]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet6/nd6.c
xnu-2050.18.24.tar.gz
[apple/xnu.git] / bsd / netinet6 / nd6.c
index a7b5cb3a3dc5284b69f22c87aa69449e08ec9e2f..7cae1d7e7255af7edc204ec3eaa5802a4fbb5db1 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008-2009 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
@@ -79,6 +79,8 @@
 #include <sys/syslog.h>
 #include <sys/protosw.h>
 #include <sys/proc.h>
+#include <sys/mcache.h>
+
 #include <kern/queue.h>
 #include <kern/zalloc.h>
 
 #include <net/if.h>
 #include <net/if_dl.h>
 #include <net/if_types.h>
-#include <net/if_atm.h>
+#include <net/if_llreach.h>
 #include <net/route.h>
 #include <net/dlil.h>
+#include <net/ntstat.h>
 
 #include <netinet/in.h>
 #include <netinet/in_arp.h>
 #include <netinet/if_ether.h>
-#include <netinet/if_fddi.h>
 #include <netinet6/in6_var.h>
 #include <netinet/ip6.h>
 #include <netinet6/ip6_var.h>
 #include <netinet6/nd6.h>
-#include <netinet6/in6_prefix.h>
+#include <netinet6/scope6_var.h>
 #include <netinet/icmp6.h>
 
 #include "loop.h"
 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
 
-#define        SA(p) ((struct sockaddr *)(p))
-#define SIN6(s) ((struct sockaddr_in6 *)s)
-#define SDL(s) ((struct sockaddr_dl *)s)
 #define        equal(a1, a2) (bcmp((caddr_t)(a1), (caddr_t)(a2), (a1)->sa_len) == 0)
 
 /* timer values */
@@ -125,6 +124,7 @@ int nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
 int nd6_maxndopt = 10; /* max # of ND options allowed */
 
 int nd6_maxnudhint = 0;        /* max # of subsequent upper layer hints */
+int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
 
 #if ND6_DEBUG
 int nd6_debug = 1;
@@ -132,6 +132,12 @@ int nd6_debug = 1;
 int nd6_debug = 0;
 #endif
 
+int nd6_optimistic_dad =
+       (ND6_OPTIMISTIC_DAD_LINKLOCAL|ND6_OPTIMISTIC_DAD_AUTOCONF|
+       ND6_OPTIMISTIC_DAD_TEMPORARY|ND6_OPTIMISTIC_DAD_DYNAMIC);
+
+static int nd6_is_new_addr_neighbor (struct sockaddr_in6 *, struct ifnet *);
+
 /* for debugging? */
 static int nd6_inuse, nd6_allocated;
 
@@ -151,7 +157,8 @@ static int nd6_inuse, nd6_allocated;
  *
  *     - Routing lock (rnh_lock)
  *
- * ln_hold, ln_asked, ln_expire, ln_state, ln_router, ln_byhint, ln_flags
+ * ln_hold, ln_asked, ln_expire, ln_state, ln_router, ln_byhint, ln_flags,
+ * ln_llreach, ln_lastused
  *
  *     - Routing entry lock (rt_lock)
  *
@@ -161,35 +168,43 @@ static int nd6_inuse, nd6_allocated;
  * freed until the route itself is freed.
  */
 struct llinfo_nd6 llinfo_nd6 = {
-       &llinfo_nd6, &llinfo_nd6, NULL, NULL, 0, 0, 0, 0, 0, 0
+       &llinfo_nd6, &llinfo_nd6, NULL, NULL, 0, 0, 0, 0, 0, 0, NULL, 0
 };
 
 /* Protected by nd_if_rwlock */
 size_t nd_ifinfo_indexlim = 32; /* increased for 5589193 */
 struct nd_ifinfo *nd_ifinfo = NULL;
 
-static lck_grp_attr_t  *nd_if_rwlock_grp_attr;
-static lck_grp_t       *nd_if_rwlock_grp;
-static lck_attr_t      *nd_if_rwlock_attr;
-lck_rw_t               *nd_if_rwlock;
+static lck_grp_attr_t  *nd_if_lock_grp_attr;
+static lck_grp_t       *nd_if_lock_grp;
+static lck_attr_t      *nd_if_lock_attr;
+decl_lck_rw_data(, nd_if_rwlock_data);
+lck_rw_t               *nd_if_rwlock = &nd_if_rwlock_data;
 
 /* Protected by nd6_mutex */
 struct nd_drhead nd_defrouter;
 struct nd_prhead nd_prefix = { 0 };
 
+/* Serialization variables for nd6_drain() */
+static boolean_t nd6_drain_busy;
+static void *nd6_drain_waitchan = &nd6_drain_busy;
+static int nd6_drain_waiters = 0;
+
 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
 static struct sockaddr_in6 all1_sa;
 
 static int regen_tmpaddr(struct in6_ifaddr *);
-extern lck_mtx_t *ip6_mutex;
 extern lck_mtx_t *nd6_mutex;
 
 static void nd6_slowtimo(void *ignored_arg);
 static struct llinfo_nd6 *nd6_llinfo_alloc(void);
 static void nd6_llinfo_free(void *);
+static void nd6_llinfo_purge(struct rtentry *);
+static void nd6_llinfo_get_ri(struct rtentry *, struct rt_reach_info *);
+static void nd6_llinfo_get_iflri(struct rtentry *, struct ifnet_llreach_info *);
 
-static void nd6_siocgdrlst(void *, int);
-static void nd6_siocgprlst(void *, int);
+static int nd6_siocgdrlst(void *, int);
+static int nd6_siocgprlst(void *, int);
 
 /*
  * Insertion and removal from llinfo_nd6 must be done with rnh_lock held.
@@ -236,11 +251,10 @@ nd6_init()
        /* initialization of the default router list */
        TAILQ_INIT(&nd_defrouter);
 
-       nd_if_rwlock_grp_attr = lck_grp_attr_alloc_init();
-       nd_if_rwlock_grp = lck_grp_alloc_init("nd_if_rwlock",
-           nd_if_rwlock_grp_attr);
-       nd_if_rwlock_attr = lck_attr_alloc_init();
-       nd_if_rwlock = lck_rw_alloc_init(nd_if_rwlock_grp, nd_if_rwlock_attr);
+       nd_if_lock_grp_attr = lck_grp_attr_alloc_init();
+       nd_if_lock_grp = lck_grp_alloc_init("nd_if_lock", nd_if_lock_grp_attr);
+       nd_if_lock_attr = lck_attr_alloc_init();
+       lck_rw_init(nd_if_rwlock, nd_if_lock_grp, nd_if_lock_attr);
 
        llinfo_nd6_zone = zinit(sizeof (struct llinfo_nd6),
            LLINFO_ND6_ZONE_MAX * sizeof (struct llinfo_nd6), 0,
@@ -249,6 +263,11 @@ nd6_init()
                panic("%s: failed allocating llinfo_nd6_zone", __func__);
 
        zone_change(llinfo_nd6_zone, Z_EXPAND, TRUE);
+       zone_change(llinfo_nd6_zone, Z_CALLERACCT, FALSE);
+
+       nd6_nbr_init();
+       nd6_rtr_init();
+       nd6_prproxy_init();
 
        nd6_init_done = 1;
 
@@ -278,13 +297,77 @@ nd6_llinfo_free(void *arg)
                ln->ln_hold = NULL;
        }
 
+       /* Purge any link-layer info caching */
+       VERIFY(ln->ln_rt->rt_llinfo == ln);
+       if (ln->ln_rt->rt_llinfo_purge != NULL)
+               ln->ln_rt->rt_llinfo_purge(ln->ln_rt);
+
        zfree(llinfo_nd6_zone, ln);
 }
 
+static void
+nd6_llinfo_purge(struct rtentry *rt)
+{
+       struct llinfo_nd6 *ln = rt->rt_llinfo;
+
+       RT_LOCK_ASSERT_HELD(rt);
+       VERIFY(rt->rt_llinfo_purge == nd6_llinfo_purge && ln != NULL);
+
+       if (ln->ln_llreach != NULL) {
+               RT_CONVERT_LOCK(rt);
+               ifnet_llreach_free(ln->ln_llreach);
+               ln->ln_llreach = NULL;
+       }
+       ln->ln_lastused = 0;
+}
+
+static void
+nd6_llinfo_get_ri(struct rtentry *rt, struct rt_reach_info *ri)
+{
+       struct llinfo_nd6 *ln = rt->rt_llinfo;
+       struct if_llreach *lr = ln->ln_llreach;
+
+       if (lr == NULL) {
+               bzero(ri, sizeof (*ri));
+               ri->ri_rssi = IFNET_RSSI_UNKNOWN;
+               ri->ri_lqm = IFNET_LQM_THRESH_OFF;
+               ri->ri_npm = IFNET_NPM_THRESH_UNKNOWN;
+       } else {
+               IFLR_LOCK(lr);
+               /* Export to rt_reach_info structure */
+               ifnet_lr2ri(lr, ri);
+               /* Export ND6 send expiration (calendar) time */
+               ri->ri_snd_expire =
+                   ifnet_llreach_up2calexp(lr, ln->ln_lastused);
+               IFLR_UNLOCK(lr);
+       }
+}
+
+static void
+nd6_llinfo_get_iflri(struct rtentry *rt, struct ifnet_llreach_info *iflri)
+{
+       struct llinfo_nd6 *ln = rt->rt_llinfo;
+       struct if_llreach *lr = ln->ln_llreach;
+
+       if (lr == NULL) {
+               bzero(iflri, sizeof (*iflri));
+               iflri->iflri_rssi = IFNET_RSSI_UNKNOWN;
+               iflri->iflri_lqm = IFNET_LQM_THRESH_OFF;
+               iflri->iflri_npm = IFNET_NPM_THRESH_UNKNOWN;
+       } else {
+               IFLR_LOCK(lr);
+               /* Export to ifnet_llreach_info structure */
+               ifnet_lr2iflri(lr, iflri);
+               /* Export ND6 send expiration (uptime) time */
+               iflri->iflri_snd_expire =
+                   ifnet_llreach_up2upexp(lr, ln->ln_lastused);
+               IFLR_UNLOCK(lr);
+       }
+}
+
 int
 nd6_ifattach(struct ifnet *ifp)
 {
-
        /*
         * We have some arrays that should be indexed by if_index.
         * since if_index will grow dynamically, they should grow too.
@@ -303,7 +386,7 @@ nd6_ifattach(struct ifnet *ifp)
                q = (caddr_t)_MALLOC(n, M_IP6NDP, M_WAITOK);
                if (q == NULL) {
                        lck_rw_done(nd_if_rwlock);
-                       return ENOBUFS;
+                       return (ENOBUFS);
                }
                bzero(q, n);
                nd_ifinfo_indexlim = newlim;
@@ -315,36 +398,31 @@ nd6_ifattach(struct ifnet *ifp)
                         */
                        FREE((caddr_t)nd_ifinfo, M_IP6NDP);
                }
-               nd_ifinfo = (struct nd_ifinfo *)q;
+               nd_ifinfo = (struct nd_ifinfo *)(void *)q;
        }
-       lck_rw_done(nd_if_rwlock);
 
 #define ND nd_ifinfo[ifp->if_index]
-
        /*
         * Don't initialize if called twice.
-        * XXX: to detect this, we should choose a member that is never set
-        * before initialization of the ND structure itself.  We formaly used
-        * the linkmtu member, which was not suitable because it could be 
-        * initialized via "ifconfig mtu".
         */
-       lck_rw_lock_shared(nd_if_rwlock);
-       if (ND.basereachable) {
+       if (ND.initialized) {
                lck_rw_done(nd_if_rwlock);
-               return 0;
+               return (0);
        }
+       lck_mtx_init(&ND.lock, nd_if_lock_grp, nd_if_lock_attr);
+       ND.initialized = TRUE;
        ND.linkmtu = ifp->if_mtu;
        ND.chlim = IPV6_DEFHLIM;
        ND.basereachable = REACHABLE_TIME;
        ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
        ND.retrans = RETRANS_TIMER;
-       ND.receivedra = 0;
        ND.flags = ND6_IFF_PERFORMNUD;
        lck_rw_done(nd_if_rwlock);
-       nd6_setmtu(ifp);
 #undef ND
-       
-       return 0;
+
+       nd6_setmtu(ifp);
+
+       return (0);
 }
 
 /*
@@ -362,12 +440,15 @@ nd6_setmtu(struct ifnet *ifp)
         * because this can be called directly from SIOCSIFMTU for IPv4
         */
        lck_rw_lock_shared(nd_if_rwlock);
-       if (ifp->if_index >= nd_ifinfo_indexlim) {
+       if (ifp->if_index >= nd_ifinfo_indexlim ||
+           !nd_ifinfo[ifp->if_index].initialized) {
                lck_rw_done(nd_if_rwlock);
-               return; /* we're  out of bound for nd_ifinfo */
+               return; /* nd_ifinfo out of bound, or not yet initialized */
        }
 
        ndi = &nd_ifinfo[ifp->if_index];
+       VERIFY(ndi->initialized);
+       lck_mtx_lock(&ndi->lock);
        oldmaxmtu = ndi->maxmtu;
 
        /*
@@ -378,8 +459,8 @@ nd6_setmtu(struct ifnet *ifp)
         * the sanity checks related to the maximum MTU allowed for the
         * interface (a value that is known only by the interface layer),
         * by sending the request down via ifnet_ioctl().  The use of the
-        * ND level maxmtu and linkmtu (the latter obtained via RA) are done
-        * via IN6_LINKMTU() which does further checking against if_mtu.
+        * ND level maxmtu and linkmtu are done via IN6_LINKMTU() which
+        * does further checking against if_mtu.
         */
        maxmtu = ndi->maxmtu = ifp->if_mtu;
 
@@ -394,6 +475,8 @@ nd6_setmtu(struct ifnet *ifp)
                    "new link MTU on %s%d (%u) is too small for IPv6\n",
                    ifp->if_name, ifp->if_unit, (uint32_t)ndi->maxmtu);
        }
+       ndi->linkmtu = ifp->if_mtu;
+       lck_mtx_unlock(&ndi->lock);
        lck_rw_done(nd_if_rwlock);
 
        /* also adjust in6_maxmtu if necessary. */
@@ -480,16 +563,16 @@ nd6_options(
        struct nd_opt_hdr *nd_opt;
        int i = 0;
 
-       if (!ndopts)
-               panic("ndopts == NULL in nd6_options\n");
-       if (!ndopts->nd_opts_last)
-               panic("uninitialized ndopts in nd6_options\n");
-       if (!ndopts->nd_opts_search)
+       if (ndopts == NULL)
+               panic("ndopts == NULL in nd6_options");
+       if (ndopts->nd_opts_last == NULL)
+               panic("uninitialized ndopts in nd6_options");
+       if (ndopts->nd_opts_search == NULL)
                return 0;
 
        while (1) {
                nd_opt = nd6_option(ndopts);
-               if (!nd_opt && !ndopts->nd_opts_last) {
+               if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
                        /*
                         * Message validation requires that all included
                         * options have a length that is greater than zero.
@@ -499,7 +582,7 @@ nd6_options(
                        return -1;
                }
 
-               if (!nd_opt)
+               if (nd_opt == NULL)
                        goto skip1;
 
                switch (nd_opt->nd_opt_type) {
@@ -525,6 +608,9 @@ nd6_options(
                        ndopts->nd_opts_pi_end =
                                (struct nd_opt_prefix_info *)nd_opt;
                        break;
+               case ND_OPT_RDNSS:
+                       /* ignore */
+                       break;
                default:
                        /*
                         * Unknown options must be silently ignored,
@@ -550,12 +636,8 @@ skip1:
        return 0;
 }
 
-/*
- * ND6 timer routine to expire default route list and prefix list
- */
 void
-nd6_timer(
-       __unused void   *ignored_arg)
+nd6_drain(__unused void        *ignored_arg)
 {
        struct llinfo_nd6 *ln;
        struct nd_defrouter *dr;
@@ -585,7 +667,8 @@ again:
                struct rtentry *rt;
                struct sockaddr_in6 *dst;
                struct llinfo_nd6 *next;
-               struct nd_ifinfo ndi;
+               struct nd_ifinfo *ndi;
+               u_int32_t retrans, flags;
 
                /* ln_next/prev/rt is protected by rnh_lock */
                next = ln->ln_next;
@@ -614,7 +697,7 @@ again:
                }
 
                /* rt_key should never be NULL */
-               dst = (struct sockaddr_in6 *)rt_key(rt);
+               dst = (struct sockaddr_in6 *)(void *)rt_key(rt);
                if (dst == NULL) {
                        panic("%s: rt(%p) key is NULL ln(%p)", __func__,
                            rt, ln);
@@ -630,7 +713,6 @@ again:
                        continue;
                }
 
-               /* Make a copy (we're using it read-only anyway) */
                lck_rw_lock_shared(nd_if_rwlock);
                if (ifp->if_index >= nd_ifinfo_indexlim) {
                        lck_rw_done(nd_if_rwlock);
@@ -638,7 +720,12 @@ again:
                        ln = next;
                        continue;
                }
-               ndi = nd_ifinfo[ifp->if_index];
+               ndi = ND_IFINFO(ifp);
+               VERIFY(ndi->initialized);
+               lck_mtx_lock(&ndi->lock);
+               retrans = ndi->retrans;
+               flags = ndi->flags;
+               lck_mtx_unlock(&ndi->lock);
                lck_rw_done(nd_if_rwlock);
 
                RT_LOCK_ASSERT_HELD(rt);
@@ -647,13 +734,17 @@ again:
                case ND6_LLINFO_INCOMPLETE:
                        if (ln->ln_asked < nd6_mmaxtries) {
                                ln->ln_asked++;
-                               ln->ln_expire = timenow.tv_sec +
-                                   ndi.retrans / 1000;
+                               ln->ln_expire = timenow.tv_sec + retrans / 1000;
                                RT_ADDREF_LOCKED(rt);
                                RT_UNLOCK(rt);
                                lck_mtx_unlock(rnh_lock);
-                               nd6_ns_output(ifp, NULL, &dst->sin6_addr,
-                                       ln, 0, 0);
+                               if (ip6_forwarding) {
+                                       nd6_prproxy_ns_output(ifp, NULL,
+                                           &dst->sin6_addr, ln);
+                               } else {
+                                       nd6_ns_output(ifp, NULL,
+                                           &dst->sin6_addr, ln, 0);
+                               }
                                RT_REMREF(rt);
                        } else {
                                struct mbuf *m = ln->ln_hold;
@@ -683,7 +774,8 @@ again:
                case ND6_LLINFO_REACHABLE:
                        if (ln->ln_expire) {
                                ln->ln_state = ND6_LLINFO_STALE;
-                               ln->ln_expire = timenow.tv_sec + nd6_gctimer;
+                               ln->ln_expire = rt_expiry(rt, timenow.tv_sec,
+                                   nd6_gctimer);
                        }
                        RT_UNLOCK(rt);
                        break;
@@ -704,37 +796,36 @@ again:
                        break;
 
                case ND6_LLINFO_DELAY:
-                       if ((ndi.flags & ND6_IFF_PERFORMNUD) != 0) {
+                       if ((flags & ND6_IFF_PERFORMNUD) != 0) {
                                /* We need NUD */
                                ln->ln_asked = 1;
                                ln->ln_state = ND6_LLINFO_PROBE;
-                               ln->ln_expire = timenow.tv_sec +
-                                       ndi.retrans / 1000;
+                               ln->ln_expire = timenow.tv_sec + retrans / 1000;
                                RT_ADDREF_LOCKED(rt);
                                RT_UNLOCK(rt);
                                lck_mtx_unlock(rnh_lock);
                                nd6_ns_output(ifp, &dst->sin6_addr,
-                                   &dst->sin6_addr, ln, 0, 0);
+                                   &dst->sin6_addr, ln, 0);
                                lck_mtx_assert(rnh_lock,
                                    LCK_MTX_ASSERT_NOTOWNED);
                                RT_REMREF(rt);
                                goto again;
                        }
                        ln->ln_state = ND6_LLINFO_STALE; /* XXX */
-                       ln->ln_expire = timenow.tv_sec + nd6_gctimer;
+                       ln->ln_expire = rt_expiry(rt, timenow.tv_sec,
+                           nd6_gctimer);
                        RT_UNLOCK(rt);
                        break;
 
                case ND6_LLINFO_PROBE:
                        if (ln->ln_asked < nd6_umaxtries) {
                                ln->ln_asked++;
-                               ln->ln_expire = timenow.tv_sec +
-                                   ndi.retrans / 1000;
+                               ln->ln_expire = timenow.tv_sec + retrans / 1000;
                                RT_ADDREF_LOCKED(rt);
                                RT_UNLOCK(rt);
                                lck_mtx_unlock(rnh_lock);
                                nd6_ns_output(ifp, &dst->sin6_addr,
-                                   &dst->sin6_addr, ln, 0, 0);
+                                   &dst->sin6_addr, ln, 0);
                                RT_REMREF(rt);
                        } else {
                                RT_UNLOCK(rt);
@@ -773,12 +864,13 @@ again:
                if (dr->expire && dr->expire < timenow.tv_sec) {
                        struct nd_defrouter *t;
                        t = TAILQ_NEXT(dr, dr_entry);
-                       defrtrlist_del(dr, 1);
+                       defrtrlist_del(dr);
                        dr = t;
                } else {
                        dr = TAILQ_NEXT(dr, dr_entry);
                }
        }
+       lck_mtx_unlock(nd6_mutex);
 
        /*
         * expire interface addresses.
@@ -786,22 +878,21 @@ again:
         * However, from a stricter speci-confrmance standpoint, we should
         * rather separate address lifetimes and prefix lifetimes.
         */
-  addrloop:
+addrloop:
+       lck_rw_lock_exclusive(&in6_ifaddr_rwlock);
        for (ia6 = in6_ifaddrs; ia6; ia6 = nia6) {
                nia6 = ia6->ia_next;
+               IFA_LOCK(&ia6->ia_ifa);
+               /*
+                * Extra reference for ourselves; it's no-op if
+                * we don't have to regenerate temporary address,
+                * otherwise it protects the address from going
+                * away since we drop in6_ifaddr_rwlock below.
+                */
+               IFA_ADDREF_LOCKED(&ia6->ia_ifa);
                /* check address lifetime */
                lt6 = &ia6->ia6_lifetime;
                if (IFA6_IS_INVALID(ia6)) {
-                       int regen = 0;
-
-                       /*
-                        * Extra reference for ourselves; it's no-op if
-                        * we don't have to regenerate temporary address,
-                        * otherwise it protects the address from going
-                        * away since we drop nd6_mutex below.
-                        */
-                       ifaref(&ia6->ia_ifa);
-
                        /*
                         * If the expiring address is temporary, try
                         * regenerating a new one.  This would be useful when
@@ -810,30 +901,39 @@ again:
                         * addresses.  Although we may have to restart the
                         * loop (see below), it must be after purging the
                         * address.  Otherwise, we'd see an infinite loop of
-                        * regeneration. 
+                        * regeneration.
                         */
                        if (ip6_use_tempaddr &&
                            (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
-                               /* NOTE: We have to drop the lock here because 
-                                * regen_tmpaddr() eventually calls in6_update_ifa(),  
-                                * which must take the lock and would otherwise cause a 
-                                * hang. This is safe because the goto addrloop 
-                                * leads to a reevaluation of the in6_ifaddrs list
+                               /*
+                                * NOTE: We have to drop the lock here
+                                * because regen_tmpaddr() eventually calls
+                                * in6_update_ifa(), which must take the lock
+                                * and would otherwise cause a hang.  This is
+                                * safe because the goto addrloop leads to a
+                                * re-evaluation of the in6_ifaddrs list
                                 */
-                               lck_mtx_unlock(nd6_mutex);
-                               if (regen_tmpaddr(ia6) == 0) 
-                                       regen = 1;
-                               lck_mtx_lock(nd6_mutex);
+                               IFA_UNLOCK(&ia6->ia_ifa);
+                               lck_rw_done(&in6_ifaddr_rwlock);
+                               (void) regen_tmpaddr(ia6);
+                       } else {
+                               IFA_UNLOCK(&ia6->ia_ifa);
+                               lck_rw_done(&in6_ifaddr_rwlock);
                        }
 
-                       in6_purgeaddr(&ia6->ia_ifa, 1);
+                       /*
+                        * Purging the address would have caused
+                        * in6_ifaddr_rwlock to be dropped and reacquired;
+                        * therefore search again from the beginning
+                        * of in6_ifaddrs list.
+                        */
+                       in6_purgeaddr(&ia6->ia_ifa);
 
                        /* Release extra reference taken above */
-                       ifafree(&ia6->ia_ifa);
-
-                       if (regen)
-                               goto addrloop; /* XXX: see below */
+                       IFA_REMREF(&ia6->ia_ifa);
+                       goto addrloop;
                }
+               IFA_LOCK_ASSERT_HELD(&ia6->ia_ifa);
                if (IFA6_IS_DEPRECATED(ia6)) {
                        int oldflags = ia6->ia6_flags;
 
@@ -848,7 +948,8 @@ again:
                            (oldflags & IN6_IFF_DEPRECATED) == 0) {
 
                                /* see NOTE above */
-                               lck_mtx_unlock(nd6_mutex);
+                               IFA_UNLOCK(&ia6->ia_ifa);
+                               lck_rw_done(&in6_ifaddr_rwlock);
                                if (regen_tmpaddr(ia6) == 0) {
                                        /*
                                         * A new temporary address is
@@ -859,13 +960,16 @@ again:
                                         * would not cause disaster (because
                                         * it's not a deletion, but an
                                         * addition,) we'd rather restart the
-                                        * loop just for safety.  Or does this 
+                                        * loop just for safety.  Or does this
                                         * significantly reduce performance??
                                         */
-                                       lck_mtx_lock(nd6_mutex);
+                                       /* Release extra reference */
+                                       IFA_REMREF(&ia6->ia_ifa);
                                        goto addrloop;
                                }
-                               lck_mtx_lock(nd6_mutex);
+                               lck_rw_lock_exclusive(&in6_ifaddr_rwlock);
+                       } else {
+                               IFA_UNLOCK(&ia6->ia_ifa);
                        }
                } else {
                        /*
@@ -873,8 +977,26 @@ again:
                         * preferred.
                         */
                        ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
+                       IFA_UNLOCK(&ia6->ia_ifa);
                }
+               lck_rw_assert(&in6_ifaddr_rwlock, LCK_RW_ASSERT_EXCLUSIVE);
+               /* Release extra reference taken above */
+               IFA_REMREF(&ia6->ia_ifa);
        }
+       lck_rw_done(&in6_ifaddr_rwlock);
+
+       lck_mtx_lock(nd6_mutex);
+       /*
+        * Since we drop the nd6_mutex in prelist_remove, we want to run this
+        * section single threaded.
+        */
+       while (nd6_drain_busy) {
+               nd6_drain_waiters++;
+               msleep(nd6_drain_waitchan, nd6_mutex, (PZERO-1),
+                   __func__, NULL);
+               lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED);
+       }
+       nd6_drain_busy = TRUE;
 
