FreeBSD previously provided route caching for TCP (and UDP). Re-add

route caching for TCP, with some improvements. In particular, invalidate
the route cache if a new route is added, which might be a better match.
The cache is automatically invalidated if the old route is deleted.

Submitted by:	Mike Karels
Reviewed by:	gnn
Differential Revision:	https://reviews.freebsd.org/D4306

7 files changed, 98 insertions, 29 deletions

diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index ec42e67..50e5d6e 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -1298,6 +1298,11 @@ in_pcbfree(struct inpcb *inp)
 	if (inp->inp_moptions != NULL)
 		inp_freemoptions(inp->inp_moptions);
 #endif
+	if (inp->inp_route.ro_rt) {
+		RTFREE(inp->inp_route.ro_rt);
+		inp->inp_route.ro_rt = (struct rtentry *)NULL;
+	}
+
 	inp->inp_vflag = 0;
 	inp->inp_flags2 |= INP_FREED;
 	crfree(inp->inp_cred);
@@ -2225,6 +2230,23 @@ in_pcbremlists(struct inpcb *inp)
 }
 
 /*
+ * Check for alternatives when higher level complains
+ * about service problems.  For now, invalidate cached
+ * routing information.  If the route was created dynamically
+ * (by a redirect), time to try a default gateway again.
+ */
+void
+in_losing(struct inpcb *inp)
+{
+
+	if (inp->inp_route.ro_rt) {
+		RTFREE(inp->inp_route.ro_rt);
+		inp->inp_route.ro_rt = (struct rtentry *)NULL;
+	}
+	return;
+}
+
+/*
  * A set label operation has occurred at the socket layer, propagate the
  * label change into the in_pcb for the socket.
  */
diff --git a/sys/netinet/in_pcb.h b/sys/netinet/in_pcb.h
index 54eab88..9c77a51 100644
--- a/sys/netinet/in_pcb.h
+++ b/sys/netinet/in_pcb.h
@@ -42,6 +42,7 @@
 #include <sys/_lock.h>
 #include <sys/_mutex.h>
 #include <sys/_rwlock.h>
+#include <net/route.h>
 
 #ifdef _KERNEL
 #include <sys/lock.h>
@@ -238,8 +239,14 @@ struct inpcb {
 #define inp_zero_size offsetof(struct inpcb, inp_gencnt)
 	inp_gen_t	inp_gencnt;	/* (c) generation count */
 	struct llentry	*inp_lle;	/* cached L2 information */
-	struct rtentry	*inp_rt;	/* cached L3 information */
 	struct rwlock	inp_lock;
+	rt_gen_t	inp_rt_cookie;	/* generation for route entry */
+	union {				/* cached L3 information */
+		struct route inpu_route;
+		struct route_in6 inpu_route6;
+	} inp_rtu;
+#define inp_route inp_rtu.inpu_route
+#define inp_route6 inp_rtu.inpu_route6
 };
 #define	inp_fport	inp_inc.inc_fport
 #define	inp_lport	inp_inc.inc_lport
@@ -722,6 +729,7 @@ void	in_pcbrehash_mbuf(struct inpcb *, struct mbuf *);
 int	in_pcbrele(struct inpcb *);
 int	in_pcbrele_rlocked(struct inpcb *);
 int	in_pcbrele_wlocked(struct inpcb *);
+void	in_losing(struct inpcb *);
 void	in_pcbsetsolabel(struct socket *so);
 int	in_getpeeraddr(struct socket *so, struct sockaddr **nam);
 int	in_getsockaddr(struct socket *so, struct sockaddr **nam);
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index fa225fc..afdf1a7 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -282,17 +282,36 @@ ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 	gw = dst = (struct sockaddr_in *)&ro->ro_dst;
 	fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m);
 	rte = ro->ro_rt;
-	/*
-	 * The address family should also be checked in case of sharing
-	 * the cache with IPv6.
-	 */
-	if (rte == NULL || dst->sin_family != AF_INET) {
+	if (rte == NULL) {
 		bzero(dst, sizeof(*dst));
 		dst->sin_family = AF_INET;
 		dst->sin_len = sizeof(*dst);
 		dst->sin_addr = ip->ip_dst;
 	}
 again:
+	/*
+	 * Validate route against routing table additions;
+	 * a better/more specific route might have been added.
+	 */
+	if (inp)
+		RT_VALIDATE(ro, &inp->inp_rt_cookie, fibnum);
+	/*
+	 * If there is a cached route,
+	 * check that it is to the same destination
+	 * and is still up.  If not, free it and try again.
+	 * The address family should also be checked in case of sharing the
+	 * cache with IPv6.
+	 * Also check whether routing cache needs invalidation.
+	 */
+	rte = ro->ro_rt;
+	if (rte && ((rte->rt_flags & RTF_UP) == 0 ||
+		    rte->rt_ifp == NULL ||
+		    !RT_LINK_IS_UP(rte->rt_ifp) ||
+			  dst->sin_family != AF_INET ||
+			  dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
+		RTFREE(rte);
+		rte = ro->ro_rt = (struct rtentry *)NULL;
+	}
 	ia = NULL;
 	have_ia_ref = 0;
 	/*
diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c
index d7fb75d..2043fc9 100644
--- a/sys/netinet/tcp_output.c
+++ b/sys/netinet/tcp_output.c
@@ -1379,9 +1379,6 @@ send:
 #endif
 #ifdef INET
     {
-	struct route ro;
-
-	bzero(&ro, sizeof(ro));
 	ip->ip_len = htons(m->m_pkthdr.len);
 #ifdef INET6
 	if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO)
@@ -1412,13 +1409,12 @@ send:
 	tcp_pcap_add(th, m, &(tp->t_outpkts));
 #endif
 
