Re: vfs/xfs: directio updates to ease COW handling V2

From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: Jeff Moyer <jmoyer@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>,
	linux-fsdevel@vger.kernel.org, linux-ext4@vger.kernel.org,
	ocfs2-devel@oss.oracle.com, xfs@oss.sgi.com
Subject: Re: vfs/xfs: directio updates to ease COW handling V2
Date: Wed, 3 Feb 2016 12:01:59 -0800	[thread overview]
Message-ID: <20160203200159.GK20038@birch.djwong.org> (raw)
In-Reply-To: <x49zivhy3zt.fsf@segfault.boston.devel.redhat.com>

On Wed, Feb 03, 2016 at 02:43:02PM -0500, Jeff Moyer wrote:
> Hi, Christoph,
> 
> Can you explain a bit what you mean by easing COW handling?  Whenever I
> see COW referenced near DIO, my mind always turns to g_u_p vs. fork.

Just to clarify, I'm talking about copy on write for disk blocks, not for
memory pages.

Basically, XFS implements (disk block) copy on write for (perfectly block
aligned) directio writes by allocating a set of replacement blocks, mapping the
dio writes to the new blocks, and playing a punch/remap trick to map the new
blocks into the file at the appropriate offset.

If the write fails then we don't want do the remap, so the dio_complete handler
has to know whether or not the IO succeeded, hence the new parameter.
Furthermore, if the write succeeds but the remap fails, we also want to be able
to report that to userspace, hence the change of return value from void to int.

(If the dio write isn't block aligned, we fall back to the page cache.)

Hope that helps,

--D

> 
> Thanks!
> Jeff
> 
> Christoph Hellwig <hch@lst.de> writes:
> 
> > See http://www.infradead.org/rpr.html
> >
> > The first patch ensures ->end_io is always called for direct I/O requests
> > that pass it in, even if there was a zero length write, or if an error
> > occured.  The existing users have been updated to ignore it, but XFS
> > will make use of it in the future, and a comment in ext4 suggests it
> > might be useful for it as well.
> >
> > The other two simplify the XFS direct I/O code.
> >
> > Changes since V1:
> >  - allow ->end_io to return errors
> >  - a comment spelling fix
> >
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs