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, 34 insertions, 66 deletions

diff --git a/sys/netinet/tcp_syncache.c b/sys/netinet/tcp_syncache.c
index 822ffeb..e2d96e9 100644
--- a/sys/netinet/tcp_syncache.c
+++ b/sys/netinet/tcp_syncache.c
@@ -202,29 +202,9 @@ static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache");
 static void
 syncache_free(struct syncache *sc)
 {
-	struct rtentry *rt;
-
 	if (sc->sc_ipopts)
 		(void) m_free(sc->sc_ipopts);
-#ifdef INET6
-	if (sc->sc_inc.inc_isipv6)
-		rt = sc->sc_route6.ro_rt;
-	else
-#endif
-		rt = sc->sc_route.ro_rt;
-	if (rt != NULL) {
-		/*
-		 * If this is the only reference to a protocol cloned 
-		 * route, remove it immediately.
-		 */
-		if (rt->rt_flags & RTF_WASCLONED &&
-		    (sc->sc_flags & SCF_KEEPROUTE) == 0 &&
-		    rt->rt_refcnt == 1)
-			rtrequest(RTM_DELETE, rt_key(rt),
-			    rt->rt_gateway, rt_mask(rt),
-			    rt->rt_flags, NULL);
-		RTFREE(rt);
-	}
+
 	uma_zfree(tcp_syncache.zone, sc);
 }
 
@@ -644,8 +624,6 @@ syncache_socket(sc, lso, m)
 		if (oinp->in6p_outputopts)
 			inp->in6p_outputopts =
 			    ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT);
-		inp->in6p_route = sc->sc_route6;
-		sc->sc_route6.ro_rt = NULL;
 
 		MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6,
 		    M_SONAME, M_NOWAIT | M_ZERO);
@@ -675,8 +653,6 @@ syncache_socket(sc, lso, m)
 			inp->inp_options = sc->sc_ipopts;
 			sc->sc_ipopts = NULL;
 		}
-		inp->inp_route = sc->sc_route;
-		sc->sc_route.ro_rt = NULL;
 
 		MALLOC(sin, struct sockaddr_in *, sizeof *sin,
 		    M_SONAME, M_NOWAIT | M_ZERO);
@@ -733,6 +709,10 @@ syncache_socket(sc, lso, m)
 		tp->cc_recv = sc->sc_cc_recv;
 	}
 
+	/*
+	 * Set up MSS and get cached values from tcp_hostcache.
+	 * This might overwrite some of the defaults we just set.
+	 */
 	tcp_mss(tp, sc->sc_peer_mss);
 
 	/*
@@ -811,10 +791,9 @@ resetandabort:
 #endif
 		m_freem(m);			/* XXX only needed for above */
 		tcpstat.tcps_sc_aborted++;
-	} else {
-		sc->sc_flags |= SCF_KEEPROUTE;
+	} else
 		tcpstat.tcps_sc_completed++;
-	}
+
 	if (sch == NULL)
 		syncache_free(sc);
 	else
@@ -849,13 +828,14 @@ syncache_add(inc, to, th, sop, m)
 	struct syncache *sc = NULL;
 	struct syncache_head *sch;
 	struct mbuf *ipopts = NULL;
-	struct rmxp_tao *taop;
+	struct rmxp_tao tao;
 	int i, win;
 
 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
 
 	so = *sop;
 	tp = sototcpcb(so);
+	bzero(&tao, sizeof(tao));
 
