SUMMARY: How to clean up a CLOSE_WAIT socket

From: Pedro Jose Serrahima Bartolome <serrahim_at_lulu.iic.uam.es>
Date: Thu, 06 Feb 1997 09:54:41 -0600

Thanks to all osf managers who replied to my question,
specially to Charles M. Richmond <cmr_at_iisc.com>, who
has dedicated to my problem more time than I could expect.

These are the persons who replied:

Charles M. Richmond <cmr_at_iisc.com>
John Hascall <john_at_iastate.edu>
Arnaud Valeix <fnet_at_ifh.sncf.fr>
Jeffrey G. Micono 6533 (Ktech) <jgmicon_at_sandia.gov>
Peter Hudson <peter_at_osm.co.uk>

My question was:

------
I'm programming a client/server application that sometimes crash
abnormally, leaving one socket in CLOSE_WAIT state, as indicated
by netstat command.

My problem is not why the application crashed, but the fact that
I don't know how to clean up this socket, and then, I cannot restart
the program because the bind C function fails due to the error :
 'address alrady in use' caused by the presence of this CLOSE_WAIT
state socket.

Up to this moment, the only way I've found to solve this is restarting
the computer, but, of course, this is not a good solution.

the O.S is Dunix 4.0 and 3.2.

----
Not considering the the cause of my program crash, that may be 
hard to find as Charles knows, John Hascall and Peter Hudson provided me
some directions to avoid the problem caused to me when the program is
crashed, allowing me to restart the application again without receiving
the 'address already in use' error. 
These are their replies: 
>From John:
--------------------------
Add a call to set the SO_REUSEADDR flag on the socket -- something
like this:
      int on = 1;
      s = socket( ... );
      setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (char *)&on, sizeof(on));
          :
      bind(s, ... );
------------------------
>From Peter:
-------------------------
There is only one way to do this: you must close() the socket.
I am in the same situation, and I had to install a signal handler for
SIGTERM, SIGBUS, SIGSEGV, and so on... which closed the socket
gracefully.
Alternatively, in the code which opens/connects the socket, you could
have a retry loop checking errno == EADDRINUSE.  A socket in CLOSE_WAIT
should timeout after about 2 minutes.
----------------------
Pedro Serrahima
IIC Madrid
Spain

Received on Thu Feb 06 1997 - 10:15:37 NZDT