Re: [PATCH bpf v3] libbpf: Apply map_set_def_max_entries() for inner_maps on creation

[Yonghong Song <yonghong.song@linux.dev>]

On 1/15/24 6:50 AM, Hou Tao wrote:
> Hi,
>
> On 1/15/2024 8:59 PM, Andrey Grafin wrote:
>> This patch allows to create BPF_MAP_TYPE_ARRAY_OF_MAPS and
>> BPF_MAP_TYPE_HASH_OF_MAPS with values of BPF_MAP_TYPE_PERF_EVENT_ARRAY.
>>
>> 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.
>>
>> Signed-off-by: Andrey Grafin <conquistador@yandex-team.ru>
>> ---
>>   tools/lib/bpf/libbpf.c                        |  4 +++
>>   .../selftests/bpf/progs/test_map_in_map.c     | 23 +++++++++++++++
>>   tools/testing/selftests/bpf/test_maps.c       | 29 ++++++++++++++++++-
> It would be better to move the selftest into a separated patch, so the
> fix patch for libbpf could be backported alone.

+1. Please put the selftest into the second patch.
Also for the second patch commit message, please mention
the test run result. For example, command line './test_maps',
without the first patch:
   ...
   err ...
with the first patch:
   ...
   ... // everything is okay

>>   3 files changed, 55 insertions(+), 1 deletion(-)
>>
>> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
>> index e067be95da3c..8f4d580187aa 100644
>> --- a/tools/lib/bpf/libbpf.c
>> +++ b/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",
>> @@ -5212,6 +5213,9 @@ 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",
>> diff --git a/tools/testing/selftests/bpf/progs/test_map_in_map.c b/tools/testing/selftests/bpf/progs/test_map_in_map.c
>> index f416032ba858..b393d2b8bd6f 100644
>> --- a/tools/testing/selftests/bpf/progs/test_map_in_map.c
>> +++ b/tools/testing/selftests/bpf/progs/test_map_in_map.c
>> @@ -21,6 +21,29 @@ struct {
>>   	__type(value, __u32);
>>   } mim_hash SEC(".maps");
>>   
>> +struct perf_event_array {
>> +	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
>> +	__type(key, __u32);
>> +	__type(value, __u32);
>> +} inner_map0 SEC(".maps"), inner_map1 SEC(".maps");
>> +
>> +struct {
>> +	__uint(type, BPF_MAP_TYPE_ARRAY_OF_MAPS);
>> +	__uint(max_entries, 2);
>> +	__type(key, __u32);
>> +	__array(values, struct perf_event_array);
>> +} mim_array_pe SEC(".maps") = {
>> +	.values = {&inner_map0, &inner_map1}};
>> +
>> +struct {
>> +	__uint(type, BPF_MAP_TYPE_HASH_OF_MAPS);
>> +	__uint(max_entries, 2);
>> +	__type(key, __u32);
>> +	__array(values, struct perf_event_array);
>> +} mim_hash_pe SEC(".maps") = {
>> +	.values = {&inner_map0, &inner_map1}};

This is tricky. I suggest to have max_entries to be
1 for BPF_MAP_TYPE_HASH_OF_MAPS and put only &inner_map0
in 'values'. If you have two buckets for the
hash table, actually it is not clear eventually
which map in which bucket.

>> +
>> +
>>   SEC("xdp")
>>   int xdp_mimtest0(struct xdp_md *ctx)
>>   {
>> diff --git a/tools/testing/selftests/bpf/test_maps.c b/tools/testing/selftests/bpf/test_maps.c
>> index 7fc00e423e4d..03f4d448fd3b 100644
>> --- a/tools/testing/selftests/bpf/test_maps.c
>> +++ b/tools/testing/selftests/bpf/test_maps.c
>> @@ -1159,6 +1159,7 @@ static void test_map_in_map(void)
>>   	__u32 len = sizeof(info);
>>   	__u32 id = 0;
>>   	libbpf_print_fn_t old_print_fn;
>> +	int ret;
> Why not use err instead ?
>>   
>>   	obj = bpf_object__open(MAPINMAP_PROG);
>>   
>> @@ -1190,7 +1191,11 @@ static void test_map_in_map(void)
>>   		goto out_map_in_map;
>>   	}
>>   
>> -	bpf_object__load(obj);
>> +	ret = bpf_object__load(obj);
>> +	if (ret) {
>> +		printf("Failed to load test prog\n");
>> +		goto out_map_in_map;
>> +	}
>>   
>>   	map = bpf_object__find_map_by_name(obj, "mim_array");
>>   	if (!map) {
>> @@ -1226,6 +1231,28 @@ static void test_map_in_map(void)
>>   		goto out_map_in_map;
>>   	}
>>   
>> +	map = bpf_object__find_map_by_name(obj, "mim_array_pe");
>> +	if (!map) {
>> +		printf("Failed to load array of perf event array maps\n");
>> +		goto out_map_in_map;
>> +	}
>> +	mim_fd = bpf_map__fd(map);
>> +	if (mim_fd < 0) {
>> +		printf("Failed to get descriptor for array of perf event array maps\n");
>> +		goto out_map_in_map;
>> +	}
>> +
>> +	map = bpf_object__find_map_by_name(obj, "mim_hash_pe");
>> +	if (!map) {
>> +		printf("Failed to load hash of perf event array maps\n");
>> +		goto out_map_in_map;
>> +	}
>> +	mim_fd = bpf_map__fd(map);
>> +	if (mim_fd < 0) {
>> +		printf("Failed to get descriptor for array of perf event array maps\n");
> array -> hash ?
>> +		goto out_map_in_map;
>> +	}

The above change is really not necessary. If program loading is successful, all the
maps should be created properly. I think your above bpf_object__load test should be
good enough.

>> +
>>   	close(fd);
>>   	fd = -1;
>>   	bpf_object__close(obj);
>