Re: [PATCH 2/6] security/keys/encrypted: Clean up request_trusted_key()

[Dan Williams]

On Tue, Mar 19, 2019 at 6:11 PM Mimi Zohar <zohar@xxxxxxxxxxxxx> wrote:
>
> On Tue, 2019-03-19 at 17:20 -0700, Dan Williams wrote:
> > On Tue, Mar 19, 2019 at 5:07 PM Mimi Zohar <zohar@xxxxxxxxxxxxx> wrote:
> > > On Mon, 2019-03-18 at 23:06 -0700, Dan Williams wrote:
> >
> > > > diff --git a/security/keys/key.c b/security/keys/key.c
> > > > index 696f1c092c50..9045b62afb04 100644
> > > > --- a/security/keys/key.c
> > > > +++ b/security/keys/key.c
> > > > @@ -706,6 +706,7 @@ struct key_type *key_type_lookup(const char *type)
> > > >  found_kernel_type:
> > > >       return ktype;
> > > >  }
> > > > +EXPORT_SYMBOL_GPL(key_type_lookup);
> >
> > This needs to be moved to patch1.
> >
> > > Only the kernel is calling key_type_lookup().  Why does
> > > key_type_lookup() need to be exported?
> >
> > This patch series adds several new callers outside of keys-subsystem
> > core that need this export, the first one being encrypted-keys itself
> > in patch1.
>
> It's needed, because they could be compiled as kernel modules, not
> builtin (eg. EVM).
>

Right, but now I realize a problem. The side effect of having direct
module dependencies to the key_type_{encrypted,trusted} symbols is
that module reference counting is handled automatically.

So, I need to respin this with some explicit try_module_get() and
module_put() added otherwise the encrypted_keys facility can be
removed while active keys are instantiated.

> > drivers/nvdimm/security.c:57:   type = key_type_lookup("encrypted");
> > fs/ecryptfs/keystore.c:1627:    type = key_type_lookup("encrypted");
> > security/integrity/evm/evm_crypto.c:361:        type =
> > key_type_lookup("encrypted");
> > security/keys/encrypted-keys/encrypted.c:440:   type =
> > key_type_lookup("trusted");
> >
>