MFC r261708, r261847, r268525, r274316, r274347, r275593,

    r276844, r276847, r279531, r279559, r279564, r279676

A bunch of IPv6 fixes by melifaro, hrs and ae

Major changes:
Simplify nd6_output_lle()
Add refcounting to DAD and fix races and other errors
Implement Enhanced DAD algorithm for IPv6

Suggested by: 	ae
Tested by: 	Jason Wolfe <j at nitrology.com>
Sponsored by:	Limelight Networks

8 files changed, 481 insertions, 418 deletions

diff --git a/sbin/ifconfig/af_inet6.c b/sbin/ifconfig/af_inet6.c
index 0f8688a..321bf5b 100644
--- a/sbin/ifconfig/af_inet6.c
+++ b/sbin/ifconfig/af_inet6.c
@@ -483,6 +483,10 @@ static struct cmd inet6_cmds[] = {
 	DEF_CMD("-auto_linklocal",-ND6_IFF_AUTO_LINKLOCAL,setnd6flags),
 	DEF_CMD("no_prefer_iface",ND6_IFF_NO_PREFER_IFACE,setnd6flags),
 	DEF_CMD("-no_prefer_iface",-ND6_IFF_NO_PREFER_IFACE,setnd6flags),
+	DEF_CMD("no_dad",	ND6_IFF_NO_DAD,		setnd6flags),
+	DEF_CMD("-no_dad",	-ND6_IFF_NO_DAD,	setnd6flags),
+	DEF_CMD("ignoreloop",	ND6_IFF_IGNORELOOP,	setnd6flags),
+	DEF_CMD("-ignoreloop",	-ND6_IFF_IGNORELOOP,	setnd6flags),
 	DEF_CMD_ARG("pltime",        			setip6pltime),
 	DEF_CMD_ARG("vltime",        			setip6vltime),
 	DEF_CMD("eui64",	0,			setip6eui64),
diff --git a/sbin/ifconfig/af_nd6.c b/sbin/ifconfig/af_nd6.c
index b3db0a8..3a510a5 100644
--- a/sbin/ifconfig/af_nd6.c
+++ b/sbin/ifconfig/af_nd6.c
@@ -58,7 +58,8 @@ static const char rcsid[] =
 #define	MAX_SYSCTL_TRY	5
 #define	ND6BITS	"\020\001PERFORMNUD\002ACCEPT_RTADV\003PREFER_SOURCE" \
 		"\004IFDISABLED\005DONT_SET_IFROUTE\006AUTO_LINKLOCAL" \
-		"\007NO_RADR\010NO_PREFER_IFACE\020DEFAULTIF"
+		"\007NO_RADR\010NO_PREFER_IFACE\011IGNORELOOP\012NO_DAD" \
+		"\020DEFAULTIF"
 
 static int isnd6defif(int);
 void setnd6flags(const char *, int, int, const struct afswtch *);
diff --git a/sbin/ifconfig/ifconfig.8 b/sbin/ifconfig/ifconfig.8
index ff97c85..82835fa 100644
--- a/sbin/ifconfig/ifconfig.8
+++ b/sbin/ifconfig/ifconfig.8
@@ -28,7 +28,7 @@
 .\"     From: @(#)ifconfig.8	8.3 (Berkeley) 1/5/94
 .\" $FreeBSD$
 .\"
-.Dd September 9, 2014
+.Dd March 6, 2015
 .Dt IFCONFIG 8
 .Os
 .Sh NAME
@@ -736,6 +736,20 @@ outgoing interface.
 .It Cm -no_prefer_iface
 Clear a flag
 .Cm no_prefer_iface .
+.It Cm no_dad
+Set a flag to disable Duplicate Address Detection.
+.It Cm -no_dad
+Clear a flag
+.Cm no_dad .
+.It Cm ignoreloop
+Set a flag to disable loopback detection in Enhanced Duplicate Address
+Detection Algorithm.
+When this flag is set,
+Duplicate Address Detection will stop in a finite number of probings
+even if a loopback configuration is detected.
+.It Cm -ignoreloop
+Clear a flag
+.Cm ignoreloop .
 .El
 .Pp
 The following parameters are specific for IPv6 addresses.
diff --git a/sys/netinet/icmp6.h b/sys/netinet/icmp6.h
index 15f4c2d..979c8fd 100644
--- a/sys/netinet/icmp6.h
+++ b/sys/netinet/icmp6.h
@@ -297,9 +297,11 @@ struct nd_opt_hdr {		/* Neighbor discovery option header */
 #define ND_OPT_PREFIX_INFORMATION	3
 #define ND_OPT_REDIRECTED_HEADER	4
 #define ND_OPT_MTU			5
+#define ND_OPT_NONCE			14	/* RFC 3971 */
 #define ND_OPT_ROUTE_INFO		24	/* RFC 4191 */
 #define ND_OPT_RDNSS			25	/* RFC 6106 */
 #define ND_OPT_DNSSL			31	/* RFC 6106 */
+#define ND_OPT_MAX			31
 
 struct nd_opt_prefix_info {	/* prefix information */
 	u_int8_t	nd_opt_pi_type;
@@ -330,6 +332,16 @@ struct nd_opt_mtu {		/* MTU option */
 	u_int32_t	nd_opt_mtu_mtu;
 } __packed;
 
+#define	ND_OPT_NONCE_LEN	((1 * 8) - 2)
+#if ((ND_OPT_NONCE_LEN + 2) % 8) != 0
+#error "(ND_OPT_NONCE_LEN + 2) must be a multiple of 8."
+#endif 
+struct nd_opt_nonce {		/* nonce option */
+	u_int8_t	nd_opt_nonce_type;
+	u_int8_t	nd_opt_nonce_len;
+	u_int8_t	nd_opt_nonce[ND_OPT_NONCE_LEN];
+} __packed;
+
 struct nd_opt_route_info {	/* route info */
 	u_int8_t	nd_opt_rti_type;
 	u_int8_t	nd_opt_rti_len;
diff --git a/sys/netinet6/in6.c b/sys/netinet6/in6.c
index 609efa7..9a5a6ed 100644
--- a/sys/netinet6/in6.c
+++ b/sys/netinet6/in6.c
@@ -2370,7 +2370,8 @@ in6if_do_dad(struct ifnet *ifp)
 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
 		return (0);
 
-	if (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)
+	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) ||
+	    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD))
 		return (0);
 
 	switch (ifp->if_type) {
diff --git a/sys/netinet6/nd6.c b/sys/netinet6/nd6.c
index 8052e8f..94add9a 100644
--- a/sys/netinet6/nd6.c
+++ b/sys/netinet6/nd6.c
@@ -133,6 +133,10 @@ static int regen_tmpaddr(struct in6_ifaddr *);
 static struct llentry *nd6_free(struct llentry *, int);
 static void nd6_llinfo_timer(void *);
 static void clear_llinfo_pqueue(struct llentry *);
+static int nd6_output_lle(struct ifnet *, struct ifnet *, struct mbuf *,
+	struct sockaddr_in6 *);
+static int nd6_output_ifp(struct ifnet *, struct ifnet *, struct mbuf *,
+    struct sockaddr_in6 *);
 
 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
@@ -152,6 +156,8 @@ nd6_init(void)
 	callout_init(&V_nd6_slowtimo_ch, 0);
 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
 	    nd6_slowtimo, curvnet);
+
+	nd6_dad_init();
 }
 
