Extend the protosw pr_ctlinput function to include the rdomain. This is

needed so that the route and inp lookups done in TCP and UDP know where
to look. Additionally in_pcbnotifyall() and tcp_respond() got a rdomain
argument as well for similar reasons. With this tcp seems to be now
fully rdomain save and no longer leaks single packets into the main domain.
Looks good markus@, henning@

11 files changed, 59 insertions, 43 deletions

diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index 0e72496653d..333e5b3f071 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in_pcb.c,v 1.107 2009/11/03 10:59:04 claudio Exp $	*/
+/*	$OpenBSD: in_pcb.c,v 1.108 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: in_pcb.c,v 1.25 1996/02/13 23:41:53 christos Exp $	*/
 
 /*
@@ -570,9 +570,10 @@ in_setpeeraddr(inp, nam)
  * Must be called at splsoftnet.
  */
 void
-in_pcbnotifyall(table, dst, errno, notify)
+in_pcbnotifyall(table, dst, rdomain, errno, notify)
 	struct inpcbtable *table;
 	struct sockaddr *dst;
+	u_int rdomain;
 	int errno;
 	void (*notify)(struct inpcb *, int);
 {
@@ -604,6 +605,7 @@ in_pcbnotifyall(table, dst, errno, notify)
 		}
 #endif
 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
+		    inp->inp_rdomain != rdomain ||
 		    inp->inp_socket == 0) {
 			inp = CIRCLEQ_NEXT(inp, inp_queue);
 			continue;
@@ -788,7 +790,8 @@ in_pcbrtentry(inp)
 			ro->ro_dst.sa_family = AF_INET;
 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
 			satosin(&ro->ro_dst)->sin_addr = inp->inp_faddr;
-			rtalloc_mpath(ro, &inp->inp_laddr.s_addr, 0);
+			rtalloc_mpath(ro, &inp->inp_laddr.s_addr,
+			    inp->inp_rdomain);
 			break;
 		}
 	}
diff --git a/sys/netinet/in_pcb.h b/sys/netinet/in_pcb.h
index 7e75d8294c6..da1df0acdcc 100644
--- a/sys/netinet/in_pcb.h
+++ b/sys/netinet/in_pcb.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in_pcb.h,v 1.67 2009/06/05 00:05:22 claudio Exp $	*/
+/*	$OpenBSD: in_pcb.h,v 1.68 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: in_pcb.h,v 1.14 1996/02/13 23:42:00 christos Exp $	*/
 
 /*
@@ -272,7 +272,7 @@ struct inpcb *
 	 in_pcblookup(struct inpcbtable *, void *, u_int, void *,
 	    u_int, int, u_int);
 void	 in_pcbnotifyall(struct inpcbtable *, struct sockaddr *,
-	    int, void (*)(struct inpcb *, int));
+	    u_int, int, void (*)(struct inpcb *, int));
 void	 in_pcbrehash(struct inpcb *);
 void	 in_rtchange(struct inpcb *, int);
 void	 in_setpeeraddr(struct inpcb *, struct mbuf *);
diff --git a/sys/netinet/ip_icmp.c b/sys/netinet/ip_icmp.c
index c0cc5dbb89d..e1aeed69ade 100644
--- a/sys/netinet/ip_icmp.c
+++ b/sys/netinet/ip_icmp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_icmp.c,v 1.85 2009/11/03 10:59:04 claudio Exp $	*/
+/*	$OpenBSD: ip_icmp.c,v 1.86 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: ip_icmp.c,v 1.19 1996/02/13 23:42:22 christos Exp $	*/
 
