Re: [PATCH bpf-next v2 2/3] bpf, netns: Keep attached programs in bpf_prog_array

On Tue, Jun 23, 2020 at 09:33 PM CEST, Martin KaFai Lau wrote:

On Tue, Jun 23, 2020 at 12:34:58PM +0200, Jakub Sitnicki wrote:

[ ... ]

quoted

@@ -93,8 +108,16 @@ static int bpf_netns_link_update_prog(struct bpf_link *link,
 		goto out_unlock;
 	}

+	run_array = rcu_dereference_protected(net->bpf.run_array[type],
+					      lockdep_is_held(&netns_bpf_mutex));
+	if (run_array)
+		ret = bpf_prog_array_replace_item(run_array, link->prog, new_prog);
+	else

When will this happen?

This will never happen, unless there is a bug. As long as there is a
link attached, run_array should never be detached (null). Because it can
be handled gracefully, we fail the bpf(LINK_UPDATE) syscall.

Your question makes me think that perhaps it should trigger a warning,
with WARN_ON_ONCE, to signal clearly to the reader that this is an
unexpected state.

WDYT?

quoted

+		ret = -ENOENT;
+	if (ret)
+		goto out_unlock;
+
 	old_prog = xchg(&link->prog, new_prog);
-	rcu_assign_pointer(net->bpf.progs[type], new_prog);
 	bpf_prog_put(old_prog);

 out_unlock:

@@ -142,14 +165,38 @@ static const struct bpf_link_ops bpf_netns_link_ops = {
 	.show_fdinfo = bpf_netns_link_show_fdinfo,
 };