L!TF Bulletin #4: The state of the horrors

L!TF Bulletin #4: The state of the horrors

[Thomas Gleixner]

[-- Attachment #1: Type: text/plain, Size: 1011 bytes --]

Hi!

The repository has been updated with the following changes since bulletin #3:

  - Online sibling threads when SMT control is switched from off to on

  - Expose the VMX mitigation state properly in the l1tf vulnerability file

  - Fix EPT off handling

  - Make the kvm L1D flush parameter runtime controllable

  - Add the 'l1tf' kernel command line option

  - Add documentation about the whole mess

The master branch is still based on 4.18-rc1 and merges almost cleanly into
4.18-rc4+ with a very trivial conflict. It's not going to be rebased
because the commit SHA1s are used in the stable branches and are also
already used in distro backports.

The stable branches linux-4.14.y, linux-4.16.y and linux-4.17.y have been
rebased to the latest stable versions. David Woodhouse said he's looking
into updating the linux-4.9.y stable branch in the next days.

Git bundle against v4.18-rc1 is attached.

Thanks everyone involved for patches, testing, review and entertaining
discussions!

Thanks,

	tglx

[-- Attachment #2: Type: application/octet-stream, Size: 81836 bytes --]

[MODERATED] Re: L!TF Bulletin #4: The state of the horrors

[Konrad Rzeszutek Wilk]

On Fri, Jul 13, 2018 at 05:08:35PM +0200, speck for Thomas Gleixner wrote:
> Hi!
> 
> The repository has been updated with the following changes since bulletin #3:
> 
>   - Online sibling threads when SMT control is switched from off to on
> 
>   - Expose the VMX mitigation state properly in the l1tf vulnerability file
> 
>   - Fix EPT off handling
> 
>   - Make the kvm L1D flush parameter runtime controllable
> 
>   - Add the 'l1tf' kernel command line option
> 
>   - Add documentation about the whole mess
> 
> The master branch is still based on 4.18-rc1 and merges almost cleanly into
> 4.18-rc4+ with a very trivial conflict. It's not going to be rebased
> because the commit SHA1s are used in the stable branches and are also
> already used in distro backports.
> 
> The stable branches linux-4.14.y, linux-4.16.y and linux-4.17.y have been
> rebased to the latest stable versions. David Woodhouse said he's looking
> into updating the linux-4.9.y stable branch in the next days.
> 
> Git bundle against v4.18-rc1 is attached.
> 
> Thanks everyone involved for patches, testing, review and entertaining
> discussions!

You may also want to backport 

git cherry-pick 84676c1f21e8
git cherry-pick adbe552349f2d

As in the stable trees. The reason being that with 'nosmt' the megasas controller
goes bonky and stops working.

> 
> Thanks,
> 
> 	tglx

Re: L!TF Bulletin #4: The state of the horrors

[Thomas Gleixner]

On Fri, 13 Jul 2018, speck for Konrad Rzeszutek Wilk wrote:
> On Fri, Jul 13, 2018 at 05:08:35PM +0200, speck for Thomas Gleixner wrote:
> > Hi!
> > 
> > The repository has been updated with the following changes since bulletin #3:
> > 
> >   - Online sibling threads when SMT control is switched from off to on
> > 
> >   - Expose the VMX mitigation state properly in the l1tf vulnerability file
> > 
> >   - Fix EPT off handling
> > 
> >   - Make the kvm L1D flush parameter runtime controllable
> > 
> >   - Add the 'l1tf' kernel command line option
> > 
> >   - Add documentation about the whole mess
> > 
> > The master branch is still based on 4.18-rc1 and merges almost cleanly into
> > 4.18-rc4+ with a very trivial conflict. It's not going to be rebased
> > because the commit SHA1s are used in the stable branches and are also
> > already used in distro backports.
> > 
> > The stable branches linux-4.14.y, linux-4.16.y and linux-4.17.y have been
> > rebased to the latest stable versions. David Woodhouse said he's looking
> > into updating the linux-4.9.y stable branch in the next days.
> > 
> > Git bundle against v4.18-rc1 is attached.
> > 
> > Thanks everyone involved for patches, testing, review and entertaining
> > discussions!
> 
> You may also want to backport 
> 
> git cherry-pick 84676c1f21e8
> git cherry-pick adbe552349f2d
> 
> As in the stable trees. The reason being that with 'nosmt' the megasas controller
> goes bonky and stops working.

