]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/net/if.c
xnu-1699.32.7.tar.gz
[apple/xnu.git] / bsd / net / if.c
index 1f10ed96f61bf5cfe87b263e32166bf35bc58cb1..595fcaea994c6467ab3ad9585a3a9b411135ba9c 100644 (file)
@@ -1,16 +1,19 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
@@ -20,7 +23,7 @@
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * Copyright (c) 1980, 1986, 1993
  *     @(#)if.c        8.3 (Berkeley) 1/4/94
  * $FreeBSD: src/sys/net/if.c,v 1.85.2.9 2001/07/24 19:10:17 brooks Exp $
  */
+/*
+ * NOTICE: This file was modified by SPARTA, Inc. in 2006 to introduce
+ * support for mandatory and extensible security protections.  This notice
+ * is included in support of clause 2.2 (b) of the Apple Public License,
+ * Version 2.0.
+ */
+
+#include <kern/locks.h>
 
 #include <sys/param.h>
 #include <sys/malloc.h>
 #include <sys/sockio.h>
 #include <sys/syslog.h>
 #include <sys/sysctl.h>
+#include <sys/mcache.h>
+#include <kern/zalloc.h>
+
+#include <machine/endian.h>
+
+#include <pexpert/pexpert.h>
 
 #include <net/if.h>
 #include <net/if_arp.h>
 #include <net/if_dl.h>
 #include <net/if_types.h>
 #include <net/if_var.h>
+#include <net/net_osdep.h>
+#include <net/ethernet.h>
+
 #include <net/radix.h>
 #include <net/route.h>
 #ifdef __APPLE__
 #include <net/dlil.h>
 //#include <string.h>
 #include <sys/domain.h>
+#include <libkern/OSAtomic.h>
 #endif
 
-#if defined(INET) || defined(INET6)
+#if INET || INET6
 /*XXX*/
 #include <netinet/in.h>
 #include <netinet/in_var.h>
+#include <netinet/ip_var.h>
+#include <netinet/ip6.h>
 #if INET6
 #include <netinet6/in6_var.h>
 #include <netinet6/in6_ifattach.h>
+#include <netinet6/ip6_var.h>
 #endif
 #endif
 
+#if CONFIG_MACF_NET 
+#include <security/mac_framework.h>
+#endif
+
+
 /*
  * System initialization
  */
 
-static int ifconf __P((u_long, caddr_t));
-static void if_qflush __P((struct ifqueue *));
-static void link_rtrequest __P((int, struct rtentry *, struct sockaddr *));
+/* Lock group and attribute for ifaddr lock */
+lck_attr_t     *ifa_mtx_attr;
+lck_grp_t      *ifa_mtx_grp;
+static lck_grp_attr_t  *ifa_mtx_grp_attr;
+
+static int ifconf(u_long cmd, user_addr_t ifrp, int * ret_space);
+static void if_qflush(struct ifqueue *);
+__private_extern__ void link_rtrequest(int, struct rtentry *, struct sockaddr *);
+void if_rtproto_del(struct ifnet *ifp, int protocol);
+
+static int if_addmulti_common(struct ifnet *, const struct sockaddr *,
+    struct ifmultiaddr **, int);
+static int if_delmulti_common(struct ifmultiaddr *, struct ifnet *,
+    const struct sockaddr *, int);
+
+static int if_rtmtu(struct radix_node *, void *);
+static void if_rtmtu_update(struct ifnet *);
+
+#if IF_CLONE_LIST
+static int     if_clone_list(int count, int * total, user_addr_t dst);
+#endif /* IF_CLONE_LIST */
 
 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
-MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 
 int    ifqmaxlen = IFQ_MAXLEN;
-struct ifnethead ifnet;        /* depend on static init XXX */
+struct ifnethead ifnet_head = TAILQ_HEAD_INITIALIZER(ifnet_head);
+
+static int     if_cloners_count;
+LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
+
+static struct ifaddr *ifa_ifwithnet_common(const struct sockaddr *,
+    unsigned int);
+static void if_attach_ifa_common(struct ifnet *, struct ifaddr *, int);
+static void if_detach_ifa_common(struct ifnet *, struct ifaddr *, int);
+
+static void if_attach_ifma(struct ifnet *, struct ifmultiaddr *, int);
+static int if_detach_ifma(struct ifnet *, struct ifmultiaddr *, int);
+
+static struct ifmultiaddr *ifma_alloc(int);
+static void ifma_free(struct ifmultiaddr *);
+static void ifma_trace(struct ifmultiaddr *, int);
+
+#if DEBUG
+static unsigned int ifma_debug = 1;    /* debugging (enabled) */
+#else
+static unsigned int ifma_debug;                /* debugging (disabled) */
+#endif /* !DEBUG */
+static unsigned int ifma_size;         /* size of zone element */
+static struct zone *ifma_zone;         /* zone for ifmultiaddr */
+
+#define        IFMA_TRACE_HIST_SIZE    32      /* size of trace history */
+
+/* For gdb */
+__private_extern__ unsigned int ifma_trace_hist_size = IFMA_TRACE_HIST_SIZE;
+
+struct ifmultiaddr_dbg {
+       struct ifmultiaddr      ifma;                   /* ifmultiaddr */
+       u_int16_t               ifma_refhold_cnt;       /* # of ref */
+       u_int16_t               ifma_refrele_cnt;       /* # of rele */
+       /*
+        * Circular lists of IFA_ADDREF and IFA_REMREF callers.
+        */
+       ctrace_t                ifma_refhold[IFMA_TRACE_HIST_SIZE];
+       ctrace_t                ifma_refrele[IFMA_TRACE_HIST_SIZE];
+       /*
+        * Trash list linkage
+        */
+       TAILQ_ENTRY(ifmultiaddr_dbg) ifma_trash_link;
+};
+
+/* List of trash ifmultiaddr entries protected by ifma_trash_lock */
+static TAILQ_HEAD(, ifmultiaddr_dbg) ifma_trash_head;
+static decl_lck_mtx_data(, ifma_trash_lock);
+
+#define        IFMA_ZONE_MAX           64              /* maximum elements in zone */
+#define        IFMA_ZONE_NAME          "ifmultiaddr"   /* zone name */
 
 #if INET6
 /*
  * XXX: declare here to avoid to include many inet6 related files..
  * should be more generalized?
  */
-extern void    nd6_setmtu __P((struct ifnet *));
-extern int ip6_auto_on;
+extern void    nd6_setmtu(struct ifnet *);
+extern lck_mtx_t *nd6_mutex;
 #endif
 
+
+void
+ifa_init(void)
+{
+       /* Setup lock group and attribute for ifaddr */
+       ifa_mtx_grp_attr = lck_grp_attr_alloc_init();
+       ifa_mtx_grp = lck_grp_alloc_init("ifaddr", ifa_mtx_grp_attr);
+       ifa_mtx_attr = lck_attr_alloc_init();
+
+       PE_parse_boot_argn("ifa_debug", &ifma_debug, sizeof (ifma_debug));
+
+       ifma_size = (ifma_debug == 0) ? sizeof (struct ifmultiaddr) :
+           sizeof (struct ifmultiaddr_dbg);
+
+       ifma_zone = zinit(ifma_size, IFMA_ZONE_MAX * ifma_size, 0,
+           IFMA_ZONE_NAME);
+       if (ifma_zone == NULL) {
+               panic("%s: failed allocating %s", __func__, IFMA_ZONE_NAME);
+               /* NOTREACHED */
+       }
+       zone_change(ifma_zone, Z_EXPAND, TRUE);
+       zone_change(ifma_zone, Z_CALLERACCT, FALSE);
+
+       lck_mtx_init(&ifma_trash_lock, ifa_mtx_grp, ifa_mtx_attr);
+       TAILQ_INIT(&ifma_trash_head);
+}
+
 /*
  * Network interface utility routines.
  *
@@ -124,428 +247,1069 @@ extern int ip6_auto_on;
  * parameters.
  */
 
-int if_index = 0;
+int if_index;
 struct ifaddr **ifnet_addrs;
-struct ifnet **ifindex2ifnet = NULL;
+struct ifnet **ifindex2ifnet;
 
-
-/*
- * Attach an interface to the
- * list of "active" interfaces.
- */
-void
-old_if_attach(ifp)
-       struct ifnet *ifp;
+__private_extern__ void
+if_attach_ifa(struct ifnet *ifp, struct ifaddr *ifa)
 {
-       unsigned socksize, ifasize;
-       int namelen, masklen;
-       char workbuf[64];
-       register struct sockaddr_dl *sdl;
-       register struct ifaddr *ifa;
-       static int if_indexlim = 8;
-       static int inited;
+       if_attach_ifa_common(ifp, ifa, 0);
+}
 
-       if (ifp->if_snd.ifq_maxlen == 0)
-           ifp->if_snd.ifq_maxlen = ifqmaxlen;
+__private_extern__ void
+if_attach_link_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_attach_ifa_common(ifp, ifa, 1);
+}
 
-       if (!inited) {
-               TAILQ_INIT(&ifnet);
-               inited = 1;
+static void
+if_attach_ifa_common(struct ifnet *ifp, struct ifaddr *ifa, int link)
+{
+       ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_EXCLUSIVE);
+       IFA_LOCK_ASSERT_HELD(ifa);
+
+       if (ifa->ifa_ifp != ifp) {
+               panic("%s: Mismatch ifa_ifp=%p != ifp=%p", __func__,
+                   ifa->ifa_ifp, ifp);
+               /* NOTREACHED */
+       } else if (ifa->ifa_debug & IFD_ATTACHED) {
+               panic("%s: Attempt to attach an already attached ifa=%p",
+                   __func__, ifa);
+               /* NOTREACHED */
+       } else if (link && !(ifa->ifa_debug & IFD_LINK)) {
+               panic("%s: Unexpected non-link address ifa=%p", __func__, ifa);
+               /* NOTREACHED */
+       } else if (!link && (ifa->ifa_debug & IFD_LINK)) {
+               panic("%s: Unexpected link address ifa=%p", __func__, ifa);
+               /* NOTREACHED */
        }
+       IFA_ADDREF_LOCKED(ifa);
+       ifa->ifa_debug |= IFD_ATTACHED;
+       if (link)
+               TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
+       else
+               TAILQ_INSERT_TAIL(&ifp->if_addrhead, ifa, ifa_link);
 
-       TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
-       ifp->if_index = ++if_index;
-       /*
-        * XXX -
-        * The old code would work if the interface passed a pre-existing
-        * chain of ifaddrs to this code.  We don't trust our callers to
-        * properly initialize the tailq, however, so we no longer allow
-        * this unlikely case.
-        */
-       TAILQ_INIT(&ifp->if_addrhead);
-       TAILQ_INIT(&ifp->if_prefixhead);
-       LIST_INIT(&ifp->if_multiaddrs);
-       getmicrotime(&ifp->if_lastchange);
-       if (ifnet_addrs == 0 || if_index >= if_indexlim) {
-               unsigned n = (if_indexlim <<= 1) * sizeof(ifa);
-               struct ifaddr **q = (struct ifaddr **)
-                                       _MALLOC(n, M_IFADDR, M_WAITOK);
-               bzero((caddr_t)q, n);
-               if (ifnet_addrs) {
-                       bcopy((caddr_t)ifnet_addrs, (caddr_t)q, n/2);
-                       FREE((caddr_t)ifnet_addrs, M_IFADDR);
-               }
-               ifnet_addrs = (struct ifaddr **)q;
-
-               /* grow ifindex2ifnet */
-               n = if_indexlim * sizeof(struct ifaddr *);
-               q = (struct ifaddr **)_MALLOC(n, M_IFADDR, M_WAITOK);
-               bzero(q, n);
-               if (ifindex2ifnet) {
-                       bcopy((caddr_t)ifindex2ifnet, q, n/2);
-                       _FREE((caddr_t)ifindex2ifnet, M_IFADDR);
-               }
-               ifindex2ifnet = (struct ifnet **)q;
-       }
+       if (ifa->ifa_attached != NULL)
+               (*ifa->ifa_attached)(ifa);
+}
 
-       ifindex2ifnet[if_index] = ifp;
+__private_extern__ void
+if_detach_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_detach_ifa_common(ifp, ifa, 0);
+}
 
-       /*
-        * create a Link Level name for this device
-        */
-       namelen = snprintf(workbuf, sizeof(workbuf),
-           "%s%d", ifp->if_name, ifp->if_unit);
-#define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
-       masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
-       socksize = masklen + ifp->if_addrlen;
-#define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
-       if (socksize < sizeof(*sdl))
-               socksize = sizeof(*sdl);
-       socksize = ROUNDUP(socksize);
-       ifasize = sizeof(*ifa) + 2 * socksize;
-       ifa = (struct ifaddr *) _MALLOC(ifasize, M_IFADDR, M_WAITOK);
-       if (ifa) {
-               bzero((caddr_t)ifa, ifasize);
-               sdl = (struct sockaddr_dl *)(ifa + 1);
-               sdl->sdl_len = socksize;
-               sdl->sdl_family = AF_LINK;
-               bcopy(workbuf, sdl->sdl_data, namelen);
-               sdl->sdl_nlen = namelen;
-               sdl->sdl_index = ifp->if_index;
-               sdl->sdl_type = ifp->if_type;
-               ifnet_addrs[if_index - 1] = ifa;
-               ifa->ifa_ifp = ifp;
-               ifa->ifa_rtrequest = link_rtrequest;
-               ifa->ifa_addr = (struct sockaddr *)sdl;
-               sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
-               ifa->ifa_netmask = (struct sockaddr *)sdl;
-               sdl->sdl_len = masklen;
-               while (namelen != 0)
-                       sdl->sdl_data[--namelen] = 0xff;
-               TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
-       }
+__private_extern__ void
+if_detach_link_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_detach_ifa_common(ifp, ifa, 1);
 }
 
-/*
- * Locate an interface based on a complete address.
- */
-/*ARGSUSED*/
-struct ifaddr *
-ifa_ifwithaddr(addr)
-       register struct sockaddr *addr;
-{
-       register struct ifnet *ifp;
-       register struct ifaddr *ifa;
-
-#define        equal(a1, a2) \
-  (bcmp((caddr_t)(a1), (caddr_t)(a2), ((struct sockaddr *)(a1))->sa_len) == 0)
-       for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next)
-           for (ifa = ifp->if_addrhead.tqh_first; ifa;
-                ifa = ifa->ifa_link.tqe_next) {
-               if (ifa->ifa_addr->sa_family != addr->sa_family)
-                       continue;
-               if (equal(addr, ifa->ifa_addr))
-                       return (ifa);
-               if ((ifp->if_flags & IFF_BROADCAST) && ifa->ifa_broadaddr &&
-                   /* IP6 doesn't have broadcast */
-                   ifa->ifa_broadaddr->sa_len != 0 &&
-                   equal(ifa->ifa_broadaddr, addr))
-                       return (ifa);
+static void
+if_detach_ifa_common(struct ifnet *ifp, struct ifaddr *ifa, int link)
+{
+       ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_EXCLUSIVE);
+       IFA_LOCK_ASSERT_HELD(ifa);
+
+       if (link && !(ifa->ifa_debug & IFD_LINK)) {
+               panic("%s: Unexpected non-link address ifa=%p", __func__, ifa);
+               /* NOTREACHED */
+       } else if (link && ifa != TAILQ_FIRST(&ifp->if_addrhead)) {
+               panic("%s: Link address ifa=%p not first", __func__, ifa);
+               /* NOTREACHED */
+       } else if (!link && (ifa->ifa_debug & IFD_LINK)) {
+               panic("%s: Unexpected link address ifa=%p", __func__, ifa);
+               /* NOTREACHED */
+       } else if (!(ifa->ifa_debug & IFD_ATTACHED)) {
+               panic("%s: Attempt to detach an unattached address ifa=%p",
+                   __func__, ifa);
+               /* NOTREACHED */
+       } else if (ifa->ifa_ifp != ifp) {
+               panic("%s: Mismatch ifa_ifp=%p, ifp=%p", __func__,
+                   ifa->ifa_ifp, ifp);
+               /* NOTREACHED */
+       } else if (ifa->ifa_debug & IFD_DEBUG) {
+               struct ifaddr *ifa2;
+               TAILQ_FOREACH(ifa2, &ifp->if_addrhead, ifa_link) {
+                       if (ifa2 == ifa)
+                               break;
+               }
+               if (ifa2 != ifa) {
+                       panic("%s: Attempt to detach a stray address ifa=%p",
+                           __func__, ifa);
+                       /* NOTREACHED */
+               }
+       }
+       TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
+       /* This must not be the last reference to the ifaddr */
+       if (IFA_REMREF_LOCKED(ifa) == NULL) {
+               panic("%s: unexpected (missing) refcnt ifa=%p", __func__, ifa);
+               /* NOTREACHED */
        }
-       return ((struct ifaddr *)0);
+       ifa->ifa_debug &= ~IFD_ATTACHED;
+
+       if (ifa->ifa_detached != NULL)
+               (*ifa->ifa_detached)(ifa);
 }
+
+#define INITIAL_IF_INDEXLIM    8
+
 /*
- * Locate the point to point interface with a given destination address.
+ * Function: if_next_index
+ * Purpose:
+ *   Return the next available interface index.  
+ *   Grow the ifnet_addrs[] and ifindex2ifnet[] arrays to accomodate the 
+ *   added entry when necessary.
+ *
+ * Note:
+ *   ifnet_addrs[] is indexed by (if_index - 1), whereas
+ *   ifindex2ifnet[] is indexed by ifp->if_index.  That requires us to
+ *   always allocate one extra element to hold ifindex2ifnet[0], which
+ *   is unused.
  */
-/*ARGSUSED*/
-struct ifaddr *
-ifa_ifwithdstaddr(addr)
-       register struct sockaddr *addr;
+int if_next_index(void);
+
+__private_extern__ int
+if_next_index(void)
 {
-       register struct ifnet *ifp;
-       register struct ifaddr *ifa;
+       static int      if_indexlim = 0;
+       int             new_index;
+
+       new_index = ++if_index;
+       if (if_index > if_indexlim) {
+               unsigned        n;
+               int             new_if_indexlim;
+               caddr_t         new_ifnet_addrs;
+               caddr_t         new_ifindex2ifnet;
+               caddr_t         old_ifnet_addrs;
+
+               old_ifnet_addrs = (caddr_t)ifnet_addrs;
+               if (ifnet_addrs == NULL) {
+                       new_if_indexlim = INITIAL_IF_INDEXLIM;
+               } else {
+                       new_if_indexlim = if_indexlim << 1;
+               }
 
