Re: [PATCH net-next v5 02/15] net-timestamp: prepare for bpf prog use

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

On 1/14/25 4:15 PM, Jason Xing wrote:
> On Wed, Jan 15, 2025 at 8:09 AM Jason Xing <kerneljasonxing@gmail.com> wrote:
>>
>> On Wed, Jan 15, 2025 at 7:40 AM Martin KaFai Lau <martin.lau@linux.dev> wrote:
>>>
>>> On 1/12/25 3:37 AM, Jason Xing wrote:
>>>> Later, I would introduce three points to report some information
>>>> to user space based on this.
>>>>
>>>> Signed-off-by: Jason Xing <kerneljasonxing@gmail.com>
>>>> ---
>>>>    include/net/sock.h |  7 +++++++
>>>>    net/core/sock.c    | 14 ++++++++++++++
>>>>    2 files changed, 21 insertions(+)
>>>>
>>>> diff --git a/include/net/sock.h b/include/net/sock.h
>>>> index f5447b4b78fd..dd874e8337c0 100644
>>>> --- a/include/net/sock.h
>>>> +++ b/include/net/sock.h
>>>> @@ -2930,6 +2930,13 @@ int sock_set_timestamping(struct sock *sk, int optname,
>>>>                          struct so_timestamping timestamping);
>>>>
>>>>    void sock_enable_timestamps(struct sock *sk);
>>>> +#if defined(CONFIG_CGROUP_BPF) && defined(CONFIG_BPF_SYSCALL)
>>>> +void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op);
>>>> +#else
>>>> +static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
>>>> +{
>>>> +}
>>>> +#endif
>>>>    void sock_no_linger(struct sock *sk);
>>>>    void sock_set_keepalive(struct sock *sk);
>>>>    void sock_set_priority(struct sock *sk, u32 priority);
>>>> diff --git a/net/core/sock.c b/net/core/sock.c
>>>> index eae2ae70a2e0..e06bcafb1b2d 100644
>>>> --- a/net/core/sock.c
>>>> +++ b/net/core/sock.c
>>>> @@ -948,6 +948,20 @@ int sock_set_timestamping(struct sock *sk, int optname,
>>>>        return 0;
>>>>    }
>>>>
>>>> +#if defined(CONFIG_CGROUP_BPF) && defined(CONFIG_BPF_SYSCALL)
>>>> +void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
>>>> +{
>>>> +     struct bpf_sock_ops_kern sock_ops;
>>>> +
>>>> +     memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
>>>> +     sock_ops.op = op;
>>>> +     if (sk_is_tcp(sk) && sk_fullsock(sk))
>>>> +             sock_ops.is_fullsock = 1;
>>>> +     sock_ops.sk = sk;
>>>> +     __cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS);
>>>
>>> hmm... I think I have already mentioned it in the earlier revision
>>> (https://lore.kernel.org/bpf/f8e9ab4a-38b9-43a5-aaf4-15f95a3463d0@linux.dev/).
>>
>> Right, sorry, but I deleted it intentionally.
>>
>>>
>>> __cgroup_bpf_run_filter_sock_ops(sk, ...) requires sk to be fullsock.
>>
>> Well, I don't understand it, BPF_CGROUP_RUN_PROG_SOCK_OPS_SK() don't
>> need to check whether it is fullsock or not.

It is because the callers of BPF_CGROUP_RUN_PROG_SOCK_OPS_SK guarantees it is 
fullsock.

>>
>>> Take a look at how BPF_CGROUP_RUN_PROG_SOCK_OPS does it.
>>> sk_to_full_sk() is used to get back the listener. For other mini socks,
>>> it needs to skip calling the cgroup bpf prog. I still don't understand
>>> why BPF_CGROUP_RUN_PROG_SOCK_OPS cannot be used here because of udp.
>>
>> Sorry, I got lost here. BPF_CGROUP_RUN_PROG_SOCK_OPS cannot support
>> udp, right? And I think we've discussed that we have to get rid of the
>> limitation of fullsock.

It is the part I am missing. Why BPF_CGROUP_RUN_PROG_SOCK_OPS cannot support 
udp? UDP is not a fullsock?

> 
> To support udp case, I think I can add the following check for
> __cgroup_bpf_run_filter_sock_ops() instead of directly calling
> BPF_CGROUP_RUN_PROG_SOCK_OPS():
> 1) if the socket belongs to tcp type, it should be fullsock.
> 2) or if it is a udp type socket. Then no need to check and use the fullsock.
> 
> Above lines/policies should be applied to the rest of the series, right?
> 
> According to the existing callbacks, the tcp socket is indeed fullsock.