Re: Trying the bpf trace a bpf xdp program

"Eelco Chaudron" <echaudro@xxxxxxxxxx> · Wed, 04 Dec 2019 19:52:17 +0100

On 4 Dec 2019, at 19:01, Yonghong Song wrote:

<SNIP>

I’ve put my code on GitHub, maybe it’s just something stupid…

Thanks for the test case. This indeed a kernel bug.
The following change fixed the issue:


-bash-4.4$ git diff

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index a0482e1c4a77..034ef81f935b 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -9636,7 +9636,10 @@ static int check_attach_btf_id(struct
bpf_verifier_env *env)
                                 ret = -EINVAL;
                                 goto out;
                         }
-                       addr = (long)
tgt_prog->aux->func[subprog]->bpf_func;
+                       if (subprog == 0)
+                               addr = (long) tgt_prog->bpf_func;
+                       else
+                               addr = (long)
tgt_prog->aux->func[subprog]->bpf_func;
                 } else {
                         addr = kallsyms_lookup_name(tname);
                         if (!addr) {
-bash-4.4$

The reason is for a bpf program without any additional subprogram
(callees), tgt_prog->aux->func is not populated and is a NULL pointer,
so the access tgt_prog->aux->func[0]->bpf_func will segfault.

With the above change, your test works properly.

Thanks for the quick response, and as you mention the test passes with 
the patch above.

I will continue my experiments later this week, and let you know if I 
run into any other problems.

Cheers,

Eelco









