Re: [proc/sysctl] 1dd38979b2: BUG:kernel_NULL_pointer_dereference,address

[Luis Chamberlain]

On Mon, Apr 25, 2022 at 04:57:26PM -0700, Luis Chamberlain wrote:
> On Mon, Apr 25, 2022 at 04:34:34PM -0700, Andrew Morton wrote:
> > On Mon, 25 Apr 2022 16:08:57 -0700 Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> > 
> > > On Mon, Apr 25, 2022 at 02:46:07PM -0700, Andrew Morton wrote:
> > > > On Mon, 25 Apr 2022 12:00:21 -0700 Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> > > > 
> > > > > Andrew, can we drop this patch for now?
> > > > 
> > > > I've been sitting on (ie, forgotten about) this patch
> > > > (https://lore.kernel.org/all/20200709235115.56954-1-jpitti@xxxxxxxxx/T/#u)
> > > 
> > > Jesh, yeah I see.
> > > 
> > > > for two years.  Evidently waiting for you/Kees/Ingo to provide
> > > > guidance.  So sure, the need seems very unurgent so I can drop it.
> > > 
> > > Well Keew as OK with it, but I yeah I can't decipher the issue at this
> > > point in time.
> > > 
> > > > However I fail to see how that patch could have caused this crash.  I'm
> > > > suspecting a bisection error?
> > > > 
> > > > Maybe something is unwell in drivers/vdpa/vdpa_user/vduse_dev.c.
> > > 
> > > At a quick glance, yes it could very well by vduse_init() is messy and
> > > races somehow with init, but if a race does lurk here my instincts tell
> > > me this can't be the only place.
> > > 
> > > Not sure if leaving a patch in place more time to see how else things
> > > can explode is worth it.
> > 
> > Confused.  Are you thinking that the above-linked patch was somehow
> > involved in this crash?  If so, but how?  All it does it to permit
> > unprivileged reads to four ints via proc_dointvec_minmax()?
> 
> If the priv allows for it access to dereferencing a pointer is allowed.
> How that race happens though, indeed is beyond the patch's fault.

Even so, I am 99.99% sure this is a driver bug then. And so if enabling
this patch just enables more driver bugs so be it.

So driver maintainers: please take a look, I tried to check but
the driver bug is not obvious to me yet.

  Luis