Re: [PATCH v2 3/3] tpm: vtpm_proxy: Add ioctl to request locality prepended to command

[Jarkko Sakkinen]

On Wed, May 10, 2017 at 09:20:14AM -0400, Stefan Berger wrote:
> On 05/10/2017 08:47 AM, Jarkko Sakkinen wrote:
> > On Tue, May 09, 2017 at 11:49:05AM -0400, Stefan Berger wrote:
> > > On 05/08/2017 07:43 PM, Jarkko Sakkinen wrote:
> > > > On Thu, May 04, 2017 at 04:03:18PM -0400, Stefan Berger wrote:
> > > > > On 05/04/2017 02:40 PM, Jarkko Sakkinen wrote:
> > > > > > On Thu, May 04, 2017 at 07:14:27AM -0400, Stefan Berger wrote:
> > > > > > > On 05/04/2017 05:17 AM, Jarkko Sakkinen wrote:
> > > > > > > > On Wed, May 03, 2017 at 07:40:48PM -0400, Stefan Berger wrote:
> > > > > > > > > On 05/03/2017 06:37 PM, Jarkko Sakkinen wrote:
> > > > > > > > > > On Fri, Apr 28, 2017 at 09:02:18AM -0400, Stefan Berger wrote:
> > > > > > > > > > > Add an ioctl to request that the locality be prepended to every TPM
> > > > > > > > > > > command.
> > > > > > > > > > Don't really understand this change. Why locality is prenpended?
> > > > > > > > > Commands can be executed under locality 0-3 and for some commands it is
> > > > > > > > > important to know which locality a user may have chosen. How else should we
> > > > > > > > > convey that locality to the TPM emulator ?
> > > > > > > > Why this is not in the commit message?
> > > > > > > > 
> > > > > > > > More scalable way to do this would be to have a set of vtpm proxy
> > > > > > > > commands. There could be a command for requesting and releasing
> > > > > > > > locality. That would be more clean.
> > > > > > > I would think that if someone wanted to use locality it's the client using
> > > > > > > /dev/tpm(rm)0 calling an ioctl or so and the vtpm proxy then merely passing
> > > > > > > that locality to the backend (TPM emulator). I suppose the intention is to
> > > > > > > support something like that following the addition of the new functions
> > > > > > > request_locality and release_locality?
> > > > > > What if we later on want to pass something else than locality to the
> > > > > > backend? How that will work out?
> > > > > 'push' more data in front. 'pop' off by recipient. We could wrap the command
> > > > > in some form.
> > > > > 
> > > > >       Stefan
> > > > I would find having a set of special commands cleaner. Prepending sounds
> > > > like a quick hack to me, not really something that should exist in the
> > > > mainline.
> > > Along the lines of this here?
> > > 
> > > uint32_2 command
> > > uint32_2 totlength
> > > uint8_t locality
> > > uint8_t buffer[]  <- the actual TPM command
> > > 
> > > 
> > > With a command code like VTPM_PROXY_CMD_TPM_CMD = 1.
> > > 
> > >     Stefan
> > That would break binary compability.
> 
> That's why I am adding that additional flag that allows a client to choose
> whether it wants the TPM command wrapped (or locality prepended) so that it
> knows what to expect from the driver. I don't think that breaks
> compatibility.

I think having TPM command codes for control messages would be a
better idea and it is trivial to filter them so that client cannot
use those commands.

/Jarkko