Re: [RFC PATCH 0/3] Allow access to confidential computing secret area

From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
To: Andi Kleen <ak@linux.intel.com>, Dov Murik <dovmurik@linux.ibm.com>
Cc: Brijesh Singh <brijesh.singh@amd.com>,
	linux-efi@vger.kernel.org,
	Tobin Feldman-Fitzthum <tobin@linux.ibm.com>,
	Tobin Feldman-Fitzthum <tobin@ibm.com>,
	Jim Cadden <jcadden@ibm.com>,
	James Bottomley <jejb@linux.ibm.com>,
	Hubertus Franke <frankeh@us.ibm.com>,
	Mike Rapoport <rppt@linux.ibm.com>,
	Laszlo Ersek <lersek@redhat.com>,
	Ashish Kalra <ashish.kalra@amd.com>,
	Tom Lendacky <thomas.lendacky@amd.com>,
	Ard Biesheuvel <ardb@kernel.org>,
	James Morris <jmorris@namei.org>,
	"Serge E. Hallyn" <serge@hallyn.com>,
	linux-security-module@vger.kernel.org,
	linux-kernel@vger.kernel.org,
	Kuppuswamy Sathyanarayanan 
	<sathyanarayanan.kuppuswamy@linux.intel.com>
Subject: Re: [RFC PATCH 0/3] Allow access to confidential computing secret area
Date: Mon, 24 May 2021 13:08:03 +0100	[thread overview]
Message-ID: <YKuXI9TUBa3sjY3e@work-vm> (raw)
In-Reply-To: <45842efd-7b6b-496f-d161-e5786760078d@linux.intel.com>

* Andi Kleen (ak@linux.intel.com) wrote:
> 
> > The SEV-SNP attestation approach is very similar to what Andi described
> > for the TDX. However, in the case of legacy SEV and ES, the attestation
> > verification is performed before the guest is booted. In this case, the
> > hyervisor puts the secret provided by the guest owner (after the
> > attestation) at a fixed location. Dov's driver is simply reading that
> > fixed location and making it available through the simple text file.
> 
> That's the same as our SVKL model.
> 
> The (not yet posted) driver is here:
> 
> https://github.com/intel/tdx/commit/62c2d9fae275d5bf50f869e8cfb71d2ca1f71363
> 

Is there any way we could merge these two so that the TDX/SVKL would
look similar to SEV/ES to userspace?  If we needed some initrd glue here
for luks it would be great if we could have one piece of glue.
[I'm not sure if the numbering/naming of the secrets, and their format
are defined in the same way]

> We opted to use ioctls, with the idea that it should be just read and
> cleared once to not let the secret lying around. Typically you would just
> use it to set up dmcrypt or similar once. I think read-and-clear with
> explicit operations is a better model than some virtual file because of the
> security properties.

Do you think the ioctl is preferable to read+ftruncate/unlink ?
And if it was an ioctl, again could we get some standardisation here -
i.e.
maybe a /dev/confguest with a CONF_COMP_GET_KEY etc ?

Dave

> -Andi
> 
> 
-- 
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK