The patch provides the same socket option as Linux IP_ORIGDSTADDR.

Unfortunately they will have different integer value due to Linux value being already assigned in FreeBSD.

The patch is similar to IP_RECVDSTADDR but also provides the destination port value to the application.

This allows/improves implementation of transparent proxies on UDP sockets due to having the whole information on forwarded packets.

Reviewed by:	adrian, aw
Approved by:	ae (mentor)
Sponsored by:	rsync.net
Differential Revision:	D9235

share/man/man4/ip.4

@@ -136,6 +136,37 @@ determined by the destination address, returns an
 error.
 .Pp
 If the
+.Dv IP_ORIGDSTADDR
+option is enabled on a
+.Dv SOCK_DGRAM
+socket,
+the
+.Xr recvmsg 2
+call will return the destination
+.Tn IP
+address and destination port or a
+.Tn UDP
+datagram.
+The
+.Vt msg_control
+field in the
+.Vt msghdr
+structure points to a buffer
+that contains a
+.Vt cmsghdr
+structure followed by the
+.Tn in_sockkaddr
+structre.
+The
+.Vt cmsghdr
+fields have the following values:
+.Bd -literal
+cmsg_len = CMSG_LEN(sizeof(struct in_sockaddr))
+cmsg_level = IPPROTO_IP
+cmsg_type = IP_ORIGDSTADDR
+.Ed
+.Pp
+If the
 .Dv IP_RECVDSTADDR
 option is enabled on a
 .Dv SOCK_DGRAM

share/man/man4/ip6.4

@@ -156,6 +156,9 @@ datagrams sent on this socket.
 .\" .It Dv IPV6_RECVDSTADDR Fa "int *"
 .\" Get or set the status of whether datagrams are received with
 .\" destination addresses.
+.\" .It Dv IPV6_ORIGDSTADDR Fa "int *"
+.\" Get or set the status of whether datagrams are received with
+.\" destination addresses and destination ports.
 .\" .It Dv IPV6_RETOPTS
 .\" Get or set IPv6 options.
 .It Dv IPV6_MULTICAST_IF Fa "u_int *"

sys/netinet/in.h

@@ -433,6 +433,8 @@ __END_DECLS
 #define	IP_BINDANY		24   /* bool: allow bind to any address */
 #define	IP_BINDMULTI		25   /* bool: allow multiple listeners on a tuple */
 #define	IP_RSS_LISTEN_BUCKET	26   /* int; set RSS listen bucket */
+#define	IP_ORIGDSTADDR		27   /* bool: receive IP dst addr/port w/dgram */
+#define	IP_RECVORIGDSTADDR      IP_ORIGDSTADDR
 
 /*
  * Options for controlling the firewall and dummynet.

sys/netinet/in_pcb.c

@@ -2492,6 +2492,10 @@ db_print_inpflags(int inp_flags)
 		db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
 		comma = 1;
 	}
+	if (inp_flags & INP_ORIGDSTADDR) {
+		db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
+		comma = 1;
+	}
 	if (inp_flags & INP_HDRINCL) {
 		db_printf("%sINP_HDRINCL", comma ? ", " : "");
 		comma = 1;

sys/netinet/in_pcb.h

@@ -618,6 +618,7 @@ short	inp_so_options(const struct inpcb *inp);
 #define	INP_RECVFLOWID		0x00000100 /* populate recv datagram with flow info */
 #define	INP_RECVRSSBUCKETID	0x00000200 /* populate recv datagram with bucket id */
 #define	INP_RATE_LIMIT_CHANGED	0x00000400 /* rate limit needs attention */
+#define	INP_ORIGDSTADDR		0x00000800 /* receive IP dst address/port */
 
 /*
  * Flags passed to in_pcblookup*() functions.

sys/netinet/ip_output.c

@@ -1065,6 +1065,7 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 		case IP_MINTTL:
 		case IP_RECVOPTS:
 		case IP_RECVRETOPTS:
+		case IP_ORIGDSTADDR:
 		case IP_RECVDSTADDR:
 		case IP_RECVTTL:
 		case IP_RECVIF:
@@ -1126,6 +1127,10 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 				OPTSET(INP_RECVDSTADDR);
 				break;
 
+			case IP_ORIGDSTADDR:
+				OPTSET2(INP_ORIGDSTADDR, optval);
+				break;
+
 			case IP_RECVTTL:
 				OPTSET(INP_RECVTTL);
 				break;
@@ -1258,6 +1263,7 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 		case IP_MINTTL:
 		case IP_RECVOPTS:
 		case IP_RECVRETOPTS:
+		case IP_ORIGDSTADDR:
 		case IP_RECVDSTADDR:
 		case IP_RECVTTL:
 		case IP_RECVIF:
@@ -1303,6 +1309,10 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 				optval = OPTBIT(INP_RECVDSTADDR);
 				break;
 
