Re: [PATCH bpf v2] selftests/bpf: fix endianness issues in test_sysctl

[Daniel Borkmann <daniel@iogearbox.net>]

On 8/19/19 12:59 PM, Ilya Leoshkevich wrote:
> A lot of test_sysctl sub-tests fail due to handling strings as a bunch
> of immediate values in a little-endian-specific manner.
> 
> Fix by wrapping all immediates in bpf_ntohl and the new bpf_be64_to_cpu.
> 
> Also, sometimes tests fail because sysctl() unexpectedly succeeds with
> an inappropriate "Unexpected failure" message and a random errno. Zero
> out errno before calling sysctl() and replace the message with
> "Unexpected success".
> 
> Fixes: 1f5fa9ab6e2e ("selftests/bpf: Test BPF_CGROUP_SYSCTL")
> Fixes: 9a1027e52535 ("selftests/bpf: Test file_pos field in bpf_sysctl ctx")
> Fixes: 6041c67f28d8 ("selftests/bpf: Test bpf_sysctl_get_name helper")
> Fixes: 11ff34f74e32 ("selftests/bpf: Test sysctl_get_current_value helper")
> Fixes: 786047dd08de ("selftests/bpf: Test bpf_sysctl_{get,set}_new_value helpers")
> Fixes: 8549ddc832d6 ("selftests/bpf: Test bpf_strtol and bpf_strtoul helpers")
> Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
> ---
> v1->v2: Use bpf_ntohl and bpf_be64_to_cpu, drop __bpf_le64_to_cpu.
> 
>   tools/testing/selftests/bpf/bpf_endian.h  |   7 ++
>   tools/testing/selftests/bpf/test_sysctl.c | 130 ++++++++++++++--------
>   2 files changed, 92 insertions(+), 45 deletions(-)
> 
> diff --git a/tools/testing/selftests/bpf/bpf_endian.h b/tools/testing/selftests/bpf/bpf_endian.h
> index 05f036df8a4c..f3be9322e89c 100644
> --- a/tools/testing/selftests/bpf/bpf_endian.h
> +++ b/tools/testing/selftests/bpf/bpf_endian.h
> @@ -29,6 +29,8 @@
>   # define __bpf_htonl(x)			__builtin_bswap32(x)
>   # define __bpf_constant_ntohl(x)	___constant_swab32(x)
>   # define __bpf_constant_htonl(x)	___constant_swab32(x)
> +# define __bpf_be64_to_cpu(x)		__builtin_bswap64(x)
> +# define __bpf_constant_be64_to_cpu(x)	___constant_swab64(x)
>   #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
>   # define __bpf_ntohs(x)			(x)
>   # define __bpf_htons(x)			(x)
> @@ -38,6 +40,8 @@
>   # define __bpf_htonl(x)			(x)
>   # define __bpf_constant_ntohl(x)	(x)
>   # define __bpf_constant_htonl(x)	(x)
> +# define __bpf_be64_to_cpu(x)		(x)
> +# define __bpf_constant_be64_to_cpu(x)  (x)
>   #else
>   # error "Fix your compiler's __BYTE_ORDER__?!"
>   #endif
> @@ -54,5 +58,8 @@
>   #define bpf_ntohl(x)				\
>   	(__builtin_constant_p(x) ?		\
>   	 __bpf_constant_ntohl(x) : __bpf_ntohl(x))
> +#define bpf_be64_to_cpu(x)			\
> +	(__builtin_constant_p(x) ?		\
> +	 __bpf_constant_be64_to_cpu(x) : __bpf_be64_to_cpu(x))
>   
>   #endif /* __BPF_ENDIAN__ */

By the way, looks like progs/test_lwt_seg6local.c defines its own ntohll()/htonll(),
could we make this two patches: i) consolidating the former by adding proper variants
to bpf_endian.h (I mean generically) and ii) fixing test_sysctl.c by using them? Thinking
addition of i) as plan for bpf-next would be to move bpf_endian.h as proper API into
libbpf in near future such that BPF progs don't need to redefine them.

Thanks,
Daniel