? netbsd.netinet.ip_pktinfo.patch
Index: in.h
===================================================================
RCS file: /src/cvs/cvsroot/netbsd/src/sys/netinet/in.h,v
retrieving revision 1.83
diff -a -u -r1.83 in.h
--- in.h	25 Jan 2008 21:12:14 -0000	1.83
+++ in.h	24 Jul 2008 17:08:57 -0000
@@ -282,6 +282,8 @@
 #if 1 /*IPSEC*/
 #define	IP_IPSEC_POLICY		22 /* struct; get/set security policy */
 #endif
+#define	IP_RECVPKTINFO		23   /* bool; recv if, dst addr */
+#define	IP_PKTINFO		24   /* in_pktinfo; send if, src addr */
 
 /*
  * Defaults and limits for options
@@ -306,6 +308,15 @@
 #define	IP_PORTRANGE_HIGH	1	/* same as DEFAULT (FreeBSD compat) */
 #define	IP_PORTRANGE_LOW	2	/* use privileged range */
 
+/*
+ * Argument structure for IP_PKTINFO
+ */
+struct in_pktinfo {
+	int ipi_ifindex;		/* send interface index (XXX: doesn't work) */
+	struct in_addr ipi_spec_dst;	/* local address for lookup rtentry (XXX: doesn't work) */
+	struct in_addr ipi_addr;	/* src IP address */
+};
+
 #if defined(_NETBSD_SOURCE)
 /*
  * Definitions for inet sysctl operations.
Index: in_pcb.h
===================================================================
RCS file: /src/cvs/cvsroot/netbsd/src/sys/netinet/in_pcb.h,v
retrieving revision 1.45
diff -a -u -r1.45 in_pcb.h
--- in_pcb.h	16 Dec 2007 18:39:57 -0000	1.45
+++ in_pcb.h	24 Jul 2008 17:08:57 -0000
@@ -108,8 +108,9 @@
 /* XXX should move to an UDP control block */
 #define INP_ESPINUDP		0x100	/* ESP over UDP for NAT-T */
 #define INP_ESPINUDP_NON_IKE	0x200	/* ESP over UDP for NAT-T */
+#define INP_PKTINFO		0x800	/* receive with ip_pktinfo */
 #define	INP_CONTROLOPTS		(INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\
-				INP_RECVIF)
+				INP_RECVIF|INP_PKTINFO)
 #define INP_ESPINUDP_ALL	(INP_ESPINUDP|INP_ESPINUDP_NON_IKE)
 #define INP_NOHEADER		0x400	/* Kernel removes IP header
 					 * before feeding a packet
Index: ip_input.c
===================================================================
RCS file: /src/cvs/cvsroot/netbsd/src/sys/netinet/ip_input.c,v
retrieving revision 1.272
diff -a -u -r1.272 ip_input.c
--- ip_input.c	5 May 2008 17:11:17 -0000	1.272
+++ ip_input.c	24 Jul 2008 17:08:57 -0000
@@ -2097,6 +2097,18 @@
 		if (*mp)
 			mp = &(*mp)->m_next;
 	}
+	if (inp->inp_flags & INP_PKTINFO) {
+		struct in_pktinfo pinfo;
+
+		pinfo.ipi_ifindex = (m->m_pkthdr.rcvif != NULL) ?
+		    m->m_pkthdr.rcvif->if_index : 0;
+		pinfo.ipi_spec_dst = pinfo.ipi_addr = ip->ip_dst;
+
+		*mp = sbcreatecontrol((void *) &pinfo,
+		    sizeof(pinfo), IP_PKTINFO, IPPROTO_IP);
+		if (*mp)
+			mp = &(*mp)->m_next;
+	}
 }
 
