]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/net/if.c
xnu-4570.71.2.tar.gz
[apple/xnu.git] / bsd / net / if.c
index 5659b1c9d327b73264743ab15916797f9bee39e8..0fe5872eaf537bf4058971028b9dd61f5baa8941 100644 (file)
@@ -1,17 +1,20 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2018 Apple Inc. All rights reserved.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
- * 
  * 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
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
@@ -19,8 +22,8 @@
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * 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 <sys/kauth.h>
+#include <sys/priv.h>
+#include <kern/zalloc.h>
+#include <mach/boolean.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/if_ppp.h>
+#include <net/ethernet.h>
+#include <net/network_agent.h>
 #include <net/radix.h>
 #include <net/route.h>
-#ifdef __APPLE__
 #include <net/dlil.h>
-//#include <string.h>
+#include <net/nwk_wq.h>
+
 #include <sys/domain.h>
-#endif
+#include <libkern/OSAtomic.h>
 
-#if defined(INET) || defined(INET6)
-/*XXX*/
+#if INET || INET6
 #include <netinet/in.h>
 #include <netinet/in_var.h>
+#include <netinet/in_tclass.h>
+#include <netinet/ip_var.h>
+#include <netinet/ip6.h>
+#include <netinet/ip_var.h>
+#include <netinet/tcp.h>
+#include <netinet/tcp_var.h>
+#include <netinet/udp.h>
+#include <netinet/udp_var.h>
 #if INET6
 #include <netinet6/in6_var.h>
 #include <netinet6/in6_ifattach.h>
-#endif
+#include <netinet6/ip6_var.h>
+#include <netinet6/nd6.h>
+#endif /* INET6 */
+#endif /* INET || INET6 */
+
+#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 *));
-
-static struct  if_clone *if_clone_lookup(const char *, int *);
-static int     if_clone_list(struct if_clonereq *);
+extern char *proc_name_address(void *);
+
+/* 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 ifioctl_ifreq(struct socket *, u_long, struct ifreq *,
+    struct proc *);
+static int ifioctl_ifconf(u_long, caddr_t);
+static int ifioctl_ifclone(u_long, caddr_t);
+static int ifioctl_iforder(u_long, caddr_t);
+static int ifioctl_ifdesc(struct ifnet *, u_long, caddr_t, struct proc *);
+static int ifioctl_linkparams(struct ifnet *, u_long, caddr_t, struct proc *);
+static int ifioctl_qstats(struct ifnet *, u_long, caddr_t);
+static int ifioctl_throttle(struct ifnet *, u_long, caddr_t, struct proc *);
+static int ifioctl_netsignature(struct ifnet *, u_long, caddr_t);
+static int ifconf(u_long cmd, user_addr_t ifrp, int * ret_space);
+__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 struct ifnet *ifunit_common(const char *, boolean_t);
+
+static int if_rtmtu(struct radix_node *, void *);
+static void if_rtmtu_update(struct ifnet *);
+
+static int if_clone_list(int, int *, user_addr_t);
 
 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
-MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
 
-int    ifqmaxlen = IFQ_MAXLEN;
-struct ifnethead ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
-struct ifmultihead ifma_lostlist = LIST_HEAD_INITIALIZER(ifma_lostlist);
+struct ifnethead ifnet_head = TAILQ_HEAD_INITIALIZER(ifnet_head);
+
+/* ifnet_ordered_head and if_ordered_count are protected by the ifnet_head lock */
+struct ifnethead ifnet_ordered_head = TAILQ_HEAD_INITIALIZER(ifnet_ordered_head);
+static u_int32_t if_ordered_count = 0;
 
 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 void    nd6_setmtu(struct ifnet *);
+extern lck_mtx_t *nd6_mutex;
 #endif
 
-#define M_CLONE                M_IFADDR
+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");
+
+SYSCTL_DECL(_net_link_generic_system);
+
+static uint32_t if_verbose = 0;
+SYSCTL_INT(_net_link_generic_system, OID_AUTO, if_verbose,
+    CTLFLAG_RW | CTLFLAG_LOCKED, &if_verbose, 0, "");
+
+boolean_t intcoproc_unrestricted;
+
+/* Eventhandler context for interface events */
+struct eventhandler_lists_ctxt ifnet_evhdlr_ctxt;
+
+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);
+
+       PE_parse_boot_argn("intcoproc_unrestricted", &intcoproc_unrestricted,
+           sizeof (intcoproc_unrestricted));
+}
 
 /*
  * Network interface utility routines.
@@ -136,13 +288,118 @@ int if_index;
 struct ifaddr **ifnet_addrs;
 struct ifnet **ifindex2ifnet;
 
-#define INITIAL_IF_INDEXLIM    8
+__private_extern__ void
+if_attach_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_attach_ifa_common(ifp, ifa, 0);
+}
+
+__private_extern__ void
+if_attach_link_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_attach_ifa_common(ifp, ifa, 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);
+
+       if (ifa->ifa_attached != NULL)
+               (*ifa->ifa_attached)(ifa);
+
+}
+
+__private_extern__ void
+if_detach_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_detach_ifa_common(ifp, ifa, 0);
+}
+
+__private_extern__ void
+if_detach_link_ifa(struct ifnet *ifp, struct ifaddr *ifa)
+{
+       if_detach_ifa_common(ifp, ifa, 1);
+}
+
+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 */
+       }
+       ifa->ifa_debug &= ~(IFD_ATTACHED | IFD_DETACHING);
+
+       if (ifa->ifa_detached != NULL)
+               (*ifa->ifa_detached)(ifa);
+
+}
+
+#define        INITIAL_IF_INDEXLIM     8
 
 /*
  * Function: if_next_index
  * Purpose:
- *   Return the next available interface index.  
- *   Grow the ifnet_addrs[] and ifindex2ifnet[] arrays to accomodate the 
+ *   Return the next available interface index.
+ *   Grow the ifnet_addrs[] and ifindex2ifnet[] arrays to accomodate the
  *   added entry when necessary.
  *
  * Note:
@@ -151,17 +408,14 @@ struct ifnet **ifindex2ifnet;
  *   always allocate one extra element to hold ifindex2ifnet[0], which
  *   is unused.
  */
-static int
+int if_next_index(void);
+
+__private_extern__ int
 if_next_index(void)
 {
        static int      if_indexlim = 0;
-       static int      if_list_growing = 0;
        int             new_index;
 
-       while (if_list_growing) {
-               /* wait until list is done growing */
-               (void)tsleep((caddr_t)&ifnet_addrs, PZERO, "if_next_index", 0);
-       }
        new_index = ++if_index;
        if (if_index > if_indexlim) {
                unsigned        n;
@@ -170,9 +424,6 @@ if_next_index(void)
                caddr_t         new_ifindex2ifnet;
                caddr_t         old_ifnet_addrs;
 
-               /* mark list as growing */
-               if_list_growing = 1;
-
                old_ifnet_addrs = (caddr_t)ifnet_addrs;
                if (ifnet_addrs == NULL) {
                        new_if_indexlim = INITIAL_IF_INDEXLIM;
@@ -182,123 +433,47 @@ if_next_index(void)
 
                /* allocate space for the larger arrays */
                n = (2 * new_if_indexlim + 1) * sizeof(caddr_t);
-               new_ifnet_addrs = _MALLOC(n, M_IFADDR, M_WAITOK);
-               new_ifindex2ifnet = new_ifnet_addrs 
+               new_ifnet_addrs = _MALLOC(n, M_IFADDR, M_WAITOK | M_ZERO);
+               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));
+                           if_indexlim * sizeof(caddr_t));
                        bcopy((caddr_t)ifindex2ifnet,
-                             new_ifindex2ifnet,
-                             (if_indexlim + 1) * sizeof(caddr_t));
+                           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;
+               ifnet_addrs = (struct ifaddr **)(void *)new_ifnet_addrs;
+               ifindex2ifnet = (struct ifnet **)(void *)new_ifindex2ifnet;
                if_indexlim = new_if_indexlim;
 
                /* release the old data */
                if (old_ifnet_addrs != NULL) {
                        _FREE((caddr_t)old_ifnet_addrs, M_IFADDR);
                }
-
-               /* wake up others that might be blocked */
-               if_list_growing = 0;
-               wakeup((caddr_t)&ifnet_addrs);
        }
        return (new_index);
-
-}
-
-/*
- * Attach an interface to the
- * list of "active" interfaces.
- */
-void
-old_if_attach(ifp)
-       struct ifnet *ifp;
-{
-       unsigned socksize, ifasize;
-       int namelen, masklen;
-       char workbuf[64];
-       register struct sockaddr_dl *sdl;
-       register struct ifaddr *ifa;
-
-       if (ifp->if_snd.ifq_maxlen == 0)
-           ifp->if_snd.ifq_maxlen = ifqmaxlen;
-
-       /*
-        * 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 ((ifp->if_eflags & IFEF_REUSE) == 0 || ifp->if_index == 0) {
-               /* allocate a new entry */
-               ifp->if_index = if_next_index();
-               ifindex2ifnet[ifp->if_index] = ifp;
-
-               /*
-                * 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[ifp->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;
-               }
-       } else {
-               ifa = ifnet_addrs[ifp->if_index - 1];
-       }
-       if (ifa != NULL) {
-               TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
-       }
-       TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
 }
 
 /*
  * Create a clone network interface.
  */
 static int
-if_clone_create(char *name, int len)
+if_clone_create(char *name, int len, void *params)
 {
        struct if_clone *ifc;
        char *dp;
-       int wildcard, bytoff, bitoff;
-       int unit;
+       int wildcard;
+       u_int32_t bytoff, bitoff;
+       u_int32_t unit;
        int err;
 
        ifc = if_clone_lookup(name, &unit);
@@ -309,13 +484,13 @@ if_clone_create(char *name, int len)
                return (EEXIST);
 
        bytoff = bitoff = 0;
-       wildcard = (unit < 0);
+       wildcard = (unit == UINT32_MAX);
        /*
         * Find a free unit if none was given.
         */
        if (wildcard) {
-               while ((bytoff < ifc->ifc_bmlen)
-                   && (ifc->ifc_units[bytoff] == 0xff))
+               while ((bytoff < ifc->ifc_bmlen) &&
+                   (ifc->ifc_units[bytoff] == 0xff))
                        bytoff++;
                if (bytoff >= ifc->ifc_bmlen)
                        return (ENOSPC);
@@ -327,7 +502,7 @@ if_clone_create(char *name, int len)
        if (unit > ifc->ifc_maxunit)
                return (ENXIO);
 
-       err = (*ifc->ifc_create)(ifc, unit);
+       err = (*ifc->ifc_create)(ifc, unit, params);
        if (err != 0)
                return (err);
 
@@ -353,7 +528,8 @@ if_clone_create(char *name, int len)
                         * there's no straightforward way to recover if
                         * it happens.
                         */
-                       panic("if_clone_create(): interface name too long");
+                       panic("%s: interface name too long", __func__);
+                       /* NOTREACHED */
                }
 
        }
@@ -364,13 +540,13 @@ if_clone_create(char *name, int len)
 /*
  * Destroy a clone network interface.
  */
-int
+static int
 if_clone_destroy(const char *name)
 {
        struct if_clone *ifc;
        struct ifnet *ifp;
        int bytoff, bitoff;
-       int unit;
+       u_int32_t unit;
 
        ifc = if_clone_lookup(name, &unit);
        if (ifc == NULL)
@@ -403,29 +579,29 @@ if_clone_destroy(const char *name)
  * Look up a network interface cloner.
  */
 
-static struct if_clone *
-if_clone_lookup(const char *name, int *unitp)
+__private_extern__ struct if_clone *
+if_clone_lookup(const char *name, u_int32_t *unitp)
 {
        struct if_clone *ifc;
        const char *cp;
-       int i;
+       u_int32_t i;
 
-       for (ifc = LIST_FIRST(&if_cloners); ifc != NULL;) {
+       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:
+next_ifc:
                ifc = LIST_NEXT(ifc, ifc_list);
        }
 
        /* No match. */
        return ((struct if_clone *)NULL);
 
- found_name:
+found_name:
        if (*cp == '\0') {
-               i = -1;
+               i = UINT32_MAX;
        } else {
                for (i = 0; *cp != '\0'; cp++) {
                        if (*cp < '0' || *cp > '9') {
@@ -444,13 +620,13 @@ if_clone_lookup(const char *name, int *unitp)
 /*
  * Register a network interface cloner.
  */
-void
+int
 if_clone_attach(struct if_clone *ifc)
 {
        int bytoff, bitoff;
        int err;
        int len, maxclone;
-       int unit;
+       u_int32_t unit;
 
        KASSERT(ifc->ifc_minifs - 1 <= ifc->ifc_maxunit,
            ("%s: %s requested more units then allowed (%d > %d)",
@@ -464,14 +640,15 @@ if_clone_attach(struct if_clone *ifc)
        if ((len << 3) < maxclone)
                len++;
        ifc->ifc_units = _MALLOC(len, M_CLONE, M_WAITOK | M_ZERO);
-       bzero(ifc->ifc_units, len);
+       if (ifc->ifc_units == NULL)
+               return (ENOBUFS);
        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);
+               err = (*ifc->ifc_create)(ifc, unit, NULL);
                KASSERT(err == 0,
                    ("%s: failed to create required interface %s%d",
                    __func__, ifc->ifc_name, unit));
@@ -481,6 +658,8 @@ if_clone_attach(struct if_clone *ifc)
                bitoff = unit - (bytoff << 3);
                ifc->ifc_units[bytoff] |= (1 << bitoff);
        }
+
+       return (0);
 }
 
 /*
@@ -489,7 +668,6 @@ if_clone_attach(struct if_clone *ifc)
 void
 if_clone_detach(struct if_clone *ifc)
 {
-
        LIST_REMOVE(ifc, ifc_list);
        FREE(ifc->ifc_units, M_CLONE);
        if_cloners_count--;
@@ -499,27 +677,27 @@ if_clone_detach(struct if_clone *ifc)
  * Provide list of interface cloners to userspace.
  */
 static int
-if_clone_list(struct if_clonereq *ifcr)
+if_clone_list(int count, int *ret_total, user_addr_t dst)
 {
-       char outbuf[IFNAMSIZ], *dst;
+       char outbuf[IFNAMSIZ];
        struct if_clone *ifc;
-       int count, error = 0;
+       int error = 0;
 
-       ifcr->ifcr_total = if_cloners_count;
-       if ((dst = ifcr->ifcr_buffer) == NULL) {
+       *ret_total = if_cloners_count;
+       if (dst == USER_ADDR_NULL) {
                /* Just asking how many there are. */
                return (0);
        }
 
-       if (ifcr->ifcr_count < 0)
+       if (count < 0)
                return (EINVAL);
 
-       count = (if_cloners_count < ifcr->ifcr_count) ?
-           if_cloners_count : ifcr->ifcr_count;
+       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) {
-               strncpy(outbuf, ifc->ifc_name, IFNAMSIZ - 1);
+           ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
+               bzero(outbuf, sizeof(outbuf));
+               strlcpy(outbuf, ifc->ifc_name, IFNAMSIZ);
                error = copyout(outbuf, dst, IFNAMSIZ);
                if (error)
                        break;
@@ -528,723 +706,2681 @@ if_clone_list(struct if_clonereq *ifcr)
        return (error);
 }
 
-__private_extern__ int
-ifa_foraddr(addr)
-       unsigned int addr;
-{
-       register struct ifnet *ifp;
-       register struct ifaddr *ifa;
-       register unsigned int addr2;
-       
-
-       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 != AF_INET)
-                       continue;
-               addr2 = IA_SIN(ifa)->sin_addr.s_addr;
-
-               if (addr == addr2)
-                       return (1);
+u_int32_t
+if_functional_type(struct ifnet *ifp, bool exclude_delegate)
+{
+       u_int32_t ret = IFRTYPE_FUNCTIONAL_UNKNOWN;
+       if (ifp != NULL) {
+               if (ifp->if_flags & IFF_LOOPBACK) {
+                       ret = IFRTYPE_FUNCTIONAL_LOOPBACK;
+               } else if ((exclude_delegate &&
+                   (ifp->if_subfamily == IFNET_SUBFAMILY_WIFI)) ||
+                   (!exclude_delegate && IFNET_IS_WIFI(ifp))) {
+                       if (ifp->if_eflags & IFEF_AWDL)
+                               ret = IFRTYPE_FUNCTIONAL_WIFI_AWDL;
+                       else
+                               ret = IFRTYPE_FUNCTIONAL_WIFI_INFRA;
+               } else if ((exclude_delegate &&
+                   (ifp->if_type == IFT_CELLULAR)) ||
+                   (!exclude_delegate && IFNET_IS_CELLULAR(ifp))) {
+                       ret = IFRTYPE_FUNCTIONAL_CELLULAR;
+               } else if (IFNET_IS_INTCOPROC(ifp)) {
+                       ret = IFRTYPE_FUNCTIONAL_INTCOPROC;
+               } else if ((exclude_delegate &&
+                   (ifp->if_family == IFNET_FAMILY_ETHERNET ||
+                   ifp->if_family == IFNET_FAMILY_FIREWIRE)) ||
+                   (!exclude_delegate && IFNET_IS_WIRED(ifp))) {
+                       ret = IFRTYPE_FUNCTIONAL_WIRED;
+               }
        }
-       return (0);
+
+       return (ret);
 }
 
 /*
- * Locate an interface based on a complete address.
+ * Similar to ifa_ifwithaddr, except that this is IPv4 specific
+ * and that it matches only the local (not broadcast) 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);
-       }
-       return ((struct ifaddr *)0);
+__private_extern__ struct in_ifaddr *
+ifa_foraddr(unsigned int addr)
+{
+       return (ifa_foraddr_scoped(addr, IFSCOPE_NONE));
 }
+
 /*
- * Locate the point to point interface with a given destination address.
+ * 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).
  */
-/*ARGSUSED*/
-struct ifaddr *
-ifa_ifwithdstaddr(addr)
-       register struct sockaddr *addr;
+__private_extern__ struct in_ifaddr *
+ifa_foraddr_scoped(unsigned int addr, unsigned int scope)
 {
-       register struct ifnet *ifp;
-       register struct ifaddr *ifa;
-
-       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);
+       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);
        }
-       return ((struct ifaddr *)0);
+       lck_rw_done(in_ifaddr_rwlock);
+       return (ia);
 }
 
+#if INET6
 /*
- * Find an interface on a specific network.  If many, choice
- * is most specific found.
+ * Similar to ifa_foraddr, except that this for IPv6.
  */
-struct ifaddr *
-ifa_ifwithnet(addr)
-       struct sockaddr *addr;
+__private_extern__ struct in6_ifaddr *
+ifa_foraddr6(struct in6_addr *addr6)
 {
-       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;
-
-       /*
-        * AF_LINK addresses can be looked up directly by their index number,
-        * so do that if we can.
-        */
-       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]);
-       }
-
-       /*
-        * 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 (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;
-                                       }
-                               }
+       return (ifa_foraddr6_scoped(addr6, IFSCOPE_NONE));
+}
 
-                               /*
-                                * 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;
-                       }
+__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);
        }
-       return (ifa_maybe);
+       lck_rw_done(&in6_ifaddr_rwlock);
+
+       return (ia);
 }
+#endif /* INET6 */
 
 /*
- * Find an interface address specific to an interface best matching
- * a given address.
+ * Return the first (primary) address of a given family on an interface.
  */
-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);
+__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);
        }
-       return (ifa_maybe);
+       ifnet_lock_done(ifp);
+
+       return (ifa);
 }
 
-#include <net/route.h>
+static inline int
+ifa_equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
+{
+
+       if (!sa1 || !sa2)
+               return 0;
+       if (sa1->sa_len != sa2->sa_len)
+               return 0;
+
+       return (bcmp(sa1, sa2, sa1->sa_len) == 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.
+ * Locate an interface based on a complete address.
  */
-static void
-link_rtrequest(cmd, rt, sa)
-       int cmd;
-       register struct rtentry *rt;
-       struct sockaddr *sa;
+struct ifaddr *
+ifa_ifwithaddr_locked(const struct sockaddr *addr)
 {
-       register struct ifaddr *ifa;
-       struct sockaddr *dst;
        struct ifnet *ifp;
+       struct ifaddr *ifa;
+       struct ifaddr *result = NULL;
 
-       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);
+       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 (ifa_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 &&
+                           ifa_equal(ifa->ifa_broadaddr, addr)) {
+                               result = ifa;
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       IFA_UNLOCK(ifa);
+               }
+               ifnet_lock_done(ifp);
        }
-}
 
-/*
- * 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);
+       return (result);
 }
 
-/*
- * Mark an interface up and notify protocols of
- * the transition.
- * NOTE: must be called at splnet or eqivalent.
- */
-void
-if_route(ifp, flag, fam)
-       register struct ifnet *ifp;
-       int flag, fam;
+struct ifaddr *
+ifa_ifwithaddr(const struct sockaddr *addr)
 {
-       register struct ifaddr *ifa;
+       struct ifaddr *result = NULL;
 
-       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);
+       ifnet_head_lock_shared();
 
-}
+       result = ifa_ifwithaddr_locked(addr);
 
+       ifnet_head_done();
+
+       return (result);
+}
 /*
- * Mark an interface down and notify protocols of
- * the transition.
- * NOTE: must be called at splnet or eqivalent.
+ * Locate the point to point interface with a given destination address.
  */
-void
-if_down(ifp)
-       register struct ifnet *ifp;
+/*ARGSUSED*/
+struct ifaddr *
+ifa_ifwithdstaddr(const struct sockaddr *addr)
 {
-
-       if_unroute(ifp, IFF_UP, AF_UNSPEC);
+       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_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);
 }
 
 /*
- * Mark an interface up and notify protocols of
+ * Locate the source address of an interface based on a complete address.
+ */
+struct ifaddr *
+ifa_ifwithaddr_scoped_locked(const struct sockaddr *addr, unsigned int ifscope)
+{
+       struct ifaddr *result = NULL;
+       struct ifnet *ifp;
+
+       if (ifscope == IFSCOPE_NONE)
+               return (ifa_ifwithaddr_locked(addr));
+
+       if (ifscope > (unsigned int)if_index) {
+               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 (ifa_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 &&
+                           ifa_equal(ifa->ifa_broadaddr, addr)) {
+                               result = ifa;
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+                       IFA_UNLOCK(ifa);
+               }
+               ifnet_lock_done(ifp);
+       }
+
+       return (result);
+}
+
+struct ifaddr *
+ifa_ifwithaddr_scoped(const struct sockaddr *addr, unsigned int ifscope)
+{
+       struct ifaddr *result = NULL;
+
+       ifnet_head_lock_shared();
+
+       result = ifa_ifwithaddr_scoped_locked(addr, ifscope);
+
+       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)
+#else
+       if (af != AF_INET)
+#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 *)(uintptr_t)(size_t)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;
+                       }
+                       /*
+                        * 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 applying same source address selection rules
+ * as done in the kernel for implicit source address binding
+ */
+struct ifaddr *
+ifaof_ifpforaddr_select(const struct sockaddr *addr, struct ifnet *ifp)
+{
+       u_int af = addr->sa_family;
+
+       if (af == AF_INET6)
+               return (in6_selectsrc_core_ifa(__DECONST(struct sockaddr_in6 *, addr), ifp, 0));
+
+       return (ifaof_ifpforaddr(addr, ifp));
+}
+
+/*
+ * Find an interface address specific to an interface best matching
+ * a given address without regards to source address selection.
+ *
+ * This is appropriate for use-cases where we just want to update/init
+ * some data structure like routing table entries.
+ */
+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 (ifa_equal(addr, ifa->ifa_addr) ||
+                           ifa_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_equal(addr, ifa->ifa_dstaddr)) {
+                               IFA_ADDREF_LOCKED(ifa); /* for caller */
+                               IFA_UNLOCK(ifa);
+                               break;
+                       }
+               } else {
+                       if (ifa_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;
+       struct ifclassq *ifq = &ifp->if_snd;
+
+       /* 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);
+       ifnet_touch_lastupdown(ifp);
+
+       /* Drop the lock to notify addresses and route */
+       ifnet_lock_done(ifp);
+
+       IFCQ_LOCK(ifq);
+       if_qflush(ifp, 1);
+
+       /* Inform all transmit queues about the new link state */
+       ifnet_update_sndq(ifq, up ? CLASSQ_EV_LINK_UP : CLASSQ_EV_LINK_DOWN);
+       IFCQ_UNLOCK(ifq);
+
+       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 */
+       ifnet_lock_exclusive(ifp);
+       ifp->if_eflags &= ~IFEF_UPDOWNCHANGE;
+       wakeup(&ifp->if_eflags);
+}
+
+/*
+ * Mark an interface down and notify protocols of
  * the transition.
- * NOTE: must be called at splnet or eqivalent.
  */
 void
-if_up(ifp)
-       register struct ifnet *ifp;
+if_down(
+       struct ifnet *ifp)
 {
+       ifnet_lock_exclusive(ifp);
+       if_updown(ifp, 0);
+       ifnet_lock_done(ifp);
+}
 
-       if_route(ifp, IFF_UP, AF_UNSPEC);
+/*
+ * 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(ifq)
-       register struct ifqueue *ifq;
+void
+if_qflush(struct ifnet *ifp, int ifq_locked)
+{
+       struct ifclassq *ifq = &ifp->if_snd;
+
+       if (!ifq_locked)
+               IFCQ_LOCK(ifq);
+
+       if (IFCQ_IS_ENABLED(ifq))
+               IFCQ_PURGE(ifq);
+
+       VERIFY(IFCQ_IS_EMPTY(ifq));
+
+       if (!ifq_locked)
+               IFCQ_UNLOCK(ifq);
+}
+
+void
+if_qflush_sc(struct ifnet *ifp, mbuf_svc_class_t sc, u_int32_t flow,
+    u_int32_t *packets, u_int32_t *bytes, int ifq_locked)
+{
+       struct ifclassq *ifq = &ifp->if_snd;
+       u_int32_t cnt = 0, len = 0;
+       u_int32_t a_cnt = 0, a_len = 0;
+
+       VERIFY(sc == MBUF_SC_UNSPEC || MBUF_VALID_SC(sc));
+       VERIFY(flow != 0);
+
+       if (!ifq_locked)
+               IFCQ_LOCK(ifq);
+
+       if (IFCQ_IS_ENABLED(ifq))
+               IFCQ_PURGE_SC(ifq, sc, flow, cnt, len);
+
+       if (!ifq_locked)
+               IFCQ_UNLOCK(ifq);
+
+       if (packets != NULL)
+               *packets = cnt + a_cnt;
+       if (bytes != NULL)
+               *bytes = len + a_len;
+}
+
+/*
+ * Extracts interface unit number and name from string, returns -1 upon failure.
+ * Upon success, returns extracted unit number, and interface name in dst.
+ */
+int
+ifunit_extract(const char *src, char *dst, size_t dstlen, int *unit)
+{
+       const char *cp;
+       size_t len, m;
+       char c;
+       int u;
+
+       if (src == NULL || dst == NULL || dstlen == 0 || unit == NULL)
+               return (-1);
+
+       len = strlen(src);
+       if (len < 2 || len > dstlen)
+               return (-1);
+       cp = src + len - 1;
+       c = *cp;
+       if (c < '0' || c > '9')
+               return (-1);            /* trailing garbage */
+       u = 0;
+       m = 1;
+       do {
+               if (cp == src)
+                       return (-1);    /* no interface name */
+               u += (c - '0') * m;
+               if (u > 1000000)
+                       return (-1);    /* number is unreasonable */
+               m *= 10;
+               c = *--cp;
+       } while (c >= '0' && c <= '9');
+       len = cp - src + 1;
+       bcopy(src, dst, len);
+       dst[len] = '\0';
+       *unit = u;
+
+       return (0);
+}
+
+/*
+ * Map interface name to
+ * interface structure pointer.
+ */
+static struct ifnet *
+ifunit_common(const char *name, boolean_t hold)
 {
-       register struct mbuf *m, *n;
+       char namebuf[IFNAMSIZ + 1];
+       struct ifnet *ifp;
+       int unit;
+
+       if (ifunit_extract(name, namebuf, sizeof (namebuf), &unit) < 0)
+               return (NULL);
+
+       /* for safety, since we use strcmp() below */
+       namebuf[sizeof (namebuf) - 1] = '\0';
+
+       /*
+        * Now search all the interfaces for this name/number
+        */
+       ifnet_head_lock_shared();
+       TAILQ_FOREACH(ifp, &ifnet_head, if_link) {
+               /*
+                * Use strcmp() rather than strncmp() here,
+                * since we want to match the entire string.
+                */
+               if (strcmp(ifp->if_name, namebuf))
+                       continue;
+               if (unit == ifp->if_unit)
+                       break;
+       }
+
+       /* if called from ifunit_ref() and ifnet is not attached, bail */
+       if (hold && ifp != NULL && !ifnet_is_attached(ifp, 1))
+               ifp = NULL;
+
+       ifnet_head_done();
+       return (ifp);
+}
+
+struct ifnet *
+ifunit(const char *name)
+{
+       return (ifunit_common(name, FALSE));
+}
+
+/*
+ * Similar to ifunit(), except that we hold an I/O reference count on an
+ * attached interface, which must later be released via ifnet_decr_iorefcnt().
+ * Will return NULL unless interface exists and is fully attached.
+ */
+struct ifnet *
+ifunit_ref(const char *name)
+{
+       return (ifunit_common(name, TRUE));
+}
+
+/*
+ * Map interface name in a sockaddr_dl to
+ * interface structure pointer.
+ */
+struct ifnet *
+if_withname(struct sockaddr *sa)
+{
+       char ifname[IFNAMSIZ+1];
+       struct sockaddr_dl *sdl = (struct sockaddr_dl *)(void *)sa;
+
+       if ((sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
+           (sdl->sdl_nlen > IFNAMSIZ))
+               return (NULL);
+
+       /*
+        * ifunit wants a null-terminated name.  It may not be null-terminated
+        * in the sockaddr.  We don't want to change the caller's sockaddr,
+        * and there might not be room to put the trailing null anyway, so we
+        * make a local copy that we know we can null terminate safely.
+        */
+
+       bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
+       ifname[sdl->sdl_nlen] = '\0';
+       return (ifunit(ifname));
+}
+
+static __attribute__((noinline)) int
+ifioctl_ifconf(u_long cmd, caddr_t data)
+{
+       int error = 0;
+
+       switch (cmd) {
+       case OSIOCGIFCONF32:                    /* struct ifconf32 */
+       case SIOCGIFCONF32: {                   /* struct ifconf32 */
+               struct ifconf32 ifc;
+               bcopy(data, &ifc, sizeof (ifc));
+               error = ifconf(cmd, CAST_USER_ADDR_T(ifc.ifc_req),
+                   &ifc.ifc_len);
+               bcopy(&ifc, data, sizeof (ifc));
+               break;
+       }
+
+       case SIOCGIFCONF64:                     /* struct ifconf64 */
+       case OSIOCGIFCONF64: {                  /* struct ifconf64 */
+               struct ifconf64 ifc;
+               bcopy(data, &ifc, sizeof (ifc));
+               error = ifconf(cmd, ifc.ifc_req, &ifc.ifc_len);
+               bcopy(&ifc, data, sizeof (ifc));
+               break;
+       }
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_ifclone(u_long cmd, caddr_t data)
+{
+       int error = 0;
+
+       switch (cmd) {
+       case SIOCIFGCLONERS32: {                /* struct if_clonereq32 */
+               struct if_clonereq32 ifcr;
+               bcopy(data, &ifcr, sizeof (ifcr));
+               error = if_clone_list(ifcr.ifcr_count, &ifcr.ifcr_total,
+                   CAST_USER_ADDR_T(ifcr.ifcru_buffer));
+               bcopy(&ifcr, data, sizeof (ifcr));
+               break;
+       }
+
+       case SIOCIFGCLONERS64: {                /* struct if_clonereq64 */
+               struct if_clonereq64 ifcr;
+               bcopy(data, &ifcr, sizeof (ifcr));
+               error = if_clone_list(ifcr.ifcr_count, &ifcr.ifcr_total,
+                   ifcr.ifcru_buffer);
+               bcopy(&ifcr, data, sizeof (ifcr));
+               break;
+       }
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_ifdesc(struct ifnet *ifp, u_long cmd, caddr_t data, struct proc *p)
+{
+       struct if_descreq *ifdr = (struct if_descreq *)(void *)data;
+       u_int32_t ifdr_len;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+
+       switch (cmd) {
+       case SIOCSIFDESC: {                     /* struct if_descreq */
+               if ((error = proc_suser(p)) != 0)
+                       break;
+
+               ifnet_lock_exclusive(ifp);
+               bcopy(&ifdr->ifdr_len, &ifdr_len, sizeof (ifdr_len));
+               if (ifdr_len > sizeof (ifdr->ifdr_desc) ||
+                   ifdr_len > ifp->if_desc.ifd_maxlen) {
+                       error = EINVAL;
+                       ifnet_lock_done(ifp);
+                       break;
+               }
+
+               bzero(ifp->if_desc.ifd_desc, ifp->if_desc.ifd_maxlen);
+               if ((ifp->if_desc.ifd_len = ifdr_len) > 0) {
+                       bcopy(ifdr->ifdr_desc, ifp->if_desc.ifd_desc,
+                           MIN(ifdr_len, ifp->if_desc.ifd_maxlen));
+               }
+               ifnet_lock_done(ifp);
+               break;
+       }
+
+       case SIOCGIFDESC: {                     /* struct if_descreq */
+               ifnet_lock_shared(ifp);
+               ifdr_len = MIN(ifp->if_desc.ifd_len, sizeof (ifdr->ifdr_desc));
+               bcopy(&ifdr_len, &ifdr->ifdr_len, sizeof (ifdr_len));
+               bzero(&ifdr->ifdr_desc, sizeof (ifdr->ifdr_desc));
+               if (ifdr_len > 0) {
+                       bcopy(ifp->if_desc.ifd_desc, ifdr->ifdr_desc, ifdr_len);
+               }
+               ifnet_lock_done(ifp);
+               break;
+       }
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_linkparams(struct ifnet *ifp, u_long cmd, caddr_t data, struct proc *p)
+{
+       struct if_linkparamsreq *iflpr =
+           (struct if_linkparamsreq *)(void *)data;
+       struct ifclassq *ifq;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+       ifq = &ifp->if_snd;
+
+       switch (cmd) {
+       case SIOCSIFLINKPARAMS: {               /* struct if_linkparamsreq */
+               struct tb_profile tb = { 0, 0, 0 };
+
+               if ((error = proc_suser(p)) != 0)
+                       break;
+
+
+               IFCQ_LOCK(ifq);
+               if (!IFCQ_IS_READY(ifq)) {
+                       error = ENXIO;
+                       IFCQ_UNLOCK(ifq);
+                       break;
+               }
+               bcopy(&iflpr->iflpr_output_tbr_rate, &tb.rate,
+                   sizeof (tb.rate));
+               bcopy(&iflpr->iflpr_output_tbr_percent, &tb.percent,
+                   sizeof (tb.percent));
+               error = ifclassq_tbr_set(ifq, &tb, TRUE);
+               IFCQ_UNLOCK(ifq);
+               break;
+       }
+
+       case SIOCGIFLINKPARAMS: {               /* struct if_linkparamsreq */
+               u_int32_t sched_type = PKTSCHEDT_NONE, flags = 0;
+               u_int64_t tbr_bw = 0, tbr_pct = 0;
+
+               IFCQ_LOCK(ifq);
+
+               if (IFCQ_IS_ENABLED(ifq))
+                       sched_type = ifq->ifcq_type;
+
+               bcopy(&sched_type, &iflpr->iflpr_output_sched,
+                   sizeof (iflpr->iflpr_output_sched));
+
+               if (IFCQ_TBR_IS_ENABLED(ifq)) {
+                       tbr_bw = ifq->ifcq_tbr.tbr_rate_raw;
+                       tbr_pct = ifq->ifcq_tbr.tbr_percent;
+               }
+               bcopy(&tbr_bw, &iflpr->iflpr_output_tbr_rate,
+                   sizeof (iflpr->iflpr_output_tbr_rate));
+               bcopy(&tbr_pct, &iflpr->iflpr_output_tbr_percent,
+                   sizeof (iflpr->iflpr_output_tbr_percent));
+               IFCQ_UNLOCK(ifq);
+
+               if (ifp->if_output_sched_model ==
+                   IFNET_SCHED_MODEL_DRIVER_MANAGED)
+                       flags |= IFLPRF_DRVMANAGED;
+               bcopy(&flags, &iflpr->iflpr_flags, sizeof (iflpr->iflpr_flags));
+               bcopy(&ifp->if_output_bw, &iflpr->iflpr_output_bw,
+                   sizeof (iflpr->iflpr_output_bw));
+               bcopy(&ifp->if_input_bw, &iflpr->iflpr_input_bw,
+                   sizeof (iflpr->iflpr_input_bw));
+               bcopy(&ifp->if_output_lt, &iflpr->iflpr_output_lt,
+                   sizeof (iflpr->iflpr_output_lt));
+               bcopy(&ifp->if_input_lt, &iflpr->iflpr_input_lt,
+                   sizeof (iflpr->iflpr_input_lt));
+               break;
+       }
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_qstats(struct ifnet *ifp, u_long cmd, caddr_t data)
+{
+       struct if_qstatsreq *ifqr = (struct if_qstatsreq *)(void *)data;
+       u_int32_t ifqr_len, ifqr_slot;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+
+       switch (cmd) {
+       case SIOCGIFQUEUESTATS: {               /* struct if_qstatsreq */
+               bcopy(&ifqr->ifqr_slot, &ifqr_slot, sizeof (ifqr_slot));
+               bcopy(&ifqr->ifqr_len, &ifqr_len, sizeof (ifqr_len));
+               error = ifclassq_getqstats(&ifp->if_snd, ifqr_slot,
+                   ifqr->ifqr_buf, &ifqr_len);
+               if (error != 0)
+                       ifqr_len = 0;
+               bcopy(&ifqr_len, &ifqr->ifqr_len, sizeof (ifqr_len));
+               break;
+       }
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_throttle(struct ifnet *ifp, u_long cmd, caddr_t data, struct proc *p)
+{
+       struct if_throttlereq *ifthr = (struct if_throttlereq *)(void *)data;
+       u_int32_t ifthr_level;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+
+       switch (cmd) {
+       case SIOCSIFTHROTTLE: {                 /* struct if_throttlereq */
+               /*
+                * XXX: Use priv_check_cred() instead of root check?
+                */
+               if ((error = proc_suser(p)) != 0)
+                       break;
+
+               bcopy(&ifthr->ifthr_level, &ifthr_level, sizeof (ifthr_level));
+               error = ifnet_set_throttle(ifp, ifthr_level);
+               if (error == EALREADY)
+                       error = 0;
+               break;
+       }
+
+       case SIOCGIFTHROTTLE: {                 /* struct if_throttlereq */
+               if ((error = ifnet_get_throttle(ifp, &ifthr_level)) == 0) {
+                       bcopy(&ifthr_level, &ifthr->ifthr_level,
+                           sizeof (ifthr_level));
+               }
+               break;
+       }
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+static int
+ifioctl_getnetagents(struct ifnet *ifp, u_int32_t *count, user_addr_t uuid_p)
+{
+       int error = 0;
+       u_int32_t index = 0;
+       u_int32_t valid_netagent_count = 0;
+       *count = 0;
+
+       ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_SHARED);
+
+       if (ifp->if_agentids != NULL) {
+               for (index = 0; index < ifp->if_agentcount; index++) {
+                       uuid_t *netagent_uuid = &(ifp->if_agentids[index]);
+                       if (!uuid_is_null(*netagent_uuid)) {
+                               if (uuid_p != USER_ADDR_NULL) {
+                                       error = copyout(netagent_uuid,
+                                               uuid_p + sizeof(uuid_t) * valid_netagent_count,
+                                               sizeof(uuid_t));
+                                       if (error != 0) {
+                                               return (error);
+                                       }
+                               }
+                               valid_netagent_count++;
+                       }
+               }
+       }
+       *count = valid_netagent_count;
+
+       return (0);
+}
+
+#define        IF_MAXAGENTS            64
+#define        IF_AGENT_INCREMENT      8
+static int
+if_add_netagent_locked(struct ifnet *ifp, uuid_t new_agent_uuid)
+{
+       uuid_t *first_empty_slot = NULL;
+       u_int32_t index = 0;
+       bool already_added = FALSE;
+
+       if (ifp->if_agentids != NULL) {
+               for (index = 0; index < ifp->if_agentcount; index++) {
+                       uuid_t *netagent_uuid = &(ifp->if_agentids[index]);
+                       if (uuid_compare(*netagent_uuid, new_agent_uuid) == 0) {
+                               /* Already present, ignore */
+                               already_added = TRUE;
+                               break;
+                       }
+                       if (first_empty_slot == NULL &&
+                               uuid_is_null(*netagent_uuid)) {
+                               first_empty_slot = netagent_uuid;
+                       }
+               }
+       }
+       if (already_added) {
+               /* Already added agent, don't return an error */
+               return (0);
+       }
+       if (first_empty_slot == NULL) {
+               if (ifp->if_agentcount >= IF_MAXAGENTS) {
+                       /* No room for another netagent UUID, bail */
+                       return (ENOMEM);
+               } else {
+                       /* Calculate new array size */
+                       u_int32_t new_agent_count =
+                       MIN(ifp->if_agentcount + IF_AGENT_INCREMENT,
+                           IF_MAXAGENTS);
+
+                       /* Reallocate array */
+                       uuid_t *new_agent_array = _REALLOC(ifp->if_agentids,
+                           sizeof(uuid_t) * new_agent_count, M_NETAGENT,
+                           M_WAITOK | M_ZERO);
+                       if (new_agent_array == NULL) {
+                               return (ENOMEM);
+                       }
+
+                       /* Save new array */
+                       ifp->if_agentids = new_agent_array;
+
+                       /* Set first empty slot */
+                       first_empty_slot =
+                           &(ifp->if_agentids[ifp->if_agentcount]);
+
+                       /* Save new array length */
+                       ifp->if_agentcount = new_agent_count;
+               }
+       }
+       uuid_copy(*first_empty_slot, new_agent_uuid);
+       netagent_post_updated_interfaces(new_agent_uuid);
+       return (0);
+}
+
+int
+if_add_netagent(struct ifnet *ifp, uuid_t new_agent_uuid)
+{
