Index | Thread | Search

From:
Vitaliy Makkoveev <mvs@openbsd.org>
Subject:
Re: sysctl: unlock IP{,V6}CTL_MULTIPATH
To:
Alexander Bluhm <bluhm@openbsd.org>
Cc:
tech@openbsd.org
Date:
Sun, 22 Jun 2025 01:29:56 +0300

Download raw body.

Thread
On Sun, Jun 22, 2025 at 12:07:50AM +0200, Alexander Bluhm wrote:
> On Sun, Jun 22, 2025 at 12:52:45AM +0300, Vitaliy Makkoveev wrote:
> > On Sat, Jun 21, 2025 at 11:30:32PM +0200, Alexander Bluhm wrote:
> > > On Sat, Jun 21, 2025 at 10:21:02PM +0300, Vitaliy Makkoveev wrote:
> > > > They both are the same and both do `rtgeneration' update. As I
> > > > understand the code, we don't need to serialize `ipmultipath' and
> > > > `rtgeneration' update with the netstack, however, want to increment
> > > > `rtgeneration' before `ipmultipath' update. IIRC, atomic operations will
> > > > not be reordered and we don't need any extra barriers here.
> > > 
> > > This is architecture dependent.  membar_enter_after_atomic() is not
> > > defined everywhere.  I think membar_consumer() is needed here.  See
> > > comments inline.
> > > 
> > > bluhm
> > > 
> > >     atomic_store_int(&ipmultipath, newval);
> > >     membar_producer();
> > >     atomic_inc_long(&rtgeneration);
> > > 
> > > This ensures that route cache gets invalid after ipmultipath changes.
> > > You need that as rt_hash() result is based on ipmultipath.  That
> > > way the ipmultipath loads in rt_hash() and route_cache() can be
> > > kept independent.  No lock needed.
> > 
> > Could we have the side effects while two threads set `ipmultipath' to
> > opposing values? I mean
> > 
> > 	/* ipmultipath is 1 */
> > 	atomic_store_int(&ipmultipath, 0); /* thread A */
> > 	atomic_store_int(&ipmultipath, 1); /* thread B */
> > 	atomic_inc_long(&rtgeneration); /* thread B */
> > 	atomic_inc_long(&rtgeneration); /* thread A */
> > 
> > If so, I like to keep patch serialized.
> 
> As usual with concurrent sysctl, last write wins.  There is no point
> in serializing interger writes of userland threads.  Important is
> consistent kernel view.  And in your example atomic_inc_long(&rtgeneration)
> reloads the routing cache at next use.  It does not matter which
> thread does it.  It just has to happen after writing ipmultipath
> on the current CPU.
> 
> bluhm
> 

Ok. Updated diff, with barriers and without locks.

Index: sys/net/pf.c
===================================================================
RCS file: /cvs/src/sys/net/pf.c,v
retrieving revision 1.1213
diff -u -p -r1.1213 pf.c
--- sys/net/pf.c	9 Jun 2025 20:34:08 -0000	1.1213
+++ sys/net/pf.c	21 Jun 2025 22:25:52 -0000
@@ -6501,7 +6501,7 @@ pf_routable(struct pf_addr *addr, sa_fam
 		dst->sin_family = AF_INET;
 		dst->sin_len = sizeof(*dst);
 		dst->sin_addr = addr->v4;
-		if (ipmultipath)
+		if (atomic_load_int(&ipmultipath))
 			check_mpath = 1;
 		break;
 #ifdef INET6
@@ -6516,7 +6516,7 @@ pf_routable(struct pf_addr *addr, sa_fam
 		dst6->sin6_family = AF_INET6;
 		dst6->sin6_len = sizeof(*dst6);
 		dst6->sin6_addr = addr->v6;
-		if (ip6_multipath)
+		if (atomic_load_int(&ip6_multipath))
 			check_mpath = 1;
 		break;
 #endif /* INET6 */
Index: sys/net/route.c
===================================================================
RCS file: /cvs/src/sys/net/route.c,v
retrieving revision 1.444
diff -u -p -r1.444 route.c
--- sys/net/route.c	16 Mar 2025 23:45:06 -0000	1.444
+++ sys/net/route.c	21 Jun 2025 22:25:52 -0000
@@ -215,7 +215,7 @@ route_cache(struct route *ro, const stru
 	    ro->ro_tableid == rtableid &&
 	    ro->ro_dstsa.sa_family == AF_INET &&
 	    ro->ro_dstsin.sin_addr.s_addr == dst->s_addr) {
-		if (src == NULL || !ipmultipath ||
+		if (src == NULL || !atomic_load_int(&ipmultipath) ||
 		    !ISSET(ro->ro_rt->rt_flags, RTF_MPATH) ||
 		    (ro->ro_srcin.s_addr != INADDR_ANY &&
 		    ro->ro_srcin.s_addr == src->s_addr)) {
@@ -272,7 +272,7 @@ route6_cache(struct route *ro, const str
 	    ro->ro_tableid == rtableid &&
 	    ro->ro_dstsa.sa_family == AF_INET6 &&
 	    IN6_ARE_ADDR_EQUAL(&ro->ro_dstsin6.sin6_addr, dst)) {
-		if (src == NULL || !ip6_multipath ||
+		if (src == NULL || !atomic_load_int(&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))) {
@@ -426,7 +426,7 @@ rt_hash(struct rtentry *rt, const struct
 	    {
 		const struct sockaddr_in *sin;
 
-		if (!ipmultipath)
+		if (!atomic_load_int(&ipmultipath))
 			return (-1);
 
 		sin = satosin_const(dst);
@@ -440,7 +440,7 @@ rt_hash(struct rtentry *rt, const struct
 	    {
 		const struct sockaddr_in6 *sin6;
 
-		if (!ip6_multipath)
+		if (!atomic_load_int(&ip6_multipath))
 			return (-1);
 
 		sin6 = satosin6_const(dst);
Index: sys/netinet/ip_input.c
===================================================================
RCS file: /cvs/src/sys/netinet/ip_input.c,v
retrieving revision 1.410
diff -u -p -r1.410 ip_input.c
--- sys/netinet/ip_input.c	21 Jun 2025 14:21:17 -0000	1.410
+++ sys/netinet/ip_input.c	21 Jun 2025 22:25:52 -0000
@@ -93,7 +93,7 @@
 /* values controllable via sysctl */
 int	ip_forwarding = 0;			/* [a] */
 int	ipmforwarding = 0;
-int	ipmultipath = 0;
+int	ipmultipath = 0;			/* [a] */
 int	ip_sendredirects = 1;			/* [a] */
 int	ip_dosourceroute = 0;			/* [a] */
 int	ip_defttl = IPDEFTTL;
@@ -1817,13 +1817,15 @@ ip_sysctl(int *name, u_int namelen, void
 		return (EOPNOTSUPP);
 #endif
 	case IPCTL_MULTIPATH:
-		NET_LOCK();
-		oldval = ipmultipath;
+		oldval = newval = atomic_load_int(&ipmultipath);
 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
-		    &ipmultipath, 0, 1);
-		if (oldval != ipmultipath)
+		    &newval, 0, 1);
+		if (error == 0 && oldval != newval) {
+			atomic_store_int(&ipmultipath, newval);
+			membar_producer();
 			atomic_inc_long(&rtgeneration);
-		NET_UNLOCK();
+		}
+
 		return (error);
 	case IPCTL_FORWARDING:
 	case IPCTL_SENDREDIRECTS:
Index: sys/netinet6/in6_proto.c
===================================================================
RCS file: /cvs/src/sys/netinet6/in6_proto.c,v
retrieving revision 1.130
diff -u -p -r1.130 in6_proto.c
--- sys/netinet6/in6_proto.c	21 Jun 2025 14:21:17 -0000	1.130
+++ sys/netinet6/in6_proto.c	21 Jun 2025 22:25:52 -0000
@@ -359,7 +359,7 @@ const struct domain inet6domain = {
  */
 int	ip6_forwarding = 0;	/* [a] no forwarding unless sysctl to enable */
 int	ip6_mforwarding = 0;	/* no multicast forwarding unless ... */
-int	ip6_multipath = 0;	/* no using multipath routes unless ... */
+int	ip6_multipath = 0;	/* [a] no using multipath routes unless ... */
 int	ip6_sendredirects = 1;	/* [a] */
 int	ip6_defhlim = IPV6_DEFHLIM;
 int	ip6_defmcasthlim = IPV6_DEFAULT_MULTICAST_HOPS;
Index: sys/netinet6/ip6_input.c
===================================================================
RCS file: /cvs/src/sys/netinet6/ip6_input.c,v
retrieving revision 1.274
diff -u -p -r1.274 ip6_input.c
--- sys/netinet6/ip6_input.c	21 Jun 2025 14:21:17 -0000	1.274
+++ sys/netinet6/ip6_input.c	21 Jun 2025 22:25:52 -0000
@@ -1566,15 +1566,19 @@ ip6_sysctl(int *name, u_int namelen, voi
 		return (sysctl_niq(name + 1, namelen - 1,
 		    oldp, oldlenp, newp, newlen, &ip6intrq));
 	case IPV6CTL_MULTIPATH: {
-		NET_LOCK();
-		int oldval = ip6_multipath;
+		int oldval, newval;
+
+		oldval = newval = atomic_load_int(&ip6_multipath);
 		error = sysctl_int_bounded(oldp, oldlenp, newp, newlen,
-		    &ip6_multipath, 0, 1);
-		if (oldval != ip6_multipath)
+		    &newval, 0, 1);
+		if (error == 0 && oldval != newval) {
+			atomic_store_int(&ip6_multipath, newval);
+			membar_producer();
 			atomic_inc_long(&rtgeneration);
-		NET_UNLOCK();
+		}
+
 		return (error);
-	    }
+	}
 	case IPV6CTL_FORWARDING:
 	case IPV6CTL_SENDREDIRECTS:
 		return (sysctl_bounded_arr(