On Tue, Aug 29, 2017 at 2:31 PM, Dan Williams [off-list ref] wrote:

On Tue, Aug 29, 2017 at 2:26 PM, Christoph Hellwig [off-list ref] wrote:

quoted

Call me nitpicky, but..

First this really should be three patches, one for the DAX code, one
for the VFS code and one for XFS.  The DAX and XFS bits looks fine to
me:

Reviewed-by: Christoph Hellwig <redacted>

But I'm a little worried about stuffing more DAX knowledge into the
block mount_bdev helper.  For now it's probably ok as everything
else would involve a lot of refactoring and/or duplication, but
I'm generally not too happy about it.

I think this is why we ended up with calling dax_get_by_host() in
->iomap_begin() because the mount_bdev() touches I started with back
when this was first introduced were not very palatable. I agree with
the direction to move to mount_dax() in the future. I can respin this
into three patches and a TODO comment about how we want to kill the
dax knowledge in mount_bdev() going forward.

Actually, why not just do this directly in xfs_fs_mount()? I think I
can refactor this to not touch mount_bdev() and put all the details in
the per-fs mount/umount paths.