revert "Retire kernel support for SO_DONTROUTE" diff, which does bad things

for localhost connections. discussed with deraadt@

13 files changed, 123 insertions, 49 deletions

diff --git a/lib/libc/sys/getsockopt.2 b/lib/libc/sys/getsockopt.2
index cefce112404..556ae778d73 100644
--- a/lib/libc/sys/getsockopt.2
+++ b/lib/libc/sys/getsockopt.2
@@ -1,4 +1,4 @@
-.\"	$OpenBSD: getsockopt.2,v 1.44 2014/03/27 13:27:28 mpi Exp $
+.\"	$OpenBSD: getsockopt.2,v 1.45 2014/03/28 08:33:51 sthen Exp $
 .\"	$NetBSD: getsockopt.2,v 1.7 1995/02/27 12:33:29 cgd Exp $
 .\"
 .\" Copyright (c) 1983, 1991, 1993
@@ -30,7 +30,7 @@
 .\"
 .\"     @(#)getsockopt.2	8.3 (Berkeley) 4/19/94
 .\"
-.Dd $Mdocdate: March 27 2014 $
+.Dd $Mdocdate: March 28 2014 $
 .Dt GETSOCKOPT 2
 .Os
 .Sh NAME
@@ -142,7 +142,7 @@ enables duplicate address and port bindings
 .It Dv SO_KEEPALIVE
 enables keep connections alive
 .It Dv SO_DONTROUTE
-enables routing bypass; not supported
+enables routing bypass for outgoing messages
 .It Dv SO_LINGER
 linger on close if data present
 .It Dv SO_BROADCAST
@@ -195,6 +195,11 @@ Should the connected party fail to respond to these messages, the connection
 is considered broken and processes using the socket are notified via a
 .Dv SIGPIPE
 signal when attempting to send data.
+.Dv SO_DONTROUTE
+indicates that outgoing messages should
+bypass the standard routing facilities.
+Instead, messages are directed to the appropriate network interface
+according to the network portion of the destination address.
 .Pp
 .Dv SO_LINGER
 controls the action taken when unsent messages
@@ -466,8 +471,6 @@ The argument
 is a file, not a socket.
 .It Bq Er ENOPROTOOPT
 The option is unknown at the level indicated.
-.It Bq Er EOPNOTSUPP
-The option is unsupported.
 .It Bq Er EFAULT
 The address pointed to by
 .Fa optval
diff --git a/lib/libc/sys/send.2 b/lib/libc/sys/send.2
index 2884dd658ed..53c30094a68 100644
--- a/lib/libc/sys/send.2
+++ b/lib/libc/sys/send.2
@@ -1,4 +1,4 @@
-.\"	$OpenBSD: send.2,v 1.28 2014/03/27 13:27:28 mpi Exp $
+.\"	$OpenBSD: send.2,v 1.29 2014/03/28 08:33:51 sthen Exp $
 .\"	$NetBSD: send.2,v 1.6 1996/01/15 01:17:18 thorpej Exp $
 .\"
 .\" Copyright (c) 1983, 1991, 1993
@@ -30,7 +30,7 @@
 .\"
 .\"     @(#)send.2	8.2 (Berkeley) 2/21/94
 .\"
-.Dd $Mdocdate: March 27 2014 $
+.Dd $Mdocdate: March 28 2014 $
 .Dt SEND 2
 .Os
 .Sh NAME
@@ -99,7 +99,7 @@ parameter may include one or more of the following:
 .It Dv MSG_OOB
 process out-of-band data
 .It Dv MSG_DONTROUTE
-bypass routing tables, silently ignored
+bypass routing, use direct interface
 .It Dv MSG_NOSIGNAL
 don't send
 .Dv SIGPIPE
@@ -116,6 +116,8 @@ data on sockets that support this notion (e.g.,
 the underlying protocol must also support
 .Dq out-of-band
 data.
+.Dv MSG_DONTROUTE
+is usually used only by diagnostic or routing programs.
 .Dv MSG_NOSIGNAL
 is used to request not to send the
 .Dv SIGPIPE
diff --git a/sys/compat/linux/linux_socket.c b/sys/compat/linux/linux_socket.c
index 8df363627cf..be4a6dd24ef 100644
--- a/sys/compat/linux/linux_socket.c
+++ b/sys/compat/linux/linux_socket.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: linux_socket.c,v 1.50 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: linux_socket.c,v 1.51 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: linux_socket.c,v 1.14 1996/04/05 00:01:50 christos Exp $	*/
 
