Re: [PATCH v4 1/3] mm: Shuffle initial free memory

[Michal Hocko <mhocko@kernel.org>]

On Mon 15-10-18 15:25:47, Kees Cook wrote:
> On Wed, Oct 10, 2018 at 6:36 PM, Dan Williams <dan.j.williams@intel.com> wrote:
> > While SLAB_FREELIST_RANDOM reduces the predictability of some local slab
> > caches it leaves vast bulk of memory to be predictably in order
> > allocated. That ordering can be detected by a memory side-cache.
> >
> > The shuffling is done in terms of CONFIG_SHUFFLE_PAGE_ORDER sized free
> > pages where the default CONFIG_SHUFFLE_PAGE_ORDER is MAX_ORDER-1 i.e.
> > 10, 4MB this trades off randomization granularity for time spent
> > shuffling.  MAX_ORDER-1 was chosen to be minimally invasive to the page
> > allocator while still showing memory-side cache behavior improvements,
> > and the expectation that the security implications of finer granularity
> > randomization is mitigated by CONFIG_SLAB_FREELIST_RANDOM.
> 
> Perhaps it would help some of the detractors of this feature to make
> this a runtime choice? Some benchmarks show improvements, some show
> regressions. It could just be up to the admin to turn this on/off
> given their paranoia levels? (i.e. the shuffling could become a no-op
> with a given specific boot param?)

Sure, making this a opt-in is really necessary but it would be even
_better_ to actually evaluate how much security relevance it has as
well. If for nothing else then to allow an educated decision rather than
a fear driven one. And that pretty much involves evaluation on how hard
it is to bypass the randomness. If I am going to pay some overhead I
would like to know how much hardening I get in return, right? Something
completely missing in the current evaluation so far.
-- 
Michal Hocko
SUSE Labs