 /*
Index: ip_output.c
===================================================================
RCS file: /src/cvs/cvsroot/netbsd/src/sys/netinet/ip_output.c,v
retrieving revision 1.196
diff -a -u -r1.196 ip_output.c
--- ip_output.c	28 Apr 2008 20:24:09 -0000	1.196
+++ ip_output.c	24 Jul 2008 17:08:57 -0000
@@ -1232,6 +1232,8 @@
 		case IP_RECVRETOPTS:
 		case IP_RECVDSTADDR:
 		case IP_RECVIF:
+		case IP_RECVPKTINFO:
+		case IP_PKTINFO:	/* for compatibile linux */
 			if (m == NULL || m->m_len != sizeof(int))
 				error = EINVAL;
 			else {
@@ -1266,6 +1268,11 @@
 				case IP_RECVIF:
 					OPTSET(INP_RECVIF);
 					break;
+
+				case IP_RECVPKTINFO:
+				case IP_PKTINFO:	/* for compatibile linux */
+					OPTSET(INP_PKTINFO);
+					break;
 				}
 			}
 			break;
@@ -1356,6 +1363,8 @@
 		case IP_RECVRETOPTS:
 		case IP_RECVDSTADDR:
 		case IP_RECVIF:
+		case IP_RECVPKTINFO:
+		case IP_PKTINFO:	/* for compatible linux */
 		case IP_ERRORMTU:
 			*mp = m = m_get(M_WAIT, MT_SOOPTS);
 			MCLAIM(m, so->so_mowner);
@@ -1391,6 +1400,11 @@
 			case IP_RECVIF:
 				optval = OPTBIT(INP_RECVIF);
 				break;
+
+			case IP_RECVPKTINFO:
+			case IP_PKTINFO:	/* for compatible linux */
+				optval = OPTBIT(INP_PKTINFO);
+				break;
 			}
 			*mtod(m, int *) = optval;
 			break;
@@ -1548,6 +1562,106 @@
 	return (EINVAL);
 }
 
+void
+ip_initpktopts(struct ip_pktopts *pktopt)
+{
+	memset(pktopt, 0, sizeof(*pktopt));
+}
+
+int
+ip_copypktopts(struct ip_pktopts *dst, struct ip_pktopts *src, int canwait)
+{
+	memcpy(dst, src, sizeof(*dst));
+	return 0;
+}
+
+void
+ip_clearpktopts(struct ip_pktopts *pktopt, int optname)
+{
+	if (optname == -1 || optname == IP_PKTINFO) {
+		memset(&pktopt->ipo_pktinfo, 0, sizeof(pktopt->ipo_pktinfo));
+	}
+}
+
+int
+ip_setpktopts(struct mbuf *control, struct ip_pktopts *opt,
+    struct ip_pktopts *stickyopt, int priv, int uproto)
+{
+	struct cmsghdr *cm = 0;
+
+	if (control == NULL || opt == NULL)
+		return EINVAL;
+
+	ip_initpktopts(opt);
+	if (stickyopt) {
+		int error;
+		if ((error = ip_copypktopts(opt, stickyopt, M_NOWAIT)) != 0)
+			return error;
+	}
+
+	/*
+	 * XXX: Currently, we assume all the optional information is stored
+	 * in a single mbuf.
+	 */
+	if (control->m_next)
+		return (EINVAL);
+
+	for (; control->m_len; control->m_data += CMSG_ALIGN(cm->cmsg_len),
+	    control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
+		int error;
+
+		if (control->m_len < CMSG_LEN(0))
+			return (EINVAL);
+
+		cm = mtod(control, struct cmsghdr *);
+		if (cm->cmsg_len == 0 || cm->cmsg_len > control->m_len)
+			return (EINVAL);
+		if (cm->cmsg_level != IPPROTO_IP)
+			continue;
+
+		error = ip_setpktopt(cm->cmsg_type, CMSG_DATA(cm),
+		    cm->cmsg_len - CMSG_LEN(0), opt, priv, 0, 1, uproto);
+		if (error)
+			return (error);
+	}
+
+	return (0);
+
+}
+
+int
+ip_setpktopt(int optname, u_char *buf, int len, struct ip_pktopts *pktopt,
+    int priv, int sticky, int cmsg, int uproto)
+{
+	if (!sticky && !cmsg) {
+#ifdef DIAGNOSTIC
+		printf("ip_setpktopt: impossible case\n");
+#endif
+		return EINVAL;
+	}
+
+	if (sticky && cmsg) {
+		switch (optname) {
+		case IP_PKTINFO:
+			return ENOPROTOOPT;
+		}
+	}
+
+	switch (optname) {
+	case IP_PKTINFO:
+		{
+			struct in_pktinfo *p;
+			p = (struct in_pktinfo *)buf;
+			pktopt->ipo_pktinfo = *p;
+		}
+		break;
+	default:
+		return ENOPROTOOPT;
+	}
+
+	return 0;
+}
+
 /*
  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
  */
Index: ip_var.h
===================================================================
RCS file: /src/cvs/cvsroot/netbsd/src/sys/netinet/ip_var.h,v
retrieving revision 1.87
diff -a -u -r1.87 ip_var.h
--- ip_var.h	12 Apr 2008 05:58:22 -0000	1.87
+++ ip_var.h	24 Jul 2008 17:08:57 -0000
@@ -110,6 +110,14 @@
 	int8_t	ipopt_list[MAX_IPOPTLEN];	/* options proper */
 };
 
