From caa7f414ec6e1ce31591b7ba9152b8bdb34ac389 Mon Sep 17 00:00:00 2001
From: bluhm <bluhm@openbsd.org>
Date: Thu, 22 Feb 2024 14:25:58 +0000
Subject: [PATCH] Make the route cache aware of multipath routing.

Pass source address to route_cache() and store it in struct route.
Cached multipath routes are only valid if source address matches.
If sysctl multipath changes, increase route generation number.

OK claudio@
---
 sys/net/route.c            | 44 +++++++++++++++++++++++++-------------
 sys/net/route.h            | 23 +++++++++++---------
 sys/netinet/in_pcb.c       |  7 +++---
 sys/netinet/ip_input.c     | 16 ++++++++++----
 sys/netinet/ip_output.c    |  4 ++--
 sys/netinet6/in6_pcb.c     |  5 +++--
 sys/netinet6/in6_src.c     |  8 +++----
 sys/netinet6/ip6_forward.c |  5 +++--
 sys/netinet6/ip6_input.c   | 14 +++++++++---
 sys/netinet6/ip6_output.c  |  4 ++--
 10 files changed, 83 insertions(+), 47 deletions(-)

diff --git a/sys/net/route.c b/sys/net/route.c
index f9915cdf7c1..9188b6c4cfc 100644
--- a/sys/net/route.c
+++ b/sys/net/route.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: route.c,v 1.432 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: route.c,v 1.433 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$NetBSD: route.c,v 1.14 1996/02/13 22:00:46 christos Exp $	*/
 
 /*
@@ -202,7 +202,8 @@ route_init(void)
 }
 
 int
-route_cache(struct route *ro, struct in_addr addr, u_int rtableid)
+route_cache(struct route *ro, const struct in_addr *dst,
+    const struct in_addr *src, u_int rtableid)
 {
 	u_long gen;
 
@@ -213,28 +214,35 @@ route_cache(struct route *ro, struct in_addr addr, u_int rtableid)
 	    ro->ro_generation == gen &&
 	    ro->ro_tableid == rtableid &&
 	    ro->ro_dstsa.sa_family == AF_INET &&
-	    ro->ro_dstsin.sin_addr.s_addr == addr.s_addr) {
-		ipstat_inc(ips_rtcachehit);
-		return (0);
+	    ro->ro_dstsin.sin_addr.s_addr == dst->s_addr) {
+		if (src == NULL || !ipmultipath ||
+		    !ISSET(ro->ro_rt->rt_flags, RTF_MPATH) ||
+		    (ro->ro_srcin.s_addr != INADDR_ANY &&
+		    ro->ro_srcin.s_addr == src->s_addr)) {
+			ipstat_inc(ips_rtcachehit);
+			return (0);
+		}
 	}
 
 	ipstat_inc(ips_rtcachemiss);
 	rtfree(ro->ro_rt);
-	ro->ro_rt = NULL;
+	memset(ro, 0, sizeof(*ro));
 	ro->ro_generation = gen;
 	ro->ro_tableid = rtableid;
 
-	memset(&ro->ro_dst, 0, sizeof(ro->ro_dst));
 	ro->ro_dstsin.sin_family = AF_INET;
 	ro->ro_dstsin.sin_len = sizeof(struct sockaddr_in);
-	ro->ro_dstsin.sin_addr = addr;
+	ro->ro_dstsin.sin_addr = *dst;
+	if (src != NULL)
+		ro->ro_srcin = *src;
 
 	return (ESRCH);
 }
 
 #ifdef INET6
 int
-route6_cache(struct route *ro, const struct in6_addr *addr, u_int rtableid)
+route6_cache(struct route *ro, const struct in6_addr *dst,
+    const struct in6_addr *src, u_int rtableid)
 {
 	u_long gen;
 
@@ -245,21 +253,27 @@ route6_cache(struct route *ro, const struct in6_addr *addr, u_int rtableid)
 	    ro->ro_generation == gen &&
 	    ro->ro_tableid == rtableid &&
 	    ro->ro_dstsa.sa_family == AF_INET6 &&
-	    IN6_ARE_ADDR_EQUAL(&ro->ro_dstsin6.sin6_addr, addr)) {
-		ip6stat_inc(ip6s_rtcachehit);
-		return (0);
+	    IN6_ARE_ADDR_EQUAL(&ro->ro_dstsin6.sin6_addr, dst)) {
+		if (src == NULL || !ip6_multipath ||
+		    !ISSET(ro->ro_rt->rt_flags, RTF_MPATH) ||
+		    (!IN6_IS_ADDR_UNSPECIFIED(&ro->ro_srcin6) &&
+		    IN6_ARE_ADDR_EQUAL(&ro->ro_srcin6, src))) {
+			ip6stat_inc(ip6s_rtcachehit);
+			return (0);
+		}
 	}
 
 	ip6stat_inc(ip6s_rtcachemiss);
 	rtfree(ro->ro_rt);
-	ro->ro_rt = NULL;
+	memset(ro, 0, sizeof(*ro));
 	ro->ro_generation = gen;
 	ro->ro_tableid = rtableid;
 
