[PATCH net-next] tcp: fix RFS vs lockless listeners

[PATCH net-next] tcp: fix RFS vs lockless listeners

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

Before recent TCP listener patches, we were updating listener
sk->sk_rxhash before the cloning of master socket.

children sk_rxhash was therefore correct after the normal 3WHS.

But with lockless listener, we no longer dirty/change listener sk_rxhash
as it would be racy.

We need to correctly update the child sk_rxhash, otherwise first data
packet wont hit correct cpu if RFS is used.

Fixes: 079096f103fa ("tcp/dccp: install syn_recv requests into ehash table")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Willem de Bruijn <willemb@google.com>
Cc: Tom Herbert <tom@herbertland.com>
---
 net/ipv4/syncookies.c    |    1 +
 net/ipv4/tcp_minisocks.c |    1 +
 2 files changed, 2 insertions(+)

diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index 8113c30ccf96..2dbb11331f6c 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -225,6 +225,7 @@ struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 	child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
 	if (child) {
 		atomic_set(&req->rsk_refcnt, 1);
+		sock_rps_save_rxhash(child, skb);
 		inet_csk_reqsk_queue_add(sk, req, child);
 	} else {
 		reqsk_free(req);
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 9adf1e2c3170..1079e6ad77fe 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -768,6 +768,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 	if (!child)
 		goto listen_overflow;
 
+	sock_rps_save_rxhash(child, skb);
 	tcp_synack_rtt_meas(child, req);
 	inet_csk_reqsk_queue_drop(sk, req);
 	inet_csk_reqsk_queue_add(sk, req, child);

Re: [PATCH net-next] tcp: fix RFS vs lockless listeners

[Tom Herbert]

On Thu, Oct 8, 2015 at 11:16 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> From: Eric Dumazet <edumazet@google.com>
>
> Before recent TCP listener patches, we were updating listener
> sk->sk_rxhash before the cloning of master socket.
>
> children sk_rxhash was therefore correct after the normal 3WHS.
>
> But with lockless listener, we no longer dirty/change listener sk_rxhash
> as it would be racy.
>
> We need to correctly update the child sk_rxhash, otherwise first data
> packet wont hit correct cpu if RFS is used.
>
> Fixes: 079096f103fa ("tcp/dccp: install syn_recv requests into ehash table")
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Reported-by: Willem de Bruijn <willemb@google.com>
> Cc: Tom Herbert <tom@herbertland.com>
> ---
>  net/ipv4/syncookies.c    |    1 +
>  net/ipv4/tcp_minisocks.c |    1 +
>  2 files changed, 2 insertions(+)
>
> diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
> index 8113c30ccf96..2dbb11331f6c 100644
> --- a/net/ipv4/syncookies.c
> +++ b/net/ipv4/syncookies.c
> @@ -225,6 +225,7 @@ struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
>         child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
>         if (child) {
>                 atomic_set(&req->rsk_refcnt, 1);
> +               sock_rps_save_rxhash(child, skb);
>                 inet_csk_reqsk_queue_add(sk, req, child);
>         } else {
>                 reqsk_free(req);
> diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
> index 9adf1e2c3170..1079e6ad77fe 100644
> --- a/net/ipv4/tcp_minisocks.c
> +++ b/net/ipv4/tcp_minisocks.c
> @@ -768,6 +768,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
>         if (!child)
>                 goto listen_overflow;
>
> +       sock_rps_save_rxhash(child, skb);
>         tcp_synack_rtt_meas(child, req);
>         inet_csk_reqsk_queue_drop(sk, req);
>         inet_csk_reqsk_queue_add(sk, req, child);
>
>
Acked-by: Tom Herbert <tom@herbertland.com>

Re: [PATCH net-next] tcp: fix RFS vs lockless listeners

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 08 Oct 2015 11:16:48 -0700

> From: Eric Dumazet <edumazet@google.com>
> 
> Before recent TCP listener patches, we were updating listener
> sk->sk_rxhash before the cloning of master socket.
> 
> children sk_rxhash was therefore correct after the normal 3WHS.
> 
> But with lockless listener, we no longer dirty/change listener sk_rxhash
> as it would be racy.
> 
> We need to correctly update the child sk_rxhash, otherwise first data
> packet wont hit correct cpu if RFS is used.
> 
> Fixes: 079096f103fa ("tcp/dccp: install syn_recv requests into ehash table")
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Reported-by: Willem de Bruijn <willemb@google.com>
> Cc: Tom Herbert <tom@herbertland.com>

Applied, thanks Eric.