Re: [PATCH bpf-next v2] bpf: Fix bpf_get/setsockopt to tos not take effect when TCP over IPv4 via INET6 API

[Martin KaFai Lau <martin.lau@linux.dev>]

On 8/23/24 6:35 AM, Eric Dumazet wrote:
> On Fri, Aug 23, 2024 at 10:53 AM Feng zhou <zhoufeng.zf@bytedance.com> wrote:
>>
>> From: Feng Zhou <zhoufeng.zf@bytedance.com>
>>
>> when TCP over IPv4 via INET6 API, bpf_get/setsockopt with ipv4 will
>> fail, because sk->sk_family is AF_INET6. With ipv6 will success, not
>> take effect, because inet_csk(sk)->icsk_af_ops is ipv6_mapped and
>> use ip_queue_xmit, inet_sk(sk)->tos.
>>
>> So bpf_get/setsockopt needs add the judgment of this case. Just check
>> "inet_csk(sk)->icsk_af_ops == &ipv6_mapped".
>>
>> | Reported-by: kernel test robot <lkp@intel.com>
>> | Closes: https://lore.kernel.org/oe-kbuild-all/202408152034.lw9Ilsj6-lkp@intel.com/
>> Signed-off-by: Feng Zhou <zhoufeng.zf@bytedance.com>
>> ---
>> Changelog:
>> v1->v2: Addressed comments from kernel test robot
>> - Fix compilation error
>> Details in here:
>> https://lore.kernel.org/bpf/202408152058.YXAnhLgZ-lkp@intel.com/T/
>>
>>   include/net/tcp.h   | 2 ++
>>   net/core/filter.c   | 6 +++++-
>>   net/ipv6/tcp_ipv6.c | 6 ++++++
>>   3 files changed, 13 insertions(+), 1 deletion(-)
>>
>> diff --git a/include/net/tcp.h b/include/net/tcp.h
>> index 2aac11e7e1cc..ea673f88c900 100644
>> --- a/include/net/tcp.h
>> +++ b/include/net/tcp.h
>> @@ -493,6 +493,8 @@ struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
>>                                              struct tcp_options_received *tcp_opt,
>>                                              int mss, u32 tsoff);
>>
>> +bool is_tcp_sock_ipv6_mapped(struct sock *sk);
>> +
>>   #if IS_ENABLED(CONFIG_BPF)
>>   struct bpf_tcp_req_attrs {
>>          u32 rcv_tsval;
>> diff --git a/net/core/filter.c b/net/core/filter.c
>> index ecf2ddf633bf..02a825e35c4d 100644
>> --- a/net/core/filter.c
>> +++ b/net/core/filter.c
>> @@ -5399,7 +5399,11 @@ static int sol_ip_sockopt(struct sock *sk, int optname,
>>                            char *optval, int *optlen,
>>                            bool getopt)
>>   {
>> -       if (sk->sk_family != AF_INET)
>> +       if (sk->sk_family != AF_INET
>> +#if IS_BUILTIN(CONFIG_IPV6)
>> +           && !is_tcp_sock_ipv6_mapped(sk)
>> +#endif
>> +           )
>>                  return -EINVAL;
> 
> This does not look right to me.
> 
> I would remove the test completely.
> 
> SOL_IP socket options are available on AF_INET6 sockets just fine.

Good point on the SOL_IP options.

The sk could be neither AF_INET nor AF_INET6. e.g. the bpf_get/setsockopt 
calling from the bpf_lsm's socket_post_create). so the AF_INET test is still needed.

Adding "&& sk->sk_family != AF_INET6" should do. From ipv6_setsockopt, I think 
it also needs to consider the "sk->sk_type != SOCK_RAW".

Please add a test in the next re-spin.

pw-bot: cr