 #ifdef VIMAGE
@@ -365,6 +371,7 @@ nd6_options(union nd_opts *ndopts)
 		case ND_OPT_TARGET_LINKADDR:
 		case ND_OPT_MTU:
 		case ND_OPT_REDIRECTED_HEADER:
+		case ND_OPT_NONCE:
 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
 				nd6log((LOG_INFO,
 				    "duplicated ND6 option found (type=%d)\n",
@@ -519,7 +526,7 @@ nd6_llinfo_timer(void *arg)
 			ln->la_asked++;
 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
 			LLE_WUNLOCK(ln);
-			nd6_ns_output(ifp, NULL, dst, ln, 0);
+			nd6_ns_output(ifp, NULL, dst, ln, NULL);
 			LLE_WLOCK(ln);
 		} else {
 			struct mbuf *m = ln->la_hold;
@@ -566,7 +573,7 @@ nd6_llinfo_timer(void *arg)
 			ln->ln_state = ND6_LLINFO_PROBE;
 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
 			LLE_WUNLOCK(ln);
-			nd6_ns_output(ifp, dst, dst, ln, 0);
+			nd6_ns_output(ifp, dst, dst, ln, NULL);
 			LLE_WLOCK(ln);
 		} else {
 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
@@ -578,7 +585,7 @@ nd6_llinfo_timer(void *arg)
 			ln->la_asked++;
 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
 			LLE_WUNLOCK(ln);
-			nd6_ns_output(ifp, dst, dst, ln, 0);
+			nd6_ns_output(ifp, dst, dst, ln, NULL);
 			LLE_WLOCK(ln);
 		} else {
 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
@@ -1662,42 +1669,8 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
 		ln->ln_state = newstate;
 
 		if (ln->ln_state == ND6_LLINFO_STALE) {
-			/*
-			 * XXX: since nd6_output() below will cause
-			 * state tansition to DELAY and reset the timer,
-			 * we must set the timer now, although it is actually
-			 * meaningless.
-			 */
-			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
-
-			if (ln->la_hold) {
-				struct mbuf *m_hold, *m_hold_next;
-
-				/*
-				 * reset the la_hold in advance, to explicitly
-				 * prevent a la_hold lookup in nd6_output()
-				 * (wouldn't happen, though...)
-				 */
-				for (m_hold = ln->la_hold, ln->la_hold = NULL;
-				    m_hold; m_hold = m_hold_next) {
-					m_hold_next = m_hold->m_nextpkt;
-					m_hold->m_nextpkt = NULL;
-
-					/*
-					 * we assume ifp is not a p2p here, so
-					 * just set the 2nd argument as the
-					 * 1st one.
-					 */
-					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
-				}
-				/*
-				 * If we have mbufs in the chain we need to do
-				 * deferred transmit. Copy the address from the
-				 * llentry before dropping the lock down below.
-				 */
-				if (chain != NULL)
-					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
-			}
+			if (ln->la_hold != NULL)
+				nd6_grab_holdchain(ln, &chain, &sin6);
 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
 			/* probe right away */
 			nd6_llinfo_settimer_locked((void *)ln, 0);
@@ -1780,8 +1753,8 @@ nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
 		if (static_route)
 			ln = NULL;
 	}
-	if (chain)
-		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
+	if (chain != NULL)
+		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
 	
 	/*
 	 * When the link-layer address of a router changes, select the
@@ -1849,55 +1822,159 @@ nd6_slowtimo(void *arg)
 	CURVNET_RESTORE();
 }
 
-int
-nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
-    struct sockaddr_in6 *dst, struct rtentry *rt0)
+void
+nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
+    struct sockaddr_in6 *sin6)
 {
 
-	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
+	LLE_WLOCK_ASSERT(ln);
+
+	*chain = ln->la_hold;
+	ln->la_hold = NULL;
+	memcpy(sin6, L3_ADDR_SIN6(ln), sizeof(*sin6));
+
+	if (ln->ln_state == ND6_LLINFO_STALE) {
+
+		/*
+		 * The first time we send a packet to a
+		 * neighbor whose entry is STALE, we have
+		 * to change the state to DELAY and a sets
+		 * a timer to expire in DELAY_FIRST_PROBE_TIME
+		 * seconds to ensure do neighbor unreachability
+		 * detection on expiration.
+		 * (RFC 2461 7.3.3)
+		 */
+		ln->la_asked = 0;
+		ln->ln_state = ND6_LLINFO_DELAY;
+		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
+	}
 }
 
+static int
+nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
+    struct sockaddr_in6 *dst)
+{
+	int error;
+	int ip6len;
+	struct ip6_hdr *ip6;
+	struct m_tag *mtag;
+
+#ifdef MAC
+	mac_netinet6_nd6_send(ifp, m);
+#endif
+
+	/*
+	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
+	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
+	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
+	 * to be diverted to user space.  When re-injected into the kernel,
+	 * send_output() will directly dispatch them to the outgoing interface.
+	 */
+	if (send_sendso_input_hook != NULL) {
+		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
+		if (mtag != NULL) {
+			ip6 = mtod(m, struct ip6_hdr *);
+			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
+			/* Use the SEND socket */
+			error = send_sendso_input_hook(m, ifp, SND_OUT,
+			    ip6len);
+			/* -1 == no app on SEND socket */
+			if (error == 0 || error != -1)
+			    return (error);
+		}
+	}
+
+	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
+	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
+	    mtod(m, struct ip6_hdr *));
+
+	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
+		origifp = ifp;
+
+	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
+	return (error);
+}
 
 /*
- * Note that I'm not enforcing any global serialization
- * lle state or asked changes here as the logic is too
- * complicated to avoid having to always acquire an exclusive
- * lock
- * KMM
- *
+ * IPv6 packet output - light version.
+ * Checks if destination LLE exists and is in proper state
+ * (e.g no modification required). If not true, fall back to
+ * "heavy" version.
  */
-#define senderr(e) { error = (e); goto bad;}
-
 int
-nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
-    struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
-	struct mbuf **chain)
+nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
+    struct sockaddr_in6 *dst, struct rtentry *rt0)
 {
-	struct mbuf *m = m0;
-	struct m_tag *mtag;
-	struct llentry *ln = lle;
-	struct ip6_hdr *ip6;
-	int error = 0;
-	int flags = 0;
-	int ip6len;
-
-#ifdef INVARIANTS
-	if (lle != NULL) {
-		
-		LLE_WLOCK_ASSERT(lle);
+	struct llentry *ln = NULL;
 
-		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
+	/* discard the packet if IPv6 operation is disabled on the interface */
+	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
+		m_freem(m);
+		return (ENETDOWN); /* better error? */
 	}
-#endif
+
 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
 		goto sendpkt;
 
 	if (nd6_need_cache(ifp) == 0)
 		goto sendpkt;
 