 /*
@@ -308,7 +308,7 @@ icmp_input(struct mbuf *m, ...)
 	int icmplen;
 	int i;
 	struct in_ifaddr *ia;
-	void *(*ctlfunc)(int, struct sockaddr *, void *);
+	void *(*ctlfunc)(int, struct sockaddr *, u_int, void *);
 	int code;
 	extern u_char ip_protox[];
 	int hlen;
@@ -469,7 +469,8 @@ icmp_input(struct mbuf *m, ...)
 		 */
 		ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput;
 		if (ctlfunc)
-			(*ctlfunc)(code, sintosa(&icmpsrc), &icp->icmp_ip);
+			(*ctlfunc)(code, sintosa(&icmpsrc), m->m_pkthdr.rdomain,
+			    &icp->icmp_ip);
 		break;
 
 	badcode:
@@ -980,7 +981,7 @@ icmp_mtudisc_timeout(struct rtentry *rt, struct rttimer *r)
 		panic("icmp_mtudisc_timeout:  bad route to timeout");
 	if ((rt->rt_flags & (RTF_DYNAMIC | RTF_HOST)) ==
 	    (RTF_DYNAMIC | RTF_HOST)) {
-		void *(*ctlfunc)(int, struct sockaddr *, void *);
+		void *(*ctlfunc)(int, struct sockaddr *, u_int, void *);
 		extern u_char ip_protox[];
 		struct sockaddr_in sa;
 		struct rt_addrinfo info;
@@ -997,7 +998,7 @@ icmp_mtudisc_timeout(struct rtentry *rt, struct rttimer *r)
 		/* Notify TCP layer of increased Path MTU estimate */
 		ctlfunc = inetsw[ip_protox[IPPROTO_TCP]].pr_ctlinput;
 		if (ctlfunc)
-			(*ctlfunc)(PRC_MTUINC,(struct sockaddr *)&sa, NULL);
+			(*ctlfunc)(PRC_MTUINC,(struct sockaddr *)&sa, 0, NULL);
 	} else
 		if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
 			rt->rt_rmx.rmx_mtu = 0;
diff --git a/sys/netinet/ip_ipsp.h b/sys/netinet/ip_ipsp.h
index d58575fb9e6..9b214c4ea5c 100644
--- a/sys/netinet/ip_ipsp.h
+++ b/sys/netinet/ip_ipsp.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_ipsp.h,v 1.138 2009/06/02 21:28:36 blambert Exp $	*/
+/*	$OpenBSD: ip_ipsp.h,v 1.139 2009/11/13 20:54:05 claudio Exp $	*/
 /*
  * The authors of this code are John Ioannidis (ji@tla.org),
  * Angelos D. Keromytis (kermit@csd.uch.gr),
@@ -562,8 +562,8 @@ extern int ah_massage_headers(struct mbuf **, int, int, int, int);
 #ifdef INET
 extern void ah4_input(struct mbuf *, ...);
 extern int ah4_input_cb(struct mbuf *, ...);
-extern void *ah4_ctlinput(int, struct sockaddr *, void *);
-extern void *udpencap_ctlinput(int, struct sockaddr *, void *);
+extern void *ah4_ctlinput(int, struct sockaddr *, u_int, void *);
+extern void *udpencap_ctlinput(int, struct sockaddr *, u_int, void *);
 #endif /* INET */
 
 #ifdef INET6
@@ -584,7 +584,7 @@ extern int esp_sysctl(int *, u_int, void *, size_t *, void *, size_t);
 #ifdef INET
 extern void esp4_input(struct mbuf *, ...);
 extern int esp4_input_cb(struct mbuf *, ...);
-extern void *esp4_ctlinput(int, struct sockaddr *, void *);
+extern void *esp4_ctlinput(int, struct sockaddr *, u_int, void *);
 #endif /* INET */
 
 #ifdef INET6
diff --git a/sys/netinet/ipsec_input.c b/sys/netinet/ipsec_input.c
index d371367d458..acc28c6b43b 100644
--- a/sys/netinet/ipsec_input.c
+++ b/sys/netinet/ipsec_input.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ipsec_input.c,v 1.92 2009/08/09 12:47:50 henning Exp $	*/
+/*	$OpenBSD: ipsec_input.c,v 1.93 2009/11/13 20:54:05 claudio Exp $	*/
 /*
  * The authors of this code are John Ioannidis (ji@tla.org),
  * Angelos D. Keromytis (kermit@csd.uch.gr) and
@@ -746,8 +746,9 @@ ah4_input_cb(struct mbuf *m, ...)
 }
 
 
+/* XXX rdomain */
 void *
-ah4_ctlinput(int cmd, struct sockaddr *sa, void *v)
+ah4_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
 {
 	if (sa->sa_family != AF_INET ||
 	    sa->sa_len != sizeof(struct sockaddr_in))
@@ -907,8 +908,9 @@ ipsec_common_ctlinput(int cmd, struct sockaddr *sa, void *v, int proto)
 	return (NULL);
 }
 
+/* XXX rdomain */
 void *
-udpencap_ctlinput(int cmd, struct sockaddr *sa, void *v)
+udpencap_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
 {
 	struct ip *ip = v;
 	struct tdb *tdbp;
@@ -965,8 +967,9 @@ udpencap_ctlinput(int cmd, struct sockaddr *sa, void *v)
 	return (NULL);
 }
 
+/* XXX rdomain */
 void *
-esp4_ctlinput(int cmd, struct sockaddr *sa, void *v)
+esp4_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
 {
 	if (sa->sa_family != AF_INET ||
 	    sa->sa_len != sizeof(struct sockaddr_in))
diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c
index dea6cdf2923..1d112c9f5de 100644
--- a/sys/netinet/tcp_input.c
+++ b/sys/netinet/tcp_input.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_input.c,v 1.229 2009/11/03 10:59:04 claudio Exp $	*/
+/*	$OpenBSD: tcp_input.c,v 1.230 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: tcp_input.c,v 1.23 1996/02/13 23:43:44 christos Exp $	*/
 
 /*
@@ -2218,12 +2218,12 @@ dropwithreset:
 		goto drop;
 	if (tiflags & TH_ACK) {
 		tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack,
-		    TH_RST);
+		    TH_RST, 0);
 	} else {
 		if (tiflags & TH_SYN)
 			tlen++;
 		tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen,
-		    (tcp_seq)0, TH_RST|TH_ACK);
+		    (tcp_seq)0, TH_RST|TH_ACK, 0);
 	}
 	m_freem(m);
 	return;
@@ -3845,7 +3845,8 @@ syn_cache_get(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
 	return (so);
 
 resetandabort:
-	tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST);
+	tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST,
+	    m->m_pkthdr.rdomain);
 	m_freem(m);
 abort:
 	if (so != NULL)
diff --git a/sys/netinet/tcp_subr.c b/sys/netinet/tcp_subr.c
index b5d91f7253b..26cba61037b 100644
--- a/sys/netinet/tcp_subr.c
+++ b/sys/netinet/tcp_subr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_subr.c,v 1.108 2009/11/03 10:59:04 claudio Exp $	*/
+/*	$OpenBSD: tcp_subr.c,v 1.109 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: tcp_subr.c,v 1.22 1996/02/13 23:44:00 christos Exp $	*/
 
 /*
@@ -318,12 +318,13 @@ tcp_template(tp)
 /* This function looks hairy, because it was so IPv4-dependent. */
 #endif /* INET6 */
 void
-tcp_respond(tp, template, th0, ack, seq, flags)
+tcp_respond(tp, template, th0, ack, seq, flags, rdomain)
 	struct tcpcb *tp;
 	caddr_t template;
 	struct tcphdr *th0;
 	tcp_seq ack, seq;
 	int flags;
+	u_int rdomain;
 {
 	int tlen;
 	int win = 0;
@@ -409,6 +410,12 @@ tcp_respond(tp, template, th0, ack, seq, flags)
 	th->th_win = htons((u_int16_t)win);
 	th->th_urp = 0;
 
+	/* force routing domain */
+	if (tp)
+		m->m_pkthdr.rdomain = tp->t_inpcb->inp_rdomain;
+	else
+		m->m_pkthdr.rdomain = rdomain;
+
 	switch (af) {
 #ifdef INET6
 	case AF_INET6:
@@ -768,9 +775,10 @@ tcp6_ctlinput(cmd, sa, d)
 #endif
 
 void *
-tcp_ctlinput(cmd, sa, v)
+tcp_ctlinput(cmd, sa, rdomain, v)
 	int cmd;
 	struct sockaddr *sa;
+	u_int rdomain;
 	void *v;
 {
 	struct ip *ip = v;
@@ -810,7 +818,7 @@ tcp_ctlinput(cmd, sa, v)
 		seq = ntohl(th->th_seq);
 		inp = in_pcbhashlookup(&tcbtable,
 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
-		    /* XXX */ 0);
+		    rdomain);
 		if (inp && (tp = intotcpcb(inp)) &&
 		    SEQ_GEQ(seq, tp->snd_una) &&
 		    SEQ_LT(seq, tp->snd_max)) {
@@ -868,7 +876,7 @@ tcp_ctlinput(cmd, sa, v)
 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
 		inp = in_pcbhashlookup(&tcbtable,
 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
-		    /* XXX */ 0);
+		    rdomain);
 		if (inp) {
 			seq = ntohl(th->th_seq);
 			if (inp->inp_socket &&
@@ -888,10 +896,10 @@ tcp_ctlinput(cmd, sa, v)
 			sin.sin_port = th->th_sport;
 			sin.sin_addr = ip->ip_src;
 			syn_cache_unreach((struct sockaddr *)&sin,
-			    sa, th, /* XXX */ 0);
+			    sa, th, rdomain);
 		}
 	} else
-		in_pcbnotifyall(&tcbtable, sa, errno, notify);
+		in_pcbnotifyall(&tcbtable, sa, rdomain, errno, notify);
 
 	return NULL;
 }
diff --git a/sys/netinet/tcp_timer.c b/sys/netinet/tcp_timer.c
index f939e3faa0a..8253775bfa3 100644
--- a/sys/netinet/tcp_timer.c
+++ b/sys/netinet/tcp_timer.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_timer.c,v 1.43 2009/06/05 00:05:22 claudio Exp $	*/
+/*	$OpenBSD: tcp_timer.c,v 1.44 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: tcp_timer.c,v 1.14 1996/02/13 23:44:09 christos Exp $	*/
 
 /*
@@ -219,8 +219,8 @@ tcp_timer_rexmt(void *arg)
 		 * Notify all connections to the same peer about
 		 * new mss and trigger retransmit.
 		 */
-		in_pcbnotifyall(&tcbtable, sintosa(&icmpsrc), EMSGSIZE,
-		    tcp_mtudisc);
+		in_pcbnotifyall(&tcbtable, sintosa(&icmpsrc),
+		    tp->t_inpcb->inp_rdomain, EMSGSIZE, tcp_mtudisc);
 		splx(s);
 		return;
 	}
@@ -454,7 +454,7 @@ tcp_timer_keep(void *arg)
 		 */
 		tcpstat.tcps_keepprobe++;
 		tcp_respond(tp, mtod(tp->t_template, caddr_t),
-		    NULL, tp->rcv_nxt, tp->snd_una - 1, 0);
+		    NULL, tp->rcv_nxt, tp->snd_una - 1, 0, 0);
 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
 	} else
 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
diff --git a/sys/netinet/tcp_var.h b/sys/netinet/tcp_var.h
index 090ba4f6d3e..2650017254f 100644
--- a/sys/netinet/tcp_var.h
+++ b/sys/netinet/tcp_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_var.h,v 1.92 2009/08/10 10:13:43 claudio Exp $	*/
+/*	$OpenBSD: tcp_var.h,v 1.93 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: tcp_var.h,v 1.17 1996/02/13 23:44:24 christos Exp $	*/
 