        /* expire prefix list */
        pr = nd_prefix.lh_first;
@@ -884,21 +1006,107 @@ again:
                 * since pltime is just for autoconf, pltime processing for
                 * prefix is not necessary.
                 */
+               NDPR_LOCK(pr);
+               if (pr->ndpr_stateflags & NDPRF_PROCESSED) {
+                       NDPR_UNLOCK(pr);
+                       pr = pr->ndpr_next;
+                       continue;
+               }
                if (pr->ndpr_expire && pr->ndpr_expire < timenow.tv_sec) {
-                       struct nd_prefix *t;
-                       t = pr->ndpr_next;
-
                        /*
                         * address expiration and prefix expiration are
                         * separate.  NEVER perform in6_purgeaddr here.
                         */
-
-                       prelist_remove(pr, 1);
-                       pr = t;
-               } else
+                       pr->ndpr_stateflags |= NDPRF_PROCESSED;
+                       NDPR_ADDREF_LOCKED(pr);
+                       prelist_remove(pr);
+                       NDPR_UNLOCK(pr);
+                       NDPR_REMREF(pr);
+                       pr = nd_prefix.lh_first;
+               } else {
+                       pr->ndpr_stateflags |= NDPRF_PROCESSED;
+                       NDPR_UNLOCK(pr);
                        pr = pr->ndpr_next;
+               }
+       }
+       LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
+               NDPR_LOCK(pr);
+               pr->ndpr_stateflags &= ~NDPRF_PROCESSED;
+               NDPR_UNLOCK(pr);
+       }
+       nd6_drain_busy = FALSE;
+       if (nd6_drain_waiters > 0) {
+               nd6_drain_waiters = 0;
+               wakeup(nd6_drain_waitchan);
        }
        lck_mtx_unlock(nd6_mutex);
+}
+
+/*
+ * ND6 router advertisement kernel notification
+ */
+void
+nd6_post_msg(u_int32_t code, struct nd_prefix_list *prefix_list,
+    u_int32_t list_length, u_int32_t mtu, char *dl_addr, u_int32_t dl_addr_len)
+{
+       struct kev_msg ev_msg;
+       struct kev_nd6_ra_data nd6_ra_msg_data;
+       struct nd_prefix_list *itr = prefix_list;
+
+       bzero(&ev_msg, sizeof(struct kev_msg));
+       ev_msg.vendor_code    = KEV_VENDOR_APPLE;
+       ev_msg.kev_class      = KEV_NETWORK_CLASS;
+       ev_msg.kev_subclass   = KEV_ND6_SUBCLASS;
+       ev_msg.event_code     = code;
+
+       bzero(&nd6_ra_msg_data, sizeof(nd6_ra_msg_data));
+       nd6_ra_msg_data.lladdrlen = (dl_addr_len <= ND6_ROUTER_LL_SIZE) ?
+           dl_addr_len : ND6_ROUTER_LL_SIZE;
+       bcopy(dl_addr, &nd6_ra_msg_data.lladdr, nd6_ra_msg_data.lladdrlen);
+
+       if (mtu > 0 && mtu >= IPV6_MMTU) {
+               nd6_ra_msg_data.mtu = mtu;
+               nd6_ra_msg_data.flags |= KEV_ND6_DATA_VALID_MTU;
+       }
+
+       if (list_length > 0 && prefix_list != NULL) {
+               nd6_ra_msg_data.list_length = list_length;
+               nd6_ra_msg_data.flags |= KEV_ND6_DATA_VALID_PREFIX;
+       }
+
+       while (itr != NULL && nd6_ra_msg_data.list_index < list_length) {
+               bcopy(&itr->pr.ndpr_prefix, &nd6_ra_msg_data.prefix.prefix,
+                   sizeof (nd6_ra_msg_data.prefix.prefix));
+               nd6_ra_msg_data.prefix.raflags = itr->pr.ndpr_raf;
+               nd6_ra_msg_data.prefix.prefixlen = itr->pr.ndpr_plen;
+               nd6_ra_msg_data.prefix.origin = PR_ORIG_RA;
+               nd6_ra_msg_data.prefix.vltime = itr->pr.ndpr_vltime;
+               nd6_ra_msg_data.prefix.pltime = itr->pr.ndpr_pltime;
+               nd6_ra_msg_data.prefix.expire = itr->pr.ndpr_expire;
+               nd6_ra_msg_data.prefix.flags = itr->pr.ndpr_stateflags;
+               nd6_ra_msg_data.prefix.refcnt = itr->pr.ndpr_addrcnt;
+               nd6_ra_msg_data.prefix.if_index = itr->pr.ndpr_ifp->if_index;
+
+               /* send the message up */
+               ev_msg.dv[0].data_ptr       = &nd6_ra_msg_data;
+               ev_msg.dv[0].data_length    = sizeof(nd6_ra_msg_data);
+               ev_msg.dv[1].data_length    = 0;
+               kev_post_msg(&ev_msg);
+
+               /* clean up for the next prefix */
+               bzero(&nd6_ra_msg_data.prefix, sizeof(nd6_ra_msg_data.prefix));
+               itr = itr->next;
+               nd6_ra_msg_data.list_index++;
+       }
+}
+
+/*
+ * ND6 timer routine to expire default route list and prefix list
+ */
+void
+nd6_timer(__unused void        *ignored_arg)
+{
+       nd6_drain(NULL);
        timeout(nd6_timer, (caddr_t)0, nd6_prune * hz);
 }
 
@@ -914,25 +1122,29 @@ regen_tmpaddr(
        getmicrotime(&timenow);
 
        ifp = ia6->ia_ifa.ifa_ifp;
-       ifnet_lock_exclusive(ifp);
+       ifnet_lock_shared(ifp);
        for (ifa = ifp->if_addrlist.tqh_first; ifa;
             ifa = ifa->ifa_list.tqe_next)
        {
                struct in6_ifaddr *it6;
 
-               if (ifa->ifa_addr->sa_family != AF_INET6)
+               IFA_LOCK(ifa);
+               if (ifa->ifa_addr->sa_family != AF_INET6) {
+                       IFA_UNLOCK(ifa);
                        continue;
-
+               }
                it6 = (struct in6_ifaddr *)ifa;
 
                /* ignore no autoconf addresses. */
-               if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
+               if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) {
+                       IFA_UNLOCK(ifa);
                        continue;
-
+               }
                /* ignore autoconf addresses with different prefixes. */
-               if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
+               if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) {
+                       IFA_UNLOCK(ifa);
                        continue;
-
+               }
                /*
                 * Now we are looking at an autoconf address with the same
                 * prefix as ours.  If the address is temporary and is still
@@ -942,6 +1154,9 @@ regen_tmpaddr(
                 */
                if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
                    !IFA6_IS_DEPRECATED(it6)) {
+                       IFA_UNLOCK(ifa);
+                       if (public_ifa6 != NULL)
+                               IFA_REMREF(&public_ifa6->ia_ifa);
                        public_ifa6 = NULL;
                        break;
                }
@@ -952,8 +1167,15 @@ regen_tmpaddr(
                 * loop here, because there may be a still-preferred temporary
                 * address with the prefix.
                 */
-               if (!IFA6_IS_DEPRECATED(it6))
-                   public_ifa6 = it6;
+               if (!IFA6_IS_DEPRECATED(it6)) {
+                       IFA_ADDREF_LOCKED(ifa); /* for public_ifa6 */
+                       IFA_UNLOCK(ifa);
+                       if (public_ifa6 != NULL)
+                               IFA_REMREF(&public_ifa6->ia_ifa);
+                       public_ifa6 = it6;
+               } else {
+                       IFA_UNLOCK(ifa);
+               }
        }
        ifnet_lock_done(ifp);
 
@@ -963,8 +1185,10 @@ regen_tmpaddr(
                if ((e = in6_tmpifadd(public_ifa6, 0, M_WAITOK)) != 0) {
                        log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
                            " tmp addr,errno=%d\n", e);
+                       IFA_REMREF(&public_ifa6->ia_ifa);
                        return(-1);
                }
+               IFA_REMREF(&public_ifa6->ia_ifa);
                return(0);
        }
 
@@ -980,7 +1204,7 @@ nd6_purge(
        struct ifnet *ifp)
 {
        struct llinfo_nd6 *ln;
-       struct nd_defrouter *dr, *ndr, drany;
+       struct nd_defrouter *dr, *ndr;
        struct nd_prefix *pr, *npr;
 
        /* Nuke default router list entries toward ifp */
@@ -992,18 +1216,38 @@ nd6_purge(
                 */
                for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
                        ndr = TAILQ_NEXT(dr, dr_entry);
+                       if (dr->stateflags & NDDRF_INSTALLED)
+                               continue;
                        if (dr->ifp == ifp)
-                               defrtrlist_del(dr, 1);
+                               defrtrlist_del(dr);
                }
                dr = TAILQ_FIRST(&nd_defrouter);
                if (dr->ifp == ifp)
-                       defrtrlist_del(dr, 1);
+                       defrtrlist_del(dr);
+       }
+
+       for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
+               ndr = TAILQ_NEXT(dr, dr_entry);
+               if (!(dr->stateflags & NDDRF_INSTALLED))
+                       continue;
+
+               if (dr->ifp == ifp)
+                       defrtrlist_del(dr);
        }
 
        /* Nuke prefix list entries toward ifp */
        for (pr = nd_prefix.lh_first; pr; pr = npr) {
                npr = pr->ndpr_next;
+               NDPR_LOCK(pr);
                if (pr->ndpr_ifp == ifp) {
+                       /*
+                        * Because if_detach() does *not* release prefixes
+                        * while purging addresses the reference count will
+                        * still be above zero. We therefore reset it to
+                        * make sure that the prefix really gets purged.
+                        */
+                       pr->ndpr_addrcnt = 0;
+
                        /*
                         * Previously, pr->ndpr_addr is removed as well,
                         * but I strongly believe we don't have to do it.
@@ -1012,27 +1256,31 @@ nd6_purge(
                         * by itself.
                         * (jinmei@kame.net 20010129)
                         */
-                       prelist_remove(pr, 1);
+                       NDPR_ADDREF_LOCKED(pr);
+                       prelist_remove(pr);
+                       NDPR_UNLOCK(pr);
+                       NDPR_REMREF(pr);
+               } else {
+                       NDPR_UNLOCK(pr);
                }
        }
+       lck_mtx_unlock(nd6_mutex);
 
        /* cancel default outgoing interface setting */
        if (nd6_defifindex == ifp->if_index) {
-               /* Release nd6_mutex as it will be acquired 
-                * during nd6_setdefaultiface again
-                */ 
-               lck_mtx_unlock(nd6_mutex);
                nd6_setdefaultiface(0);
-               lck_mtx_lock(nd6_mutex);
        }
 
-       if (!ip6_forwarding && (ip6_accept_rtadv || (ifp->if_eflags & IFEF_ACCEPT_RTADVD))) { 
+       /*
+        * Perform default router selection even when we are a router,
+        * if Scoped Routing is enabled.
+        */
+       if (ip6_doscopedroute || !ip6_forwarding) {
+               lck_mtx_lock(nd6_mutex);
                /* refresh default router list */
-               bzero(&drany, sizeof(drany));
-               defrouter_delreq(&drany, 0);
-               defrouter_select();
+               defrouter_select(ifp);
+               lck_mtx_unlock(nd6_mutex);
        }
-       lck_mtx_unlock(nd6_mutex);
 
        /*
         * Nuke neighbor cache entries for the ifp.
@@ -1091,28 +1339,31 @@ nd6_lookup(
 {
        struct rtentry *rt;
        struct sockaddr_in6 sin6;
+       unsigned int ifscope;
 
        bzero(&sin6, sizeof(sin6));
        sin6.sin6_len = sizeof(struct sockaddr_in6);
        sin6.sin6_family = AF_INET6;
        sin6.sin6_addr = *addr6;
-#if SCOPEDROUTING
-       sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
-#endif
-       if (rt_locked)
-               lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED);
 
-       rt = rt_locked ? rtalloc1_locked((struct sockaddr *)&sin6, create, 0) :
-           rtalloc1((struct sockaddr *)&sin6, create, 0);
+       ifscope = (ifp != NULL) ? ifp->if_index : IFSCOPE_NONE;
+       if (rt_locked) {
+               lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED);
+               rt = rtalloc1_scoped_locked((struct sockaddr *)&sin6,
+                   create, 0, ifscope);
+       } else {
+               rt = rtalloc1_scoped((struct sockaddr *)&sin6,
+                   create, 0, ifscope);
+       }
 
        if (rt != NULL) {
                RT_LOCK(rt);
                if ((rt->rt_flags & RTF_LLINFO) == 0) {
                        /*
-                        * This is the case for the default route.  If we
-                        * want to create a neighbor cache for the address,
-                        * we should free the route for the destination and
-                        * allocate an interface route.
+                        * This is the case for the default route.
+                        * If we want to create a neighbor cache for the
+                        * address, we should free the route for the
+                        * destination and allocate an interface route.
                         */
                        if (create) {
                                RT_UNLOCK(rt);
@@ -1127,6 +1378,7 @@ nd6_lookup(
        if (rt == NULL) {
                if (create && ifp) {
                        struct ifaddr *ifa;
+                       u_int32_t ifa_flags;
                        int e;
 
                        /*
@@ -1148,11 +1400,14 @@ nd6_lookup(
                         */
                        if (!rt_locked)
                                lck_mtx_lock(rnh_lock);
-                       if ((e = rtrequest_locked(RTM_ADD,
+                       IFA_LOCK_SPIN(ifa);
+                       ifa_flags = ifa->ifa_flags;
+                       IFA_UNLOCK(ifa);
+                       if ((e = rtrequest_scoped_locked(RTM_ADD,
                            (struct sockaddr *)&sin6, ifa->ifa_addr,
                            (struct sockaddr *)&all1_sa,
-                           (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
-                           ~RTF_CLONING, &rt)) != 0) {
+                           (ifa_flags | RTF_HOST | RTF_LLINFO) &
+                           ~RTF_CLONING, &rt, ifscope)) != 0) {
                                if (e != EEXIST)
                                        log(LOG_ERR, "%s: failed to add route "
                                            "for a neighbor(%s), errno=%d\n",
@@ -1160,7 +1415,7 @@ nd6_lookup(
                        }
                        if (!rt_locked)
                                lck_mtx_unlock(rnh_lock);
-                       ifafree(ifa);
+                       IFA_REMREF(ifa);
                        if (rt == NULL)
                                return(NULL);
 
@@ -1184,12 +1439,21 @@ nd6_lookup(
         *      it might be the loopback interface if the entry is for our
         *      own address on a non-loopback interface. Instead, we should
         *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
-        *      interface.
+        *      interface.
+        * Note also that ifa_ifp and ifp may differ when we connect two
+        * interfaces to a same link, install a link prefix to an interface,
+        * and try to install a neighbor cache on an interface that does not
+        * have a route to the prefix.
+        *
+        * If the address is from a proxied prefix, the ifa_ifp and ifp might
+        * not match, because nd6_na_input() could have modified the ifp
+        * of the route to point to the interface where the NA arrived on,
+        * hence the test for RTF_PROXY.
         */
-       if (((ifp && (ifp->if_type != IFT_PPP)) && ((ifp->if_eflags & IFEF_NOAUTOIPV6LL) == 0)) &&
-           ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
+       if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
            rt->rt_gateway->sa_family != AF_LINK ||  rt->rt_llinfo == NULL ||
-           (ifp && rt->rt_ifa->ifa_ifp != ifp))) {
+           (ifp && rt->rt_ifa->ifa_ifp != ifp &&
+           !(rt->rt_flags & RTF_PROXY))) {
                RT_REMREF_LOCKED(rt);
                RT_UNLOCK(rt);
                if (create) {
@@ -1207,73 +1471,136 @@ nd6_lookup(
 }
 
