FreeBSD Security Advisory FreeBSD-SA-16:15.sysarch [REVISED]

[Pawel Jakub Dawidek]

On Wed, Oct 26, 2016 at 11:18:35AM +0300, Konstantin Belousov wrote:
> On Wed, Oct 26, 2016 at 07:53:44AM +0100, Robert N. M. Watson wrote:
> > Hi Pawel:
> > 
> > In general, my strong recommendation is against issuing advisories for local denial-of-service attacks, in part because it suggests we consider it a security guarantee of the system that those problems can be reliably prevented. At least in current operating-system designs, preventing local DoS is a very hard problem (not quite up there with covert channels, but certainly not something we can do reliably) ??? and so I think it would be misleading to suggest to our users that they can expect them not to exist at all. If something is being widely exploited, then it might be appropriate to issue an errata update, but I think if it???s something obscure where a local user to trigger a panic (and there really is no escalation path to kernel privilege, for example), then I think an advisory would generally be a mistake. Otherwise we???d find that a huge number of our ordinary kernel bug fixes get reclassified as security patches requiring advisories, if nothing else!
> > 
> > (In this case, I???m not passing judgement one way or the other ??? zeroing of arbitrary kernel memory can have more broad implications than a panic ??? for example, you can imagine that if it were to zero a process credential, a process might start running unexpectedly as root. And what were once thought to be innocuous crashes due to NULL-pointer dereferences turn out not to be!)
> 
> It is not quite arbitrary kernel memory, memory is adjanced to the
> region allocated with kmem_malloc(kernel_arena), which puts the
> allocated chunk aside from the typical kernel allocations. In fact, most
> likely the allocated chunk is followed by an unmapped page, which means
> that attempt to zero past the end of legit chunk traps. In other words,
> I consider the escalation of the issue unlikely or, at least, hard.
> 
> FWIW, I asked the same question as Pawel when initial SA was created.
> The answer was not technical but satisfactory.

I'd definiately prefer to have strict rules in this area that we always
follow. "Sometimes" publishing SAs for local DoS sends wrong message to
the users who may start to depend on them and to the people who report
those kind of bugs, who can accuse us of trying to lower number of SAs
on purpose. Note that the latter already happend in the past, AFAIR,
where we were committing fixes and later publish SAs - the time between
commit and SA publication gave a room to people to make those kind of
accusations. This is the reason we now try to commit fix and publish SA
at the same time, isn't it?

-- 
Pawel Jakub Dawidek                       http://www.wheelsystems.com
FreeBSD committer                         http://www.FreeBSD.org
Am I Evil? Yes, I Am!                     http://mobter.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-security/attachments/20161026/f1b6d4cf/attachment.sig>