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)

6 files changed, 79 insertions, 81 deletions

diff --git a/sys/netinet6/icmp6.c b/sys/netinet6/icmp6.c
index 997474e..6baa2db 100644
--- a/sys/netinet6/icmp6.c
+++ b/sys/netinet6/icmp6.c
@@ -94,6 +94,7 @@
 #include <netinet/in_var.h>
 #include <netinet/ip6.h>
 #include <netinet/icmp6.h>
+#include <netinet/tcp_var.h>
 #include <netinet6/in6_ifattach.h>
 #include <netinet6/in6_pcb.h>
 #include <netinet6/ip6protosw.h>
@@ -1105,8 +1106,7 @@ icmp6_mtudisc_update(ip6cp, validated)
 	struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6;
 	struct mbuf *m = ip6cp->ip6c_m;	/* will be necessary for scope issue */
 	u_int mtu = ntohl(icmp6->icmp6_mtu);
-	struct rtentry *rt = NULL;
-	struct sockaddr_in6 sin6;
+	struct in_conninfo inc;
 
 #if 0
 	/*
@@ -1131,31 +1131,19 @@ icmp6_mtudisc_update(ip6cp, validated)
 	if (!validated)
 		return;
 
-	bzero(&sin6, sizeof(sin6));
-	sin6.sin6_family = PF_INET6;
-	sin6.sin6_len = sizeof(struct sockaddr_in6);
-	sin6.sin6_addr = *dst;
+	bzero(&inc, sizeof(inc));
+	inc.inc_flags = 1; /* IPv6 */
+	inc.inc6_faddr = *dst;
 	/* XXX normally, this won't happen */
 	if (IN6_IS_ADDR_LINKLOCAL(dst)) {
-		sin6.sin6_addr.s6_addr16[1] =
+		inc.inc6_faddr.s6_addr16[1] =
 		    htons(m->m_pkthdr.rcvif->if_index);
 	}
-	/* sin6.sin6_scope_id = XXX: should be set if DST is a scoped addr */
-	rt = rtalloc1((struct sockaddr *)&sin6, 0, RTF_CLONING);
-
-	if (rt && (rt->rt_flags & RTF_HOST) &&
-	    !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
-		if (mtu < IPV6_MMTU) {
-				/* xxx */
-			rt->rt_rmx.rmx_locks |= RTV_MTU;
-		} else if (mtu < rt->rt_ifp->if_mtu &&
-			   rt->rt_rmx.rmx_mtu > mtu) {
-			icmp6stat.icp6s_pmtuchg++;
-			rt->rt_rmx.rmx_mtu = mtu;
-		}
+
+	if (mtu >= IPV6_MMTU) {
+		tcp_hc_updatemtu(&inc, mtu);
+		icmp6stat.icp6s_pmtuchg++;
 	}
-	if (rt)
-		rtfree(rt);
 }
 
 /*
diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index 5c7f1f2..b3d58e8 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -337,8 +337,7 @@ in6_pcbladdr(inp, nam, plocal_addr6)
 		 * Is it the intended behavior?
 		 */
 		*plocal_addr6 = in6_selectsrc(sin6, inp->in6p_outputopts,
-					      inp->in6p_moptions,
-					      &inp->in6p_route,
+					      inp->in6p_moptions, NULL,
 					      &inp->in6p_laddr, &error);
 		if (*plocal_addr6 == 0) {
 			if (error == 0)
@@ -351,10 +350,6 @@ in6_pcbladdr(inp, nam, plocal_addr6)
 		 * and exit to caller, that will do the lookup.
 		 */
 	}
-
-	if (inp->in6p_route.ro_rt)
-		ifp = inp->in6p_route.ro_rt->rt_ifp;
-
 	return (0);
 }
 
