[e-lang] Side Channels (was: E-on-Common-Lisp now available)

[Eric Northup]

On Sat, 2005-05-21 at 18:32, David Wagner wrote:
> I'm unconvinced.  I'm happy to discount intentional covert channels.
> Instead, I want to focus on unintentional side channels.  Suppose Alice
> and Bob, two mutually distrusting objects, are sharing the same cache.
> If the cache uses weak pointers, then information about Alice's use
> of the cache can leak to Bob.  That might violate Alice's security, if
> Alice's access pattern was intended to be secret.  This is true even if
> the cache is non-malicious.
>
> Saying that the attack doesn't work if Bob has no access to
> non-determinism is not a useful answer in practice, because that is
> not something Alice has control over.  At a minimum, we want to give
> Alice everything she needs to protect herself; at a maximum, we want
> to minimize the number of unexpected security pitfalls associated with
> natural programming idioms (such as caches, weak pointers, etc.).

I know David Wagner is already aware of this, but I thought I'd put in a
link for the archives, since MarkM asked me to:

Colin Percival has recently (<2 weeks ago) presented an attack to steal
crypto keys by timing memory accesses to see which are in the CPU's
cache.  See
http://www.daemonology.net/hyperthreading-considered-harmful/ or the
usenet discussion on comp.arch and sci.crypt rooted in a message
"Hyper-Threading Considered Harmful" by Colin sent Fri, 13 May 2005
00:00:56 +0000 (UTC) ) for details.  In principle, these attacks do
not depend on HyperThreading, and would apply even to a single-CPU
machines, though it would be much more difficult to pull it off there.

> I don't know whether E ever made any explicit claims about this,
> but I don't think that gets us off the hook.  It is a security risk.
> The question is: Should E protect against this?  What are the guidelines
> we give to programmers to enable them to protect themselves against
> this risk?

E is not in good a position to make claims when running on information-
leaking hardware or software.

--Eric