 /*
@@ -215,6 +215,8 @@ linux_to_bsd_msg_flags(int lflags)
 		flags |= MSG_OOB;
 	if (lflags & LINUX_MSG_PEEK)
 		flags |= MSG_PEEK;
+	if (lflags & LINUX_MSG_DONTROUTE)
+		flags |= MSG_DONTROUTE;
 	if (lflags & LINUX_MSG_DONTWAIT)
 		flags |= MSG_DONTWAIT;
 	if (lflags & LINUX_MSG_WAITALL)
diff --git a/sys/kern/uipc_socket.c b/sys/kern/uipc_socket.c
index 22b23951459..583c6514ed1 100644
--- a/sys/kern/uipc_socket.c
+++ b/sys/kern/uipc_socket.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uipc_socket.c,v 1.124 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: uipc_socket.c,v 1.125 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: uipc_socket.c,v 1.21 1996/02/04 02:17:52 christos Exp $	*/
 
 /*
@@ -385,7 +385,7 @@ sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
 	struct mbuf *m;
 	long space, len, mlen, clen = 0;
 	quad_t resid;
-	int error, s;
+	int error, s, dontroute;
 	int atomic = sosendallatonce(so) || top;
 
 	if (uio)
@@ -405,6 +405,9 @@ sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
 		error = EINVAL;
 		goto out;
 	}
+	dontroute =
+	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
+	    (so->so_proto->pr_flags & PR_ATOMIC);
 	if (uio && uio->uio_procp)
 		uio->uio_procp->p_ru.ru_msgsnd++;
 	if (control) {
@@ -519,6 +522,8 @@ nopages:
 					break;
 				}
 			} while (space > 0 && atomic);
+			if (dontroute)
+				so->so_options |= SO_DONTROUTE;
 			s = splsoftnet();		/* XXX */
 			if (resid <= 0)
 				so->so_state &= ~SS_ISSENDING;
@@ -526,6 +531,8 @@ nopages:
 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
 			    top, addr, control, curproc);
 			splx(s);
+			if (dontroute)
+				so->so_options &= ~SO_DONTROUTE;
 			clen = 0;
 			control = 0;
 			top = 0;
@@ -1476,6 +1483,7 @@ sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
 		case SO_BINDANY:
 		case SO_DEBUG:
 		case SO_KEEPALIVE:
+		case SO_DONTROUTE:
 		case SO_USELOOPBACK:
 		case SO_BROADCAST:
 		case SO_REUSEADDR:
@@ -1492,15 +1500,6 @@ sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
 				so->so_options &= ~optname;
 			break;
 
-		case SO_DONTROUTE:
-			if (m == NULL || m->m_len < sizeof (int)) {
-				error = EINVAL;
-				goto bad;
-			}
-			if (*mtod(m, int *))
-				error = EOPNOTSUPP;
-			break;
-
 		case SO_SNDBUF:
 		case SO_RCVBUF:
 		case SO_SNDLOWAT:
@@ -1659,6 +1658,7 @@ sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
 
 		case SO_BINDANY:
 		case SO_USELOOPBACK:
+		case SO_DONTROUTE:
 		case SO_DEBUG:
 		case SO_KEEPALIVE:
 		case SO_REUSEADDR:
@@ -1669,10 +1669,6 @@ sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
 			*mtod(m, int *) = so->so_options & optname;
 			break;
 
-		case SO_DONTROUTE:
-			*mtod(m, int *) = 0;
-			break;
-
 		case SO_TYPE:
 			*mtod(m, int *) = so->so_type;
 			break;
diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index 28012438f98..db67fbaa95c 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in_pcb.c,v 1.148 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: in_pcb.c,v 1.149 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: in_pcb.c,v 1.25 1996/02/13 23:41:53 christos Exp $	*/
 
