From: Anton Arapov <aarapov@redhat.com>
Date: Tue, 25 Sep 2007 10:36:03 +0200
Subject: [net] kernel needs to support TCP_RTO_MIN
Message-id: h8d4w7m9ak.fsf@pepelac.englab.brq.redhat.com
O-Subject: [RHEL5.2 PATCH] BZ303011: kernel needs to support TCP_RTO_MIN
Bugzilla: 303011

BZ#303011:
  https://bugzilla.redhat.com/show_bug.cgi?id=303011

Description:
  Customer application for KDDI requires a kernel parameter
TCP_RTO_MIN to either be tunable (such as it is for HP-UX, Solaris,
etc.) or set to = 3000 milli-seconds to enable the customers
application to restart a transmission in accordance to wireless
transmission (aka cell phone transmission rates).
  Version-Release number of selected component (if applicable):
RHEL-4U5 (hotfix)or patch.  Customer is also planning to roll to RHEL
5 over time so this fix must exist there as well, but that is less
pressing.
  With the value hardcoded = 200 milli-seconds, it is not possible to
have the application align properly with mobile phone TCPIP data traffic.
  This problem has potentially a 28 million dollar impact! =)

Upstream status:
  commit# 05bb1fad1cde025a864a90cfeb98dcbefe78a44a
  - Cell phone networks do link layer retransmissions and other things
    that cause unnecessary timeout retransmits.  So allow the minimum
    RTO to be inflated per-route to deal with this.
  commit# 5c127c58ae9bf196d787815b1bd6b0aec5aee816
  - (fix)  'dst' can be NULL in tcp_rto_min()

Test status:
  Patch has been tested for compilation, boot, and has been tested by
modified iproute user-space tool.

==

Acked-by: "David S. Miller" <davem@redhat.com>
Acked-by: Alan Cox <alan@redhat.com>

diff --git a/include/linux/rtnetlink.h b/include/linux/rtnetlink.h
index 41b8df5..b3f3c6b 100644
--- a/include/linux/rtnetlink.h
+++ b/include/linux/rtnetlink.h
@@ -350,6 +350,8 @@ enum
 #define RTAX_INITCWND RTAX_INITCWND
 	RTAX_FEATURES,
 #define RTAX_FEATURES RTAX_FEATURES
+	RTAX_RTO_MIN,
+#define RTAX_RTO_MIN RTAX_RTO_MIN
 	__RTAX_MAX
 };
 
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index aaeed37..684fc4d 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -544,6 +544,16 @@ static void tcp_event_data_recv(struct sock *sk, struct tcp_sock *tp, struct sk_
 		tcp_grow_window(sk, tp, skb);
 }
 
+static u32 tcp_rto_min(struct sock *sk)
+{
+	struct dst_entry *dst = __sk_dst_get(sk);
+	u32 rto_min = TCP_RTO_MIN;
+	
+	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
+		rto_min = dst->metrics[RTAX_RTO_MIN-1];
+	return rto_min;
+}
+
 /* Called to compute a smoothed rtt estimate. The data fed to this
  * routine either comes from timestamps, or from segments that were
  * known _not_ to have been retransmitted [see Karn/Partridge
@@ -605,13 +615,13 @@ static void tcp_rtt_estimator(struct sock *sk, const __u32 mrtt)
 			if (tp->mdev_max < tp->rttvar)
 				tp->rttvar -= (tp->rttvar-tp->mdev_max)>>2;
 			tp->rtt_seq = tp->snd_nxt;
-			tp->mdev_max = TCP_RTO_MIN;
+			tp->mdev_max = tcp_rto_min(sk);
 		}
 	} else {
 		/* no previous measure. */
 		tp->srtt = m<<3;	/* take the measured time to be rtt */
 		tp->mdev = m<<1;	/* make sure rto = 3*rtt */
-		tp->mdev_max = tp->rttvar = max(tp->mdev, TCP_RTO_MIN);
+		tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
 		tp->rtt_seq = tp->snd_nxt;
 	}
 }