+	IF_AFDATA_RLOCK(ifp);
+	ln = nd6_lookup(&dst->sin6_addr, 0, ifp);
+	IF_AFDATA_RUNLOCK(ifp);
+
 	/*
-	 * next hop determination.  This routine is derived from ether_output.
+	 * Perform fast path for the following cases:
+	 * 1) lle state is REACHABLE
+	 * 2) lle state is DELAY (NS message sentNS message sent)
+	 *
+	 * Every other case involves lle modification, so we handle
+	 * them separately.
 	 */
+	if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
+	    ln->ln_state != ND6_LLINFO_DELAY)) {
+		/* Fall back to slow processing path */
+		if (ln != NULL)
+			LLE_RUNLOCK(ln);
+		return (nd6_output_lle(ifp, origifp, m, dst));
+	}
+
+sendpkt:
+	if (ln != NULL)
+		LLE_RUNLOCK(ln);
+
+	return (nd6_output_ifp(ifp, origifp, m, dst));
+}
+
+
+/*
+ * Output IPv6 packet - heavy version.
+ * Function assume that either
+ * 1) destination LLE does not exist, is invalid or stale, so
+ *   ND6_EXCLUSIVE lock needs to be acquired
+ * 2) destination lle is provided (with ND6_EXCLUSIVE lock),
+ *   in that case packets are queued in &chain.
+ *
+ */
+static int
+nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
+    struct sockaddr_in6 *dst)
+{
+	struct llentry *lle = NULL;
+	int flags = 0;
+
+	KASSERT(m != NULL, ("NULL mbuf, nothing to send"));
+	/* discard the packet if IPv6 operation is disabled on the interface */
+	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
+		m_freem(m);
+		return (ENETDOWN); /* better error? */
+	}
+
+	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
+		goto sendpkt;
+
+	if (nd6_need_cache(ifp) == 0)
+		goto sendpkt;
 
 	/*
 	 * Address resolution or Neighbor Unreachability Detection
@@ -1905,48 +1982,43 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
 	 * At this point, the destination of the packet must be a unicast
 	 * or an anycast address(i.e. not a multicast).
 	 */
-
-	flags = (lle != NULL) ? LLE_EXCLUSIVE : 0;
-	if (ln == NULL) {
-	retry:
+	if (lle == NULL) {
 		IF_AFDATA_RLOCK(ifp);
-		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
+		lle = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
 		IF_AFDATA_RUNLOCK(ifp);
-		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
+		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
 			/*
 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
 			 * the condition below is not very efficient.  But we believe
 			 * it is tolerable, because this should be a rare case.
 			 */
-			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
+			flags = ND6_CREATE | ND6_EXCLUSIVE;
 			IF_AFDATA_LOCK(ifp);
-			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
+			lle = nd6_lookup(&dst->sin6_addr, flags, ifp);
 			IF_AFDATA_UNLOCK(ifp);
 		}
 	} 
-	if (ln == NULL) {
+	if (lle == NULL) {
 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
 			char ip6buf[INET6_ADDRSTRLEN];
 			log(LOG_DEBUG,
 			    "nd6_output: can't allocate llinfo for %s "
 			    "(ln=%p)\n",
-			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
-			senderr(EIO);	/* XXX: good error? */
+			    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
+			m_freem(m);
+			return (ENOBUFS);
 		}
 		goto sendpkt;	/* send anyway */
 	}
 
+	LLE_WLOCK_ASSERT(lle);
+
 	/* We don't have to do link-layer address resolution on a p2p link. */
 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
-	    ln->ln_state < ND6_LLINFO_REACHABLE) {
-		if ((flags & LLE_EXCLUSIVE) == 0) {
-			flags |= LLE_EXCLUSIVE;
-			LLE_RUNLOCK(ln);
-			goto retry;
-		}
-		ln->ln_state = ND6_LLINFO_STALE;
-		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
+	    lle->ln_state < ND6_LLINFO_REACHABLE) {
+		lle->ln_state = ND6_LLINFO_STALE;
+		nd6_llinfo_settimer_locked(lle, (long)V_nd6_gctimer * hz);
 	}
 
 	/*
@@ -1956,15 +2028,10 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
 	 * neighbor unreachability detection on expiration.
 	 * (RFC 2461 7.3.3)
 	 */
-	if (ln->ln_state == ND6_LLINFO_STALE) {
-		if ((flags & LLE_EXCLUSIVE) == 0) {
-			flags |= LLE_EXCLUSIVE;
-			LLE_RUNLOCK(ln);
-			goto retry;
-		}
-		ln->la_asked = 0;
-		ln->ln_state = ND6_LLINFO_DELAY;
-		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
+	if (lle->ln_state == ND6_LLINFO_STALE) {
+		lle->la_asked = 0;
+		lle->ln_state = ND6_LLINFO_DELAY;
+		nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
 	}
 
 	/*
@@ -1972,7 +2039,7 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
 	 * (i.e. its link-layer address is already resolved), just
 	 * send the packet.
 	 */
-	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
+	if (lle->ln_state > ND6_LLINFO_INCOMPLETE)
 		goto sendpkt;
 
 	/*
@@ -1982,23 +2049,15 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
 	 * the oldest packet in the queue will be removed.
 	 */
-	if (ln->ln_state == ND6_LLINFO_NOSTATE)
-		ln->ln_state = ND6_LLINFO_INCOMPLETE;
-
-	if ((flags & LLE_EXCLUSIVE) == 0) {
-		flags |= LLE_EXCLUSIVE;
-		LLE_RUNLOCK(ln);
-		goto retry;
-	}
-
-	LLE_WLOCK_ASSERT(ln);
+	if (lle->ln_state == ND6_LLINFO_NOSTATE)
+		lle->ln_state = ND6_LLINFO_INCOMPLETE;
 
-	if (ln->la_hold) {
+	if (lle->la_hold != NULL) {
 		struct mbuf *m_hold;
 		int i;
 		
 		i = 0;
-		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
+		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
 			i++;
 			if (m_hold->m_nextpkt == NULL) {
 				m_hold->m_nextpkt = m;
@@ -2006,135 +2065,44 @@ nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
 			}
 		}
 		while (i >= V_nd6_maxqueuelen) {
-			m_hold = ln->la_hold;
-			ln->la_hold = ln->la_hold->m_nextpkt;
+			m_hold = lle->la_hold;
+			lle->la_hold = lle->la_hold->m_nextpkt;
 			m_freem(m_hold);
 			i--;
 		}
 	} else {
-		ln->la_hold = m;
+		lle->la_hold = m;
 	}
 
 	/*
 	 * If there has been no NS for the neighbor after entering the
 	 * INCOMPLETE state, send the first solicitation.
 	 */
-	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
-		ln->la_asked++;
+	if (!ND6_LLINFO_PERMANENT(lle) && lle->la_asked == 0) {
+		lle->la_asked++;
 		
-		nd6_llinfo_settimer_locked(ln,
+		nd6_llinfo_settimer_locked(lle,
 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
-		LLE_WUNLOCK(ln);
-		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
-		if (lle != NULL && ln == lle)
-			LLE_WLOCK(lle);
-
-	} else if (lle == NULL || ln != lle) {
-		/*
-		 * We did the lookup (no lle arg) so we
-		 * need to do the unlock here.
-		 */
-		LLE_WUNLOCK(ln);
+		LLE_WUNLOCK(lle);
+		nd6_ns_output(ifp, NULL, &dst->sin6_addr, lle, NULL);
+	} else {
+		/* We did the lookup so we need to do the unlock here. */
+		LLE_WUNLOCK(lle);
 	}
 
 	return (0);
 
   sendpkt:
-	/* discard the packet if IPv6 operation is disabled on the interface */
-	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
-		error = ENETDOWN; /* better error? */
-		goto bad;
-	}
-	/*
-	 * ln is valid and the caller did not pass in 
-	 * an llentry
-	 */
-	if ((ln != NULL) && (lle == NULL)) {
-		if (flags & LLE_EXCLUSIVE)
-			LLE_WUNLOCK(ln);
-		else
-			LLE_RUNLOCK(ln);
-	}
+	if (lle != NULL)
+		LLE_WUNLOCK(lle);
 
-#ifdef MAC
-	mac_netinet6_nd6_send(ifp, m);
-#endif
-
-	/*
-	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
-	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
-	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
-	 * to be diverted to user space.  When re-injected into the kernel,
-	 * send_output() will directly dispatch them to the outgoing interface.
-	 */
-	if (send_sendso_input_hook != NULL) {
-		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
-		if (mtag != NULL) {
-			ip6 = mtod(m, struct ip6_hdr *);
-			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
-			/* Use the SEND socket */
-			error = send_sendso_input_hook(m, ifp, SND_OUT,
-			    ip6len);
-			/* -1 == no app on SEND socket */
-			if (error == 0 || error != -1)
-			    return (error);
-		}
-	}
-
-	/*
-	 * We were passed in a pointer to an lle with the lock held 
-	 * this means that we can't call if_output as we will
-	 * recurse on the lle lock - so what we do is we create
-	 * a list of mbufs to send and transmit them in the caller
-	 * after the lock is dropped
-	 */
-	if (lle != NULL) {
-		if (*chain == NULL)
-			*chain = m;
-		else {
-			struct mbuf *mb;
-
-			/*
-			 * append mbuf to end of deferred chain
-			 */
-			mb = *chain;
-			while (mb->m_nextpkt != NULL)
-				mb = mb->m_nextpkt;
-			mb->m_nextpkt = m;
-		}
-		return (error);
-	}
-	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
-	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
-	    mtod(m, struct ip6_hdr *));
-	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
-		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
-		    NULL));
-	}
-	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
-	return (error);
-
-  bad:
-	/*
-	 * ln is valid and the caller did not pass in 
-	 * an llentry
-	 */
-	if ((ln != NULL) && (lle == NULL)) {
-		if (flags & LLE_EXCLUSIVE)
-			LLE_WUNLOCK(ln);
-		else
-			LLE_RUNLOCK(ln);
-	}
-	if (m)
-		m_freem(m);
-	return (error);
+	return (nd6_output_ifp(ifp, origifp, m, dst));
 }
-#undef senderr
 
 
 int
-nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
-    struct sockaddr_in6 *dst, struct route *ro)
+nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
+    struct sockaddr_in6 *dst)
 {
 	struct mbuf *m, *m_head;
 	struct ifnet *outifp;
@@ -2149,7 +2117,7 @@ nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
 	while (m_head) {
 		m = m_head;
 		m_head = m_head->m_nextpkt;
-		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);			       
+		error = nd6_output_ifp(ifp, origifp, m, dst);
 	}
 
 	/*
diff --git a/sys/netinet6/nd6.h b/sys/netinet6/nd6.h
index 25d8c5d..d664a94 100644
--- a/sys/netinet6/nd6.h
+++ b/sys/netinet6/nd6.h
@@ -87,6 +87,8 @@ struct nd_ifinfo {
 #define ND6_IFF_AUTO_LINKLOCAL	0x20
 #define	ND6_IFF_NO_RADR		0x40
 #define ND6_IFF_NO_PREFER_IFACE	0x80 /* XXX: not related to ND. */
+#define ND6_IFF_IGNORELOOP	0x100
+#define ND6_IFF_NO_DAD		0x200
 
 #define	ND6_CREATE		LLE_CREATE
 #define	ND6_EXCLUSIVE		LLE_EXCLUSIVE
@@ -359,7 +361,7 @@ VNET_DECLARE(int, ip6_temp_regen_advance); /* seconds */
 #define	V_ip6_temp_regen_advance	VNET(ip6_temp_regen_advance)
 
 union nd_opts {
-	struct nd_opt_hdr *nd_opt_array[8];	/* max = target address list */
+	struct nd_opt_hdr *nd_opt_array[16];	/* max = ND_OPT_NONCE */
 	struct {
 		struct nd_opt_hdr *zero;
 		struct nd_opt_hdr *src_lladdr;
@@ -367,6 +369,16 @@ union nd_opts {
 		struct nd_opt_prefix_info *pi_beg; /* multiple opts, start */
 		struct nd_opt_rd_hdr *rh;
 		struct nd_opt_mtu *mtu;
+		struct nd_opt_hdr *__res6;
+		struct nd_opt_hdr *__res7;
+		struct nd_opt_hdr *__res8;
+		struct nd_opt_hdr *__res9;
+		struct nd_opt_hdr *__res10;
+		struct nd_opt_hdr *__res11;
+		struct nd_opt_hdr *__res12;
+		struct nd_opt_hdr *__res13;
+		struct nd_opt_nonce *nonce;
+		struct nd_opt_hdr *__res15;
 		struct nd_opt_hdr *search;	/* multiple opts */
 		struct nd_opt_hdr *last;	/* multiple opts */
 		int done;
@@ -379,6 +391,7 @@ union nd_opts {
 #define nd_opts_pi_end		nd_opt_each.pi_end
 #define nd_opts_rh		nd_opt_each.rh
 #define nd_opts_mtu		nd_opt_each.mtu
+#define nd_opts_nonce		nd_opt_each.nonce
 #define nd_opts_search		nd_opt_each.search
 #define nd_opts_last		nd_opt_each.last
 #define nd_opts_done		nd_opt_each.done
@@ -410,11 +423,10 @@ struct llentry *nd6_cache_lladdr(struct ifnet *, struct in6_addr *,
 	char *, int, int, int);
 int nd6_output(struct ifnet *, struct ifnet *, struct mbuf *,
 	struct sockaddr_in6 *, struct rtentry *);
-int nd6_output_lle(struct ifnet *, struct ifnet *, struct mbuf *,
-	struct sockaddr_in6 *, struct rtentry *, struct llentry *,
-	struct mbuf **);
-int nd6_output_flush(struct ifnet *, struct ifnet *, struct mbuf *,
-	struct sockaddr_in6 *, struct route *);
+void nd6_grab_holdchain(struct llentry *, struct mbuf **,
+    struct sockaddr_in6 *);
+int nd6_flush_holdchain(struct ifnet *, struct ifnet *, struct mbuf *,
+    struct sockaddr_in6 *);
 int nd6_need_cache(struct ifnet *);
 int nd6_storelladdr(struct ifnet *, struct mbuf *,
 	const struct sockaddr *, u_char *, struct llentry **);
@@ -425,11 +437,11 @@ void nd6_na_output(struct ifnet *, const struct in6_addr *,
 	const struct in6_addr *, u_long, int, struct sockaddr *);
 void nd6_ns_input(struct mbuf *, int, int);
 void nd6_ns_output(struct ifnet *, const struct in6_addr *,
-	const struct in6_addr *, struct llentry *, int);
+	const struct in6_addr *, struct llentry *, uint8_t *);
 caddr_t nd6_ifptomac(struct ifnet *);
+void nd6_dad_init(void);
 void nd6_dad_start(struct ifaddr *, int);
 void nd6_dad_stop(struct ifaddr *);
-void nd6_dad_duplicated(struct ifaddr *);
 
 /* nd6_rtr.c */
 void nd6_rs_input(struct mbuf *, int, int);
diff --git a/sys/netinet6/nd6_nbr.c b/sys/netinet6/nd6_nbr.c
index 8e317af..d5f40f9 100644
--- a/sys/netinet6/nd6_nbr.c
+++ b/sys/netinet6/nd6_nbr.c
@@ -40,6 +40,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/malloc.h>
+#include <sys/libkern.h>
 #include <sys/lock.h>
 #include <sys/rwlock.h>
 #include <sys/mbuf.h>
@@ -48,9 +49,11 @@ __FBSDID("$FreeBSD$");
 #include <sys/time.h>
 #include <sys/kernel.h>
 #include <sys/errno.h>
+#include <sys/sysctl.h>
 #include <sys/syslog.h>
 #include <sys/queue.h>
 #include <sys/callout.h>
+#include <sys/refcount.h>
 
