[PATCH] lsm: drop LSM_ID_IMA

[Paul Moore]

On Wed, Oct 25, 2023 at 10:37 AM Roberto Sassu
<roberto.sassu at huaweicloud.com> wrote:
> On 10/25/2023 4:06 PM, Roberto Sassu wrote:
> > On 10/25/2023 3:14 PM, Paul Moore wrote:
> >> On Wed, Oct 25, 2023 at 6:36 AM Roberto Sassu
> >> <roberto.sassu at huaweicloud.com> wrote:
> >>> On 10/24/2023 11:18 PM, Paul Moore wrote:
> >>>> On Mon, Oct 23, 2023 at 11:48 AM Casey Schaufler
> >>>> <casey at schaufler-ca.com> wrote:
> >>>>> On 10/23/2023 8:20 AM, Roberto Sassu wrote:
> >>>>>> On 10/20/2023 11:56 PM, Casey Schaufler wrote:
> >>>>>>> On 10/19/2023 1:08 AM, Roberto Sassu wrote:
> >>>>>>>> On Wed, 2023-10-18 at 17:50 -0400, Paul Moore wrote:
> >>>>>>>>> When IMA becomes a proper LSM we will reintroduce an appropriate
> >>>>>>>>> LSM ID, but drop it from the userspace API for now in an effort
> >>>>>>>>> to put an end to debates around the naming of the LSM ID macro.
> >>>>>>>>>
> >>>>>>>>> Signed-off-by: Paul Moore <paul at paul-moore.com>
> >>>>>>>> Reviewed-by: Roberto Sassu <roberto.sassu at huawei.com>
> >>>>>>>>
> >>>>>>>> This makes sense according to the new goal of making 'ima' and
> >>>>>>>> 'evm' as
> >>>>>>>> standalone LSMs.
> >>>>>>>>
> >>>>>>>> Otherwise, if we took existing LSMs, we should have defined
> >>>>>>>> LSM_ID_INTEGRITY, associated to DEFINE_LSM(integrity).
> >>>>>>>>
> >>>>>>>> If we proceed with the new direction, I will add the new LSM IDs as
> >>>>>>>> soon as IMA and EVM become LSMs.
> >>>>>>>
> >>>>>>> This seems right to me. Thank You.
> >>>>>>
> >>>>>> Perfect! Is it fine to assign an LSM ID to 'ima' and 'evm' and keep
> >>>>>> the 'integrity' LSM to reserve space in the security blob without LSM
> >>>>>> ID (as long as it does not register any hook)?
> >>>>>
> >>>>> That will work, although it makes me wonder if all the data in the
> >>>>> 'integrity' blob
> >>>>> is used by both IMA and EVM. If these are going to be separate LSMs
> >>>>> they should probably
> >>>>> have their own security blobs. If there is data in common then an
> >>>>> 'integrity' blob can
> >>>>> still makes sense.
> >>>>
> >>>> Users interact with IMA and EVM, not the "integrity" layer, yes?  If
> >>>> so, I'm not sure it makes sense to have an "integrity" LSM, we should
> >>>> just leave it at "IMA" and "EVM".
> >>>
> >>> The problem is who reserves and manages the shared integrity metadata.
> >>> For now, it is still the 'integrity' LSM. If not, it would be IMA or EVM
> >>> on behalf of the other (depending on which ones are enabled). Probably
> >>> the second would not be a good idea.
> >>
> >> I'm not certain that managing kernel metadata alone necessitates a LSM
> >> ID token value.  Does "integrity" have any user visible "things" that
> >> it would want to expose to userspace?
> >
> > No, it doesn't. I already moved the LSM hook registration to 'ima' and
> > 'evm'. Also the old 'integrity' initialization is done by 'ima' and 'evm'.
> >
> > DEFINE_LSM(integrity) exists only to reserve space in the security blob
> > and to provide the security blob offset to the get/set functions.
> >
> > Maybe I send the patch set, so that you get a better idea.
>
> Uhm, we should have updated security.c and removed:
>
>          (IS_ENABLED(CONFIG_IMA) ? 1 : 0) + \

With LSM_CONFIG_COUNT only being used inside security_add_hooks() I
believe you are correct.  Do you want to send a patch against
lsm/dev-staging?

-- 
paul-moore.com