net: tcp: add sysctl_tcp_wnd_shrink

Add the 'sysctl_tcp_wnd_shrink' to control the enable/disable of TCP
window shrink. By default, it is disabled.

Signed-off-by: Menglong Dong <imagedong@tencent.com>

include/net/tcp.h

@@ -255,6 +255,7 @@ extern int sysctl_tcp_synack_rto_interval;
 extern int sysctl_tcp_rto_min;
 extern int sysctl_tcp_rto_max;
 extern int sysctl_tcp_proc_sched;
+extern int sysctl_tcp_wnd_shrink;
 
 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
@@ -1853,6 +1854,9 @@ static inline bool tcp_probe0_needed(const struct sock *sk)
 	if (__tcp_probe0_needed(sk))
 		return true;
 
+	if (!sysctl_tcp_wnd_shrink)
+		return false;
+
 	/* for the window shrink case */
 	return tcp_rtx_overflow(sk);
 }

net/ipv4/sysctl_net_ipv4.c

@@ -713,6 +713,15 @@ static struct ctl_table ipv4_table[] = {
 		.proc_handler   = proc_dointvec_minmax,
 		.extra1         = &four
 	},
+	{
+		.procname       = "tcp_wnd_shrink",
+		.data           = &sysctl_tcp_wnd_shrink,
+		.maxlen         = sizeof(int),
+		.mode           = 0644,
+		.proc_handler   = proc_dointvec_minmax,
+		.extra1         = SYSCTL_ZERO,
+		.extra2         = SYSCTL_ONE
+	},
 	{ }
 };
 

net/ipv4/tcp.c

@@ -288,6 +288,11 @@ EXPORT_SYMBOL(sysctl_tcp_mem);
 atomic_long_t tcp_memory_allocated;	/* Current allocated memory. */
 EXPORT_SYMBOL(tcp_memory_allocated);
 
+#ifdef CONFIG_TCP_WND_SHRINK
+int sysctl_tcp_wnd_shrink;
+EXPORT_SYMBOL(sysctl_tcp_wnd_shrink);
+#endif
+
 #if IS_ENABLED(CONFIG_SMC)
 DEFINE_STATIC_KEY_FALSE(tcp_have_smc);
 EXPORT_SYMBOL(tcp_have_smc);

net/ipv4/tcp_input.c

@@ -3035,7 +3035,8 @@ static void tcp_set_xmit_timer(struct sock *sk)
 	 * not needed to schedule the RTO. The normal probe0 can't reach
 	 * here, so it must be window-shrink probe0 case here.
 	 */
-	if (icsk->icsk_pending == ICSK_TIME_PROBE0)
+	if (sysctl_tcp_wnd_shrink &&
+	    icsk->icsk_pending == ICSK_TIME_PROBE0)
 		return;
 #endif
 
@@ -3327,6 +3328,9 @@ static void tcp_ack_probe_shrink(struct sock *sk)
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	unsigned long when;
 
+	if (!sysctl_tcp_wnd_shrink)
+		return;
+
 	if (tcp_rtx_overflow(sk)) {
 		when = tcp_probe0_when(sk, TCP_RTO_MAX);
 
@@ -4871,6 +4875,14 @@ queue_and_out:
 			sk_forced_mem_schedule(sk, skb->truesize);
 		else if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {
 #ifdef CONFIG_TCP_WND_SHRINK
+			if (sysctl_tcp_wnd_shrink)
+				goto do_wnd_shrink;
+#endif
+			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVQDROP);
+			sk->sk_data_ready(sk);
+			goto drop;
+#ifdef CONFIG_TCP_WND_SHRINK
+do_wnd_shrink:
 			if (sock_flag(sk, SOCK_NO_MEM)) {
 				NET_INC_STATS(sock_net(sk),
 					      LINUX_MIB_TCPRCVQDROP);
@@ -4879,10 +4891,6 @@ queue_and_out:
 			}
 			sk_forced_mem_schedule(sk, skb->truesize);
 			sock_set_flag(sk, SOCK_NO_MEM);
-#else
-			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVQDROP);
-			sk->sk_data_ready(sk);
-			goto drop;
 #endif
 		}
 
@@ -4924,9 +4932,7 @@ out_of_window:
 		NET_INC_DROPSTATS(sock_net(sk), LINUX_MIB_TCPOOWDROP);
 		tcp_enter_quickack_mode(sk, TCP_MAX_QUICKACKS);
 		inet_csk_schedule_ack(sk);
-#ifndef CONFIG_TCP_WND_SHRINK
 drop:
-#endif
 		tcp_drop(sk, skb);
 		return;
 	}

net/ipv4/tcp_output.c

@@ -302,7 +302,7 @@ static u16 tcp_select_window(struct sock *sk)
 	}
 
 #ifdef CONFIG_TCP_WND_SHRINK
-	if (sock_flag(sk, SOCK_NO_MEM)) {
+	if (sysctl_tcp_wnd_shrink && sock_flag(sk, SOCK_NO_MEM)) {
 		if (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 2))
 			sock_reset_flag(sk, SOCK_NO_MEM);
 		else