Subject: Re: shutdown(8) Heisenbug?
To: John Darrow <John.P.Darrow@wheaton.edu>
From: Frederick Bruckman <email@example.com>
Date: 09/29/1999 01:31:48
On Tue, 28 Sep 1999, John Darrow wrote:
> However, I've been running into a different problem. It will take forever
> to get to the 'syncing disks' stage, and then be unable to flush the last
> few buffers, thus 'giving up' and forcing a fsck of all partitions upon
> boot. How many buffers would differ between machines, but would be
> consistent from reboot to reboot on the same machine. I eventually
> tracked it down to having an mfs mounted - the mount_mfs process would
> hang in a wait state something like 'vflushbuf' (sorry, I don't remember
> exact state, and I don't have a machine I can reboot right not to check).
> If I unmounted my mfs /tmp as part of rc.shutdown (after xdm is killed,
> which uses a socket in /tmp), then the sync completes, and the reboot
> procedes quickly, leaving the disks marked clean.
I have seen that, too, but not lately (I'm running 1.4.1. now). After
a "make build" I got into the habit of doing a "shutdown now", which
unmounts the mfs, and then "sync; sync; sync; shutdown -r now".