index > Winsock Kernel (WSK) > Keepalive Timeout Detection

Keepalive Timeout Detection

Hi

I seem to be seeing some strange behavior in a half-open connection where keepalives stop receiving acknowledgements after a while.

I have a simple client and server talking to each other through WinSock apis. After establishing a connection from my client to the server, the server uses setsockopt to send keepalives to the client every 20 seconds. At this state, I see the keepalives being sent by the server and these packets are being acknowledged by the client. Subsequently, I issue a half-close through my client (send a FIN to close the send side of the connection, but leave the receive side open). Following this, I see that the server sends one set of keepalives that are acknowledged. But after some time, the keepalives on the server mysteriously stop receiving acknowledgements. The server sends about 5 such keepalives and then gives up. When I did a netstat, the connection was already gone on the server. However, my server application was not notified about this event. Is there a way by which I can be notified in my application of failed keepalives on the socket? I tried to do a getsockopt(sock, SOL_SOCKET, SO_ERROR...), but that did not return with any errors on this socket even after the keepalives started timing out. I tried a couple of other functions like getpeername, etc and even these returned success.

Any ideas on how I'd be able to detect keepalive timeouts on windows?

Thanks!

iknownoone

If your server application has issued a blocking read using API recv(...), which should be in blocking mode before this problem happens, then, as soon as max retries, 5, is reached, the tcp/ip driver will reset that connection. You'll see that the recv() operation returned with errno 10054. This reset will come from local TCP/IP stack not from remote peer. The following example shows a thread doing the recv() , but this could be done anywhere in your code, (e.g main function).

DWORD WINAPI MyReceiver(VOID threadArg)

{

INT nRet = 0;

while(1)

{

nRet = recv(......); // blocks if no data available

if(nRet <= 0)

{

// handle error

printf("recv() failed, errno %d\n",WSAGetLastError());

}

} // end of while

}

If keep alive max retries was reached, you "should" see something like this:

recv() failed, errno 10054

Hope it helps.

mauro007

mauro,

Thanks for your reply.

True that receives will return with an error, but if you observe, I dont have the option of calling a "recv" in my server code. The server has just received a FIN from the client to close one end of the connection (half-close) and so, I'll need some other way of detecting this error. Also, if you do a tcpdump, you will observe that the server does not send a "RST" for this connection (I tested this case).

iknownoone

reply 3