-	error = ip_output(m, tp->t_inpcb->inp_options, &ro,
+	error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
 	    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
 	    tp->t_inpcb);
 
-	if (error == EMSGSIZE && ro.ro_rt != NULL)
-		mtu = ro.ro_rt->rt_mtu;
-	RO_RTFREE(&ro);
+	if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL)
+		mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu;
     }
 #endif /* INET */
 
diff --git a/sys/netinet/tcp_subr.c b/sys/netinet/tcp_subr.c
index 6635bb7..877f7f8 100644
--- a/sys/netinet/tcp_subr.c
+++ b/sys/netinet/tcp_subr.c
@@ -1632,6 +1632,10 @@ tcp_notify(struct inpcb *inp, int error)
 	if (tp->t_state == TCPS_ESTABLISHED &&
 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
 	     error == EHOSTDOWN)) {
+		if (inp->inp_route.ro_rt) {
+			RTFREE(inp->inp_route.ro_rt);
+			inp->inp_route.ro_rt = (struct rtentry *)NULL;
+		}
 		return (inp);
 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
 	    tp->t_softerror) {
@@ -1926,11 +1930,11 @@ tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
 		cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
 		notify = tcp_drop_syn_sent;
-	/*
-	 * Redirects don't need to be handled up here.
-	 */
-	else if (PRC_IS_REDIRECT(cmd))
+	else if (PRC_IS_REDIRECT(cmd)) {
+		/* signal EHOSTDOWN, as it flushes the cached route */
+		in_pcbnotifyall(&tcbinfo, faddr, EHOSTDOWN, notify);
 		return;
+	}
 	/*
 	 * Hostdead is ugly because it goes linearly through all PCBs.
 	 * XXX: We never get this from ICMP, otherwise it makes an
diff --git a/sys/netinet/tcp_timer.c b/sys/netinet/tcp_timer.c
index 046feba..172c394 100644
--- a/sys/netinet/tcp_timer.c
+++ b/sys/netinet/tcp_timer.c
@@ -786,7 +786,9 @@ tcp_timer_rexmt(void * xtp)
 #ifdef INET6
 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 			in6_losing(tp->t_inpcb);
+		else
 #endif
+			in_losing(tp->t_inpcb);
 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
 		tp->t_srtt = 0;
 	}
diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index f474a54..ecd1e0c 100644
--- a/sys/netinet/udp_usrreq.c
+++ b/sys/netinet/udp_usrreq.c
@@ -740,6 +740,11 @@ udp_notify(struct inpcb *inp, int errno)
 	 * or a write lock, but a read lock is sufficient.
 	 */
 	INP_LOCK_ASSERT(inp);
+	if ((errno == EHOSTUNREACH || errno == ENETUNREACH ||
+	     errno == EHOSTDOWN) && inp->inp_route.ro_rt) {
+		RTFREE(inp->inp_route.ro_rt);
+		inp->inp_route.ro_rt = (struct rtentry *)NULL;
+	}
 