I'd prefer to route them directly to stable as the stuff is already broken
with maxcpus=N. Greg?

Thanks,

	tglx

[MODERATED] Re: L!TF Bulletin #4: The state of the horrors

[Greg KH]

On Sat, Jul 14, 2018 at 02:57:00PM +0200, speck for Thomas Gleixner wrote:
> On Fri, 13 Jul 2018, speck for Konrad Rzeszutek Wilk wrote:
> > On Fri, Jul 13, 2018 at 05:08:35PM +0200, speck for Thomas Gleixner wrote:
> > > Hi!
> > > 
> > > The repository has been updated with the following changes since bulletin #3:
> > > 
> > >   - Online sibling threads when SMT control is switched from off to on
> > > 
> > >   - Expose the VMX mitigation state properly in the l1tf vulnerability file
> > > 
> > >   - Fix EPT off handling
> > > 
> > >   - Make the kvm L1D flush parameter runtime controllable
> > > 
> > >   - Add the 'l1tf' kernel command line option
> > > 
> > >   - Add documentation about the whole mess
> > > 
> > > The master branch is still based on 4.18-rc1 and merges almost cleanly into
> > > 4.18-rc4+ with a very trivial conflict. It's not going to be rebased
> > > because the commit SHA1s are used in the stable branches and are also
> > > already used in distro backports.
> > > 
> > > The stable branches linux-4.14.y, linux-4.16.y and linux-4.17.y have been
> > > rebased to the latest stable versions. David Woodhouse said he's looking
> > > into updating the linux-4.9.y stable branch in the next days.
> > > 
> > > Git bundle against v4.18-rc1 is attached.
> > > 
> > > Thanks everyone involved for patches, testing, review and entertaining
> > > discussions!
> > 
> > You may also want to backport 
> > 
> > git cherry-pick 84676c1f21e8
> > git cherry-pick adbe552349f2d
> > 
> > As in the stable trees. The reason being that with 'nosmt' the megasas controller
> > goes bonky and stops working.
> 
> I'd prefer to route them directly to stable as the stuff is already broken
> with maxcpus=N. Greg?

Good idea, I'll queue these up directly, thanks.

greg k-h

[MODERATED] Re: L!TF Bulletin #4: The state of the horrors

[Greg KH]

On Sat, Jul 14, 2018 at 09:08:58PM +0200, speck for Greg KH wrote:
> On Sat, Jul 14, 2018 at 02:57:00PM +0200, speck for Thomas Gleixner wrote:
> > On Fri, 13 Jul 2018, speck for Konrad Rzeszutek Wilk wrote:
> > > On Fri, Jul 13, 2018 at 05:08:35PM +0200, speck for Thomas Gleixner wrote:
> > > > Hi!
> > > > 
> > > > The repository has been updated with the following changes since bulletin #3:
> > > > 
> > > >   - Online sibling threads when SMT control is switched from off to on
> > > > 
> > > >   - Expose the VMX mitigation state properly in the l1tf vulnerability file
> > > > 
> > > >   - Fix EPT off handling
> > > > 
> > > >   - Make the kvm L1D flush parameter runtime controllable
> > > > 
> > > >   - Add the 'l1tf' kernel command line option
> > > > 
> > > >   - Add documentation about the whole mess
> > > > 
> > > > The master branch is still based on 4.18-rc1 and merges almost cleanly into
> > > > 4.18-rc4+ with a very trivial conflict. It's not going to be rebased
> > > > because the commit SHA1s are used in the stable branches and are also
> > > > already used in distro backports.
> > > > 
> > > > The stable branches linux-4.14.y, linux-4.16.y and linux-4.17.y have been
> > > > rebased to the latest stable versions. David Woodhouse said he's looking
> > > > into updating the linux-4.9.y stable branch in the next days.
> > > > 
> > > > Git bundle against v4.18-rc1 is attached.
> > > > 
> > > > Thanks everyone involved for patches, testing, review and entertaining
> > > > discussions!
> > > 
> > > You may also want to backport 
> > > 
> > > git cherry-pick 84676c1f21e8
> > > git cherry-pick adbe552349f2d
> > > 
> > > As in the stable trees. The reason being that with 'nosmt' the megasas controller
> > > goes bonky and stops working.
> > 
> > I'd prefer to route them directly to stable as the stuff is already broken
> > with maxcpus=N. Greg?
> 
> Good idea, I'll queue these up directly, thanks.

