I am doing non-blocking TCP socket programming using select().
After normal set up,
socket()
make socket non-blocking
bind()
listen()
FD_ZERO
FD_SET
...
select()
FD_ISSET
...
s = accept()
For certain reason, after FD_ISSET the server does some preparation but then chooses not to accept. Easy way is just to close s after accept(). But before accept(), is there any way to tell the client that this connection should fail? Thank you.
No. If you get an incoming connection and you don’t want it after all, you will have to
accept()is andclose()it right away. This is not a consequence of the way TCP (or other network protocols) work, but it is inherent in the socket API.Assuming we’re talking about TCP/IP, it’s instructive to note that by the time
select()(orpoll()or whatever) has notified you that the listening socket is ready for reading, it’s already too late to not accept the connection: the kernel has already accepted the connection on your behalf. That is, it has already sent out aSYN|ACKin response to the incomingSYNpacket. That’s actually the effect of thelisten()system call. After that, the call toaccept()is just a formality: it allocates a new file descriptor for the accepted connection.The most notable piece of software that could potentially take advantage of a different way of doing things is TCP wrappers. TCP wrappers generally makes a decision as to whether or not to accept an incoming connection based solely on the remote IP address. If it were given the opportunity to make this decision and refuse the connection before it was even accepted, then it would enable clients whose connections have been denied by TCP wrappers to get a true “Connection refused” response (a
RSTresponse to theSYNpacket, just the same as if nothing were listening on the port). But with the sockets interface, there is no way for it to do that. Instead the client’s connection is opened and then immediately closed.Not all operating systems are the same this way. For example on Cisco IOS, when you configure an access-list of on a vty port to permit incoming connections from some IP addresses and deny them from others, the denied connections get a true “Connection refused”. But it’s not implemented using sockets. (It’s probably implemented using something very proprietary.)