[RFC PATCH v3 04/12] x86/sgx: Require userspace to define enclave pages' protection bits

[Jarkko Sakkinen]

On Sun, Aug 04, 2019 at 03:20:24PM -0700, Andy Lutomirski wrote:
> On Thu, Aug 1, 2019 at 9:38 AM Jarkko Sakkinen
> <jarkko.sakkinen at linux.intel.com> wrote:
> >
> > On Mon, Jul 15, 2019 at 03:29:23PM -0700, Andy Lutomirski wrote:
> > > I would say it differently: regardless of exactly how /dev/sgx/enclave
> > > is wired up under the hood, we want a way that a process can be
> > > granted permission to usefully run enclaves without being granted
> > > permission to execute whatever bytes of code it wants.  Preferably
> > > without requiring LSMs to maintain some form of enclave signature
> > > whitelist.
> >
> > Would it be better to have a signer whitelist instead or some
> > combination? E.g. you could whiteliste either by signer or
> > enclave signature.
> >
> 
> I'm not sure, and also don't really think we need to commit to an
> answer right now.  I do think that the eventual solution should be
> more flexible than just whitelisting the signers.  In particular, it
> should be possible to make secure enclaves, open-source or otherwise,
> that are reproducibly buildable.  This more or less requires that the
> signing private key not be a secret, which means that no one would
> want to whitelist the signing key.  The enclave would be trusted, and
> would seal data, on the basis of its MRENCLAVE, and the policy, if
> any, would want to whitelist the MRENCLAVE or perhaps the whole
> SIGSTRUCT.
> 
> But my overall point is that it should be possible to have a conherent
> policy that allows any enclave whatsoever to run but that still
> respects EXECMEM and such.

So could kernel embed a fixed signing key that would be made available
through sysfs for signing? Already have one for my selftest.

/Jarkko