 #include <net/if.h>
 #include <net/if_types.h>
@@ -60,6 +63,7 @@ __FBSDID("$FreeBSD$");
 #ifdef RADIX_MPATH
 #include <net/radix_mpath.h>
 #endif
+#include <net/vnet.h>
 
 #include <netinet/in.h>
 #include <netinet/in_var.h>
@@ -78,21 +82,32 @@ __FBSDID("$FreeBSD$");
 #define SDL(s) ((struct sockaddr_dl *)s)
 
 struct dadq;
-static struct dadq *nd6_dad_find(struct ifaddr *);
+static struct dadq *nd6_dad_find(struct ifaddr *, struct nd_opt_nonce *);
+static void nd6_dad_add(struct dadq *dp);
+static void nd6_dad_del(struct dadq *dp);
+static void nd6_dad_rele(struct dadq *);
 static void nd6_dad_starttimer(struct dadq *, int);
 static void nd6_dad_stoptimer(struct dadq *);
 static void nd6_dad_timer(struct dadq *);
+static void nd6_dad_duplicated(struct ifaddr *, struct dadq *);
 static void nd6_dad_ns_output(struct dadq *, struct ifaddr *);
-static void nd6_dad_ns_input(struct ifaddr *);
+static void nd6_dad_ns_input(struct ifaddr *, struct nd_opt_nonce *);
 static void nd6_dad_na_input(struct ifaddr *);
 static void nd6_na_output_fib(struct ifnet *, const struct in6_addr *,
     const struct in6_addr *, u_long, int, struct sockaddr *, u_int);
+static void nd6_ns_output_fib(struct ifnet *, const struct in6_addr *,
+    const struct in6_addr *, struct llentry *, uint8_t *, u_int);
+
+static VNET_DEFINE(int, dad_enhanced) = 1;
+#define	V_dad_enhanced			VNET(dad_enhanced)
+
+SYSCTL_DECL(_net_inet6_ip6);
+SYSCTL_INT(_net_inet6_ip6, OID_AUTO, dad_enhanced, CTLFLAG_VNET | CTLFLAG_RW,
+    &VNET_NAME(dad_enhanced), 0,
+    "Enable Enhanced DAD, which adds a random nonce to NS messages for DAD.");
 
-static VNET_DEFINE(int, dad_ignore_ns) = 0;	/* ignore NS in DAD
-						   - specwise incorrect */
 static VNET_DEFINE(int, dad_maxtry) = 15;	/* max # of *tries* to
 						   transmit DAD packet */
-#define	V_dad_ignore_ns			VNET(dad_ignore_ns)
 #define	V_dad_maxtry			VNET(dad_maxtry)
 
 /*
@@ -316,7 +331,7 @@ nd6_ns_input(struct mbuf *m, int off, int icmp6len)
 		 * silently ignore it.
 		 */
 		if (IN6_IS_ADDR_UNSPECIFIED(&saddr6))
-			nd6_dad_ns_input(ifa);
+			nd6_dad_ns_input(ifa, ndopts.nd_opts_nonce);
 
 		goto freeit;
 	}
@@ -377,12 +392,14 @@ nd6_ns_input(struct mbuf *m, int off, int icmp6len)
  * Based on RFC 2461
  * Based on RFC 2462 (duplicate address detection)
  *
- *   ln - for source address determination
- *  dad - duplicate address detection
+ *    ln - for source address determination
+ * nonce - If non-NULL, NS is used for duplicate address detection and
+ *         the value (length is ND_OPT_NONCE_LEN) is used as a random nonce.
  */
-void
-nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6, 
-    const struct in6_addr *taddr6, struct llentry *ln, int dad)
+static void
+nd6_ns_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6,
+    const struct in6_addr *taddr6, struct llentry *ln, uint8_t *nonce,
+    u_int fibnum)
 {
 	struct mbuf *m;
 	struct m_tag *mtag;
@@ -404,12 +421,14 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 	    "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)",
 	    __func__, max_linkhdr, maxlen, MCLBYTES));
 
+
 	if (max_linkhdr + maxlen > MHLEN)
 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 	else
 		m = m_gethdr(M_NOWAIT, MT_DATA);
 	if (m == NULL)
 		return;
+	M_SETFIB(m, fibnum);
 
 	bzero(&ro, sizeof(ro));
 
@@ -444,7 +463,7 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 		if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0)
 			goto bad;
 	}
-	if (!dad) {
+	if (nonce == NULL) {
 		struct ifaddr *ifa;
 
 		/*
@@ -503,9 +522,8 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 			    NULL, &ro, NULL, &oifp, &src_in);
 			if (error) {
 				char ip6buf[INET6_ADDRSTRLEN];
-				nd6log((LOG_DEBUG,
-				    "nd6_ns_output: source can't be "
-				    "determined: dst=%s, error=%d\n",
+				nd6log((LOG_DEBUG, "%s: source can't be "
+				    "determined: dst=%s, error=%d\n", __func__,
 				    ip6_sprintf(ip6buf, &dst_sa.sin6_addr),
 				    error));
 				goto bad;
@@ -541,7 +559,7 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 	 *	Multicast NS		MUST add one	add the option
 	 *	Unicast NS		SHOULD add one	add the option
 	 */
-	if (!dad && (mac = nd6_ifptomac(ifp))) {
+	if (nonce == NULL && (mac = nd6_ifptomac(ifp))) {
 		int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen;
 		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
 		/* 8 byte alignments... */
@@ -555,7 +573,26 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 		nd_opt->nd_opt_len = optlen >> 3;
 		bcopy(mac, (caddr_t)(nd_opt + 1), ifp->if_addrlen);
 	}
+	/*
+	 * Add a Nonce option (RFC 3971) to detect looped back NS messages.
+	 * This behavior is documented as Enhanced Duplicate Address
+	 * Detection in draft-ietf-6man-enhanced-dad-13.
+	 * net.inet6.ip6.dad_enhanced=0 disables this.
+	 */
+	if (V_dad_enhanced != 0 && nonce != NULL) {
+		int optlen = sizeof(struct nd_opt_hdr) + ND_OPT_NONCE_LEN;
+		struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1);
+		/* 8-byte alignment is required. */
+		optlen = (optlen + 7) & ~7;
 
+		m->m_pkthdr.len += optlen;
+		m->m_len += optlen;
+		icmp6len += optlen;
+		bzero((caddr_t)nd_opt, optlen);
+		nd_opt->nd_opt_type = ND_OPT_NONCE;
+		nd_opt->nd_opt_len = optlen >> 3;
+		bcopy(nonce, (caddr_t)(nd_opt + 1), ND_OPT_NONCE_LEN);
+	}
 	ip6->ip6_plen = htons((u_short)icmp6len);
 	nd_ns->nd_ns_cksum = 0;
 	nd_ns->nd_ns_cksum =
@@ -570,7 +607,8 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 		m_tag_prepend(m, mtag);
 	}
 
-	ip6_output(m, NULL, &ro, dad ? IPV6_UNSPECSRC : 0, &im6o, NULL, NULL);
+	ip6_output(m, NULL, &ro, (nonce != NULL) ? IPV6_UNSPECSRC : 0,
+	    &im6o, NULL, NULL);
 	icmp6_ifstat_inc(ifp, ifs6_out_msg);
 	icmp6_ifstat_inc(ifp, ifs6_out_neighborsolicit);
 	ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_SOLICIT]);
@@ -588,6 +626,15 @@ nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
 	return;
 }
 