 /*
@@ -817,12 +817,14 @@ in_selectsrc(struct sockaddr_in *sin, struct route *ro, int soopts,
 	 * If route is known or can be allocated now,
 	 * our src addr is taken from the i/f, else punt.
 	 */
-	if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
-	    (satosin(&ro->ro_dst)->sin_addr.s_addr != sin->sin_addr.s_addr))) {
+	if (ro->ro_rt && (!(ro->ro_rt->rt_flags & RTF_UP) ||
+	    (satosin(&ro->ro_dst)->sin_addr.s_addr != sin->sin_addr.s_addr ||
+	    soopts & SO_DONTROUTE))) {
 		RTFREE(ro->ro_rt);
 		ro->ro_rt = NULL;
 	}
-	if ((ro->ro_rt == NULL || ro->ro_rt->rt_ifp == NULL)) {
+	if ((soopts & SO_DONTROUTE) == 0 && /*XXX*/
+	    (ro->ro_rt == NULL || ro->ro_rt->rt_ifp == NULL)) {
 		/* No route yet, so try to acquire one */
 		ro->ro_dst.sa_family = AF_INET;
 		ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
@@ -847,8 +849,19 @@ in_selectsrc(struct sockaddr_in *sin, struct route *ro, int soopts,
 	    !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK))
 		ia = ifatoia(ro->ro_rt->rt_ifa);
 	if (ia == 0) {
-		*errorp = EADDRNOTAVAIL;
-		return (NULL);
+		u_int16_t fport = sin->sin_port;
+
+		sin->sin_port = 0;
+		ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin), rtableid));
+		if (ia == 0)
+			ia = ifatoia(ifa_ifwithnet(sintosa(sin), rtableid));
+		sin->sin_port = fport;
+		if (ia == 0)
+			ia = TAILQ_FIRST(&in_ifaddr);
+		if (ia == 0) {
+			*errorp = EADDRNOTAVAIL;
+			return NULL;
+		}
 	}
 	return (&ia->ia_addr);
 }
diff --git a/sys/netinet/ip_divert.c b/sys/netinet/ip_divert.c
index 947f5205b2e..6ac2f0688e2 100644
--- a/sys/netinet/ip_divert.c
+++ b/sys/netinet/ip_divert.c
@@ -1,4 +1,4 @@
-/*      $OpenBSD: ip_divert.c,v 1.17 2014/03/27 13:27:28 mpi Exp $ */
+/*      $OpenBSD: ip_divert.c,v 1.18 2014/03/28 08:33:51 sthen Exp $ */
 
 /*
  * Copyright (c) 2009 Michele Marchetto <michele@openbsd.org>
@@ -180,8 +180,10 @@ divert_output(struct mbuf *m, ...)
 		schednetisr(NETISR_IP);
 		splx(s);
 	} else {
-		error = ip_output(m, NULL, &inp->inp_route,
-		    IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL);
+		error = ip_output(m, (void *)NULL, &inp->inp_route,
+		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
+		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, (void *)NULL,
+		    (void *)NULL);
 		if (error == EACCES)	/* translate pf(4) error for userland */
 			error = EHOSTUNREACH;
 	}
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index 6d490bc5700..03a6a782028 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_output.c,v 1.258 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: ip_output.c,v 1.259 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: ip_output.c,v 1.28 1996/02/13 23:43:07 christos Exp $	*/
 
 /*
@@ -194,7 +194,23 @@ ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro, int flags,
 			ro->ro_tableid = m->m_pkthdr.rdomain;
 		}
 
-		if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
+		/*
+		 * If routing to interface only, short-circuit routing lookup.
+		 */
+		if (flags & IP_ROUTETOIF) {
+			if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
+			    m->m_pkthdr.rdomain))) == 0 &&
+			    (ia = ifatoia(ifa_ifwithnet(sintosa(dst),
+			    m->m_pkthdr.rdomain))) == 0) {
+				ipstat.ips_noroute++;
+				error = ENETUNREACH;
+				goto bad;
+			}
+
+			ifp = ia->ia_ifp;
+			mtu = ifp->if_mtu;
+			ip->ip_ttl = 1;
+		} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
 		    (ip->ip_dst.s_addr == INADDR_BROADCAST)) &&
 		    imo != NULL && imo->imo_multicast_ifp != NULL) {
 			ifp = imo->imo_multicast_ifp;
@@ -344,7 +360,23 @@ reroute:
 			ro->ro_tableid = m->m_pkthdr.rdomain;
 		}
 
