Re: [Security] [PATCH] proc: avoid information leaks to non-privileged processes

[Ingo Molnar <mingo@elte.hu>]

* Matt Mackall <mpm@selenic.com> wrote:

> On Tue, May 05, 2009 at 09:52:46PM +0200, Ingo Molnar wrote:
> > 
> > * Eric W. Biederman <ebiederm@xmission.com> wrote:
> > 
> > > Ingo Molnar <mingo@elte.hu> writes:
> > > 
> > > > * Matt Mackall <mpm@selenic.com> wrote:
> > > >
> > > >> As to what's the appropriate sort of RNG for ASLR to use, finding 
> > > >> a balance between too strong and too weak is tricky. [...]
> > > >
> > > > In exec-shield i mixed 'easily accessible and fast' semi-random 
> > > > state to the get_random_int() result: xor-ed the cycle counter, the 
> > > > pid and a kernel address to it. That strengthened the result in a 
> > > > pretty practical way (without strengthening the theoretical 
> > > > randomless - each of those items are considered guessable) and does 
> > > > so without weakening the entropy of the random pool.
> > > 
> > > The trouble is, that thinking completely misses the problem, and I 
> > > expect that is why we have a problem.  Throwing a bunch of 
> > > possibly truly random values into the pot for luck is nice.  But 
> > > you didn't throw in a pseudo random number generator.  An 
> > > unpredictable sequence that is guaranteed to change from one 
> > > invocation to the next.
> > 
> > Alas, i did - it got 'reviewed' out of existence ;)
> > 
> > I still have the backups, here's the original exec-shield RNG:
> > 
> > +/*
> > + * Get a random word:
> > + */
> > +static inline unsigned int get_random_int(void)
> > +{
> > +       unsigned int val = 0;
> > +
> > +       if (!exec_shield_randomize)
> > +               return 0;
> > +
> > +#ifdef CONFIG_X86_HAS_TSC
> > +       rdtscl(val);
> > +#endif
> > +       val += current->pid + jiffies + (int)&val;
> > +
> > +       /*
> > +        * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
> > +        * every second, from the entropy pool (and thus creates a limited
> > +        * drain on it), and uses halfMD4Transform within the second. We
> > +        * also spice it with the TSC (if available), jiffies, PID and the
> > +        * stack address:
> > +        */
> > +       return secure_ip_id(val);
> > +}
> 
> Ingo, what are you on about? On every architecture but X86 with 
> TSC this is identical to the broken code.

Note that this was the exec-shield arch/*86/mm/mmap.c code.

(Also, obviously "only" covering 95% of the Linux systems has its 
use as well. Most other architectures have their own cycle counters 
as well.)

> TSC only helps matters slightly - the timescales involved in 
> process creation are very short and we can probably brute-force 
> attack it with a useful probability of success. ie:
> 
> a) record TSC
> b) fork target process
> c) record TSC
> d) guess TSC value 
> e) attempt attack
> f) repeat

Try that one day and see how much jitter there is in that sequence, 
even on a completely quiescent system.

	Ingo