-       for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next)
-           if (ifp->if_flags & IFF_POINTOPOINT)
-               for (ifa = ifp->if_addrhead.tqh_first; ifa;
-                    ifa = ifa->ifa_link.tqe_next) {
-                       if (ifa->ifa_addr->sa_family != addr->sa_family)
-                               continue;
-                       if (ifa->ifa_dstaddr && equal(addr, ifa->ifa_dstaddr))
-                               return (ifa);
+               /* allocate space for the larger arrays */
+               n = (2 * new_if_indexlim + 1) * sizeof(caddr_t);
+               new_ifnet_addrs = _MALLOC(n, M_IFADDR, M_WAITOK);
+               if (new_ifnet_addrs == NULL) {
+                       --if_index;
+                       return -1;
+               }
+
+               new_ifindex2ifnet = new_ifnet_addrs 
+                       + new_if_indexlim * sizeof(caddr_t);
+               bzero(new_ifnet_addrs, n);
+               if (ifnet_addrs != NULL) {
+                       /* copy the existing data */
+                       bcopy((caddr_t)ifnet_addrs, new_ifnet_addrs,
+                             if_indexlim * sizeof(caddr_t));
+                       bcopy((caddr_t)ifindex2ifnet,
+                             new_ifindex2ifnet,
+                             (if_indexlim + 1) * sizeof(caddr_t));
+               }
+
+               /* switch to the new tables and size */
+               ifnet_addrs = (struct ifaddr **)new_ifnet_addrs;
+               ifindex2ifnet = (struct ifnet **)new_ifindex2ifnet;
+               if_indexlim = new_if_indexlim;
+
+               /* release the old data */
+               if (old_ifnet_addrs != NULL) {
+                       _FREE((caddr_t)old_ifnet_addrs, M_IFADDR);
+               }
        }
-       return ((struct ifaddr *)0);
+       return (new_index);
 }
 
 /*
- * Find an interface on a specific network.  If many, choice
- * is most specific found.
+ * Create a clone network interface.
  */
-struct ifaddr *
-ifa_ifwithnet(addr)
-       struct sockaddr *addr;
+static int
+if_clone_create(char *name, int len, void *params)
 {
-       register struct ifnet *ifp;
-       register struct ifaddr *ifa;
-       struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
-       u_int af = addr->sa_family;
-       char *addr_data = addr->sa_data, *cplim;
-
+       struct if_clone *ifc;
+       char *dp;
+       int wildcard;
+       u_int32_t bytoff, bitoff;
+       u_int32_t unit;
+       int err;
+
+       ifc = if_clone_lookup(name, &unit);
+       if (ifc == NULL)
+               return (EINVAL);
+
+       if (ifunit(name) != NULL)
+               return (EEXIST);
+
+       bytoff = bitoff = 0;
+       wildcard = (unit == UINT32_MAX);
        /*
-        * AF_LINK addresses can be looked up directly by their index number,
-        * so do that if we can.
+        * Find a free unit if none was given.
         */
-       if (af == AF_LINK) {
-           register struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
-           if (sdl->sdl_index && sdl->sdl_index <= if_index)
-               return (ifnet_addrs[sdl->sdl_index - 1]);
+       if (wildcard) {
+               while ((bytoff < ifc->ifc_bmlen)
+                   && (ifc->ifc_units[bytoff] == 0xff))
+                       bytoff++;
+               if (bytoff >= ifc->ifc_bmlen)
+                       return (ENOSPC);
+               while ((ifc->ifc_units[bytoff] & (1 << bitoff)) != 0)
+                       bitoff++;
+               unit = (bytoff << 3) + bitoff;
        }
 
-       /*
-        * Scan though each interface, looking for ones that have
-        * addresses in this address family.
-        */
-       for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next) {
-               for (ifa = ifp->if_addrhead.tqh_first; ifa;
-                    ifa = ifa->ifa_link.tqe_next) {
-                       register char *cp, *cp2, *cp3;
+       if (unit > ifc->ifc_maxunit)
+               return (ENXIO);
 
-                       if (ifa->ifa_addr->sa_family != af)
-next:                          continue;
-#ifndef __APPLE__
-/* This breaks tunneling application trying to install a route with
- * a specific subnet and the local address as the destination
- * It's breaks binary compatibility with previous version of MacOS X
- */
-                       if (
-#if INET6 /* XXX: for maching gif tunnel dst as routing entry gateway */
-                           addr->sa_family != AF_INET6 &&
-#endif
-                           ifp->if_flags & IFF_POINTOPOINT) {
-                               /*
-                                * This is a bit broken as it doesn't
-                                * take into account that the remote end may
-                                * be a single node in the network we are
-                                * looking for.
-                                * The trouble is that we don't know the
-                                * netmask for the remote end.
-                                */
-                               if (ifa->ifa_dstaddr != 0
-                                   && equal(addr, ifa->ifa_dstaddr))
-                                       return (ifa);
-                       } else
-#endif /* __APPLE__*/
-                       {
-                               /*
-                                * if we have a special address handler,
-                                * then use it instead of the generic one.
-                                */
-                               if (ifa->ifa_claim_addr) {
-                                       if ((*ifa->ifa_claim_addr)(ifa, addr)) {
-                                               return (ifa);
-                                       } else {
-                                               continue;
-                                       }
-                               }
+       err = (*ifc->ifc_create)(ifc, unit, params);
+       if (err != 0)
+               return (err);
 
-                               /*
-                                * Scan all the bits in the ifa's address.
-                                * If a bit dissagrees with what we are
-                                * looking for, mask it with the netmask
-                                * to see if it really matters.
-                                * (A byte at a time)
-                                */
-                               if (ifa->ifa_netmask == 0)
-                                       continue;
-                               cp = addr_data;
-                               cp2 = ifa->ifa_addr->sa_data;
-                               cp3 = ifa->ifa_netmask->sa_data;
-                               cplim = ifa->ifa_netmask->sa_len
-                                       + (char *)ifa->ifa_netmask;
-                               while (cp3 < cplim)
-                                       if ((*cp++ ^ *cp2++) & *cp3++)
-                                               goto next; /* next address! */
-                               /*
-                                * If the netmask of what we just found
-                                * is more specific than what we had before
-                                * (if we had one) then remember the new one
-                                * before continuing to search
-                                * for an even better one.
-                                */
-                               if (ifa_maybe == 0 ||
-                                   rn_refines((caddr_t)ifa->ifa_netmask,
-                                   (caddr_t)ifa_maybe->ifa_netmask))
-                                       ifa_maybe = ifa;
-                       }
-               }
+       if (!wildcard) {
+               bytoff = unit >> 3;
+               bitoff = unit - (bytoff << 3);
        }
-       return (ifa_maybe);
-}
-
-/*
- * Find an interface address specific to an interface best matching
- * a given address.
- */
-struct ifaddr *
-ifaof_ifpforaddr(addr, ifp)
-       struct sockaddr *addr;
-       register struct ifnet *ifp;
-{
-       register struct ifaddr *ifa;
-       register char *cp, *cp2, *cp3;
-       register char *cplim;
-       struct ifaddr *ifa_maybe = 0;
-       u_int af = addr->sa_family;
 
-       if (af >= AF_MAX)
-               return (0);
-       for (ifa = ifp->if_addrhead.tqh_first; ifa;
-            ifa = ifa->ifa_link.tqe_next) {
-               if (ifa->ifa_addr->sa_family != af)
-                       continue;
-               if (ifa_maybe == 0)
-                       ifa_maybe = ifa;
-               if (ifa->ifa_netmask == 0) {
-                       if (equal(addr, ifa->ifa_addr) ||
-                           (ifa->ifa_dstaddr && equal(addr, ifa->ifa_dstaddr)))
-                               return (ifa);
-                       continue;
-               }
-               if (ifp->if_flags & IFF_POINTOPOINT) {
-                       if (equal(addr, ifa->ifa_dstaddr))
-                               return (ifa);
-               } else {
-                       cp = addr->sa_data;
-                       cp2 = ifa->ifa_addr->sa_data;
-                       cp3 = ifa->ifa_netmask->sa_data;
-                       cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
-                       for (; cp3 < cplim; cp3++)
-                               if ((*cp++ ^ *cp2++) & *cp3)
-                                       break;
-                       if (cp3 == cplim)
-                               return (ifa);
+       /*
+        * Allocate the unit in the bitmap.
+        */
+       KASSERT((ifc->ifc_units[bytoff] & (1 << bitoff)) == 0,
+           ("%s: bit is already set", __func__));
+       ifc->ifc_units[bytoff] |= (1 << bitoff);
+
+       /* In the wildcard case, we need to update the name. */
+       if (wildcard) {
+               for (dp = name; *dp != '\0'; dp++);
+               if (snprintf(dp, len - (dp-name), "%d", unit) >
+                   len - (dp-name) - 1) {
+                       /*
+                        * This can only be a programmer error and
+                        * there's no straightforward way to recover if
+                        * it happens.
+                        */
+                       panic("%s: interface name too long", __func__);
+                       /* NOTREACHED */
                }
+
        }
-       return (ifa_maybe);
-}
 
-#include <net/route.h>
+       return (0);
+}
 
 /*
- * Default action when installing a route with a Link Level gateway.
- * Lookup an appropriate real ifa to point to.
- * This should be moved to /sys/net/link.c eventually.
+ * Destroy a clone network interface.
  */
-static void
-link_rtrequest(cmd, rt, sa)
-       int cmd;
-       register struct rtentry *rt;
-       struct sockaddr *sa;
+static int
+if_clone_destroy(const char *name)
 {
-       register struct ifaddr *ifa;
-       struct sockaddr *dst;
+       struct if_clone *ifc;
        struct ifnet *ifp;
+       int bytoff, bitoff;
+       u_int32_t unit;
 
-       if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
-           ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
-               return;
-       ifa = ifaof_ifpforaddr(dst, ifp);
-       if (ifa) {
-               rtsetifa(rt, ifa);
-               if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
-                       ifa->ifa_rtrequest(cmd, rt, sa);
-       }
-}
+       ifc = if_clone_lookup(name, &unit);
+       if (ifc == NULL)
+               return (EINVAL);
 
-/*
- * Mark an interface down and notify protocols of
- * the transition.
- * NOTE: must be called at splnet or eqivalent.
- */
-void
-if_unroute(ifp, flag, fam)
-       register struct ifnet *ifp;
-       int flag, fam;
-{
-       register struct ifaddr *ifa;
-
-       ifp->if_flags &= ~flag;
-       getmicrotime(&ifp->if_lastchange);
-       TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
-               if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
-                       pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
-       if_qflush(&ifp->if_snd);
-       rt_ifmsg(ifp);
+       if (unit < ifc->ifc_minifs)
+               return (EINVAL);
+
+       ifp = ifunit(name);
+       if (ifp == NULL)
+               return (ENXIO);
+
+       if (ifc->ifc_destroy == NULL)
+               return (EOPNOTSUPP);
+
+       (*ifc->ifc_destroy)(ifp);
+
+       /*
+        * Compute offset in the bitmap and deallocate the unit.
+        */
+       bytoff = unit >> 3;
+       bitoff = unit - (bytoff << 3);
+       KASSERT((ifc->ifc_units[bytoff] & (1 << bitoff)) != 0,
+           ("%s: bit is already cleared", __func__));
+       ifc->ifc_units[bytoff] &= ~(1 << bitoff);
+       return (0);
 }
 
 /*
- * Mark an interface up and notify protocols of
- * the transition.
- * NOTE: must be called at splnet or eqivalent.
+ * Look up a network interface cloner.
  */
-void
-if_route(ifp, flag, fam)
-       register struct ifnet *ifp;
-       int flag, fam;
+
+__private_extern__ struct if_clone *
+if_clone_lookup(const char *name, u_int32_t *unitp)
 {
-       register struct ifaddr *ifa;
+       struct if_clone *ifc;
+       const char *cp;
+       size_t i;
 
-       ifp->if_flags |= flag;
-       getmicrotime(&ifp->if_lastchange);
-       TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
-               if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
-                       pfctlinput(PRC_IFUP, ifa->ifa_addr);
-       rt_ifmsg(ifp);
+       for (ifc = LIST_FIRST(&if_cloners); ifc != NULL;) {
+               for (cp = name, i = 0; i < ifc->ifc_namelen; i++, cp++) {
+                       if (ifc->ifc_name[i] != *cp)
+                               goto next_ifc;
+               }
+               goto found_name;
+ next_ifc:
+               ifc = LIST_NEXT(ifc, ifc_list);
+       }
 
-#if INET6
-       if (ip6_auto_on) /* Only if IPv6 is on on configured on on all ifs */
-               in6_if_up(ifp);
-#endif
+       /* No match. */
+       return ((struct if_clone *)NULL);
+
+ found_name:
+       if (*cp == '\0') {
+               i = 0xffff;
+       } else {
+               for (i = 0; *cp != '\0'; cp++) {
+                       if (*cp < '0' || *cp > '9') {
+                               /* Bogus unit number. */
+                               return (NULL);
+                       }
+                       i = (i * 10) + (*cp - '0');
+               }
+       }
+
+       if (unitp != NULL)
+               *unitp = i;
+       return (ifc);
 }
 
 /*
- * Mark an interface down and notify protocols of
- * the transition.
- * NOTE: must be called at splnet or eqivalent.
+ * Register a network interface cloner.
  */
-void
-if_down(ifp)
-       register struct ifnet *ifp;
+int
+if_clone_attach(struct if_clone *ifc)
 {
+       int bytoff, bitoff;
+       int err;
+       int len, maxclone;
+       u_int32_t unit;
+
+       KASSERT(ifc->ifc_minifs - 1 <= ifc->ifc_maxunit,
+           ("%s: %s requested more units then allowed (%d > %d)",
+           __func__, ifc->ifc_name, ifc->ifc_minifs,
+           ifc->ifc_maxunit + 1));
+       /*
+        * Compute bitmap size and allocate it.
+        */
+       maxclone = ifc->ifc_maxunit + 1;
+       len = maxclone >> 3;
+       if ((len << 3) < maxclone)
+               len++;
+       ifc->ifc_units = _MALLOC(len, M_CLONE, M_WAITOK | M_ZERO);
+       if (ifc->ifc_units == NULL)
+               return ENOBUFS;
+       bzero(ifc->ifc_units, len);
+       ifc->ifc_bmlen = len;
+
+       LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
+       if_cloners_count++;
+
+       for (unit = 0; unit < ifc->ifc_minifs; unit++) {
+               err = (*ifc->ifc_create)(ifc, unit, NULL);
+               KASSERT(err == 0,
+                   ("%s: failed to create required interface %s%d",
+                   __func__, ifc->ifc_name, unit));
+
+               /* Allocate the unit in the bitmap. */
+               bytoff = unit >> 3;
+               bitoff = unit - (bytoff << 3);
+               ifc->ifc_units[bytoff] |= (1 << bitoff);
+       }
 
-       if_unroute(ifp, IFF_UP, AF_UNSPEC);
+       return 0;
 }
 
 /*
- * Mark an interface up and notify protocols of
- * the transition.
- * NOTE: must be called at splnet or eqivalent.
+ * Unregister a network interface cloner.
  */
 void
-if_up(ifp)
-       register struct ifnet *ifp;
+if_clone_detach(struct if_clone *ifc)
 {
 
