Merge: tcp: allow to control max RTO

CKI KWF Bot · CKI KWF Bot · commit 8e91b845a6f1 · 2025-10-09T15:40:49.000Z
MR: https://gitlab.com/redhat/centos-stream/src/kernel/centos-stream-9/-/merge_requests/7381 Adding support for controlling the TCP max RTO, through a sysctl or a sockopt. Tested using a packetdrill script (see the linked issue). JIRA: https://issues.redhat.com/browse/RHEL-115191 Signed-off-by: Antoine Tenart <atenart@redhat.com> Approved-by: Paolo Abeni <pabeni@redhat.com> Approved-by: Florian Westphal <fwestpha@redhat.com> Approved-by: CKI KWF Bot <cki-ci-bot+kwf-gitlab-com@redhat.com> Merged-by: CKI GitLab Kmaint Pipeline Bot <26919896-cki-kmaint-pipeline-bot@users.noreply.gitlab.com>
diff --git a/Documentation/networking/ip-sysctl.rst b/Documentation/networking/ip-sysctl.rst
@@ -697,6 +697,8 @@ tcp_retries2 - INTEGER
 	seconds and is a lower bound for the effective timeout.
 	TCP will effectively time out at the first RTO which exceeds the
 	hypothetical timeout.
+	If tcp_rto_max_ms is decreased, it is recommended to also
+	change tcp_retries2.
 
 	RFC 1122 recommends at least 100 seconds for the timeout,
 	which corresponds to a value of at least 8.
@@ -1041,6 +1043,17 @@ tcp_challenge_ack_limit - INTEGER
 	in RFC 5961 (Improving TCP's Robustness to Blind In-Window Attacks)
 	Default: 1000
 
+tcp_rto_max_ms - INTEGER
+	Maximal TCP retransmission timeout (in ms).
+	Note that TCP_RTO_MAX_MS socket option has higher precedence.
+
+	When changing tcp_rto_max_ms, it is important to understand
+	that tcp_retries2 might need a change.
+
+	Possible Values: 1000 - 120,000
+
+	Default: 120,000
+
 UDP variables
 =============
 
diff --git a/Documentation/networking/net_cachelines/inet_connection_sock.rst b/Documentation/networking/net_cachelines/inet_connection_sock.rst
@@ -16,6 +16,7 @@ struct_timer_list                   icsk_retransmit_timer  read_mostly         -
 struct_timer_list                   icsk_delack_timer      read_mostly         -                   inet_csk_reset_xmit_timer,tcp_connect
 u32                                 icsk_rto               read_write          -                   tcp_cwnd_validate,tcp_schedule_loss_probe,tcp_connect_init,tcp_connect,tcp_write_xmit,tcp_push_one
 u32                                 icsk_rto_min           -                   -                   
+u32                                 icsk_rto_max           read_mostly         -                   tcp_reset_xmit_timer
 u32                                 icsk_delack_max        -                   -                   
 u32                                 icsk_pmtu_cookie       read_write          -                   tcp_sync_mss,tcp_current_mss,tcp_send_syn_data,tcp_connect_init,tcp_connect
 struct_tcp_congestion_ops           icsk_ca_ops            read_write          -                   tcp_cwnd_validate,tcp_tso_segs,tcp_ca_dst_init,tcp_connect_init,tcp_connect,tcp_write_xmit
diff --git a/Documentation/networking/net_cachelines/netns_ipv4_sysctl.rst b/Documentation/networking/net_cachelines/netns_ipv4_sysctl.rst
@@ -84,6 +84,7 @@ u8                              sysctl_tcp_sack                              -
 u8                              sysctl_tcp_window_scaling                    -                   -                   tcp_syn_options,tcp_parse_options
 u8                              sysctl_tcp_timestamps                                                                
 u8                              sysctl_tcp_early_retrans                     read_mostly         -                   tcp_schedule_loss_probe(tcp_write_xmit)
+u32                             sysctl_tcp_rto_max_ms                        -                   -
 u8                              sysctl_tcp_recovery                          -                   -                   tcp_fastretrans_alert
 u8                              sysctl_tcp_thin_linear_timeouts              -                   -                   tcp_retrans_timer(on_thin_streams)
 u8                              sysctl_tcp_slow_start_after_idle             -                   -                   unlikely(tcp_cwnd_validate-network-not-starved)
diff --git a/include/net/inet_connection_sock.h b/include/net/inet_connection_sock.h
@@ -88,6 +88,7 @@ struct inet_connection_sock {
  	struct timer_list	  icsk_delack_timer;
 	__u32			  icsk_rto;
 	__u32                     icsk_rto_min;
+	u32			  icsk_rto_max;
 	__u32                     icsk_delack_max;
 	__u32			  icsk_pmtu_cookie;
 	const struct tcp_congestion_ops *icsk_ca_ops;
diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
@@ -171,6 +171,7 @@ struct netns_ipv4 {
 	u8 sysctl_tcp_sack;
 	u8 sysctl_tcp_window_scaling;
 	u8 sysctl_tcp_timestamps;
+	int sysctl_tcp_rto_max_ms;
 	u8 sysctl_tcp_recovery;
 	u8 sysctl_tcp_thin_linear_timeouts;
 	u8 sysctl_tcp_slow_start_after_idle;
diff --git a/include/net/tcp.h b/include/net/tcp.h
@@ -141,8 +141,9 @@ void tcp_time_wait(struct sock *sk, int state, int timeo);
 #define TCP_DELACK_MIN	4U
 #define TCP_ATO_MIN	4U
 #endif
-#define TCP_RTO_MAX	((unsigned)(120*HZ))
-#define TCP_RTO_MIN	((unsigned)(HZ/5))
+#define TCP_RTO_MAX_SEC 120
+#define TCP_RTO_MAX	((unsigned)(TCP_RTO_MAX_SEC * HZ))
+#define TCP_RTO_MIN	((unsigned)(HZ / 5))
 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
 