-		if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
+		/*
+		 * If routing to interface only, short-circuit routing lookup.
+		 */
+		if (flags & IP_ROUTETOIF) {
+			if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
+			    m->m_pkthdr.rdomain))) == 0 &&
+			    (ia = ifatoia(ifa_ifwithnet(sintosa(dst),
+			    m->m_pkthdr.rdomain))) == 0) {
+				ipstat.ips_noroute++;
+				error = ENETUNREACH;
+				goto bad;
+			}
+
+			ifp = ia->ia_ifp;
+			mtu = ifp->if_mtu;
+			ip->ip_ttl = 1;
+		} else if ((IN_MULTICAST(ip->ip_dst.s_addr) ||
 		    (ip->ip_dst.s_addr == INADDR_BROADCAST)) &&
 		    imo != NULL && imo->imo_multicast_ifp != NULL) {
 			ifp = imo->imo_multicast_ifp;
@@ -720,7 +752,7 @@ sendit:
 		ipstat.ips_fragmented++;
 
 done:
-	if (ro == &iproute && ro->ro_rt)
+	if (ro == &iproute && (flags & IP_ROUTETOIF) == 0 && ro->ro_rt)
 		RTFREE(ro->ro_rt);
 	return (error);
 bad:
diff --git a/sys/netinet/ip_var.h b/sys/netinet/ip_var.h
index 8abdd451a27..0d90a82969c 100644
--- a/sys/netinet/ip_var.h
+++ b/sys/netinet/ip_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_var.h,v 1.53 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: ip_var.h,v 1.54 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: ip_var.h,v 1.16 1996/02/13 23:43:20 christos Exp $	*/
 
 /*
@@ -140,6 +140,7 @@ struct ipq {
 /* flags passed to ip_output as last parameter */
 #define	IP_FORWARDING		0x1		/* most of ip header exists */
 #define	IP_RAWOUTPUT		0x2		/* raw ip header exists */
+#define	IP_ROUTETOIF		SO_DONTROUTE	/* bypass routing tables */
 #define	IP_ALLOWBROADCAST	SO_BROADCAST	/* can send broadcast packets */
 #define	IP_MTUDISC		0x0800		/* pmtu discovery, set DF */
 #define IP_ROUTETOETHER		0x1000		/* ether addresses given */
diff --git a/sys/netinet/raw_ip.c b/sys/netinet/raw_ip.c
index 3b6cbe590f7..91d22a49987 100644
--- a/sys/netinet/raw_ip.c
+++ b/sys/netinet/raw_ip.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: raw_ip.c,v 1.68 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: raw_ip.c,v 1.69 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: raw_ip.c,v 1.25 1996/02/18 18:58:33 christos Exp $	*/
 
 /*
@@ -218,7 +218,7 @@ rip_output(struct mbuf *m, ...)
 	va_end(ap);
 
 	inp = sotoinpcb(so);
-	flags = IP_ALLOWBROADCAST;
+	flags = (so->so_options & SO_DONTROUTE) | IP_ALLOWBROADCAST;
 
 	/*
 	 * If the user handed us a complete IP packet, use it.
diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c
index d0633905556..7eec9e2d4c6 100644
--- a/sys/netinet/tcp_output.c
+++ b/sys/netinet/tcp_output.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_output.c,v 1.102 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: tcp_output.c,v 1.103 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: tcp_output.c,v 1.16 1997/06/03 16:17:09 kml Exp $	*/
 
