Re: [PATCH v3 1/3] tpm: vtpm_proxy: Implement new ioctl to get supported flags

From: Jason Gunthorpe
Date: Thu May 04 2017 - 13:20:55 EST

On Thu, May 04, 2017 at 01:13:18PM -0400, Stefan Berger wrote:
> On 05/04/2017 11:34 AM, Jason Gunthorpe wrote:
> >On Thu, May 04, 2017 at 10:56:25AM -0400, Stefan Berger wrote:
> >>Implement VTPM_PROXY_IOC_GET_SUPT_FLAGS ioctl to get the bitmask
> >>of flags that the vtpm_proxy driver supports in the
> >>VTPM_PROXY_IOC_NEW_DEV ioctl. This helps user space in deciding
> >>which flags to set in that ioctl.
> >you might be better off just having a VTPM_PROXY_IO_ENABLE_FEATURE
> >.feature = LOCALITY
> Do you have an example driver that shows how to do this ? Can user space
> query that feature?

Try and enable the feature, if it fails then there is no feature in
the kernel.

This is the usual way to add new syscalls..

> >If that fails then the feature is not supported, no real need for the
> >query in that case.
> >
> >Not sure about Jarkko's point on request/release locality.. Is there a
> >scenario where the emulator should fail the request locality?
> We could filter localities 5 and higher on the level of the driver (patch
> 2/3) since basically there are only 5 localities (0-4) in any TPM interface
> today. The typical hardware locality 4 would be filtered by the emulator per
> policy passed via command line, but I would allow it on the level of this
> driver. An error message would be returned for any command executed in that
> locality, unless the 'policy' allows it. Localities 0-3 should just be
> selectable. The TPM TIS (in the hardware) implements some complicated scheme
> when it comes to allowing the selection of a locality and I would say we
> need none of that but just tell the vTPM proxy driver the locality (patch
> 2/3) in which the next command will be executed.

Well, if TIS hardware has some scheme I feel like the emulator uAPI should
have enough fidelity to ecompass existing hardware, even if your
current emulator does not need it.

So allowing request_locality to fail from userspace seems reasonable.