Re: [Xen-devel] [PATCH] gnttab: lock the left grant table earlier in __gnttab_unmap_common()

[Anthony Liguori <anthony@xxxxxxxxxxxxx>]

On Mon, Nov 11, 2013 at 8:23 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
>>>> On 11.11.13 at 15:52, Anthony Liguori <anthony@xxxxxxxxxxxxx> wrote:
>> I agree, it's not at all practical and extremely unlikely to happen in
>> practice.  That's why I qualified the comment with "in theory at
>> least".
>
> Extremely unlikely still means not impossible. And hence necessary
> to be fixed. Yet my question stands - can this really happen with a
> well behaved guest?

I'm not sure to be perfectly honest.  It seems like it would be
necessary to predict a future grant handle to trigger this.

> As said, I think we're fine with a guest issuing
> racing calls getting what it deserves, so long as the hypervisor or
> other guests aren't affected.

I'm less concerned about what happens to the guest than I am about
having the hypervisor be confused.

I can also see someone looking at the code and seeing this check not
guarded by a lock and drawing the conclusion that locking is not
necessary in other places where the result could be a security
concern.

Regards,

Anthony Liguori

> Jan
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel