[libvirt] [Qemu-devel] libvirt/QEMU/SEV interaction

[Laszlo Ersek]

On 09/29/17 23:16, Michael S. Tsirkin wrote:
> On Fri, Sep 29, 2017 at 02:48:45PM -0500, Richard Relph wrote:
>> On 9/29/17 2:34 PM, Michael S. Tsirkin wrote:
>>> On Wed, Sep 27, 2017 at 02:06:10PM -0500, Richard Relph wrote:
>>>> Whether the "BIOS" is a "static shim" as Michael suggests, or a full BIOS,
>>>> or even a BIOS+kernel+initrd is really not too significant. What is
>>>> significant is that the GO has a basis for trusting all code that is
>>>> imported in to their VM by the CP. And that NONE of the code provided by the
>>>> CP is "unknown" and unauditable by the GO. If the CP has a way to inject
>>>> code unknown to the GO in to the guest VM, the trust model is broken and
>>>> both GO and CP suffer the consequences.
>>>
>>> Absolutely.
>>>
>>>> When the CP needs to update the BIOS image, they will have to inform the GO
>>>> and allow the GO to establish trust in the CP's new BIOS image somehow.
>>>
>>> This GO update on every BIOS change is imho is not a workable model. You
>>> want something like checking the BIOS signature instead. And since
>>> hardware is all hash based, you need the shim to do it in software.
>>
>> A BIOS "signed" by the CP doesn't meet the security requirement. It is code
>> that is "unknown" to the GO.
> 
> There is a misunderstanding here.
> 
> BIOS would not be signed by a CP. It would be signed by a trusted
> software vendor e.g. by Red Hat.
> 
>> The (legitimate) CP does NOT want to be in that position of trust. If they
>> are, then some government somewhere is going to insist that they sign a BIOS
>> that allows the government to spy on the GO's VMs, and steal secrets from
>> it. Or some hacker admin will do it "for fun".
>>
>> How often do large public CPs really change their BIOSes? My sense is that
>> large public CPs prefer stability over "latest and greatest".
> 
> CPs just do dnf update. Software vendors change BIOSes.
> 
> And we do change them. Look at number of revisions for seabios in e.g.
> Fedora. More importantly we might need to change them quickly e.g.
> because of a security issue. Adding the need to coordinate with all GOs
> is not going to work. Neither can QEMU support booting old BIOS
> versions on new machine types indefinitely.
> 
>> And, perhaps more importantly, if a CP are able to sell a "more secure" VM,
>> one that justifies a higher price per vCPU hour, wouldn't that warrant some
>> changes in the "insecure" model being used today?
>>
>> Richard
> 
> Absolutely. CPs have no business signing images. But it is just not
> really feasible for software vendors to distribute hashes.

Can this be helped by "reproducible builds"? Like,

- guest firmware vendor publishes the source package,
- GO asynchronously audits the source package (most likely
incrementally, reviewing the new, broken-out patches in the source package),
- GO builds the binary package,
- GO verifies that it bit-wise matches the guest fw vendor's binary package,
- GO adds the fw binary's hash to their own whitelist.

By virtue of releasing the source package of the guest firmware (and by
announcing it via another erratum), the guest fw vendor implicitly
notifies all GOs. It is then up to the individual GOs to evaluate the
changes on their own time, and to switch to the new fw binary.

It would even be OK if the GO built their own fw binary from the source
package, and submitted that to the CP, as part of the guest payload. The
guest firmware vendor would still be able to support this build, since
it would match the vendor's own binary 100%.

Thanks
Laszlo