 /*
- * Detect if a given IPv6 address identifies a neighbor on a given link.
- * XXX: should take care of the destination of a p2p link?
+ * Test whether a given IPv6 address is a neighbor or not, ignoring
+ * the actual neighbor cache.  The neighbor cache is ignored in order
+ * to not reenter the routing code from within itself.
  */
-int
-nd6_is_addr_neighbor(
+static int
+nd6_is_new_addr_neighbor(
        struct sockaddr_in6 *addr,
-       struct ifnet *ifp,
-       int    rt_locked)
+       struct ifnet *ifp)
 {
-       struct ifaddr *ifa;
-       struct rtentry *rt;
-       int i;
+       struct nd_prefix *pr;
+       struct ifaddr *dstaddr;
 
-#define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
-#define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
+       lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED);
 
        /*
         * A link-local address is always a neighbor.
-        * XXX: we should use the sin6_scope_id field rather than the embedded
-        * interface index.
+        * XXX: a link does not necessarily specify a single interface.
         */
-       if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
-           ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
-               return(1);
+       if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
+               struct sockaddr_in6 sin6_copy;
+               u_int32_t zone;
+
+               /*
+                * We need sin6_copy since sa6_recoverscope() may modify the
+                * content (XXX).
+                */
+               sin6_copy = *addr;
+               if (sa6_recoverscope(&sin6_copy, FALSE))
+                       return (0); /* XXX: should be impossible */
+               if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
+                       return (0);
+               if (sin6_copy.sin6_scope_id == zone)
+                       return (1);
+               else
+                       return (0);
+       }
 
        /*
         * If the address matches one of our addresses,
         * it should be a neighbor.
+        * If the address matches one of our on-link prefixes, it should be a
+        * neighbor.
         */
-       ifnet_lock_shared(ifp);
-       for (ifa = ifp->if_addrlist.tqh_first;
-            ifa;
-            ifa = ifa->ifa_list.tqe_next)
-       {
-               if (ifa->ifa_addr->sa_family != AF_INET6)
+       for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
+               NDPR_LOCK(pr);
+               if (pr->ndpr_ifp != ifp) {
+                       NDPR_UNLOCK(pr);
+                       continue;
+               }
+               if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
+                       NDPR_UNLOCK(pr);
                        continue;
+               }
+               if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
+                   &addr->sin6_addr, &pr->ndpr_mask)) {
+                       NDPR_UNLOCK(pr);
+                       return (1);
+               }
+               NDPR_UNLOCK(pr);
+       }
 
-               for (i = 0; i < 4; i++) {
-                       if ((IFADDR6(ifa).s6_addr32[i] ^
-                            addr->sin6_addr.s6_addr32[i]) &
-                               IFMASK6(ifa).s6_addr32[i])
-                               continue;
+       /*
+        * If the address is assigned on the node of the other side of
+        * a p2p interface, the address should be a neighbor.
+        */
+       dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
+       if (dstaddr != NULL) {
+               if (dstaddr->ifa_ifp == ifp) {
+                       IFA_REMREF(dstaddr);
+                       return (1);
                }
-               ifnet_lock_done(ifp);
-               return(1);
+               IFA_REMREF(dstaddr);
+               dstaddr = NULL;
        }
-       ifnet_lock_done(ifp);
+
+       /*
+        * If the default router list is empty, all addresses are regarded
+        * as on-link, and thus, as a neighbor.
+        * XXX: we restrict the condition to hosts, because routers usually do
+        * not have the "default router list".
+        * XXX: this block should eventually be removed (it is disabled when
+        * Scoped Routing is in effect); treating all destinations as on-link
+        * in the absence of a router is rather harmful.
+        */
+       if (!ip6_doscopedroute && !ip6_forwarding &&
+           TAILQ_FIRST(&nd_defrouter) == NULL &&
+           nd6_defifindex == ifp->if_index) {
+               return (1);
+       }
+
+       return (0);
+}
+
+
+/*
+ * Detect if a given IPv6 address identifies a neighbor on a given link.
+ * XXX: should take care of the destination of a p2p link?
+ */
+int
+nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp, int rt_locked)
+{
+       struct rtentry *rt;
+
+       lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_NOTOWNED);
+       lck_mtx_lock(nd6_mutex);
+       if (nd6_is_new_addr_neighbor(addr, ifp)) {
+               lck_mtx_unlock(nd6_mutex);
+               return (1);
+       }
+       lck_mtx_unlock(nd6_mutex);
 
        /*
         * Even if the address matches none of our addresses, it might be
-        * in the neighbor cache.  Callee returns a locked route upon
-        * success.
+        * in the neighbor cache.
         */
        if ((rt = nd6_lookup(&addr->sin6_addr, 0, ifp, rt_locked)) != NULL) {
                RT_LOCK_ASSERT_HELD(rt);
                RT_REMREF_LOCKED(rt);
                RT_UNLOCK(rt);
-               return(1);
+               return (1);
        }
 
-       return(0);
-#undef IFADDR6
-#undef IFMASK6
+       return (0);
 }
 
 /*
  * Free an nd6 llinfo entry.
+ * Since the function would cause significant changes in the kernel, DO NOT
+ * make it global, unless you have a strong reason for the change, and are sure
+ * that the change is safe.
  */
 void
 nd6_free(
@@ -1290,7 +1617,7 @@ nd6_free(
        RT_LOCK(rt);
        RT_ADDREF_LOCKED(rt);   /* Extra ref */
        ln = rt->rt_llinfo;
-       in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
+       in6 = ((struct sockaddr_in6 *)(void *)rt_key(rt))->sin6_addr;
 
        /*
         * Prevent another thread from modifying rt_key, rt_gateway
@@ -1300,14 +1627,13 @@ nd6_free(
        rt->rt_flags |= RTF_CONDEMNED;
 
        /*
-        * we used to have pfctlinput(PRC_HOSTDEAD) here. 
-        * even though it is not harmful, it was not really necessary.
+        * We used to have pfctlinput(PRC_HOSTDEAD) here.  Even though it is
+        * not harmful, it was not really necessary.  Perform default router
+        * selection even when we are a router, if Scoped Routing is enabled.
         */
-
-       if (!ip6_forwarding && (ip6_accept_rtadv ||
-           (rt->rt_ifp->if_eflags & IFEF_ACCEPT_RTADVD))) {
-               dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->
-                   sin6_addr, rt->rt_ifp);
+       if (ip6_doscopedroute || !ip6_forwarding) {
+               dr = defrouter_lookup(&((struct sockaddr_in6 *)(void *)
+                   rt_key(rt))->sin6_addr, rt->rt_ifp);
 
                if ((ln && ln->ln_router) || dr) {
                        /*
@@ -1316,12 +1642,15 @@ nd6_free(
                         * See a corresponding comment in nd6_na_input().
                         */
                        RT_UNLOCK(rt);
+                       lck_mtx_unlock(nd6_mutex);
                        rt6_flush(&in6, rt->rt_ifp);
+                       lck_mtx_lock(nd6_mutex);
                } else {
                        RT_UNLOCK(rt);
                }
 
                if (dr) {
+                       NDDR_REMREF(dr);
                        /*
                         * Unreachablity of a router might affect the default
                         * router selection and on-link detection of advertised
@@ -1345,21 +1674,12 @@ nd6_free(
                         * the check now.
                         */
                        RT_UNLOCK(rt);
-                       pfxlist_onlink_check(1);
+                       pfxlist_onlink_check();
 
-                       if (dr == TAILQ_FIRST(&nd_defrouter)) {
-                               /*
-                                * It is used as the current default router,
-                                * so we have to move it to the end of the
-                                * list and choose a new one.
-                                * XXX: it is not very efficient if this is
-                                *      the only router.
-                                */
-                               TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
-                               TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
-
-                               defrouter_select();
-                       }
+                       /*
+                        * refresh default router list
+                        */
+                       defrouter_select(rt->rt_ifp);
                }
                RT_LOCK_ASSERT_NOTHELD(rt);
        } else {
@@ -1382,7 +1702,7 @@ nd6_free(
 /*
  * Upper-layer reachability hint for Neighbor Unreachability Detection.
  *
- * XXX cost-effective metods?
+ * XXX cost-effective methods?
  */
 void
 nd6_nud_hint(
@@ -1435,9 +1755,14 @@ nd6_nud_hint(
 
        ln->ln_state = ND6_LLINFO_REACHABLE;
        if (ln->ln_expire) {
+               struct nd_ifinfo *ndi;
+
                lck_rw_lock_shared(nd_if_rwlock);
-               ln->ln_expire = timenow.tv_sec +
-                       nd_ifinfo[rt->rt_ifp->if_index].reachable;
+               ndi = ND_IFINFO(rt->rt_ifp);
+               VERIFY(ndi != NULL && ndi->initialized);
+               lck_mtx_lock(&ndi->lock);
+               ln->ln_expire = timenow.tv_sec + ndi->reachable;
+               lck_mtx_unlock(&ndi->lock);
                lck_rw_done(nd_if_rwlock);
        }
 done:
@@ -1482,17 +1807,17 @@ nd6_rtrequest(
                if (!nd6_need_cache(ifp)) {     /* stf case */
                        no_nd_cache = 1;
                } else {
+                       struct sockaddr_in6 sin6;
+
+                       rtkey_to_sa6(rt, &sin6);
                        /*
                         * nd6_is_addr_neighbor() may call nd6_lookup(),
                         * therefore we drop rt_lock to avoid deadlock
-                        * during the lookup.  Using rt_key(rt) is still
-                        * safe because it won't change while rnh_lock
-                        * is held.
+                        * during the lookup.
                         */
                        RT_ADDREF_LOCKED(rt);
                        RT_UNLOCK(rt);
-                       no_nd_cache = !nd6_is_addr_neighbor(
-                           (struct sockaddr_in6 *)rt_key(rt), ifp, 1);
+                       no_nd_cache = !nd6_is_addr_neighbor(&sin6, ifp, 1);
                        RT_LOCK(rt);
                        RT_REMREF_LOCKED(rt);
                }
@@ -1527,12 +1852,13 @@ nd6_rtrequest(
                 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
                 *         rt->rt_flags |= RTF_CLONING;
                 */
-               if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
+               if ((rt->rt_flags & RTF_CLONING) ||
+                   ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) {
                        /*
-                        * Case 1: This route should come from
-                        * a route to interface.  RTF_LLINFO flag is set
-                        * for a host route whose destination should be
-                        * treated as on-link.
+                        * Case 1: This route should come from a route to
+                        * interface (RTF_CLONING case) or the route should be
+                        * treated as on-link but is currently not
+                        * (RTF_LLINFO && ln == NULL case).
                         */
                        if (rt_setgate(rt, rt_key(rt),
                            (struct sockaddr *)&null_sdl) == 0) {
@@ -1567,15 +1893,6 @@ nd6_rtrequest(
                 *   (or should we allow proxy ND configuration only for
                 *   routers?  there's no mention about proxy ND from hosts)
                 */
-#if 0
-               /* XXX it does not work */
-               if (rt->rt_flags & RTF_ANNOUNCE)
-                       nd6_na_output(ifp,
-                             &SIN6(rt_key(rt))->sin6_addr,
-                             &SIN6(rt_key(rt))->sin6_addr,
-                             ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
-                             1, NULL);
-#endif
                /* FALLTHROUGH */
        case RTM_RESOLVE:
                if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
@@ -1585,9 +1902,12 @@ nd6_rtrequest(
                         */
                        if (gate->sa_family != AF_LINK ||
                            gate->sa_len < sizeof(null_sdl)) {
-                               log(LOG_DEBUG,
-                                   "nd6_rtrequest: bad gateway value: %s\n",
-                                   if_name(ifp));
+                               /* Don't complain in case of RTM_ADD */
+                               if (req == RTM_RESOLVE) {
+                                       log(LOG_DEBUG,
+                                           "nd6_rtrequest: bad gateway "
+                                               "value: %s\n", if_name(ifp));
+                               }
                                break;
                        }
                        SDL(gate)->sdl_type = ifp->if_type;
@@ -1604,6 +1924,9 @@ nd6_rtrequest(
                        log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
                        break;
                }
+               rt->rt_llinfo_get_ri = nd6_llinfo_get_ri;
+               rt->rt_llinfo_get_iflri = nd6_llinfo_get_iflri;
+               rt->rt_llinfo_purge = nd6_llinfo_purge;
                rt->rt_llinfo_free = nd6_llinfo_free;
 
                nd6_inuse++;
@@ -1680,6 +2003,22 @@ nd6_rtrequest(
                                SDL(gate)->sdl_alen = ifp->if_addrlen;
                        }
                        if (nd6_useloopback) {
+                               if (rt->rt_ifp != lo_ifp) {
+                                       /*
+                                        * Purge any link-layer info caching.
+                                        */
+                                       if (rt->rt_llinfo_purge != NULL)
+                                               rt->rt_llinfo_purge(rt);
+
+                                       /*
+                                        * Adjust route ref count for the
+                                        * interfaces.
+                                        */
+                                       if (rt->rt_if_ref_fn != NULL) {
+                                               rt->rt_if_ref_fn(lo_ifp, 1);
+                                               rt->rt_if_ref_fn(rt->rt_ifp, -1);
+                                       }
+                               }
                                rt->rt_ifp = lo_ifp;    /* XXX */
                                /*
                                 * Make sure rt_ifa be equal to the ifaddr
@@ -1693,7 +2032,7 @@ nd6_rtrequest(
                                        rtsetifa(rt, ifa);
                                }
                        }
-                       ifafree(ifa);
+                       IFA_REMREF(ifa);
                } else if (rt->rt_flags & RTF_ANNOUNCE) {
                        ln->ln_expire = 0;
                        ln->ln_state = ND6_LLINFO_REACHABLE;
@@ -1702,26 +2041,30 @@ nd6_rtrequest(
                        /* join solicited node multicast for proxy ND */
                        if (ifp->if_flags & IFF_MULTICAST) {
                                struct in6_addr llsol;
+                               struct in6_multi *in6m;
                                int error;
 
