Re: [PATCH bpf-next] selftests/bpf: make libbpf_probe_prog_types testcase aware of kernel configuration

[Jiri Olsa]

On Fri, Sep 30, 2022 at 01:09:00PM +0200, Artem Savkov wrote:
> At the moment libbpf_probe_prog_types test iterates over all available
> BPF_PROG_TYPE regardless of kernel configuration which can exclude some
> of those. Unfortunately there is no direct way to tell which types are
> available, but we can look at struct bpf_ctx_onvert to tell which ones
> are available.
> 
> Signed-off-by: Artem Savkov <asavkov@xxxxxxxxxx>
> ---
>  .../selftests/bpf/prog_tests/libbpf_probes.c  | 33 +++++++++++++++++--
>  1 file changed, 31 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/libbpf_probes.c b/tools/testing/selftests/bpf/prog_tests/libbpf_probes.c
> index 9f766ddd946ab..753ddf79cf5e0 100644
> --- a/tools/testing/selftests/bpf/prog_tests/libbpf_probes.c
> +++ b/tools/testing/selftests/bpf/prog_tests/libbpf_probes.c
> @@ -4,12 +4,29 @@
>  #include <test_progs.h>
>  #include <bpf/btf.h>
>  
> +static int find_type_in_ctx_convert(struct btf *btf,
> +				    const char *prog_type_name,
> +				    const struct btf_type *t)
> +{
> +	const struct btf_member *m;
> +	size_t cmplen = strlen(prog_type_name);
> +	int i, n;
> +
> +	for (m = btf_members(t), i = 0, n = btf_vlen(t); i < n; m++, i++) {
> +		const char *member_name = btf__str_by_offset(btf, m->name_off);
> +
> +		if (!strncmp(prog_type_name, member_name, cmplen))
> +			return 1;
> +	}
> +	return 0;
> +}
> +
>  void test_libbpf_probe_prog_types(void)
>  {
>  	struct btf *btf;
> -	const struct btf_type *t;
> +	const struct btf_type *t, *context_convert_t;
>  	const struct btf_enum *e;
> -	int i, n, id;
> +	int i, n, id, context_convert_id;
>  
>  	btf = btf__parse("/sys/kernel/btf/vmlinux", NULL);
>  	if (!ASSERT_OK_PTR(btf, "btf_parse"))
> @@ -23,6 +40,14 @@ void test_libbpf_probe_prog_types(void)
>  	if (!ASSERT_OK_PTR(t, "bpf_prog_type_enum"))
>  		goto cleanup;
>  
> +	context_convert_id = btf__find_by_name_kind(btf, "bpf_ctx_convert",
> +						    BTF_KIND_STRUCT);
> +	if (!ASSERT_GT(context_convert_id, 0, "bpf_ctx_convert_id"))
> +		goto cleanup;
> +	context_convert_t = btf__type_by_id(btf, context_convert_id);
> +	if (!ASSERT_OK_PTR(t, "bpf_ctx_convert_type"))

ASSERT_OK_PTR should check context_convert_t ?

I wonder if we could traverse bpf_ctx_convert members directly
instead of bpf_prog_type enum, but maybe there'd be other issues

jirka

> +		goto cleanup;
> +
>  	for (e = btf_enum(t), i = 0, n = btf_vlen(t); i < n; e++, i++) {
>  		const char *prog_type_name = btf__str_by_offset(btf, e->name_off);
>  		enum bpf_prog_type prog_type = (enum bpf_prog_type)e->val;
> @@ -31,6 +56,10 @@ void test_libbpf_probe_prog_types(void)
>  		if (prog_type == BPF_PROG_TYPE_UNSPEC)
>  			continue;
>  
> +		if (!find_type_in_ctx_convert(btf, prog_type_name,
> +					      context_convert_t))
> +			continue;
> +
>  		if (!test__start_subtest(prog_type_name))
>  			continue;
>  
> -- 
> 2.37.3
>