-       if_route(ifp, IFF_UP, AF_UNSPEC);
+       LIST_REMOVE(ifc, ifc_list);
+       FREE(ifc->ifc_units, M_CLONE);
+       if_cloners_count--;
 }
 
+#if IF_CLONE_LIST
 /*
- * Flush an interface queue.
+ * Provide list of interface cloners to userspace.
  */
-static void
-if_qflush(ifq)
-       register struct ifqueue *ifq;
+static int
+if_clone_list(int count, int * total, user_addr_t dst)
 {
-       register struct mbuf *m, *n;
+       char outbuf[IFNAMSIZ];
+       struct if_clone *ifc;
+       int error = 0;
+
+       *total = if_cloners_count;
+       if (dst == USER_ADDR_NULL) {
+               /* Just asking how many there are. */
+               return (0);
+       }
+
+       if (count < 0)
+               return (EINVAL);
+
+       count = (if_cloners_count < count) ? if_cloners_count : count;
+
+       for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
+            ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
+               strlcpy(outbuf, ifc->ifc_name, IFNAMSIZ);
+               error = copyout(outbuf, dst, IFNAMSIZ);
+               if (error)
+                       break;
+       }
+
+       return (error);
+}
+#endif /* IF_CLONE_LIST */
+
+/*
+ * Similar to ifa_ifwithaddr, except that this is IPv4 specific
+ * and that it matches only the local (not broadcast) address.
+ */
+__private_extern__ struct in_ifaddr *
+ifa_foraddr(unsigned int addr)
+{
+       return (ifa_foraddr_scoped(addr, IFSCOPE_NONE));
+}
+
+/*
+ * Similar to ifa_foraddr, except with the added interface scope
+ * constraint (unless the caller passes in IFSCOPE_NONE in which
+ * case there is no scope restriction).
+ */
+__private_extern__ struct in_ifaddr *
+ifa_foraddr_scoped(unsigned int addr, unsigned int scope)
+{
+       struct in_ifaddr *ia = NULL;
+
+       lck_rw_lock_shared(in_ifaddr_rwlock);
+       TAILQ_FOREACH(ia, INADDR_HASH(addr), ia_hash) {
+               IFA_LOCK_SPIN(&ia->ia_ifa);
+               if (ia->ia_addr.sin_addr.s_addr == addr &&
+                   (scope == IFSCOPE_NONE || ia->ia_ifp->if_index == scope)) {
+                       IFA_ADDREF_LOCKED(&ia->ia_ifa); /* for caller */
+                       IFA_UNLOCK(&ia->ia_ifa);
+                       break;
+               }
+               IFA_UNLOCK(&ia->ia_ifa);
+       }
+       lck_rw_done(in_ifaddr_rwlock);
+       return (ia);
+}
+
+#if INET6
+/*
+ * Similar to ifa_foraddr, except that this for IPv6.
+ */
+__private_extern__ struct in6_ifaddr *
+ifa_foraddr6(struct in6_addr *addr6)
+{
+       return (ifa_foraddr6_scoped(addr6, IFSCOPE_NONE));
+}
+
+__private_extern__ struct in6_ifaddr *
+ifa_foraddr6_scoped(struct in6_addr *addr6, unsigned int scope)
+{
+       struct in6_ifaddr *ia = NULL;
+
+       lck_rw_lock_shared(&in6_ifaddr_rwlock);
+       for (ia = in6_ifaddrs; ia; ia = ia->ia_next) {
+               IFA_LOCK(&ia->ia_ifa);
+               if (IN6_ARE_ADDR_EQUAL(&ia->ia_addr.sin6_addr, addr6) &&
+                   (scope == IFSCOPE_NONE || ia->ia_ifp->if_index == scope)) {
+                       IFA_ADDREF_LOCKED(&ia->ia_ifa); /* for caller */
+                       IFA_UNLOCK(&ia->ia_ifa);
+                       break;
+               }
+               IFA_UNLOCK(&ia->ia_ifa);
+       }
+       lck_rw_done(&in6_ifaddr_rwlock);
+
+       return (ia);
+}
+#endif /* INET6 */
+
+/*
+ * Return the first (primary) address of a given family on an interface.
+ */
+__private_extern__ struct ifaddr *
+ifa_ifpgetprimary(struct ifnet *ifp, int family)
+{
+       struct ifaddr *ifa;
+
+       ifnet_lock_shared(ifp);
+       TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
+               IFA_LOCK_SPIN(ifa);
+               if (ifa->ifa_addr->sa_family == family) {
+                       IFA_ADDREF_LOCKED(ifa); /* for caller */
+                       IFA_UNLOCK(ifa);
+                       break;
+               }
+               IFA_UNLOCK(ifa);
+       }
+       ifnet_lock_done(ifp);
+
+       return (ifa);
+}
+
+/*
+ * Locate an interface based on a complete address.
+ */
+/*ARGSUSED*/
+struct ifaddr *
+ifa_ifwithaddr(const struct sockaddr *addr)
+{
+       struct ifnet *ifp;
+       struct ifaddr *ifa;
+       struct ifaddr *result = NULL;
+
+#define        equal(a1, a2)                                                   \
+       (bcmp((const void*)(a1), (const void*)(a2),                     \
+           ((const struct sockaddr *)(a1))->sa_len) == 0)
+
+       ifnet_head_lock_shared();
+       for (ifp = ifnet_head.tqh_first; ifp && !result;
+           ifp = ifp->if_link.tqe_next) {
+               ifnet_lock_shared(ifp);
+               for (ifa = ifp->if_addrhead.tqh_first; ifa;
+                   ifa = ifa->ifa_link.tqe_next) {
+                       IFA_LOCK_SPIN(ifa);
+                       if (ifa->ifa_addr->sa_family != addr->sa_family) {
+                               IFA_UNLOCK(ifa);
+                               continue;
+                       }
+                       if (equal(addr, ifa->ifa_addr)) {
+                               result = ifa;
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       if ((ifp->if_flags & IFF_BROADCAST) &&
+                           ifa->ifa_broadaddr != NULL &&
+                           /* IP6 doesn't have broadcast */
+                           ifa->ifa_broadaddr->sa_len != 0 &&
+                           equal(ifa->ifa_broadaddr, addr)) {
+                               result = ifa;
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       IFA_UNLOCK(ifa);
+               }
+               ifnet_lock_done(ifp);
+       }
+       ifnet_head_done();
+
+       return (result);
+}
+/*
+ * Locate the point to point interface with a given destination address.
+ */
+/*ARGSUSED*/
+struct ifaddr *
+ifa_ifwithdstaddr(const struct sockaddr *addr)
+{
+       struct ifnet *ifp;
+       struct ifaddr *ifa;
+       struct ifaddr *result = NULL;
+
+       ifnet_head_lock_shared();
+       for (ifp = ifnet_head.tqh_first; ifp && !result;
+           ifp = ifp->if_link.tqe_next) {
+           if ((ifp->if_flags & IFF_POINTOPOINT)) {
+                       ifnet_lock_shared(ifp);
+                       for (ifa = ifp->if_addrhead.tqh_first; ifa;
+                           ifa = ifa->ifa_link.tqe_next) {
+                               IFA_LOCK_SPIN(ifa);
+                               if (ifa->ifa_addr->sa_family !=
+                                   addr->sa_family) {
+                                       IFA_UNLOCK(ifa);
+                                       continue;
+                               }
+                               if (ifa->ifa_dstaddr &&
+                                   equal(addr, ifa->ifa_dstaddr)) {
+                                       result = ifa;
+                                       IFA_ADDREF_LOCKED(ifa); /* for caller */
+                                       IFA_UNLOCK(ifa);
+                                       break;
+                               }
+                               IFA_UNLOCK(ifa);
+                       }
+                       ifnet_lock_done(ifp);
+               }
+       }
+       ifnet_head_done();
+       return (result);
+}
+
+/*
+ * Locate the source address of an interface based on a complete address.
+ */
+struct ifaddr *
+ifa_ifwithaddr_scoped(const struct sockaddr *addr, unsigned int ifscope)
+{
+       struct ifaddr *result = NULL;
+       struct ifnet *ifp;
+
+       if (ifscope == IFSCOPE_NONE)
+               return (ifa_ifwithaddr(addr));
+
+       ifnet_head_lock_shared();
+       if (ifscope > (unsigned int)if_index) {
+               ifnet_head_done();
+               return (NULL);
+       }
+
+       ifp = ifindex2ifnet[ifscope];
+       if (ifp != NULL) {
+               struct ifaddr *ifa = NULL;
+
+               /*
+                * This is suboptimal; there should be a better way
+                * to search for a given address of an interface
+                * for any given address family.
+                */
+               ifnet_lock_shared(ifp);
+               for (ifa = ifp->if_addrhead.tqh_first; ifa != NULL;
+                   ifa = ifa->ifa_link.tqe_next) {
+                       IFA_LOCK_SPIN(ifa);
+                       if (ifa->ifa_addr->sa_family != addr->sa_family) {
+                               IFA_UNLOCK(ifa);
+                               continue;
+                       }
+                       if (equal(addr, ifa->ifa_addr)) {
+                               result = ifa;
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       if ((ifp->if_flags & IFF_BROADCAST) &&
+                           ifa->ifa_broadaddr != NULL &&
+                           /* IP6 doesn't have broadcast */
+                           ifa->ifa_broadaddr->sa_len != 0 &&
+                           equal(ifa->ifa_broadaddr, addr)) {
+                               result = ifa;
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       IFA_UNLOCK(ifa);
+               }
+               ifnet_lock_done(ifp);
+       }
+       ifnet_head_done();
+
+       return (result);
+}
+
+struct ifaddr *
+ifa_ifwithnet(const struct sockaddr *addr)
+{
+       return (ifa_ifwithnet_common(addr, IFSCOPE_NONE));
+}
+
+struct ifaddr *
+ifa_ifwithnet_scoped(const struct sockaddr *addr, unsigned int ifscope)
+{
+       return (ifa_ifwithnet_common(addr, ifscope));
+}
+
+/*
+ * Find an interface on a specific network.  If many, choice
+ * is most specific found.
+ */
+static struct ifaddr *
+ifa_ifwithnet_common(const struct sockaddr *addr, unsigned int ifscope)
+{
+       struct ifnet *ifp;
+       struct ifaddr *ifa = NULL;
+       struct ifaddr *ifa_maybe = NULL;
+       u_int af = addr->sa_family;
+       const char *addr_data = addr->sa_data, *cplim;
+
+#if INET6
+       if ((af != AF_INET && af != AF_INET6) ||
+           (af == AF_INET && !ip_doscopedroute) ||
+           (af == AF_INET6 && !ip6_doscopedroute))
+#else
+       if (af != AF_INET || !ip_doscopedroute)
+#endif /* !INET6 */
+               ifscope = IFSCOPE_NONE;
+
+       ifnet_head_lock_shared();
+       /*
+        * AF_LINK addresses can be looked up directly by their index number,
+        * so do that if we can.
+        */
+       if (af == AF_LINK) {
+               const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)addr;
+               if (sdl->sdl_index && sdl->sdl_index <= if_index) {
+                       ifa = ifnet_addrs[sdl->sdl_index - 1];
+                       if (ifa != NULL)
+                               IFA_ADDREF(ifa);
+
+                       ifnet_head_done();
+                       return (ifa);
+               }
+       }
+
+       /*
+        * Scan though each interface, looking for ones that have
+        * addresses in this address family.
+        */
+       for (ifp = ifnet_head.tqh_first; ifp; ifp = ifp->if_link.tqe_next) {
+               ifnet_lock_shared(ifp);
+               for (ifa = ifp->if_addrhead.tqh_first; ifa;
+                    ifa = ifa->ifa_link.tqe_next) {
+                       const char *cp, *cp2, *cp3;
+
+                       IFA_LOCK(ifa);
+                       if (ifa->ifa_addr == NULL ||
+                           ifa->ifa_addr->sa_family != af) {
+next:
+                               IFA_UNLOCK(ifa);
+                               continue;
+                       }
+#ifndef __APPLE__
+/* This breaks tunneling application trying to install a route with
+ * a specific subnet and the local address as the destination
+ * It's breaks binary compatibility with previous version of MacOS X
+ */
+                       if (
+#if INET6 /* XXX: for maching gif tunnel dst as routing entry gateway */
+                           addr->sa_family != AF_INET6 &&
+#endif
+                           ifp->if_flags & IFF_POINTOPOINT) {
+                               /*
+                                * This is a bit broken as it doesn't
+                                * take into account that the remote end may
+                                * be a single node in the network we are
+                                * looking for.
+                                * The trouble is that we don't know the
+                                * netmask for the remote end.
+                                */
+                               if (ifa->ifa_dstaddr != 0 &&
+                                   equal(addr, ifa->ifa_dstaddr)) {
+                                       IFA_ADDREF_LOCKED(ifa);
+                                       IFA_UNLOCK(ifa);
+                                       break;
+                               }
+                               IFA_UNLOCK(ifa);
+                       } else
+#endif /* __APPLE__*/
+                       {
+                               /*
+                                * If we're looking up with a scope,
+                                * find using a matching interface.
+                                */
+                               if (ifscope != IFSCOPE_NONE &&
+                                   ifp->if_index != ifscope) {
+                                       IFA_UNLOCK(ifa);
+                                       continue;
+                               }
+
+                               /*
+                                * Scan all the bits in the ifa's address.
+                                * If a bit dissagrees with what we are
+                                * looking for, mask it with the netmask
+                                * to see if it really matters.
+                                * (A byte at a time)
+                                */
+                               if (ifa->ifa_netmask == 0) {
+                                       IFA_UNLOCK(ifa);
+                                       continue;
+                               }
+                               cp = addr_data;
+                               cp2 = ifa->ifa_addr->sa_data;
+                               cp3 = ifa->ifa_netmask->sa_data;
+                               cplim = ifa->ifa_netmask->sa_len
+                                       + (char *)ifa->ifa_netmask;
+                               while (cp3 < cplim)
+                                       if ((*cp++ ^ *cp2++) & *cp3++)
+                                               goto next; /* next address! */
+                               /*
+                                * If the netmask of what we just found
+                                * is more specific than what we had before
+                                * (if we had one) then remember the new one
+                                * before continuing to search
+                                * for an even better one.
+                                */
+                               if (ifa_maybe == NULL ||
+                                   rn_refines((caddr_t)ifa->ifa_netmask,
+                                   (caddr_t)ifa_maybe->ifa_netmask)) {
+                                       IFA_ADDREF_LOCKED(ifa); /* ifa_maybe */
+                                       IFA_UNLOCK(ifa);
+                                       if (ifa_maybe != NULL)
+                                               IFA_REMREF(ifa_maybe);
+                                       ifa_maybe = ifa;
+                               } else {
+                                       IFA_UNLOCK(ifa);
+                               }
+                       }
+                       IFA_LOCK_ASSERT_NOTHELD(ifa);
+               }
+               ifnet_lock_done(ifp);
+
+               if (ifa != NULL)
+                       break;
+       }
+       ifnet_head_done();
+
+       if (ifa == NULL)
+               ifa = ifa_maybe;
+       else if (ifa_maybe != NULL)
+               IFA_REMREF(ifa_maybe);
+
+       return (ifa);
+}
+
+/*
+ * Find an interface address specific to an interface best matching
+ * a given address.
+ */
+struct ifaddr *
+ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
+{
+       struct ifaddr *ifa = NULL;
+       const char *cp, *cp2, *cp3;
+       char *cplim;
+       struct ifaddr *ifa_maybe = NULL;
+       struct ifaddr *better_ifa_maybe = NULL;
+       u_int af = addr->sa_family;
+
+       if (af >= AF_MAX)
+               return (NULL);
+
+       ifnet_lock_shared(ifp);
+       for (ifa = ifp->if_addrhead.tqh_first; ifa;
+            ifa = ifa->ifa_link.tqe_next) {
+               IFA_LOCK(ifa);
+               if (ifa->ifa_addr->sa_family != af) {
+                       IFA_UNLOCK(ifa);
+                       continue;
+               }
+               if (ifa_maybe == NULL) {
+                       IFA_ADDREF_LOCKED(ifa); /* for ifa_maybe */
+                       ifa_maybe = ifa;
+               }
+               if (ifa->ifa_netmask == 0) {
+                       if (equal(addr, ifa->ifa_addr) || (ifa->ifa_dstaddr &&
+                           equal(addr, ifa->ifa_dstaddr))) {
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       IFA_UNLOCK(ifa);
+                       continue;
+               }
+               if (ifp->if_flags & IFF_POINTOPOINT) {
+                       if (ifa->ifa_dstaddr && equal(addr, ifa->ifa_dstaddr)) {
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+               } else {
+                       if (equal(addr, ifa->ifa_addr)) {
+                               /* exact match */
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       cp = addr->sa_data;
+                       cp2 = ifa->ifa_addr->sa_data;
+                       cp3 = ifa->ifa_netmask->sa_data;
+                       cplim = ifa->ifa_netmask->sa_len +
+                           (char *)ifa->ifa_netmask;
+                       for (; cp3 < cplim; cp3++)
+                               if ((*cp++ ^ *cp2++) & *cp3)
+                                       break;
+                       if (cp3 == cplim) {
+                               /* subnet match */
+                               if (better_ifa_maybe == NULL) {
+                                       /* for better_ifa_maybe */
+                                       IFA_ADDREF_LOCKED(ifa);
+                                       better_ifa_maybe = ifa;
+                               }
+                       }
+               }
+               IFA_UNLOCK(ifa);
+       }
+
+       if (ifa == NULL) {
+               if (better_ifa_maybe != NULL) {
+                       ifa = better_ifa_maybe;
+                       better_ifa_maybe = NULL;
+               } else {
+                       ifa = ifa_maybe;
+                       ifa_maybe = NULL;
+               }
+       }
+
+       ifnet_lock_done(ifp);
+
+       if (better_ifa_maybe != NULL)
+               IFA_REMREF(better_ifa_maybe);
+       if (ifa_maybe != NULL)
+               IFA_REMREF(ifa_maybe);
+
+       return (ifa);
+}
+
+#include <net/route.h>
+
+/*
+ * Default action when installing a route with a Link Level gateway.
+ * Lookup an appropriate real ifa to point to.
+ * This should be moved to /sys/net/link.c eventually.
+ */
+void
+link_rtrequest(int cmd, struct rtentry *rt, struct sockaddr *sa)
+{
+       struct ifaddr *ifa;
+       struct sockaddr *dst;
+       struct ifnet *ifp;
+       void (*ifa_rtrequest)(int, struct rtentry *, struct sockaddr *);
+
+       lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_OWNED);
+       RT_LOCK_ASSERT_HELD(rt);
+
+       if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
+           ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
+               return;
+
+       /* Become a regular mutex, just in case */
+       RT_CONVERT_LOCK(rt);
+
+       ifa = ifaof_ifpforaddr(dst, ifp);
+       if (ifa) {
+               rtsetifa(rt, ifa);
+               IFA_LOCK_SPIN(ifa);
+               ifa_rtrequest = ifa->ifa_rtrequest;
+               IFA_UNLOCK(ifa);
+               if (ifa_rtrequest != NULL && ifa_rtrequest != link_rtrequest)
+                       ifa_rtrequest(cmd, rt, sa);
+               IFA_REMREF(ifa);
+       }
+}
+
+/*
+ * if_updown will set the interface up or down. It will
+ * prevent other up/down events from occurring until this
+ * up/down event has completed.
+ *
+ * Caller must lock ifnet. This function will drop the
+ * lock. This allows ifnet_set_flags to set the rest of
+ * the flags after we change the up/down state without
+ * dropping the interface lock between setting the
+ * up/down state and updating the rest of the flags.
+ */
+__private_extern__ void
+if_updown(
+       struct ifnet    *ifp,
+       int                             up)
+{
+       int i;
+       struct ifaddr **ifa;
+       struct timespec tv;
+
+       /* Wait until no one else is changing the up/down state */
+       while ((ifp->if_eflags & IFEF_UPDOWNCHANGE) != 0) {
+               tv.tv_sec = 0;
+               tv.tv_nsec = NSEC_PER_SEC / 10;
+               ifnet_lock_done(ifp);
+               msleep(&ifp->if_eflags, NULL, 0, "if_updown", &tv);
+               ifnet_lock_exclusive(ifp);
+       }
+       
+       /* Verify that the interface isn't already in the right state */
+       if ((!up && (ifp->if_flags & IFF_UP) == 0) ||
+               (up && (ifp->if_flags & IFF_UP) == IFF_UP)) {
+               return;
+       }
+       
+       /* Indicate that the up/down state is changing */
+       ifp->if_eflags |= IFEF_UPDOWNCHANGE;
+       
+       /* Mark interface up or down */
+       if (up) {
+               ifp->if_flags |= IFF_UP;
+       }
+       else {
+               ifp->if_flags &= ~IFF_UP;
+       }
+       
+       ifnet_touch_lastchange(ifp);
+       
+       /* Drop the lock to notify addresses and route */
+       ifnet_lock_done(ifp);
+       if (ifnet_get_address_list(ifp, &ifa) == 0) {
+               for (i = 0; ifa[i] != 0; i++) {
+                       pfctlinput(up ? PRC_IFUP : PRC_IFDOWN, ifa[i]->ifa_addr);
+               }
+               ifnet_free_address_list(ifa);
+       }
+       rt_ifmsg(ifp);
+       
+       /* Aquire the lock to clear the changing flag and flush the send queue */
+       ifnet_lock_exclusive(ifp);
+       if (!up)
+               if_qflush(&ifp->if_snd);
+       ifp->if_eflags &= ~IFEF_UPDOWNCHANGE;
+       wakeup(&ifp->if_eflags);
+       
+       return;
+}
+
+/*
+ * Mark an interface down and notify protocols of
+ * the transition.
+ */
+void
+if_down(
+       struct ifnet *ifp)
+{
+       ifnet_lock_exclusive(ifp);
+       if_updown(ifp, 0);
+       ifnet_lock_done(ifp);
+}
+
+/*
+ * Mark an interface up and notify protocols of
+ * the transition.
+ */
+void
+if_up(
+       struct ifnet *ifp)
+{
+       ifnet_lock_exclusive(ifp);
+       if_updown(ifp, 1);
+       ifnet_lock_done(ifp);
+}
+
+/*
+ * Flush an interface queue.
+ */
+static void
+if_qflush(struct ifqueue *ifq)
+{
+       struct mbuf *m, *n;
 
        n = ifq->ifq_head;
        while ((m = n) != 0) {
                n = m->m_act;
                m_freem(m);
        }
-       ifq->ifq_head = 0;
-       ifq->ifq_tail = 0;
+       ifq->ifq_head = NULL;
+       ifq->ifq_tail = NULL;
        ifq->ifq_len = 0;
 }
 
@@ -565,19 +1329,19 @@ ifunit(const char *name)
 
        len = strlen(name);
        if (len < 2 || len > IFNAMSIZ)
-               return NULL;
+               return (NULL);
        cp = name + len - 1;
        c = *cp;
        if (c < '0' || c > '9')
-               return NULL;            /* trailing garbage */
+               return (NULL);          /* trailing garbage */
        unit = 0;
        m = 1;
        do {
                if (cp == name)
-                       return NULL;    /* no interface name */
+                       return (NULL);  /* no interface name */
                unit += (c - '0') * m;
                if (unit > 1000000)
-                       return NULL;    /* number is unreasonable */
+                       return (NULL);  /* number is unreasonable */
                m *= 10;
                c = *--cp;
        } while (c >= '0' && c <= '9');
@@ -587,12 +1351,14 @@ ifunit(const char *name)
        /*
         * Now search all the interfaces for this name/number
         */
-       for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next) {
-               if (strcmp(ifp->if_name, namebuf))
+       ifnet_head_lock_shared();
+       TAILQ_FOREACH(ifp, &ifnet_head, if_link) {
+               if (strncmp(ifp->if_name, namebuf, len))
                        continue;
                if (unit == ifp->if_unit)
                        break;
        }
+       ifnet_head_done();
        return (ifp);
 }
 
@@ -602,15 +1368,14 @@ ifunit(const char *name)
  * interface structure pointer.
  */
 struct ifnet *
-if_withname(sa)
-       struct sockaddr *sa;
+if_withname(struct sockaddr *sa)
 {
        char ifname[IFNAMSIZ+1];
        struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
 
        if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
             (sdl->sdl_nlen > IFNAMSIZ) )
-               return NULL;
+               return (NULL);
 
        /*
         * ifunit wants a null-terminated name.  It may not be null-terminated
@@ -621,7 +1386,7 @@ if_withname(sa)
 
        bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
        ifname[sdl->sdl_nlen] = '\0';
-       return ifunit(ifname);
+       return (ifunit(ifname));
 }
 
 
@@ -629,272 +1394,390 @@ if_withname(sa)
  * Interface ioctls.
  */
 int
-ifioctl(so, cmd, data, p)
-       struct socket *so;
-       u_long cmd;
-       caddr_t data;
-       struct proc *p;
-{
-       register struct ifnet *ifp;
-       register struct ifreq *ifr;
+ifioctl(struct socket *so, u_long cmd, caddr_t data, struct proc *p)
+{
+       struct ifnet *ifp;
+       struct ifreq *ifr;
        struct ifstat *ifs;
        int error = 0;
        short oif_flags;
        struct kev_msg        ev_msg;
        struct net_event_data ev_data;
 
+       bzero(&ev_data, sizeof(struct net_event_data));
+       bzero(&ev_msg, sizeof(struct kev_msg));
        switch (cmd) {
-
-       case SIOCGIFCONF:
-       case OSIOCGIFCONF:
-               return (ifconf(cmd, data));
+       case OSIOCGIFCONF32:
+       case SIOCGIFCONF32: {
+               struct ifconf32 *ifc = (struct ifconf32 *)data;
+               return (ifconf(cmd, CAST_USER_ADDR_T(ifc->ifc_req),
+                   &ifc->ifc_len));
+               /* NOTREACHED */
+       }
+       case SIOCGIFCONF64:
+       case OSIOCGIFCONF64: {
+               struct ifconf64 *ifc = (struct ifconf64 *)data;
+               return (ifconf(cmd, ifc->ifc_req, &ifc->ifc_len));
+               /* NOTREACHED */
+       }
        }
        ifr = (struct ifreq *)data;
+       switch (cmd) {
+       case SIOCIFCREATE:
+       case SIOCIFCREATE2:
+                error = proc_suser(p);
+                if (error)
+                        return (error);
+                return if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
+                        cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
+       case SIOCIFDESTROY:
+               error = proc_suser(p);
+               if (error)
+                       return (error);
+               return if_clone_destroy(ifr->ifr_name);
+#if IF_CLONE_LIST
+       case SIOCIFGCLONERS32: {
+               struct if_clonereq32 *ifcr = (struct if_clonereq32 *)data;
+               return (if_clone_list(ifcr->ifcr_count, &ifcr->ifcr_total,
+                   CAST_USER_ADDR_T(ifcr->ifcru_buffer)));
+               /* NOTREACHED */
+
+       }
+       case SIOCIFGCLONERS64: {
+               struct if_clonereq64 *ifcr = (struct if_clonereq64 *)data;
+               return (if_clone_list(ifcr->ifcr_count, &ifcr->ifcr_total,
+                   ifcr->ifcru_buffer));
+               /* NOTREACHED */
+           }
+#endif /* IF_CLONE_LIST */
+       }
+
+       /*
+        * ioctls which require ifp.  Note that we acquire dlil_ifnet_lock
+        * here to ensure that the ifnet, if found, has been fully attached.
+        */
+       dlil_if_lock();
        ifp = ifunit(ifr->ifr_name);
-       if (ifp == 0)
+       dlil_if_unlock();
+       if (ifp == NULL)
                return (ENXIO);
-       switch (cmd) {
 
+       switch (cmd) {
        case SIOCGIFFLAGS:
+               ifnet_lock_shared(ifp);
                ifr->ifr_flags = ifp->if_flags;
+               ifnet_lock_done(ifp);
                break;
 
+       case SIOCGIFCAP:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_reqcap = ifp->if_capabilities;
+               ifr->ifr_curcap = ifp->if_capenable;
+               ifnet_lock_done(ifp);
+               break;
+
+#if CONFIG_MACF_NET
+       case SIOCGIFMAC:
+               error = mac_ifnet_label_get(kauth_cred_get(), ifr, ifp);
+               break;
+#endif
        case SIOCGIFMETRIC:
+               ifnet_lock_shared(ifp);
                ifr->ifr_metric = ifp->if_metric;
+               ifnet_lock_done(ifp);
                break;
 
        case SIOCGIFMTU:
+               ifnet_lock_shared(ifp);
                ifr->ifr_mtu = ifp->if_mtu;
+               ifnet_lock_done(ifp);
                break;
 
        case SIOCGIFPHYS:
+               ifnet_lock_shared(ifp);
                ifr->ifr_phys = ifp->if_physical;
+               ifnet_lock_done(ifp);
                break;
 
        case SIOCSIFFLAGS:
-               error = suser(p->p_ucred, &p->p_acflag);
-               if (error)
-                       return (error);
-#ifndef __APPLE__
-               if (ifp->if_flags & IFF_SMART) {
-                       /* Smart drivers twiddle their own routes */
-               } else
-#endif
-               if (ifp->if_flags & IFF_UP &&
-                   (ifr->ifr_flags & IFF_UP) == 0) {
-                       int s = splimp();
-                       if_down(ifp);
-                       splx(s);
-               } else if (ifr->ifr_flags & IFF_UP &&
-                   (ifp->if_flags & IFF_UP) == 0) {
-                       int s = splimp();
-                       if_up(ifp);
-                       splx(s);
-               }
-               ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
-                       (ifr->ifr_flags &~ IFF_CANTCHANGE);
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
-               error = dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                  ifp, cmd, (caddr_t) data);
+               (void) ifnet_set_flags(ifp, ifr->ifr_flags,
+                   (u_int16_t)~IFF_CANTCHANGE);
 
-               if (error == 0) {
-                        ev_msg.vendor_code    = KEV_VENDOR_APPLE;
-                        ev_msg.kev_class      = KEV_NETWORK_CLASS;
-                        ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
-
-                        ev_msg.event_code = KEV_DL_SIFFLAGS;
-                        strncpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
-                        ev_data.if_family = ifp->if_family;
-                        ev_data.if_unit   = (unsigned long) ifp->if_unit;
-                        ev_msg.dv[0].data_length = sizeof(struct net_event_data);
-                        ev_msg.dv[0].data_ptr    = &ev_data;
-                        ev_msg.dv[1].data_length = 0;
-                        kev_post_msg(&ev_msg);
+               /*
+                * Note that we intentionally ignore any error from below
+                * for the SIOCSIFFLAGS case.
+                */
+               (void) ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+
+               /*
+                * Send the event even upon error from the driver because
+                * we changed the flags.
+                */
+               ev_msg.vendor_code    = KEV_VENDOR_APPLE;
+               ev_msg.kev_class      = KEV_NETWORK_CLASS;
+               ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
+
+               ev_msg.event_code = KEV_DL_SIFFLAGS;
+               strlcpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
+               ev_data.if_family = ifp->if_family;
+               ev_data.if_unit   = (u_int32_t) ifp->if_unit;
+               ev_msg.dv[0].data_length = sizeof(struct net_event_data);
+               ev_msg.dv[0].data_ptr    = &ev_data;
+               ev_msg.dv[1].data_length = 0;
+               kev_post_msg(&ev_msg);
+
+               ifnet_touch_lastchange(ifp);
+               break;
+
+       case SIOCSIFCAP:
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
+
+               if ((ifr->ifr_reqcap & ~ifp->if_capabilities)) {
+                       error = EINVAL;
+                       break;
                }
-               getmicrotime(&ifp->if_lastchange);
+               error = ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+
+               ifnet_touch_lastchange(ifp);
                break;
 
+#if CONFIG_MACF_NET
+       case SIOCSIFMAC:
+               error = mac_ifnet_label_set(kauth_cred_get(), ifr, ifp);
+               break;
+#endif
        case SIOCSIFMETRIC:
-               error = suser(p->p_ucred, &p->p_acflag);
-               if (error)
-                       return (error);
-               ifp->if_metric = ifr->ifr_metric;
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
+               ifp->if_metric = ifr->ifr_metric;
 
                ev_msg.vendor_code    = KEV_VENDOR_APPLE;
                ev_msg.kev_class      = KEV_NETWORK_CLASS;
                ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
-       
+
                ev_msg.event_code = KEV_DL_SIFMETRICS;
-               strncpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
+               strlcpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
+               ev_data.if_family = ifp->if_family;
+               ev_data.if_unit   = (u_int32_t) ifp->if_unit;
+               ev_msg.dv[0].data_length = sizeof(struct net_event_data);
+               ev_msg.dv[0].data_ptr    = &ev_data;
+
+               ev_msg.dv[1].data_length = 0;
+               kev_post_msg(&ev_msg);
+
+               ifnet_touch_lastchange(ifp);
+               break;
+
+       case SIOCSIFPHYS:
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
+
+               error = ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+               if (error != 0)
+                       break;
+
+               ev_msg.vendor_code    = KEV_VENDOR_APPLE;
+               ev_msg.kev_class      = KEV_NETWORK_CLASS;
+               ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
+
+               ev_msg.event_code = KEV_DL_SIFPHYS;
+               strlcpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
                ev_data.if_family = ifp->if_family;
-               ev_data.if_unit   = (unsigned long) ifp->if_unit;
+               ev_data.if_unit   = (u_int32_t) ifp->if_unit;
                ev_msg.dv[0].data_length = sizeof(struct net_event_data);
                ev_msg.dv[0].data_ptr    = &ev_data;
-
                ev_msg.dv[1].data_length = 0;
                kev_post_msg(&ev_msg);
 
-               getmicrotime(&ifp->if_lastchange);
+               ifnet_touch_lastchange(ifp);
                break;
 
-       case SIOCSIFPHYS:
-               error = suser(p->p_ucred, &p->p_acflag);
-               if (error)
-                       return error;
+       case SIOCSIFMTU:
+       {
+               u_int32_t oldmtu = ifp->if_mtu;
 
-               error = dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                       ifp, cmd, (caddr_t) data);
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
-               if (error == 0) {
-                       ev_msg.vendor_code    = KEV_VENDOR_APPLE;
-                       ev_msg.kev_class      = KEV_NETWORK_CLASS;
-                       ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
-
-                       ev_msg.event_code = KEV_DL_SIFPHYS;
-                       strncpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
-                       ev_data.if_family = ifp->if_family;
-                       ev_data.if_unit   = (unsigned long) ifp->if_unit;
-                       ev_msg.dv[0].data_length = sizeof(struct net_event_data);
-                       ev_msg.dv[0].data_ptr    = &ev_data;
-                       ev_msg.dv[1].data_length = 0;
-                       kev_post_msg(&ev_msg);
-
-                       getmicrotime(&ifp->if_lastchange);
+               if (ifp->if_ioctl == NULL) {
+                       error = EOPNOTSUPP;
+                       break;
                }
-               return(error);
+               if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
+                       error = EINVAL;
+                       break;
+               }
+               error = ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+               if (error != 0)
+                       break;
 
-       case SIOCSIFMTU:
-       {
-               u_long oldmtu = ifp->if_mtu;
+               ev_msg.vendor_code    = KEV_VENDOR_APPLE;
+               ev_msg.kev_class      = KEV_NETWORK_CLASS;
+               ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
 
-               error = suser(p->p_ucred, &p->p_acflag);
-               if (error)
-                       return (error);
-               if (ifp->if_ioctl == NULL)
-                       return (EOPNOTSUPP);
-               if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
-                       return (EINVAL);
+               ev_msg.event_code = KEV_DL_SIFMTU;
+               strlcpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
+               ev_data.if_family = ifp->if_family;
+               ev_data.if_unit   = (u_int32_t) ifp->if_unit;
+               ev_msg.dv[0].data_length = sizeof(struct net_event_data);
+               ev_msg.dv[0].data_ptr    = &ev_data;
+               ev_msg.dv[1].data_length = 0;
+               kev_post_msg(&ev_msg);
 
-               error = dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                  ifp, cmd, (caddr_t) data);
+               ifnet_touch_lastchange(ifp);
+               rt_ifmsg(ifp);
 
-               if (error == 0) {
-                    ev_msg.vendor_code    = KEV_VENDOR_APPLE;
-                    ev_msg.kev_class      = KEV_NETWORK_CLASS;
-                    ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
-       
-                    ev_msg.event_code = KEV_DL_SIFMTU;
-                    strncpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
-                    ev_data.if_family = ifp->if_family;
-                    ev_data.if_unit   = (unsigned long) ifp->if_unit;
-                    ev_msg.dv[0].data_length = sizeof(struct net_event_data);
-                    ev_msg.dv[0].data_ptr    = &ev_data;
-                    ev_msg.dv[1].data_length = 0;
-                    kev_post_msg(&ev_msg);
-
-                       getmicrotime(&ifp->if_lastchange);
-                       rt_ifmsg(ifp);
-               }
                /*
-                * If the link MTU changed, do network layer specific procedure.
+                * If the link MTU changed, do network layer specific procedure
+                * and update all route entries associated with the interface,
+                * so that their MTU metric gets updated.
                 */
                if (ifp->if_mtu != oldmtu) {
+                       if_rtmtu_update(ifp);
 #if INET6
                        nd6_setmtu(ifp);
 #endif
                }
-               return (error);
+               break;
        }
 
        case SIOCADDMULTI:
        case SIOCDELMULTI:
-               error = suser(p->p_ucred, &p->p_acflag);
-               if (error)
-                       return (error);
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
                /* Don't allow group membership on non-multicast interfaces. */
-               if ((ifp->if_flags & IFF_MULTICAST) == 0)
-                       return EOPNOTSUPP;
+               if ((ifp->if_flags & IFF_MULTICAST) == 0) {
+                       error = EOPNOTSUPP;
+                       break;
+               }
 
-#ifndef __APPLE__
                /* Don't let users screw up protocols' entries. */
-               if (ifr->ifr_addr.sa_family != AF_LINK)
-                       return EINVAL;
-#endif
+               if (ifr->ifr_addr.sa_family != AF_UNSPEC &&
+                   ifr->ifr_addr.sa_family != AF_LINK) {
+                       error = EINVAL;
+                       break;
+               }
 
+               /*
+                * User is permitted to anonymously join a particular link
+                * multicast group via SIOCADDMULTI.  Subsequent join requested
+                * for the same record which has an outstanding refcnt from a
+                * past if_addmulti_anon() will not result in EADDRINUSE error
+                * (unlike other BSDs.)  Anonymously leaving a group is also
+                * allowed only as long as there is an outstanding refcnt held
+                * by a previous anonymous request, or else ENOENT (even if the
+                * link-layer multicast membership exists for a network-layer
+                * membership.)
+                */
                if (cmd == SIOCADDMULTI) {
-                       struct ifmultiaddr *ifma;
-                       error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
+                       error = if_addmulti_anon(ifp, &ifr->ifr_addr, NULL);
                        ev_msg.event_code = KEV_DL_ADDMULTI;
                } else {
-                       error = if_delmulti(ifp, &ifr->ifr_addr);
+                       error = if_delmulti_anon(ifp, &ifr->ifr_addr);
                        ev_msg.event_code = KEV_DL_DELMULTI;
                }
-               if (error == 0) {
-                    ev_msg.vendor_code    = KEV_VENDOR_APPLE;
-                    ev_msg.kev_class      = KEV_NETWORK_CLASS;
-                    ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
-                    strncpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
-       
-                    ev_data.if_family = ifp->if_family;
-                    ev_data.if_unit   = (unsigned long) ifp->if_unit;
-                    ev_msg.dv[0].data_length = sizeof(struct net_event_data);
-                    ev_msg.dv[0].data_ptr    = &ev_data;
-                    ev_msg.dv[1].data_length = 0;
-                    kev_post_msg(&ev_msg);
-
-                    getmicrotime(&ifp->if_lastchange);
-               }
-               return error;
+               if (error != 0)
+                       break;
+
+               ev_msg.vendor_code    = KEV_VENDOR_APPLE;
+               ev_msg.kev_class      = KEV_NETWORK_CLASS;
+               ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
+               strlcpy(&ev_data.if_name[0], ifp->if_name, IFNAMSIZ);
+
+               ev_data.if_family = ifp->if_family;
+               ev_data.if_unit   = (u_int32_t) ifp->if_unit;
+               ev_msg.dv[0].data_length = sizeof(struct net_event_data);
+               ev_msg.dv[0].data_ptr    = &ev_data;
+               ev_msg.dv[1].data_length = 0;
+               kev_post_msg(&ev_msg);
+
+               ifnet_touch_lastchange(ifp);
+               break;
 
        case SIOCSIFPHYADDR:
        case SIOCDIFPHYADDR:
-#ifdef INET6
-       case SIOCSIFPHYADDR_IN6:
+#if INET6
+       case SIOCSIFPHYADDR_IN6_32:
+       case SIOCSIFPHYADDR_IN6_64:
 #endif
        case SIOCSLIFPHYADDR:
        case SIOCSIFMEDIA:
        case SIOCSIFGENERIC:
        case SIOCSIFLLADDR:
-               error = suser(p->p_ucred, &p->p_acflag);
-               if (error)
-                       return (error);
+       case SIOCSIFALTMTU:
+       case SIOCSIFVLAN:
+       case SIOCSIFBOND:
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
-               error = dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                  ifp, cmd, (caddr_t) data);
+               error = ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+               if (error != 0)
+                       break;
 
-               if (error == 0)
-                       getmicrotime(&ifp->if_lastchange);
-               return error;
+               ifnet_touch_lastchange(ifp);
+               break;
 
        case SIOCGIFSTATUS:
                ifs = (struct ifstat *)data;
                ifs->ascii[0] = '\0';
-               
+
        case SIOCGIFPSRCADDR:
        case SIOCGIFPDSTADDR:
        case SIOCGLIFPHYADDR:
-       case SIOCGIFMEDIA:
+       case SIOCGIFMEDIA32:
+       case SIOCGIFMEDIA64:
        case SIOCGIFGENERIC:
+       case SIOCGIFDEVMTU:
+               error = ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+               break;
+
+       case SIOCGIFVLAN:
+       case SIOCGIFBOND:
+               error = ifnet_ioctl(ifp, so->so_proto->pr_domain->dom_family,
+                   cmd, data);
+               break;
+
+       case SIOCGIFWAKEFLAGS:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_wake_flags = ifnet_get_wake_flags(ifp);
+               ifnet_lock_done(ifp);
+               break;
 
-               return dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                  ifp, cmd, (caddr_t) data);
+       case SIOCGIFGETRTREFCNT:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_route_refcnt = ifp->if_route_refcnt;
+               ifnet_lock_done(ifp);
+               break;
 
        default:
                oif_flags = ifp->if_flags;
-               if (so->so_proto == 0)
-                       return (EOPNOTSUPP);
-#if !COMPAT_43
-               return ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
-                                                                data,
-                                                                ifp, p));
-#else
+               if (so->so_proto == NULL) {
+                       error = EOPNOTSUPP;
+                       break;
+               }
            {
-               int ocmd = cmd;
+               u_long ocmd = cmd;
 
                switch (cmd) {
-
                case SIOCSIFDSTADDR:
                case SIOCSIFADDR:
                case SIOCSIFBRDADDR:
@@ -926,12 +1809,13 @@ ifioctl(so, cmd, data, p)
                case OSIOCGIFNETMASK:
                        cmd = SIOCGIFNETMASK;
                }
-               error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
-                                                                  cmd,
-                                                                  data,
-                                                                  ifp, p));
-               switch (ocmd) {
 
+               socket_lock(so, 1);
+               error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
+                   data, ifp, p));
+               socket_unlock(so, 1);
+
+               switch (ocmd) {
                case OSIOCGIFADDR:
                case OSIOCGIFDSTADDR:
                case OSIOCGIFBRDADDR:
@@ -940,58 +1824,81 @@ ifioctl(so, cmd, data, p)
 
                }
            }
-#endif /* COMPAT_43 */
+               if (cmd == SIOCSIFKPI) {
+                       int temperr = proc_suser(p);
+                       if (temperr != 0)
+                               error = temperr;
+               }
 
-               if (error == EOPNOTSUPP)
-                       error = dlil_ioctl(so->so_proto->pr_domain->dom_family,
-                                                               ifp, cmd, (caddr_t) data);
+               if (error == EOPNOTSUPP || error == ENOTSUP)
+                       error = ifnet_ioctl(ifp,
+                           so->so_proto->pr_domain->dom_family, cmd, data);
 
-               return (error);
+               break;
        }
-       return (0);
+       return (error);
 }
 
+int
+ifioctllocked(struct socket *so, u_long cmd, caddr_t data, struct proc *p)
+{
+       int error;
+
+       socket_unlock(so, 0);
+       error = ifioctl(so, cmd, data, p);
+       socket_lock(so, 0);
+       return(error);
+}
+       
 /*
  * Set/clear promiscuous mode on interface ifp based on the truth value
  * of pswitch.  The calls are reference counted so that only the first
  * "on" request actually has an effect, as does the final "off" request.
  * Results are undefined if the "off" and "on" requests are not matched.
  */
-int
-ifpromisc(ifp, pswitch)
-       struct ifnet *ifp;
-       int pswitch;
+errno_t
+ifnet_set_promiscuous(
+       ifnet_t ifp,
+       int pswitch)
 {
-       struct ifreq ifr;
-       int error;
-       int oldflags;
+       int error = 0;
+       int oldflags = 0;
+       int newflags = 0;
 
+       ifnet_lock_exclusive(ifp);
        oldflags = ifp->if_flags;
-       if (pswitch) {
-               /*
-                * If the device is not configured up, we cannot put it in
-                * promiscuous mode.
-                */
-               if ((ifp->if_flags & IFF_UP) == 0)
-                       return (ENETDOWN);
-               if (ifp->if_pcount++ != 0)
-                       return (0);
+       ifp->if_pcount += pswitch ? 1 : -1;
+       
+       if (ifp->if_pcount > 0)
                ifp->if_flags |= IFF_PROMISC;
-               log(LOG_INFO, "%s%d: promiscuous mode enabled\n",
-                   ifp->if_name, ifp->if_unit);
-       } else {
-               if (--ifp->if_pcount > 0)
-                       return (0);
+       else
                ifp->if_flags &= ~IFF_PROMISC;
-               log(LOG_INFO, "%s%d: promiscuous mode disabled\n",
-                   ifp->if_name, ifp->if_unit);
+       
+       newflags = ifp->if_flags;
+       ifnet_lock_done(ifp);
+       
+       if (newflags != oldflags && (newflags & IFF_UP) != 0) {
+               error = ifnet_ioctl(ifp, 0, SIOCSIFFLAGS, NULL);
+               if (error == 0) {
+                       rt_ifmsg(ifp);
+               } else {
+                       ifnet_lock_exclusive(ifp);
+                       // revert the flags
+                       ifp->if_pcount -= pswitch ? 1 : -1;
+                       if (ifp->if_pcount > 0)
+                           ifp->if_flags |= IFF_PROMISC;
+                       else
+                           ifp->if_flags &= ~IFF_PROMISC;
+                       ifnet_lock_done(ifp);
+               }
+       }
+       
+       if (newflags != oldflags) {
+               log(LOG_INFO, "%s%d: promiscuous mode %s%s\n",
+                   ifp->if_name, ifp->if_unit,
+                   (newflags & IFF_PROMISC) != 0 ? "enable" : "disable",
+                   error != 0 ? " failed" : " succeeded");
        }
-       ifr.ifr_flags = ifp->if_flags;
-       error = dlil_ioctl(0, ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
-       if (error == 0)
-               rt_ifmsg(ifp);
-       else
-               ifp->if_flags = oldflags;
        return error;
 }
 
@@ -1003,20 +1910,26 @@ ifpromisc(ifp, pswitch)
  */
 /*ARGSUSED*/
 static int
-ifconf(cmd, data)
-       u_long cmd;
-       caddr_t data;
-{
-       register struct ifconf *ifc = (struct ifconf *)data;
-       register struct ifnet *ifp = ifnet.tqh_first;
-       register struct ifaddr *ifa;
-       struct ifreq ifr, *ifrp;
-       int space = ifc->ifc_len, error = 0;
-
-       ifrp = ifc->ifc_req;
-       for (; space > sizeof (ifr) && ifp; ifp = ifp->if_link.tqe_next) {
+ifconf(u_long cmd, user_addr_t ifrp, int * ret_space)
+{
+       struct ifnet *ifp = NULL;
+       struct ifaddr *ifa;
+       struct ifreq ifr;
+       int error = 0;
+       size_t space;
+
+       /*
+        * Zero the ifr buffer to make sure we don't
+        * disclose the contents of the stack.
+        */
+       bzero(&ifr, sizeof(struct ifreq));
+
+       space = *ret_space;
+       ifnet_head_lock_shared();
+       for (ifp = ifnet_head.tqh_first; space > sizeof(ifr) &&
+           ifp; ifp = ifp->if_link.tqe_next) {
                char workbuf[64];
-               int ifnlen, addrs;
+               size_t ifnlen, addrs;
 
