Thread: uncertainty in tcp data arrival?

i have been fidgeting with winsock tcp ( i created simple client and server applications with only a main function )

(i know this isn't the code but i'm trying to get the idea across i can post code if necessary)

server

Code:

accept
recv ( 17 bytes )
shutdown          // listning and connected socket
closesocket       // listning and connected socket
exit application

client

Code:

connect
send ( 17 bytes )
Sleep( 5000 )
send ( 17 bytes )
shutdown
closesocket
exit application

i started the server and then started the client. ( i'm using blocking sockets )
so basicly what i tried to do was:

- make the server wait for a connection
- recieve the 17 bytes sent by the client
- and exit the server application ( closing the sockets properly )

- see the error on the client side on the second attempt to send 17 bytes

i was suprised to see that the second attemp to send returned 17 bytes ( expected a SOCKET_ERROR ) eventhough the server application was no longer running. of course i'm a bit disturbed. can someone explain what's happening?

The TCP socket layer is "optimistic" in that it reports success if the packet physically left your host or was queued for sending during packet coalesce. It just happens that at the moment you send the next 17 bytes, the server side has shut down. It should respond to the client packet with a TCP reset.

But in the meantime, your socket layer happily continues along. It won't be until after you receive the RST, and after you attempt yet AGAIN to send a packet, that you will see the error condition.

All is normal...

thanks for all the suggestions. i thought i'd close the topic properly because i already resolved this a while ago.

the problem was that i didn't understand the tcp shutdown sequence. i was slamming the connection shut by calling shutdown with the SD_BOTH flag. i didn't realize that both sides have to to agree that the connection is shut instead of one side suddenly sending RST packets.

so i'm calling shutdown with the SD_SEND flag instead now so that each sender shuts down their own sending. everything is working fine now ( no socket erros ). i guess the uncertainty was caused by me not shutting down properly.

this page's micro faq confirmed it
Winsock Programmer's FAQ: Debugging TCP/IP

Greeting,
symbiote