[PATCH net-next] ipv4: tcp: fix ACK/RST sent with a transmit delay

[PATCH net-next] ipv4: tcp: fix ACK/RST sent with a transmit delay

[Eric Dumazet]

If we want to set a EDT time for the skb we want to send
via ip_send_unicast_reply(), we have to pass a new parameter
and initialize ipc.sockc.transmit_time with it.

This fixes the EDT time for ACK/RST packets sent on behalf of
a TIME_WAIT socket.

Fixes: a842fe1425cb ("tcp: add optional per socket transmit delay")
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/ip.h     |  2 +-
 include/net/tcp.h    |  9 ++++++---
 net/ipv4/ip_output.c |  3 ++-
 net/ipv4/tcp_ipv4.c  | 14 +++++++++-----
 net/ipv6/tcp_ipv6.c  |  2 +-
 5 files changed, 19 insertions(+), 11 deletions(-)

diff --git a/include/net/ip.h b/include/net/ip.h
index 6dbf88ea07f1a99a1f9e58d7e5c22e3b5da5df96..29d89de39822288f7b61d20c96374cfed5f6b8fb 100644
--- a/include/net/ip.h
+++ b/include/net/ip.h
@@ -279,7 +279,7 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
 			   const struct ip_options *sopt,
 			   __be32 daddr, __be32 saddr,
 			   const struct ip_reply_arg *arg,
-			   unsigned int len);
+			   unsigned int len, u64 transmit_time);
 
 #define IP_INC_STATS(net, field)	SNMP_INC_STATS64((net)->mib.ip_statistics, field)
 #define __IP_INC_STATS(net, field)	__SNMP_INC_STATS64((net)->mib.ip_statistics, field)
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 49a178b8d5b2fafcbbeda53506ea38e3adb0f9dd..96e0e53ff4408d13d4b062f0f81f94552660b68e 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -2240,15 +2240,18 @@ static inline void tcp_add_tx_delay(struct sk_buff *skb,
 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
 }
 
-static inline void tcp_set_tx_time(struct sk_buff *skb,
-				   const struct sock *sk)
+/* Compute Earliest Departure Time for some control packets
+ * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
+ */
+static inline u64 tcp_transmit_time(const struct sock *sk)
 {
 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
 
-		skb->skb_mstamp_ns = tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
+		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
 	}
+	return 0;
 }
 
 #endif	/* _TCP_H */
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index f5636ab0b9c32fa47eaa5417fec7cdc47e4dabc2..e0ac39072a9c8fc755d2904d5f61ae49ac3c416f 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1632,7 +1632,7 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
 			   const struct ip_options *sopt,
 			   __be32 daddr, __be32 saddr,
 			   const struct ip_reply_arg *arg,
-			   unsigned int len)
+			   unsigned int len, u64 transmit_time)
 {
 	struct ip_options_data replyopts;
 	struct ipcm_cookie ipc;
@@ -1648,6 +1648,7 @@ void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
 
 	ipcm_init(&ipc);
 	ipc.addr = daddr;
+	ipc.sockc.transmit_time = transmit_time;
 
 	if (replyopts.opt.opt.optlen) {
 		ipc.opt = &replyopts.opt;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 1b7e9e1fbd3be3670a7fe9da4978f7a2e0959f58..633e8244ed5b6abe4964f8f078961dcbef17a296 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -662,8 +662,9 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
 	int genhash;
 	struct sock *sk1 = NULL;
 #endif
-	struct net *net;
+	u64 transmit_time = 0;
 	struct sock *ctl_sk;
+	struct net *net;
 
 	/* Never send a reset in response to a reset. */
 	if (th->rst)
@@ -770,12 +771,13 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
 	if (sk) {
 		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 				   inet_twsk(sk)->tw_mark : sk->sk_mark;
-		tcp_set_tx_time(skb, sk);
+		transmit_time = tcp_transmit_time(sk);
 	}
 	ip_send_unicast_reply(ctl_sk,
 			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
-			      &arg, arg.iov[0].iov_len);
+			      &arg, arg.iov[0].iov_len,
+			      transmit_time);
 
 	ctl_sk->sk_mark = 0;
 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
@@ -810,6 +812,7 @@ static void tcp_v4_send_ack(const struct sock *sk,
 	struct net *net = sock_net(sk);
 	struct ip_reply_arg arg;
 	struct sock *ctl_sk;
+	u64 transmit_time;
 
 	memset(&rep.th, 0, sizeof(struct tcphdr));
 	memset(&arg, 0, sizeof(arg));
@@ -863,11 +866,12 @@ static void tcp_v4_send_ack(const struct sock *sk,
 	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 			   inet_twsk(sk)->tw_mark : sk->sk_mark;
-	tcp_set_tx_time(skb, sk);
+	transmit_time = tcp_transmit_time(sk);
 	ip_send_unicast_reply(ctl_sk,
 			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
-			      &arg, arg.iov[0].iov_len);
+			      &arg, arg.iov[0].iov_len,
+			      transmit_time);
 
 	ctl_sk->sk_mark = 0;
 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 5606b2131b653572f8ef6cdb6af5a118d5f4934d..408d9ec2697154e840a26675765e8a9c1636ada4 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -892,7 +892,7 @@ static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32
 		} else {
 			mark = sk->sk_mark;
 		}
-		tcp_set_tx_time(buff, sk);
+		buff->tstamp = tcp_transmit_time(sk);
 	}
 	fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
 	fl6.fl6_dport = t1->dest;
-- 
2.22.0.rc2.383.gf4fbbf30c2-goog

Re: [PATCH net-next] ipv4: tcp: fix ACK/RST sent with a transmit delay

[David Miller]

From: Eric Dumazet <edumazet@google.com>
Date: Thu, 13 Jun 2019 21:22:35 -0700

> If we want to set a EDT time for the skb we want to send
> via ip_send_unicast_reply(), we have to pass a new parameter
> and initialize ipc.sockc.transmit_time with it.
> 
> This fixes the EDT time for ACK/RST packets sent on behalf of
> a TIME_WAIT socket.
> 
> Fixes: a842fe1425cb ("tcp: add optional per socket transmit delay")
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied.