Re: [PATCH 1/6] dquot: move remount handling into the filesystem

[Christoph Hellwig <hch@infradead.org>]

On Tue, May 18, 2010 at 12:34:18AM +0200, Jan Kara wrote:
>   It seems you have missed out UDF from the conversion (maybe you could
> setup a checklist of filesystem to convert? :) OTOH looking at it more in
> detail quota support for UDF is broken (i.e. quotaon returns EINVAL because
> .quota_write is not set) for several years now and noone has complained so
> I'm starting to wonder whether fixing it is worth the effort.

It's probably not worth it.  Given that quota is not part of the UDF
spec, and the linux quota formats are non-portable is's probably rather
worthless.  The same applies to UFS which has also beend disabled for
long.  Do you want me to send patches to remove UDF and UFS quota
support?

>   Also I'm slightly concerned that previous vfs_dq_quota_on_remount was
> called only after
>   sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
> so in particular MS_RDONLY has been cleared. Now it is called before so
> some filesystem could possibly barf when it sees writes from quota system
> before MS_RDONLY gets cleared. I've checked and only JFS could have this
> problem since others already clear MS_RDONLY inside their foo_remount()
> functions but still...

I've added explicit clears of MS_RDONLY to the next revision of the
patch in jfs and udf.

> >  		spin_unlock(&sbi->s_lock);
> > +
> >  		ext2_write_super(sb);
> > +		enable_quota = 1;
> >  	}
> > +
> > +	if (enable_quota)
> > +		vfs_dq_quota_on_remount(sb);
>   I kind of miss the purpose of "enable_quota" in the above...
> Also the ENOSYS check was there only for filesystems which do not support
> quotas. Since all the filesystems that call vfs_dq_off now obviously do
> support quotas, you can just drop it.

Indeed.  It was left over from an earlier version of the patch that had
locking constrainst that made it nessecary.  I've removed it for the
next version.