-	memset(&ro->ro_dst, 0, sizeof(ro->ro_dst));
 	ro->ro_dstsin6.sin6_family = AF_INET6;
 	ro->ro_dstsin6.sin6_len = sizeof(struct sockaddr_in6);
-	ro->ro_dstsin6.sin6_addr = *addr;
+	ro->ro_dstsin6.sin6_addr = *dst;
+	if (src != NULL)
+		ro->ro_srcin6 = *src;
 
 	return (ESRCH);
 }
diff --git a/sys/net/route.h b/sys/net/route.h
index 7833f7ca9a8..fe652bd0925 100644
--- a/sys/net/route.h
+++ b/sys/net/route.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: route.h,v 1.206 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: route.h,v 1.207 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$NetBSD: route.h,v 1.9 1996/02/13 22:00:49 christos Exp $	*/
 
 /*
@@ -393,13 +393,14 @@ struct route {
 	u_long		 ro_generation;
 	u_long		 ro_tableid;	/* u_long because of alignment */
 	union {
-		struct	sockaddr	rod_sa;
-		struct	sockaddr_in	rod_sin;
-		struct	sockaddr_in6	rod_sin6;
-	} ro_dst;
-#define ro_dstsa	ro_dst.rod_sa
-#define ro_dstsin	ro_dst.rod_sin
-#define ro_dstsin6	ro_dst.rod_sin6
+		struct	sockaddr	ro_dstsa;
+		struct	sockaddr_in	ro_dstsin;
+		struct	sockaddr_in6	ro_dstsin6;
+	};
+	union {
+		struct	in_addr		ro_srcin;
+		struct	in6_addr	ro_srcin6;
+	};
 };
 
 #endif /* __BSD_VISIBLE */
@@ -462,8 +463,10 @@ struct if_ieee80211_data;
 struct bfd_config;
 
 void	 route_init(void);
-int	 route_cache(struct route *, struct in_addr, u_int);
-int	 route6_cache(struct route *, const struct in6_addr *, u_int);
+int	 route_cache(struct route *, const struct in_addr *,
+	    const struct in_addr *, u_int);
+int	 route6_cache(struct route *, const struct in6_addr *,
+	    const struct in6_addr *, u_int);
 void	 rtm_ifchg(struct ifnet *);
 void	 rtm_ifannounce(struct ifnet *, int);
 void	 rtm_bfd(struct bfd_config *);
diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index 87241c03e7a..cc08e62925f 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in_pcb.c,v 1.293 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: in_pcb.c,v 1.294 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$NetBSD: in_pcb.c,v 1.25 1996/02/13 23:41:53 christos Exp $	*/
 
 /*
@@ -919,7 +919,8 @@ in_pcbrtentry(struct inpcb *inp)
 
 	if (inp->inp_faddr.s_addr == INADDR_ANY)
 		return (NULL);
-	if (route_cache(ro, inp->inp_faddr, inp->inp_rtableid)) {
+	if (route_cache(ro, &inp->inp_faddr, &inp->inp_laddr,
+	    inp->inp_rtableid)) {
 		ro->ro_rt = rtalloc_mpath(&ro->ro_dstsa,
 		    &inp->inp_laddr.s_addr, ro->ro_tableid);
 	}
@@ -982,7 +983,7 @@ in_pcbselsrc(struct in_addr *insrc, struct sockaddr_in *sin,
 	 * If route is known or can be allocated now,
 	 * our src addr is taken from the i/f, else punt.
 	 */
-	if (route_cache(ro, sin->sin_addr, rtableid)) {
+	if (route_cache(ro, &sin->sin_addr, NULL, rtableid)) {
 		/* No route yet, so try to acquire one */
 		ro->ro_rt = rtalloc_mpath(&ro->ro_dstsa, NULL, ro->ro_tableid);
 	}
diff --git a/sys/netinet/ip_input.c b/sys/netinet/ip_input.c
index 60c6e6a422b..d8709c94cf7 100644
--- a/sys/netinet/ip_input.c
+++ b/sys/netinet/ip_input.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_input.c,v 1.389 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: ip_input.c,v 1.390 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$NetBSD: ip_input.c,v 1.30 1996/03/16 23:53:58 christos Exp $	*/
 
