[Cerowrt-devel] KASLR: Do we have to worry about other arches than x86?

dpreed at deepplum.com dpreed at deepplum.com
Thu Jan 4 17:09:19 EST 2018


I don't disagree that anyone who can run code in the hypervisor itself can attack the guest instances.
 
But that has nothing to do with KALSR or Meltdown or Sceptre. That's just bad security design - the rule is "the principle of least privilege", which comes from the 1970's work on secure operating systems.
 
I should point out here that I was one of the researchers that helped develop the original multi-level security systems then. Those "colored books" come from us.
 
-----Original Message-----
From: "Joel Wirāmu Pauling" <joel at aenertia.net>
Sent: Thursday, January 4, 2018 5:00pm
To: "dpreed at deepplum.com" <dpreed at deepplum.com>
Cc: "Jonathan Morton" <chromatix99 at gmail.com>, cerowrt-devel at lists.bufferbloat.net
Subject: Re: [Cerowrt-devel] KASLR: Do we have to worry about other arches than x86?




SRIOV ports and Vendor NIC optimizations wrt Latencies.


Whilst these heavy hitting NVF appliances tend to be large and span multiple compute hosts (and therefore are the only tenannts on those computes) - this isn't always the case. 


It's a problem in that if you can get onto the hypervisor even as an unprivileged user you can read out guest stores. .... Big Problem. 


On 5 January 2018 at 10:57, [ dpreed at deepplum.com ]( mailto:dpreed at deepplum.com ) <[ dpreed at deepplum.com ]( mailto:dpreed at deepplum.com )> wrote:

Hmm... protection datacentres tend to require lower latencies than can be achieved running on hypervisors.
 
Which doesn't mean that some datacenters don't do that.
 
As far as NFV is concerned, Meltdown only breaks security if a userspace application is running on a machine where another user has data running through kernel address space. NFV environments don't tend to run NFV in userspace under an OS that has kernel data in the page tables that are reachable from CR3.
 
The key issue in Meltdown is that CR3 is not changed between userspace and kernelspace. Which means that the memory access pipeline in userspace can use a kernelspace address (what Intel calls a "linear" address) without a check that the pagetables enable userspace access. The check happens after the speculative execution of the memory access.
 
I repeat this, because many pseudo-experts who love to be quoted in the press as saying "be afraid, be very afraid" are saying a lot of nonsense about Meltdown and Sceptre. It seems to be an echo chamber effect - the papers were released yesterday afternoon, but in a rush to get "quoted", all the wannabe-quoted people are saying things that are just plain NOT TRUE.
 
 
-----Original Message-----
From: "Joel Wirāmu Pauling" <[ joel at aenertia.net ]( mailto:joel at aenertia.net )>
Sent: Thursday, January 4, 2018 4:44pm
To: "Jonathan Morton" <[ chromatix99 at gmail.com ]( mailto:chromatix99 at gmail.com )>
Cc: [ cerowrt-devel at lists.bufferbloat.net ]( mailto:cerowrt-devel at lists.bufferbloat.net )
Subject: Re: [Cerowrt-devel] KASLR: Do we have to worry about other arches than x86?








On 5 January 2018 at 01:09, Jonathan Morton <[ chromatix99 at gmail.com ]( mailto:chromatix99 at gmail.com )> wrote:


 I don't think we need to worry about it too much in a router context.  Virtual server folks, OTOH...



  - Jonathan Morton



​Disagree - The Router is pretty much synonymous with NFV​ 
​; I run my lede instances at home on hypervisors - and this is definitely the norm in Datacentres now. We need to work through this quite carefully. ​
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/cerowrt-devel/attachments/20180104/d05fde71/attachment-0001.html>


More information about the Cerowrt-devel mailing list