Re: NFS over-quota not detected if utimes() called before fsync()/close()

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Sun, Jul 07, 2013 at 10:51:30PM -0400, Thor Lancelot Simon wrote:
> On Sun, Jul 07, 2013 at 07:55:12PM +0200, Manuel Bouyer wrote:
> > On Sun, Jul 07, 2013 at 10:18:29AM -0700, Erik Fair wrote:
> > > The problem of close(2) returning an "unexpected" error showed up in the 
> > > original Andrew Filesystem (AFS) papers from CMU, too - AFS syncs on 
> > > close, which can fail, but they found most applications not checking the 
> > > return code from close(2).
> > 
> > Yes, it's the same issue as NFS: write(2) can't return the error, unless
> > it really does the write (in this case the writes are synchrounous).
> 
> I don't think that's quite true.  It could do the quota accounting for the
> write, but not actually do the write.  If all the quota datastructures
> are in-core, that might be fast enough.

How could it do that, when all informations is on the remote server ?
When writing a block, the NFS client doesn't know if it's already allocated
(which which case it's a NOP for quotas) or if it's a new block.
Not to mention that writes can come from different clients ...

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--