+			case IP_ORIGDSTADDR:
+				optval = OPTBIT2(INP_ORIGDSTADDR);
+				break;
+
 			case IP_RECVTTL:
 				optval = OPTBIT(INP_RECVTTL);
 				break;

sys/netinet/udp_usrreq.c

@@ -304,7 +304,7 @@ udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 {
 	struct sockaddr *append_sa;
 	struct socket *so;
-	struct mbuf *opts = NULL;
+	struct mbuf *tmpopts, *opts = NULL;
 #ifdef INET6
 	struct sockaddr_in6 udp_in6;
 #endif
@@ -319,7 +319,7 @@ udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 	if (up->u_tun_func != NULL) {
 		in_pcbref(inp);
 		INP_RUNLOCK(inp);
-		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in,
+		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0],
 		    up->u_tun_ctx);
 		INP_RLOCK(inp);
 		return (in_pcbrele_rlocked(inp));
@@ -355,16 +355,27 @@ udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off,
 #endif /* INET6 */
 			ip_savecontrol(inp, &opts, ip, n);
 	}
+	if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
+		tmpopts = sbcreatecontrol((caddr_t)&udp_in[1],
+			sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP);
+		if (tmpopts) {
+			if (opts) {
+				tmpopts->m_next = opts;
+				opts = tmpopts;
+			} else
+				opts = tmpopts;
+		}
+	}
 #ifdef INET6
 	if (inp->inp_vflag & INP_IPV6) {
 		bzero(&udp_in6, sizeof(udp_in6));
 		udp_in6.sin6_len = sizeof(udp_in6);
 		udp_in6.sin6_family = AF_INET6;
-		in6_sin_2_v4mapsin6(udp_in, &udp_in6);
+		in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6);
 		append_sa = (struct sockaddr *)&udp_in6;
 	} else
 #endif /* INET6 */
-		append_sa = (struct sockaddr *)udp_in;
+		append_sa = (struct sockaddr *)&udp_in[0];
 	m_adj(n, off);
 
 	so = inp->inp_socket;
@@ -390,7 +401,7 @@ udp_input(struct mbuf **mp, int *offp, int proto)
 	uint16_t len, ip_len;
 	struct inpcbinfo *pcbinfo;
 	struct ip save_ip;
-	struct sockaddr_in udp_in;
+	struct sockaddr_in udp_in[2];
 	struct mbuf *m;
 	struct m_tag *fwd_tag;
 	int cscov_partial, iphlen;
@@ -435,11 +446,15 @@ udp_input(struct mbuf **mp, int *offp, int proto)
 	 * Construct sockaddr format source address.  Stuff source address
 	 * and datagram in user buffer.
 	 */
-	bzero(&udp_in, sizeof(udp_in));
+	bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2);
-	udp_in.sin_len = sizeof(udp_in);
+	udp_in[0].sin_len = sizeof(struct sockaddr_in);
-	udp_in.sin_family = AF_INET;
+	udp_in[0].sin_family = AF_INET;
-	udp_in.sin_port = uh->uh_sport;
+	udp_in[0].sin_port = uh->uh_sport;
-	udp_in.sin_addr = ip->ip_src;
+	udp_in[0].sin_addr = ip->ip_src;
+	udp_in[1].sin_len = sizeof(struct sockaddr_in);
+	udp_in[1].sin_family = AF_INET;
+	udp_in[1].sin_port = uh->uh_dport;
+	udp_in[1].sin_addr = ip->ip_dst;
 
 	/*
 	 * Make mbuf data length reflect UDP length.  If not enough data to
@@ -568,7 +583,7 @@ udp_input(struct mbuf **mp, int *offp, int proto)
 
