Re: [PATCH v15 00/11] LSM: Three basic syscalls

[Casey Schaufler]

On 10/19/2023 12:45 AM, Roberto Sassu wrote:
> On Wed, 2023-10-18 at 16:40 -0400, Paul Moore wrote:
>> On Wed, Oct 18, 2023 at 4:23 PM Mimi Zohar <zohar@xxxxxxxxxxxxx> wrote:
>>> On Wed, 2023-10-18 at 12:35 -0400, Paul Moore wrote:
>>>> On Wed, Oct 18, 2023 at 10:15 AM Roberto Sassu
>>>> <roberto.sassu@xxxxxxxxxxxxxxx> wrote:
>>>>> On 10/18/2023 3:09 PM, Mimi Zohar wrote:
>>>> ...
>>>>
>>>>>> I agree with Roberto.  All three should be defined: LSM_ID_INTEGRITY,
>>>>>> LSM_ID_IMA, LSM_ID_EVM.
>>>>> I did not try yet, but the 'integrity' LSM does not need an LSM ID. With
>>>>> the last version adding hooks to 'ima' or 'evm', it should be sufficient
>>>>> to keep DEFINE_LSM(integrity) with the request to store a pointer in the
>>>>> security blob (even the init function can be a dummy function).
>>>> First off, this *really* should have been brought up way, way, *way*
>>>> before now.  This patchset has been discussed for months, and bringing
>>>> up concerns in the eleventh hour is borderline rude.
>>> As everyone knows IMA and EVM are not LSMs at this point.
>> Considering all the work Roberto has been doing to make that happen,
>> not to mention the discussions we've had on this topic, that's an
>> awfully small technicality to use as the basis of an argument.
> Sorry Paul, but I've been working on this patch set for a long time and
> you were also involved in the prerequisites (like making the
> 'integrity' LSM as the last).
>
> I thought it was clear at this point that we were going to add the
> hooks to the 'integrity' LSM.

There's a chicken/egg issue here. You can hold up the syscalls patch
forever if you insist on it accommodating every patch set that's in the
pipeline. I understand that you've been working on the integrity rework
for some time. I understand that it's frustrating when things change out
from under you. Believe me, I do.

>
> I really have no problem to rebase my work on top of other work, but I
> was very surprised to see LSM_ID_IMA instead of LSM_ID_INTEGRITY, and
> at minimum this should have been agreed with Mimi. And also, I was not
> convinced with the argument that LSM_ID_IMA should represent IMA+EVM.
>
> Roberto
>