 /*
@@ -558,7 +558,7 @@ int	 tcp_freeq(struct tcpcb *);
 #if defined(INET6) && !defined(TCP6)
 void	 tcp6_ctlinput(int, struct sockaddr *, void *);
 #endif
-void	 *tcp_ctlinput(int, struct sockaddr *, void *);
+void	 *tcp_ctlinput(int, struct sockaddr *, u_int, void *);
 int	 tcp_ctloutput(int, struct socket *, int, int, struct mbuf **);
 struct tcpcb *
 	 tcp_disconnect(struct tcpcb *);
@@ -588,7 +588,7 @@ void	 tcp_pulloutofband(struct socket *, u_int, struct mbuf *, int);
 int	 tcp_reass(struct tcpcb *, struct tcphdr *, struct mbuf *, int *);
 void	 tcp_rscale(struct tcpcb *, u_long);
 void	 tcp_respond(struct tcpcb *, caddr_t, struct tcphdr *, tcp_seq,
-		tcp_seq, int);
+		tcp_seq, int, u_int);
 void	 tcp_setpersist(struct tcpcb *);
 void	 tcp_slowtimo(void);
 struct mbuf *
diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index 53f253f9037..6dd612a5efe 100644
--- a/sys/netinet/udp_usrreq.c
+++ b/sys/netinet/udp_usrreq.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: udp_usrreq.c,v 1.131 2009/11/03 10:59:04 claudio Exp $	*/
+/*	$OpenBSD: udp_usrreq.c,v 1.132 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: udp_usrreq.c,v 1.28 1996/03/16 23:54:03 christos Exp $	*/
 
 /*
@@ -864,7 +864,7 @@ udp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
 #endif
 
 void *
-udp_ctlinput(int cmd, struct sockaddr *sa, void *v)
+udp_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
 {
 	struct ip *ip = v;
 	struct udphdr *uhp;
@@ -899,17 +899,17 @@ udp_ctlinput(int cmd, struct sockaddr *sa, void *v)
 		/* PMTU discovery for udpencap */
 		if (cmd == PRC_MSGSIZE && ip_mtudisc && udpencap_enable &&
 		    udpencap_port && uhp->uh_sport == htons(udpencap_port)) {
-			udpencap_ctlinput(cmd, sa, v);
+			udpencap_ctlinput(cmd, sa, rdomain, v);
 			return (NULL);
 		}
 #endif
 		inp = in_pcbhashlookup(&udbtable,
 		    ip->ip_dst, uhp->uh_dport, ip->ip_src, uhp->uh_sport,
-		    /* XXX */ 0);
+		    rdomain);
 		if (inp && inp->inp_socket != NULL)
 			notify(inp, errno);
 	} else
-		in_pcbnotifyall(&udbtable, sa, errno, notify);
+		in_pcbnotifyall(&udbtable, sa, rdomain, errno, notify);
 	return (NULL);
 }
 
diff --git a/sys/netinet/udp_var.h b/sys/netinet/udp_var.h
index 5241be127db..68f18b0e70d 100644
--- a/sys/netinet/udp_var.h
+++ b/sys/netinet/udp_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: udp_var.h,v 1.19 2008/05/24 19:48:32 thib Exp $	*/
+/*	$OpenBSD: udp_var.h,v 1.20 2009/11/13 20:54:05 claudio Exp $	*/
 /*	$NetBSD: udp_var.h,v 1.12 1996/02/13 23:44:41 christos Exp $	*/
 
 /*
@@ -106,7 +106,7 @@ extern struct	udpstat udpstat;
 void	udp6_ctlinput(int, struct sockaddr *, void *);
 int	udp6_input(struct mbuf **, int *, int);
 #endif /* INET6 && !TCP6 */
-void	 *udp_ctlinput(int, struct sockaddr *, void *);
+void	 *udp_ctlinput(int, struct sockaddr *, u_int, void *);
 void	 udp_init(void);
 void	 udp_input(struct mbuf *, ...);
 #ifdef INET6