libbpf: Apply map_set_def_max_entries() for inner_maps on creation

[ Upstream commit f04deb90e516e8e48bf8693397529bc942a9e80b ]

This patch allows to auto create BPF_MAP_TYPE_ARRAY_OF_MAPS and
BPF_MAP_TYPE_HASH_OF_MAPS with values of BPF_MAP_TYPE_PERF_EVENT_ARRAY
by bpf_object__load().

Previous behaviour created a zero filled btf_map_def for inner maps and
tried to use it for a map creation but the linux kernel forbids to create
a BPF_MAP_TYPE_PERF_EVENT_ARRAY map with max_entries=0.

Fixes: 646f02f ("libbpf: Add BTF-defined map-in-map support")
Signed-off-by: Andrey Grafin <conquistador@yandex-team.ru>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Acked-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/bpf/20240117130619.9403-1-conquistador@yandex-team.ru
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

tools/lib/bpf/libbpf.c

@@ -70,6 +70,7 @@
 
 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
+static int map_set_def_max_entries(struct bpf_map *map);
 
 static const char * const attach_type_name[] = {
 	[BPF_CGROUP_INET_INGRESS]	= "cgroup_inet_ingress",
@@ -5172,6 +5173,9 @@ static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, b
 
 	if (bpf_map_type__is_map_in_map(def->type)) {
 		if (map->inner_map) {
+			err = map_set_def_max_entries(map->inner_map);
+			if (err)
+				return err;
 			err = bpf_object__create_map(obj, map->inner_map, true);
 			if (err) {
 				pr_warn("map '%s': failed to create inner map: %d\n",