                                llsol = SIN6(rt_key(rt))->sin6_addr;
-                               llsol.s6_addr16[0] = htons(0xff02);
-                               llsol.s6_addr16[1] = htons(ifp->if_index);
+                               llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL;
                                llsol.s6_addr32[1] = 0;
                                llsol.s6_addr32[2] = htonl(1);
                                llsol.s6_addr8[12] = 0xff;
-
-                               if (!in6_addmulti(&llsol, ifp, &error, 0)) {
+                               if (in6_setscope(&llsol, ifp, NULL))
+                                       break;
+                               error = in6_mc_join(ifp, &llsol, NULL, &in6m, 0);
+                               if (error) {
                                        nd6log((LOG_ERR, "%s: failed to join "
                                            "%s (errno=%d)\n", if_name(ifp),
                                            ip6_sprintf(&llsol), error));
+                               } else {
+                                       IN6M_REMREF(in6m);
                                }
                        }
                }
                break;
 
        case RTM_DELETE:
-               if (!ln)
+               if (ln == NULL)
                        break;
                /* leave from solicited node multicast for proxy ND */
                if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
@@ -1730,17 +2073,19 @@ nd6_rtrequest(
                        struct in6_multi *in6m;
 
                        llsol = SIN6(rt_key(rt))->sin6_addr;
-                       llsol.s6_addr16[0] = htons(0xff02);
-                       llsol.s6_addr16[1] = htons(ifp->if_index);
+                       llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL;
                        llsol.s6_addr32[1] = 0;
                        llsol.s6_addr32[2] = htonl(1);
                        llsol.s6_addr8[12] = 0xff;
-
-                       ifnet_lock_shared(ifp);
-                       IN6_LOOKUP_MULTI(llsol, ifp, in6m);
-                       ifnet_lock_done(ifp);
-                       if (in6m)
-                               in6_delmulti(in6m, 0);
+                       if (in6_setscope(&llsol, ifp, NULL) == 0) {
+                               in6_multihead_lock_shared();
+                               IN6_LOOKUP_MULTI(&llsol, ifp, in6m);
+                               in6_multihead_lock_done();
+                               if (in6m != NULL) {
+                                       in6_mc_leave(in6m, NULL);
+                                       IN6M_REMREF(in6m);
+                               }
+                       }
                }
                nd6_inuse--;
                /*
@@ -1751,27 +2096,43 @@ nd6_rtrequest(
                 */
                if (ln->ln_flags & ND6_LNF_IN_USE)
                        LN_DEQUEUE(ln);
+
+               /*
+                * Purge any link-layer info caching.
+                */
+               if (rt->rt_llinfo_purge != NULL)
+                       rt->rt_llinfo_purge(rt);
+
                rt->rt_flags &= ~RTF_LLINFO;
-               if (ln->ln_hold != NULL)
+               if (ln->ln_hold != NULL) {
                        m_freem(ln->ln_hold);
-               ln->ln_hold = NULL;
+                       ln->ln_hold = NULL;
+               }
        }
 }
 
-static void
+static int
 nd6_siocgdrlst(void *data, int data_is_64)
 {
-       struct in6_drlist_64 *drl_64 = (struct in6_drlist_64 *)data;
-       struct in6_drlist_32 *drl_32 = (struct in6_drlist_32 *)data;
+       struct in6_drlist_32 *drl_32;
        struct nd_defrouter *dr;
        int i = 0;
 
        lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED);
 
-       bzero(data, data_is_64 ? sizeof (*drl_64) : sizeof (*drl_32));
        dr = TAILQ_FIRST(&nd_defrouter);
+
+       /* For 64-bit process */
        if (data_is_64) {
-               /* For 64-bit process */
+               struct in6_drlist_64 *drl_64;
+
+               drl_64 = _MALLOC(sizeof (*drl_64), M_TEMP, M_WAITOK|M_ZERO);
+               if (drl_64 == NULL)
+                       return (ENOMEM);
+
+               /* preserve the interface name */
+               bcopy(data, drl_64, sizeof (drl_64->ifname));
+
                while (dr && i < DRLSTSIZ) {
                        drl_64->defrouter[i].rtaddr = dr->rtaddr;
                        if (IN6_IS_ADDR_LINKLOCAL(&drl_64->defrouter[i].rtaddr)) {
@@ -1790,9 +2151,19 @@ nd6_siocgdrlst(void *data, int data_is_64)
                        i++;
                        dr = TAILQ_NEXT(dr, dr_entry);
                }
-               return;
+               bcopy(drl_64, data, sizeof (*drl_64));
+               _FREE(drl_64, M_TEMP);
+               return (0);
        }
+
        /* For 32-bit process */
+       drl_32 = _MALLOC(sizeof (*drl_32), M_TEMP, M_WAITOK|M_ZERO);
+       if (drl_32 == NULL)
+               return (ENOMEM);
+
+       /* preserve the interface name */
+       bcopy(data, drl_32, sizeof (drl_32->ifname));
+
        while (dr && i < DRLSTSIZ) {
                drl_32->defrouter[i].rtaddr = dr->rtaddr;
                if (IN6_IS_ADDR_LINKLOCAL(&drl_32->defrouter[i].rtaddr)) {
@@ -1811,33 +2182,45 @@ nd6_siocgdrlst(void *data, int data_is_64)
                i++;
                dr = TAILQ_NEXT(dr, dr_entry);
        }
+       bcopy(drl_32, data, sizeof (*drl_32));
+       _FREE(drl_32, M_TEMP);
+       return (0);
 }
 
-static void
+/*
+ * XXX meaning of fields, especialy "raflags", is very
+ * differnet between RA prefix list and RR/static prefix list.
+ * how about separating ioctls into two?
+ */
+static int
 nd6_siocgprlst(void *data, int data_is_64)
 {
-       struct in6_prlist_64 *prl_64 = (struct in6_prlist_64 *)data;
-       struct in6_prlist_32 *prl_32 = (struct in6_prlist_32 *)data;
+       struct in6_prlist_32 *prl_32;
        struct nd_prefix *pr;
-       struct rr_prefix *rpp;
        int i = 0;
 
        lck_mtx_assert(nd6_mutex, LCK_MTX_ASSERT_OWNED);
-       /*
-        * XXX meaning of fields, especialy "raflags", is very
-        * differnet between RA prefix list and RR/static prefix list.
-        * how about separating ioctls into two?
-        */
-       bzero(data, data_is_64 ? sizeof (*prl_64) : sizeof (*prl_32));
+
        pr = nd_prefix.lh_first;
+
+       /* For 64-bit process */
        if (data_is_64) {
-               /* For 64-bit process */
+               struct in6_prlist_64 *prl_64;
+
+               prl_64 = _MALLOC(sizeof (*prl_64), M_TEMP, M_WAITOK|M_ZERO);
+               if (prl_64 == NULL)
+                       return (ENOMEM);
+
+               /* preserve the interface name */
+               bcopy(data, prl_64, sizeof (prl_64->ifname));
+
                while (pr && i < PRLSTSIZ) {
                        struct nd_pfxrouter *pfr;
                        int j;
 
+                       NDPR_LOCK(pr);
                        (void) in6_embedscope(&prl_64->prefix[i].prefix,
-                           &pr->ndpr_prefix, NULL, NULL);
+                           &pr->ndpr_prefix, NULL, NULL, NULL);
                        prl_64->prefix[i].raflags = pr->ndpr_raf;
                        prl_64->prefix[i].prefixlen = pr->ndpr_plen;
                        prl_64->prefix[i].vltime = pr->ndpr_vltime;
@@ -1868,36 +2251,31 @@ nd6_siocgprlst(void *data, int data_is_64)
                        }
                        prl_64->prefix[i].advrtrs = j;
                        prl_64->prefix[i].origin = PR_ORIG_RA;
+                       NDPR_UNLOCK(pr);
 
                        i++;
                        pr = pr->ndpr_next;
                }
-
-               for (rpp = LIST_FIRST(&rr_prefix); rpp;
-                    rpp = LIST_NEXT(rpp, rp_entry)) {
-                       if (i >= PRLSTSIZ)
-                               break;
-                       (void) in6_embedscope(&prl_64->prefix[i].prefix,
-                           &pr->ndpr_prefix, NULL, NULL);
-                       prl_64->prefix[i].raflags = rpp->rp_raf;
-                       prl_64->prefix[i].prefixlen = rpp->rp_plen;
-                       prl_64->prefix[i].vltime = rpp->rp_vltime;
-                       prl_64->prefix[i].pltime = rpp->rp_pltime;
-                       prl_64->prefix[i].if_index = rpp->rp_ifp->if_index;
-                       prl_64->prefix[i].expire = rpp->rp_expire;
-                       prl_64->prefix[i].advrtrs = 0;
-                       prl_64->prefix[i].origin = rpp->rp_origin;
-                       i++;
-               }
-               return;
+               bcopy(prl_64, data, sizeof (*prl_64));
+               _FREE(prl_64, M_TEMP);
+               return (0);
        }
+
        /* For 32-bit process */
+       prl_32 = _MALLOC(sizeof (*prl_32), M_TEMP, M_WAITOK|M_ZERO);
+       if (prl_32 == NULL)
+               return (ENOMEM);
+
+       /* preserve the interface name */
+       bcopy(data, prl_32, sizeof (prl_32->ifname));
+
        while (pr && i < PRLSTSIZ) {
                struct nd_pfxrouter *pfr;
                int j;
 
+               NDPR_LOCK(pr);
                (void) in6_embedscope(&prl_32->prefix[i].prefix,
-                   &pr->ndpr_prefix, NULL, NULL);
+                   &pr->ndpr_prefix, NULL, NULL, NULL);
                prl_32->prefix[i].raflags = pr->ndpr_raf;
                prl_32->prefix[i].prefixlen = pr->ndpr_plen;
                prl_32->prefix[i].vltime = pr->ndpr_vltime;
@@ -1928,141 +2306,203 @@ nd6_siocgprlst(void *data, int data_is_64)
                }
                prl_32->prefix[i].advrtrs = j;
                prl_32->prefix[i].origin = PR_ORIG_RA;
+               NDPR_UNLOCK(pr);
 
                i++;
                pr = pr->ndpr_next;
        }
-
-       for (rpp = LIST_FIRST(&rr_prefix); rpp;
-            rpp = LIST_NEXT(rpp, rp_entry)) {
-               if (i >= PRLSTSIZ)
-                       break;
-               (void) in6_embedscope(&prl_32->prefix[i].prefix,
-                   &pr->ndpr_prefix, NULL, NULL);
-               prl_32->prefix[i].raflags = rpp->rp_raf;
-               prl_32->prefix[i].prefixlen = rpp->rp_plen;
-               prl_32->prefix[i].vltime = rpp->rp_vltime;
-               prl_32->prefix[i].pltime = rpp->rp_pltime;
-               prl_32->prefix[i].if_index = rpp->rp_ifp->if_index;
-               prl_32->prefix[i].expire = rpp->rp_expire;
-               prl_32->prefix[i].advrtrs = 0;
-               prl_32->prefix[i].origin = rpp->rp_origin;
-               i++;
-       }
+       bcopy(prl_32, data, sizeof (*prl_32));
+       _FREE(prl_32, M_TEMP);
+       return (0);
 }
 
 int
 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
 {
-       struct in6_ndireq *ndi = (struct in6_ndireq *)data;
-       struct in6_ondireq *ondi = (struct in6_ondireq *)data;
-       struct nd_defrouter *dr, any;
+       struct nd_defrouter *dr;
        struct nd_prefix *pr;
        struct rtentry *rt;
        int i = ifp->if_index, error = 0;
 
        switch (cmd) {
-       case SIOCGDRLST_IN6_32:
-       case SIOCGDRLST_IN6_64:
+       case SIOCGDRLST_IN6_32:         /* struct in6_drlist_32 */
+       case SIOCGDRLST_IN6_64:         /* struct in6_drlist_64 */
                /*
                 * obsolete API, use sysctl under net.inet6.icmp6
                 */
                lck_mtx_lock(nd6_mutex);
-               nd6_siocgdrlst(data, cmd == SIOCGDRLST_IN6_64);
+               error = nd6_siocgdrlst(data, cmd == SIOCGDRLST_IN6_64);
                lck_mtx_unlock(nd6_mutex);
                break;
 
-       case SIOCGPRLST_IN6_32:
-       case SIOCGPRLST_IN6_64:
+       case SIOCGPRLST_IN6_32:         /* struct in6_prlist_32 */
+       case SIOCGPRLST_IN6_64:         /* struct in6_prlist_64 */
                /*
                 * obsolete API, use sysctl under net.inet6.icmp6
                 */
                lck_mtx_lock(nd6_mutex);
-               nd6_siocgprlst(data, cmd == SIOCGPRLST_IN6_64);
+               error = nd6_siocgprlst(data, cmd == SIOCGPRLST_IN6_64);
                lck_mtx_unlock(nd6_mutex);
                break;
 
