Re: Linux/QEMU issue

["Steven M. Christey" <coley () linus mitre org>]

On Tue, 8 Dec 2009, Jamie Strandboge wrote:

> The bug is really two parts though: the qemu issue which crashes the
> guest, and the guest kernel writing garbage to the virtio net backend.

There are three options for how to handle this:

- create separate CVEs, one for QEMU and one for virtio net
- create one CVE, concentrating on QEMU, with virtio net as an example
- create one CVE, concentrating on virtio new, mentioning the impact on
  QEMU

In the discussion in http://patchwork.kernel.org/patch/56479/, I don't see 
any clear resolution about whether it's a security problem that qemu exits 
in situations other than an ACPI shutdown.  I don't know qemu/kvm so I 
can't be sure; if qemu can support multiple hosts at the same time, and a 
crash in one host triggers a qemu crash/exit that affects all the hosts, 
then I would call this an issue in qemu.  If there's an attack angle to 
prevent forensics analysis, then that might be a reasonable argument, too.

If virtio net (whatever that is) exhibits this behavior in "normal" non-VM 
situations, then it should get a CVE itself.  But if it's closely tied to 
qemu, or to VMs in general, then I'm less certain.

VMs can stretch our older notions about threat models, and I suspect 
that's why this is difficult to handle.

- Steve


> We decided to fix it as a security update in qemu since a remote
> attacker could DoS an Ubuntu 8.04 LTS guest, possibly leading to data
> corruption within the guest. 2.6.26 and later kernels should not be
> affected.
>
> Jamie
>
> --
> Jamie Strandboge             | http://www.canonical.com