[PATCH RFC bpf-next 0/3] bpf_sk_lookup.remote_port fixes

[Ilya Leoshkevich <iii@linux.ibm.com>]

Hi,

These are my current patches for the discussion from [1] and [2].

With these changes tests pass on both x86_64 and s390x, but there are
quite a few things I'm not sure about, hence it's just an RFC:

- Can moving size < target_size check lead to incorrect shifts in some
  corner cases that I missed?

- Are different layouts of remote_port on little- and big-endian a bug
  or a feature? Do we want things to be this way? If not, are we bound
  by the ABI anyway?

- Is there any way to make uapi changes look nicer? A wall of nested
  structs, unions and ifdefs in an otherwise clean struct definition
  isn't looking particularly good.

- What is the Officially Approved way to access the remote_port field
  from C code? I'm leaning towards bpf_ntohs((__u16)remote_port), like
  in [3], and I adjusted the test accordingly.

[1] https://lore.kernel.org/bpf/CAEf4BzaRNLw9_EnaMo5e46CdEkzbJiVU3j9oxnsemBKjNFf3wQ@mail.gmail.com/
[2] https://lore.kernel.org/bpf/20220221180358.169101-1-jakub@cloudflare.com/
[3] https://lore.kernel.org/bpf/20220113070245.791577-1-imagedong@tencent.com/

Ilya Leoshkevich (3):
  bpf: Fix certain narrow loads with offsets
  bpf: Fix bpf_sk_lookup.remote_port on big-endian
  selftests/bpf: Adapt bpf_sk_lookup.remote_port loads

 include/uapi/linux/bpf.h                        | 17 +++++++++++++++--
 kernel/bpf/verifier.c                           | 14 +++++++++-----
 net/core/filter.c                               |  5 ++---
 tools/include/uapi/linux/bpf.h                  | 17 +++++++++++++++--
 .../selftests/bpf/progs/test_sk_lookup.c        | 17 +++++++++++------
 5 files changed, 52 insertions(+), 18 deletions(-)

-- 
2.34.1