                ifnlen = snprintf(workbuf, sizeof(workbuf),
                    "%s%d", ifp->if_name, ifp->if_unit);
@@ -1024,276 +1937,799 @@ ifconf(cmd, data)
                        error = ENAMETOOLONG;
                        break;
                } else {
-                       strcpy(ifr.ifr_name, workbuf);
+                       strlcpy(ifr.ifr_name, workbuf, IFNAMSIZ);
+               }
+
+               ifnet_lock_shared(ifp);
+
+               addrs = 0;
+               ifa = ifp->if_addrhead.tqh_first;
+               for ( ; space > sizeof (ifr) && ifa;
+                   ifa = ifa->ifa_link.tqe_next) {
+                       struct sockaddr *sa;
+
+                       IFA_LOCK(ifa);
+                       sa = ifa->ifa_addr;
+#ifndef __APPLE__
+                       if (curproc->p_prison && prison_if(curproc, sa)) {
+                               IFA_UNLOCK(ifa);
+                               continue;
+                       }
+#endif
+                       addrs++;
+                       if (cmd == OSIOCGIFCONF32 || cmd == OSIOCGIFCONF64) {
+                               struct osockaddr *osa =
+                                        (struct osockaddr *)&ifr.ifr_addr;
+                               ifr.ifr_addr = *sa;
+                               osa->sa_family = sa->sa_family;
+                               error = copyout((caddr_t)&ifr, ifrp,
+                                   sizeof (ifr));
+                               ifrp += sizeof(struct ifreq);
+                       } else if (sa->sa_len <= sizeof(*sa)) {
+                               ifr.ifr_addr = *sa;
+                               error = copyout((caddr_t)&ifr, ifrp,
+                                   sizeof (ifr));
+                               ifrp += sizeof(struct ifreq);
+                       } else {
+                               if (space <
+                                   sizeof (ifr) + sa->sa_len - sizeof(*sa)) {
+                                       IFA_UNLOCK(ifa);
+                                       break;
+                               }
+                               space -= sa->sa_len - sizeof(*sa);
+                               error = copyout((caddr_t)&ifr, ifrp,
+                                   sizeof (ifr.ifr_name));
+                               if (error == 0) {
+                                   error = copyout((caddr_t)sa, (ifrp +
+                                       offsetof(struct ifreq, ifr_addr)),
+                                       sa->sa_len);
+                               }
+                               ifrp += (sa->sa_len + offsetof(struct ifreq,
+                                   ifr_addr));
+                       }
+                       IFA_UNLOCK(ifa);
+                       if (error)
+                               break;
+                       space -= sizeof (ifr);
+               }
+               ifnet_lock_done(ifp);
+
+               if (error)
+                       break;
+               if (!addrs) {
+                       bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
+                       error = copyout((caddr_t)&ifr, ifrp, sizeof (ifr));
+                       if (error)
+                               break;
+                       space -= sizeof (ifr);
+                       ifrp += sizeof(struct ifreq);
+               }
+       }
+       ifnet_head_done();
+       *ret_space -= space;
+       return (error);
+}
+
+/*
+ * Just like if_promisc(), but for all-multicast-reception mode.
+ */
+int
+if_allmulti(struct ifnet *ifp, int onswitch)
+{
+       int error = 0;
+       int     modified = 0;
+       
+       ifnet_lock_exclusive(ifp);
+
+       if (onswitch) {
+               if (ifp->if_amcount++ == 0) {
+                       ifp->if_flags |= IFF_ALLMULTI;
+                       modified = 1;
+               }
+       } else {
+               if (ifp->if_amcount > 1) {
+                       ifp->if_amcount--;
+               } else {
+                       ifp->if_amcount = 0;
+                       ifp->if_flags &= ~IFF_ALLMULTI;
+                       modified = 1;
                }
+       }
+       ifnet_lock_done(ifp);
+       
+       if (modified)
+               error = ifnet_ioctl(ifp, 0, SIOCSIFFLAGS, NULL);
+
+       if (error == 0)
+               rt_ifmsg(ifp);
+       return error;
+}
+
+static struct ifmultiaddr *
+ifma_alloc(int how)
+{
+       struct ifmultiaddr *ifma;
+
+       ifma = (how == M_WAITOK) ? zalloc(ifma_zone) :
+           zalloc_noblock(ifma_zone);
+
+       if (ifma != NULL) {
+               bzero(ifma, ifma_size);
+               lck_mtx_init(&ifma->ifma_lock, ifa_mtx_grp, ifa_mtx_attr);
+               ifma->ifma_debug |= IFD_ALLOC;
+               if (ifma_debug != 0) {
+                       ifma->ifma_debug |= IFD_DEBUG;
+                       ifma->ifma_trace = ifma_trace;
+               }
+       }
+       return (ifma);
+}
+
+static void
+ifma_free(struct ifmultiaddr *ifma)
+{
+       IFMA_LOCK(ifma);
+
+       if (ifma->ifma_protospec != NULL) {
+               panic("%s: Protospec not NULL for ifma=%p", __func__, ifma);
+               /* NOTREACHED */
+       } else if ((ifma->ifma_flags & IFMAF_ANONYMOUS) ||
+           ifma->ifma_anoncnt != 0) {
+               panic("%s: Freeing ifma=%p with outstanding anon req",
+                   __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_debug & IFD_ATTACHED) {
+               panic("%s: ifma=%p attached to ifma_ifp=%p is being freed",
+                   __func__, ifma, ifma->ifma_ifp);
+               /* NOTREACHED */
+       } else if (!(ifma->ifma_debug & IFD_ALLOC)) {
+               panic("%s: ifma %p cannot be freed", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_refcount != 0) {
+               panic("%s: non-zero refcount ifma=%p", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_reqcnt != 0) {
+               panic("%s: non-zero reqcnt ifma=%p", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_ifp != NULL) {
+               panic("%s: non-NULL ifma_ifp=%p for ifma=%p", __func__,
+                   ifma->ifma_ifp, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_ll != NULL) {
+               panic("%s: non-NULL ifma_ll=%p for ifma=%p", __func__,
+                   ifma->ifma_ll, ifma);
+               /* NOTREACHED */
+       }
+       ifma->ifma_debug &= ~IFD_ALLOC;
+       if ((ifma->ifma_debug & (IFD_DEBUG | IFD_TRASHED)) ==
+           (IFD_DEBUG | IFD_TRASHED)) {
+               lck_mtx_lock(&ifma_trash_lock);
+               TAILQ_REMOVE(&ifma_trash_head, (struct ifmultiaddr_dbg *)ifma,
+                   ifma_trash_link);
+               lck_mtx_unlock(&ifma_trash_lock);
+               ifma->ifma_debug &= ~IFD_TRASHED;
+       }
+       IFMA_UNLOCK(ifma);
+
+       if (ifma->ifma_addr != NULL) {
+               FREE(ifma->ifma_addr, M_IFADDR);
+               ifma->ifma_addr = NULL;
+       }
+       lck_mtx_destroy(&ifma->ifma_lock, ifa_mtx_grp);
+       zfree(ifma_zone, ifma);
+}
+
+static void
+ifma_trace(struct ifmultiaddr *ifma, int refhold)
+{
+       struct ifmultiaddr_dbg *ifma_dbg = (struct ifmultiaddr_dbg *)ifma;
+       ctrace_t *tr;
+       u_int32_t idx;
+       u_int16_t *cnt;
+
+       if (!(ifma->ifma_debug & IFD_DEBUG)) {
+               panic("%s: ifma %p has no debug structure", __func__, ifma);
+               /* NOTREACHED */
+       }
+       if (refhold) {
+               cnt = &ifma_dbg->ifma_refhold_cnt;
+               tr = ifma_dbg->ifma_refhold;
+       } else {
+               cnt = &ifma_dbg->ifma_refrele_cnt;
+               tr = ifma_dbg->ifma_refrele;
+       }
+
+       idx = atomic_add_16_ov(cnt, 1) % IFMA_TRACE_HIST_SIZE;
+       ctrace_record(&tr[idx]);
+}
+
+void
+ifma_addref(struct ifmultiaddr *ifma, int locked)
+{
+       if (!locked)
+               IFMA_LOCK(ifma);
+       else
+               IFMA_LOCK_ASSERT_HELD(ifma);
+
+       if (++ifma->ifma_refcount == 0) {
+               panic("%s: ifma=%p wraparound refcnt", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_trace != NULL) {
+               (*ifma->ifma_trace)(ifma, TRUE);
+       }
+       if (!locked)
+               IFMA_UNLOCK(ifma);
+}
+
+void
+ifma_remref(struct ifmultiaddr *ifma)
+{
+       struct ifmultiaddr *ll;
+
+       IFMA_LOCK(ifma);
+
+       if (ifma->ifma_refcount == 0) {
+               panic("%s: ifma=%p negative refcnt", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_trace != NULL) {
+               (*ifma->ifma_trace)(ifma, FALSE);
+       }
+
+       --ifma->ifma_refcount;
+       if (ifma->ifma_refcount > 0) {
+               IFMA_UNLOCK(ifma);
+               return;
+       }
+
+       ll = ifma->ifma_ll;
+       ifma->ifma_ifp = NULL;
+       ifma->ifma_ll = NULL;
+       IFMA_UNLOCK(ifma);
+       ifma_free(ifma);        /* deallocate it */
+
+       if (ll != NULL)
+               IFMA_REMREF(ll);
+}
+
+static void
+if_attach_ifma(struct ifnet *ifp, struct ifmultiaddr *ifma, int anon)
+{
+       ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_EXCLUSIVE);
+       IFMA_LOCK_ASSERT_HELD(ifma);
+
+       if (ifma->ifma_ifp != ifp) {
+               panic("%s: Mismatch ifma_ifp=%p != ifp=%p", __func__,
+                   ifma->ifma_ifp, ifp);
+               /* NOTREACHED */
+       } else if (ifma->ifma_debug & IFD_ATTACHED) {
+               panic("%s: Attempt to attach an already attached ifma=%p",
+                   __func__, ifma);
+               /* NOTREACHED */
+       } else if (anon && (ifma->ifma_flags & IFMAF_ANONYMOUS)) {
+               panic("%s: ifma=%p unexpected IFMAF_ANONYMOUS", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_debug & IFD_TRASHED) {
+               panic("%s: Attempt to reattach a detached ifma=%p",
+                   __func__, ifma);
+               /* NOTREACHED */
+       }
+
+       ifma->ifma_reqcnt++;
+       VERIFY(ifma->ifma_reqcnt == 1);
+       IFMA_ADDREF_LOCKED(ifma);
+       ifma->ifma_debug |= IFD_ATTACHED;
+       if (anon) {
+               ifma->ifma_anoncnt++;
+               VERIFY(ifma->ifma_anoncnt == 1);
+               ifma->ifma_flags |= IFMAF_ANONYMOUS;
+       }
+
+       LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
+}
+
+static int
+if_detach_ifma(struct ifnet *ifp, struct ifmultiaddr *ifma, int anon)
+{
+       ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_EXCLUSIVE);
+       IFMA_LOCK_ASSERT_HELD(ifma);
+
+       if (ifma->ifma_reqcnt == 0) {
+               panic("%s: ifma=%p negative reqcnt", __func__, ifma);
+               /* NOTREACHED */
+       } else if (anon && !(ifma->ifma_flags & IFMAF_ANONYMOUS)) {
+               panic("%s: ifma=%p missing IFMAF_ANONYMOUS", __func__, ifma);
+               /* NOTREACHED */
+       } else if (anon && ifma->ifma_anoncnt == 0) {
+               panic("%s: ifma=%p negative anonreqcnt", __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_ifp != ifp) {
+               panic("%s: Mismatch ifma_ifp=%p, ifp=%p", __func__,
+                   ifma->ifma_ifp, ifp);
+               /* NOTREACHED */
+       }
+
+       if (anon) {
+               --ifma->ifma_anoncnt;
+               if (ifma->ifma_anoncnt > 0)
+                       return (0);
+               ifma->ifma_flags &= ~IFMAF_ANONYMOUS;
+       }
+
+       --ifma->ifma_reqcnt;
+       if (ifma->ifma_reqcnt > 0)
+               return (0);
+
+       if (ifma->ifma_protospec != NULL) {
+               panic("%s: Protospec not NULL for ifma=%p", __func__, ifma);
+               /* NOTREACHED */
+       } else if ((ifma->ifma_flags & IFMAF_ANONYMOUS) ||
+           ifma->ifma_anoncnt != 0) {
+               panic("%s: Detaching ifma=%p with outstanding anon req",
+                   __func__, ifma);
+               /* NOTREACHED */
+       } else if (!(ifma->ifma_debug & IFD_ATTACHED)) {
+               panic("%s: Attempt to detach an unattached address ifma=%p",
+                   __func__, ifma);
+               /* NOTREACHED */
+       } else if (ifma->ifma_debug & IFD_TRASHED) {
+               panic("%s: ifma %p is already in trash list", __func__, ifma);
+               /* NOTREACHED */
+       }
+
+       /*
+        * NOTE: Caller calls IFMA_REMREF
+        */
+       ifma->ifma_debug &= ~IFD_ATTACHED;
+       LIST_REMOVE(ifma, ifma_link);
+       if (LIST_EMPTY(&ifp->if_multiaddrs))
+               ifp->if_updatemcasts = 0;
+
+       if (ifma->ifma_debug & IFD_DEBUG) {
+               /* Become a regular mutex, just in case */
+               IFMA_CONVERT_LOCK(ifma);
+               lck_mtx_lock(&ifma_trash_lock);
+               TAILQ_INSERT_TAIL(&ifma_trash_head,
+                   (struct ifmultiaddr_dbg *)ifma, ifma_trash_link);
+               lck_mtx_unlock(&ifma_trash_lock);
+               ifma->ifma_debug |= IFD_TRASHED;
+       }
+
+       return (1);
+}
+
+/*
+ * Find an ifmultiaddr that matches a socket address on an interface. 
+ *
+ * Caller is responsible for holding the ifnet_lock while calling
+ * this function.
+ */
+static int
+if_addmulti_doesexist(struct ifnet *ifp, const struct sockaddr *sa,
+    struct ifmultiaddr **retifma, int anon)
+{
+       struct ifmultiaddr *ifma;
 
-               addrs = 0;
-               ifa = ifp->if_addrhead.tqh_first;
-               for ( ; space > sizeof (ifr) && ifa;
-                   ifa = ifa->ifa_link.tqe_next) {
-                       register struct sockaddr *sa = ifa->ifa_addr;
-#ifndef __APPLE__
-                       if (curproc->p_prison && prison_if(curproc, sa))
-                               continue;
-#endif
-                       addrs++;
-#ifdef COMPAT_43
-                       if (cmd == OSIOCGIFCONF) {
-                               struct osockaddr *osa =
-                                        (struct osockaddr *)&ifr.ifr_addr;
-                               ifr.ifr_addr = *sa;
-                               osa->sa_family = sa->sa_family;
-                               error = copyout((caddr_t)&ifr, (caddr_t)ifrp,
-                                               sizeof (ifr));
-                               ifrp++;
-                       } else
-#endif
-                       if (sa->sa_len <= sizeof(*sa)) {
-                               ifr.ifr_addr = *sa;
-                               error = copyout((caddr_t)&ifr, (caddr_t)ifrp,
-                                               sizeof (ifr));
-                               ifrp++;
-                       } else {
-                               if (space < sizeof (ifr) + sa->sa_len -
-                                           sizeof(*sa))
-                                       break;
-                               space -= sa->sa_len - sizeof(*sa);
-                               error = copyout((caddr_t)&ifr, (caddr_t)ifrp,
-                                               sizeof (ifr.ifr_name));
-                               if (error == 0)
-                                   error = copyout((caddr_t)sa,
-                                     (caddr_t)&ifrp->ifr_addr, sa->sa_len);
-                               ifrp = (struct ifreq *)
-                                       (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
+       for (ifma = LIST_FIRST(&ifp->if_multiaddrs); ifma != NULL;
+            ifma = LIST_NEXT(ifma, ifma_link)) {
+               IFMA_LOCK_SPIN(ifma);
+               if (!equal(sa, ifma->ifma_addr)) {
+                       IFMA_UNLOCK(ifma);
+                       continue;
+               }
+               if (anon) {
+                       VERIFY(!(ifma->ifma_flags & IFMAF_ANONYMOUS) ||
+                           ifma->ifma_anoncnt != 0);
+                       VERIFY((ifma->ifma_flags & IFMAF_ANONYMOUS) ||
+                           ifma->ifma_anoncnt == 0);
+                       ifma->ifma_anoncnt++;
+                       if (!(ifma->ifma_flags & IFMAF_ANONYMOUS)) {
+                               VERIFY(ifma->ifma_anoncnt == 1);
+                               ifma->ifma_flags |= IFMAF_ANONYMOUS;
                        }
-                       if (error)
-                               break;
-                       space -= sizeof (ifr);
                }
-               if (error)
-                       break;
-               if (!addrs) {
-                       bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
-                       error = copyout((caddr_t)&ifr, (caddr_t)ifrp,
-                           sizeof (ifr));
-                       if (error)
-                               break;
-                       space -= sizeof (ifr);
-                       ifrp++;
+               if (!anon || ifma->ifma_anoncnt == 1) {
+                       ifma->ifma_reqcnt++;
+                       VERIFY(ifma->ifma_reqcnt > 1);
                }
+               if (retifma != NULL) {
+                       *retifma = ifma;
+                       IFMA_ADDREF_LOCKED(ifma);
+               }
+               IFMA_UNLOCK(ifma);
+               return (0);
        }
-       ifc->ifc_len -= space;
-       return (error);
+       return (ENOENT);
 }
 
 /*
- * Just like if_promisc(), but for all-multicast-reception mode.
+ * Radar 3642395, make sure all multicasts are in a standard format.
  */
-int
-if_allmulti(ifp, onswitch)
-       struct ifnet *ifp;
-       int onswitch;
+static struct sockaddr*
+copy_and_normalize(
+       const struct sockaddr   *original)
 {
-       int error = 0;
-       int s = splimp();
-
-       if (onswitch) {
-               if (ifp->if_amcount++ == 0) {
-                       ifp->if_flags |= IFF_ALLMULTI;
-                       error = dlil_ioctl(0, ifp, SIOCSIFFLAGS, (caddr_t) 0);
+       int                                     alen = 0;
+       const u_char            *aptr = NULL;
+       struct sockaddr         *copy = NULL;
+       struct sockaddr_dl      *sdl_new = NULL;
+       int                                     len = 0;
+       
+       if (original->sa_family != AF_LINK &&
+               original->sa_family != AF_UNSPEC) {
+               /* Just make a copy */
+               MALLOC(copy, struct sockaddr*, original->sa_len, M_IFADDR, M_WAITOK);
+               if (copy != NULL)
+                       bcopy(original, copy, original->sa_len);
+               return copy;
+       }
+       
+       switch (original->sa_family) {
+               case AF_LINK: {
+                       const struct sockaddr_dl        *sdl_original =
+                                                                                       (const struct sockaddr_dl*)original;
+                       
+                       if (sdl_original->sdl_nlen + sdl_original->sdl_alen + sdl_original->sdl_slen +
+                               offsetof(struct sockaddr_dl, sdl_data) > sdl_original->sdl_len)
+                               return NULL;
+                       
+                       alen = sdl_original->sdl_alen;
+                       aptr = CONST_LLADDR(sdl_original);
                }
-       } else {
-               if (ifp->if_amcount > 1) {
-                       ifp->if_amcount--;
-               } else {
-                       ifp->if_amcount = 0;
-                       ifp->if_flags &= ~IFF_ALLMULTI;
-                       error = dlil_ioctl(0, ifp, SIOCSIFFLAGS, (caddr_t) 0);
+               break;
+               
+               case AF_UNSPEC: {
+                       if (original->sa_len < ETHER_ADDR_LEN +
+                               offsetof(struct sockaddr, sa_data)) {
+                               return NULL;
+                       }
+                       
+                       alen = ETHER_ADDR_LEN;
+                       aptr = (const u_char*)original->sa_data;
                }
+               break;
+       }
+       
+       if (alen == 0 || aptr == NULL)
+               return NULL;
+       
+       len = alen + offsetof(struct sockaddr_dl, sdl_data);
+       MALLOC(sdl_new, struct sockaddr_dl*, len, M_IFADDR, M_WAITOK);
+       
+       if (sdl_new != NULL) {
+               bzero(sdl_new, len);
+               sdl_new->sdl_len = len;
+               sdl_new->sdl_family = AF_LINK;
+               sdl_new->sdl_alen = alen;
+               bcopy(aptr, LLADDR(sdl_new), alen);
        }
-       splx(s);
+       
+       return (struct sockaddr*)sdl_new;
+}
 
-       if (error == 0)
-               rt_ifmsg(ifp);
-       return error;
+/*
+ * Network-layer protocol domains which hold references to the underlying
+ * link-layer record must use this routine.
+ */
+int
+if_addmulti(struct ifnet *ifp, const struct sockaddr *sa,
+    struct ifmultiaddr **retifma)
+{
+       return (if_addmulti_common(ifp, sa, retifma, 0));
 }
 
 /*
- * Add a multicast listenership to the interface in question.
- * The link layer provides a routine which converts
+ * Anything other than network-layer protocol domains which hold references
+ * to the underlying link-layer record must use this routine: SIOCADDMULTI
+ * ioctl, ifnet_add_multicast(), AppleTalk, if_bond.
  */
 int
-if_addmulti(ifp, sa, retifma)
-       struct ifnet *ifp;      /* interface to manipulate */
-       struct sockaddr *sa;    /* address to add */
-       struct ifmultiaddr **retifma;
-{
-       struct sockaddr *llsa = 0;
-       struct sockaddr *dupsa;
-       int error, s;
-       struct ifmultiaddr *ifma;
-       struct rslvmulti_req   rsreq;
+if_addmulti_anon(struct ifnet *ifp, const struct sockaddr *sa,
+    struct ifmultiaddr **retifma)
+{
+       return (if_addmulti_common(ifp, sa, retifma, 1));
+}
 
-       /*
-        * If the matching multicast address already exists
-        * then don't add a new one, just add a reference
-        */
-       for (ifma = ifp->if_multiaddrs.lh_first; ifma;
-            ifma = ifma->ifma_link.le_next) {
-               if (equal(sa, ifma->ifma_addr)) {
-                       ifma->ifma_refcount++;
-                       if (retifma)
-                               *retifma = ifma;
-                       return 0;
+/*
+ * Register an additional multicast address with a network interface.
+ *
+ * - If the address is already present, bump the reference count on the
+ *   address and return.
+ * - If the address is not link-layer, look up a link layer address.
+ * - Allocate address structures for one or both addresses, and attach to the
+ *   multicast address list on the interface.  If automatically adding a link
+ *   layer address, the protocol address will own a reference to the link
+ *   layer address, to be freed when it is freed.
+ * - Notify the network device driver of an addition to the multicast address
+ *   list.
+ *
+ * 'sa' points to caller-owned memory with the desired multicast address.
+ *
+ * 'retifma' will be used to return a pointer to the resulting multicast
+ * address reference, if desired.
+ *
+ * 'anon' indicates a link-layer address with no protocol address reference
+ * made to it.  Anything other than network-layer protocol domain requests
+ * are considered as anonymous.
+ */
+static int
+if_addmulti_common(struct ifnet *ifp, const struct sockaddr *sa,
+    struct ifmultiaddr **retifma, int anon)
+{
+       struct sockaddr_storage storage;
+       struct sockaddr *llsa = NULL;
+       struct sockaddr *dupsa = NULL;
+       int error = 0, ll_firstref = 0, lladdr;
+       struct ifmultiaddr *ifma = NULL;
+       struct ifmultiaddr *llifma = NULL;
+
+       /* Only AF_UNSPEC/AF_LINK is allowed for an "anonymous" address */
+       VERIFY(!anon || sa->sa_family == AF_UNSPEC ||
+           sa->sa_family == AF_LINK);
+
+       /* If sa is a AF_LINK or AF_UNSPEC, duplicate and normalize it */
+       if (sa->sa_family == AF_LINK || sa->sa_family == AF_UNSPEC) {
+               dupsa = copy_and_normalize(sa);
+               if (dupsa == NULL) {
+                       error = ENOMEM;
+                       goto cleanup;
                }
+               sa = dupsa;
+       }
+
+       ifnet_lock_exclusive(ifp);
+       if (!(ifp->if_flags & IFF_MULTICAST)) {
+               error = EADDRNOTAVAIL;
+               ifnet_lock_done(ifp);
+               goto cleanup;
        }
 
+       /* If the address is already present, return a new reference to it */
+       error = if_addmulti_doesexist(ifp, sa, retifma, anon);
+       ifnet_lock_done(ifp);
+       if (error == 0)
+               goto cleanup;
+
        /*
-        * Give the link layer a chance to accept/reject it, and also
-        * find out which AF_LINK address this maps to, if it isn't one
-        * already.
+        * The address isn't already present; give the link layer a chance
+        * to accept/reject it, and also find out which AF_LINK address this
+        * maps to, if it isn't one already.
         */
-       rsreq.sa = sa;
-       rsreq.llsa = &llsa;
+       error = dlil_resolve_multi(ifp, sa, (struct sockaddr *)&storage,
+           sizeof (storage));
+       if (error == 0 && storage.ss_len != 0) {
+               llsa = copy_and_normalize((struct sockaddr *)&storage);
+               if (llsa == NULL) {
+                       error = ENOMEM;
+                       goto cleanup;
+               }
+
+               llifma = ifma_alloc(M_WAITOK);
+               if (llifma == NULL) {
+                       error = ENOMEM;
+                       goto cleanup;
+               }
+       }
 
-       error = dlil_ioctl(sa->sa_family, ifp, SIOCRSLVMULTI, (caddr_t) &rsreq);
-       
        /* to be similar to FreeBSD */
        if (error == EOPNOTSUPP)
                error = 0;
+       else if (error != 0)
+               goto cleanup;
+
+       /* Allocate while we aren't holding any locks */
+       if (dupsa == NULL) {
+               dupsa = copy_and_normalize(sa);
+               if (dupsa == NULL) {
+                       error = ENOMEM;
+                       goto cleanup;
+               }
+       }
+       ifma = ifma_alloc(M_WAITOK);
+       if (ifma == NULL) {
+               error = ENOMEM;
+               goto cleanup;
+       }
+
+       ifnet_lock_exclusive(ifp);
+       /*
+        * Check again for the matching multicast.
+        */
+       error = if_addmulti_doesexist(ifp, sa, retifma, anon);
+       if (error == 0) {
+               ifnet_lock_done(ifp);
+               goto cleanup;
+       }
 
-       if (error) 
-            return error;
+       if (llifma != NULL) {
+               VERIFY(!anon);  /* must not get here if "anonymous" */
+               if (if_addmulti_doesexist(ifp, llsa, &ifma->ifma_ll, 0) == 0) {
+                       FREE(llsa, M_IFADDR);
+                       llsa = NULL;
+                       ifma_free(llifma);
+                       llifma = NULL;
+                       VERIFY(ifma->ifma_ll->ifma_ifp == ifp);
+               } else {
+                       ll_firstref = 1;
+                       llifma->ifma_addr = llsa;
+                       llifma->ifma_ifp = ifp;
+                       IFMA_LOCK(llifma);
+                       if_attach_ifma(ifp, llifma, 0);
+                       /* add extra refcnt for ifma */
+                       IFMA_ADDREF_LOCKED(llifma);
+                       IFMA_UNLOCK(llifma);
+                       ifma->ifma_ll = llifma;
+               }
+       }
 
-       MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
-       MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
-       bcopy(sa, dupsa, sa->sa_len);
+       /* "anonymous" request should not result in network address */
+       VERIFY(!anon || ifma->ifma_ll == NULL);
 
        ifma->ifma_addr = dupsa;
-       ifma->ifma_lladdr = llsa;
        ifma->ifma_ifp = ifp;
-       ifma->ifma_refcount = 1;
-       ifma->ifma_protospec = 0;
+       IFMA_LOCK(ifma);
+       if_attach_ifma(ifp, ifma, anon);
+       IFMA_ADDREF_LOCKED(ifma);               /* for this routine */
+       if (retifma != NULL) {
+               *retifma = ifma;
+               IFMA_ADDREF_LOCKED(*retifma);   /* for caller */
+       }
+       lladdr = (ifma->ifma_addr->sa_family == AF_UNSPEC ||
+           ifma->ifma_addr->sa_family == AF_LINK);
+       IFMA_UNLOCK(ifma);
+       ifnet_lock_done(ifp);
+
        rt_newmaddrmsg(RTM_NEWMADDR, ifma);
+       IFMA_REMREF(ifma);                      /* for this routine */
 
-       /*
-        * Some network interfaces can scan the address list at
-        * interrupt time; lock them out.
-        */
-       s = splimp();
-       LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
-       splx(s);
-       if (retifma)
-           *retifma = ifma;
-
-       if (llsa != 0) {
-               for (ifma = ifp->if_multiaddrs.lh_first; ifma;
-                    ifma = ifma->ifma_link.le_next) {
-                       if (equal(ifma->ifma_addr, llsa))
-                               break;
-               }
-               if (ifma) {
-                       ifma->ifma_refcount++;
-               } else {
-                       MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
-                              M_IFMADDR, M_WAITOK);
-                       MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
-                              M_IFMADDR, M_WAITOK);
-                       bcopy(llsa, dupsa, llsa->sa_len);
-                       ifma->ifma_addr = dupsa;
-                       ifma->ifma_lladdr = 0;
-                       ifma->ifma_ifp = ifp;
-                       ifma->ifma_refcount = 1;
-                       s = splimp();
-                       LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
-                       splx(s);
-               }
-       }
        /*
         * We are certain we have added something, so call down to the
-        * interface to let them know about it.
+        * interface to let them know about it.  Do this only for newly-
+        * added AF_LINK/AF_UNSPEC address in the if_multiaddrs set.
         */
-       s = splimp();
-       dlil_ioctl(0, ifp, SIOCADDMULTI, (caddr_t) 0);
-       splx(s);
+       if (lladdr || ll_firstref)
+               (void) ifnet_ioctl(ifp, 0, SIOCADDMULTI, NULL);
 
-       return 0;
+       if (ifp->if_updatemcasts > 0)
+               ifp->if_updatemcasts = 0;
+
+       return (0);
+
+cleanup:
+       if (ifma != NULL)
+               ifma_free(ifma);
+       if (dupsa != NULL)
+               FREE(dupsa, M_IFADDR);
+       if (llifma != NULL)
+               ifma_free(llifma);
+       if (llsa != NULL)
+               FREE(llsa, M_IFADDR);
+
+       return (error);
 }
 
 /*
- * Remove a reference to a multicast address on this interface.  Yell
- * if the request does not match an existing membership.
+ * Delete a multicast group membership by network-layer group address.
+ * This routine is deprecated.
  */
 int
-if_delmulti(ifp, sa)
-       struct ifnet *ifp;
-       struct sockaddr *sa;
+if_delmulti(struct ifnet *ifp, const struct sockaddr *sa)
 {
-       struct ifmultiaddr *ifma;
-       int s;
+       return (if_delmulti_common(NULL, ifp, sa, 0));
+}
 
-       for (ifma = ifp->if_multiaddrs.lh_first; ifma;
-            ifma = ifma->ifma_link.le_next)
-               if (equal(sa, ifma->ifma_addr))
-                       break;
-       if (ifma == 0)
-               return ENOENT;
+/*
+ * Delete a multicast group membership by group membership pointer.
+ * Network-layer protocol domains must use this routine.
+ */
+int
+if_delmulti_ifma(struct ifmultiaddr *ifma)
+{
+       return (if_delmulti_common(ifma, NULL, NULL, 0));
+}
 
-       if (ifma->ifma_refcount > 1) {
-               ifma->ifma_refcount--;
-               return 0;
-       }
+/*
+ * Anything other than network-layer protocol domains which hold references
+ * to the underlying link-layer record must use this routine: SIOCDELMULTI
+ * ioctl, ifnet_remove_multicast(), AppleTalk, if_bond.
+ */
+int
+if_delmulti_anon(struct ifnet *ifp, const struct sockaddr *sa)
+{
+       return (if_delmulti_common(NULL, ifp, sa, 1));
+}
 
-       rt_newmaddrmsg(RTM_DELMADDR, ifma);
-       sa = ifma->ifma_lladdr;
-       s = splimp();
-       LIST_REMOVE(ifma, ifma_link);
-       /*
-        * Make sure the interface driver is notified
-        * in the case of a link layer mcast group being left.
-        */
-       if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0)
-               dlil_ioctl(0, ifp, SIOCDELMULTI, 0);
-       splx(s);
-       FREE(ifma->ifma_addr, M_IFMADDR);
-       FREE(ifma, M_IFMADDR);
-       if (sa == 0)
-               return 0;
+/*
+ * Delete a multicast group membership by network-layer group address.
+ *
+ * Returns ENOENT if the entry could not be found.
+ */
+static int
+if_delmulti_common(struct ifmultiaddr *ifma, struct ifnet *ifp,
+    const struct sockaddr *sa, int anon)
+{
+       struct sockaddr         *dupsa = NULL;
+       int                     lastref, ll_lastref = 0, lladdr;
+       struct ifmultiaddr      *ll = NULL;
+
+       /* sanity check for callers */
+       VERIFY(ifma != NULL || (ifp != NULL && sa != NULL));
+
+       if (ifma != NULL)
+               ifp = ifma->ifma_ifp;
+
+       if (sa != NULL &&
+           (sa->sa_family == AF_LINK || sa->sa_family == AF_UNSPEC)) {
+               dupsa = copy_and_normalize(sa);
+               if (dupsa == NULL)
+                       return (ENOMEM);
+               sa = dupsa;
+       }
 
