Re: [PATCH net-next v2 3/4] cgroup: bpf: Add bpf_skb_in_cgroup_proto

[Daniel Borkmann]

On 06/23/2016 06:54 PM, Martin KaFai Lau wrote:
> On Thu, Jun 23, 2016 at 11:53:50AM +0200, Daniel Borkmann wrote:
> > > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > > index 668e079..68753e0 100644
> > > --- a/kernel/bpf/verifier.c
> > > +++ b/kernel/bpf/verifier.c
> > > @@ -1062,6 +1062,10 @@ static int check_map_func_compatibility(struct bpf_map *map, int func_id)
> > >   		if (func_id != BPF_FUNC_get_stackid)
> > >   			goto error;
> > >   		break;
> > > +	case BPF_MAP_TYPE_CGROUP_ARRAY:
> > > +		if (func_id != BPF_FUNC_skb_in_cgroup)
> > > +			goto error;
> > > +		break;
> >
> > I think the BPF_MAP_TYPE_CGROUP_ARRAY case should have been fist here in
> > patch 2/4, but with unconditional goto error. And this one only adds the
> > 'func_id != BPF_FUNC_skb_in_cgroup' test.
> I am not sure I understand.  Can you elaborate? I am probably missing
> something here.

If someone backports patch 2/4 as-is, but for some reason not 3/4, then you
could craft a program that calls f.e. bpf_map_update_elem() on a cgroup array
and would thus cause a NULL pointer deref, since verifier doesn't prevent it.
I'm just trying to say that it would probably make sense to add the above 'case
BPF_MAP_TYPE_CGROUP_ARRAY:' with an unconditional 'goto error' in patch 2/4
and extend upon it in patch 3/4 so result looks like here, so that the patches
are fine/complete each as stand-alone.