Introduce tcp_hostcache and remove the tcp specific metrics from

the routing table.  Move all usage and references in the tcp stack
from the routing table metrics to the tcp hostcache.

It caches measured parameters of past tcp sessions to provide better
initial start values for following connections from or to the same
source or destination.  Depending on the network parameters to/from
the remote host this can lead to significant speedups for new tcp
connections after the first one because they inherit and shortcut
the learning curve.

tcp_hostcache is designed for multiple concurrent access in SMP
environments with high contention and is hash indexed by remote
ip address.

It removes significant locking requirements from the tcp stack with
regard to the routing table.

Reviewed by:	sam (mentor), bms
Reviewed by:	-net, -current, core@kame.net (IPv6 parts)
Approved by:	re (scottl)

4 files changed, 33 insertions, 41 deletions

diff --git a/sys/net/if_faith.c b/sys/net/if_faith.c
index 07216b5..a8da4ad 100644
--- a/sys/net/if_faith.c
+++ b/sys/net/if_faith.c
@@ -270,17 +270,8 @@ faithrtrequest(cmd, rt, info)
 	struct rt_addrinfo *info;
 {
 	RT_LOCK_ASSERT(rt);
-
-	if (rt) {
-		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; /* for ISO */
-		/*
-		 * For optimal performance, the send and receive buffers
-		 * should be at least twice the MTU plus a little more for
-		 * overhead.
-		 */
-		rt->rt_rmx.rmx_recvpipe =
-			rt->rt_rmx.rmx_sendpipe = 3 * FAITHMTU;
-	}
+	if (rt)
+		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 }
 
 /*
diff --git a/sys/net/if_loop.c b/sys/net/if_loop.c
index afe0a73..9a54af4 100644
--- a/sys/net/if_loop.c
+++ b/sys/net/if_loop.c
@@ -329,17 +329,8 @@ lortrequest(cmd, rt, info)
 	struct rt_addrinfo *info;
 {
 	RT_LOCK_ASSERT(rt);
-
-	if (rt) {
-		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; /* for ISO */
-		/*
-		 * For optimal performance, the send and receive buffers
-		 * should be at least twice the MTU plus a little more for
-		 * overhead.
-		 */
-		rt->rt_rmx.rmx_recvpipe =
-			rt->rt_rmx.rmx_sendpipe = 3 * LOMTU;
-	}
+	if (rt)
+		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 }
 
 /*
diff --git a/sys/net/route.h b/sys/net/route.h
index 8fff560..34c33eb 100644
--- a/sys/net/route.h
+++ b/sys/net/route.h
@@ -58,6 +58,12 @@ struct route {
  * These numbers are used by reliable protocols for determining
  * retransmission behavior and are included in the routing structure.
  */
+struct rt_metrics_lite {
+	u_long	rmx_mtu;	/* MTU for this path */
+	u_long	rmx_expire;	/* lifetime for route, e.g. redirect */
+	u_long	rmx_pksent;	/* packets sent using this route */
+};
+
 struct rt_metrics {
 	u_long	rmx_locks;	/* Kernel must leave these values alone */
 	u_long	rmx_mtu;	/* MTU for this path */
@@ -104,10 +110,10 @@ struct rtentry {
 	long	rt_refcnt;		/* # held references */
 	u_long	rt_flags;		/* up/down?, host/net */
 	struct	ifnet *rt_ifp;		/* the answer: interface to use */
-	struct	ifaddr *rt_ifa;		/* the answer: interface to use */
+	struct	ifaddr *rt_ifa;		/* the answer: interface address to use */
 	struct	sockaddr *rt_genmask;	/* for generation of cloned routes */
 	caddr_t	rt_llinfo;		/* pointer to link level info cache */
-	struct	rt_metrics rt_rmx;	/* metrics used by rx'ing protocols */
+	struct	rt_metrics_lite rt_rmx;	/* metrics used by rx'ing protocols */
 	struct	rtentry *rt_gwroute;	/* implied entry for gatewayed routes */
 	int	(*rt_output)(struct ifnet *, struct mbuf *, struct sockaddr *,
 		    struct rtentry *);
diff --git a/sys/net/rtsock.c b/sys/net/rtsock.c
index 4fba1a2..3290c0c 100644
--- a/sys/net/rtsock.c
+++ b/sys/net/rtsock.c
@@ -87,7 +87,8 @@ static int	sysctl_dumpentry(struct radix_node *rn, void *vw);
 static int	sysctl_iflist(int af, struct walkarg *w);
 static int	sysctl_ifmalist(int af, struct walkarg *w);
 static int	route_output(struct mbuf *, struct socket *);
-static void	rt_setmetrics(u_long, struct rt_metrics *, struct rt_metrics *);
+static void	rt_setmetrics(u_long, struct rt_metrics *, struct rt_metrics_lite *);
+static void	rt_getmetrics(struct rt_metrics_lite *, struct rt_metrics *);
 static void	rt_dispatch(struct mbuf *, struct sockaddr *);
 
