[PATCH v3 net-next 1/2] tcp: use data length instead of skb->len in tcp_probe

[PATCH v3 net-next 1/2] tcp: use data length instead of skb->len in tcp_probe

[Yafang Shao]

At this point skb->len is including tcp header length, so it is meaningless
to user. data length could be more helpful, with which we can easily filter
out the packet without payload.

Cc: Eric Dumazet <edumazet@google.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>

---
v3: tcp_hdr() is a little expensive than skb->data, so replace it with
    skb->data.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
---
 include/trace/events/tcp.h | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/include/trace/events/tcp.h b/include/trace/events/tcp.h
index c1a5284..7ff0446 100644
--- a/include/trace/events/tcp.h
+++ b/include/trace/events/tcp.h
@@ -236,7 +236,7 @@
 		__field(__u16, sport)
 		__field(__u16, dport)
 		__field(__u32, mark)
-		__field(__u16, length)
+		__field(__u16, data_len)
 		__field(__u32, snd_nxt)
 		__field(__u32, snd_una)
 		__field(__u32, snd_cwnd)
@@ -250,6 +250,7 @@
 	TP_fast_assign(
 		const struct tcp_sock *tp = tcp_sk(sk);
 		const struct inet_sock *inet = inet_sk(sk);
+		const struct tcphdr *th = (const struct tcphdr *)skb->data;
 
 		memset(__entry->saddr, 0, sizeof(struct sockaddr_in6));
 		memset(__entry->daddr, 0, sizeof(struct sockaddr_in6));
@@ -261,7 +262,7 @@
 		__entry->dport = ntohs(inet->inet_dport);
 		__entry->mark = skb->mark;
 
-		__entry->length = skb->len;
+		__entry->data_len = skb->len - __tcp_hdrlen(th);
 		__entry->snd_nxt = tp->snd_nxt;
 		__entry->snd_una = tp->snd_una;
 		__entry->snd_cwnd = tp->snd_cwnd;
@@ -272,9 +273,9 @@
 		__entry->sock_cookie = sock_gen_cookie(sk);
 	),
 
-	TP_printk("src=%pISpc dest=%pISpc mark=%#x length=%d snd_nxt=%#x snd_una=%#x snd_cwnd=%u ssthresh=%u snd_wnd=%u srtt=%u rcv_wnd=%u sock_cookie=%llx",
+	TP_printk("src=%pISpc dest=%pISpc mark=%#x data_len=%d snd_nxt=%#x snd_una=%#x snd_cwnd=%u ssthresh=%u snd_wnd=%u srtt=%u rcv_wnd=%u sock_cookie=%llx",
 		  __entry->saddr, __entry->daddr, __entry->mark,
-		  __entry->length, __entry->snd_nxt, __entry->snd_una,
+		  __entry->data_len, __entry->snd_nxt, __entry->snd_una,
 		  __entry->snd_cwnd, __entry->ssthresh, __entry->snd_wnd,
 		  __entry->srtt, __entry->rcv_wnd, __entry->sock_cookie)
 );
-- 
1.8.3.1

[PATCH v3 net-next 2/2] tcp: minor optimization around tcp_hdr() usage in tcp receive path

[Yafang Shao]

This is additional to the commit ea1627c20c34 ("tcp: minor optimizations around tcp_hdr() usage").
At this point, skb->data is same with tcp_hdr() as tcp header has not
been pulled yet.

Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
---
 net/ipv4/tcp_ipv4.c | 2 +-
 net/ipv6/tcp_ipv6.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index adbdb50..d179386 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1486,7 +1486,7 @@ int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 				sk->sk_rx_dst = NULL;
 			}
 		}
-		tcp_rcv_established(sk, skb, tcp_hdr(skb));
+		tcp_rcv_established(sk, skb, (const struct tcphdr *)skb->data);
 		return 0;
 	}
 
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 7d47c2b..1c633ff 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1322,7 +1322,7 @@ static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 			}
 		}
 
-		tcp_rcv_established(sk, skb, tcp_hdr(skb));
+		tcp_rcv_established(sk, skb, (const struct tcphdr *)skb->data);
 		if (opt_skb)
 			goto ipv6_pktoptions;
 		return 0;
-- 
1.8.3.1

Re: [PATCH v3 net-next 2/2] tcp: minor optimization around tcp_hdr() usage in tcp receive path

[Eric Dumazet]

On Mon, May 28, 2018 at 8:36 AM Yafang Shao <laoar.shao@gmail.com> wrote:

> This is additional to the commit ea1627c20c34 ("tcp: minor optimizations
around tcp_hdr() usage").
> At this point, skb->data is same with tcp_hdr() as tcp header has not
> been pulled yet.

