Re: [PATCH v2 1/2] IMA: Define workqueue for early boot "key" measurements

From: Mimi Zohar <zohar@linux.ibm.com>
To: Lakshmi Ramasubramanian <nramas@linux.microsoft.com>,
	linux-integrity@vger.kernel.org
Cc: eric.snowberg@oracle.com, dhowells@redhat.com,
	mathew.j.martineau@linux.intel.com, matthewgarrett@google.com,
	sashal@kernel.org, jamorris@linux.microsoft.com,
	linux-kernel@vger.kernel.org, keyrings@vger.kernel.org
Subject: Re: [PATCH v2 1/2] IMA: Define workqueue for early boot "key" measurements
Date: Thu, 12 Dec 2019 22:54:18 +0000	[thread overview]
Message-ID: <1576191258.4579.181.camel@linux.ibm.com> (raw)
In-Reply-To: <b4ff3607-076e-7b90-24d1-9a129d9ce720@linux.microsoft.com>

On Thu, 2019-12-12 at 13:59 -0800, Lakshmi Ramasubramanian wrote:
> On 12/12/19 1:13 PM, Mimi Zohar wrote:
> 
> > 
> > Looking at this again, something seems off or at least the comment
> > doesn't match the code.
> > 
> >         /*
> >           * To avoid holding the mutex while processing queued keys,
> >           * transfer the queued keys with the mutex held to a temp list,
> >           * release the mutex, and then process the queued keys from
> >           * the temp list.
> >           *
> >           * Since ima_process_keys is set to true above, any new key will
> >           * be processed immediately and not queued.
> >           */
> > 
> > Setting ima_process_key before taking the lock won't prevent the race.
> >   I think you want to test ima_process_keys before taking the lock and
> > again immediately afterward taking the lock, before setting it.  Then
> > the comment would match the code.
> > 
> > Shouldn't ima_process_keys be defined as static to limit the scope to
> > this file?
> > 
> > Mimi
> > 
> 
> In IMA hook, ima_process_key is checked without lock. If it is false, 
> ima_queue_key is called. If the key was queued (by ima_queue_key()) then 
> the hook defers measurement. Else, it processes it immediately.
> 
> In ima_queue_key() function the check for ima_process_key is done after 
> taking the lock and the key queued if the flag is false.
> 
> In ima_process_keys() ima_process_key is set without lock and then the 
> queued keys are moved to a temp list after taking the lock.
> 
> I have reviewed the changes myself and also with a few of my colleagues. 
> I don't think there is a race condition. Please let me know if you do 
> see a problem.
> 
> I can move the setting of ima_process_key flag inside the lock. But 
> honestly I don't think that is necessary.
> 
> I agree that ima_process_keys should be static since it is used in this 
> file one. I'll make that change.
> 
> I can also move the setting of ima_process_key flag inside the lock 
> along with the above change.

My concern is with the last sentence "Since ima_process_keys is set to
true above, any new key will be processed immediately and not queued."
  It's unlikely, but possible, that a second process will wait for the
ima_keys_mutex.  Either we remove this sentence or move setting
ima_process_keys to after taking the lock.

Mimi 