@@ -447,8 +442,6 @@ in6_pcbdetach(inp)
 
  	ip6_freepcbopts(inp->in6p_outputopts);
  	ip6_freemoptions(inp->in6p_moptions);
-	if (inp->in6p_route.ro_rt)
-		RTFREE(inp->in6p_route.ro_rt);
 	/* Check and free IPv4 related resources in case of mapped addr */
 	if (inp->inp_options)
 		(void)m_free(inp->inp_options);
@@ -830,26 +823,10 @@ void
 in6_losing(in6p)
 	struct inpcb *in6p;
 {
-	struct rtentry *rt;
-	struct rt_addrinfo info;
-
-	if ((rt = in6p->in6p_route.ro_rt) != NULL) {
-		RT_LOCK(rt);
-		in6p->in6p_route.ro_rt = NULL;
-		bzero((caddr_t)&info, sizeof(info));
-		info.rti_flags = rt->rt_flags;
-		info.rti_info[RTAX_DST] = rt_key(rt);
-		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
-		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
-		rt_missmsg(RTM_LOSING, &info, rt->rt_flags, 0);
-		if (rt->rt_flags & RTF_DYNAMIC)
-			rtexpunge(rt);
-		RTFREE_LOCKED(rt);
-		/*
-		 * A new route can be allocated
-		 * the next time output is attempted.
-		 */
-	}
+	/*
+	 * We don't store route pointers in the routing table anymore
+	 */
+	return;
 }
 
 /*
@@ -861,14 +838,9 @@ in6_rtchange(inp, errno)
 	struct inpcb *inp;
 	int errno;
 {
-	if (inp->in6p_route.ro_rt) {
-		RTFREE(inp->in6p_route.ro_rt);
-		inp->in6p_route.ro_rt = 0;
-		/*
-		 * A new route can be allocated the next time
-		 * output is attempted.
-		 */
-	}
+	/*
+	 * We don't store route pointers in the routing table anymore
+	 */
 	return inp;
 }
 
diff --git a/sys/netinet6/in6_rmx.c b/sys/netinet6/in6_rmx.c
index 09526b2..b68852d 100644
--- a/sys/netinet6/in6_rmx.c
+++ b/sys/netinet6/in6_rmx.c
@@ -141,8 +141,7 @@ in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
 		}
 	}
 
-	if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
-	    && rt->rt_ifp)
+	if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
 		rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp);
 
 	ret = rn_addroute(v_arg, n_arg, head, treenodes);
diff --git a/sys/netinet6/in6_src.c b/sys/netinet6/in6_src.c
index d584956..88ace1c 100644
--- a/sys/netinet6/in6_src.c
+++ b/sys/netinet6/in6_src.c
@@ -211,7 +211,6 @@ in6_selectsrc(dstsock, opts, mopts, ro, laddr, errorp)
 		    != 0) {
 			return (NULL);
 		}
-
 		/*
 		 * determine the appropriate zone id of the source based on
 		 * the zone of the destination and the outgoing interface.
@@ -449,12 +448,19 @@ in6_selectif(dstsock, opts, mopts, ro, retifp)
 	struct route_in6 *ro;
 	struct ifnet **retifp;
 {
-	int error, clone;
+	int error;
+	struct route_in6 sro;
 	struct rtentry *rt = NULL;
 
-	clone = IN6_IS_ADDR_MULTICAST(&dstsock->sin6_addr) ? 0 : 1;
+	if (ro == NULL) {
+		bzero(&sro, sizeof(sro));
+		ro = &sro;
+	}
+
 	if ((error = in6_selectroute(dstsock, opts, mopts, ro, retifp,
-				     &rt, clone)) != 0) {
+				     &rt, 0)) != 0) {
+		if (rt && rt == sro.ro_rt)
+			RTFREE(rt);
 		return (error);
 	}
 
@@ -476,7 +482,11 @@ in6_selectif(dstsock, opts, mopts, ro, retifp)
 	 * We thus reject the case here.
 	 */
 	if (rt && (rt->rt_flags & (RTF_REJECT | RTF_BLACKHOLE))) {
-		return (rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
+		int flags = (rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH);
+
+		if (rt && rt == sro.ro_rt)
+			RTFREE(rt);
+		return (flags);
 	}
 