-       case OSIOCGIFINFO_IN6:
-       case SIOCGIFINFO_IN6:
+       case OSIOCGIFINFO_IN6:          /* struct in6_ondireq */
+       case SIOCGIFINFO_IN6: {         /* struct in6_ondireq */
+               u_int32_t linkmtu;
+               struct in6_ondireq *ondi = (struct in6_ondireq *)(void *)data;
+               struct nd_ifinfo *ndi;
                /*
                 * SIOCGIFINFO_IN6 ioctl is encoded with in6_ondireq
                 * instead of in6_ndireq, so we treat it as such.
                 */
                lck_rw_lock_shared(nd_if_rwlock);
-               if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
+               ndi = ND_IFINFO(ifp);
+               if (!nd_ifinfo || i >= nd_ifinfo_indexlim ||
+                   !ndi->initialized) {
                        lck_rw_done(nd_if_rwlock);
                        error = EINVAL;
                        break;
                }
-               ondi->ndi.linkmtu = IN6_LINKMTU(ifp);
-               ondi->ndi.maxmtu = nd_ifinfo[i].maxmtu;
-               ondi->ndi.basereachable = nd_ifinfo[i].basereachable;
-               ondi->ndi.reachable = nd_ifinfo[i].reachable;
-               ondi->ndi.retrans = nd_ifinfo[i].retrans;
-               ondi->ndi.flags = nd_ifinfo[i].flags;
-               ondi->ndi.recalctm = nd_ifinfo[i].recalctm;
+               lck_mtx_lock(&ndi->lock);
+               linkmtu = IN6_LINKMTU(ifp);
+               bcopy(&linkmtu, &ondi->ndi.linkmtu, sizeof (linkmtu));
+               bcopy(&nd_ifinfo[i].maxmtu, &ondi->ndi.maxmtu,
+                   sizeof (u_int32_t));
+               bcopy(&nd_ifinfo[i].basereachable, &ondi->ndi.basereachable,
+                   sizeof (u_int32_t));
+               bcopy(&nd_ifinfo[i].reachable, &ondi->ndi.reachable,
+                   sizeof (u_int32_t));
+               bcopy(&nd_ifinfo[i].retrans, &ondi->ndi.retrans,
+                   sizeof (u_int32_t));
+               bcopy(&nd_ifinfo[i].flags, &ondi->ndi.flags,
+                   sizeof (u_int32_t));
+               bcopy(&nd_ifinfo[i].recalctm, &ondi->ndi.recalctm,
+                   sizeof (int));
                ondi->ndi.chlim = nd_ifinfo[i].chlim;
-               ondi->ndi.receivedra = nd_ifinfo[i].receivedra;
+               ondi->ndi.receivedra = 0;
+               lck_mtx_unlock(&ndi->lock);
                lck_rw_done(nd_if_rwlock);
                break;
+       }
+
+       case SIOCSIFINFO_FLAGS: {       /* struct in6_ndireq */
+               struct in6_ndireq *cndi = (struct in6_ndireq *)(void *)data;
+               u_int32_t oflags, flags;
+               struct nd_ifinfo *ndi;
 
-       case SIOCSIFINFO_FLAGS:
-               /* XXX: almost all other fields of ndi->ndi is unused */
+               /* XXX: almost all other fields of cndi->ndi is unused */
                lck_rw_lock_shared(nd_if_rwlock);
-               if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
+               ndi = ND_IFINFO(ifp);
+               if (!nd_ifinfo || i >= nd_ifinfo_indexlim ||
+                   !ndi->initialized) {
                        lck_rw_done(nd_if_rwlock);
                        error = EINVAL;
                        break;
                }
-               nd_ifinfo[i].flags = ndi->ndi.flags;
+               lck_mtx_lock(&ndi->lock);
+               oflags = nd_ifinfo[i].flags;
+               bcopy(&cndi->ndi.flags, &nd_ifinfo[i].flags, sizeof (flags));
+               flags = nd_ifinfo[i].flags;
+               lck_mtx_unlock(&ndi->lock);
                lck_rw_done(nd_if_rwlock);
+
+               if (oflags == flags)
+                       break;
+
+               error = nd6_setifinfo(ifp, oflags, flags);
                break;
+       }
 
-       case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
+       case SIOCSNDFLUSH_IN6:          /* struct in6_ifreq */
                /* flush default router list */
                /*
                 * xxx sumikawa: should not delete route if default
                 * route equals to the top of default router list
                 */
-               bzero(&any, sizeof(any));
                lck_mtx_lock(nd6_mutex);
-               defrouter_delreq(&any, 1);
-               defrouter_select();
+               defrouter_reset();
+               defrouter_select(ifp);
                lck_mtx_unlock(nd6_mutex);
                /* xxx sumikawa: flush prefix list */
                break;
 
-       case SIOCSPFXFLUSH_IN6: {
+       case SIOCSPFXFLUSH_IN6: {       /* struct in6_ifreq */
                /* flush all the prefix advertised by routers */
                struct nd_prefix *next;
-               lck_mtx_lock(nd6_mutex);
 
+               lck_mtx_lock(nd6_mutex);
                for (pr = nd_prefix.lh_first; pr; pr = next) {
-                       struct in6_ifaddr *ia, *ia_next;
+                       struct in6_ifaddr *ia;
 
                        next = pr->ndpr_next;
 
-                       if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
+                       NDPR_LOCK(pr);
+                       if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) {
+                               NDPR_UNLOCK(pr);
                                continue; /* XXX */
-
+                       }
+                       if (ifp != lo_ifp && pr->ndpr_ifp != ifp) {
+                               NDPR_UNLOCK(pr);
+                               continue;
+                       }
                        /* do we really have to remove addresses as well? */
-                       for (ia = in6_ifaddrs; ia; ia = ia_next) {
-                               /* ia might be removed.  keep the next ptr. */
-                               ia_next = ia->ia_next;
+                       NDPR_ADDREF_LOCKED(pr);
+                       NDPR_UNLOCK(pr);
+                       lck_rw_lock_exclusive(&in6_ifaddr_rwlock);
+                       ia = in6_ifaddrs;
+                       while (ia != NULL) {
+                               IFA_LOCK(&ia->ia_ifa);
+                               if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) {
+                                       IFA_UNLOCK(&ia->ia_ifa);
+                                       ia = ia->ia_next;
+                                       continue;
+                               }
 
-                               if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
+                               if (ia->ia6_ndpr == pr) {
+                                       IFA_ADDREF_LOCKED(&ia->ia_ifa);
+                                       IFA_UNLOCK(&ia->ia_ifa);
+                                       lck_rw_done(&in6_ifaddr_rwlock);
+                                       lck_mtx_unlock(nd6_mutex);
+                                       in6_purgeaddr(&ia->ia_ifa);
+                                       IFA_REMREF(&ia->ia_ifa);
+                                       lck_mtx_lock(nd6_mutex);
+                                       lck_rw_lock_exclusive(&in6_ifaddr_rwlock);
+                                       /*
+                                        * Purging the address caused
+                                        * in6_ifaddr_rwlock to be
+                                        * dropped and
+                                        * reacquired; therefore search again
+                                        * from the beginning of in6_ifaddrs.
+                                        * The same applies for the prefix list.
+                                        */
+                                       ia = in6_ifaddrs;
+                                       next = nd_prefix.lh_first;
                                        continue;
 
-                               if (ia->ia6_ndpr == pr)
-                                       in6_purgeaddr(&ia->ia_ifa, 1);
+                               }
+                               IFA_UNLOCK(&ia->ia_ifa);
+                               ia = ia->ia_next;
                        }
-                       prelist_remove(pr, 1);
+                       lck_rw_done(&in6_ifaddr_rwlock);
+                       NDPR_LOCK(pr);
+                       prelist_remove(pr);
+                       NDPR_UNLOCK(pr);
+                       /*
+                        * If we were trying to restart this loop
+                        * above by changing the value of 'next', we might
+                        * end up freeing the only element on the list
+                        * when we call NDPR_REMREF().
+                        * When this happens, we also have get out of this
+                        * loop because we have nothing else to do.
+                        */
+                       if (pr == next)
+                               next = NULL;
+                       NDPR_REMREF(pr);
                }
                lck_mtx_unlock(nd6_mutex);
                break;
        }
 
-       case SIOCSRTRFLUSH_IN6: {
+       case SIOCSRTRFLUSH_IN6: {       /* struct in6_ifreq */
                /* flush all the default routers */
                struct nd_defrouter *next;
 
@@ -2074,27 +2514,32 @@ nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
                         */
                        for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
                                next = TAILQ_NEXT(dr, dr_entry);
-                               defrtrlist_del(dr, 1);
+                               if (ifp == lo_ifp || dr->ifp == ifp)
+                                       defrtrlist_del(dr);
                        }
-                       defrtrlist_del(TAILQ_FIRST(&nd_defrouter), 1);
+                       if (ifp == lo_ifp ||
+                           TAILQ_FIRST(&nd_defrouter)->ifp == ifp)
+                               defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
                }
                lck_mtx_unlock(nd6_mutex);
                break;
        }
 
-       case SIOCGNBRINFO_IN6_32: {
+       case SIOCGNBRINFO_IN6_32: {     /* struct in6_nbrinfo_32 */
                struct llinfo_nd6 *ln;
-               struct in6_nbrinfo_32 *nbi_32 = (struct in6_nbrinfo_32 *)data;
-               /* make local for safety */
-               struct in6_addr nb_addr = nbi_32->addr;
+               struct in6_nbrinfo_32 nbi_32;
+               struct in6_addr nb_addr; /* make local for safety */
 
+               bcopy(data, &nbi_32, sizeof (nbi_32));
+               nb_addr = nbi_32.addr;
                /*
                 * XXX: KAME specific hack for scoped addresses
                 *      XXXX: for other scopes than link-local?
                 */
-               if (IN6_IS_ADDR_LINKLOCAL(&nbi_32->addr) ||
-                   IN6_IS_ADDR_MC_LINKLOCAL(&nbi_32->addr)) {
-                       u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
+               if (IN6_IS_ADDR_LINKLOCAL(&nbi_32.addr) ||
+                   IN6_IS_ADDR_MC_LINKLOCAL(&nbi_32.addr)) {
+                       u_int16_t *idp =
+                           (u_int16_t *)(void *)&nb_addr.s6_addr[2];
 
                        if (*idp == 0)
                                *idp = htons(ifp->if_index);
@@ -2107,28 +2552,31 @@ nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
                }
                RT_LOCK_ASSERT_HELD(rt);
                ln = rt->rt_llinfo;
-               nbi_32->state = ln->ln_state;
-               nbi_32->asked = ln->ln_asked;
-               nbi_32->isrouter = ln->ln_router;
-               nbi_32->expire = ln->ln_expire;
+               nbi_32.state = ln->ln_state;
+               nbi_32.asked = ln->ln_asked;
+               nbi_32.isrouter = ln->ln_router;
+               nbi_32.expire = ln->ln_expire;
                RT_REMREF_LOCKED(rt);
                RT_UNLOCK(rt);
+               bcopy(&nbi_32, data, sizeof (nbi_32));
                break;
        }
 
-       case SIOCGNBRINFO_IN6_64: {
+       case SIOCGNBRINFO_IN6_64: {     /* struct in6_nbrinfo_64 */
                struct llinfo_nd6 *ln;
-               struct in6_nbrinfo_64 *nbi_64 = (struct in6_nbrinfo_64 *)data;
-               /* make local for safety */
-               struct in6_addr nb_addr = nbi_64->addr;
+               struct in6_nbrinfo_64 nbi_64;
+               struct in6_addr nb_addr; /* make local for safety */
 
+               bcopy(data, &nbi_64, sizeof (nbi_64));
+               nb_addr = nbi_64.addr;
                /*
                 * XXX: KAME specific hack for scoped addresses
                 *      XXXX: for other scopes than link-local?
                 */
-               if (IN6_IS_ADDR_LINKLOCAL(&nbi_64->addr) ||
-                   IN6_IS_ADDR_MC_LINKLOCAL(&nbi_64->addr)) {
-                       u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
+               if (IN6_IS_ADDR_LINKLOCAL(&nbi_64.addr) ||
+                   IN6_IS_ADDR_MC_LINKLOCAL(&nbi_64.addr)) {
+                       u_int16_t *idp =
+                           (u_int16_t *)(void *)&nb_addr.s6_addr[2];
 
                        if (*idp == 0)
                                *idp = htons(ifp->if_index);
@@ -2141,34 +2589,51 @@ nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
                }
                RT_LOCK_ASSERT_HELD(rt);
                ln = rt->rt_llinfo;
-               nbi_64->state = ln->ln_state;
-               nbi_64->asked = ln->ln_asked;
-               nbi_64->isrouter = ln->ln_router;
-               nbi_64->expire = ln->ln_expire;
+               nbi_64.state = ln->ln_state;
+               nbi_64.asked = ln->ln_asked;
+               nbi_64.isrouter = ln->ln_router;
+               nbi_64.expire = ln->ln_expire;
                RT_REMREF_LOCKED(rt);
                RT_UNLOCK(rt);
+               bcopy(&nbi_64, data, sizeof (nbi_64));
                break;
        }
 
-       case SIOCGDEFIFACE_IN6_32: /* XXX: should be implemented as a sysctl? */
-       case SIOCGDEFIFACE_IN6_64: {
-               struct in6_ndifreq_64 *ndif_64 = (struct in6_ndifreq_64 *)data;
-               struct in6_ndifreq_32 *ndif_32 = (struct in6_ndifreq_32 *)data;
+       case SIOCGDEFIFACE_IN6_32:      /* struct in6_ndifreq_32 */
+       case SIOCGDEFIFACE_IN6_64: {    /* struct in6_ndifreq_64 */
+               struct in6_ndifreq_64 *ndif_64 =
+                   (struct in6_ndifreq_64 *)(void *)data;
+               struct in6_ndifreq_32 *ndif_32 =
+                   (struct in6_ndifreq_32 *)(void *)data;
 
-               if (cmd == SIOCGDEFIFACE_IN6_64)
-                       ndif_64->ifindex = nd6_defifindex;
-               else
-                       ndif_32->ifindex = nd6_defifindex;
+               if (cmd == SIOCGDEFIFACE_IN6_64) {
+                       u_int64_t j = nd6_defifindex;
+                       bcopy(&j, &ndif_64->ifindex, sizeof (j));
+               } else {
+                       bcopy(&nd6_defifindex, &ndif_32->ifindex,
+                           sizeof (u_int32_t));
+               }
                break;
        }
 
-       case SIOCSDEFIFACE_IN6_32: /* XXX: should be implemented as a sysctl? */
-       case SIOCSDEFIFACE_IN6_64: {
-               struct in6_ndifreq_64 *ndif_64 = (struct in6_ndifreq_64 *)data;
-               struct in6_ndifreq_32 *ndif_32 = (struct in6_ndifreq_32 *)data;
+       case SIOCSDEFIFACE_IN6_32:      /* struct in6_ndifreq_32 */
+       case SIOCSDEFIFACE_IN6_64: {    /* struct in6_ndifreq_64 */
+               struct in6_ndifreq_64 *ndif_64 =
+                   (struct in6_ndifreq_64 *)(void *)data;
+               struct in6_ndifreq_32 *ndif_32 =
+                   (struct in6_ndifreq_32 *)(void *)data;
+               u_int32_t idx;
+
+               if (cmd == SIOCSDEFIFACE_IN6_64) {
+                       u_int64_t j;
+                       bcopy(&ndif_64->ifindex, &j, sizeof (j));
+                       idx = (u_int32_t)j;
+               } else {
+                       bcopy(&ndif_32->ifindex, &idx, sizeof (idx));
+               }
 
