Re: [tpmdd-devel] [PATCH RFC v3 5/5] tpm2: expose resource manager via a device link /dev/tpms<n>

[Jarkko Sakkinen]

On Sun, Jan 22, 2017 at 01:36:28PM -0800, James Bottomley wrote:
> On Sun, 2017-01-22 at 23:04 +0200, Jarkko Sakkinen wrote:
> > On Sun, Jan 22, 2017 at 11:01:07PM +0200, Jarkko Sakkinen wrote:
> > > On Sun, Jan 22, 2017 at 10:30:55PM +0200, Jarkko Sakkinen wrote:
> > > > On Sun, Jan 22, 2017 at 10:48:12AM -0800, James Bottomley wrote:
> > > > > On Sun, 2017-01-22 at 09:49 -0800, James Bottomley wrote:
> > > > > > On Fri, 2017-01-20 at 23:05 +0200, Jarkko Sakkinen wrote:
> > > > > > > 'tabrm4' branch has been now rebased. It's now on top of
> > > > > > > master
> > > > > > > branch that contains Stefan's latest patch (min body length
> > > > > > > check) 
> > > > > > > that I've reviewed and tested. It also contains your
> > > > > > > updated
> > > > > > > /dev/tpms patch.
> > > > > > > 
> > > > > > > I guess the 5 commits that are there now are such that we
> > > > > > > have 
> > > > > > > fairly good consensus, don't we? If so, can I add your
> > > > > > > reviewed-by 
> > > > > > > and tested-by to my commits and vice versa?
> > > > > > 
> > > > > > We're still failing my test_transients.  This is the full
> > > > > > python of 
> > > > > > the test case:
> > > > > > 
> > > > > > 
> > > > > >     def test_transients(self):
> > > > > >         k = self.open_transients()
> > > > > >         self.c.flush_context(k[0])
> > > > > >         self.c.change_auth(self.c.SRK, k[1], None, pwd1)
> > > > > >         ...
> > > > > > 
> > > > > > It's failing at self.c.flush_context(k[0]) with TPM_RC_VALUE.
> > > > > >   It's 
> > > > > > the same problem Ken complained about: TPM2_FlushContext
> > > > > > doesn't have 
> > > > > > a declared handle area so we don't translate the handle being
> > > > > > sent
> > > > > > down.  We have to fix this either by intercepting the flush
> > > > > > and 
> > > > > > manually translating the context, or by being dangerously
> > > > > > clever and 
> > > > > > marking flush as a command which takes one handle.
> > > > > 
> > > > > This is what the dangerously clever fix looks like.  With this
> > > > > and a
> > > > > few other changes, my smoke tests now pass.
> > > > > 
> > > > > James
> > > > 
> > > > I don't want to be clever here. I will rather intercept the body
> > > > and
> > > > try to keep the core code simple and easy to understand.
> > > 
> > > It came out quite clean actually.
> > > 
> > > I just encapsulated handle mapping and have this in the beginning
> > > of
> > > tpm2_map_command:
> > > 
> > > if (cc == TPM2_CC_FLUSH_CONTEXT)
> > > 	return tpm2_map_to_phandle(space, &cmd[TPM_HEADER_SIZE]);
> > > 
> > > I think this documents better what is actually going on than
> > > tinkering
> > > cc_attr_tbl.
> > > 
> > > /Jarkko
> > 
> > Actually what you suggested is much better idea because it will also
> > take care of validation.
> 
> Yes, that's why it's clever ... I'm just always wary of clever code
> because of the Kernighan principle.
> 
> >  I'm still going to keep tpm2_map_to_phandle because it makes the 
> > code flow a lot cleaner and probably sessions have to anyway make it
> > even more complicated.
> 
> OK, there's one more thing that seems to be causing problems: when
> tpm2_save_context fails because the handle no longer exists (like it's
> been flushed) it returns TPM_RC_REFERENCE_H0 not TPM_RC_HANDLE (the
> session code does seem to return TPM_RC_HANDLE under some
> circumstances).
> 
> James

What is your way for reproducing this issue? Just want to add
a test case for my smoke test suite so that I can verify that
the issue is fixed once I've fixed it.

/Jarkko