 	/*
@@ -489,6 +499,8 @@ in6_selectif(dstsock, opts, mopts, ro, retifp)
 	if (rt && rt->rt_ifa && rt->rt_ifa->ifa_ifp)
 		*retifp = rt->rt_ifa->ifa_ifp;
 
+	if (rt && rt == sro.ro_rt)
+		RTFREE(rt);
 	return (0);
 }
 
@@ -623,6 +635,7 @@ in6_selectroute(dstsock, opts, mopts, ro, retifp, retrt, clone)
 			sa6 = (struct sockaddr_in6 *)&ro->ro_dst;
 			*sa6 = *dstsock;
 			sa6->sin6_scope_id = 0;
+
 			if (clone) {
 				rtalloc((struct route *)ro);
 			} else {
@@ -695,7 +708,7 @@ in6_selectroute(dstsock, opts, mopts, ro, retifp, retrt, clone)
  * 2. (If the outgoing interface is detected) the current
  *     hop limit of the interface specified by router advertisement.
  * 3. The system default hoplimit.
-*/
+ */
 int
 in6_selecthlim(in6p, ifp)
 	struct in6pcb *in6p;
@@ -705,8 +718,24 @@ in6_selecthlim(in6p, ifp)
 		return (in6p->in6p_hops);
 	else if (ifp)
 		return (ND_IFINFO(ifp)->chlim);
-	else
-		return (ip6_defhlim);
+	else if (in6p && !IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_faddr)) {
+		struct route_in6 ro6;
+		struct ifnet *lifp;
+
+		bzero(&ro6, sizeof(ro6));
+		ro6.ro_dst.sin6_family = AF_INET6;
+		ro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
+		ro6.ro_dst.sin6_addr = in6p->in6p_faddr;
+		rtalloc((struct route *)&ro6);
+		if (ro6.ro_rt) {
+			lifp = ro6.ro_rt->rt_ifp;
+			RTFREE(ro6.ro_rt);
+			if (lifp)
+				return (ND_IFINFO(lifp)->chlim);
+		} else
+			return (ip6_defhlim);
+	}
+	return (ip6_defhlim);
 }
 
 /*
diff --git a/sys/netinet6/ip6_output.c b/sys/netinet6/ip6_output.c
index b95b197..3072851 100644
--- a/sys/netinet6/ip6_output.c
+++ b/sys/netinet6/ip6_output.c
@@ -96,6 +96,7 @@
 #include <netinet/icmp6.h>
 #include <netinet6/ip6_var.h>
 #include <netinet/in_pcb.h>
+#include <netinet/tcp_var.h>
 #include <netinet6/nd6.h>
 
 #ifdef IPSEC
@@ -661,7 +662,7 @@ skip_ipsec2:;
 		/* XXX rt not locked */
 		ia = ifatoia6(ro->ro_rt->rt_ifa);
 		ifp = ro->ro_rt->rt_ifp;
-		ro->ro_rt->rt_use++;
+		ro->ro_rt->rt_rmx.rmx_pksent++;
 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 			dst = (struct sockaddr_in6 *)ro->ro_rt->rt_gateway;
 		m->m_flags &= ~(M_BCAST | M_MCAST);	/* just in case */
@@ -757,7 +758,7 @@ skip_ipsec2:;
 			}
 			ia = ifatoia6(ro->ro_rt->rt_ifa);
 			ifp = ro->ro_rt->rt_ifp;
-			ro->ro_rt->rt_use++;
+			ro->ro_rt->rt_rmx.rmx_pksent++;
 			RT_UNLOCK(ro->ro_rt);
 		}
 