> Cc: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> ---
>   net/ipv4/tcp_ipv4.c | 2 +-
>   net/ipv6/tcp_ipv6.c | 2 +-
>   2 files changed, 2 insertions(+), 2 deletions(-)

> diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
> index adbdb50..d179386 100644
> --- a/net/ipv4/tcp_ipv4.c
> +++ b/net/ipv4/tcp_ipv4.c
> @@ -1486,7 +1486,7 @@ int tcp_v4_do_rcv(struct sock *sk, struct sk_buff
*skb)
>                                  sk->sk_rx_dst = NULL;
>                          }
>                  }
> -               tcp_rcv_established(sk, skb, tcp_hdr(skb));
> +               tcp_rcv_established(sk, skb, (const struct tcphdr
*)skb->data);
>                  return 0;
>          }

> diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
> index 7d47c2b..1c633ff 100644
> --- a/net/ipv6/tcp_ipv6.c
> +++ b/net/ipv6/tcp_ipv6.c
> @@ -1322,7 +1322,7 @@ static int tcp_v6_do_rcv(struct sock *sk, struct
sk_buff *skb)
>                          }
>                  }

> -               tcp_rcv_established(sk, skb, tcp_hdr(skb));
> +               tcp_rcv_established(sk, skb, (const struct tcphdr
*)skb->data);
>                  if (opt_skb)
>                          goto ipv6_pktoptions;
>                  return 0;
> --
> 1.8.3.1


I would rather remove the third parameter of tcp_rcv_established() instead
of duplicating the cast.

Re: [PATCH v3 net-next 2/2] tcp: minor optimization around tcp_hdr() usage in tcp receive path

[Yafang Shao]

On Tue, May 29, 2018 at 12:36 AM, Eric Dumazet <edumazet@google.com> wrote:
> On Mon, May 28, 2018 at 8:36 AM Yafang Shao <laoar.shao@gmail.com> wrote:
>
>> This is additional to the commit ea1627c20c34 ("tcp: minor optimizations
> around tcp_hdr() usage").
>> At this point, skb->data is same with tcp_hdr() as tcp header has not
>> been pulled yet.
>
>> Cc: Eric Dumazet <edumazet@google.com>
>> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
>> ---
>>   net/ipv4/tcp_ipv4.c | 2 +-
>>   net/ipv6/tcp_ipv6.c | 2 +-
>>   2 files changed, 2 insertions(+), 2 deletions(-)
>
>> diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
>> index adbdb50..d179386 100644
>> --- a/net/ipv4/tcp_ipv4.c
>> +++ b/net/ipv4/tcp_ipv4.c
>> @@ -1486,7 +1486,7 @@ int tcp_v4_do_rcv(struct sock *sk, struct sk_buff
> *skb)
>>                                  sk->sk_rx_dst = NULL;
>>                          }
>>                  }
>> -               tcp_rcv_established(sk, skb, tcp_hdr(skb));
>> +               tcp_rcv_established(sk, skb, (const struct tcphdr
> *)skb->data);
>>                  return 0;
>>          }
>
>> diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
>> index 7d47c2b..1c633ff 100644
>> --- a/net/ipv6/tcp_ipv6.c
>> +++ b/net/ipv6/tcp_ipv6.c
>> @@ -1322,7 +1322,7 @@ static int tcp_v6_do_rcv(struct sock *sk, struct
> sk_buff *skb)
>>                          }
>>                  }
>
>> -               tcp_rcv_established(sk, skb, tcp_hdr(skb));
>> +               tcp_rcv_established(sk, skb, (const struct tcphdr
> *)skb->data);
>>                  if (opt_skb)
>>                          goto ipv6_pktoptions;
>>                  return 0;
>> --
>> 1.8.3.1
>
>
> I would rather remove the third parameter of tcp_rcv_established() instead
> of duplicating the cast.

OK.

And what about introducing a new helper tcp_hdr_fast() ?

/* use it when tcp header has not been pulled yet */
static inline struct tcphdr *tcp_hdr_fast(const struct sk_buff *skb)

{

        return (const struct tcphdr *)skb->data;

}


That could help us to use this optimized one instead of the original
one if possilbe.

Thanks
Yafang

Re: [PATCH v3 net-next 2/2] tcp: minor optimization around tcp_hdr() usage in tcp receive path

[Eric Dumazet]

On 05/28/2018 05:41 PM, Yafang Shao wrote:

> OK.
> 
> And what about introducing a new helper tcp_hdr_fast() ?
> 
> /* use it when tcp header has not been pulled yet */
> static inline struct tcphdr *tcp_hdr_fast(const struct sk_buff *skb)
> 
> {
> 
>         return (const struct tcphdr *)skb->data;
> 
> }
> 
> 
> That could help us to use this optimized one instead of the original
> one if possilbe.


I would rather not add such macro...

The call site needs to know what is going on, so having a macro like that would not help.

Re: [PATCH v3 net-next 2/2] tcp: minor optimization around tcp_hdr() usage in tcp receive path

[Yafang Shao]

On Tue, May 29, 2018 at 12:53 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
>
> On 05/28/2018 05:41 PM, Yafang Shao wrote:
>
>> OK.
>>
>> And what about introducing a new helper tcp_hdr_fast() ?
>>
>> /* use it when tcp header has not been pulled yet */
>> static inline struct tcphdr *tcp_hdr_fast(const struct sk_buff *skb)
>>
>> {
>>
>>         return (const struct tcphdr *)skb->data;
>>
>> }
>>
>>
>> That could help us to use this optimized one instead of the original
>> one if possilbe.
>
>
> I would rather not add such macro...
>
> The call site needs to know what is going on, so having a macro like that would not help.

OK