[dm-devel] [PATCH 1/7] dm: measure data on table load

[Tushar Sugandhi]

Hi Mimi,

On 7/21/21 2:17 PM, Mimi Zohar wrote:
> On Wed, 2021-07-21 at 12:07 -0400, Mimi Zohar wrote:
>> On Wed, 2021-07-21 at 11:42 -0400, Mike Snitzer wrote:
>>> On Tue, Jul 20 2021 at 10:12P -0400,
>>> Mimi Zohar <zohar at linux.ibm.com> wrote:
>>>
>>>> Hi Tushar, Mike,
>>>>
>>>> On Mon, 2021-07-12 at 17:48 -0700, Tushar Sugandhi wrote:
>>>>> +struct dm_ima_device_table_metadata {
>>>>> +       /*
>>>>> +        * Contains data specific to the device which is common across
>>>>> +        * all the targets in the table.e.g. name, uuid, major, minor etc.
>>>>> +        * The values are stored in comma separated list of key1=val1,key2=val2; pairs
>>>>> +        * delimited by a semicolon at the end of the list.
>>>>> +        */
>>>>> +       char *device_metadata;
>>>>> +       unsigned int device_metadata_len;
>>>>> +       unsigned int num_targets;
>>>>> +
>>>>> +       /*
>>>>> +        * Contains the sha256 hashs of the IMA measurements of the
>>>>> +        * target attributes key-value pairs from the active/inactive tables.
>>>>> +        */
>>>>
>>>>  From past experience hard coding the hash algorithm is really not a
>>>> good idea.
>>>>
>>>> Mimi
>>>>
>>>>> +       char *hash;
>>>>> +       unsigned int hash_len;
>>>>> +
>>>>> +};
>>>
>>> Hi Mimi,
>>>
>>> The dm-ima.c code is using SHASH_DESC_ON_STACK and then storing the
>>> more opaque result via 'hash' and 'hash_len'.
>>>
>>> So if/when the dm-ima.c hash algorithm were to change this detail
>>> won't change the dm_ima_device_table_metadata structure at all right?
>>> But even if changes were needed this is purely an implementation
>>> detail correct?  Why might users care which algorithm is used by
>>> dm-ima to generate the hashes?
>>>
>>> Assuming there is a valid reason for users to care about this, we can
>>> improve this aspect as follow-on work.. so I don't consider this a
>>> blocker for this patchset at this point.  Please clarify if you feel
>>> it should be a blocker.
>>
>> This goes back to my question as to if or how the template data in
>> these DM critical data records are to be validated by the attestation
>> server.   Asumming the hash/hash_len is being stored in the IMA
>> measurement list, the less the attestation should need to know about
>> the specific kernel version the better.
> 
> Hi Mike, Tushar,  Laskshmi,
> 
> Perhaps, when defining a new IMA "critical data" record, especially if
> you know it's going to change, the critical data should contain a
> version number.
>
Just to close the loop on this thread:

As I explained on the other thread in this patch series -

@the hash verification:
the clear-text for the active/inactive table hashes is already logged in
the 'table_load' event. And we will prefix the active/inactive table
hashes with hash_algo.  (e.g. sha256:<hash>) in the remaining events.
Together it should be sufficient for the attestation server to verify
the active/inactive table hashes without maintaining any list of
expected hashes or referring to kernel version etc.

@versioning:
We are already logging versions for individual targets. What was missing
was some versioning at device level. So thanks again for the suggestion.
We will add a version at device level in each of the events. Together
that should help attestation server to determine what attributes to 
expect in the event data - without relying on specific kernel version.

Thanks again for your feedback.

Regards,
Tushar

> thanks,
> 
> Mimi
>