@@ -1387,11 +1388,20 @@ ip6_getpmtu(ro_pmtu, ro, ifp, dst, mtup, alwaysfragp)
 	}
 	if (ro_pmtu->ro_rt) {
 		u_int32_t ifmtu;
+		struct in_conninfo inc;
+
+		bzero(&inc, sizeof(inc));
+		inc.inc_flags = 1; /* IPv6 */
+		inc.inc6_faddr = *dst;
 
 		if (ifp == NULL)
 			ifp = ro_pmtu->ro_rt->rt_ifp;
 		ifmtu = IN6_LINKMTU(ifp);
-		mtu = ro_pmtu->ro_rt->rt_rmx.rmx_mtu;
+		mtu = tcp_hc_getmtu(&inc);
+		if (mtu)
+			mtu = min(mtu, ro_pmtu->ro_rt->rt_rmx.rmx_mtu);
+		else
+			mtu = ro_pmtu->ro_rt->rt_rmx.rmx_mtu;
 		if (mtu == 0)
 			mtu = ifmtu;
 		else if (mtu < IPV6_MMTU) {
@@ -1415,8 +1425,7 @@ ip6_getpmtu(ro_pmtu, ro, ifp, dst, mtup, alwaysfragp)
 			 * field isn't locked).
 			 */
 			mtu = ifmtu;
-			if (!(ro_pmtu->ro_rt->rt_rmx.rmx_locks & RTV_MTU))
-				ro_pmtu->ro_rt->rt_rmx.rmx_mtu = mtu;
+			ro_pmtu->ro_rt->rt_rmx.rmx_mtu = mtu;
 		}
 	} else if (ifp) {
 		mtu = IN6_LINKMTU(ifp);
@@ -1993,7 +2002,9 @@ do { \
 			{
 				u_long pmtu = 0;
 				struct ip6_mtuinfo mtuinfo;
-				struct route_in6 *ro = (struct route_in6 *)&in6p->in6p_route;
+				struct route_in6 sro;
+
+				bzero(&sro, sizeof(sro));
 
 				if (!(so->so_state & SS_ISCONNECTED))
 					return (ENOTCONN);
@@ -2002,8 +2013,10 @@ do { \
 				 * routing, or optional information to specify
 				 * the outgoing interface.
 				 */
-				error = ip6_getpmtu(ro, NULL, NULL,
+				error = ip6_getpmtu(&sro, NULL, NULL,
 				    &in6p->in6p_faddr, &pmtu, NULL);
+				if (sro.ro_rt)
+					RTFREE(sro.ro_rt);
 				if (error)
 					break;
 				if (pmtu > IPV6_MAXPACKET)
diff --git a/sys/netinet6/udp6_output.c b/sys/netinet6/udp6_output.c
index 36a7fba..d905e84 100644
--- a/sys/netinet6/udp6_output.c
+++ b/sys/netinet6/udp6_output.c
@@ -203,8 +203,7 @@ udp6_output(in6p, m, addr6, control, td)
 
 		if (!IN6_IS_ADDR_V4MAPPED(faddr)) {
 			laddr = in6_selectsrc(sin6, in6p->in6p_outputopts,
-					      in6p->in6p_moptions,
-					      &in6p->in6p_route,
+					      in6p->in6p_moptions, NULL,
 					      &in6p->in6p_laddr, &error);
 		} else
 			laddr = &in6p->in6p_laddr;	/* XXX */
@@ -277,9 +276,7 @@ udp6_output(in6p, m, addr6, control, td)
 		ip6->ip6_plen	= htons((u_short)plen);
 #endif
 		ip6->ip6_nxt	= IPPROTO_UDP;
-		ip6->ip6_hlim	= in6_selecthlim(in6p,
-						 in6p->in6p_route.ro_rt ?
-						 in6p->in6p_route.ro_rt->rt_ifp : NULL);
+		ip6->ip6_hlim	= in6_selecthlim(in6p, NULL);
 		ip6->ip6_src	= *laddr;
 		ip6->ip6_dst	= *faddr;
 
@@ -297,7 +294,7 @@ udp6_output(in6p, m, addr6, control, td)
 			goto release;
 		}
 #endif /* IPSEC */
-		error = ip6_output(m, in6p->in6p_outputopts, &in6p->in6p_route,
+		error = ip6_output(m, in6p->in6p_outputopts, NULL,
 		    flags, in6p->in6p_moptions, NULL, in6p);
 		break;
 	case AF_INET: