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

[Song Liu <songliubraving@fb.com>]

> On Feb 12, 2020, at 10:34 AM, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
> 
> On Wed, Feb 12, 2020 at 10:29 AM Song Liu <songliubraving@fb.com> wrote:
>> 
>> 
>> 
>>> On Feb 12, 2020, at 10:14 AM, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
>>> 
>>> On Wed, Feb 12, 2020 at 10:07 AM Song Liu <songliubraving@fb.com> wrote:
>>>> 
>>>> 
>>>> 
>>>>> On Feb 12, 2020, at 9:34 AM, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
>>>>> 
>>>>> On Wed, Feb 12, 2020 at 4:32 AM Eelco Chaudron <echaudro@redhat.com> 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_attach_target(prog, xdp_fd,
>>>>>>                               "fentry/xdpfilt_blk_all");
>>>>>> bpf_object__load(trace_obj)
>>>>>> 
>>>>>> Signed-off-by: Eelco Chaudron <echaudro@redhat.com>
>>>> 
>>>> 
>>>> I am trying to solve the same problem with slightly different approach.
>>>> 
>>>> It works as the following (with skeleton):
>>>> 
>>>>       obj = myobject_bpf__open_opts(&opts);
>>>>       bpf_object__for_each_program(prog, obj->obj)
>>>>               bpf_program__overwrite_section_name(prog, new_names[id++]);
>>>>       err = myobject_bpf__load(obj);
>>>> 
>>>> I don't have very strong preference. But I think my approach is simpler?
>>> 
>>> I prefer bpf_program__set_attach_target() approach. Section name is a
>>> program identifier and a *hint* for libbpf to determine program type,
>>> attach type, and whatever else makes sense. But there still should be
>>> an API to set all that manually at runtime, thus
>>> bpf_program__set_attach_target(). Doing same by overriding section
>>> name feels like a hack, plus it doesn't handle overriding
>>> attach_program_fd at all.
>> 
>> We already have bpf_object_open_opts to handle different attach_program_fd.
> 
> Not really, because open_opts apply to bpf_object and all its
> bpf_programs, not to individual bpf_program. So it works only if BPF
> application has only one BPF program. If you have many, you can only
> set the same attach_program_fd for all of them. Basically, open_opts'
> attach_prog_fd should be treated as a default or fallback
> attach_prog_fd.

Fair enough. I will use set_attach_target in my code. 

> 
>> Can we depreciate bpf_object_open_opts.attach_prog_fd with the
>> bpf_program__set_attach_target() approach?
> 
> bpf_program__set_attach_target() overrides attach_prog_fd, yes. But we
> can't just deprecate that option because it's part of an API already,
> even though adding it to open opts was probably a mistake. But for
> simple BPF apps with single BPF program it does work fine, so...

Maybe add a warning saying "attach_prog_fd is deprecated, xxx"?

Thanks,
Song