Re: [PATCH 0/4] random,x86,kvm: Add and use MSR_KVM_GET_RNG_SEED

From: Andy Lutomirski
Date: Wed Jul 16 2014 - 10:08:29 EST


On Wed, Jul 16, 2014 at 12:36 AM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
> Il 16/07/2014 09:10, Daniel Borkmann ha scritto:
>
>> On 07/16/2014 08:41 AM, Gleb Natapov wrote:
>>>
>>> On Tue, Jul 15, 2014 at 07:48:06PM -0700, Andy Lutomirski wrote:
>>>>
>>>> virtio-rng is both too complicated and insufficient for initial rng
>>>> seeding. It's far too complicated to use for KASLR or any other
>>>> early boot random number needs. It also provides /dev/random-style
>>>> bits, which means that making guest boot wait for virtio-rng is
>>>> unacceptably slow, and doing it asynchronously means that
>>>> /dev/urandom might be predictable when userspace starts.
>>>>
>>>> This introduces a very simple synchronous mechanism to get
>>>> /dev/urandom-style bits.
>>>
>>>
>>> Why can't you use RDRAND instruction for that?
>>
>>
>> You mean using it directly? I think simply for the very same reasons
>> as in c2557a303a ...
>
>
> No, this is very different. This mechanism "provides no guarantee that the
> result contains any actual entropy". In fact, patch 3 adds a call to the
> new arch_get_slow_rng_u64 just below a call to arch_get_random_lang aka
> RDRAND. I agree with Gleb that it's simpler to just expect a relatively
> recent processor and use RDRAND.
>
> BTW, the logic for crediting entropy to RDSEED but not RDRAND escapes me.
> If you trust the processor, you could use Intel's algorithm to force
> reseeding of RDRAND. If you don't trust the processor, the same paranoia
> applies to RDRAND and RDSEED.
>
> In a guest you must trust the hypervisor anyway to use RDRAND or RDSEED,
> since the hypervisor can trap it. A malicious hypervisor is no different
> from a malicious processor.
>

This patch has nothing whatsoever to do with how much I trust the CPU
vs the hypervisor. It's for the enormous installed base of machines
without RDRAND.

hpa suggested emulating RDRAND awhile ago, but I think that'll
unusably slow -- the kernel uses RDRAND in various places where it's
expected to be fast, and not using it at all will be preferable to
causing a VM exit for every few bytes. I've been careful to only use
this in the guest in places where a few hundred to a few thousand
cycles per 64 bits of RNG seed is acceptable.

> In any case, is there a matching QEMU patch somewhere?

What QEMU change is needed? I admit I'm a bit vague on how QEMU and
KVM cooperate here, but there's no state to save and restore. I guess
that QEMU wants the ability to turn this on and off for migration.
How does that work? I couldn't spot the KVM code that allows this
type of control.

--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/