 /*
@@ -118,7 +118,6 @@ const struct sysctl_bounded_args ipctl_vars[] = {
 	{ IPCTL_IPPORT_HILASTAUTO, &ipport_hilastauto, 0, 65535 },
 	{ IPCTL_IPPORT_MAXQUEUE, &ip_maxqueue, 0, 10000 },
 	{ IPCTL_MFORWARDING, &ipmforwarding, 0, 1 },
-	{ IPCTL_MULTIPATH, &ipmultipath, 0, 1 },
 	{ IPCTL_ARPTIMEOUT, &arpt_keep, 0, INT_MAX },
 	{ IPCTL_ARPDOWN, &arpt_down, 0, INT_MAX },
 };
@@ -1491,7 +1490,7 @@ ip_forward(struct mbuf *m, struct ifnet *ifp, struct rtentry *rt, int srcrt)
 	}
 
 	ro.ro_rt = NULL;
-	route_cache(&ro, ip->ip_dst, m->m_pkthdr.ph_rtableid);
+	route_cache(&ro, &ip->ip_dst, &ip->ip_src, m->m_pkthdr.ph_rtableid);
 	if (!rtisvalid(rt)) {
 		rtfree(rt);
 		rt = rtalloc_mpath(&ro.ro_dstsa, &ip->ip_src.s_addr,
@@ -1633,10 +1632,10 @@ int
 ip_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
     size_t newlen)
 {
-	int error;
 #ifdef MROUTING
 	extern struct mrtstat mrtstat;
 #endif
+	int oldval, error;
 
 	/* Almost all sysctl names at this level are terminal. */
 	if (namelen != 1 && name[0] != IPCTL_IFQUEUE &&
@@ -1721,6 +1720,15 @@ ip_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
 	case IPCTL_MRTVIF:
 		return (EOPNOTSUPP);
 #endif
+	case IPCTL_MULTIPATH:
+		NET_LOCK();
+		oldval = ipmultipath;
+		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
+		    &ipmultipath, 0, 1);
+		if (oldval != ipmultipath)
+			atomic_inc_long(&rtgeneration);
+		NET_UNLOCK();
+		return (error);
 	default:
 		NET_LOCK();
 		error = sysctl_bounded_arr(ipctl_vars, nitems(ipctl_vars),
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index 561b690c765..0a606960376 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_output.c,v 1.395 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: ip_output.c,v 1.396 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$NetBSD: ip_output.c,v 1.28 1996/02/13 23:43:07 christos Exp $	*/
 
 /*
@@ -166,7 +166,7 @@ reroute:
 	 * 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.
 	 */
-	route_cache(ro, ip->ip_dst, m->m_pkthdr.ph_rtableid);
+	route_cache(ro, &ip->ip_dst, &ip->ip_src, m->m_pkthdr.ph_rtableid);
 	dst = &ro->ro_dstsin;
 
 	if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index d622c5dcc43..cea4415c17d 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in6_pcb.c,v 1.138 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: in6_pcb.c,v 1.139 2024/02/22 14:25:58 bluhm Exp $	*/
 
 /*
  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -565,7 +565,8 @@ in6_pcbrtentry(struct inpcb *inp)
 
 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->inp_faddr6))
 		return (NULL);
-	if (route6_cache(ro, &inp->inp_faddr6, inp->inp_rtableid)) {
+	if (route6_cache(ro, &inp->inp_faddr6, &inp->inp_laddr6,
+	    inp->inp_rtableid)) {
 		ro->ro_rt = rtalloc_mpath(&ro->ro_dstsa,
 		    &inp->inp_laddr6.s6_addr32[0], ro->ro_tableid);
 	}
diff --git a/sys/netinet6/in6_src.c b/sys/netinet6/in6_src.c
index 470ad67ebbc..ac35960a0f7 100644
--- a/sys/netinet6/in6_src.c
+++ b/sys/netinet6/in6_src.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in6_src.c,v 1.94 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: in6_src.c,v 1.95 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$KAME: in6_src.c,v 1.36 2001/02/06 04:08:17 itojun Exp $	*/
 
 /*
@@ -179,8 +179,8 @@ in6_pcbselsrc(const struct in6_addr **in6src, struct sockaddr_in6 *dstsock,
 	 * If route is known or can be allocated now,
 	 * our src addr is taken from the i/f, else punt.
 	 */
-	if (route6_cache(ro, dst, rtableid)) {
-		ro->ro_rt = rtalloc(&ro->ro_dstsa, RT_RESOLVE, ro->ro_tableid);
+	if (route6_cache(ro, dst, NULL, rtableid)) {
+		ro->ro_rt = rtalloc_mpath(&ro->ro_dstsa, NULL, ro->ro_tableid);
 	}
 
 	/*
@@ -304,7 +304,7 @@ in6_selectroute(const struct in6_addr *dst, struct ip6_pktopts *opts,
 	 * a new one.
 	 */
 	if (ro) {
-		if (route6_cache(ro, dst, rtableid)) {
+		if (route6_cache(ro, dst, NULL, rtableid)) {
 			/* No route yet, so try to acquire one */
 			ro->ro_rt = rtalloc_mpath(&ro->ro_dstsa, NULL,
 			    ro->ro_tableid);
diff --git a/sys/netinet6/ip6_forward.c b/sys/netinet6/ip6_forward.c
index cdb2fb417d1..65a51d52dac 100644
--- a/sys/netinet6/ip6_forward.c
+++ b/sys/netinet6/ip6_forward.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip6_forward.c,v 1.114 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: ip6_forward.c,v 1.115 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$KAME: ip6_forward.c,v 1.75 2001/06/29 12:42:13 jinmei Exp $	*/
 
 /*
@@ -166,7 +166,8 @@ reroute:
 #endif /* IPSEC */
 
 	ro.ro_rt = NULL;
-	route6_cache(&ro, &ip6->ip6_dst, m->m_pkthdr.ph_rtableid);
+	route6_cache(&ro, &ip6->ip6_dst, &ip6->ip6_src,
+	    m->m_pkthdr.ph_rtableid);
 	dst = &ro.ro_dstsa;
 	if (!rtisvalid(rt)) {
 		rtfree(rt);
diff --git a/sys/netinet6/ip6_input.c b/sys/netinet6/ip6_input.c
index 21bd2848c31..5d78407442b 100644
--- a/sys/netinet6/ip6_input.c
+++ b/sys/netinet6/ip6_input.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip6_input.c,v 1.257 2023/12/03 20:36:24 bluhm Exp $	*/
+/*	$OpenBSD: ip6_input.c,v 1.258 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$KAME: ip6_input.c,v 1.188 2001/03/29 05:34:31 itojun Exp $	*/
 
 /*
@@ -1451,7 +1451,6 @@ const struct sysctl_bounded_args ipv6ctl_vars[] = {
 	{ IPV6CTL_USE_DEPRECATED, &ip6_use_deprecated, 0, 1 },
 	{ IPV6CTL_MAXFRAGS, &ip6_maxfrags, 0, 1000 },
 	{ IPV6CTL_MFORWARDING, &ip6_mforwarding, 0, 1 },
-	{ IPV6CTL_MULTIPATH, &ip6_multipath, 0, 1 },
 	{ IPV6CTL_MCAST_PMTU, &ip6_mcast_pmtu, 0, 1 },
 	{ IPV6CTL_NEIGHBORGCTHRESH, &ip6_neighborgcthresh, -1, 5 * 2048 },
 	{ IPV6CTL_MAXDYNROUTES, &ip6_maxdynroutes, -1, 5 * 4096 },
@@ -1499,7 +1498,7 @@ ip6_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
 #ifdef MROUTING
 	extern struct mrt6stat mrt6stat;
 #endif
-	int error;
+	int oldval, error;
 
 	/* Almost all sysctl names at this level are terminal. */
 	if (namelen != 1 && name[0] != IPV6CTL_IFQUEUE)
@@ -1551,6 +1550,15 @@ ip6_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
 		    oldp, oldlenp, newp, newlen, &ip6intrq));
 	case IPV6CTL_SOIIKEY:
 		return (ip6_sysctl_soiikey(oldp, oldlenp, newp, newlen));
+	case IPV6CTL_MULTIPATH:
+		NET_LOCK();
+		oldval = ip6_multipath;
+		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
+		    &ip6_multipath, 0, 1);
+		if (oldval != ip6_multipath)
+			atomic_inc_long(&rtgeneration);
+		NET_UNLOCK();
+		return (error);
 	default:
 		NET_LOCK();
 		error = sysctl_bounded_arr(ipv6ctl_vars, nitems(ipv6ctl_vars),
diff --git a/sys/netinet6/ip6_output.c b/sys/netinet6/ip6_output.c
index adb5f30554d..e277d49599f 100644
--- a/sys/netinet6/ip6_output.c
+++ b/sys/netinet6/ip6_output.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip6_output.c,v 1.286 2024/02/13 12:22:09 bluhm Exp $	*/
+/*	$OpenBSD: ip6_output.c,v 1.287 2024/02/22 14:25:58 bluhm Exp $	*/
 /*	$KAME: ip6_output.c,v 1.172 2001/03/25 09:55:56 itojun Exp $	*/
 
 /*
@@ -480,7 +480,7 @@ reroute:
 			goto bad;
 		}
 	} else {
-		route6_cache(ro, &ip6->ip6_dst, m->m_pkthdr.ph_rtableid);
+		route6_cache(ro, &ip6->ip6_dst, NULL, m->m_pkthdr.ph_rtableid);
 	}
 
 	if (rt && (rt->rt_flags & RTF_GATEWAY) &&
-- 
2.20.1