When TCP over IPv4 via INET6 API, sk->sk_family is AF_INET6, but it is a
v4 pkt. inet_csk(sk)->icsk_af_ops is ipv6_mapped and use ip_queue_xmit.
The tos sockopt does not work for bpf [get,set]sockopt() helpers.

Changelog:
v3 -> v4:
* Add 'sk->sk_type != SOCK_RAW && !ipv6_only_sock(sk)' check.
* Re-implement test with LLM assistance.
* v3: https://lore.kernel.org/all/20240914103226.71109-1-zhoufeng.zf@bytedance.com/ (local)

v2->v3:
* Use sk_is_inet() helper. (Eric Dumazet)
* https://lore.kernel.org/bpf/CANn89i+9GmBLCdgsfH=WWe-tyFYpiO27wONyxaxiU6aOBC6G8g@mail.gmail.com/T/ (local)

v1->v2:
* Fix compilation error. (kernel test robot)
* https://lore.kernel.org/bpf/202408152058.YXAnhLgZ-lkp@intel.com/T/ (local)

Leon Hwang (2):
  bpf: Fix bpf_get/setsockopt to tos for ipv4-mapped ipv6 socket
  selftests/bpf: Add test to verify the fix for bpf_setsockopt() helper

 net/core/filter.c                             | 15 +++-
 .../selftests/bpf/prog_tests/setget_sockopt.c | 78 +++++++++++++++++++
 .../selftests/bpf/progs/setget_sockopt.c      | 23 ++++++
 3 files changed, 115 insertions(+), 1 deletion(-)

-- 
2.54.0