Merge branch 'tcp-minor-adjustments-for-low-pacing-rates'

Eric Dumazet says:

====================
tcp: minor adjustments for low pacing rates

After pacing horizon addition, we have to adjust how we arm rto
timer, otherwise we might freeze very low pacing rate flows.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 22 insertions, 25 deletions

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 1beed50522b1..43b87a8d4790 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1289,26 +1289,22 @@ static inline bool tcp_needs_internal_pacing(const struct sock *sk)
 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
 }
 
-/* Return in jiffies the delay before one skb is sent.
- * If @skb is NULL, we look at EDT for next packet being sent on the socket.
+/* Estimates in how many jiffies next packet for this flow can be sent.
+ * Scheduling a retransmit timer too early would be silly.
  */
-static inline unsigned long tcp_pacing_delay(const struct sock *sk,
-					     const struct sk_buff *skb)
+static inline unsigned long tcp_pacing_delay(const struct sock *sk)
 {
-	s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
+	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
 
-	pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
-
-	return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
+	return delay > 0 ? nsecs_to_jiffies(delay) : 0;
 }
 
 static inline void tcp_reset_xmit_timer(struct sock *sk,
 					const int what,
 					unsigned long when,
-					const unsigned long max_when,
-					const struct sk_buff *skb)
+					const unsigned long max_when)
 {
-	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
+	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
 				  max_when);
 }
 
@@ -1336,8 +1332,7 @@ static inline void tcp_check_probe_timer(struct sock *sk)
 {
 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
-				     tcp_probe0_base(sk), TCP_RTO_MAX,
-				     NULL);
+				     tcp_probe0_base(sk), TCP_RTO_MAX);
 }
 
 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index d68128a672ab..7d205b2a733c 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -3014,7 +3014,7 @@ void tcp_rearm_rto(struct sock *sk)
 			rto = usecs_to_jiffies(max_t(int, delta_us, 1));
 		}
 		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS, rto,
-				     TCP_RTO_MAX, tcp_rtx_queue_head(sk));
+				     TCP_RTO_MAX);
 	}
 }
 
@@ -3291,7 +3291,7 @@ static void tcp_ack_probe(struct sock *sk)
 		unsigned long when = tcp_probe0_when(sk, TCP_RTO_MAX);
 
 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
-				     when, TCP_RTO_MAX, NULL);
+				     when, TCP_RTO_MAX);
 	}
 }
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index c414aeb1efa9..a50e1990a845 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2593,8 +2593,7 @@ bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto)
 	if (rto_delta_us > 0)
 		timeout = min_t(u32, timeout, usecs_to_jiffies(rto_delta_us));
 
-	tcp_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout,
-			     TCP_RTO_MAX, NULL);
+	tcp_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout, TCP_RTO_MAX);
 	return true;
 }
 
@@ -3113,6 +3112,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 	const struct inet_connection_sock *icsk = inet_csk(sk);
 	struct sk_buff *skb, *rtx_head, *hole = NULL;
 	struct tcp_sock *tp = tcp_sk(sk);
+	bool rearm_timer = false;
 	u32 max_segs;
 	int mib_idx;
 
@@ -3135,7 +3135,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 
 		segs = tp->snd_cwnd - tcp_packets_in_flight(tp);
 		if (segs <= 0)
-			return;
+			break;
 		sacked = TCP_SKB_CB(skb)->sacked;
 		/* In case tcp_shift_skb_data() have aggregated large skbs,
 		 * we need to make sure not sending too bigs TSO packets
@@ -3160,10 +3160,10 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 			continue;
 
 		if (tcp_small_queue_check(sk, skb, 1))
-			return;
+			break;
 
 		if (tcp_retransmit_skb(sk, skb, segs))
-			return;
+			break;
 
 		NET_ADD_STATS(sock_net(sk), mib_idx, tcp_skb_pcount(skb));
 
@@ -3172,11 +3172,13 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 
 		if (skb == rtx_head &&
 		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
-			tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-					     inet_csk(sk)->icsk_rto,
-					     TCP_RTO_MAX,
-					     skb);
+			rearm_timer = true;
+
 	}
+	if (rearm_timer)
+		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+				     inet_csk(sk)->icsk_rto,
+				     TCP_RTO_MAX);
 }
 
 /* We allow to exceed memory limits for FIN packets to expedite
@@ -3907,7 +3909,7 @@ void tcp_send_probe0(struct sock *sk)
 		 */
 		timeout = TCP_RESOURCE_PROBE_INTERVAL;
 	}
-	tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, timeout, TCP_RTO_MAX, NULL);
+	tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, timeout, TCP_RTO_MAX);
 }
 
 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)