[PATCH v3] KEYS: trusted: Fix trusted key backends when building as module

[Mimi Zohar]

On Mon, 2021-09-27 at 22:55 +0200, Andreas Rammhold wrote:
> On 16:33 27.09.21, Mimi Zohar wrote:
> > On Mon, 2021-09-27 at 22:08 +0200, Andreas Rammhold wrote:
> > > On 07:27 27.09.21, Mimi Zohar wrote:
> > > > On Mon, 2021-09-27 at 10:51 +0200, Andreas Rammhold wrote:
> > > > > On 09:47 13.09.21, Ahmad Fatoum wrote:
> > > > > > Dear trusted key maintainers,
> > > > > > 
> > > > > > On 30.07.21 03:28, Andreas Rammhold wrote:
> > > > > > > Before this commit the kernel could end up with no trusted key sources
> > > > > > > even though both of the currently supported backends (TPM and TEE) were
> > > > > > > compiled as modules. This manifested in the trusted key type not being
> > > > > > > registered at all.
> > > > > > > 
> > > > > > > When checking if a CONFIG_… preprocessor variable is defined we only
> > > > > > > test for the builtin (=y) case and not the module (=m) case. By using
> > > > > > > the IS_REACHABLE() macro we do test for both cases.
> > > > > > > 
> > > > > > > Fixes: 5d0682be3189 ("KEYS: trusted: Add generic trusted keys framework")
> > > > > > > Signed-off-by: Andreas Rammhold <andreas at rammhold.de>
> > > > > > > Reviewed-by: Jarkko Sakkinen <jarkko at kernel.org>
> > > > > > Does anyone intend to pick this up?
> > > > > 
> > > > > Did this end up in any tree by now? I am wondering if I should resend
> > > > > the patch instead. Perhaps it was just overlooked?
> > > > 
> > > > For EVM environments only using trusted and encrypted keys, not file
> > > > signatures, the trusted key is needed to decrypt the "master" key in
> > > > order to verify kernel modules.
> > > 
> > > So what you are saying is that right now (before this patch & after this
> > > patch) you could compile a kernel that wouldn't be able to load any
> > > modules when the trusted keychain part is built as module?
> > 
> > Before this patch, trusted and encrypted keys are builtin, so verifying
> > kernel modules with security.evm containing an EVM hmac would succeed. 
> > Afterwards it would fail, as there's a dependency on the trusted key to
> > verify the integrity of the trusted key module.
> 
> But building with =m was a valid configuration which is the original
> reason for me submitting the patch. So perhaps this should not be
> allowed to be a module then?

My mistake.  Trusted and encrypted key types have always been defined
as tristate.  Only when EVM selects encrypted keys, and by extension
trusted keys, are they builtin.

Mimi