Re: [PATCH] tpm: Detach page allocation from tpm_buf

[Jarkko Sakkinen]

On Thu, Oct 03, 2019 at 08:39:10AM -0400, Mimi Zohar wrote:
> On Thu, 2019-10-03 at 14:32 +0300, Jarkko Sakkinen wrote:
> > On Wed, Oct 02, 2019 at 08:40:24AM -0400, Mimi Zohar wrote:
> > > On Thu, 2019-09-26 at 16:12 +0300, Jarkko Sakkinen wrote:
> > > > On Thu, Sep 26, 2019 at 03:46:35PM +0300, Jarkko Sakkinen wrote:
> > > > > On Wed, Sep 25, 2019 at 04:48:41PM +0300, Jarkko Sakkinen wrote:
> > > > > > -		tpm_buf_reset(&buf, TPM2_ST_NO_SESSIONS, TPM2_CC_GET_RANDOM);
> > > > > > +		tpm_buf_reset(&buf, data_ptr, PAGE_SIZE,
> > > > > > +			      TPM2_ST_NO_SESSIONS, TPM2_CC_PCR_EXTEND);
> > > > > 
> > > > > Oops.
> > > > 
> > > > Maybe we could use random as the probe for TPM version since we anyway
> > > > send a TPM command as a probe for TPM version:
> > > > 
> > > > 1. Try TPM2 get random.
> > > > 2. If fail, try TPM1 get random.
> > > > 3. Output random number to klog.
> > > > 
> > > > Something like 8 bytes would be sufficient. This would make sure that
> > > > no new change breaks tpm_get_random() and also this would give some
> > > > feedback that TPM is at least somewhat working.
> > > 
> > > That involves sending 2 TPM commands.  At what point does this occur?
> > >  On registration?  Whenever getting a random number?  Is the result
> > > cached in chip->flags?
> > 
> > On registeration. It is just printed to klog.
> 
> What sets "TPM_CHIP_FLAG_TPM2" in chip->flags?  And when?
> 
> > 
> > > Will this delay the TPM initialization, causing IMA to go into "TPM
> > > bypass mode"?
> > 
> > Of course it will delay the init.
> 
> Delaying the init will most likely cause regressions on systems with
> TPM 1.2 systems.
> 
> Instead of sending the TPM 2.0 command and on failure sending the TPM
> 1.2 version of the command, could chip->flags be tested?  And if not
> chip->flags, then provide the TPM version as part of registration.

No rush pushing this forward. I got your point.

> > As I've stated before the real fix for the bypass issue would be
> > to make TPM as part of the core but this has not received much
> > appeal. I think I've sent patch for this once.
> 
> I must have missed this discussion.

Yeah, I think that'd be a great idea. We need a better control on
TPM core as multiple subsystem's depend on it in API level. Something
to reconsider in future.

/Jarkko