On Tue, May 26, 2020 at 7:59 AM Jesper Dangaard Brouer
<brouer@xxxxxxxxxx> wrote:
>
> When a BPF-map type doesn't support having a BTF info associated, the
> bpf_map_ops->map_check_btf is set to map_check_no_btf(). This function
> map_check_no_btf() currently returns -ENOTSUPP, which result in a very
> confusing error message in libbpf, see below.
>
> The errno ENOTSUPP is part of the kernels internal errno in file
> include/linux/errno.h. As is stated in the file, these "should never be seen
> by user programs."
>
> Choosing errno EUCLEAN instead, which translated to "Structure needs
> cleaning" by strerror(3). This hopefully leads people to think about data
> structures which BTF is all about.
How about instead of tweaking error code, we actually just add support
for BTF key/values for all maps. For special maps, we can just enforce
that BTF is 4-byte integer (or typedef of that), so that in practice
you'll be defining it as:
struct {
__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
__type(key, u32);
__type(value, u32);
} my_map SEC(".maps");
and it will just work?
>
> Before this change end-users of libbpf will see:
> libbpf: Error in bpf_create_map_xattr(cpu_map):ERROR: strerror_r(-524)=22(-524). Retrying without BTF.
>
> After this change end-users of libbpf will see:
> libbpf: Error in bpf_create_map_xattr(cpu_map):Structure needs cleaning(-117). Retrying without BTF.
>
> Fixes: e8d2bec04579 ("bpf: decouple btf from seq bpf fs dump and enable more maps")
> Signed-off-by: Jesper Dangaard Brouer <brouer@xxxxxxxxxx>
> ---
> kernel/bpf/syscall.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index d13b804ff045..ecde7d938421 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -732,7 +732,7 @@ int map_check_no_btf(const struct bpf_map *map,
> const struct btf_type *key_type,
> const struct btf_type *value_type)
> {
> - return -ENOTSUPP;
> + return -EUCLEAN;
> }
>
> static int map_check_btf(struct bpf_map *map, const struct btf *btf,
>
>