+#ifndef BURN_BRIDGES
+void
+nd6_ns_output(struct ifnet *ifp, const struct in6_addr *daddr6,
+    const struct in6_addr *taddr6, struct llentry *ln, uint8_t *nonce)
+{
+
+	nd6_ns_output_fib(ifp, daddr6, taddr6, ln, nonce, RT_DEFAULT_FIB);
+}
+#endif
 /*
  * Neighbor advertisement input handling.
  *
@@ -617,7 +664,6 @@ nd6_na_input(struct mbuf *m, int off, int icmp6len)
 	struct llentry *ln = NULL;
 	union nd_opts ndopts;
 	struct mbuf *chain = NULL;
-	struct m_tag *mtag;
 	struct sockaddr_in6 sin6;
 	char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN];
 
@@ -644,6 +690,7 @@ nd6_na_input(struct mbuf *m, int off, int icmp6len)
 	is_router = ((flags & ND_NA_FLAG_ROUTER) != 0);
 	is_solicited = ((flags & ND_NA_FLAG_SOLICITED) != 0);
 	is_override = ((flags & ND_NA_FLAG_OVERRIDE) != 0);
+	memset(&sin6, 0, sizeof(sin6));
 
 	taddr6 = nd_na->nd_na_target;
 	if (in6_setscope(&taddr6, ifp, NULL))
@@ -882,43 +929,15 @@ nd6_na_input(struct mbuf *m, int off, int icmp6len)
 	 *  rt->rt_flags &= ~RTF_REJECT;
 	 */
 	ln->la_asked = 0;
-	if (ln->la_hold) {
-		struct mbuf *m_hold, *m_hold_next;
-
-		/*
-		 * reset the la_hold in advance, to explicitly
-		 * prevent a la_hold lookup in nd6_output()
-		 * (wouldn't happen, though...)
-		 */
-		for (m_hold = ln->la_hold, ln->la_hold = NULL;
-		    m_hold; m_hold = m_hold_next) {
-			m_hold_next = m_hold->m_nextpkt;
-			m_hold->m_nextpkt = NULL;
-			/*
-			 * we assume ifp is not a loopback here, so just set
-			 * the 2nd argument as the 1st one.
-			 */
-
-			if (send_sendso_input_hook != NULL) {
-				mtag = m_tag_get(PACKET_TAG_ND_OUTGOING,
-				    sizeof(unsigned short), M_NOWAIT);
-				if (mtag == NULL)
-					goto bad;
-				m_tag_prepend(m, mtag);
-			}
-
-			nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
-		}
-	}
+	if (ln->la_hold != NULL)
+		nd6_grab_holdchain(ln, &chain, &sin6);
  freeit:
-	if (ln != NULL) {
-		if (chain)
-			memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
+	if (ln != NULL)
 		LLE_WUNLOCK(ln);
 
-		if (chain)
-			nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
-	}
+	if (chain != NULL)
+		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
+
 	if (checklink)
 		pfxlist_onlink_check();
 
@@ -1154,8 +1173,14 @@ struct dadq {
 	int dad_ns_ocount;	/* NS sent so far */
 	int dad_ns_icount;
 	int dad_na_icount;
+	int dad_ns_lcount;	/* looped back NS */
+	int dad_loopbackprobe;	/* probing state for loopback detection */
 	struct callout dad_timer_ch;
 	struct vnet *dad_vnet;
+	u_int dad_refcnt;
+#define	ND_OPT_NONCE_LEN32 \
+		((ND_OPT_NONCE_LEN + sizeof(uint32_t) - 1)/sizeof(uint32_t))
+	uint32_t dad_nonce[ND_OPT_NONCE_LEN32];
 };
 
 static VNET_DEFINE(TAILQ_HEAD(, dadq), dadq);
@@ -1163,23 +1188,54 @@ static VNET_DEFINE(struct rwlock, dad_rwlock);
 #define	V_dadq			VNET(dadq)
 #define	V_dad_rwlock		VNET(dad_rwlock)
 
-#define	DADQ_LOCK_INIT()	rw_init(&V_dad_rwlock, "nd6 DAD queue")	
-#define	DADQ_LOCK_DESTROY()	rw_destroy(&V_dad_rwlock)	
-#define	DADQ_LOCK_INITIALIZED()	rw_initialized(&V_dad_rwlock)	
 #define	DADQ_RLOCK()		rw_rlock(&V_dad_rwlock)	
 #define	DADQ_RUNLOCK()		rw_runlock(&V_dad_rwlock)	
 #define	DADQ_WLOCK()		rw_wlock(&V_dad_rwlock)	
 #define	DADQ_WUNLOCK()		rw_wunlock(&V_dad_rwlock)	
 
+static void
+nd6_dad_add(struct dadq *dp)
+{
+
+	DADQ_WLOCK();
+	TAILQ_INSERT_TAIL(&V_dadq, dp, dad_list);
+	DADQ_WUNLOCK();
+}
+
+static void
+nd6_dad_del(struct dadq *dp)
+{
+
+	DADQ_WLOCK();
+	TAILQ_REMOVE(&V_dadq, dp, dad_list);
+	DADQ_WUNLOCK();
+	nd6_dad_rele(dp);
+}
+
 static struct dadq *
-nd6_dad_find(struct ifaddr *ifa)
+nd6_dad_find(struct ifaddr *ifa, struct nd_opt_nonce *n)
 {
 	struct dadq *dp;
 
 	DADQ_RLOCK();
-	TAILQ_FOREACH(dp, &V_dadq, dad_list)
-		if (dp->dad_ifa == ifa)
-			break;
+	TAILQ_FOREACH(dp, &V_dadq, dad_list) {
+		if (dp->dad_ifa != ifa)
+			continue;
+		/*
+		 * Skip if the nonce matches the received one.
+		 * +2 in the length is required because of type and
+		 * length fields are included in a header.
+		 */
+		if (n != NULL &&
+		    n->nd_opt_nonce_len == (ND_OPT_NONCE_LEN + 2) / 8 &&
+		    memcmp(&n->nd_opt_nonce[0], &dp->dad_nonce[0],
+		        ND_OPT_NONCE_LEN) == 0) {
+			dp->dad_ns_lcount++;
+			continue;
+		}
+		refcount_acquire(&dp->dad_refcnt);
+		break;
+	}
 	DADQ_RUNLOCK();
 
 	return (dp);
@@ -1197,7 +1253,25 @@ static void
 nd6_dad_stoptimer(struct dadq *dp)
 {
 
-	callout_stop(&dp->dad_timer_ch);
+	callout_drain(&dp->dad_timer_ch);
+}
+
+static void
+nd6_dad_rele(struct dadq *dp)
+{
+
+	if (refcount_release(&dp->dad_refcnt)) {
+		ifa_free(dp->dad_ifa);
+		free(dp, M_IP6NDP);
+	}
+}
+
+void
+nd6_dad_init(void)
+{
+
+	rw_init(&V_dad_rwlock, "nd6 DAD queue");
+	TAILQ_INIT(&V_dadq);
 }
 
 /*
@@ -1210,11 +1284,6 @@ nd6_dad_start(struct ifaddr *ifa, int delay)
 	struct dadq *dp;
 	char ip6buf[INET6_ADDRSTRLEN];
 
