[PATCH v20 05/23] net: Prepare UDS for security module stacking

[John Johansen]

On 9/9/20 11:19 AM, Casey Schaufler wrote:
> On 9/9/2020 6:19 AM, Stephen Smalley wrote:
>> On Tue, Sep 8, 2020 at 8:21 PM John Johansen
>> <john.johansen at canonical.com> wrote:
>>> On 9/8/20 4:37 PM, Casey Schaufler wrote:
>>>> On 9/8/2020 6:35 AM, Stephen Smalley wrote:
>>>>> On Mon, Sep 7, 2020 at 9:28 PM Stephen Smalley
>>>>> <stephen.smalley.work at gmail.com> wrote:
>>>>>> On Sat, Sep 5, 2020 at 3:07 PM John Johansen
>>>>>> <john.johansen at canonical.com> wrote:
>>>>>>> On 9/5/20 11:13 AM, Casey Schaufler wrote:
>>>>>>>> On 9/5/2020 6:25 AM, Paul Moore wrote:
>>>>>>>>> On Fri, Sep 4, 2020 at 7:58 PM Casey Schaufler <casey at schaufler-ca.com> wrote:
>>>>>>>>>> On 9/4/2020 2:53 PM, Paul Moore wrote:
>>>>>>>>>>> On Fri, Sep 4, 2020 at 5:35 PM Casey Schaufler <casey at schaufler-ca.com> wrote:
>>>>>>>>>>>> On 9/4/2020 1:08 PM, Paul Moore wrote:
>>>>>>>>> ...
>>>>>>>>>
>>>>>>>>>>> I understand the concerns you mention, they are all valid as far as
>>>>>>>>>>> I'm concerned, but I think we are going to get burned by this code as
>>>>>>>>>>> it currently stands.
>>>>>>>>>> Yes, I can see that. We're getting burned by the non-extensibility
>>>>>>>>>> of secids. It will take someone smarter than me to figure out how to
>>>>>>>>>> fit N secids into 32bits without danger of either failure or memory
>>>>>>>>>> allocation.
>>>>>>>>> Sooo what are the next steps here?  It sounds like there is some
>>>>>>>>> agreement that the currently proposed unix_skb_params approach is a
>>>>>>>>> problem, but it also sounds like you just want to merge it anyway?
>>>>>>>> There are real problems with all the approaches. This is by far the
>>>>>>>> least invasive of the lot. If this is acceptable for now I will commit
>>>>>>>> to including the dynamic allocation version in the full stacking
>>>>>>>> (e.g. Smack + SELinux) stage. If it isn't, well, this stage is going
>>>>>>>> to take even longer than it already has. Sigh.
>>>>>>>>
>>>>>>>>
>>>>>>>>> I was sorta hoping for something a bit better.
>>>>>>>> I will be looking at alternatives. I am very much open to suggestions.
>>>>>>>> I'm not even 100% convinced that Stephen's objections to my separate
>>>>>>>> allocation strategy outweigh its advantages. If you have an opinion on
>>>>>>>> that, I'd love to hear it.
>>>>>>>>
>>>>>>> fwiw I prefer the separate allocation strategy, but as you have already
>>>>>>> said it trading off one set of problems for another. I would rather see
>>>>>>> this move forward and one set of trade offs isn't significantly worse
>>>>>>> than the other to me so, either wfm.
>>>>>> I remain unclear that AppArmor needs this patch at all even when
>>>>>> support for SO_PEERSEC lands.
>>>>>> Contrary to the patch description, it is about supporting SCM_SECURITY
>>>>>> for datagram not SO_PEERSEC.  And I don't know of any actual users of
>>>>>> SCM_SECURITY even for SELinux, just SO_PEERSEC.
>>>>> I remembered that systemd once tried using SCM_SECURITY but that was a
>>>>> bug since systemd was using it with stream sockets and that wasn't
>>>>> supported by the kernel at the time,
>>>>> https://bugzilla.redhat.com/show_bug.cgi?id=1224211, so systemd
>>>>> switched over to using SO_PEERSEC.  Subsequently I did fix
>>>>> SCM_SECURITY to work with stream sockets via kernel commit
>>>>> 37a9a8df8ce9de6ea73349c9ac8bdf6ba4ec4f70 but SO_PEERSEC is still
>>>>> preferred.  Looking around, I see that there is still one usage of
>>>>> SCM_SECURITY in systemd-journald but it doesn't seem to be required
>>>>> (if provided, journald will pass the label along but nothing seems to
>>>>> depend on it AFAICT).  In any event, I don't believe this patch is
>>>>> needed to support stacking AppArmor.
>>>> Stephen is, as is so often the case, correct. AppArmor has a stub
>>>> socket_getpeersec_dgram() that gets removed in patch 23. If I remove
>>> right but as I said before this is coming, I have been playing with
>>> it and have code. So the series doesn't need it today but sooner than
>>> later it will be needed
> 
> Is sooner like 5.10, or 5.15? It matters.
> 

I can split SCM_SECURITY off from the rest of the unix mediation and
push it off for a while. So lets call it 5.15 or later.

>> I don't understand why.  Is there a userspace component that relies on
>> SCM_SECURITY today for anything real (more than just blindly passing
>> it along and maybe writing to a log somewhere)?  And this doesn't
>> provide support for a composite SCM_SECURITY or SCM_CONTEXT, so it
>> doesn't really solve the stacking problem for it anyway.  What am I
>> missing?  Why do you care about this patch?