Re: [vfs PATCH v3 1/4] VFS: move iomap from exportfs.h to iomap.h

From: Dave Chinner <david@fromorbit.com>
To: Christoph Hellwig <hch@infradead.org>
Cc: Bob Peterson <rpeterso@redhat.com>,
	linux-fsdevel@vger.kernel.org, Jan Kara <jack@suse.cz>,
	Al Viro <viro@ZenIV.linux.org.uk>
Subject: Re: [vfs PATCH v3 1/4] VFS: move iomap from exportfs.h to iomap.h
Date: Mon, 11 Apr 2016 16:17:40 +1000	[thread overview]
Message-ID: <20160411061740.GH567@dastard> (raw)
In-Reply-To: <20160410171528.GA9118@infradead.org>

On Sun, Apr 10, 2016 at 10:15:28AM -0700, Christoph Hellwig wrote:
> On Wed, Mar 30, 2016 at 09:20:36AM +1100, Dave Chinner wrote:
> > QA over the past couple of days has indicatedno regressions, so I'm
> > going to seriously consider the multipage write code for the next
> > XFS merge. I'd be happy to also take fiemap code based on the same
> > iomap interfaces, especially if we can make it a generic, fully
> > functional fiemap implementation and use it in XFS, too.
> 
> Now that's I've spent some more time with it: it depend on your
> idea of fully functional.  For the actual file data the fiemap
> implementation is simple and beautiful.  But XFS also supports the
> odd FIEMAP_FLAG_XATTR flags, which makes no sense at all for an
> fiemap based implementation.

In this case, I'd let xfs_vn_fiemap handle that. i.e. if the
FIEMAP_FLAG_XATTR is set, run the old code, otherwise call straight
into the new generic iomap based code.

I'm not sure if how your new code is structured, but n this version
__generic_iomap_fiemap() is passed an iomap_get_t callback function.
We could simply have XFS pass a different callback function that
looks up the attribute fork extent list rather than the data fork
extent list to support FIEMAP_FLAG_XATTR appropriately because other
than the different initial extent list root it seems to me like the
implementation should be identical....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com