Subject: Re: Quota, Softdeps, and Raid - and the Quality of NetBSD Releases
To: None <netbsd-users@NetBSD.org>
From: Lasse Kliemann <lasse-list-netbsd-users-2006@plastictree.net>
List: netbsd-users
Date: 09/14/2006 12:58:42
* I wrote:::
> * Manuel Bouyer:::
> > On Sat, Sep 09, 2006 at 01:29:03PM +0200, Lasse Kliemann wrote:
> > > 
> > > as some of you might have noticed already, I ran into serious 
> > > trouble using all these three together on one of my machines. 
> > > The issue is known as kern/33642. So far, this problem seems 
> > > to have received little attention. As I already lined out in 
> > > an additional post to netbsd-bugs last week, this problem has 
> > > not gone away since RC1 for 3.1. It is still there, and it can 
> > > lead to serious damage of the system (loss of /etc/passwd). I 
> > > write to this high-traffic list now in the hope to receive 
> > > answers to the following questions:
> > > 
> > > - Is anyone using quotas, softdeps, and raid at the same time 
> > >   on some machine, and what are the experiences?
> > 
> > I have quota+raid without problems, and I can't see how raid could affects
> > this. Also, I have the quota files in /var/quota and not on the
> > quota-enbabled filesystems, maybe this is what causes troubles
> > 
> > I would try:
> > turning off softdep on /var
> > putting the quota files in /var
> > 
> > also make sure nothing writes to the filesystems with quota enabled while
> > quotacheck is running. I once tried enabling the NFS server before
> > quotacheck and got a similar panic.
> 
> Thanks a lot for your advice! I have set up my system 
> accordingly and will now see what happens during the next 
> days. I will report my findings here.

I now received during a lot of disk activity:

uvm_fault(0xcee9ca84, 0, 0, 2) -> 0xe
kernel: page fault trap, code=0
Stopped in pid 23164.1 (tar) at netbsd:dqget+0x396: movl %edx,0(%eax)

I recall getting similar messages even when quotas were turned 
off (but softdeps on). I do not recall, however, whether there 
was a reference to 'dqget' in those messages, as it is here 
now and suggests (to me) a connection to quotas.

Any ideas?

Lasse