Re: [PATCH bpf 0/5] bpf: allow wide (u64) aligned loads for some fields of bpf_sock_addr

[Yonghong Song <yhs@fb.com>]

On 7/15/19 9:39 AM, Stanislav Fomichev wrote:
> When fixing selftests by adding support for wide stores, Yonghong
> reported that he had seen some examples where clang generates
> single u64 loads for two adjacent u32s as well:
> http://lore.kernel.org/netdev/a66c937f-94c0-eaf8-5b37-8587d66c0c62@fb.com
> 
> Let's support aligned u64 reads for some bpf_sock_addr fields
> as well.
> 
> (This can probably wait for bpf-next, I'll defer to Younhong and the
> maintainers.)
> 
> Cc: Yonghong Song <yhs@fb.com>
> 
> Stanislav Fomichev (5):
>    bpf: rename bpf_ctx_wide_store_ok to bpf_ctx_wide_access_ok
>    bpf: allow wide aligned loads for bpf_sock_addr user_ip6 and
>      msg_src_ip6
>    selftests/bpf: rename verifier/wide_store.c to verifier/wide_access.c
>    selftests/bpf: add selftests for wide loads
>    bpf: sync bpf.h to tools/

Thanks for fixing. Maybe getting into bpf is better as this indeed
a potential issue? I do not have strong feeling either as the
issue can be easily workarounded with "volatile" tricks.

Acked-by: Yonghong Song <yhs@fb.com>

> 
>   include/linux/filter.h                        |  2 +-
>   include/uapi/linux/bpf.h                      |  4 +-
>   net/core/filter.c                             | 24 ++++--
>   tools/include/uapi/linux/bpf.h                |  4 +-
>   .../selftests/bpf/verifier/wide_access.c      | 73 +++++++++++++++++++
>   .../selftests/bpf/verifier/wide_store.c       | 36 ---------
>   6 files changed, 95 insertions(+), 48 deletions(-)
>   create mode 100644 tools/testing/selftests/bpf/verifier/wide_access.c
>   delete mode 100644 tools/testing/selftests/bpf/verifier/wide_store.c
>