Re: [PATCH v2] tracepoint: add new `tcp:tcp_ca_event` trace event

[Eric Dumazet]

On Tue, Aug 8, 2023 at 7:59 AM Manjusaka <me@xxxxxxxxxxxx> wrote:
>
> In normal use case, the tcp_ca_event would be changed in high frequency.
>
> It's a good indicator to represent the network quanlity.

quality ?

Honestly, it is more about TCP stack tracing than 'network quality'

>
> So I propose to add a `tcp:tcp_ca_event` trace event
> like `tcp:tcp_cong_state_set` to help the people to
> trace the TCP connection status
>
> Signed-off-by: Manjusaka <me@xxxxxxxxxxxx>
> ---
>  include/net/tcp.h          |  9 ++------
>  include/trace/events/tcp.h | 45 ++++++++++++++++++++++++++++++++++++++
>  net/ipv4/tcp_cong.c        | 10 +++++++++
>  3 files changed, 57 insertions(+), 7 deletions(-)
>
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index 0ca972ebd3dd..a68c5b61889c 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -1154,13 +1154,8 @@ static inline bool tcp_ca_needs_ecn(const struct sock *sk)
>         return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
>  }
>
> -static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
> -{
> -       const struct inet_connection_sock *icsk = inet_csk(sk);
> -
> -       if (icsk->icsk_ca_ops->cwnd_event)
> -               icsk->icsk_ca_ops->cwnd_event(sk, event);
> -}
> +/* from tcp_cong.c */
> +void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event);
>
>  /* From tcp_cong.c */
>  void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
> diff --git a/include/trace/events/tcp.h b/include/trace/events/tcp.h
> index bf06db8d2046..b374eb636af9 100644
> --- a/include/trace/events/tcp.h
> +++ b/include/trace/events/tcp.h
> @@ -416,6 +416,51 @@ TRACE_EVENT(tcp_cong_state_set,
>                   __entry->cong_state)
>  );
>
> +TRACE_EVENT(tcp_ca_event,
> +
> +       TP_PROTO(struct sock *sk, const u8 ca_event),
> +
> +       TP_ARGS(sk, ca_event),
> +
> +       TP_STRUCT__entry(
> +               __field(const void *, skaddr)
> +               __field(__u16, sport)
> +               __field(__u16, dport)
> +               __array(__u8, saddr, 4)
> +               __array(__u8, daddr, 4)
> +               __array(__u8, saddr_v6, 16)
> +               __array(__u8, daddr_v6, 16)
> +               __field(__u8, ca_event)
> +       ),
> +

Please add the family (look at commit 3dd344ea84e1 ("net: tracepoint:
exposing sk_family in all tcp:tracepoints"))



> +       TP_fast_assign(
> +               struct inet_sock *inet = inet_sk(sk);
> +               __be32 *p32;
> +
> +               __entry->skaddr = sk;
> +
> +               __entry->sport = ntohs(inet->inet_sport);
> +               __entry->dport = ntohs(inet->inet_dport);
> +
> +               p32 = (__be32 *) __entry->saddr;
> +               *p32 = inet->inet_saddr;
> +
> +               p32 = (__be32 *) __entry->daddr;
> +               *p32 =  inet->inet_daddr;

We keep copying IPv4 addresses that might contain garbage for IPv6 sockets :/

> +
> +               TP_STORE_ADDRS(__entry, inet->inet_saddr, inet->inet_daddr,
> +                          sk->sk_v6_rcv_saddr, sk->sk_v6_daddr);

I will send a cleanup, because IP_STORE_ADDRS() should really take
care of all details.


> +
> +               __entry->ca_event = ca_event;
> +       ),
> +
> +       TP_printk("sport=%hu dport=%hu saddr=%pI4 daddr=%pI4 saddrv6=%pI6c daddrv6=%pI6c ca_event=%u",
> +                 __entry->sport, __entry->dport,
> +                 __entry->saddr, __entry->daddr,
> +                 __entry->saddr_v6, __entry->daddr_v6,
> +                 __entry->ca_event)

Please print the symbol instead of numeric ca_event.

Look at show_tcp_state_name() for instance.