On Wed, Mar 24, 2010 at 03:28:54PM -0600, Jonathan Corbet wrote:
quoted text
> On Wed, 24 Mar 2010 22:19:24 +0100
> Andrea Arcangeli <aarcange@redhat.com> wrote:
> 
> > > > It's a bit strange to test this when we're about to oops anyway.  The
> > > > oops will tell us the same thing.  
> > > 
> > > ...except that we've seen a fair number of null pointer dereference
> > > exploits that have told us something altogether different.  Are we
> > > *sure* we don't want to test for null pointers...?  
> > 
> > Examples? Maybe WARN_ON != oops, but VM_BUG_ON still an oops that is
> > and without serial console it would go lost too. I personally don't
> > see how it's needed.
> 
> I don't quite understand the question; are you asking for examples of
> exploits?
> 
> 	http://lwn.net/Articles/347006/
> 	http://lwn.net/Articles/360328/
> 	http://lwn.net/Articles/342330/
> 	...

As far as I can tell, VM_BUG_ON would make _zero_ differences there.

I think you mistaken a VM_BUG_ON for a:

  if (could_be_null->something) {
     WARN_ON(1);
     return -ESOMETHING;
  }

adding a VM_BUG_ON(inode->something) would _still_ be as exploitable
as the null pointer deference, because it's a DoS. It's not really a
big deal of an exploit but it _sure_ need fixing.

The whole point is that VM_BUG_ON(!something) before something->else
won't move the needle as far as your null pointer deference exploits
are concerned.

quoted text
> As to whether this particular test makes sense, I don't know.  But the
> idea that we never need to test about-to-be-dereferenced pointers for
> NULL does worry me a bit.

Being worried is good idea, as we don't want DoS bugs ;). It's just
that VM_BUG_ON isn't a solution to the problem (and the really
important thing, it's not improving its detectability either), fixing
the actual bug is the solution.
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/