lkml.org 
[lkml]   [1996]   [Sep]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: patch-2.0.1[67] kills automount daemon (Berkeley amd)
>>>>> "Systemkennung" == Systemkennung Linux <linux@informatik.uni-koblenz.de> writes:

>> > I'm going to release a 2.0.18 reasonably soon, but in the
>> meantime I'd like > to know if this small fix fixes the NFS
>> problems.. Now that the kernel > mailing list seems to work
>> well again, maybe we can use it for what it was > meant for.
>>
>> Master,
>>
>> it seems to work. I had no more hangs with amd, but I could not
>> deterministically force a hang before, therefore "seems". But
>> I've tried hard (with run-parts /etc/cron.daily, this used to
>> hang the automounter).
Systemkennung> A very reliable way to hang amd is to use elm for
Systemkennung> reading mail on a automounted /var/spool/mail/.
Systemkennung> Start elm, then let it sit around idle for some
Systemkennung> time. Amd then tried to unmount the fs and NFS
Systemkennung> hangs.

Amazing! I have almost exactly the same problem, in a different
way. After installing amd for the first time last week, I got used to
having to reboot my machine every morning to get the NFS/amd system
unfrozen. Sometimes, I could simply restart amd, and all the processes
waiting for NFS I/O would error out and return. Until then, they would
all freeze. Only a reboot would fully clear it.

The one NFS mount that was always solidly hung was the one I get my
mail from. I automount it in a directory under /var/spool, and I
symlink my mailbox from /var/spool/mail into it. This works until
about three in the morning, when something happens to the NFS mount
(or service), and all amd/NFS freezes.

What I did to fix it was to uncomment my old /etc/fstab NFS mount of
the mail directory. Now amd doesn't ever get called, so the amd system
hasn't frozen since. The "hard" NFS mount of the mail directory hasn't
frozen either.

Maybe this has something to do with amd and/or NFS not being able to
detect that a process still has a file handle open in the automounted
directory, and they freeze during the unmount, or during the remount
when that handle is accessed again.

--
Andrew C. Esh mailto:andrew_esh@cnt.com
Computer Network Technology andrewes@mtn.org (finger for PGP key)
6500 Wedgwood Road 612.550.8000 (main)
Maple Grove MN 55311 612.550.8229 (direct)
http://www.cnt.com - CNT Inc. Home Page
http://www.mtn.org/~andrewes - ACE Home Page

\
 
 \ /
  Last update: 2005-03-22 13:38    [from the cache]
©2003-2011 Jasper Spaans