Release of 2.0.7 due soon (quotas)

[David Lee]

On Sat, 8 Apr 2000, Jeremy Allison wrote:

> Norbert Püschel wrote:
> > 
> > 
> > There is a long standing bug in Solaris quota handling: If the soft
> > quota is zero,
> > you need to check the hard quota instead. This bug may affect other
> > systems as well.
> 
> Thanks for that fix, but I need to confirm it before including
> in the 2.0.7 source.
> 
> Can someone with access to Solaris confirm that the soft
> quota being zero means check hard quota ?

To confirm:

Background:  Our systems have many users per partition, user has order of
several *mega*bytes, paritions of order several *giga*bytes.  So quotas
are very important to us, and were one of the very first things we checked
when we introduced Samba last year.  What our users need to see in the
drive's "Properties" is the space available to them (quota) not the
df-like free space on the UNIX parition (shared with thousands of others).

The current Solaris behaviour attempts to map the soft limit to the PC
"capacity" field (seen on the drive's "Properties").  If the user's
quantity of data exceeds the soft limit, then the capacity gets adjusted
(see "rubber latex" comments in source code) to this figure, so the free
space stays at 0.  My own view is that this is all good behaviour.  (If it
had been bad for us, you would have seen a patch from me many many months
ago!) 

Now, suppose that the soft limit is unset.  The quota processing fails
and the PC's "Properties" view shows disk partition space (order of
several *giga*bytes).  My own view is that this is bad (Norbert, indeed,
classifies it as a bug, which has my sympathy!).

I then applied Norbert's patch.  Now if the soft limit is zero, the PC
Properties show reasonable figures: very similar to the "soft"-based
figures, just slightly higher, being the "hard"-based figures.  This
is (in my view) much better than the old behaviour of falling back to the
order(gigabyte) figures.

Count this as a vote for Norbert's patch.

Also:

o  The new Veritas quotas code is very similar.  Indeed, it was originally
   cloned from the Solaris code.  This functionality would be highly
   desirable there, too.

o  Various other systems may benefit from similar attention: skim-reading
   the source suggests OSF1, FreeBDS/OpenBSD

o  Assuming the LINUX code is correct, the IRIX code seems similar but
   I think may have a bug.

I don't have any of those non-Solaris versions, so cannot test.  I have
the Veritas version.  It is in major user-service, so I cannot do many
tests, but we could do one or perhaps two.  Any thoughts, Jeremy?


-- 

:  David Lee                                I.T. Service          :
:  Systems Programmer                       Computer Centre       :
:                                           University of Durham  :
:  http://www.dur.ac.uk/~dcl0tdl            South Road            :
:                                           Durham                :
:  Phone: +44 191 374 2882                  U.K.                  :