Re: [Ext2-devel] [RFC] [PATCH 3/4] Add extended attributes to ext2/3

[Andreas Dilger <adilger@clusterfs.com>]

On Oct 08, 2002  21:47 +0100, Christoph Hellwig wrote:
> On Tue, Oct 08, 2002 at 09:41:43PM +0100, Stephen C. Tweedie wrote:
> > On Tue, Oct 08, 2002 at 08:20:38PM +0100, Christoph Hellwig wrote:
> > > On Tue, Oct 08, 2002 at 09:14:00PM +0200, Andreas Gruenbacher wrote:
> > > > Users might just fill up all xattr space leaving no space for ACLs (or 
> > > > similar). If user xattrs are disabled this can no longer occur, so some 
> > > > administrators might be happy to have a choice.
> > > 
> > > Umm, that's why we have quota..
> > 
> > It's the per-inode extended attribute space that's at risk here,
> > quotas don't help.
> 
> Well, that's a more important problem.  But I doubt a hack to just turn off
> user xattrs is the right fix then.  A static reservation for ACLs or just
> totally separating them (like in XFS) seems more m?ture.

Yes, we have made proposals in the past to change the way ext2/3 EAs are
stored on disk, but nobody has had time to work on it yet.  The current
limitations of 4kB of EA per inode (user+system) is bad, as is the fact
that you need a full block for each inode if you have any unique
per-inode data (basically anything other than ACLs).

That said, the benefits of the current EA implementation far outweigh the 
limitations, and since this is an internal-to-ext2/3 issue, we can
change it at some later date without affecting the rest of the kernel.

Cheers, Andreas
--
Andreas Dilger
http://www-mddsp.enel.ucalgary.ca/People/adilger/
http://sourceforge.net/projects/ext2resize/