[Ocfs2-tools-devel] [PATCH 0/11] Quota support for ocfs2-tools and a couple of fixes

[Jan Kara]

On Tue 01-09-09 12:26:25, tristan.ye wrote:
> On Wed, 2009-08-26 at 15:17 +0200, Jan Kara wrote:
> > Hi,
> > 
> >   here comes probably the final version of quota support for ocfs2-tools.  I've
> > heard no further comments from reviewers and fixed one or two minor bugs
> > Tristan has found during his testing. So Sunil, can you merge the patches to
> > ocfs2-tools tree? Thanks everybody for testing and useful comments.
> 
> Jan,
> 
> This v4 patches set for quota support on ocfs2-tools was awesome! all
> tests in my current quota-tools-test passed, and all open bugs get
> fixed!
> 
> One concern is something about inline-data and xattr.
> 
> 1) Assume we have inline-data and usrquota,grpquota all enabled in a
> ocfs2 volume, when disabling the inline-data feature, all inlined files
> are going to be extented to clusters, and it begins to consume space, I
> think such space consumption caused by expansion from inline to extents
> deserves our re-calculation for quota usage. I knew disabling 'sparse'
> feature by tunefs.ocfs2 was doing a great job on this part...so we do
> want disabling 'inline-data' also do the same thing.
  Ah, true, this feature needs similar quota support as when disabling
sparse feature. I'll implement that. Thanks for pointing it out.

> 2) For the xattr, I'm not quite familiar with your current strategy on
> this. we maybe want to make following things clear.
> 
>    1. Xattr entries attached to a given file also consume space on a
> target volume, so should such space consumption be taken account into
> our quota calculation?
  Yes, that's a question. Currently, it is not taken into account. Partly
because xattr support has been developed in parallel with quota support
and partly because I don't know about a way how to easily find out how much
clusters xattr tree occupies. In theory, we could add clusters occupied
by xattrs to i_clusters but I'm not sure if it won't break something.
  Other filesystems such as ext3 or reiserfs do account xattrs into quota
(because they are also accounted into i_blocks) so from a consistency point
of view, we may want to implement it once.

>    2. Disabling xattr feature removes all ea entries from host file,
> right? and such space release also should cause our recalculation on
> quota usage, right?
  Yes. Once kernel starts accounting for xattrs, userspace tools should do
it as well.

								Honza
-- 
Jan Kara <jack at suse.cz>
SUSE Labs, CR