Message-ID: <CAEf4BzY73uiNE8V_zyhRuMpX8+e6JyR+BkfHqPfztOpm9Orjng@mail.gmail.com>
Date:   Thu, 29 Oct 2020 19:30:02 -0700
From:   Andrii Nakryiko <andrii.nakryiko@...il.com>
To:     Alexander Duyck <alexander.duyck@...il.com>
Cc:     bpf <bpf@...r.kernel.org>, Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Martin Lau <kafai@...com>,
        john fastabend <john.fastabend@...il.com>,
        Kernel Team <kernel-team@...com>,
        Networking <netdev@...r.kernel.org>,
        Eric Dumazet <edumazet@...gle.com>,
        Lawrence Brakmo <brakmo@...com>, alexanderduyck@...com
Subject: Re: [bpf-next PATCH 4/4] selftests/bpf: Migrate tcpbpf_user.c to use
 BPF skeleton

On Thu, Oct 29, 2020 at 5:42 PM Alexander Duyck
<alexander.duyck@...il.com> wrote:
>
> On Thu, Oct 29, 2020 at 4:20 PM Andrii Nakryiko
> <andrii.nakryiko@...il.com> wrote:
> >
> > On Thu, Oct 29, 2020 at 12:57 AM Alexander Duyck
> > <alexander.duyck@...il.com> wrote:
> > >
> > > From: Alexander Duyck <alexanderduyck@...com>
> > >
> > > Update tcpbpf_user.c to make use of the BPF skeleton. Doing this we can
> > > simplify test_tcpbpf_user and reduce the overhead involved in setting up
> > > the test.
> > >
> > > Signed-off-by: Alexander Duyck <alexanderduyck@...com>
> > > ---
> >
> > Few suggestions below, but overall looks good:
> >
> > Acked-by: Andrii Nakryiko <andrii@...nel.org>
> >
> > >  .../testing/selftests/bpf/prog_tests/tcpbpf_user.c |   48 +++++++++-----------
> > >  1 file changed, 21 insertions(+), 27 deletions(-)
> > >
> > > diff --git a/tools/testing/selftests/bpf/prog_tests/tcpbpf_user.c b/tools/testing/selftests/bpf/prog_tests/tcpbpf_user.c
> > > index 4e1190894e1e..7e92c37976ac 100644
> > > --- a/tools/testing/selftests/bpf/prog_tests/tcpbpf_user.c
> > > +++ b/tools/testing/selftests/bpf/prog_tests/tcpbpf_user.c
> > > @@ -5,6 +5,7 @@
> > >
> > >  #include "test_tcpbpf.h"
> > >  #include "cgroup_helpers.h"
> > > +#include "test_tcpbpf_kern.skel.h"
> > >
> > >  #define LO_ADDR6 "::1"
> > >  #define CG_NAME "/tcpbpf-user-test"
> > > @@ -162,39 +163,32 @@ static void run_test(int map_fd, int sock_map_fd)
> > >
> > >  void test_tcpbpf_user(void)
> > >  {
> > > -       const char *file = "test_tcpbpf_kern.o";
> > > -       int prog_fd, map_fd, sock_map_fd;
> > > -       struct bpf_object *obj;
> > > +       struct test_tcpbpf_kern *skel;
> > > +       int map_fd, sock_map_fd;
> > > +       struct bpf_link *link;
> > >         int cg_fd = -1;
> > > -       int rv;
> > > -
> > > -       cg_fd = cgroup_setup_and_join(CG_NAME);
> > > -       if (CHECK_FAIL(cg_fd < 0))
> > > -               goto err;
> > >
> > > -       if (CHECK_FAIL(bpf_prog_load(file, BPF_PROG_TYPE_SOCK_OPS, &obj, &prog_fd))) {
> > > -               fprintf(stderr, "FAILED: load_bpf_file failed for: %s\n", file);
> > > -               goto err;
> > > -       }
> > > +       skel = test_tcpbpf_kern__open_and_load();
> > > +       if (CHECK(!skel, "open and load skel", "failed"))
> > > +               return;
> > >
> > > -       rv = bpf_prog_attach(prog_fd, cg_fd, BPF_CGROUP_SOCK_OPS, 0);
> > > -       if (CHECK_FAIL(rv)) {
> > > -               fprintf(stderr, "FAILED: bpf_prog_attach: %d (%s)\n",
> > > -                      errno, strerror(errno));
> > > -               goto err;
> > > -       }
> > > +       cg_fd = test__join_cgroup(CG_NAME);
> > > +       if (CHECK_FAIL(cg_fd < 0))
> >
> > please use either CHECK() or one of the newer ASSERT_xxx() macro (also
> > defined in test_progs.h), CHECK_FAIL should be avoided in general.
>
> So the plan I had was to actually move over to the following:
>         cg_fd = test__join_cgroup(CG_NAME);
>         if (CHECK_FAIL(cg_fd < 0))
>                 goto cleanup_skel;
>
> It still makes use of CHECK_FAIL but it looks like test__join_cgroup
> already takes care of error messaging so using CHECK_FAIL in this case
> makes more sense.

CHECK (and ASSERT) leave a paper-trail in verbose test log, so it
makes it easier to debug tests, if something fails. CHECK_FAIL is
invisible, unless if fails. So CHECK_FAIL should be used only for
things that are happening on the order of hundreds of instances per
test, or more.

>
> In addition I was looking at simplifying the first patch which should
> just be the move with minimal changes to allow the functionality to
> build as a part of the test_progs framework. The end result should be
> the same it just helps to make the fact that the first patch should
> just be a move a little more clear.

sure


>
> > > +               goto cleanup_skel;
> > >
> > > -       map_fd = bpf_find_map(__func__, obj, "global_map");
> > > -       if (CHECK_FAIL(map_fd < 0))
> > > -               goto err;
> > > +       map_fd = bpf_map__fd(skel->maps.global_map);
> > > +       sock_map_fd = bpf_map__fd(skel->maps.sockopt_results);
> > >
> > > -       sock_map_fd = bpf_find_map(__func__, obj, "sockopt_results");
> > > -       if (CHECK_FAIL(sock_map_fd < 0))
> > > -               goto err;
> > > +       link = bpf_program__attach_cgroup(skel->progs.bpf_testcb, cg_fd);
> >
> > you can do skel->links.bpf_testcb = bpf_program__attach_cgroup() and
> > skeleton's __destroy() call will take care of destroying the link
>
> Okay, I can look into using that. Actually this has me wondering if
> there wouldn't be some way to make use of test_tcpbpf_kern__attach to
> achieve this in some standard way. I'll get that sorted before I
> submit v2.

cgroup BPF programs can't be auto-attached, because they expect cgroup
FD, which you can't provide at compilation time (declaratively) in BPF
code. So it has to be manually attached. skeleton's attach() will just
ignore such programs.

>
> > > +       if (CHECK(IS_ERR(link), "attach_cgroup(estab)", "err: %ld\n",
> > > +                 PTR_ERR(link)))
> >
> > there is a convenient ASSERT_OK_PTR() specifically for pointers like
> > this (and NULL ones as well, of course); saves a lot of typing and
> > encapsulates error extraction internally.
>
> I'll update the code to address that.

Powered by blists - more mailing lists