 	inp->inp_socket->so_error = errno;
 	sorwakeup(inp->inp_socket);
@@ -761,11 +766,11 @@ udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip,
 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
 		return;
 
-	/*
-	 * Redirects don't need to be handled up here.
-	 */
-	if (PRC_IS_REDIRECT(cmd))
+	if (PRC_IS_REDIRECT(cmd)) {
+		/* signal EHOSTDOWN, as it flushes the cached route */
+		in_pcbnotifyall(&udbinfo, faddr, EHOSTDOWN, udp_notify);
 		return;
+	}
 
 	/*
 	 * Hostdead is ugly because it goes linearly through all PCBs.
@@ -1116,7 +1121,7 @@ udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 	int error = 0;
 	int ipflags;
 	u_short fport, lport;
-	int unlock_udbinfo;
+	int unlock_udbinfo, unlock_inp;
 	u_char tos;
 	uint8_t pr;
 	uint16_t cscov = 0;
@@ -1137,7 +1142,15 @@ udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 	}
 
 	src.sin_family = 0;
-	INP_RLOCK(inp);
+	sin = (struct sockaddr_in *)addr;
+	if (sin == NULL ||
+	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
+		INP_WLOCK(inp);
+		unlock_inp = UH_WLOCKED;
+	} else {
+		INP_RLOCK(inp);
+		unlock_inp = UH_RLOCKED;
+	}
 	tos = inp->inp_ip_tos;
 	if (control != NULL) {
 		/*
@@ -1145,7 +1158,10 @@ udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 		 * stored in a single mbuf.
 		 */
 		if (control->m_next) {
-			INP_RUNLOCK(inp);
+			if (unlock_inp == UH_WLOCKED)
+				INP_WUNLOCK(inp);
+			else
+				INP_RUNLOCK(inp);
 			m_freem(control);
 			m_freem(m);
 			return (EINVAL);
@@ -1220,7 +1236,10 @@ udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 		m_freem(control);
 	}
 	if (error) {
-		INP_RUNLOCK(inp);
+		if (unlock_inp == UH_WLOCKED)
+			INP_WUNLOCK(inp);
+		else
+			INP_RUNLOCK(inp);
 		m_freem(m);
 		return (error);
 	}
@@ -1246,8 +1265,6 @@ udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 	sin = (struct sockaddr_in *)addr;
 	if (sin != NULL &&
 	    (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
-		INP_RUNLOCK(inp);
-		INP_WLOCK(inp);
 		INP_HASH_WLOCK(pcbinfo);
 		unlock_udbinfo = UH_WLOCKED;
 	} else if ((sin != NULL && (
@@ -1514,9 +1531,10 @@ udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
 	else if (unlock_udbinfo == UH_RLOCKED)
 		INP_HASH_RUNLOCK(pcbinfo);
 	UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
-	error = ip_output(m, inp->inp_options, NULL, ipflags,
+	error = ip_output(m, inp->inp_options,
+	    (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags,
 	    inp->inp_moptions, inp);
-	if (unlock_udbinfo == UH_WLOCKED)
+	if (unlock_inp == UH_WLOCKED)
 		INP_WUNLOCK(inp);
 	else
 		INP_RUNLOCK(inp);