Re: [Qemu-devel] [PATCH RFC 2/2] block: Warn on insecure format probing

[Markus Armbruster]

Stefan Hajnoczi <address@hidden> writes:

> On Thu, Oct 30, 2014 at 10:07:26AM +0100, Markus Armbruster wrote:
>> Stefan Hajnoczi <address@hidden> writes:
>> 
>> > On Wed, Oct 29, 2014 at 02:54:32PM +0100, Markus Armbruster wrote:
>> >> Kevin Wolf <address@hidden> writes:
>> >> 
>> >> > Am 28.10.2014 um 17:03 hat Markus Armbruster geschrieben:
>> >> > Instead, let me try once more to sell my old proposal [1] from the
>> >> > thread you mentioned:
>> >> >
>> >> >> What if we let the raw driver know that it was probed and then it
>> >> >> enables a check that returns -EIO for any write on the first 2k if that
>> >> >> write would make the image look like a different format?
>> >> >
>> >> > Attacks the problem where it arises instead of trying to detect the
>> >> > outcome of it, and works in whatever way it is nested in the BDS graph
>> >> > and whatever way is used to address the image file.
>> >
>> > I think this is too clever.  It's another thing to debug if a guest
>> > starts hitting EIO.
>> >
>> > My opinion on probing is: it's ugly but let's leave it for QEMU 3.0 at
>> > which point we implement Markus solution with exit(1).
>> 
>> I regard my patch as a necessary preliminary step for that.  Warn now,
>> change behavior a couple of releases later.  When exactly is debatable.
>> 
>> > In the meantime the CVE has been known for a long time so vulnerable
>> > users (VM hosting, cloud, etc) have the information they need.  Many are
>> > automatically protected by libvirt.
>> 
>> The warning hopefully helps libvirt developers with keeping libvirt
>> users fully protected.
>
> I'm happy with this approach (haven't reviewed the patches in detail
> yet).

PATCH 1/2 is fully baked, but it's also trivial, and got plenty of
review already.

PATCH 2/2 isn't baked, yet, and I think I know what needs to be done.  I
guess your review cycles are better spent elsewhere.