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

[Dan Williams]

On Wed, Mar 20, 2019 at 5:07 AM Mimi Zohar <zohar@xxxxxxxxxxxxx> wrote:
>
> On Tue, 2019-03-19 at 22:48 -0700, Dan Williams wrote:
> > On Tue, Mar 19, 2019 at 7:36 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:
> > > > >
> > > > > < snip >
> > > > >
> > > > > > +/*
> > > > > > + * request_trusted_key - request the trusted key
> > > > > > + *
> > > > > > + * Trusted keys are sealed to PCRs and other metadata. Although userspace
> > > > > > + * manages both trusted/encrypted key-types, like the encrypted key type
> > > > > > + * data, trusted key type data is not visible decrypted from userspace.
> > > > > > + */
> > > > > > +static struct key *request_trusted_key(const char *trusted_desc,
> > > > > > +                             const u8 **master_key, size_t *master_keylen)
> > > > > > +{
> > > > > > +     struct trusted_key_payload *tpayload;
> > > > > > +     struct key_type *type;
> > > > > > +     struct key *tkey;
> > > > > > +
> > > > > > +     type = key_type_lookup("trusted");
> > > > >
> > > > > The associated key_type_put() will need to be called.
> > > >
> > > > Yes.
> > >
> > > I don't know if defining a key_type_lookup() wrapper, perhaps named
> > > is_key_type_available(), would help.  Both key_type_lookup() and
> > > key_type_put() would be called.  The existing code could then remain
> > > the same.
> > >
> >
> > Maybe, but something still needs to pin the hosting module. I think
> > this means that the first call to key_type->instantiate() pins the
> > hosting module, and the ->destroy() of the last key for the key_type
> > unpins the module. It does mean that the ->destroy() method is no
> > longer optional.
>
> This sounds like it isn't a new problem.  Both issues need to be
> addressed, but I think we should differentiate between them and
> address them separately.
>
> In terms of the original nvdimm encrypted/trusted key problem, the
> above suggestion requires the least amount of change.  For v5.2, I
> would replace it with the full updated patch set.

I believe smallest amount of change is this single patch:

https://patchwork.kernel.org/patch/10858649/