Re: [Xen-devel] Questions about attacks on Xen

[James Harper <james.harper@xxxxxxxxxxxxxxxx>]

> 
> I have some questions about attacks on Xen. I am preparing a paper for an
> operating system we have built on top of Xen and I want to ensure we have
> certain facts straight.  Among the things I have read include "Xen and the Art
> of Virtualization" and the XOAR paper.
> 
> First, what power does Dom0 have? Of course I know that Dom0 manages
> the other domains and has direct access to hardware. I know that Dom0 can
> not directly access the Xen hypervisor code in memory (except in the case of
> attacks using DMA on IOMMU-less systems). But what about
> Dom0 accessing DomU memory once the domain is running?
> 
> For isolation, our operating system encrypts all network traffic and disk I/O.
> We have also postulated that we could do the same of keyboard/display I/O.
> We can use vTPM to ensure trusted initialization. Are there other attack
> vectors other than Dom0 handling memory destined to or from an I/O
> device? Could Dom0 violate our DomU by directly accessing its memory? Are
> there any facilities in Xen 4 for restricting this? Where could I read more
> about this?
> 
> Thank you. I appreciate any responses, especially recommended reading.
> 

Dom0 has total power over DomU. I would say that you cannot be secure if you 
run on a machine with a "hostile" dom0 that your "secure" domU does not trust. 
For a start, the 'xm save' command writes out the entire DomU memory to a disk 
file, so you can already see that Dom0 has access to all DomU memory and CPU 
state, by design.

Every time DomU does network or disk access, it must pass an entire page of 
memory to Dom0, even if only part of that page is used, so there is a constant 
potential 'leak' of information from DomU to Dom0 in terms of the parts of that 
page that belong to other processes in DomU.

James

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel