Re: [PATCH v4 01/10] IMA: Defined an IMA hook to measure keys on key create or update

[Mimi Zohar <zohar@linux.ibm.com>]

On Thu, 2019-11-07 at 10:42 -0800, Lakshmi Ramasubramanian wrote:
> On 11/6/2019 7:40 PM, Mimi Zohar wrote:
> 
> >>> I would move the patch that defines the "keyring=" policy option prior
> >>> to this one.  Include the call to process_buffer_measurement() in this
> >>> patch.  A subsequent patch would add support to defer measuring the
> >>> key, by calling a function named something like
> >>> ima_queue_key_measurement().
> >>>
> >>
> >> As I'd stated in the other response, I wanted to isolate all key related
> >> code in a separate C file and build it if and only if all CONFIG
> >> dependencies are met.
> > 
> > The basic measuring of keys shouldn't be any different than any other
> > policy rule, other than it is a key and not a file.  This is the
> > reason that I keep saying start out with the basics and then add
> > support to defer measuring keys on the trusted keyrings.
> 
> I'll make the changes, rearrange the patches and send an updated set.
> 
> I do have a few questions since I am still not fully understanding the 
> requirements you are targeting. Appreciate if you could please clarify.
> 
> As you already know, I am using the "public key" of the given asymmetric 
> key as the "buffer" to measure in process_buffer_measurement().
> 
> The measurement decision is not based on whether the keyring is a 
> trusted one or an untrusted one. As long as the IMA policy allows 
> (through the "keyrings=" option) the key will be measured.

We should be able to measure all keys being loaded onto any keyring or
onto a specific "keyring=".   This shouldn't be any different than any
other policy rule.  Once you have this basic feature working, you
would address loading keys during early boot.

> 
> Do you want only trusted keyrings to be allowed in the measurement?
> In my opinion, that decision should be deferred to whoever is setting up 
> the IMA policy.

Right, but it shouldn't be limited to just "trusted" keyrings.  This
way you can first test loading keys onto any keyring.

> 
> > Only the queueing code needed for measuring keys on the trusted
> > keyrings would be in a separate file.
> > 
> 
> The decision to process key immediately or defer (queue) is based on 
> whether IMA has been initialized or not. Keyring is not used for this 
> decision.
> 
> Could you please clarify how queuing is related to keyring's 
> trustworthiness?
> 
> The check for whether the key is an asymmetric one or not, and 
> extracting the "public key" if it is an asymmetric key needs to be in a 
> separate file to handle the CONFIG dependencies in IMA.

Queuing the keys should be independent of measuring the keys.
 Initially you would start with just measuring the key.  From a high
level it would look like:

    ima_post_key_create_or_update(...)
    {
       "measure key based on
    policy(key, keyring, ...)"
    }

This requires the IMA "keyring=" policy option support be defined
first.

Subsequently you would add key queuing support, and then update
ima_post_key_create_or_update().  It would look like:

        ima_post_key_create_or_update(...)
        {
            if (custom policy is loaded)
               "measure key based on policy(key, keyring, ...)"
            else
                "queue key(key, keyring)"
        }

Mimi