On Tue 29-03-22 08:39:21, Christoph Hellwig wrote:

On Mon, Mar 28, 2022 at 10:30:45AM +0200, Jan Kara wrote:

quoted

On Fri 25-03-22 17:23:31, Christoph Hellwig wrote:

quoted

On Fri, Mar 25, 2022 at 07:54:15PM +0900, Tetsuo Handa wrote:

quoted

quoted

But for now I'd really prefer to stop moving the goalpost further and
further.

Then, why not kill this code?

I think we should eventually do that, and I've indeed tested a patch
that is only cosmetically different.  I wasn't really convinced we
should do it in this series, but if there is consensus that we should
do it now I can respin the series with a patch like this included.

I'd defer it to a separate patchset. Because as much as the change to
disallow LOOP_CLR_FD ioctl for used loop device makes sense, I'm not sure
there isn't some framework using loop devices somewhere which relies on
this just getting magically translated to setting LO_AUTOCLEAR flag. So IMO
this has a big potential of userspace visible regression and as such I'd
prefer doing it separately from the bugfixes.

At least my idea would not be to disallow LOOP_CLR_FD on a used block
devices as that would go back to the udev problems before Dave turned
it into a magic LO_AUTOCLEAR.  But to remove the lo_refcnt check
entirely, as loop_clr_fd now is safe against concurrent users - it
has to anyway as there can be other users even without an open.

Ah, OK, so you'd always set LO_AUTOCLEAR and leave cleanup to happen
from lo_release()? That makes sense to me.

								Honza
-- 
Jan Kara [off-list ref]
SUSE Labs, CR