 				blocked = imo_multi_filter(imo, ifp,
 					(struct sockaddr *)&group,
-					(struct sockaddr *)&udp_in);
+					(struct sockaddr *)&udp_in[0]);
 				if (blocked != MCAST_PASS) {
 					if (blocked == MCAST_NOTGMEMBER)
 						IPSTAT_INC(ips_notmember);
@@ -587,7 +602,7 @@ udp_input(struct mbuf **mp, int *offp, int proto)
 					UDP_PROBE(receive, NULL, last, ip,
 					    last, uh);
 					if (udp_append(last, ip, n, iphlen,
-						&udp_in)) {
+						udp_in)) {
 						goto inp_lost;
 					}
 				}
@@ -620,7 +635,7 @@ udp_input(struct mbuf **mp, int *offp, int proto)
 			goto badunlocked;
 		}
 		UDP_PROBE(receive, NULL, last, ip, last, uh);
-		if (udp_append(last, ip, m, iphlen, &udp_in) == 0) 
+		if (udp_append(last, ip, m, iphlen, udp_in) == 0) 
 			INP_RUNLOCK(last);
 	inp_lost:
 		INP_INFO_RUNLOCK(pcbinfo);
@@ -710,7 +725,7 @@ udp_input(struct mbuf **mp, int *offp, int proto)
 	}
 
 	UDP_PROBE(receive, NULL, inp, ip, inp, uh);
-	if (udp_append(inp, ip, m, iphlen, &udp_in) == 0) 
+	if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 
 		INP_RUNLOCK(inp);
 	return (IPPROTO_DONE);
 

sys/netinet6/in6.h

@@ -497,6 +497,9 @@ struct route_in6 {
 #define	IPV6_RECVFLOWID		70 /* bool; receive IP6 flowid/flowtype w/ datagram */
 #define	IPV6_RECVRSSBUCKETID	71 /* bool; receive IP6 RSS bucket id w/ datagram */
 
+#define	IPV6_ORIGDSTADDR	72 /* bool: allow getting dstaddr /port info */
+#define	IPV6_RECVORIGDSTADDR	IPV6_ORIGDSTADDR
+
 /*
  * The following option is private; do not use it from user applications.
  * It is deliberately defined to the same value as IP_MSFILTER.

sys/netinet6/in6_pcb.c

@@ -1267,7 +1267,7 @@ in6_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in6_addr *faddr,
 }
 
 void
-init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m)
+init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m, int srcordst)
 {
 	struct ip6_hdr *ip;
 
@@ -1275,7 +1275,7 @@ init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m)
 	bzero(sin6, sizeof(*sin6));
 	sin6->sin6_len = sizeof(*sin6);
 	sin6->sin6_family = AF_INET6;
-	sin6->sin6_addr = ip->ip6_src;
+	sin6->sin6_addr = srcordst ? ip->ip6_dst : ip->ip6_src;
 
 	(void)sa6_recoverscope(sin6); /* XXX: should catch errors... */
 

sys/netinet6/in6_pcb.h

@@ -113,7 +113,7 @@ int	in6_mapped_sockaddr(struct socket *so, struct sockaddr **nam);
 int	in6_mapped_peeraddr(struct socket *so, struct sockaddr **nam);
 int	in6_selecthlim(struct in6pcb *, struct ifnet *);
 int	in6_pcbsetport(struct in6_addr *, struct inpcb *, struct ucred *);
-void	init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m);
+void	init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m, int);
 #endif /* _KERNEL */
 
 #endif /* !_NETINET6_IN6_PCB_H_ */

sys/netinet6/ip6_output.c

@@ -1545,6 +1545,7 @@ ip6_ctloutput(struct socket *so, struct sockopt *sopt)
 #endif
 			case IPV6_V6ONLY:
 			case IPV6_AUTOFLOWLABEL:
+			case IPV6_ORIGDSTADDR:
 			case IPV6_BINDANY:
 			case IPV6_BINDMULTI:
 #ifdef	RSS
@@ -1730,6 +1731,9 @@ do { \
 					OPTSET(IN6P_AUTOFLOWLABEL);
 					break;
 
+				case IPV6_ORIGDSTADDR:
+					OPTSET2(INP_ORIGDSTADDR, optval);
+					break;
 				case IPV6_BINDANY:
 					OPTSET(INP_BINDANY);
 					break;
@@ -2018,6 +2022,10 @@ do { \
 					optval = OPTBIT(IN6P_AUTOFLOWLABEL);
 					break;
 
+				case IPV6_ORIGDSTADDR:
+					optval = OPTBIT2(INP_ORIGDSTADDR);
+					break;
+
 				case IPV6_BINDANY:
 					optval = OPTBIT(INP_BINDANY);
 					break;

sys/netinet6/raw_ip6.c

@@ -166,7 +166,7 @@ rip6_input(struct mbuf **mp, int *offp, int proto)
 
 	RIP6STAT_INC(rip6s_ipackets);
 
