RE: [tpmdd-devel] [PATCH 4/4] keys, trusted: seal/unseal with TPM 2.0 chips

["Fuchs, Andreas" <andreas.fuchs@sit.fraunhofer.de>]

> > OK, I guess we got stuck in the follow-up discussions and missed the points.
> 
> Yup, don't get me wrong here. I like this discussion and am willing to
> listen to reasonable arguments.

We could not agree more. I'm always up for a good discussion... ;-)

> > My 1st point is:
> >
> > TPM1.2's 0x40000000 SRK handle was a well-known, singleton, always-present
> > key, that could be relied upon.
> >
> > TPM2.0's 0x80000000 is a temporary, TPM-assigned, context-specific handle,
> > that cannot be relied upon.
> >
> > Therefore, I think your patch should not use it.
> >
> > Instead, I'd recommend using the closest equivalent to an SRK that TPM2.0
> > has to offer, which is within the range 0x81000000 to 0x8100FFFF.
> > (see http://www.trustedcomputinggroup.org/resources/registry_of_reserved_tpm_20_handles_and_localities)
> > You might want to use TPM2_GetCapability() to find the correct one.
> >
> > Also User-Space could reference any of these handles in the
> > 0x81000000-0x81FFFFFF range. This would be fine.
> 
> Alright. How about requiring keyhandle as explicit option for TPM 2.0?
> Would that be a more reasonable solution in your opinion? That would
> acceptable for me.

You mean getting rid of the default behaviour ?
That sound reasonable to me as well. A later patch could add the possibility
to use the TPM2_GetCapability() thing at a later stage then...

> > My 2nd point is:
> >
> > It is IMHO a bad idea to allow user-space to provide transient handles as
> > parameter to the TPM, because TSS2.0 will virtualize handles and /dev/tpm0
> > is single-access only.
> > Instead I'd recommend passing context-saved-blobs to the kernel.
> >
> > Then you brought up the valid point that this requires kernel-space resource
> > broker and I provided some sketch-idea in pseudo-code for discussion of
> > general approach. I did not know that the access broker was solved already.
> 
> Yeah. I'm not against implementing the broker and how I've been thinking
> implementing it is not too far away what you just suggested.
> 
> I'm not just seeing why that couldn't be done as a separate patch set
> later on.

I should have been more clear then. I agree that it can be added later on.
Or rather I think it should be added at some later point...

I was just trying to point out that the concept is not too difficult, since
kernel-space (minimal) resource-manager makes a lot of people go "oh god,
never ever, way too big, way too complicated", which IMHO it is not.
Until then, I think that the call should just return -EBUSY when you cannot
load the sealed data if no slots are available ?

I looked at Patch 3/4 and it seems you default to -EPERM on TPM2_Create()-
and TPM2_Load()-failures ?
You might want to test against rc == TPM_RC_OBJECT_MEMORY and return -EBUSY
in those cases. Would you agree ?
(P.S. I can cross-post there if that's prefered ?)

Cheers,
Andreas