 	/*
 	 * Remember the IP options, if any.
@@ -949,13 +929,11 @@ syncache_add(inc, to, th, sop, m)
 	if (inc->inc_isipv6) {
 		sc->sc_inc.inc6_faddr = inc->inc6_faddr;
 		sc->sc_inc.inc6_laddr = inc->inc6_laddr;
-		sc->sc_route6.ro_rt = NULL;
 	} else
 #endif
 	{
 		sc->sc_inc.inc_faddr = inc->inc_faddr;
 		sc->sc_inc.inc_laddr = inc->inc_laddr;
-		sc->sc_route.ro_rt = NULL;
 	}
 	sc->sc_irs = th->th_seq;
 	sc->sc_flags = 0;
@@ -1027,17 +1005,19 @@ syncache_add(inc, to, th, sop, m)
 	 *	processing: drop SYN, process data and FIN.
 	 * - otherwise do a normal 3-way handshake.
 	 */
-	taop = tcp_gettaocache(&sc->sc_inc);
+	if (tcp_do_rfc1644)
+		tcp_hc_gettao(&sc->sc_inc, &tao);
+
 	if ((to->to_flags & TOF_CC) != 0) {
 		if (((tp->t_flags & TF_NOPUSH) != 0) &&
-		    sc->sc_flags & SCF_CC && 
-		    taop != NULL && taop->tao_cc != 0 &&
-		    CC_GT(to->to_cc, taop->tao_cc)) {
+		    sc->sc_flags & SCF_CC && tao.tao_cc != 0 &&
+		    CC_GT(to->to_cc, tao.tao_cc)) {
 			sc->sc_rxtslot = 0;
 			so = syncache_socket(sc, *sop, m);
 			if (so != NULL) {
-				sc->sc_flags |= SCF_KEEPROUTE;
-				taop->tao_cc = to->to_cc;
+				tao.tao_cc = to->to_cc;
+				tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC,
+						 tao.tao_cc, 0);
 				*sop = so;
 			}
 			syncache_free(sc);
@@ -1047,9 +1027,13 @@ syncache_add(inc, to, th, sop, m)
 		/*
 		 * No CC option, but maybe CC.NEW: invalidate cached value.
 		 */
-		if (taop != NULL)
-			taop->tao_cc = 0;
+		if (tcp_do_rfc1644) {
+			tao.tao_cc = 0;
+			tcp_hc_updatetao(&sc->sc_inc, TCP_HC_TAO_CC,
+					 tao.tao_cc, 0);
+		}
 	}
+
 	/*
 	 * TAO test failed or there was no CC option,
 	 *    do a standard 3-way handshake.
@@ -1087,33 +1071,22 @@ syncache_respond(sc, m)
 	int optlen, error;
 	u_int16_t tlen, hlen, mssopt;
 	struct ip *ip = NULL;
-	struct rtentry *rt;
 	struct tcphdr *th;
 	struct inpcb *inp;
 #ifdef INET6
 	struct ip6_hdr *ip6 = NULL;
 #endif
 
+	hlen =
 #ifdef INET6
-	if (sc->sc_inc.inc_isipv6) {
-		rt = tcp_rtlookup6(&sc->sc_inc);
-		if (rt != NULL)
-			mssopt = rt->rt_ifp->if_mtu -
-			     (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
-		else 
-			mssopt = tcp_v6mssdflt;
-		hlen = sizeof(struct ip6_hdr);
-	} else
+	       (sc->sc_inc.inc_isipv6) ? sizeof(struct ip6_hdr) : 
 #endif
-	{
-		rt = tcp_rtlookup(&sc->sc_inc);
-		if (rt != NULL)
-			mssopt = rt->rt_ifp->if_mtu -
-			     (sizeof(struct ip) + sizeof(struct tcphdr));
-		else 
-			mssopt = tcp_mssdflt;
-		hlen = sizeof(struct ip);
-	}
+		sizeof(struct ip);
+
+	KASSERT((&sc->sc_inc) != NULL, ("syncache_respond with NULL in_conninfo pointer"));
+
+	/* Determine MSS we advertize to other end of connection */
+	mssopt = tcp_mssopt(&sc->sc_inc);
 
 	/* Compute the size of the TCP options. */
 	if (sc->sc_flags & SCF_NOOPT) {
@@ -1244,13 +1217,10 @@ syncache_respond(sc, m)
 
 #ifdef INET6
 	if (sc->sc_inc.inc_isipv6) {
-		struct route_in6 *ro6 = &sc->sc_route6;
-
 		th->th_sum = 0;
 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
-		ip6->ip6_hlim = in6_selecthlim(NULL,
-		    ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
-		error = ip6_output(m, NULL, ro6, 0, NULL, NULL, inp);
+		ip6->ip6_hlim = in6_selecthlim(NULL, NULL);
+		error = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
 	} else
 #endif
 	{
@@ -1268,7 +1238,7 @@ syncache_respond(sc, m)
 			    mtod(m, void *), th, 0);
 		}
 #endif
-		error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL,inp);
+		error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, inp);
 	}
 	INP_UNLOCK(inp);
 	return (error);
@@ -1435,13 +1405,11 @@ syncookie_lookup(inc, th, so)
 	if (inc->inc_isipv6) {
 		sc->sc_inc.inc6_faddr = inc->inc6_faddr;
 		sc->sc_inc.inc6_laddr = inc->inc6_laddr;
-		sc->sc_route6.ro_rt = NULL;
 	} else
 #endif
 	{
 		sc->sc_inc.inc_faddr = inc->inc_faddr;
 		sc->sc_inc.inc_laddr = inc->inc_laddr;
-		sc->sc_route.ro_rt = NULL;
 	}
 	sc->sc_irs = th->th_seq - 1;
 	sc->sc_iss = th->th_ack - 1;