Re: [PATCH bpf-next v2] libbpf: Add getter for pointer to data area for internal maps

["Toke Høiland-Jørgensen" <toke@redhat.com>]

Alexei Starovoitov <ast@fb.com> writes:

> On 3/27/20 3:26 PM, Toke Høiland-Jørgensen wrote:
>> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>> 
>>> On Fri, Mar 27, 2020 at 5:58 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>>>>
>>>> For internal maps (most notably the maps backing global variables), libbpf
>>>> uses an internal mmaped area to store the data after opening the object.
>>>> This data is subsequently copied into the kernel map when the object is
>>>> loaded.
>>>>
>>>> This adds a getter for the pointer to that internal data store. This can be
>>>> used to modify the data before it is loaded into the kernel, which is
>>>> especially relevant for RODATA, which is frozen on load. This same pointer
>>>> is already exposed to the auto-generated skeletons, so access to it is
>>>> already API; this just adds a way to get at it without pulling in the full
>>>> skeleton infrastructure.
>>>>
>>>> Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
>>>> ---
>>>> v2:
>>>>    - Add per-map getter for data area instead of a global rodata getter for bpf_obj
>>>>
>>>> tools/lib/bpf/libbpf.c   | 9 +++++++++
>>>>   tools/lib/bpf/libbpf.h   | 1 +
>>>>   tools/lib/bpf/libbpf.map | 1 +
>>>>   3 files changed, 11 insertions(+)
>>>>
>>>> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
>>>> index 085e41f9b68e..a0055f8908fd 100644
>>>> --- a/tools/lib/bpf/libbpf.c
>>>> +++ b/tools/lib/bpf/libbpf.c
>>>> @@ -6756,6 +6756,15 @@ void *bpf_map__priv(const struct bpf_map *map)
>>>>          return map ? map->priv : ERR_PTR(-EINVAL);
>>>>   }
>>>>
>>>> +void *bpf_map__data_area(const struct bpf_map *map, size_t *size)
>>>
>>> I'm not entirely thrilled about "data_area" name. This is entirely for
>>> providing initial value for maps, so maybe something like
>>> bpf_map__init_value() or something along those lines?
>>>
>>> Actually, how about a different API altogether:
>>>
>>> bpf_map__set_init_value(struct bpf_map *map, void *data, size_t size)?
>>>
>>> Application will have to prepare data of correct size, which will be
>>> copied to libbpf's internal storage. It also doesn't expose any of
>>> internal pointer. I don't think extra memcopy is a big deal here.
>>> Thoughts?
>> 
>> Huh, yeah, that's way better. Why didn't I think of that? Think maybe I
>> was too focused on doing this the same way the skeleton code is. I'll
>> send a v3 :)
>
> Could you please add a selftest as well?
> I'm not excited about new features without tests.

Sure, will do.

-Toke