 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */
@@ -711,10 +712,14 @@ int tcp_mtu_to_mss(struct sock *sk, int pmtu);
 int tcp_mss_to_mtu(struct sock *sk, int mss);
 void tcp_mtup_init(struct sock *sk);
 
+static inline unsigned int tcp_rto_max(const struct sock *sk)
+{
+	return READ_ONCE(inet_csk(sk)->icsk_rto_max);
+}
+
 static inline void tcp_bound_rto(const struct sock *sk)
 {
-	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
-		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
+	inet_csk(sk)->icsk_rto = min(inet_csk(sk)->icsk_rto, tcp_rto_max(sk));
 }
 
 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
@@ -1355,10 +1360,12 @@ static inline unsigned long tcp_pacing_delay(const struct sock *sk)
 static inline void tcp_reset_xmit_timer(struct sock *sk,
 					const int what,
 					unsigned long when,
-					const unsigned long max_when)
+					bool pace_delay)
 {
-	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
-				  max_when);
+	if (pace_delay)
+		when += tcp_pacing_delay(sk);
+	inet_csk_reset_xmit_timer(sk, what, when,
+				  tcp_rto_max(sk));
 }
 
 /* Something is really bad, we could not queue an additional packet,
@@ -1387,7 +1394,7 @@ 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);
+				     tcp_probe0_base(sk), true);
 }
 
 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
@@ -129,6 +129,7 @@ enum {
 
 #define TCP_TX_DELAY		37	/* delay outgoing packets by XX usec */
 
+#define TCP_RTO_MAX_MS		44	/* max rto time in ms */
 
 #define TCP_REPAIR_ON		1
 #define TCP_REPAIR_OFF		0
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
@@ -41,6 +41,7 @@ static int tcp_adv_win_scale_max = 31;
 static int tcp_app_win_max = 31;
 static int tcp_min_snd_mss_min = TCP_MIN_SND_MSS;
 static int tcp_min_snd_mss_max = 65535;
+static int tcp_rto_max_max = TCP_RTO_MAX_SEC * MSEC_PER_SEC;
 static int ip_privileged_port_min;
 static int ip_privileged_port_max = 65535;
 static int ip_ttl_min = 1;
@@ -1425,6 +1426,15 @@ static struct ctl_table ipv4_net_table[] = {
 		.extra1		= SYSCTL_ZERO,
 		.extra2		= SYSCTL_ONE,
 	},
+	{
+		.procname	= "tcp_rto_max_ms",
+		.data		= &init_net.ipv4.sysctl_tcp_rto_max_ms,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= SYSCTL_ONE_THOUSAND,
+		.extra2		= &tcp_rto_max_max,
+	},
 	{ }
 };
 