+struct ip_pktopts {
+	/*
+	 * XXX: each member should be a pointer.
+	 *      and NULL means not defined
+	 */
+	struct in_pktinfo ipo_pktinfo;	/* Outgoing IF/address information */
+};
+
 /*
  * Structure attached to inpcb.ip_moptions and
  * passed to ip_output when IP multicast options are in use.
@@ -214,6 +222,15 @@
 int	 ip_output(struct mbuf *, ...);
 int	 ip_fragment(struct mbuf *, struct ifnet *, u_long);
 int	 ip_pcbopts(struct mbuf **, struct mbuf *);
+
+int	 ip_pktopt(struct ip_pktopts **, struct mbuf *, struct socket *);
+void	 ip_initpktopts(struct ip_pktopts *);
+int	 ip_copypktopts(struct ip_pktopts *, struct ip_pktopts *, int);
+void	 ip_clearpktopts(struct ip_pktopts *, int);
+int	 ip_setpktopts(struct mbuf *, struct ip_pktopts *,
+	    struct ip_pktopts *, int, int );
+int	 ip_setpktopt(int, u_char *, int, struct ip_pktopts *, int, int, int, int);
+
 struct mbuf *
 	 ip_reass(struct ipqent *, struct ipq *, struct ipqhead *);
 struct in_ifaddr *
Index: udp_usrreq.c
===================================================================
RCS file: /src/cvs/cvsroot/netbsd/src/sys/netinet/udp_usrreq.c,v
retrieving revision 1.172
diff -a -u -r1.172 udp_usrreq.c
--- udp_usrreq.c	4 May 2008 07:22:14 -0000	1.172
+++ udp_usrreq.c	24 Jul 2008 17:08:57 -0000
@@ -1275,41 +1275,57 @@
 		break;
 
 	case PRU_SEND:
-		if (control && control->m_len) {
-			m_freem(control);
-			m_freem(m);
-			error = EINVAL;
-			break;
-		}
-	{
-		struct in_addr laddr;			/* XXX */
+		{
+			struct in_addr laddr;			/* XXX */
+			struct ip_pktopts opt;
+
+			laddr.s_addr = 0;
+			if (nam || control)
+				laddr = inp->inp_laddr;		/* XXX: save */
+
+			if (control) {
+				if ((error = ip_setpktopts(control, &opt,
+				    NULL, 0, IPPROTO_UDP)) != 0)
+					goto die;
+
+				if (opt.ipo_pktinfo.ipi_addr.s_addr) {	/* XXX */
+					struct in_ifaddr *ia;
+					LIST_FOREACH(ia, &IN_IFADDR_HASH(opt.ipo_pktinfo.ipi_addr.s_addr), ia_hash) {
+						if (in_hosteq(ia->ia_addr.sin_addr, opt.ipo_pktinfo.ipi_addr)) {
+							inp->inp_laddr = opt.ipo_pktinfo.ipi_addr;	/* XXX */
+						}
+					}
+				}
+			}
 
-		if (nam) {
-			laddr = inp->inp_laddr;		/* XXX */
-			if ((so->so_state & SS_ISCONNECTED) != 0) {
-				error = EISCONN;
-				goto die;
+			if (nam) {
+				if ((so->so_state & SS_ISCONNECTED) != 0) {
+					error = EISCONN;
+					goto die;
+				}
+				error = in_pcbconnect(inp, nam, l);	/* would break inp->inp_laddr */
+				if (error)
+					goto die;
+			} else {
+				if ((so->so_state & SS_ISCONNECTED) == 0) {
+					error = ENOTCONN;
+					goto die;
+				}
 			}
-			error = in_pcbconnect(inp, nam, l);
-			if (error)
-				goto die;
-		} else {
-			if ((so->so_state & SS_ISCONNECTED) == 0) {
-				error = ENOTCONN;
-				goto die;
+			error = udp_output(m, inp);
+			m = NULL;
+			if (nam || control)
+				inp->inp_laddr = laddr;		/* XXX: restore */
+			if (nam) {
+				in_pcbdisconnect(inp);
+				in_pcbstate(inp, INP_BOUND);	/* XXX */
 			}
+		die:
+			if (control)
+				m_freem(control);
+			if (m)
+				m_freem(m);
 		}
-		error = udp_output(m, inp);
-		m = NULL;
-		if (nam) {
-			in_pcbdisconnect(inp);
-			inp->inp_laddr = laddr;		/* XXX */
-			in_pcbstate(inp, INP_BOUND);	/* XXX */
-		}
-	  die:
-		if (m)
-			m_freem(m);
-	}
 		break;
 
 	case PRU_SENSE: