On Thu, Aug 29, 2019 at 03:36:42PM +0200, Nicolas Dichtel wrote:

Le 29/08/2019 à 07:12, Alexei Starovoitov a écrit :
[snip]

quoted

CAP_BPF and CAP_NET_ADMIN together allow the following:
- Attach to cgroup-bpf hooks and query
- skb, xdp, flow_dissector test_run command

CAP_NET_ADMIN allows:
- Attach networking bpf programs to xdp, tc, lwt, flow dissector

I'm not sure to understand the difference between these last two points.
But, with the current kernel, CAP_NET_ADMIN is not enough to attach bpf prog
with tc and it's still not enough after your patch.
The following command is rejected:
$ tc filter add dev eth0 ingress matchall action bpf obj ./tc_test_kern.o sec test

Prog section 'test' rejected: Operation not permitted (1)!
 - Type:         4
 - Instructions: 22 (0 over limit)
 - License:      GPL

Verifier analysis:

Error fetching program/map!
bad action parsing
parse_action: bad value (5:bpf)!
Illegal "action"

because tc/iproute2 is doing load and attach.
Currently load needs cap_sys_admin and
attach needs cap_net_admin.