, abort if so.       if (connectFailed) {         connectionState = NoConnectionActive;         switch (signallingChannel->GetErrorNumber()) {           case ENETUNREACH :             reason = EndedByUnreachable;             break;           case ECONNREFUSED :             reason = EndedByNoEndPoint;             break;           case ETIMEDOUT :             reason =  EndedByHostOffline;             break;           default:              reason = EndedByConnectFail;         }       } 
  }  
  // Lock while checking for shutting down.   if (!Lock())     return EndedByCallerAbort;   else if(NumCallEndReasons != reason)     return reason; 
//... 
    return reason; } 
Thanks -- Iurii Gordiienko
--- Original message --- 
From: "Jan Willamowius" jan@willamowius.de 
Date: 27 December 2015, 19:47:21
Hi Iurii,
PTLib 2.10.x and 2.11.x have a method to check, if a mutex is already
locked, but they changed the API and from 2.12.x on there is no way to
check except to try to lock it.
Replace H323Connection::Unlock() in src/h323.cxx with this version:
void H323Connection::Unlock()
{
  // calling Signal() on a mutex that isn't locked is causing an assert,
  // so we make sure it is locked before the Signal()
  innerMutex.Wait(0);
  innerMutex.Signal();
  outerMutex.Wait(0);
  outerMutex.Signal();
}
Please try the change and report back how it works.
Then I'll commit the code to the CVS.
Regards,
Jan
-- 
Jan Willamowius, Founder of the GNU Gatekeeper Project
EMail  : jan@willamowius.de 
Website: http://www.gnugk.org 
Support: http://www.willamowius.com/gnugk-support.html 

Relaxed Communications GmbH
Frahmredder 91
22393 Hamburg
Geschäftsführer: Jan Willamowius
HRB 125261 (Amtsgericht Hamburg)
USt-IdNr: DE286003584


Iurii Gordiienko wrote:
> Hi, 
> I have found a bug with twice H323Connection::Unlock() call if we have an connection issue. 
> Take a look: 
> 1.We are running connection.SendSignalSetup function from H225CallThread::Main() and if it returns any reason!=H323Connection::EndedByCallerAbort we will Unlock the connection. 
> 
> void H225CallThread::Main() {   PTRACE(3, "H225\tStarted call thread"); 
>   if (connection.Lock()) {     H323Connection::CallEndReason reason = connection.SendSignalSetup(alias, address); 
>     // Special case, if we aborted the call then already will be unlocked     if (reason != H323Connection::EndedByCallerAbort)       connection.Unlock(); 
>     // bla-bla-bla } 
> 
> 2.But we have case when connection.SendSignalSetup returns non H323Connection::EndedByCallerAbort reason but has own unlocked state - for this case PTLIB will unlock the mutex twice with assert (as minimum) message. 
> H323Connection::SendSignalSetup(const PString & alias, 
>                                                               const H323TransportAddress & address) { //bla-bla-bla 
>       // Release the mutex as can deadlock trying to clear call during connect.       Unlock(); 
>       PBoolean connectFailed = false;       if (!signallingChannel->IsOpen()) {         signallingChannel->SetWriteTimeout(100);         connectFailed = !signallingChannel->Connect();       } 
>       // See if transport connect failed, abort if so.       if (connectFailed) {         connectionState = NoConnectionActive;         switch (signallingChannel->GetErrorNumber()) {           case ENETUNREACH :             return EndedByUnreachable;           case ECONNREFUSED :             return EndedByNoEndPoint;           case ETIMEDOUT :             return EndedByHostOffline;         }         return EndedByConnectFail;       } 
> } 
> Thanks -- Iurii Gordiienko