On Mon, 2017-04-24 at 08:00 -0700, Christoph Hellwig wrote:

On Mon, Apr 24, 2017 at 09:45:51AM -0400, Jeff Layton wrote:

quoted

With the patch series above, ext4 now passes. xfs and btrfs end up in
r/o mode after the test. xfs returns -EIO at that point though, and
btrfs returns -EROFS. What behavior we actually want there, I'm not
certain. We might be able to mitigate that by putting the journals on a
separate device?

This looks like XFS shut down because of a permanent write error from
dm-error.  Which seems like the expected behavior.

Oops, didn't see this message earlier...

Yeah, that's entirely reasonable when there is a write error to the
journal. The latest version of this uses $SCRATCH_LOGDEV to put the
journal on a different device, and with that I get the expected behavior
from xfs.

Thanks,
-- 
Jeff Layton [off-list ref]