I'd be curious to find out what conditions cause this as well. I have
not had a single orphaned /tmp/krb5 ticket leftover since I've patched
2004a with the 2004b.DEV code.

I can't imagine why one would want to kill the process using any kind of
signals on a regular basis, but like you said that this was just yet
another way to create orphaned kerberos tickets under /tmp.

We're using Redhat Enterprise 3, which ships with 2002d (real old). So I
installed the 2002d SRPM redhat provides, and rebuilt the binary RPM
using 2004a with all the appropriate patches along the way.

If you're using RH I'd be happy to provide you with my binary RPM to try
out. I've yet to rebuild the SRPM because of all the manual patching I
had to do.

-----Original Message-----
From: David Lee [mailto:[EMAIL PROTECTED] 
Sent: Wednesday, September 22, 2004 5:05 AM
To: Mark Crispin
Cc: [EMAIL PROTECTED]; Jason Sauve
Subject: RE: Possible IMAP Bug Causes IMAP with PAM_KRB5 to rapidly
deplete INODEs and DISK space

On Tue, 14 Sep 2004, Jason Sauve wrote:

> Your src/osdep/unix/env_unix.c patch in yesterday's DEV build 
> absolutely fixed the problem with the logout hook not being called.
> [...]
> -----Original Message-----
> From: [EMAIL PROTECTED] 
> [mailto:[EMAIL PROTECTED]
> On Behalf Of Mark Crispin
> Sent: Monday, September 13, 2004 5:34 PM
> To: Jason Sauve
> Cc: [EMAIL PROTECTED]
> Subject: RE: Possible IMAP Bug Causes IMAP with PAM_KRB5 to rapidly 
> deplete INODEs and DISK space
>
> Thank you for your investigation work; and in particular for your 
> information that the logout hook was not being called.
>
> I believe that this problem is remedied in today's
>       ftp://ftp.cac.washington.edu/mail/imap-2004b.DEV.tar.Z

I think I can further confirm that this fix is largely (order 90%)
effective.

History: A couple of years ago, we had massive quantities of /tmp/krb5*
files accumulating.  At the time, the code in "ckp_pam.c" had an "#if 0"
disabling code that would unlink these files at their creation; I
enabled this and it solved our problem.  (See also my email of 19 April
2004.)

Last week, I had installed vanilla imap-2004a: the "/tmp/krb5*" problem
reappeared with a vengeance, and we quickly had 20,000 (and rising) such
files.  I then extracted your "src/osdep/unix/env_unix.c" change from
"imap-2004b.DEV" and applied that to our "imap-2004a".  Things are now
vastly better.

However there is still a smaller-scale lurking problem causing some
/tmp/krb5* files to remain.  I don't know how the users precipate this
condition from email sessions, but I can reproduce something similar
(same?) "under the bonnet" within the server:

 o  A clean session (imap login/logout) is fine (the /tmp/krb5* file is
cleanly removed at session end).

 o  Sending a HUP to imapd (dies) is also fine (file cleanly removed).

 o  Sending a TERM or INT to imapd (dies) leaves the file hanging
around.

So that "imap-2004b.DEV" change has made a huge improvement, for which
many thanks.  But there remain some circumstances under which /tmp/krb5*
files can be left around (we are seeing a few thousand per day, so I
still need my tmp-flushing cron job).

Hope that helps.


-- 

:  David Lee                                I.T. Service          :
:  Systems Programmer                       Computer Centre       :
:                                           University of Durham  :
:  http://www.dur.ac.uk/t.d.lee/            South Road            :
:                                           Durham                :
:  Phone: +44 191 334 2752                  U.K.                  :

Reply via email to