You, Philip Hazel write:
>
> Running "netstat" on machine A shows a connection from port 25 on that
> machine to some other port on machine B. However, there are no longer
> any mail processes running on machine A, and "netstat" on machine B does
> not show this connection. The state of the connection is given as
> "CLOSE_WAIT", and it sits there for days and days. While it is present,
> attempts to restart a mail listener on machine A fail, saying "port
> already in use". Machine A is running Solaris 2.4/101945-37 and by
> coincidence, one of several machine Bs involved is running the same OS,
> though I don't believe this is relevant.
Does this seem to cover it?
Distribution: inet
From: Wyllys Ingersoll <wyllys@???>
Date: 24 Jun 1996 17:42:02 -0400
Message-ID: <rxohm8q3pw.fsf@???>
Organization: ANS, Reston VA Lines: 34
> : > machine.port daria.cdnow.com.32955 8760 0 8276 0 CLOSE_WAIT
> : >
> : > where, port is NOT being using by any program yet solaris 2.4 wont
> : > let any program start to access it because it says its in use...
> : > which it isnt because ALL processes that did use it died, termed or
> : > crashed.
> : However there is a rare race condition in Solaris 2.4 which is fixed
> : in the following bugfix:
> : > : 1223853 TCP stream may not go away when process dies
> : > : It is fixed in patches 101945-39 or higher for Solaris 2.4
> : > : From your description it seems this is probably what you are runining into.
> : > All this is because of a test web server that we use in-house here.
> We are HUP'ing it all the time when we make some changes to its configurations.
> Problem is. Sometimes it dies, but the sockets are still open. All the
> children are dead and gone. Unless.... it doesn't show up in the process
> table.. which I here is possible... so how does one "see" things that ps
> doesn't show?
> I have seen this problem ALOT under Solaris 2.3 when a very busy web
> proxy gets killed, lots of sockets get left open various states
> that cannot be cleaned up (ESTABLISHED, FIN_WAIT_1, CLOSE_WAIT),
> sometimes with data in the queues, sometimes not.
--
Artificial Intelligence stands no chance against Natural Stupidity.
GAT d- -p+(--) c++++ l++ u++ t- m--- W--- !v
b+++ e* s-/+ n-(?) h++ f+g+ w+++ y*