-	init_sin6(&fromsa, m); /* general init */
+	init_sin6(&fromsa, m, 0); /* general init */
 
 	ifp = m->m_pkthdr.rcvif;
 

sys/netinet6/udp6_usrreq.c

@@ -137,7 +137,7 @@ udp6_append(struct inpcb *inp, struct mbuf *n, int off,
     struct sockaddr_in6 *fromsa)
 {
 	struct socket *so;
-	struct mbuf *opts;
+	struct mbuf *opts = NULL, *tmp_opts;
 	struct udpcb *up;
 
 	INP_LOCK_ASSERT(inp);
@@ -149,7 +149,7 @@ udp6_append(struct inpcb *inp, struct mbuf *n, int off,
 	if (up->u_tun_func != NULL) {
 		in_pcbref(inp);
 		INP_RUNLOCK(inp);
-		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)fromsa,
+		(*up->u_tun_func)(n, off, inp, (struct sockaddr *)&fromsa[0],
 		    up->u_tun_ctx);
 		INP_RLOCK(inp);
 		return (in_pcbrele_rlocked(inp));
@@ -173,11 +173,23 @@ udp6_append(struct inpcb *inp, struct mbuf *n, int off,
 	if (inp->inp_flags & INP_CONTROLOPTS ||
 	    inp->inp_socket->so_options & SO_TIMESTAMP)
 		ip6_savecontrol(inp, n, &opts);
+	if ((inp->inp_vflag & INP_IPV6) && (inp->inp_flags2 & INP_ORIGDSTADDR)) {
+		tmp_opts = sbcreatecontrol((caddr_t)&fromsa[1],
+                        sizeof(struct sockaddr_in6), IPV6_ORIGDSTADDR, IPPROTO_IPV6);
+                if (tmp_opts) {
+                        if (opts) {
+                                tmp_opts->m_next = opts;
+                                opts = tmp_opts;
+                        } else
+                                opts = tmp_opts;
+                }
+
+	}
 	m_adj(n, off + sizeof(struct udphdr));
 
 	so = inp->inp_socket;
 	SOCKBUF_LOCK(&so->so_rcv);
-	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)fromsa, n,
+	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&fromsa[0], n,
 	    opts) == 0) {
 		SOCKBUF_UNLOCK(&so->so_rcv);
 		m_freem(n);
@@ -202,7 +214,7 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 	int off = *offp;
 	int cscov_partial;
 	int plen, ulen;
-	struct sockaddr_in6 fromsa;
+	struct sockaddr_in6 fromsa[2];
 	struct m_tag *fwd_tag;
 	uint16_t uh_sum;
 	uint8_t nxt;
@@ -277,8 +289,10 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 	/*
 	 * Construct sockaddr format source address.
 	 */
-	init_sin6(&fromsa, m);
+	init_sin6(&fromsa[0], m, 0);
-	fromsa.sin6_port = uh->uh_sport;
+	fromsa[0].sin6_port = uh->uh_sport;
+	init_sin6(&fromsa[1], m, 1);
+	fromsa[1].sin6_port = uh->uh_dport;
 
 	pcbinfo = udp_get_inpcbinfo(nxt);
 	if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
@@ -349,7 +363,7 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 
 				blocked = im6o_mc_filter(imo, ifp,
 					(struct sockaddr *)&mcaddr,
-					(struct sockaddr *)&fromsa);
+					(struct sockaddr *)&fromsa[0]);
 				if (blocked != MCAST_PASS) {
 					if (blocked == MCAST_NOTGMEMBER)
 						IP6STAT_INC(ip6s_notmember);
@@ -370,7 +384,7 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 					INP_RLOCK(last);
 					UDP_PROBE(receive, NULL, last, ip6,
 					    last, uh);
-					if (udp6_append(last, n, off, &fromsa))
+					if (udp6_append(last, n, off, fromsa))
 						goto inp_lost;
 					INP_RUNLOCK(last);
 				}
@@ -402,7 +416,7 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 		INP_RLOCK(last);
 		INP_INFO_RUNLOCK(pcbinfo);
 		UDP_PROBE(receive, NULL, last, ip6, last, uh);
-		if (udp6_append(last, m, off, &fromsa) == 0) 
+		if (udp6_append(last, m, off, fromsa) == 0) 
 			INP_RUNLOCK(last);
 	inp_lost:
 		return (IPPROTO_DONE);
@@ -482,7 +496,7 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 		}
 	}
 	UDP_PROBE(receive, NULL, inp, ip6, inp, uh);
-	if (udp6_append(inp, m, off, &fromsa) == 0)
+	if (udp6_append(inp, m, off, fromsa) == 0)
 		INP_RUNLOCK(inp);
 	return (IPPROTO_DONE);