Re: [PATCH] x86/sgx: Remove checks for platform limits from sgx_validate_secs()

[Sean Christopherson <sean.j.christopherson@xxxxxxxxx>]

On Mon, Oct 05, 2020 at 02:38:40PM +0300, Jarkko Sakkinen wrote:
> On Sun, Oct 04, 2020 at 11:00:28PM -0500, Haitao Huang wrote:
> > On Sun, 04 Oct 2020 21:08:19 -0500, Jarkko Sakkinen
> > <jarkko.sakkinen@xxxxxxxxxxxxxxx> wrote:
> > 
> > > Remove from sgx_validate_secs():
> > > 
> > > 	if (secs->miscselect & sgx_misc_reserved_mask ||
> > > 	    secs->attributes & sgx_attributes_reserved_mask ||
> > > 	    secs->xfrm & sgx_xfrm_reserved_mask)
> > > 		return -EINVAL;
> > > 
> > > SECS can surpass the platform limits because it's the SIGSTRUCT that
> > > defines the limits that are used at run-time.
> > > 
> > > What SECS does is that it defines the overall limits that must apply for
> > > any platform, i.e.  SECS limits and platform limits are orthogonal. They
> > > are not dependent.
> > > 
> > > Cc: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> > > Cc: Jethro Beekman <jethro@xxxxxxxxxxxx>
> > > Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> > > Suggested-by: Haitao Huang <haitao.huang@xxxxxxxxxxxxxxx>
> > > Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@xxxxxxxxxxxxxxx>
> > > ---
> > >  arch/x86/kernel/cpu/sgx/ioctl.c | 5 -----
> > >  1 file changed, 5 deletions(-)
> > > 
> > > diff --git a/arch/x86/kernel/cpu/sgx/ioctl.c
> > > b/arch/x86/kernel/cpu/sgx/ioctl.c
> > > index 6b3cc8483008..008752cb54f0 100644
> > > --- a/arch/x86/kernel/cpu/sgx/ioctl.c
> > > +++ b/arch/x86/kernel/cpu/sgx/ioctl.c
> > > @@ -86,11 +86,6 @@ static int sgx_validate_secs(const struct sgx_secs
> > > *secs)
> > >  	if (secs->base & (secs->size - 1))
> > >  		return -EINVAL;
> > > -	if (secs->miscselect & sgx_misc_reserved_mask ||
> > > -	    secs->attributes & sgx_attributes_reserved_mask ||
> > > -	    secs->xfrm & sgx_xfrm_reserved_mask)
> > > -		return -EINVAL;
> > > -
> > >  	if (secs->size > max_size)
> > >  		return -EINVAL;
> > > 
> > 
> > Looks good to me.
> > Thanks
> > Haitao
> 
> Thanks, I'll merge it then.

You can't remove this wholesale, the kernel should still disallow access to
features that are unknown to the kernel and/or are explicitly disallowed by
the kernel.  E.g. see SGX_ATTR_RESERVED_MASK and SGX_MISC_RESERVED_MASK.

Dropping sgx_xfrm_reserved_mask is ok because the CPU explicitly checks that
XFRM is a strict subset of the current XCR0, though that makes me wonder what
it does with XSS...