[PATCH] net/tcp: track all ipv4/tcp state transition in tcp_set_state

[PATCH] net/tcp: track all ipv4/tcp state transition in tcp_set_state

[Yafang Shao]

When I hooked the function tcp_set_state with kprobe to track the ipv4/tcp
state transistion, I found state transition from TCP_LISTEN to TCP_SYN_RECV
is missed.

I think it is better to use the helper to do state transition instead of
assigning the state to sk_state directly.
Then we can monitor the whole tcp lifespans with kprobe or ftrace easily.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
---
 net/ipv4/inet_connection_sock.c | 6 +++---
 net/ipv4/inet_hashtables.c      | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c
index c039c93..7c987c5 100644
--- a/net/ipv4/inet_connection_sock.c
+++ b/net/ipv4/inet_connection_sock.c
@@ -786,7 +786,7 @@ struct sock *inet_csk_clone_lock(const struct sock *sk,
 	if (newsk) {
 		struct inet_connection_sock *newicsk = inet_csk(newsk);
 
-		newsk->sk_state = TCP_SYN_RECV;
+		tcp_set_state(newsk, TCP_SYN_RECV);
 		newicsk->icsk_bind_hash = NULL;
 
 		inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
@@ -880,7 +880,7 @@ int inet_csk_listen_start(struct sock *sk, int backlog)
 	 * It is OK, because this socket enters to hash table only
 	 * after validation is complete.
 	 */
-	sk_state_store(sk, TCP_LISTEN);
+	tcp_set_state(sk, TCP_LISTEN);
 	if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 		inet->inet_sport = htons(inet->inet_num);
 
@@ -891,7 +891,7 @@ int inet_csk_listen_start(struct sock *sk, int backlog)
 			return 0;
 	}
 
-	sk->sk_state = TCP_CLOSE;
+	tcp_set_state(sk, TCP_CLOSE);
 	return err;
 }
 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index 597bb4c..bea8318 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -430,7 +430,7 @@ bool inet_ehash_nolisten(struct sock *sk, struct sock *osk)
 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 	} else {
 		percpu_counter_inc(sk->sk_prot->orphan_count);
-		sk->sk_state = TCP_CLOSE;
+		tcp_set_state(sk, TCP_CLOSE);
 		sock_set_flag(sk, SOCK_DEAD);
 		inet_csk_destroy_sock(sk);
 	}
-- 
1.8.3.1

Re: [PATCH] net/tcp: track all ipv4/tcp state transition in tcp_set_state

[David Miller]

From: Yafang Shao <laoar.shao@gmail.com>
Date: Tue,  7 Nov 2017 18:36:28 +0800

> When I hooked the function tcp_set_state with kprobe to track the ipv4/tcp
> state transistion, I found state transition from TCP_LISTEN to TCP_SYN_RECV
> is missed.
> 
> I think it is better to use the helper to do state transition instead of
> assigning the state to sk_state directly.
> Then we can monitor the whole tcp lifespans with kprobe or ftrace easily.
> 
> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>

This is really heavy handed and excessive for these cases.

They don't have to handle any of the issues dealt with in
tcp_set_state().

I would prefer if you made a special helper to net/tcp.h which did:

static inline void __tcp_set_state(struct sock *sk, int state)
{
	trace_tcp_set_state(sk, sk->sk_state, state);
	sk->sk_state = state;
}

Re: [PATCH] net/tcp: track all ipv4/tcp state transition in tcp_set_state

[Yafang Shao]

2017-11-08 14:51 GMT+08:00 David Miller <davem@davemloft.net>:
> From: Yafang Shao <laoar.shao@gmail.com>
> Date: Tue,  7 Nov 2017 18:36:28 +0800
>
>> When I hooked the function tcp_set_state with kprobe to track the ipv4/tcp
>> state transistion, I found state transition from TCP_LISTEN to TCP_SYN_RECV
>> is missed.
>>
>> I think it is better to use the helper to do state transition instead of
>> assigning the state to sk_state directly.
>> Then we can monitor the whole tcp lifespans with kprobe or ftrace easily.
>>
>> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
>
> This is really heavy handed and excessive for these cases.
>
> They don't have to handle any of the issues dealt with in
> tcp_set_state().
>
> I would prefer if you made a special helper to net/tcp.h which did:
>
> static inline void __tcp_set_state(struct sock *sk, int state)
> {
>         trace_tcp_set_state(sk, sk->sk_state, state);
>         sk->sk_state = state;
> }

Good idea!
I will try to implement it.

Thanks
Yafang