Re: [PATCHSET v3.1 0/7] data integrity: Stabilize pages duringwriteback for various fses

From: Jan Kara
Date: Mon May 16 2011 - 15:00:22 EST


On Mon 16-05-11 11:49:27, Darrick J. Wong wrote:
> On Thu, May 12, 2011 at 11:42:55AM +0200, Jan Kara wrote:
> > On Wed 11-05-11 11:19:01, Darrick J. Wong wrote:
> > > On Tue, May 10, 2011 at 02:51:24PM +0200, Jan Kara wrote:
> > > > On Mon 09-05-11 16:03:18, Darrick J. Wong wrote:
> > > > > I am still chasing down what exactly is broken in ext3. data=writeback mode
> > > > > passes with no failures. data=ordered, however, does not pass; my current
> > > > > suspicion is that jbd is calling submit_bh on data buffers but doesn't call
> > > > > page_mkclean to kick the userspace programs off the page before writing it.
> > > > Yes, ext3 in data=ordered mode writes pages from
> > > > journal_commit_transaction() via submit_bh() without clearing page dirty
> > > > bits thus page_mkclean() is not called for these pages. Frankly, do you
> > > > really want to bother with adding support for ext2 and ext3? People can use
> > > > ext4 as a fs driver when they want to start using blk-integrity support.
> > > > Especially ext2 patch looks really painful and just from a quick look I can
> > > > see code e.g. in fs/ext2/namei.c which isn't handled by your patch yet.
> > >
> > > Yeah, I agree that ext2 is ugly and ext3/jbd might be more painful. Are there
> > > any other code that wants stable pages that's already running with ext3? In
> > > this months-long discussion I've heard that encryption and raid also like
> > > stable pages during writes. Have those users been broken this whole time, or
> > > have they been stabilizing pages themselves?
> > I believe part of them has been broken (e.g. raid) and part of them do
> > copy-out so they were OK.
>
> A future step might be to undo all these homegrown copy-outs?
Sure but I'm not the right one to tell you where these are ;).

> > > I suppose we can cross the "ext3 fails horribly on DIF" bridge when someone
> > > complains about it. Possibly we could try to steer them to btrfs.
> > Well, btrfs might be a bit too advantageous for production servers but
> > ext4 would be definitely viable for them.
>
> Are there any distros that are going straight from ext3 to btrfs?
Most distros currently offer users a choice of xfs, ext3, ext4, btrfs
with ext4 being the default. I'm not sure if that's what you are asking
about...

Honza
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/