Re: [PATCH bpf-next] selftests/bpf: Add a test for attaching a bpf fentry/fexit trace to an XDP program

[Andrii Nakryiko <andrii.nakryiko@gmail.com>]

On Thu, Jan 2, 2020 at 6:13 AM Eelco Chaudron <echaudro@redhat.com> wrote:
>
>
>
> On 20 Dec 2019, at 0:02, Andrii Nakryiko wrote:
>
> > On Thu, Dec 19, 2019 at 3:04 AM Eelco Chaudron <echaudro@redhat.com>
> > wrote:
> >>
> >> Add a test that will attach a FENTRY and FEXIT program to the XDP
> >> test
> >> program. It will also verify data from the XDP context on FENTRY and
> >> verifies the return code on exit.
> >>
> >> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
> >> ---
> >>  .../testing/selftests/bpf/prog_tests/xdp_bpf2bpf.c |   95
> >> ++++++++++++++++++++
> >>  .../testing/selftests/bpf/progs/test_xdp_bpf2bpf.c |   44 +++++++++
> >>  2 files changed, 139 insertions(+)
> >>  create mode 100644
> >> tools/testing/selftests/bpf/prog_tests/xdp_bpf2bpf.c
> >>  create mode 100644
> >> tools/testing/selftests/bpf/progs/test_xdp_bpf2bpf.c
> >>
> >
> > [...]
> >
> >> +       /* Load XDP program to introspect */
> >> +       err = bpf_prog_load(file, BPF_PROG_TYPE_XDP, &obj, &prog_fd);
> >
> > Please use BPF skeleton for this test. It will make it significantly
> > shorter and clearer. See other fentry_fexit selftest for example.
> >
>
> Trying to do this, however, I’m getting the following when trying to
> execute the test:
>
> test_xdp_bpf2bpf:PASS:pkt_skel_load 0 nsec
> libbpf: fentry/_xdp_tx_iptunnel is not found in vmlinux BTF
> libbpf: failed to load object 'test_xdp_bpf2bpf'
> libbpf: failed to load BPF skeleton 'test_xdp_bpf2bpf': -2
> test_xdp_bpf2bpf:FAIL:ftrace_skel_load ftrace skeleton failed
>
>
> My program is straight forward following the fentry_fexit.c example:
>
>      pkt_skel = test_xdp__open_and_load()
>      if (CHECK(!pkt_skel, "pkt_skel_load", "test_xdp skeleton
> failed\n"))
>          return;
>
>      map_fd = bpf_map__fd(pkt_skel->maps.vip2tnl);
>      bpf_map_update_elem(map_fd, &key4, &value4, 0);
>
>      /* Load eBPF trace program */
>      ftrace_skel = test_xdp_bpf2bpf__open_and_load();
>      if (CHECK(!ftrace_skel, "ftrace_skel_load", "ftrace skeleton
> failed\n"))
>          goto out;
>
> I assume this is due to the missing link from the XDP program to the
> eBPF trace program.

yes, exactly

> Previously I did this trough:
>
> +       DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
> +                           .attach_prog_fd = prog_fd,
> +                          );
> +
> +       tracer_obj = bpf_object__open_file("./test_xdp_bpf2bpf.o", &opts);
>
>
> If I use this approach as before it works, i.e.:
>
>          DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
>                              .attach_prog_fd = pkt_fd,
>                             );
>
>          ftrace_skel = test_xdp_bpf2bpf__open_opts(&opts);
>          if (CHECK(!ftrace_skel, "__open_opts”, "ftrace skeleton
> failed\n"))
>            goto out;
>          if (CHECK(test_xdp_bpf2bpf__load(ftrace_skel), "__load",
> "ftrace skeleton failed\n"))
>            goto out;
>
> But I do not see this in the fentry_fexit.c example, guess I might be
> missing something that is right in front of me :(

you are not missing anything, and this second variant is what you have
to do. __open_and_load() is just an convenience wrapper around
separate __open() followed by __load(), for cases where user doesn't
have to specify any extra options.

>
>
> [...]
>