-               return (nd6_setdefaultiface(cmd == SIOCSDEFIFACE_IN6_64 ?
-                   ndif_64->ifindex : ndif_32->ifindex));
+               error = nd6_setdefaultiface(idx);
+               return (error);
                /* NOTREACHED */
        }
        }
@@ -2198,9 +2663,9 @@ nd6_cache_lladdr(
        int newstate = 0;
        struct timeval timenow;
 
-       if (!ifp)
+       if (ifp == NULL)
                panic("ifp == NULL in nd6_cache_lladdr");
-       if (!from)
+       if (from == NULL)
                panic("from == NULL in nd6_cache_lladdr");
 
        /* nothing must be updated for unspecified address */
@@ -2220,12 +2685,6 @@ nd6_cache_lladdr(
 
        rt = nd6_lookup(from, 0, ifp, 0);
        if (rt == NULL) {
-#if 0
-               /* nothing must be done if there's no lladdr */
-               if (!lladdr || !lladdrlen)
-                       return;
-#endif
-
                if ((rt = nd6_lookup(from, 1, ifp, 0)) == NULL)
                        return;
                RT_LOCK_ASSERT_HELD(rt);
@@ -2241,6 +2700,8 @@ nd6_cache_lladdr(
                is_newentry = 0;
        }
 
+       if (rt == NULL)
+               return;
        if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
 fail:
                RT_UNLOCK(rt);
@@ -2248,10 +2709,10 @@ fail:
                rtfree(rt);
                return;
        }
-       ln = rt->rt_llinfo;
-       if (!ln)
+       ln = (struct llinfo_nd6 *)rt->rt_llinfo;
+       if (ln == NULL)
                goto fail;
-       if (!rt->rt_gateway)
+       if (rt->rt_gateway == NULL)
                goto fail;
        if (rt->rt_gateway->sa_family != AF_LINK)
                goto fail;
@@ -2284,18 +2745,21 @@ fail:
                 */
                sdl->sdl_alen = ifp->if_addrlen;
                bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
+
+               /* cache the gateway (sender HW) address */
+               nd6_llreach_alloc(rt, ifp, LLADDR(sdl), sdl->sdl_alen, FALSE);
        }
 
        if (!is_newentry) {
-               if ((!olladdr && lladdr)                /* (3) */
-                || (olladdr && lladdr && llchange)) {  /* (5) */
+               if ((!olladdr && lladdr != NULL) ||     /* (3) */
+                   (olladdr && lladdr != NULL && llchange)) {  /* (5) */
                        do_update = 1;
                        newstate = ND6_LLINFO_STALE;
                } else                                  /* (1-2,4) */
                        do_update = 0;
        } else {
                do_update = 1;
-               if (!lladdr)                            /* (6) */
+               if (lladdr == NULL)                     /* (6) */
                        newstate = ND6_LLINFO_NOSTATE;
                else                                    /* (7) */
                        newstate = ND6_LLINFO_STALE;
@@ -2319,13 +2783,15 @@ fail:
                        ln->ln_hold = NULL;
 
                        if (m != NULL) {
+                               struct sockaddr_in6 sin6;
+
+                               rtkey_to_sa6(rt, &sin6);
                                /*
                                 * we assume ifp is not a p2p here, so just
                                 * set the 2nd argument as the 1st one.
                                 */
                                RT_UNLOCK(rt);
-                               nd6_output(ifp, ifp, m,
-                                   (struct sockaddr_in6 *)rt_key(rt), rt, 0);
+                               nd6_output(ifp, ifp, m, &sin6, rt, NULL);
                                RT_LOCK(rt);
                        }
                } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
@@ -2358,7 +2824,7 @@ fail:
         *      0       n       y       --      (3)     c   s     s
         *      0       y       y       n       (4)     c   s     s
         *      0       y       y       y       (5)     c   s     s
-        *      1       --      n       --      (6) c   c       c s
+        *      1       --      n       --      (6) c   c       c s
         *      1       --      y       --      (7) c   c   s   c s
         *
         *                                      (c=clear s=set)
@@ -2374,8 +2840,8 @@ fail:
        case ND_REDIRECT:
                /*
                 * If the icmp is a redirect to a better router, always set the
-                * is_router flag. Otherwise, if the entry is newly created,
-                * clear the flag. [RFC 2461, sec 8.3]
+                * is_router flag.  Otherwise, if the entry is newly created,
+                * clear the flag.  [RFC 2461, sec 8.3]
                 */
                if (code == ND_REDIRECT_ROUTER)
                        ln->ln_router = 1;
@@ -2392,8 +2858,8 @@ fail:
                /*
                 * Mark an entry with lladdr as a router.
                 */
-               if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
-                || (is_newentry && lladdr)) {                  /* (7) */
+               if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
+                   (is_newentry && lladdr)) {                  /* (7) */
                        ln->ln_router = 1;
                }
                break;
@@ -2405,21 +2871,16 @@ fail:
         * created, it might affect the selection policy.
         * Question: can we restrict the first condition to the "is_newentry"
         * case?
-        * XXX: when we hear an RA from a new router with the link-layer
-        * address option, defrouter_select() is called twice, since
-        * defrtrlist_update called the function as well.  However, I believe
-        * we can compromise the overhead, since it only happens the first
-        * time.
-        * XXX: although defrouter_select() should not have a bad effect
-        * for those are not autoconfigured hosts, we explicitly avoid such
-        * cases for safety.
+        *
+        * Note: Perform default router selection even when we are a router,
+        * if Scoped Routing is enabled.
         */
-       if (do_update && ln->ln_router && !ip6_forwarding &&
-           (ip6_accept_rtadv || (ifp->if_eflags & IFEF_ACCEPT_RTADVD))) {
+       if (do_update && ln->ln_router &&
+           (ip6_doscopedroute || !ip6_forwarding)) {
                RT_REMREF_LOCKED(rt);
                RT_UNLOCK(rt);
                lck_mtx_lock(nd6_mutex);
-               defrouter_select();
+               defrouter_select(ifp);
                lck_mtx_unlock(nd6_mutex);
        } else {
                RT_REMREF_LOCKED(rt);
@@ -2439,6 +2900,9 @@ nd6_slowtimo(
                if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
                        break;
                nd6if = &nd_ifinfo[i];
+               if (!nd6if->initialized)
+                       break;
+               lck_mtx_lock(&nd6if->lock);
                if (nd6if->basereachable && /* already initialized */
                    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
                        /*
@@ -2450,6 +2914,7 @@ nd6_slowtimo(
                        nd6if->recalctm = nd6_recalc_reachtm_interval;
                        nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
                }
+               lck_mtx_unlock(&nd6if->lock);
        }
        lck_rw_done(nd_if_rwlock);
        timeout(nd6_slowtimo, (caddr_t)0, ND6_SLOWTIMER_INTERVAL * hz);
@@ -2458,7 +2923,7 @@ nd6_slowtimo(
 #define senderr(e) { error = (e); goto bad;}
 int
 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
-    struct sockaddr_in6 *dst, struct rtentry *hint0, int locked)
+    struct sockaddr_in6 *dst, struct rtentry *hint0, struct flowadv *adv)
 {
        struct mbuf *m = m0;
        struct rtentry *rt = hint0, *hint = hint0;
@@ -2466,6 +2931,7 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
        int error = 0;
        struct timeval timenow;
        struct rtentry *rtrele = NULL;
+       struct nd_ifinfo *ndi;
 
        if (rt != NULL) {
                RT_LOCK_SPIN(rt);
@@ -2491,7 +2957,7 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
         * route with a reference held for that placeholder.
         *
         * This logic is similar to, though not exactly the same as the one
-        * used by arp_route_to_gateway_route().
+        * used by route_to_gwroute().
         */
        if (rt != NULL) {
                /*
@@ -2503,14 +2969,14 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
                if (!(rt->rt_flags & RTF_UP)) {
                        RT_REMREF_LOCKED(rt);
                        RT_UNLOCK(rt);
-                       if ((hint = rt = rtalloc1((struct sockaddr *)dst,
-                           1, 0)) != NULL) {
+                       if ((hint = rt = rtalloc1_scoped((struct sockaddr *)dst,
+                           1, 0, ifp->if_index)) != NULL) {
                                RT_LOCK_SPIN(rt);
                                if (rt->rt_ifp != ifp) {
                                        /* XXX: loop care? */
                                        RT_UNLOCK(rt);
                                        error = nd6_output(ifp, origifp, m0,
-                                           dst, rt, locked);
+                                           dst, rt, adv);
                                        rtfree(rt);
                                        return (error);
                                }
@@ -2524,7 +2990,7 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
                        struct in6_ifaddr *ia6 = NULL;
                        struct sockaddr_in6 gw6;
 
-                       gw6 = *((struct sockaddr_in6 *)rt->rt_gateway);
+                       rtgw_to_sa6(rt, &gw6);
                        /*
                         * Must drop rt_lock since nd6_is_addr_neighbor()
                         * calls nd6_lookup() and acquires rnh_lock.
@@ -2547,14 +3013,14 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
                                 * XXX: we may need a more generic rule here.
                                 */
                                if (ia6 != NULL)
-                                       ifafree(&ia6->ia_ifa);
+                                       IFA_REMREF(&ia6->ia_ifa);
                                if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
                                        senderr(EHOSTUNREACH);
                                goto sendpkt;
                        }
 
                        RT_LOCK_SPIN(rt);
-                       gw6 = *((struct sockaddr_in6 *)rt->rt_gateway);
+                       gw6 = *((struct sockaddr_in6 *)(void *)rt->rt_gateway);
 
                        /* If hint is now down, give up */
                        if (!(rt->rt_flags & RTF_UP)) {
@@ -2577,14 +3043,15 @@ nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
                         */
                        RT_LOCK_SPIN(gwrt);
                        if (!(gwrt->rt_flags & RTF_UP)) {
-                               struct rtentry *ogwrt;
-
                                rt->rt_gwroute = NULL;
                                RT_UNLOCK(gwrt);
                                RT_UNLOCK(rt);
                                rtfree(gwrt);
 lookup:
-                               gwrt = rtalloc1((struct sockaddr *)&gw6, 1, 0);
+                               lck_mtx_lock(rnh_lock);
+                               gwrt = rtalloc1_scoped_locked(
+                                   (struct sockaddr *)&gw6, 1, 0,
+                                   ifp->if_index);
 
                                RT_LOCK(rt);
                                /*
@@ -2601,57 +3068,68 @@ lookup:
                                        }
                                        RT_UNLOCK(rt);
                                        if (gwrt != NULL)
-                                               rtfree(gwrt);
+                                               rtfree_locked(gwrt);
+                                       lck_mtx_unlock(rnh_lock);
                                        senderr(EHOSTUNREACH);
                                }
-
-                               /* Remove any existing gwrt */
-                               ogwrt = rt->rt_gwroute;
-                               if ((rt->rt_gwroute = gwrt) != NULL)
-                                       RT_ADDREF(gwrt);
-
+                               VERIFY(gwrt != NULL);
+                               /*
+                                * Set gateway route; callee adds ref to gwrt;
+                                * gwrt has an extra ref from rtalloc1() for
+                                * this routine.
+                                */
+                               rt_set_gwroute(rt, rt_key(rt), gwrt);
                                RT_UNLOCK(rt);
-                               /* Now free the replaced gwrt */
-                               if (ogwrt != NULL)
-                                       rtfree(ogwrt);
-                               /* If still no route to gateway, bail out */
-                               if (gwrt == NULL)
-                                       senderr(EHOSTUNREACH);
+                               lck_mtx_unlock(rnh_lock);
                                /* Remember to release/free "rt" at the end */
                                rtrele = rt;
                                rt = gwrt;
-                               RT_LOCK_SPIN(rt);
-                               /* If gwrt is now down, give up */
-                               if (!(rt->rt_flags & RTF_UP)) {
-                                       RT_UNLOCK(rt);
-                                       rtfree(rt);
-                                       rt = NULL;
-                                       /* "rtrele" == original "rt" */
-                                       senderr(EHOSTUNREACH);
-                               }
                        } else {
                                RT_ADDREF_LOCKED(gwrt);
                                RT_UNLOCK(gwrt);
                                RT_UNLOCK(rt);
-                               RT_LOCK_SPIN(gwrt);
-                               /* If gwrt is now down, give up */
-                               if (!(gwrt->rt_flags & RTF_UP)) {
-                                       RT_UNLOCK(gwrt);
-                                       rtfree(gwrt);
-                                       senderr(EHOSTUNREACH);
-                               }
                                /* Remember to release/free "rt" at the end */
                                rtrele = rt;
                                rt = gwrt;
                        }
+                       VERIFY(rt == gwrt);
+
+                       /*
+                        * This is an opportunity to revalidate the parent
+                        * route's gwroute, in case it now points to a dead
+                        * route entry.  Parent route won't go away since the
+                        * clone (hint) holds a reference to it.  rt == gwrt.
+                        */
+                       RT_LOCK_SPIN(hint);
+                       if ((hint->rt_flags & (RTF_WASCLONED | RTF_UP)) ==
+                           (RTF_WASCLONED | RTF_UP)) {
+                               struct rtentry *prt = hint->rt_parent;
+                               VERIFY(prt != NULL);
+
+                               RT_CONVERT_LOCK(hint);
+                               RT_ADDREF(prt);
+                               RT_UNLOCK(hint);
+                               rt_revalidate_gwroute(prt, rt);
+                               RT_REMREF(prt);
+                       } else {
+                               RT_UNLOCK(hint);
+                       }
+
+                       RT_LOCK_SPIN(rt);
+                       /* rt == gwrt; if it is now down, give up */
+                       if (!(rt->rt_flags & RTF_UP)) {
+                               RT_UNLOCK(rt);
+                               rtfree(rt);
+                               rt = NULL;
+                               /* "rtrele" == original "rt" */
+                               senderr(EHOSTUNREACH);
+                       }
                }
+
                /* Become a regular mutex */
                RT_CONVERT_LOCK(rt);
        }
 
