[PATCH v23 12/24] x86/sgx: Linux Enclave Driver

[Jarkko Sakkinen]

On Fri, Nov 01, 2019 at 01:16:59PM -0400, Stephen Smalley wrote:
> On 11/1/19 11:32 AM, Sean Christopherson wrote:
> > On Fri, Nov 01, 2019 at 09:28:17AM -0400, Stephen Smalley wrote:
> > > On 11/1/19 9:16 AM, Stephen Smalley wrote:
> > > > So, IIUC, that means that merging the driver will create a regression with
> > > > respect to LSM control over executable mappings that will only be
> > > > rectified at some future point in time if/when someone submits LSM hooks
> > > > or calls to existing hooks to restore such control.  That doesn't seem
> > > > like a good idea.  Why can't you include at least that basic level of
> > > > control now?  It is one thing to defer finer grained control or
> > > > SGX-specific access controls to the future - that I can understand.  But
> > > > introducing a regression in the existing controls is not really ok.
> > > 
> > > Unless you are arguing that the existing checks on mmap/mprotect of
> > > /dev/sgx/enclave are a coarse-grained approximation (effectively requiring
> > > WX to the file or execmem for any user of SGX).
> > 
> > Yes, that's the argument as running any enclave will require RWX access to
> > /dev/sgx/enclave.  EXECMEM won't trigger for SGX users as /dev/sgx/enclave
> > must be MAP_SHARED and it's a non-private file (not backed by anonymous
> > inode, in case I got the file terminology wrong).
> 
> Ok, so for SELinux's purposes, one will need to allow :file { open ioctl map
> read write execute } to whatever type is ultimately assigned to
> /dev/sgx/enclave in order to use SGX.

AFAIK yes.

/Jarkko