On Mon, Nov 19, 2018 at 10:42:21PM -0800, Nikita V. Shirokov wrote:
> idea is pretty simple. for specified map (pointed by struct bpf_map)
> we would provide descriptor of already loaded map, which is going to be
> used as a prototype for inner map. proposed workflow:
> 1) open bpf's object (bpf_object__open)
> 2) create bpf's map which is going to be used as a prototype
> 3) find (by name) map-in-map which you want to load and update w/
> descriptor of inner map w/ a new helper from this patch
> 4) load bpf program w/ bpf_object__load
>
> inner_map_fd is ignored by any other maps aside from (hash|array) of
> maps
>
> Signed-off-by: Nikita V. Shirokov <tehn...@tehnerd.com>
> Acked-by: Yonghong Song <y...@fb.com>
> ---
> tools/lib/bpf/libbpf.c | 11 ++++++++++-
> tools/lib/bpf/libbpf.h | 2 ++
> 2 files changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index a01eb9584e52..7e130e0c8fc9 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -163,6 +163,7 @@ struct bpf_map {
> char *name;
> size_t offset;
> int map_ifindex;
> + int inner_map_fd;
> struct bpf_map_def def;
> __u32 btf_key_type_id;
> __u32 btf_value_type_id;
> @@ -653,8 +654,10 @@ bpf_object__init_maps(struct bpf_object *obj, int flags)
> * fd (fd=0 is stdin) when failure (zclose won't close
> * negative fd)).
> */
> - for (i = 0; i < nr_maps; i++)
> + for (i = 0; i < nr_maps; i++) {
> obj->maps[i].fd = -1;
> + obj->maps[i].inner_map_fd = -1;
> + }
>
> /*
> * Fill obj->maps using data in "maps" section.
> @@ -1146,6 +1149,7 @@ bpf_object__create_maps(struct bpf_object *obj)
> create_attr.btf_fd = 0;
> create_attr.btf_key_type_id = 0;
> create_attr.btf_value_type_id = 0;
> + create_attr.inner_map_fd = map->inner_map_fd;
>
> if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
> create_attr.btf_fd = btf__fd(obj->btf);
> @@ -2562,6 +2566,11 @@ void bpf_map__set_ifindex(struct bpf_map *map, __u32
> ifindex)
> map->map_ifindex = ifindex;
> }
>
> +void bpf_map__add_inner_map_fd(struct bpf_map *map, int fd)
> +{
> + map->inner_map_fd = fd;
I think the name bpf_map__set_inner_map_fd() would be more appropriate
and it should check that map->def->type == map-in-map && map->inner_map_fd == -1
before assigning new one.
Also the behavior of bpf_object__create_maps() is not great.
If nothing is set the function will be passing
create_attr.inner_map_fd == -1 to the kernel.
For regular maps that field is sadly ignored by kernel.
Only for map-in-map the value of -1 will be triggering map_create error.
Imo bpf_object__create_maps() should be doing:
if (create_attr.map_type == map-in-map && map->inner_map_fd >= 0)
create_attr.inner_map_fd = map->inner_map_fd;
// otherwise keep it zero inited
