Re: [PATCH bpf-next v4 2/3] libbpf: Add support for dynamic program attach target

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

On Wed, Feb 19, 2020 at 3:06 AM Eelco Chaudron <echaudro@redhat.com> wrote:
>
>
>
> On 18 Feb 2020, at 22:24, Andrii Nakryiko wrote:
>
> > On Tue, Feb 18, 2020 at 8:34 AM Jakub Sitnicki <jakub@cloudflare.com>
> > wrote:
> >>
> >> Hey Eelco,
> >>
> >> On Mon, Feb 17, 2020 at 12:43 PM GMT, Eelco Chaudron wrote:
> >>> Currently when you want to attach a trace program to a bpf program
> >>> the section name needs to match the tracepoint/function semantics.
> >>>
> >>> However the addition of the bpf_program__set_attach_target() API
> >>> allows you to specify the tracepoint/function dynamically.
> >>>
> >>> The call flow would look something like this:
> >>>
> >>>   xdp_fd = bpf_prog_get_fd_by_id(id);
> >>>   trace_obj = bpf_object__open_file("func.o", NULL);
> >>>   prog = bpf_object__find_program_by_title(trace_obj,
> >>>                                            "fentry/myfunc");
> >>>   bpf_program__set_expected_attach_type(prog, BPF_TRACE_FENTRY);
> >>>   bpf_program__set_attach_target(prog, xdp_fd,
> >>>                                  "xdpfilt_blk_all");
> >>>   bpf_object__load(trace_obj)
> >>>
> >>> Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
> >>> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
> >>> ---
> >>>  tools/lib/bpf/libbpf.c   |   34 ++++++++++++++++++++++++++++++----
> >>>  tools/lib/bpf/libbpf.h   |    4 ++++
> >>>  tools/lib/bpf/libbpf.map |    2 ++
> >>>  3 files changed, 36 insertions(+), 4 deletions(-)
> >>>
> >>> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> >>> index 514b1a524abb..0c25d78fb5d8 100644
> >>> --- a/tools/lib/bpf/libbpf.c
> >>> +++ b/tools/lib/bpf/libbpf.c
> >>
> >> [...]
> >>
> >>> @@ -8132,6 +8133,31 @@ void bpf_program__bpil_offs_to_addr(struct
> >>> bpf_prog_info_linear *info_linear)
> >>>       }
> >>>  }
> >>>
> >>> +int bpf_program__set_attach_target(struct bpf_program *prog,
> >>> +                                int attach_prog_fd,
> >>> +                                const char *attach_func_name)
> >>> +{
> >>> +     int btf_id;
> >>> +
> >>> +     if (!prog || attach_prog_fd < 0 || !attach_func_name)
> >>> +             return -EINVAL;
> >>> +
> >>> +     if (attach_prog_fd)
> >>> +             btf_id = libbpf_find_prog_btf_id(attach_func_name,
> >>> +                                              attach_prog_fd);
> >>> +     else
> >>> +             btf_id = __find_vmlinux_btf_id(prog->obj->btf_vmlinux,
> >>> +                                            attach_func_name,
> >>> +
> >>> prog->expected_attach_type);
> >>> +
> >>> +     if (btf_id <= 0)
> >>> +             return btf_id;
> >>
> >> Looks like we can get 0 as return value on both error and success
> >> (below)?  Is that intentional?
> >
> > Neither libbpf_find_prog_btf_id nor __find_vmlinux_btf_id are going to
> > return 0 on failure. But I do agree that if (btf_id < 0) check would
> > be better here.
>
> Is see in theory btf__find_by_name_kind() could return 0:
>
>         if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
>                 return 0;
>
> But for our case, this will not happen and is invalid, so what about
> just to make sure its future proof?:
>
>    if (btf_id <= 0)
>          return btf_id ? btf_id : -ENOENT;

I don't see how void can be the right attach type, so I'd keep it
simple: if (btf_id < 0) return btf_id.
If it so happens that 0 is returned, it will fail at attach time anyways.

>
>
> > With that minor nit:
> >
> > Acked-by: Andrii Nakryiko <andriin@fb.com>
> >
> >>
> >>> +
> >>> +     prog->attach_btf_id = btf_id;
> >>> +     prog->attach_prog_fd = attach_prog_fd;
> >>> +     return 0;
> >>> +}
> >>> +
> >>>  int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
> >>>  {
> >>>       int err = 0, n, len, start, end = -1;
> >>
> >> [...]
>