Re: [PATCH v6 0/4] LSM: Measure security module data

From: Lakshmi Ramasubramanian <nramas@linux.microsoft.com>
To: Casey Schaufler <casey@schaufler-ca.com>,
	Mimi Zohar <zohar@linux.ibm.com>,
	Tyler Hicks <tyhicks@linux.microsoft.com>
Cc: stephen.smalley.work@gmail.com, sashal@kernel.org,
	jmorris@namei.org, linux-integrity@vger.kernel.org,
	selinux@vger.kernel.org, linux-security-module@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v6 0/4] LSM: Measure security module data
Date: Wed, 5 Aug 2020 11:08:11 -0700	[thread overview]
Message-ID: <50f00ace-8d46-01c2-bf0f-d5484aafd95c@linux.microsoft.com> (raw)
In-Reply-To: <ecc97f59-c2cc-0b23-6199-925ba0d6358b@schaufler-ca.com>

On 8/5/20 10:57 AM, Casey Schaufler wrote:
> On 8/5/2020 10:25 AM, Lakshmi Ramasubramanian wrote:
>> On 8/5/20 10:03 AM, Mimi Zohar wrote:
>>> On Wed, 2020-08-05 at 10:45 -0500, Tyler Hicks wrote:
>>>
>>>> In addition to SELINUX_STATE and SELINUX_POLICY, we should also consider
>>>> the proposed LSM_STATE and LSM_POLICY func values but require an "lsm"
>>>> rule conditional.
>>>>
>>>> So the current proposed rules:
>>>>
>>>> ? measure func=LSM_STATE
>>>> ? measure func=LSM_POLICY
>>>>
>>>> Would become:
>>>>
>>>> ? measure func=LSM_STATE lsm=selinux
>>>> ? measure func=LSM_POLICY lsm=selinux
>>>>
>>>> The following rules would be rejected:
>>>>
>>>> ? measure func=LSM_STATE
>>>> ? measure func=LSM_POLICY
>>>> ? measure func=LSM_STATE lsm=apparmor
>>>> ? measure func=LSM_POLICY lsm=smack
>>>
>>> Kees is cleaning up the firmware code which differentiated between how
>>> firmware was loaded.?? There will be a single firmware enumeration.
>>>
>>> Similarly, the new IMA hook to measure critical state may be placed in
>>> multiple places.? Is there really a need from a policy perspective for
>>> differentiating the source of the critical state being measurind??? The
>>> data being measured should include some identifying information.
>>
>> Yes Mimi - SELinux is including the identifying information in the "event name" field. Please see a sample measurement of STATE and POLICY for SELinux below:
>>
>> 10 e32e...5ac3 ima-buf sha256:86e8...4594 selinux-state-1595389364:287899386 696e697469616c697a65643d313b656e61626c65643d313b656e666f7263696e673d303b636865636b72657170726f743d313b6e6574776f726b5f706565725f636f6e74726f6c733d313b6f70656e5f7065726d733d313b657874656e6465645f736f636b65745f636c6173733d313b616c776179735f636865636b5f6e6574776f726b3d303b6367726f75705f7365636c6162656c3d313b6e6e705f6e6f737569645f7472616e736974696f6e3d313b67656e66735f7365636c6162656c5f73796d6c696e6b733d303
>>
>> 10 f4a7...9408 ima-ng sha256:8d1d...1834 selinux-policy-hash-1595389353:863934271
>>
>>>
>>> I think moving away from the idea that measuring "critical" data should
>>> be limited to LSMs, will clarify this.
>>>
>>
>> Are you suggesting that instead of calling the hooks LSM_STATE and LSM_POLICY, we should keep it more generic so that it can be utilized by any subsystem to measure their "critical data"?
> 
> Policy, state, history or whim, it should be up to the security module
> to determine what data it cares about, and how it should be measured.
> Smack does not keep its policy in a single blob of data, it uses lists
> which can be modified at will. Your definition of the behavior for
> LSM_POLICY wouldn't work for Smack. That doesn't mean that there isn't
> a viable way to measure the Smack policy, it just means that IMA isn't
> the place for it. If SELinux wants its data measured, SELinux should be
> providing the mechanism to do it.
> 
> I guess that I'm agreeing with you in part. If you want a generic measurement
> of "critical data", you don't need to assign a type to it, you have the
> caller (a security module, a device driver or whatever) identify itself and
> how it is going to deal with the data. That's very different from what you've
> done to date.

Agree.

Like Stephen had stated earlier, the reason we kept separate hooks 
(STATE and POLICY) is because the data for state is usually small and 
therefore we measure the entire data. Whereas, policy data is usually 
quite large (a few MB) and hence we measure a hash of the policy.

If change to a generic measurement of "critical data", at the least IMA 
should provide a way to measure "data" and "hash(data)".

And, with the caller providing the identifying information, there would 
be no need to call it "LSM_STATE" or "SELINUX_STATE" or such.

  -lakshmi