Re: [PATCH ghak90 V6 00/10] audit: implement container identifier

From: Daniel Walsh
Date: Wed May 29 2019 - 10:11:21 EST

On 5/29/19 9:17 AM, Paul Moore wrote:
> On Wed, May 29, 2019 at 8:03 AM Daniel Walsh <dwalsh@xxxxxxxxxx> wrote:
>> On 5/28/19 8:43 PM, Richard Guy Briggs wrote:
>>> On 2019-05-28 19:00, Steve Grubb wrote:
>>>> On Tuesday, May 28, 2019 6:26:47 PM EDT Paul Moore wrote:
>>>>> On Tue, May 28, 2019 at 5:54 PM Daniel Walsh <dwalsh@xxxxxxxxxx> wrote:
>>>>>> On 4/22/19 9:49 AM, Paul Moore wrote:
>>>>>>> On Mon, Apr 22, 2019 at 7:38 AM Neil Horman <nhorman@xxxxxxxxxxxxx>
>>>> wrote:
>>>>>>>> On Mon, Apr 08, 2019 at 11:39:07PM -0400, Richard Guy Briggs wrote:
>>>>>>>>> Implement kernel audit container identifier.
>>>>>>>> I'm sorry, I've lost track of this, where have we landed on it? Are we
>>>>>>>> good for inclusion?
>>>>>>> I haven't finished going through this latest revision, but unless
>>>>>>> Richard made any significant changes outside of the feedback from the
>>>>>>> v5 patchset I'm guessing we are "close".
>>>>>>> Based on discussions Richard and I had some time ago, I have always
>>>>>>> envisioned the plan as being get the kernel patchset, tests, docs
>>>>>>> ready (which Richard has been doing) and then run the actual
>>>>>>> implemented API by the userland container folks, e.g. cri-o/lxc/etc.,
>>>>>>> to make sure the actual implementation is sane from their perspective.
>>>>>>> They've already seen the design, so I'm not expecting any real
>>>>>>> surprises here, but sometimes opinions change when they have actual
>>>>>>> code in front of them to play with and review.
>>>>>>> Beyond that, while the cri-o/lxc/etc. folks are looking it over,
>>>>>>> whatever additional testing we can do would be a big win. I'm
>>>>>>> thinking I'll pull it into a separate branch in the audit tree
>>>>>>> (audit/working-container ?) and include that in my secnext kernels
>>>>>>> that I build/test on a regular basis; this is also a handy way to keep
>>>>>>> it based against the current audit/next branch. If any changes are
>>>>>>> needed Richard can either chose to base those changes on audit/next or
>>>>>>> the separate audit container ID branch; that's up to him. I've done
>>>>>>> this with other big changes in other trees, e.g. SELinux, and it has
>>>>>>> worked well to get some extra testing in and keep the patchset "merge
>>>>>>> ready" while others outside the subsystem look things over.
>>>>>> Mrunal Patel (maintainer of CRI-O) and I have reviewed the API, and
>>>>>> believe this is something we can work on in the container runtimes team
>>>>>> to implement the container auditing code in CRI-O and Podman.
>>>>> Thanks Dan. If I pulled this into a branch and built you some test
>>>>> kernels to play with, any idea how long it might take to get a proof
>>>>> of concept working on the cri-o side?
>>>> We'd need to merge user space patches and let them use that instead of the
>>>> raw interface. I'm not going to merge user space until we are pretty sure the
>>>> patch is going into the kernel.
>>> I have an f29 test rpm of the userspace bits if that helps for testing:
>>> Here's what it contains (minus the last patch):
>>>> -Steve
>>>>> FWIW, I've also reached out to some of the LXC folks I know to get
>>>>> their take on the API. I think if we can get two different container
>>>>> runtimes to give the API a thumbs-up then I think we are in good shape
>>>>> with respect to the userspace interface.
>>>>> I just finished looking over the last of the pending audit kernel
>>>>> patches that were queued waiting for the merge window to open so this
>>>>> is next on my list to look at. I plan to start doing that
>>>>> tonight/tomorrow, and as long as the changes between v5/v6 are not
>>>>> that big, it shouldn't take too long.
>>> - RGB
>>> --
>>> Richard Guy Briggs <rgb@xxxxxxxxxx>
>>> Sr. S/W Engineer, Kernel Security, Base Operating Systems
>>> Remote, Ottawa, Red Hat Canada
>>> IRC: rgb, SunRaycer
>>> Voice: +1.647.777.2635, Internal: (81) 32635
>> Our current thoughts are to put the setting of the ID inside of conmon,
>> and then launching the OCI Runtime. In a perfect world this would
>> happen in the OCI Runtime, but we have no controls over different OCI
>> Runtimes.
>> By putting it into conmon, then CRI-O and Podman will automatically get
>> the container id support. After we have this we have to plumb it back
>> up through the contianer engines to be able to easily report the link
>> between the Container UUID and The Kernel Container Audit ID.
> I'm glad you guys have a plan, that's encouraging, but sadly I have no
> idea about the level of complexity/difficulty involved in modifying
> the various container bits for a proof-of-concept? Are we talking a
> week or two? A month? More?
If we had the kernel and the libaudit api, it would involve a small
effort in conmon, I would figure a few days for a POC. Getting the
hole wiring into CRI-O and Podman, would be a little more effort.