23

How do I make sure that a socket bound to a port is properly released on process exit such that the port can be reused without bind() failing with EADDRINUSE? I've written a tiny program which just creates a socket, binds it to a fixed port, waits for a connection and then immediately terminates. When I rerun the program, the bind() call fails with EADDRINUSE, but if I wait a few minutes, it succeeds.

Is there a way I can explicitly "unbind" the socket, thereby freeing the port number?

Jonathan Leffler
  • 730,956
  • 141
  • 904
  • 1,278
JesperE
  • 63,317
  • 21
  • 138
  • 197

3 Answers3

32

Using SO_REUSEADDR socket option will allow you to re-start the program without delay.

int iSetOption = 1;
...
sockfd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
setsockopt(_sockfd, SOL_SOCKET, SO_REUSEADDR, (char*)&iSetOption,
        sizeof(iSetOption))
...
Ivo Bosticky
  • 6,338
  • 6
  • 34
  • 35
  • 3
    But note that technically, using SO_REUSEADDR violates the TCP/IP protocol, making it possible (though unlikely) for the next program that binds that port to pick up packets intended for the original program. – j_random_hacker Feb 14 '09 at 14:09
5

TCP/IP stack keeps port busy for sometime even after close() - socket will stay in TIME_WAIT and TIME_WAIT2 state.

If I'm not mistaken, usually it takes 2 minutes so if you need to use the same port immediately set SO_REUSEADDR option on your socket before binding, just like Ivo Bosticky suggested.

qrdl
  • 34,062
  • 14
  • 56
  • 86
4

Not exactly an answer to your question, but for completeness:

On Windows you can set the TcpTimedWaitDelay registry value to set the timeout for releasing closed TCP connections to as low as 30 seconds.

Amnon
  • 7,652
  • 2
  • 26
  • 34