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)

1 files changed, 9 insertions, 16 deletions

diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index cdf8b87..0a11524 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -302,13 +302,9 @@ ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 		isbroadcast = 0;	/* fool gcc */
 	} else {
 		/*
-		 * If this is the case, we probably don't want to allocate
-		 * a protocol-cloned route since we didn't get one from the
-		 * ULP.  This lets TCP do its thing, while not burdening
-		 * forwarding or ICMP with the overhead of cloning a route.
-		 * Of course, we still want to do any cloning requested by
-		 * the link layer, as this is probably required in all cases
-		 * for correct operation (as it is for ARP).
+		 * We want to do any cloning requested by the link layer,
+		 * as this is probably required in all cases for correct
+		 * operation (as it is for ARP).
 		 */
 		if (ro->ro_rt == 0)
 			rtalloc(ro);
@@ -319,7 +315,7 @@ ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 		}
 		ia = ifatoia(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_in *)ro->ro_rt->rt_gateway;
 		if (ro->ro_rt->rt_flags & RTF_HOST)
@@ -931,16 +927,14 @@ spd_done:
 				ip_input((struct mbuf *)&tag);
 				goto done;
 			}
-			/* Some of the logic for this was
+			/*
+			 * Some of the logic for this was
 			 * nicked from above.
-			 *
-			 * This rewrites the cached route in a local PCB.
-			 * Is this what we want to do?
 			 */
 			bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst));
 
 			ro_fwd->ro_rt = 0;
-			rtalloc(ro_fwd);
+			rtalloc_ign(ro_fwd, RTF_CLONING);
 
 			if (ro_fwd->ro_rt == 0) {
 				ipstat.ips_noroute++;
@@ -950,7 +944,7 @@ spd_done:
 
 			ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 			ifp = ro_fwd->ro_rt->rt_ifp;
-			ro_fwd->ro_rt->rt_use++;
+			ro_fwd->ro_rt->rt_rmx.rmx_pksent++;
 			if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 				dst = (struct sockaddr_in *)
 					ro_fwd->ro_rt->rt_gateway;
@@ -1045,7 +1039,6 @@ pass:
 		 * routes when the MTU is changed.
 		 */
 		if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
-		    !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
 		    (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
 			ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
 		}
@@ -1983,7 +1976,7 @@ ip_setmoptions(sopt, imop)
 			dst->sin_len = sizeof(*dst);
 			dst->sin_family = AF_INET;
 			dst->sin_addr = mreq.imr_multiaddr;
-			rtalloc(&ro);
+			rtalloc_ign(&ro, RTF_CLONING);
 			if (ro.ro_rt == NULL) {
 				error = EADDRNOTAVAIL;
 				splx(s);