-	if (DADQ_LOCK_INITIALIZED() == 0) {
-		DADQ_LOCK_INIT();
-		TAILQ_INIT(&V_dadq);
-	}
-
 	/*
 	 * If we don't need DAD, don't do it.
 	 * There are several cases:
@@ -1244,12 +1313,13 @@ nd6_dad_start(struct ifaddr *ifa, int delay)
 	}
 	if (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_IFDISABLED)
 		return;
-	if (nd6_dad_find(ifa) != NULL) {
+	if ((dp = nd6_dad_find(ifa, NULL)) != NULL) {
 		/* DAD already in progress */
+		nd6_dad_rele(dp);
 		return;
 	}
 
-	dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT);
+	dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT | M_ZERO);
 	if (dp == NULL) {
 		log(LOG_ERR, "nd6_dad_start: memory allocation failed for "
 			"%s(%s)\n",
@@ -1257,15 +1327,10 @@ nd6_dad_start(struct ifaddr *ifa, int delay)
 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
 		return;
 	}
-	bzero(dp, sizeof(*dp));
 	callout_init(&dp->dad_timer_ch, 0);
 #ifdef VIMAGE
 	dp->dad_vnet = curvnet;
 #endif
-	DADQ_WLOCK();
-	TAILQ_INSERT_TAIL(&V_dadq, (struct dadq *)dp, dad_list);
-	DADQ_WUNLOCK();
-
 	nd6log((LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
 	    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
 
@@ -1276,10 +1341,13 @@ nd6_dad_start(struct ifaddr *ifa, int delay)
 	 * (re)initialization.
 	 */
 	dp->dad_ifa = ifa;
-	ifa_ref(ifa);	/* just for safety */
+	ifa_ref(dp->dad_ifa);
 	dp->dad_count = V_ip6_dad_count;
 	dp->dad_ns_icount = dp->dad_na_icount = 0;
 	dp->dad_ns_ocount = dp->dad_ns_tcount = 0;
+	dp->dad_ns_lcount = dp->dad_loopbackprobe = 0;
+	refcount_init(&dp->dad_refcnt, 1);
+	nd6_dad_add(dp);
 	if (delay == 0) {
 		nd6_dad_ns_output(dp, ifa);
 		nd6_dad_starttimer(dp,
@@ -1297,9 +1365,7 @@ nd6_dad_stop(struct ifaddr *ifa)
 {
 	struct dadq *dp;
 
-	if (DADQ_LOCK_INITIALIZED() == 0)
-		return;
-	dp = nd6_dad_find(ifa);
+	dp = nd6_dad_find(ifa, NULL);
 	if (!dp) {
 		/* DAD wasn't started yet */
 		return;
@@ -1307,12 +1373,16 @@ nd6_dad_stop(struct ifaddr *ifa)
 
 	nd6_dad_stoptimer(dp);
 
-	DADQ_WLOCK();
-	TAILQ_REMOVE(&V_dadq, (struct dadq *)dp, dad_list);
-	DADQ_WUNLOCK();
-	free(dp, M_IP6NDP);
-	dp = NULL;
-	ifa_free(ifa);
+	/*
+	 * The DAD queue entry may have been removed by nd6_dad_timer() while
+	 * we were waiting for it to stop, so re-do the lookup.
+	 */
+	nd6_dad_rele(dp);
+	if (nd6_dad_find(ifa, NULL) == NULL)
+		return;
+
+	nd6_dad_del(dp);
+	nd6_dad_rele(dp);
 }
 
 static void
@@ -1327,45 +1397,36 @@ nd6_dad_timer(struct dadq *dp)
 	/* Sanity check */
 	if (ia == NULL) {
 		log(LOG_ERR, "nd6_dad_timer: called with null parameter\n");
-		goto done;
+		goto err;
 	}
 	if (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) {
 		/* Do not need DAD for ifdisabled interface. */
-		TAILQ_REMOVE(&V_dadq, (struct dadq *)dp, dad_list);
 		log(LOG_ERR, "nd6_dad_timer: cancel DAD on %s because of "
 		    "ND6_IFF_IFDISABLED.\n", ifp->if_xname);
-		free(dp, M_IP6NDP);
-		dp = NULL;
-		ifa_free(ifa);
-		goto done;
+		goto err;
 	}
 	if (ia->ia6_flags & IN6_IFF_DUPLICATED) {
 		log(LOG_ERR, "nd6_dad_timer: called with duplicated address "
 			"%s(%s)\n",
 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
-		goto done;
+		goto err;
 	}
 	if ((ia->ia6_flags & IN6_IFF_TENTATIVE) == 0) {
 		log(LOG_ERR, "nd6_dad_timer: called with non-tentative address "
 			"%s(%s)\n",
 			ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
 			ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
-		goto done;
+		goto err;
 	}
 
-	/* timeouted with IFF_{RUNNING,UP} check */
-	if (dp->dad_ns_tcount > V_dad_maxtry) {
+	/* Stop DAD if the interface is down even after dad_maxtry attempts. */
+	if ((dp->dad_ns_tcount > V_dad_maxtry) &&
+	    (((ifp->if_flags & IFF_UP) == 0) ||
+	     ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0))) {
 		nd6log((LOG_INFO, "%s: could not run DAD, driver problem?\n",
 		    if_name(ifa->ifa_ifp)));
-
-		DADQ_WLOCK();
-		TAILQ_REMOVE(&V_dadq, (struct dadq *)dp, dad_list);
-		DADQ_WUNLOCK();
-		free(dp, M_IP6NDP);
-		dp = NULL;
-		ifa_free(ifa);
-		goto done;
+		goto err;
 	}
 
 	/* Need more checks? */
@@ -1376,32 +1437,50 @@ nd6_dad_timer(struct dadq *dp)
 		nd6_dad_ns_output(dp, ifa);
 		nd6_dad_starttimer(dp,
 		    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
+		goto done;
 	} else {
 		/*
 		 * We have transmitted sufficient number of DAD packets.
 		 * See what we've got.
 		 */
-		int duplicate;
-
-		duplicate = 0;
-
-		if (dp->dad_na_icount) {
+		if (dp->dad_ns_icount > 0 || dp->dad_na_icount > 0)
+			/* We've seen NS or NA, means DAD has failed. */
+			nd6_dad_duplicated(ifa, dp);
+		else if (V_dad_enhanced != 0 &&
+		    dp->dad_ns_lcount > 0 &&
+		    dp->dad_ns_lcount > dp->dad_loopbackprobe) {
 			/*
-			 * the check is in nd6_dad_na_input(),
-			 * but just in case
+			 * A looped back probe is detected,
+			 * Sec. 4.1 in draft-ietf-6man-enhanced-dad-13
+			 * requires transmission of additional probes until
+			 * the loopback condition becomes clear.
 			 */
-			duplicate++;
-		}
-
-		if (dp->dad_ns_icount) {
-			/* We've seen NS, means DAD has failed. */
-			duplicate++;
-		}
-
-		if (duplicate) {
-			/* (*dp) will be freed in nd6_dad_duplicated() */
-			dp = NULL;
-			nd6_dad_duplicated(ifa);
+			log(LOG_ERR, "%s: a looped back NS message is "
+			    "detected during DAD for %s.  "
+			    "Another DAD probes are being sent.\n",
+			    if_name(ifa->ifa_ifp),
+			    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
+			dp->dad_loopbackprobe = dp->dad_ns_lcount;
+			/*
+			 * An interface with IGNORELOOP is one which a
+			 * loopback is permanently expected while regular
+			 * traffic works.  In that case, stop DAD after
+			 * MAX_MULTICAST_SOLICIT number of NS messages
+			 * regardless of the number of received loopback NS
+			 * by increasing dad_loopbackprobe in advance.
+			 */
+			if (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_IGNORELOOP)
+				dp->dad_loopbackprobe += V_nd6_mmaxtries;
+			/*
+			 * Send an NS immediately and increase dad_count by
+			 * V_nd6_mmaxtries - 1.
+			 */
+			nd6_dad_ns_output(dp, ifa);
+			dp->dad_count =
+			    dp->dad_ns_ocount + V_nd6_mmaxtries - 1;
+			nd6_dad_starttimer(dp,
+			    (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000);
+			goto done;
 		} else {
 			/*
 			 * We are done with DAD.  No NA came, no NS came.
@@ -1416,45 +1495,36 @@ nd6_dad_timer(struct dadq *dp)
 			    "%s: DAD complete for %s - no duplicates found\n",
 			    if_name(ifa->ifa_ifp),
 			    ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)));
-
-			DADQ_WLOCK();
-			TAILQ_REMOVE(&V_dadq, (struct dadq *)dp, dad_list);
-			DADQ_WUNLOCK();
-			free(dp, M_IP6NDP);
-			dp = NULL;
-			ifa_free(ifa);
+			if (dp->dad_ns_lcount > 0)
+				log(LOG_ERR, "%s: DAD completed while "
+				    "a looped back NS message is detected "
+				    "during DAD for %s.\n",
+				    if_name(ifa->ifa_ifp),
+				    ip6_sprintf(ip6buf, IFA_IN6(ifa)));
 		}
 	}
-
+err:
+	nd6_dad_del(dp);
 done:
 	CURVNET_RESTORE();
 }
 