Ugh, that scsi driver patch is a pain, that's going to take more work...

Also, 4.9.y is going to take more work for this as well, I'll try to
remember this for when those patches show up too.

greg k-h

[MODERATED] Re: L!TF Bulletin #4: The state of the horrors

[Andi Kleen]

> Thanks everyone involved for patches, testing, review and entertaining
> discussions!

Just looking through the patches again. Thanks for implementing
the dynamic control.

+3.2. EPT not supported or disabled
+""""""""""""""""""""""""""""""""""
+
+  If EPT is not supported by the processor or disabled in the hypervisor,
+  the system is fully protected. SMT can stay enabled and L1D flushing on
+  VMENTER is not required.

AFAIK we're still missing a patch to ensure that the shadow page tables
are actually inverted if needed, or did I miss it? 

The code just copies the original currently I think, which means
the shadow is as attackable as the original.

Rest looks good to me.

-Andi

Re: L!TF Bulletin #4: The state of the horrors

[Thomas Gleixner]

On Wed, 18 Jul 2018, speck for Andi Kleen wrote:

> > Thanks everyone involved for patches, testing, review and entertaining
> > discussions!
> 
> Just looking through the patches again. Thanks for implementing
> the dynamic control.
> 
> +3.2. EPT not supported or disabled
> +""""""""""""""""""""""""""""""""""
> +
> +  If EPT is not supported by the processor or disabled in the hypervisor,
> +  the system is fully protected. SMT can stay enabled and L1D flushing on
> +  VMENTER is not required.
> 
> AFAIK we're still missing a patch to ensure that the shadow page tables
> are actually inverted if needed, or did I miss it? 
> 
> The code just copies the original currently I think, which means
> the shadow is as attackable as the original.

Paolo told me it works today. Paolo!?!?!

Thanks,

	tglx

[MODERATED] Re: ***UNCHECKED*** Re: L!TF Bulletin #4: The state of the horrors

[Alexander Graf]

[-- Attachment #1: Type: text/plain, Size: 1352 bytes --]



On 18.07.18 21:45, speck for Thomas Gleixner wrote:
> On Wed, 18 Jul 2018, speck for Andi Kleen wrote:
> 
>>> Thanks everyone involved for patches, testing, review and entertaining
>>> discussions!
>>
>> Just looking through the patches again. Thanks for implementing
>> the dynamic control.
>>
>> +3.2. EPT not supported or disabled
>> +""""""""""""""""""""""""""""""""""
>> +
>> +  If EPT is not supported by the processor or disabled in the hypervisor,
>> +  the system is fully protected. SMT can stay enabled and L1D flushing on
>> +  VMENTER is not required.
>>
>> AFAIK we're still missing a patch to ensure that the shadow page tables
>> are actually inverted if needed, or did I miss it? 
>>
>> The code just copies the original currently I think, which means
>> the shadow is as attackable as the original.
> 
> Paolo told me it works today. Paolo!?!?!

I don't see anything in the shadow paging code that would potentially
leak guest PTEs with !P into host PTEs. Every translation needs to go
through gpa_to_gva which again checks the access bits and that check
includes the P check.

So from what I can tell in the code, non-present pages will simply be 0
which again is safe.

Andi, if you found a path in the SPT code that could leak !P pages,
could you please point us to it?


Thanks!

Alex

[MODERATED] Re: L!TF Bulletin #4: The state of the horrors

[David Woodhouse]

On Fri, 2018-07-13 at 17:08 +0200, speck for Thomas Gleixner wrote:
> 
> The stable branches linux-4.14.y, linux-4.16.y and linux-4.17.y have been
> rebased to the latest stable versions. David Woodhouse said he's looking
> into updating the linux-4.9.y stable branch in the next days.

I've just pushed out a first version of the linux-4.9.y branch. It's
basically completely untested except that it builds and boots in
'qemu-system-x86_64 -kernel arch/x86/kernel/bzImage' and panics about
the lack of root filesystem without panicking about anything *else*
first.