tcp: add a sysctl to config the tcp_default_init_rwnd

The default initial rwnd is hardcoded to 10.

Now we allow it to be controlled via
  /proc/sys/net/ipv4/tcp_default_init_rwnd
which limits the values from 3 to 100

This is somewhat needed because ipv6 routes are
autoconfigured by the kernel.

See "An Argument for Increasing TCP's Initial Congestion Window"
in https://developers.google.com/speed/articles/tcp_initcwnd_paper.pdf

Change-Id: I386b2a9d62de0ebe05c1ebe1b4bd91b314af5c54
Signed-off-by: JP Abgrall <jpa@google.com>

4 files changed, 30 insertions, 7 deletions

diff --git a/include/net/tcp.h b/include/net/tcp.h
index dddd6137dade..2dc37818a91e 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -290,6 +290,7 @@ extern int sysctl_tcp_thin_dupack;
 extern int sysctl_tcp_early_retrans;
 extern int sysctl_tcp_limit_output_bytes;
 extern int sysctl_tcp_challenge_ack_limit;
+extern int sysctl_tcp_default_init_rwnd;
 
 extern atomic_long_t tcp_memory_allocated;
 
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 8df0d620b202..e8315d3387f5 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -139,6 +139,21 @@ static int ipv4_ping_group_range(ctl_table *table, int write,
 	return ret;
 }
 
+/* Validate changes from /proc interface. */
+static int proc_tcp_default_init_rwnd(ctl_table *ctl, int write,
+				      void __user *buffer,
+				      size_t *lenp, loff_t *ppos)
+{
+	int old_value = *(int *)ctl->data;
+	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
+	int new_value = *(int *)ctl->data;
+
+	if (write && ret == 0 && (new_value < 3 || new_value > 100))
+		*(int *)ctl->data = old_value;
+
+	return ret;
+}
+
 static int proc_tcp_congestion_control(ctl_table *ctl, int write,
 				       void __user *buffer, size_t *lenp, loff_t *ppos)
 {
@@ -736,7 +751,7 @@ static struct ctl_table ipv4_table[] = {
 		.mode           = 0644,
 		.proc_handler   = proc_dointvec
 	},
-        {
+	{
 		.procname       = "tcp_thin_dupack",
 		.data           = &sysctl_tcp_thin_dupack,
 		.maxlen         = sizeof(int),
@@ -753,6 +768,13 @@ static struct ctl_table ipv4_table[] = {
 		.extra2		= &four,
 	},
 	{
+		.procname       = "tcp_default_init_rwnd",
+		.data           = &sysctl_tcp_default_init_rwnd,
+		.maxlen         = sizeof(int),
+		.mode           = 0644,
+		.proc_handler   = proc_tcp_default_init_rwnd
+	},
+	{
 		.procname	= "udp_mem",
 		.data		= &sysctl_udp_mem,
 		.maxlen		= sizeof(sysctl_udp_mem),
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index a13efa657287..7a3d602e19b3 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -98,6 +98,7 @@ int sysctl_tcp_thin_dupack __read_mostly;
 
 int sysctl_tcp_moderate_rcvbuf __read_mostly = 1;
 int sysctl_tcp_early_retrans __read_mostly = 3;
+int sysctl_tcp_default_init_rwnd __read_mostly = TCP_DEFAULT_INIT_RCVWND;
 
 #define FLAG_DATA		0x01 /* Incoming frame contained data.		*/
 #define FLAG_WIN_UPDATE		0x02 /* Incoming ACK was a window update.	*/
@@ -351,14 +352,14 @@ static void tcp_grow_window(struct sock *sk, const struct sk_buff *skb)
 static void tcp_fixup_rcvbuf(struct sock *sk)
 {
 	u32 mss = tcp_sk(sk)->advmss;
-	u32 icwnd = TCP_DEFAULT_INIT_RCVWND;
+	u32 icwnd = sysctl_tcp_default_init_rwnd;
 	int rcvmem;
 
 	/* Limit to 10 segments if mss <= 1460,
 	 * or 14600/mss segments, with a minimum of two segments.
 	 */
 	if (mss > 1460)
-		icwnd = max_t(u32, (1460 * TCP_DEFAULT_INIT_RCVWND) / mss, 2);
+		icwnd = max_t(u32, (1460 * icwnd) / mss, 2);
 
 	rcvmem = SKB_TRUESIZE(mss + MAX_TCP_HEADER);
 	while (tcp_win_from_space(rcvmem) < mss)
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index ec335fabd5cc..084eac6a43b9 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -231,14 +231,13 @@ void tcp_select_initial_window(int __space, __u32 mss,
 	}
 
 	/* Set initial window to a value enough for senders starting with
-	 * initial congestion window of TCP_DEFAULT_INIT_RCVWND. Place
+	 * initial congestion window of sysctl_tcp_default_init_rwnd. Place
 	 * a limit on the initial window when mss is larger than 1460.
 	 */
 	if (mss > (1 << *rcv_wscale)) {
-		int init_cwnd = TCP_DEFAULT_INIT_RCVWND;
+		int init_cwnd = sysctl_tcp_default_init_rwnd;
 		if (mss > 1460)
-			init_cwnd =
-			max_t(u32, (1460 * TCP_DEFAULT_INIT_RCVWND) / mss, 2);
+			init_cwnd = max_t(u32, (1460 * init_cwnd) / mss, 2);
 		/* when initializing use the value from init_rcv_wnd
 		 * rather than the default from above
 		 */