|| ||Ingo Molnar <mingo-AT-kernel.org> |
|| ||Eric Northup <digitaleric-AT-google.com> |
|| ||Re: [PATCH 3/3] x86: kernel base offset ASLR |
|| ||Fri, 5 Apr 2013 09:55:32 +0200|
|| ||"H. Peter Anvin" <hpa-AT-zytor.com>,
Kees Cook <keescook-AT-chromium.org>,
Thomas Gleixner <tglx-AT-linutronix.de>,
Ingo Molnar <mingo-AT-redhat.com>,
Jarkko Sakkinen <jarkko.sakkinen-AT-intel.com>,
Matthew Garrett <mjg-AT-redhat.com>,
Matt Fleming <matt.fleming-AT-intel.com>,
Dan Rosenberg <drosenberg-AT-vsecurity.com>,
Julien Tinnes <jln-AT-google.com>, Will Drewry <wad-AT-chromium.org>,
Linus Torvalds <torvalds-AT-linux-foundation.org>|
|| ||Article, Thread
* Eric Northup <email@example.com> wrote:
> On Thu, Apr 4, 2013 at 1:58 PM, H. Peter Anvin <firstname.lastname@example.org> wrote:
> > It seems to me that you are assuming that the attacker is targeting a specific system, but a
bot might as well target 256 different systems and see what sticks...
> The alarm signal from the ones that don't stick is, in my opinion, the
> primary benefit from this work -- it makes certain classes of attack
> much less economical. A crash dump from a panic'd machine may include
> enough information to diagnose the exploited vulnerability - and once
> diagnosed and fixed, knowledge about the vulnerability is much less
Beyond making worm propagation and zombie collection dynamics much less
favorable, there's another aspect to randomization: attacks against high
value Linux targets often use high value exploits, where considerable
effort is spent to make sure that the attack will succeed 100%, without
alerting anyone - or will fail safely without alerting anyone.
Probabilistically crashing the kernel does not fit that requirement.
In some cases adding even a _single bit_ of randomness will change the
economics dramatically, because as time progresses and the kernel gets
(hopefully) more secure, the value of an exploitable zero-day
vulnerability becomes inevitably much higher than the value of pretty much
any system attacked.
Injecting a significant risk of detection is a powerful concept. Think of
WWII: how much effort went into making sure that the Germans did not
detect that the encryption of Enigma was broken. Or how much effort went
into making sure that the soviets did not detect that the US got hold of
one of their nukes - etc.
So this feature really seems useful across the security spectrum, for low
and high value systems alike.
to post comments)