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
@@ -417,6 +417,7 @@ void tcp_init_sock(struct sock *sk)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
+	int rto_max_ms;
 
 	tp->out_of_order_queue = RB_ROOT;
 	sk->tcp_rtx_queue = RB_ROOT;
@@ -425,7 +426,12 @@ void tcp_init_sock(struct sock *sk)
 	INIT_LIST_HEAD(&tp->tsorted_sent_queue);
 
 	icsk->icsk_rto = TCP_TIMEOUT_INIT;
+
+	rto_max_ms = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rto_max_ms);
+	icsk->icsk_rto_max = msecs_to_jiffies(rto_max_ms);
+
 	icsk->icsk_rto_min = TCP_RTO_MIN;
+
 	icsk->icsk_delack_max = TCP_DELACK_MAX;
 	tp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
 	minmax_reset(&tp->rtt_min, tcp_jiffies32, ~0U);
@@ -3669,6 +3675,11 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,
 			   secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ,
 					   TCP_RTO_MAX / HZ));
 		return 0;
+	case TCP_RTO_MAX_MS:
+		if (val < MSEC_PER_SEC || val > TCP_RTO_MAX_SEC * MSEC_PER_SEC)
+			return -EINVAL;
+		WRITE_ONCE(inet_csk(sk)->icsk_rto_max, msecs_to_jiffies(val));
+		return 0;
 	}
 
 	sockopt_lock_sock(sk);
@@ -4429,6 +4440,9 @@ int do_tcp_getsockopt(struct sock *sk, int level,
 		return err;
 	}
 #endif
+	case TCP_RTO_MAX_MS:
+		val = jiffies_to_msecs(tcp_rto_max(sk));
+		break;
 	default:
 		return -ENOPROTOOPT;
 	}
diff --git a/net/ipv4/tcp_fastopen.c b/net/ipv4/tcp_fastopen.c
@@ -289,8 +289,8 @@ static struct sock *tcp_fastopen_create_child(struct sock *sk,
 	 * The request socket is not added to the ehash
 	 * because it's been added to the accept queue directly.
 	 */
-	inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
-				  TCP_TIMEOUT_INIT, TCP_RTO_MAX);
+	tcp_reset_xmit_timer(child, ICSK_TIME_RETRANS,
+			     TCP_TIMEOUT_INIT, false);
 
 	refcount_set(&req->rsk_refcnt, 2);
 
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
@@ -2205,8 +2205,7 @@ static bool tcp_check_sack_reneging(struct sock *sk, int flag)
 		unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4),
 					  msecs_to_jiffies(10));
 
-		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-					  delay, TCP_RTO_MAX);
+		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS, delay, false);
 		return true;
 	}
 	return false;
@@ -3231,8 +3230,7 @@ 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_reset_xmit_timer(sk, ICSK_TIME_RETRANS, rto, true);
 	}
 }
 
@@ -3510,10 +3508,10 @@ static void tcp_ack_probe(struct sock *sk)
 		 * This function is not for random using!
 		 */
 	} else {
-		unsigned long when = tcp_probe0_when(sk, TCP_RTO_MAX);
+		unsigned long when = tcp_probe0_when(sk, tcp_rto_max(sk));
 
 		when = tcp_clamp_probe0_to_user_timeout(sk, when);
-		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, when, TCP_RTO_MAX);
+		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, when, true);
 	}
 }
 
