[Cryptography] Determining TLS session keys from the hypervisor

Jerry Leichter leichter at lrw.com
Mon Jun 13 19:23:29 EDT 2016


> Defenses are not clear.
> 
> Actually I would have thought the defence was pretty trivial: Don't run your
> sensitive crypto on hardware controlled by an attacker.  On the one hand the
> work they've done is pretty neat, but the overall result is a "well, duh".
Well, sure ... but the fact is that a large and probably growing percentage of Web servers *are* running "in the Cloud" - i.e., on hardware controlled by someone else.  The economics are likely to push ever more stuff "out there".

Now, you can argue that Amazon or Microsoft or Google - or the smaller providers of virtual hosting - are not going to attack you; and *in general*, that's probably true.  But they may well be forced to by government order - without telling you.  And when it comes to the smaller providers - just how much should you trust them?

It's also worth keeping in mind that bugs allowing attackers to escape from their guest OS instances and gain access to the hypervisor have occurred.  So even if you trust your provider, you do have *some* level of exposure to your "running mates" on the host you share.

As will all (properly analyzed) security issues, it's about costs and risks.  By moving out to a Cloud instance, you're generally saving money and you're mitigating many practical risks - the big data centers are much less vulnerable to power outages, fires, and all kinds of similar events than you would likely be able to afford.  This paper shows that you're not quite as secure within your VM as you might think.  You now have to go make the tradeoffs for yourself.

                                                        -- Jerry



More information about the cryptography mailing list