Re: [PATCH bpf-next 3/3] bpf/bpftool: handle libbpf_probe_prog_type errors

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

On Fri, Apr 1, 2022 at 2:33 PM Quentin Monnet <quentin@isovalent.com> wrote:
>
> On Fri, 1 Apr 2022 at 19:42, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
> >
> > On Fri, Apr 1, 2022 at 9:05 AM Quentin Monnet <quentin@isovalent.com> wrote:
> > >
> > > 2022-03-31 11:45 UTC-0400 ~ Milan Landaverde <milan@mdaverde.com>
> > > > Previously [1], we were using bpf_probe_prog_type which returned a
> > > > bool, but the new libbpf_probe_bpf_prog_type can return a negative
> > > > error code on failure. This change decides for bpftool to declare
> > > > a program type is not available on probe failure.
> > > >
> > > > [1] https://lore.kernel.org/bpf/20220202225916.3313522-3-andrii@kernel.org/
> > > >
> > > > Signed-off-by: Milan Landaverde <milan@mdaverde.com>
> > > > ---
> > > >  tools/bpf/bpftool/feature.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/tools/bpf/bpftool/feature.c b/tools/bpf/bpftool/feature.c
> > > > index c2f43a5d38e0..b2fbaa7a6b15 100644
> > > > --- a/tools/bpf/bpftool/feature.c
> > > > +++ b/tools/bpf/bpftool/feature.c
> > > > @@ -564,7 +564,7 @@ probe_prog_type(enum bpf_prog_type prog_type, bool *supported_types,
> > > >
> > > >               res = probe_prog_type_ifindex(prog_type, ifindex);
> > > >       } else {
> > > > -             res = libbpf_probe_bpf_prog_type(prog_type, NULL);
> > > > +             res = libbpf_probe_bpf_prog_type(prog_type, NULL) > 0;
> > > >       }
> > > >
> > > >  #ifdef USE_LIBCAP
> > >
> >
> > A completely unrelated question to you, Quentin. How hard is bpftool's
> > dependency on libcap? We've recently removed libcap from selftests, I
> > wonder if it would be possible to do that for bpftool as well to
> > reduce amount of shared libraries bpftool depends on.
>
> There's not a super-strong dependency on it. It's used in feature
> probing, for two things.
>
> First one is to be accurate when we check that the user has the right
> capabilities for probing efficiently the system. A workaround consists
> in checking that we run with uid=0 (root), although it's less
> accurate.
>
> Second thing is probing as an unprivileged user: if bpftool is run to
> probe as root but with the "unprivileged" keyword, libcap is used to
> drop the CAP_SYS_ADMIN and run the probes without it. I don't know if
> there's an easy alternative to libcap for that. Also I don't know how
> many people use this feature, but I remember that this was added
> because there was some demand at the time, so presumably there are
> users relying on this.
>
> This being said, libcap is optional for compiling bpftool, so you
> should be able to have it work just as well if the library is not
> available on the system? Basically you'd just lose the ability to
> probe as an unprivileged user. Do you need to remove the optional
> dependency completely?

Well, see recent patches from Martin:

82cb2b30773e bpf: selftests: Remove libcap usage from test_progs
b1c2768a82b9 bpf: selftests: Remove libcap usage from test_verifier
663af70aabb7 bpf: selftests: Add helpers to directly use the capget
and capset syscall

We completely got rid of libcap dependency and ended up with more
straightforward code. So I was wondering if this is possible for
bpftool. The less unnecessary library dependencies - the better. The
less feature detection -- the better. That's my only line of thought
here :)

>
> Quentin
>
> PS: Not directly related but since we're talking of libcap, we
> recently discovered that the lib is apparently changing errno when it
> maybe shouldn't and plays badly with batch mode:
> https://stackoverflow.com/questions/71608181/bpf-xdp-bpftool-batch-file-returns-error-reading-batch-file-failed-opera

just another argument in favor of getting rid of extra layers of dependencies