-       /*
-        * Now look for the link-layer address which corresponds to
-        * this network address.  It had been squirreled away in
-        * ifma->ifma_lladdr for this purpose (so we don't have
-        * to call SIOCRSLVMULTI again), and we saved that
-        * value in sa above.  If some nasty deleted the
-        * link-layer address out from underneath us, we can deal because
-        * the address we stored was is not the same as the one which was
-        * in the record for the link-layer address.  (So we don't complain
-        * in that case.)
-        */
-       for (ifma = ifp->if_multiaddrs.lh_first; ifma;
-            ifma = ifma->ifma_link.le_next)
-               if (equal(sa, ifma->ifma_addr))
+       ifnet_lock_exclusive(ifp);
+       if (ifma == NULL) {
+               for (ifma = LIST_FIRST(&ifp->if_multiaddrs); ifma != NULL;
+                    ifma = LIST_NEXT(ifma, ifma_link)) {
+                       IFMA_LOCK(ifma);
+                       if (!equal(sa, ifma->ifma_addr) ||
+                           (anon && !(ifma->ifma_flags & IFMAF_ANONYMOUS))) {
+                               VERIFY(!(ifma->ifma_flags & IFMAF_ANONYMOUS) ||
+                                   ifma->ifma_anoncnt != 0);
+                               IFMA_UNLOCK(ifma);
+                               continue;
+                       }
+                       /* found; keep it locked */
                        break;
-       if (ifma == 0)
-               return 0;
+               }
+               if (ifma == NULL) {
+                       if (dupsa != NULL)
+                               FREE(dupsa, M_IFADDR);
+                       ifnet_lock_done(ifp);
+                       return (ENOENT);
+               }
+       } else {
+               IFMA_LOCK(ifma);
+       }
+       IFMA_LOCK_ASSERT_HELD(ifma);
+       IFMA_ADDREF_LOCKED(ifma);       /* for this routine */
+       lastref = if_detach_ifma(ifp, ifma, anon);
+       VERIFY(!lastref || (!(ifma->ifma_debug & IFD_ATTACHED) &&
+           ifma->ifma_reqcnt == 0));
+       VERIFY(!anon || ifma->ifma_ll == NULL);
+       ll = ifma->ifma_ll;
+       lladdr = (ifma->ifma_addr->sa_family == AF_UNSPEC ||
+           ifma->ifma_addr->sa_family == AF_LINK);
+       IFMA_UNLOCK(ifma);
+       if (lastref && ll != NULL) {
+               IFMA_LOCK(ll);
+               ll_lastref = if_detach_ifma(ifp, ll, 0);
+               IFMA_UNLOCK(ll);
+       }
+       ifnet_lock_done(ifp);
+
+       if (lastref)
+               rt_newmaddrmsg(RTM_DELMADDR, ifma);
 
-       if (ifma->ifma_refcount > 1) {
-               ifma->ifma_refcount--;
-               return 0;
+       if ((ll == NULL && lastref && lladdr) || ll_lastref) {
+               /*
+                * Make sure the interface driver is notified in the
+                * case of a link layer mcast group being left.  Do
+                * this only for a AF_LINK/AF_UNSPEC address that has
+                * been removed from the if_multiaddrs set.
+                */
+               ifnet_ioctl(ifp, 0, SIOCDELMULTI, NULL);
        }
 
-       s = splimp();
-       LIST_REMOVE(ifma, ifma_link);
-       dlil_ioctl(0, ifp, SIOCDELMULTI, (caddr_t) 0);
-       splx(s);
-       FREE(ifma->ifma_addr, M_IFMADDR);
-       FREE(sa, M_IFMADDR);
-       FREE(ifma, M_IFMADDR);
+       if (lastref)
+               IFMA_REMREF(ifma);      /* for if_multiaddrs list */
+       if (ll_lastref)
+               IFMA_REMREF(ll);        /* for if_multiaddrs list */
 
-       return 0;
-}
+       IFMA_REMREF(ifma);              /* for this routine */
+       if (dupsa != NULL)
+               FREE(dupsa, M_IFADDR);
 
+       return (0);
+}
 
 /*
  * We don't use if_setlladdr, our interfaces are responsible for
@@ -1307,40 +2743,30 @@ if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
 }
 #endif
 
-struct ifmultiaddr *
-ifmaof_ifpforaddr(sa, ifp)
-       struct sockaddr *sa;
-       struct ifnet *ifp;
-{
-       struct ifmultiaddr *ifma;
-       
-       for (ifma = ifp->if_multiaddrs.lh_first; ifma;
-            ifma = ifma->ifma_link.le_next)
-               if (equal(ifma->ifma_addr, sa))
-                       break;
-
-       return ifma;
-}
-
-SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
-SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
+SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW|CTLFLAG_LOCKED, 0, "Link layers");
+SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW|CTLFLAG_LOCKED, 0, "Generic link-management");
 
 
 /*
  * Shutdown all network activity.  Used boot() when halting
  * system.
  */
-int if_down_all(void)
+int
+if_down_all(void)
 {
-       struct ifnet *ifp;
-       int s;
-
-       s = splnet();
-       TAILQ_FOREACH(ifp, &ifnet, if_link)
-               if_down(ifp);
+       struct ifnet **ifp;
+       u_int32_t       count;
+       u_int32_t       i;
+
+       if (ifnet_list_get_all(IFNET_FAMILY_ANY, &ifp, &count) == 0) {
+               for (i = 0; i < count; i++) {
+                       if_down(ifp[i]);
+                       dlil_proto_unplumb_all(ifp[i]);
+               }
+               ifnet_list_free(ifp);
+       }
 
-       splx(s);
-       return(0);              /* Sheesh */
+       return 0;
 }
 
 /*
@@ -1359,44 +2785,310 @@ int if_down_all(void)
  *
  */
 static int
-if_rtdel(rn, arg)
-       struct radix_node       *rn;
-       void                    *arg;
+if_rtdel(struct radix_node *rn, void *arg)
 {
        struct rtentry  *rt = (struct rtentry *)rn;
        struct ifnet    *ifp = arg;
        int             err;
 
-       if (rt != NULL && rt->rt_ifp == ifp) {
-               
+       if (rt == NULL)
+               return (0);
+       /*
+        * Checking against RTF_UP protects against walktree
+        * recursion problems with cloned routes.
+        */
+       RT_LOCK(rt);
+       if (rt->rt_ifp == ifp && (rt->rt_flags & RTF_UP)) {
                /*
-                * Protect (sorta) against walktree recursion problems
-                * with cloned routes
+                * Safe to drop rt_lock and use rt_key, rt_gateway,
+                * since holding rnh_lock here prevents another thread
+                * from calling rt_setgate() on this route.
                 */
-               if ((rt->rt_flags & RTF_UP) == 0)
-                       return (0);
-
-               err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
-                               rt_mask(rt), rt->rt_flags,
-                               (struct rtentry **) NULL);
+               RT_UNLOCK(rt);
+               err = rtrequest_locked(RTM_DELETE, rt_key(rt), rt->rt_gateway,
+                   rt_mask(rt), rt->rt_flags, NULL);
                if (err) {
                        log(LOG_WARNING, "if_rtdel: error %d\n", err);
                }
+       } else {
+               RT_UNLOCK(rt);
        }
-
        return (0);
 }
 
 /*
- * Removes routing table reference to a given interfacei
+ * Removes routing table reference to a given interface
  * for a given protocol family
  */
+void
+if_rtproto_del(struct ifnet *ifp, int protocol)
+{
+       struct radix_node_head  *rnh;
+
+       if (use_routegenid)
+               routegenid_update();
+       if ((protocol <= AF_MAX) && (protocol >= 0) &&
+               ((rnh = rt_tables[protocol]) != NULL) && (ifp != NULL)) {
+               lck_mtx_lock(rnh_lock);
+               (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
+               lck_mtx_unlock(rnh_lock);
+       }
+}
+
+static int
+if_rtmtu(struct radix_node *rn, void *arg)
+{
+       struct rtentry *rt = (struct rtentry *)rn;
+       struct ifnet *ifp = arg;
+
+       RT_LOCK(rt);
+       if (rt->rt_ifp == ifp) {
+               /*
+                * Update the MTU of this entry only if the MTU
+                * has not been locked (RTV_MTU is not set) and
+                * if it was non-zero to begin with.
+                */
+               if (!(rt->rt_rmx.rmx_locks & RTV_MTU) && rt->rt_rmx.rmx_mtu)
+                       rt->rt_rmx.rmx_mtu = ifp->if_mtu;
+       }
+       RT_UNLOCK(rt);
+
+       return (0);
+}
+
+/*
+ * Update the MTU metric of all route entries in all protocol tables
+ * associated with a particular interface; this is called when the
+ * MTU of that interface has changed.
+ */
+static
+void if_rtmtu_update(struct ifnet *ifp)
+{
+       struct radix_node_head *rnh;
+       int p;
+
+       for (p = 0; p < AF_MAX + 1; p++) {
+               if ((rnh = rt_tables[p]) == NULL)
+                       continue;
+
+               lck_mtx_lock(rnh_lock);
+               (void) rnh->rnh_walktree(rnh, if_rtmtu, ifp);
+               lck_mtx_unlock(rnh_lock);
+       }
+
+       if (use_routegenid)
+               routegenid_update();
+}
 
-void if_rtproto_del(struct ifnet *ifp, int protocol)
+__private_extern__ void
+if_data_internal_to_if_data(struct ifnet *ifp,
+    const struct if_data_internal *if_data_int, struct if_data *if_data)
 {
+#pragma unused(ifp)
+#define COPYFIELD(fld)         if_data->fld = if_data_int->fld
+#define COPYFIELD32(fld)       if_data->fld = (u_int32_t)(if_data_int->fld)
+/* compiler will cast down to 32-bit */
+#define        COPYFIELD32_ATOMIC(fld) do {                                            \
+       atomic_get_64(if_data->fld,                                             \
+           (u_int64_t *)(void *)(uintptr_t)&if_data_int->fld);                 \
+} while (0)
+
+       COPYFIELD(ifi_type);
+       COPYFIELD(ifi_typelen);
+       COPYFIELD(ifi_physical);
+       COPYFIELD(ifi_addrlen);
+       COPYFIELD(ifi_hdrlen);
+       COPYFIELD(ifi_recvquota);
+       COPYFIELD(ifi_xmitquota);
+       if_data->ifi_unused1 = 0;
+       COPYFIELD(ifi_mtu);
+       COPYFIELD(ifi_metric);
+       if (if_data_int->ifi_baudrate & 0xFFFFFFFF00000000LL) {
+               if_data->ifi_baudrate = 0xFFFFFFFF;
+       } else {
+               COPYFIELD32(ifi_baudrate);
+       }
+
+       COPYFIELD32_ATOMIC(ifi_ipackets);
+       COPYFIELD32_ATOMIC(ifi_ierrors);
+       COPYFIELD32_ATOMIC(ifi_opackets);
+       COPYFIELD32_ATOMIC(ifi_oerrors);
+       COPYFIELD32_ATOMIC(ifi_collisions);
+       COPYFIELD32_ATOMIC(ifi_ibytes);
+       COPYFIELD32_ATOMIC(ifi_obytes);
+       COPYFIELD32_ATOMIC(ifi_imcasts);
+       COPYFIELD32_ATOMIC(ifi_omcasts);
+       COPYFIELD32_ATOMIC(ifi_iqdrops);
+       COPYFIELD32_ATOMIC(ifi_noproto);
+
+       COPYFIELD(ifi_recvtiming);
+       COPYFIELD(ifi_xmittiming);
        
-        struct radix_node_head  *rnh;
+       if_data->ifi_lastchange.tv_sec = if_data_int->ifi_lastchange.tv_sec;
+       if_data->ifi_lastchange.tv_usec = if_data_int->ifi_lastchange.tv_usec;
 
-       if ((protocol <= AF_MAX) && ((rnh = rt_tables[protocol]) != NULL) && (ifp != NULL))
-               (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
+#if IF_LASTCHANGEUPTIME
+       if_data->ifi_lastchange.tv_sec += boottime_sec();
+#endif
+
+       if_data->ifi_unused2 = 0;
+       COPYFIELD(ifi_hwassist);
+       if_data->ifi_reserved1 = 0;
+       if_data->ifi_reserved2 = 0;
+#undef COPYFIELD32_ATOMIC
+#undef COPYFIELD32
+#undef COPYFIELD
+}
+
+__private_extern__ void
+if_data_internal_to_if_data64(struct ifnet *ifp,
+    const struct if_data_internal *if_data_int,
+    struct if_data64 *if_data64)
+{
+#pragma unused(ifp)
+#define COPYFIELD64(fld)       if_data64->fld = if_data_int->fld
+#define COPYFIELD64_ATOMIC(fld) do {                                           \
+       atomic_get_64(if_data64->fld,                                           \
+       (u_int64_t *)(void *)(uintptr_t)&if_data_int->fld);                     \
+} while (0)
+
+       COPYFIELD64(ifi_type);
+       COPYFIELD64(ifi_typelen);
+       COPYFIELD64(ifi_physical);
+       COPYFIELD64(ifi_addrlen);
+       COPYFIELD64(ifi_hdrlen);
+       COPYFIELD64(ifi_recvquota);
+       COPYFIELD64(ifi_xmitquota);
+       if_data64->ifi_unused1 = 0;
+       COPYFIELD64(ifi_mtu);
+       COPYFIELD64(ifi_metric);
+       COPYFIELD64(ifi_baudrate);
+
+       COPYFIELD64_ATOMIC(ifi_ipackets);
+       COPYFIELD64_ATOMIC(ifi_ierrors);
+       COPYFIELD64_ATOMIC(ifi_opackets);
+       COPYFIELD64_ATOMIC(ifi_oerrors);
+       COPYFIELD64_ATOMIC(ifi_collisions);
+       COPYFIELD64_ATOMIC(ifi_ibytes);
+       COPYFIELD64_ATOMIC(ifi_obytes);
+       COPYFIELD64_ATOMIC(ifi_imcasts);
+       COPYFIELD64_ATOMIC(ifi_omcasts);
+       COPYFIELD64_ATOMIC(ifi_iqdrops);
+       COPYFIELD64_ATOMIC(ifi_noproto);
+
+       /* Note these two fields are actually 32 bit, so doing COPYFIELD64_ATOMIC will
+        * cause them to be misaligned
+        */
+       COPYFIELD64(ifi_recvtiming);
+       COPYFIELD64(ifi_xmittiming);
+
+       if_data64->ifi_lastchange.tv_sec = if_data_int->ifi_lastchange.tv_sec;
+       if_data64->ifi_lastchange.tv_usec = if_data_int->ifi_lastchange.tv_usec;
+
+#if IF_LASTCHANGEUPTIME
+       if_data64->ifi_lastchange.tv_sec += boottime_sec();
+#endif
+
+#undef COPYFIELD64
+}
+
+__private_extern__ void
+if_copy_traffic_class(struct ifnet *ifp,
+    struct if_traffic_class *if_tc)
+{
+#define COPY_IF_TC_FIELD64_ATOMIC(fld) do {                            \
+       atomic_get_64(if_tc->fld,                                                       \
+       (u_int64_t *)(void *)(uintptr_t)&ifp->if_tc.fld);       \
+} while (0)
+
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ibkpackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ibkbytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_obkpackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_obkbytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ivipackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ivibytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ovipackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ovibytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ivopackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ivobytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ovopackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ovobytes);
+
+#undef COPY_IF_TC_FIELD64_ATOMIC
+}
+
+
+struct ifaddr *
+ifa_remref(struct ifaddr *ifa, int locked)
+{
+       if (!locked)
+               IFA_LOCK_SPIN(ifa);
+       else
+               IFA_LOCK_ASSERT_HELD(ifa);
+
+       if (ifa->ifa_refcnt == 0)
+               panic("%s: ifa %p negative refcnt\n", __func__, ifa);
+       else if (ifa->ifa_trace != NULL)
+               (*ifa->ifa_trace)(ifa, FALSE);
+       if (--ifa->ifa_refcnt == 0) {
+               if (ifa->ifa_debug & IFD_ATTACHED)
+                       panic("ifa %p attached to ifp is being freed\n", ifa);
+               /*
+                * Some interface addresses are allocated either statically
+                * or carved out of a larger block; e.g. AppleTalk addresses.
+                * Only free it if it was allocated via MALLOC or via the
+                * corresponding per-address family allocator.  Otherwise,
+                * leave it alone.
+                */
+               if (ifa->ifa_debug & IFD_ALLOC) {
+                       if (ifa->ifa_free == NULL) {
+                               IFA_UNLOCK(ifa);
+                               FREE(ifa, M_IFADDR);
+                       } else {
+                               /* Become a regular mutex */
+                               IFA_CONVERT_LOCK(ifa);
+                               /* callee will unlock */
+                               (*ifa->ifa_free)(ifa);
+                       }
+               } else {
+                       IFA_UNLOCK(ifa);
+               }
+               ifa = NULL;
+       }
+
+       if (!locked && ifa != NULL)
+               IFA_UNLOCK(ifa);
+
+       return (ifa);
+}
+
+void
+ifa_addref(struct ifaddr *ifa, int locked)
+{
+       if (!locked)
+               IFA_LOCK_SPIN(ifa);
+       else
+               IFA_LOCK_ASSERT_HELD(ifa);
+
+       if (++ifa->ifa_refcnt == 0) {
+               panic("%s: ifa %p wraparound refcnt\n", __func__, ifa);
+               /* NOTREACHED */
+       } else if (ifa->ifa_trace != NULL) {
+               (*ifa->ifa_trace)(ifa, TRUE);
+       }
+       if (!locked)
+               IFA_UNLOCK(ifa);
+}
+
+void
+ifa_lock_init(struct ifaddr *ifa)
+{
+       lck_mtx_init(&ifa->ifa_lock, ifa_mtx_grp, ifa_mtx_attr);
+}
+
+void
+ifa_lock_destroy(struct ifaddr *ifa)
+{
+       IFA_LOCK_ASSERT_NOTHELD(ifa);
+       lck_mtx_destroy(&ifa->ifa_lock, ifa_mtx_grp);
 }