@@ -6255,9 +6253,8 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
 		    after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) {
 			/* Previous FIN/ACK or RST/ACK might be ignored. */
 			if (icsk->icsk_retransmits == 0)
-				inet_csk_reset_xmit_timer(sk,
-						ICSK_TIME_RETRANS,
-						TCP_TIMEOUT_MIN, TCP_RTO_MAX);
+				tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+						     TCP_TIMEOUT_MIN, false);
 			SKB_DR_SET(reason, TCP_INVALID_ACK_SEQUENCE);
 			goto reset_and_undo;
 		}
@@ -6371,8 +6368,8 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
 			 */
 			inet_csk_schedule_ack(sk);
 			tcp_enter_quickack_mode(sk, TCP_MAX_QUICKACKS);
-			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
-						  TCP_DELACK_MAX, TCP_RTO_MAX);
+			tcp_reset_xmit_timer(sk, ICSK_TIME_DACK,
+					     TCP_DELACK_MAX, false);
 			goto consume;
 		}
 		tcp_send_ack(sk);
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
@@ -439,15 +439,14 @@ void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
 
 	icsk->icsk_backoff--;
 	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
-	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
+	icsk->icsk_rto = inet_csk_rto_backoff(icsk, tcp_rto_max(sk));
 
 	tcp_mstamp_refresh(tp);
 	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
 	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
 
 	if (remaining > 0) {
-		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-					  remaining, TCP_RTO_MAX);
+		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS, remaining, false);
 	} else {
 		/* RTO revert clocked out retransmission.
 		 * Will retransmit now.
@@ -3223,6 +3222,8 @@ static int __net_init tcp_sk_init(struct net *net)
 
 	net->ipv4.sysctl_tcp_shrink_window = 0;
 
+	net->ipv4.sysctl_tcp_rto_max_ms = TCP_RTO_MAX_SEC * MSEC_PER_SEC;
+
 	return 0;
 fail:
 	tcp_sk_exit(net);
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
@@ -2787,7 +2787,7 @@ 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);
+	tcp_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout, true);
 	return true;
 }
 
@@ -3424,8 +3424,7 @@ void tcp_xmit_retransmit_queue(struct sock *sk)
 	}
 	if (rearm_timer)
 		tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-				     inet_csk(sk)->icsk_rto,
-				     TCP_RTO_MAX);
+				     inet_csk(sk)->icsk_rto, true);
 }
 
 /* We allow to exceed memory limits for FIN packets to expedite
@@ -3946,8 +3945,8 @@ int tcp_connect(struct sock *sk)
 	TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
 
 	/* Timer for repeating the SYN until an answer. */
-	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-				  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
+	tcp_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+			     inet_csk(sk)->icsk_rto, false);
 	return 0;
 }
 EXPORT_SYMBOL(tcp_connect);
@@ -4028,11 +4027,11 @@ void __tcp_send_ack(struct sock *sk, u32 rcv_nxt)
 		unsigned long delay;
 
 		delay = TCP_DELACK_MAX << icsk->icsk_ack.retry;
-		if (delay < TCP_RTO_MAX)
+		if (delay < tcp_rto_max(sk))
 			icsk->icsk_ack.retry++;
 		inet_csk_schedule_ack(sk);
 		icsk->icsk_ack.ato = TCP_ATO_MIN;
-		inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, delay, TCP_RTO_MAX);
+		tcp_reset_xmit_timer(sk, ICSK_TIME_DACK, delay, false);
 		return;
 	}
 
@@ -4168,7 +4167,7 @@ void tcp_send_probe0(struct sock *sk)
 	if (err <= 0) {
 		if (icsk->icsk_backoff < READ_ONCE(net->ipv4.sysctl_tcp_retries2))
 			icsk->icsk_backoff++;
-		timeout = tcp_probe0_when(sk, TCP_RTO_MAX);
+		timeout = tcp_probe0_when(sk, tcp_rto_max(sk));
 	} else {
 		/* If packet was not sent due to local congestion,
 		 * Let senders fight for local resources conservatively.
@@ -4177,7 +4176,7 @@ void tcp_send_probe0(struct sock *sk)
 	}
 
 	timeout = tcp_clamp_probe0_to_user_timeout(sk, timeout);
-	tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, timeout, TCP_RTO_MAX);
+	tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, timeout, true);
 }
 
 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
