FreeBSD 9.1 ZFS-Related Crash

[Ronald Klop]

On Wed, 13 Feb 2013 17:37:01 +0100, Tim Gustafson <tjg at ucsc.edu> wrote:

> Hi,
>
> I'm not sure how to report this as an official bug, or if it even
> qualifies as one.
>
> We're running ZFS on a FreeBSD 9.1 machine.  Our zpool is version 28
> and our zfs file systems are version 5.
>
> Some of our users still have individual ZFS file systems for their
> home directories.  One in particular also has a bunch of cron jobs
> that run every few minutes on different systems that access his home
> directory via NFSv3.  In order to take his home directory off-line, I
> removed him from /etc/exports and re-started mountd.  When I attempted
> to change the mount point of his ZFS file system, the server crashed.
> When the server came back up and I attempted to change the mount point
> a second time, the server crashed again.
>
> I've done this hundreds of times for other users, and have had maybe 5
> or 6 other crashes on other people's home directories as well.  So,
> it's a somewhat reproducible bug, but it's not very consistent.
>
> My best guess is that if you attempt to un-mount a ZFS file system
> when there is still some sort of NFS activity, it causes the system to
> crash.  But, most of the time it seems that if you attempt to unmount
> a file system that is still in use, the unmount command returns an
> error stating that the file system is still in use.  So there's some
> certain type of NFS activity that seems to cause the issue - whether
> it's perhaps trying to open a file right as the file system is being
> unmounted, or maybe having a file locked but not open...I just don't
> know which condition is the culprit.
>
> I have some spare hardware that I could use to attempt to reproduce
> this crash, but I wanted to ping the group first to see if anyone else
> has seen this issue.  Is this a known issue?  Has anyone else seen
> this sort of behavior?
>

What do you mean by crash? Does it hang, does it reboot, does it panic?
If it panics, did you get a dump of the kernel after the panic?

Ronald.