Milan Broz wrote:
> On 05/04/2018 04:40 PM, Tetsuo Handa wrote:
> > The loop module ignores sysfs_create_group() failure and pretends that
> > LOOP_SET_FD request succeeded. I guess that the author of commit
> > ee86273062cbb310 ("loop: add some basic read-only sysfs attributes")
> > assumed that it is not a fatal error enough to abort LOOP_SET_FD request.
>
> IIRC we added sysfs attributes to easily access loop info for a regular user
> in lsblk command (and perhaps even in udev rules).
>
> The ioctl interface was still controlling the loop device, so the failure
> here was not meant to be fatal. TBH I think was not a great idea.
Thanks for reply.
>
> > Do we want to abort LOOP_SET_FD request if sysfs_create_group() failed?
>
> I would prefer failure - there are several utilities that expects attributes in
> sysfs to be valid (for example I print info from here in cryptsetup status
> if the backing image is an image), so ignoring failure put the system
> in inconsistent state.
I see. But can we for now send v1 patch for 4.17 release (and postpone making
LOOP_SET_FD request fail if sysfs_create_group() failed)? This bug has so far
crashed syzbot tests for 6432 times in 190 days.
We have a lot of bugs regarding loop module which prevent syzbot from
finding other bugs. I want to immediately squash bugs in block/loop so that
we can reduce false-positive hung task reports.
>
> Thanks for fixing this,
> Milan
>