-       if (rt != NULL)
-               RT_LOCK_ASSERT_HELD(rt);
-
        /*
         * Address resolution or Neighbor Unreachability Detection
         * for the next hop.
@@ -2663,6 +3141,12 @@ lookup:
        if (rt && (rt->rt_flags & RTF_LLINFO) != 0) {
                ln = rt->rt_llinfo;
        } else {
+               struct sockaddr_in6 sin6;
+               /*
+                * Clear out Scope ID field in case it is set.
+                */
+               sin6 = *dst;
+               sin6.sin6_scope_id = 0;
                /*
                 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
                 * the condition below is not very efficient.  But we believe
@@ -2672,7 +3156,7 @@ lookup:
                 */
                if (rt != NULL)
                        RT_UNLOCK(rt);
-               if (nd6_is_addr_neighbor(dst, ifp, 0)) {
+               if (nd6_is_addr_neighbor(&sin6, ifp, 0)) {
                        /* "rtrele" may have been used, so clean up "rt" now */
                        if (rt != NULL) {
                                /* Don't free "hint0" */
@@ -2696,8 +3180,12 @@ lookup:
                if (rt != NULL)
                        RT_UNLOCK(rt);
                lck_rw_lock_shared(nd_if_rwlock);
+               ndi = ND_IFINFO(ifp);
+               VERIFY(ndi != NULL && ndi->initialized);
+               lck_mtx_lock(&ndi->lock);
                if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
-                   !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
+                   !(ndi->flags & ND6_IFF_PERFORMNUD)) {
+                       lck_mtx_unlock(&ndi->lock);
                        lck_rw_done(nd_if_rwlock);
                        log(LOG_DEBUG,
                            "nd6_output: can't allocate llinfo for %s "
@@ -2705,6 +3193,7 @@ lookup:
                            ip6_sprintf(&dst->sin6_addr), ln, rt);
                        senderr(EIO);   /* XXX: good error? */
                }
+               lck_mtx_unlock(&ndi->lock);
                lck_rw_done(nd_if_rwlock);
 
                goto sendpkt;   /* send anyway */
@@ -2716,7 +3205,7 @@ lookup:
        if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
            ln->ln_state < ND6_LLINFO_REACHABLE) {
                ln->ln_state = ND6_LLINFO_STALE;
-               ln->ln_expire = timenow.tv_sec + nd6_gctimer;
+               ln->ln_expire = rt_expiry(rt, timenow.tv_sec, nd6_gctimer);
        }
 
        /*
@@ -2729,7 +3218,7 @@ lookup:
        if (ln->ln_state == ND6_LLINFO_STALE) {
                ln->ln_asked = 0;
                ln->ln_state = ND6_LLINFO_DELAY;
-               ln->ln_expire = timenow.tv_sec + nd6_delay;
+               ln->ln_expire = rt_expiry(rt, timenow.tv_sec, nd6_delay);
        }
 
        /*
@@ -2773,12 +3262,18 @@ lookup:
            ln->ln_expire < timenow.tv_sec) {
                ln->ln_asked++;
                lck_rw_lock_shared(nd_if_rwlock);
-               ln->ln_expire = timenow.tv_sec +
-                       nd_ifinfo[ifp->if_index].retrans / 1000;
+               ndi = ND_IFINFO(ifp);
+               VERIFY(ndi != NULL && ndi->initialized);
+               lck_mtx_lock(&ndi->lock);
+               ln->ln_expire = timenow.tv_sec + ndi->retrans / 1000;
+               lck_mtx_unlock(&ndi->lock);
                lck_rw_done(nd_if_rwlock);
                RT_UNLOCK(rt);
                /* We still have a reference on rt (for ln) */
-               nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0, locked);
+               if (ip6_forwarding)
+                       nd6_prproxy_ns_output(ifp, NULL, &dst->sin6_addr, ln);
+               else
+                       nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
        } else {
                RT_UNLOCK(rt);
        }
@@ -2811,19 +3306,23 @@ sendpkt:
        if (rt != NULL)
                RT_LOCK_ASSERT_NOTHELD(rt);
 
-       /* Clean up HW checksum flags before sending the packet */
-       m->m_pkthdr.csum_data = 0;
-       m->m_pkthdr.csum_flags = 0;
+       /* discard the packet if IPv6 operation is disabled on the interface */
+       lck_rw_lock_shared(nd_if_rwlock);
+       ndi = ND_IFINFO(ifp);
+       VERIFY(ndi != NULL && ndi->initialized);
+       /* test is done here without holding ndi lock, for performance */
+       if (ndi->flags & ND6_IFF_IFDISABLED) {
+               lck_rw_done(nd_if_rwlock);
+               error = ENETDOWN; /* better error? */
+               goto bad;
+       }
+       lck_rw_done(nd_if_rwlock);
 
        if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
                /* forwarding rules require the original scope_id */
                m->m_pkthdr.rcvif = origifp;
-               if (locked)
-                       lck_mtx_unlock(ip6_mutex);
                error = dlil_output(origifp, PF_INET6, m, (caddr_t)rt,
-                   (struct sockaddr *)dst, 0);
-               if (locked)
-                       lck_mtx_lock(ip6_mutex);
+                   (struct sockaddr *)dst, 0, adv);
                goto release;
        } else {
                /* Do not allow loopback address to wind up on a wire */
@@ -2845,13 +3344,20 @@ sendpkt:
                }
        }
 
+       if (rt != NULL) {
+               RT_LOCK_SPIN(rt);
+               /* Mark use timestamp */
+               if (rt->rt_llinfo != NULL)
+                       nd6_llreach_use(rt->rt_llinfo);
+               RT_UNLOCK(rt);
+       }
+
+       if (hint && nstat_collect)
+               nstat_route_tx(hint, 1, m->m_pkthdr.len, 0);
+
        m->m_pkthdr.rcvif = NULL;
-       if (locked)
-               lck_mtx_unlock(ip6_mutex);
        error = dlil_output(ifp, PF_INET6, m, (caddr_t)rt,
-           (struct sockaddr *)dst, 0);
-       if (locked)
-               lck_mtx_lock(ip6_mutex);
+           (struct sockaddr *)dst, 0, adv);
        goto release;
 
 bad:
@@ -2906,8 +3412,13 @@ nd6_need_cache(
 #if IFT_IEEE80211
        case IFT_IEEE80211:
 #endif
-       case IFT_BRIDGE:
        case IFT_GIF:           /* XXX need more cases? */
+       case IFT_PPP:
+#if IFT_TUNNEL
+       case IFT_TUNNEL:
+#endif
+       case IFT_BRIDGE:
+       case IFT_CELLULAR:
                return(1);
        default:
                return(0);
@@ -3000,8 +3511,8 @@ nd6_lookup_ipv6(ifnet_t    ifp, const struct sockaddr_in6 *ip6_dest,
                 * Callee holds a reference on the route and returns
                 * with the route entry locked, upon success.
                 */
-               result = arp_route_to_gateway_route(
-                   (const struct sockaddr*)ip6_dest, hint, &route);
+               result = route_to_gwroute((const struct sockaddr *)ip6_dest,
+                   hint, &route);
                if (result != 0)
                        return (result);
                if (route != NULL)
@@ -3022,7 +3533,7 @@ nd6_lookup_ipv6(ifnet_t    ifp, const struct sockaddr_in6 *ip6_dest,
        if (route == NULL) {
                /*
                 * This could happen, if we could not allocate memory or
-                * if arp_route_to_gateway_route() didn't return a route.
+                * if route_to_gwroute() didn't return a route.
                 */
                result = ENOBUFS;
                goto release;
@@ -3058,6 +3569,21 @@ release:
        return (result);
 }
 
+int
+nd6_setifinfo(struct ifnet *ifp, u_int32_t before, u_int32_t after)
+{
+       /*
+        * We only care about ND6_IFF_PROXY_PREFIXES for now.
+        */
+       before &= ND6_IFF_PROXY_PREFIXES;
+       after &= ND6_IFF_PROXY_PREFIXES;
+
+       if (before == after)
+               return (0);
+
+       return (nd6_if_prproxy(ifp, ((int32_t)(after - before) > 0)));
+}
+
 SYSCTL_DECL(_net_inet6_icmp6);
 
 static int
@@ -3079,8 +3605,9 @@ nd6_sysctl_drlist SYSCTL_HANDLER_ARGS
                for (dr = TAILQ_FIRST(&nd_defrouter);
                     dr;
                     dr = TAILQ_NEXT(dr, dr_entry)) {
-                       d = (struct in6_defrouter_64 *)buf;
-                       de = (struct in6_defrouter_64 *)(buf + sizeof (buf));
+                       d = (struct in6_defrouter_64 *)(void *)buf;
+                       de = (struct in6_defrouter_64 *)
+                           (void *)(buf + sizeof (buf));
 
                        if (d + 1 <= de) {
                                bzero(d, sizeof (*d));
@@ -3093,6 +3620,8 @@ nd6_sysctl_drlist SYSCTL_HANDLER_ARGS
                                            "default router list (%s)\n",
                                            ip6_sprintf(&dr->rtaddr));
                                d->flags = dr->flags;
+                               d->stateflags = dr->stateflags;
+                               d->stateflags &= ~NDDRF_PROCESSED;
                                d->rtlifetime = dr->rtlifetime;
                                d->expire = dr->expire;
                                d->if_index = dr->ifp->if_index;
@@ -3109,8 +3638,9 @@ nd6_sysctl_drlist SYSCTL_HANDLER_ARGS
                for (dr = TAILQ_FIRST(&nd_defrouter);
                     dr;
                     dr = TAILQ_NEXT(dr, dr_entry)) {
-                       d_32 = (struct in6_defrouter_32 *)buf;
-                       de_32 = (struct in6_defrouter_32 *)(buf + sizeof (buf));
+                       d_32 = (struct in6_defrouter_32 *)(void *)buf;
+                       de_32 = (struct in6_defrouter_32 *)
+                           (void *)(buf + sizeof (buf));
 
                        if (d_32 + 1 <= de_32) {
                                bzero(d_32, sizeof (*d_32));
@@ -3123,6 +3653,8 @@ nd6_sysctl_drlist SYSCTL_HANDLER_ARGS
                                            "default router list (%s)\n",
                                            ip6_sprintf(&dr->rtaddr));
                                d_32->flags = dr->flags;
+                               d_32->stateflags = dr->stateflags;
+                               d_32->stateflags &= ~NDDRF_PROCESSED;
                                d_32->rtlifetime = dr->rtlifetime;
                                d_32->expire = dr->expire;
                                d_32->if_index = dr->ifp->if_index;
@@ -3160,13 +3692,15 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
                        struct sockaddr_in6 *sin6, *s6;
                        struct nd_pfxrouter *pfr;
 
-                       p = (struct in6_prefix_64 *)buf;
-                       pe = (struct in6_prefix_64 *)(buf + sizeof (buf));
+                       p = (struct in6_prefix_64 *)(void *)buf;
+                       pe = (struct in6_prefix_64 *)
+                           (void *)(buf + sizeof (buf));
 
                        if (p + 1 <= pe) {
                                bzero(p, sizeof (*p));
                                sin6 = (struct sockaddr_in6 *)(p + 1);
 
+                               NDPR_LOCK(pr);
                                p->prefix = pr->ndpr_prefix;
                                if (in6_recoverscope(&p->prefix,
                                    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
@@ -3179,7 +3713,7 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
                                p->pltime = pr->ndpr_pltime;
                                p->if_index = pr->ndpr_ifp->if_index;
                                p->expire = pr->ndpr_expire;
-                               p->refcnt = pr->ndpr_refcnt;
+                               p->refcnt = pr->ndpr_addrcnt;
                                p->flags = pr->ndpr_stateflags;
                                p->origin = PR_ORIG_RA;
                                advrtrs = 0;
@@ -3205,6 +3739,7 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
                                        advrtrs++;
                                }
                                p->advrtrs = advrtrs;
+                               NDPR_UNLOCK(pr);
                        } else {
                                panic("buffer too short");
                        }
@@ -3222,13 +3757,15 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
                        struct sockaddr_in6 *sin6, *s6;
                        struct nd_pfxrouter *pfr;
 
-                       p_32 = (struct in6_prefix_32 *)buf;
-                       pe_32 = (struct in6_prefix_32 *)(buf + sizeof (buf));
+                       p_32 = (struct in6_prefix_32 *)(void *)buf;
+                       pe_32 = (struct in6_prefix_32 *)
+                           (void *)(buf + sizeof (buf));
 
                        if (p_32 + 1 <= pe_32) {
                                bzero(p_32, sizeof (*p_32));
                                sin6 = (struct sockaddr_in6 *)(p_32 + 1);
 
+                               NDPR_LOCK(pr);
                                p_32->prefix = pr->ndpr_prefix;
                                if (in6_recoverscope(&p_32->prefix,
                                    &p_32->prefix.sin6_addr, pr->ndpr_ifp) != 0)
@@ -3241,7 +3778,7 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
                                p_32->pltime = pr->ndpr_pltime;
                                p_32->if_index = pr->ndpr_ifp->if_index;
                                p_32->expire = pr->ndpr_expire;
-                               p_32->refcnt = pr->ndpr_refcnt;
+                               p_32->refcnt = pr->ndpr_addrcnt;
                                p_32->flags = pr->ndpr_stateflags;
                                p_32->origin = PR_ORIG_RA;
                                advrtrs = 0;
@@ -3267,6 +3804,7 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
                                        advrtrs++;
                                }
                                p_32->advrtrs = advrtrs;
+                               NDPR_UNLOCK(pr);
                        } else {
                                panic("buffer too short");
                        }
@@ -3280,7 +3818,7 @@ nd6_sysctl_prlist SYSCTL_HANDLER_ARGS
        return (error);
 }
 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
-       CTLFLAG_RD, 0, 0, nd6_sysctl_drlist, "S,in6_defrouter","");
+       CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, nd6_sysctl_drlist, "S,in6_defrouter","");
 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
-       CTLFLAG_RD, 0, 0, nd6_sysctl_prlist, "S,in6_defrouter","");
+       CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, nd6_sysctl_prlist, "S,in6_defrouter","");