Hi,
On Fri, Jul 06, 2001 at 11:50:34PM +1000, Adam Cassar
wrote:>
> I am about to implement ext3 on a large nfs shared partition that is
> using quotas (around 2-3000 users) with kernel v2.2.19. Are there caveats
> that I should be aware of? Every time someone submits
> a bug report Stephen's first question is always 'Are you using
Quotas?' :)
Mainly because quotas tend to be less tested than other parts of the
filesystem! There are people using quotas very successfully. I've
had one repeated report of quota dealock on older 2.2+ext3 versions,
but I haven't been able to reproduce it here.
There is one deadlock I know of which we have fixed in 2.4 but which
has not yet been back-ported to 2.2 --- if you create a file with a
uid that does not exist yet in the quota file, and the system needs to
allocate new disk space for the quota record, then the quota operation
to record the new used inode may deadlock with the locking in the
inode allocation code. That's almost impossible to trigger in
practice, though --- adding a new user normally happens via things
like chown(), which won't have the deadlock.
The deadlock is most likely if you manually truncate the quota files:
then normal file allocations will end up having to allocate quota
records too, with Bad Results. The motto is, don't do that (at least,
not until 0.0.7b when I'll try to apply the fixes from 2.4.)
Cheers,
Stephen