 /*
@@ -1076,7 +1076,9 @@ send:
 		}
 		error = ip_output(m, tp->t_inpcb->inp_options,
 			&tp->t_inpcb->inp_route,
-			(ip_mtudisc ? IP_MTUDISC : 0), NULL, tp->t_inpcb);
+			(ip_mtudisc ? IP_MTUDISC : 0) |
+				  (so->so_options & SO_DONTROUTE),
+			(void *)NULL, tp->t_inpcb);
 		break;
 #endif /* INET */
 #ifdef INET6
@@ -1097,7 +1099,8 @@ send:
 		}
 		error = ip6_output(m, tp->t_inpcb->inp_outputopts6,
 			  &tp->t_inpcb->inp_route6,
-			  0, NULL, NULL, tp->t_inpcb);
+			  (so->so_options & SO_DONTROUTE), NULL, NULL,
+			  tp->t_inpcb);
 		break;
 #endif /* INET6 */
 	}
diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index b85330eeec0..17f69795b40 100644
--- a/sys/netinet/udp_usrreq.c
+++ b/sys/netinet/udp_usrreq.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: udp_usrreq.c,v 1.175 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: udp_usrreq.c,v 1.176 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$NetBSD: udp_usrreq.c,v 1.28 1996/03/16 23:54:03 christos Exp $	*/
 
 /*
@@ -1090,8 +1090,8 @@ udp_output(struct mbuf *m, ...)
 	m->m_pkthdr.rdomain = inp->inp_rtableid;
 
 	error = ip_output(m, inp->inp_options, &inp->inp_route,
-	    (inp->inp_socket->so_options & SO_BROADCAST) | IP_IPSECFLOW,
-	    inp->inp_moptions, inp, ipsecflowinfo);
+	    (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST))
+	    |IP_IPSECFLOW, inp->inp_moptions, inp, ipsecflowinfo);
 	if (error == EACCES)	/* translate pf(4) error for userland */
 		error = EHOSTUNREACH;
 
diff --git a/sys/netinet6/in6_src.c b/sys/netinet6/in6_src.c
index 5ab4fab4bfe..127471aa920 100644
--- a/sys/netinet6/in6_src.c
+++ b/sys/netinet6/in6_src.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in6_src.c,v 1.39 2014/03/27 13:27:28 mpi Exp $	*/
+/*	$OpenBSD: in6_src.c,v 1.40 2014/03/28 08:33:51 sthen Exp $	*/
 /*	$KAME: in6_src.c,v 1.36 2001/02/06 04:08:17 itojun Exp $	*/
 
 /*
@@ -286,6 +286,25 @@ in6_selectsrc(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
 			if (ia6 == 0) /* xxx scope error ?*/
 				ia6 = ifatoia6(ro->ro_rt->rt_ifa);
 		}
+#if 0
+		/*
+		 * xxx The followings are necessary? (kazu)
+		 * I don't think so.
+		 * It's for SO_DONTROUTE option in IPv4.(jinmei)
+		 */
+		if (ia6 == 0) {
+			struct sockaddr_in6 sin6 = {sizeof(sin6), AF_INET6, 0};
+
+			sin6->sin6_addr = *dst;
+
+			ia6 = ifatoia6(ifa_ifwithdstaddr(sin6tosa(&sin6)));
+			if (ia6 == 0)
+				ia6 = ifatoia6(ifa_ifwithnet(sin6tosa(&sin6)));
+			if (ia6 == 0)
+				return (0);
+			return (&ia6->ia_addr.sin6_addr);
+		}
+#endif /* 0 */
 		if (ia6 == 0) {
 			*errorp = EHOSTUNREACH;	/* no route */
 			return (0);
diff --git a/sys/netinet6/ip6_divert.c b/sys/netinet6/ip6_divert.c
index 3d4bdbfab63..46b555ae124 100644
--- a/sys/netinet6/ip6_divert.c
+++ b/sys/netinet6/ip6_divert.c
@@ -1,4 +1,4 @@
-/*      $OpenBSD: ip6_divert.c,v 1.18 2014/03/27 13:27:28 mpi Exp $ */
+/*      $OpenBSD: ip6_divert.c,v 1.19 2014/03/28 08:33:51 sthen Exp $ */
 
 /*
  * Copyright (c) 2009 Michele Marchetto <michele@openbsd.org>
@@ -180,7 +180,8 @@ divert6_output(struct mbuf *m, ...)
 		splx(s);
 	} else {
 		error = ip6_output(m, NULL, &inp->inp_route6,
-		    IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL, NULL);
+		    ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0)
+		    | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL, NULL);
 	}
 
 	div6stat.divs_opackets++;