-void
-nd6_dad_duplicated(struct ifaddr *ifa)
+static void
+nd6_dad_duplicated(struct ifaddr *ifa, struct dadq *dp)
 {
 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
 	struct ifnet *ifp;
-	struct dadq *dp;
 	char ip6buf[INET6_ADDRSTRLEN];
 
-	dp = nd6_dad_find(ifa);
-	if (dp == NULL) {
-		log(LOG_ERR, "nd6_dad_duplicated: DAD structure not found\n");
-		return;
-	}
-
 	log(LOG_ERR, "%s: DAD detected duplicate IPv6 address %s: "
-	    "NS in/out=%d/%d, NA in=%d\n",
+	    "NS in/out/loopback=%d/%d/%d, NA in=%d\n",
 	    if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr),
-	    dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_na_icount);
+	    dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_ns_lcount,
+	    dp->dad_na_icount);
 
 	ia->ia6_flags &= ~IN6_IFF_TENTATIVE;
 	ia->ia6_flags |= IN6_IFF_DUPLICATED;
 
-	/* We are done with DAD, with duplicate address found. (failure) */
-	nd6_dad_stoptimer(dp);
-
 	ifp = ifa->ifa_ifp;
 	log(LOG_ERR, "%s: DAD complete for %s - duplicate found\n",
 	    if_name(ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr));
@@ -1495,13 +1565,6 @@ nd6_dad_duplicated(struct ifaddr *ifa)
 			break;
 		}
 	}
-
-	DADQ_WLOCK();
-	TAILQ_REMOVE(&V_dadq, (struct dadq *)dp, dad_list);
-	DADQ_WUNLOCK();
-	free(dp, M_IP6NDP);
-	dp = NULL;
-	ifa_free(ifa);
 }
 
 static void
@@ -1509,6 +1572,7 @@ nd6_dad_ns_output(struct dadq *dp, struct ifaddr *ifa)
 {
 	struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa;
 	struct ifnet *ifp = ifa->ifa_ifp;
+	int i;
 
 	dp->dad_ns_tcount++;
 	if ((ifp->if_flags & IFF_UP) == 0) {
@@ -1519,17 +1583,29 @@ nd6_dad_ns_output(struct dadq *dp, struct ifaddr *ifa)
 	}
 
 	dp->dad_ns_ocount++;
-	nd6_ns_output(ifp, NULL, &ia->ia_addr.sin6_addr, NULL, 1);
+	if (V_dad_enhanced != 0) {
+		for (i = 0; i < ND_OPT_NONCE_LEN32; i++)
+			dp->dad_nonce[i] = arc4random();
+		/*
+		 * XXXHRS: Note that in the case that
+		 * DupAddrDetectTransmits > 1, multiple NS messages with
+		 * different nonces can be looped back in an unexpected
+		 * order.  The current implementation recognizes only
+		 * the latest nonce on the sender side.  Practically it
+		 * should work well in almost all cases.
+		 */
+	}
+	nd6_ns_output(ifp, NULL, &ia->ia_addr.sin6_addr, NULL,
+	    (uint8_t *)&dp->dad_nonce[0]);
 }
 
 static void
-nd6_dad_ns_input(struct ifaddr *ifa)
+nd6_dad_ns_input(struct ifaddr *ifa, struct nd_opt_nonce *ndopt_nonce)
 {
 	struct in6_ifaddr *ia;
 	struct ifnet *ifp;
 	const struct in6_addr *taddr6;
 	struct dadq *dp;
-	int duplicate;
 
 	if (ifa == NULL)
 		panic("ifa == NULL in nd6_dad_ns_input");
@@ -1537,39 +1613,15 @@ nd6_dad_ns_input(struct ifaddr *ifa)
 	ia = (struct in6_ifaddr *)ifa;
 	ifp = ifa->ifa_ifp;
 	taddr6 = &ia->ia_addr.sin6_addr;
-	duplicate = 0;
-	dp = nd6_dad_find(ifa);
-
-	/* Quickhack - completely ignore DAD NS packets */
-	if (V_dad_ignore_ns) {
-		char ip6buf[INET6_ADDRSTRLEN];
-		nd6log((LOG_INFO,
-		    "nd6_dad_ns_input: ignoring DAD NS packet for "
-		    "address %s(%s)\n", ip6_sprintf(ip6buf, taddr6),
-		    if_name(ifa->ifa_ifp)));
+	/* Ignore Nonce option when Enhanced DAD is disabled. */
+	if (V_dad_enhanced == 0)
+		ndopt_nonce = NULL;
+	dp = nd6_dad_find(ifa, ndopt_nonce);
+	if (dp == NULL)
 		return;
-	}
-
-	/*
-	 * if I'm yet to start DAD, someone else started using this address
-	 * first.  I have a duplicate and you win.
-	 */
-	if (dp == NULL || dp->dad_ns_ocount == 0)
-		duplicate++;
-
-	/* XXX more checks for loopback situation - see nd6_dad_timer too */
 
-	if (duplicate) {
-		dp = NULL;	/* will be freed in nd6_dad_duplicated() */
-		nd6_dad_duplicated(ifa);
-	} else {
-		/*
-		 * not sure if I got a duplicate.
-		 * increment ns count and see what happens.
-		 */
-		if (dp)
-			dp->dad_ns_icount++;
-	}
+	dp->dad_ns_icount++;
+	nd6_dad_rele(dp);
 }
 
 static void
@@ -1580,10 +1632,9 @@ nd6_dad_na_input(struct ifaddr *ifa)
 	if (ifa == NULL)
 		panic("ifa == NULL in nd6_dad_na_input");
 
-	dp = nd6_dad_find(ifa);
-	if (dp)
+	dp = nd6_dad_find(ifa, NULL);
+	if (dp != NULL) {
 		dp->dad_na_icount++;
-
-	/* remove the address. */
-	nd6_dad_duplicated(ifa);
+		nd6_dad_rele(dp);
+	}
 }