Message-ID: <87499f4e-d1d1-916c-24d2-0228bcacad26@linux.microsoft.com>
Date:   Fri, 23 Oct 2020 15:50:52 -0700
From:   Tushar Sugandhi <tusharsu@...ux.microsoft.com>
To:     Mimi Zohar <zohar@...ux.ibm.com>, stephen.smalley.work@...il.com,
        casey@...aufler-ca.com, agk@...hat.com, snitzer@...hat.com,
        gmazyland@...il.com
Cc:     tyhicks@...ux.microsoft.com, sashal@...nel.org, jmorris@...ei.org,
        nramas@...ux.microsoft.com, linux-integrity@...r.kernel.org,
        selinux@...r.kernel.org, linux-security-module@...r.kernel.org,
        linux-kernel@...r.kernel.org, dm-devel@...hat.com
Subject: Re: [PATCH v4 4/6] IMA: add policy to measure critical data from
 kernel components



On 2020-10-22 2:15 p.m., Mimi Zohar wrote:
> Hi Tushar,
> 
> The above Subject line should be truncated to "IMA: add policy to
> measure critical data".
> 
> On Wed, 2020-09-23 at 12:20 -0700, Tushar Sugandhi wrote:
>> There would be several candidate kernel components suitable for IMA
>> measurement. Not all of them would have support for IMA measurement.
> 
> This intro is besides the point.  The patch description should describe
> what is meant by "critical data".
> 
Thanks. I will fix the description to address this.

>> Also, system administrators may not want to measure data for all of
>> them, even when they support IMA measurement.
>> An IMA policy option
>> specific to various kernel components is needed to measure their
>> respective critical data.
>>
>> This policy option needs to be constrained to measure data for
>> specific kernel components that are specified as input values to the
>> policy option.
>>
>> Add a new IMA policy option - "data_sources:=" to allow measuring
>> various critical kernel components. This policy option would enable the
>> system administrators to limit the measurement to the components
>> listed in "data_sources:=", if the components support IMA measurement.
>>
>> The new policy option "data_sources:=" is different from the existing
>> policy option "keyrings:=".
>>
>> In case of "keyrings:=", a policy may measure all keyrings (when
>> "keyrings:=" option is not provided for func KEY_CHECK), or may
>> constrain which keyrings need to be measured (when "keyrings:=" option
>> is provided for func KEY_CHECK).
>>
>> But unlike "keyrings:=", the entries in "data_sources:=" would have
>> different data format. Further, the components listed in
>> "data_sources:=" need to be modified to call IMA to measure their
>> data. Therefore, unlike "keyrings:=", IMA shouldn't measure all of the
>> components by default, when "data_sources:=" is not specified. Because
>> measuring non-vetted components just by specifying them as a policy
>> option value may impact the overall reliability of the system.
>>
>> To address this, "data_sources:=" should be a mandatory policy option
>> for func=CRITICAL_DATA. This func is introduced in the 5th patch in this
>> series). And the compile-time vetting functionality described above is
>> introduced in the 6th patch in this series.
>>
>> Signed-off-by: Tushar Sugandhi <tusharsu@...ux.microsoft.com>
> 
> I don't understand what you mean by "non-vetted" components, nor how
> measuring critical data may impact the overall reliability of the
> system.
> 
Tushar: Before we introduced the mechanism to check supported
data-sources at compile time (patch 6/6 of this series), there was a
back-and-forth on whether “data_sources:=” should be a mandatory policy
option, or optional like “keyrings:=”. And we decided to make the
“data_sources:=” mandatory. But now that we have the compile time check
(patch 6/6 of this series), we can switch to make “data_sources:=”
optional (with the default to allow measuring all critical data – just 
like what you suggested below). I will make the code and description 
changes accordingly.
> The system owner or adminstrator defines what they want to measure,
> including "critical data", based on the policy rules.  They might
> decide that they want to constrain which "critical data" is measured by
> specifying "data_sources:=", but that decision is their perogative.
> The default should allow measuring all critical data.
> 
Makes sense.
To summarize, we will make the decision which sources to measure- based
on the sources defined in the allow list (in patch 6) and the sources
defined in “data_sources:=”. If “data_sources:=” is not present, we will
measure all sources defined in the allow list.
Hope my this understanding is correct based on your feedback.

>   A simple example of "critical data" could be some in memory structure,
> along the lines of __ro_after_init, or hash of the memory structure.
> Once the data structure is initialized, the "critical data" measurement
> shouldn't change.    From the attestation server perspective, the IMA
> measurement list would contain a single record unless the critical data
> changes.  The attestation server doesn't need to know anything about
> the initial value, just that it has changed in order to trigger some
> sort alert.
Yes agreed. After the updates (based on your feedback) I stated above,
the behavior should remain consistent with what you described here.
> 
> thanks,
> 
> Mimi
> 

Powered by blists - more mailing lists