 /*
@@ -355,9 +356,6 @@ route_output(m, so)
 			RT_LOCK(saved_nrt);
 			rt_setmetrics(rtm->rtm_inits,
 				&rtm->rtm_rmx, &saved_nrt->rt_rmx);
-			saved_nrt->rt_rmx.rmx_locks &= ~(rtm->rtm_inits);
-			saved_nrt->rt_rmx.rmx_locks |=
-				(rtm->rtm_inits & rtm->rtm_rmx.rmx_locks);
 			RT_REMREF(saved_nrt);
 			saved_nrt->rt_genmask = info.rti_info[RTAX_GENMASK];
 			RT_UNLOCK(saved_nrt);
@@ -428,7 +426,7 @@ route_output(m, so)
 			(void)rt_msg2(rtm->rtm_type, &info, (caddr_t)rtm,
 				(struct walkarg *)0);
 			rtm->rtm_flags = rt->rt_flags;
-			rtm->rtm_rmx = rt->rt_rmx;
+			rt_getmetrics(&rt->rt_rmx, &rtm->rtm_rmx);
 			rtm->rtm_addrs = info.rti_addrs;
 			break;
 
@@ -478,9 +476,7 @@ route_output(m, so)
 				rt->rt_genmask = info.rti_info[RTAX_GENMASK];
 			/* FALLTHROUGH */
 		case RTM_LOCK:
-			rt->rt_rmx.rmx_locks &= ~(rtm->rtm_inits);
-			rt->rt_rmx.rmx_locks |=
-				(rtm->rtm_inits & rtm->rtm_rmx.rmx_locks);
+			/* We don't support locks anymore */
 			break;
 		}
 		RT_UNLOCK(rt);
@@ -542,20 +538,28 @@ flush:
 }
 
 static void
-rt_setmetrics(u_long which, struct rt_metrics *in, struct rt_metrics *out)
+rt_setmetrics(u_long which, struct rt_metrics *in, struct rt_metrics_lite *out)
 {
 #define metric(f, e) if (which & (f)) out->e = in->e;
-	metric(RTV_RPIPE, rmx_recvpipe);
-	metric(RTV_SPIPE, rmx_sendpipe);
-	metric(RTV_SSTHRESH, rmx_ssthresh);
-	metric(RTV_RTT, rmx_rtt);
-	metric(RTV_RTTVAR, rmx_rttvar);
-	metric(RTV_HOPCOUNT, rmx_hopcount);
+	/*
+	 * Only these are stored in the routing entry since introduction
+	 * of tcp hostcache. The rest is ignored.
+	 */
 	metric(RTV_MTU, rmx_mtu);
 	metric(RTV_EXPIRE, rmx_expire);
 #undef metric
 }
 
+static void
+rt_getmetrics(struct rt_metrics_lite *in, struct rt_metrics *out)
+{
+#define metric(e) out->e = in->e;
+	bzero(out, sizeof(*out));
+	metric(rmx_mtu);
+	metric(rmx_expire);
+#undef metric
+}
+
 #define ROUNDUP(a) \
 	((a) > 0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
 
@@ -948,8 +952,8 @@ sysctl_dumpentry(struct radix_node *rn, void *vw)
 		struct rt_msghdr *rtm = (struct rt_msghdr *)w->w_tmem;
 
 		rtm->rtm_flags = rt->rt_flags;
-		rtm->rtm_use = rt->rt_use;
-		rtm->rtm_rmx = rt->rt_rmx;
+		rtm->rtm_use = rt->rt_rmx.rmx_pksent;
+		rt_getmetrics(&rt->rt_rmx, &rtm->rtm_rmx);
 		rtm->rtm_index = rt->rt_ifp->if_index;
 		rtm->rtm_errno = rtm->rtm_pid = rtm->rtm_seq = 0;
 		rtm->rtm_addrs = info.rti_addrs;