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

Stephen Smalley stephen.smalley.work at gmail.com
Tue Sep 8 01:28:51 UTC 2020


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.




More information about the Linux-audit mailing list