Re: [PATCH v3 3/3] selinux: add permission names to trace event

[Paul Moore]

On Fri, Aug 21, 2020 at 8:15 AM Stephen Smalley
<stephen.smalley.work@xxxxxxxxx> wrote:
>
> On Thu, Aug 20, 2020 at 10:22 PM Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
> >
> > On Tue, Aug 18, 2020 at 8:14 AM Stephen Smalley
> > <stephen.smalley.work@xxxxxxxxx> wrote:
> > > On Tue, Aug 18, 2020 at 4:11 AM peter enderborg <peter.enderborg@xxxxxxxx> wrote:
> >
> > ...
> >
> > > > Is there any other things we need to fix? A part 1&2 now OK?
> > >
> > > They looked ok to me, but Paul should review them.
> >
> > Patches 1 and 2 look fine to me with the small nits that Stephen
> > pointed out corrected.  I'm glad to see the information in string form
> > now, I think that will be a big help for people making use of this.
> >
> > Unfortunately, I'm a little concerned about patch 3 for the reason
> > Stephen already mentioned.  While changes to the class mapping are
> > infrequent, they do happen, and I'm not very excited about adding it
> > to the userspace kAPI via a header.  Considering that the tracing
> > tools are going to be running on the same system that is being
> > inspected, perhaps the tracing tools could inspect
> > /sys/fs/selinux/class at runtime to query the permission mappings?
> > Stephen, is there a libselinux API which does this already?
>
> There is a libselinux API but both it and the /sys/fs/selinux/class
> tree is exposing the policy values for classes/permissions, not the
> kernel-private indices.  The dynamic class/perm mapping support
> introduced a layer of indirection between them.  The tracepoint is in
> the avc and therefore dealing with the kernel-private values, not the
> policy values.  The mapping occurs on entry/exit of the security
> server functions. So there is no way for userspace to read the kernel
> class/perm values.  We'd just need to keep them in sync manually.  And
> one is allowed to insert new classes or permissions before existing
> ones, thereby changing the values of existing ones, or even to remove
> them.

Ah, okay, thanks.  Can you tell I've never really had to look very
closely at that code ;)

-- 
paul moore
www.paul-moore.com