Re: [PATCH v2 1/2] fcntl: fix potential deadlocks for &fown_struct.lock

[Greg KH]

On Wed, Jul 07, 2021 at 11:34:17AM -0400, J. Bruce Fields wrote:
> On Wed, Jul 07, 2021 at 05:31:06PM +0200, Greg KH wrote:
> > On Wed, Jul 07, 2021 at 11:19:36AM -0400, J. Bruce Fields wrote:
> > > On Wed, Jul 07, 2021 at 05:06:45PM +0200, Greg KH wrote:
> > > > On Wed, Jul 07, 2021 at 09:51:29AM -0400, J. Bruce Fields wrote:
> > > > > On Wed, Jul 07, 2021 at 07:40:47AM -0400, Jeff Layton wrote:
> > > > > > On Wed, 2021-07-07 at 12:51 +0200, Greg KH wrote:
> > > > > > > On Wed, Jul 07, 2021 at 06:44:42AM -0400, Jeff Layton wrote:
> > > > > > > > On Wed, 2021-07-07 at 08:05 +0200, Greg KH wrote:
> > > > > > > > > On Wed, Jul 07, 2021 at 10:35:47AM +0800, Desmond Cheong Zhi Xi wrote:
> > > > > > > > > > +	WARN_ON_ONCE(irqs_disabled());
> > > > > > > > > 
> > > > > > > > > If this triggers, you just rebooted the box :(
> > > > > > > > > 
> > > > > > > > > Please never do this, either properly handle the problem and return an
> > > > > > > > > error, or do not check for this.  It is not any type of "fix" at all,
> > > > > > > > > and at most, a debugging aid while you work on the root problem.
> > > > > > > > > 
> > > > > > > > > thanks,
> > > > > > > > > 
> > > > > > > > > greg k-h
> > > > > > > > 
> > > > > > > > Wait, what? Why would testing for irqs being disabled and throwing a
> > > > > > > > WARN_ON in that case crash the box?
> > > > > > > 
> > > > > > > If panic-on-warn is enabled, which is a common setting for systems these
> > > > > > > days.
> > > > > > 
> > > > > > Ok, that makes some sense.
> > > > > 
> > > > > Wait, I don't get it.
> > > > > 
> > > > > How are we supposed to decide when to use WARN, when to use BUG, and
> > > > > when to panic?  Do we really want to treat them all as equivalent?  And
> > > > > who exactly is turning on panic-on-warn?
> > > > 
> > > > You never use WARN or BUG, unless the system is so messed up that you
> > > > can not possibly recover from the issue.
> > > 
> > > I've heard similar advice for BUG before, but this is the first I've
> > > heard it for WARN.  Do we have any guidelines for how to choose between
> > > WARN and BUG?
> > 
> > Never use either :)
> 
> I can't tell if you're kidding.

I am not.

> Is there some plan to remove them?

Over time, yes.  And any WARN that userspace can ever hit should be
removed today.

> There are definitely cases where I've been able to resolve a problem
> more quickly because I got a backtrace from a WARN.

If you want a backtrace, ask for that, recover from the error, and move
on.  Do not allow userspace to reboot a machine for no good reason as
again, panic-on-warn is a common setting that people use now.

This is what all of the syzbot work has been doing, it triggers things
that cause WARN() to be hit and so we have to fix them.

thanks,

greg k-h