linux-kernel - Re: Spectre mitigation doesn't seem to work at all?!

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20180604091931.sziimdeei54jgwzn@salmiak>
Date:   Mon, 4 Jun 2018 10:19:31 +0100
From:   Mark Rutland <mark.rutland@....com>
To:     Andreas Hartmann <andihartmann@...enet.de>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: Spectre mitigation doesn't seem to work at all?!

On Mon, Jun 04, 2018 at 10:50:07AM +0200, Andreas Hartmann wrote:
> Hello Peter,
> 
> thanks for your answer! I appreciate it!
> 
> On 06/04/2018 at 10:15 AM Peter Zijlstra wrote:
> > On Fri, Jun 01, 2018 at 02:19:38PM +0200, Andreas Hartmann wrote:
> > 
> >> I tested the spectre mitigation of different machines and kernels with
> >> https://github.com/crozone/SpectrePoC
> >>
> >> You can see the results below.
> > 
> >> My question: Did I miss something?
> > 
> > Yes.
> > 
> >> Build: ... INTEL_MITIGATION_DISABLED LINUX_KERNEL_MITIGATION_DISABLED
> >> Build: ... INTEL_MITIGATION_DISABLED LINUX_KERNEL_MITIGATION_DISABLED
> >> Build: ... INTEL_MITIGATION_DISABLED LINUX_KERNEL_MITIGATION_DISABLED
> > 
> >                               ^^^^^^^^                         ^^^^^^^^
> > 
> > The POC is a v1 on itself. V1 needs to be fixed for every individual
> > executable (worse, for every individual location in the code, and we're
> > still finding them). The kernel mitigation status for v1 only indicates
> > the kernel itself has mitigations (for some locations).
> > 
> > The POC is meant to test effectiveness of these mitigations, either the
> > original LFENCE or the dependent instruction thing, but you have to
> > enable one or the other.
> 
> Ok, this means every program running on the machine has to care itself
> to be spectre v1 - safe.

Correct. Primiarily this matters for things like JITs, where untrusted code may
be run in the same address space as sensitive data.

> A malicious program most probably won't care about that. Therefore, my
> next question is: which memory regions can be exploited by a malicious
> program? The complete physical memory or only the memory provided to the
> malicious program? Should be the latter if this approach should have any
> impact.

Assuming you have a CPU which is not vulnerable to meltdown / variant-3, or you
have mitigated this, (e.g. with KPTI), a malicious program can only access data
within its own address space.

Spectre variant-1 alone only gives access to memory in the address space of the
program itself.

Thanks,
Mark.