Re: [PATCH 2/2] groups: Allow unprivileged processes to use setgroups to drop groups

[josh-iaAMLnmF4UmaiuxdJuQwMA@public.gmane.org]

On Mon, Nov 17, 2014 at 02:22:59PM -0800, Andy Lutomirski wrote:
> On Mon, Nov 17, 2014 at 2:11 PM, Eric W.Biederman <ebiederm-aS9lmoZGLiVWk0Htik3J/w@public.gmane.org> wrote:
> >
> >
> > On November 17, 2014 1:07:30 PM EST, Andy Lutomirski <luto-kltTT9wpgjJwATOyAt5JVQ@public.gmane.org> wrote:
> >>On Nov 17, 2014 3:37 AM, "One Thousand Gnomes"
> >><gnomes-qBU/x9rampVanCEyBjwyrvXRex20P6io@public.gmane.org> wrote:
> >>>
> >>> > optional), I can do that too.  The security model of "having a
> >>group
> >>> > gives you less privilege than not having it" seems crazy, but
> >>> > nonetheless I can see a couple of easy ways that we can avoid
> >>breaking
> >>>
> >>> It's an old pattern of use that makes complete sense in a traditional
> >>> Unix permission world because it's the only way to do "exclude
> >>{list}"
> >>> nicely. Our default IMHO shouldn't break this.
> >>>
> >>> > that pattern, no_new_privs being one of them.  I'd like to make
> >>sure
> >>> > that nobody sees any other real-world corner case that unprivileged
> >>> > setgroups would break.
> >>>
> >>> Barring the usual risk of people doing improper error checking I
> >>don't
> >>> see one immediately.
> >>>
> >>> For containers I think it actually makes sense that the sysctl can be
> >>> applied per container anyway.
> >>
> >>We'll probably need per container sysctls some day.
> >
> > We already have a mess of per network namespace sysctls,
> > as well as few for other namespaces.
> >
> > We have the infrastructure it is just a matter of using it for whatever purpose we need.
> >
> 
> A list of group id ranges that it's permissible to drop would do the
> trick, both for setgroups and for unshare.  The downside would be that
> users in those groups (i.e. everyone by default) would not be able to
> unshare their user ns.
> 
> Better ideas welcome.

Personally, I think that seems like more flexibility than necessary to
achieve the goal.  I think a sysctl turning group-dropping on and off
would suffice; systems that know they don't use groups to exclude
specific users can enable that sysctl.

- Josh Triplett