+       VERIFY(ifp != NULL);
+
+       ifnet_lock_exclusive(ifp);
+
+       int error = if_add_netagent_locked(ifp, new_agent_uuid);
+
+       ifnet_lock_done(ifp);
+
+       return (error);
+}
+
+static int
+if_delete_netagent_locked(struct ifnet *ifp, uuid_t remove_agent_uuid)
+{
+       u_int32_t index = 0;
+       bool removed_agent_id = FALSE;
+
+       if (ifp->if_agentids != NULL) {
+               for (index = 0; index < ifp->if_agentcount; index++) {
+                       uuid_t *netagent_uuid = &(ifp->if_agentids[index]);
+                       if (uuid_compare(*netagent_uuid,
+                           remove_agent_uuid) == 0) {
+                               uuid_clear(*netagent_uuid);
+                               removed_agent_id = TRUE;
+                               break;
+                       }
+               }
+       }
+       if (removed_agent_id)
+               netagent_post_updated_interfaces(remove_agent_uuid);
+
+       return (0);
+}
+
+int
+if_delete_netagent(struct ifnet *ifp, uuid_t remove_agent_uuid)
+{
+       VERIFY(ifp != NULL);
+
+       ifnet_lock_exclusive(ifp);
+
+       int error = if_delete_netagent_locked(ifp, remove_agent_uuid);
+
+       ifnet_lock_done(ifp);
+
+       return (error);
+}
+
+boolean_t
+if_check_netagent(struct ifnet *ifp, uuid_t find_agent_uuid)
+{
+       boolean_t found = FALSE;
+
+       if (!ifp || uuid_is_null(find_agent_uuid))
+               return FALSE;
+
+       ifnet_lock_shared(ifp);
+
+       if (ifp->if_agentids != NULL) {
+               for (uint32_t index = 0; index < ifp->if_agentcount; index++) {
+                       if (uuid_compare(ifp->if_agentids[index], find_agent_uuid) == 0) {
+                               found = TRUE;
+                               break;
+                       }
+               }
+       }
+
+       ifnet_lock_done(ifp);
+
+       return found;
+}
+
+static __attribute__((noinline)) int
+ifioctl_netagent(struct ifnet *ifp, u_long cmd, caddr_t data, struct proc *p)
+{
+       struct if_agentidreq *ifar = (struct if_agentidreq *)(void *)data;
+       union {
+               struct if_agentidsreq32 s32;
+               struct if_agentidsreq64 s64;
+       } u;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+
+       /* Get an io ref count if the interface is attached */
+       if (!ifnet_is_attached(ifp, 1)) {
+               return (EOPNOTSUPP);
+       }
+
+       if (cmd == SIOCAIFAGENTID ||
+               cmd == SIOCDIFAGENTID) {
+               ifnet_lock_exclusive(ifp);
+       } else {
+               ifnet_lock_shared(ifp);
+       }
+
+       switch (cmd) {
+               case SIOCAIFAGENTID: {          /* struct if_agentidreq */
+                       // TODO: Use priv_check_cred() instead of root check
+                       if ((error = proc_suser(p)) != 0) {
+                               break;
+                       }
+                       error = if_add_netagent_locked(ifp, ifar->ifar_uuid);
+                       break;
+               }
+               case SIOCDIFAGENTID: {                  /* struct if_agentidreq */
+                       // TODO: Use priv_check_cred() instead of root check
+                       if ((error = proc_suser(p)) != 0) {
+                               break;
+                       }
+                       error = if_delete_netagent_locked(ifp, ifar->ifar_uuid);
+                       break;
+               }
+               case SIOCGIFAGENTIDS32: {       /* struct if_agentidsreq32 */
+                       bcopy(data, &u.s32, sizeof(u.s32));
+                       error = ifioctl_getnetagents(ifp, &u.s32.ifar_count,
+                           u.s32.ifar_uuids);
+                       if (error == 0) {
+                               bcopy(&u.s32, data, sizeof(u.s32));
+                       }
+                       break;
+               }
+               case SIOCGIFAGENTIDS64: {       /* struct if_agentidsreq64 */
+                       bcopy(data, &u.s64, sizeof(u.s64));
+                       error = ifioctl_getnetagents(ifp, &u.s64.ifar_count,
+                           u.s64.ifar_uuids);
+                       if (error == 0) {
+                               bcopy(&u.s64, data, sizeof(u.s64));
+                       }
+                       break;
+               }
+               default:
+                       VERIFY(0);
+                       /* NOTREACHED */
+       }
+
+       ifnet_lock_done(ifp);
+       ifnet_decr_iorefcnt(ifp);
+
+       return (error);
+}
+
+void
+ifnet_clear_netagent(uuid_t netagent_uuid)
+{
+       struct ifnet *ifp = NULL;
+       u_int32_t index = 0;
+
+       ifnet_head_lock_shared();
+
+       TAILQ_FOREACH(ifp, &ifnet_head, if_link) {
+               ifnet_lock_shared(ifp);
+               if (ifp->if_agentids != NULL) {
+                       for (index = 0; index < ifp->if_agentcount; index++) {
+                               uuid_t *ifp_netagent_uuid = &(ifp->if_agentids[index]);
+                               if (uuid_compare(*ifp_netagent_uuid, netagent_uuid) == 0) {
+                                       uuid_clear(*ifp_netagent_uuid);
+                               }
+                       }
+               }
+               ifnet_lock_done(ifp);
+       }
+
+       ifnet_head_done();
+}
+
+void
+ifnet_increment_generation(ifnet_t interface)
+{
+       OSIncrementAtomic(&interface->if_generation);
+}
+
+u_int32_t
+ifnet_get_generation(ifnet_t interface)
+{
+       return (interface->if_generation);
+}
+
+void
+ifnet_remove_from_ordered_list(struct ifnet *ifp)
+{
+       ifnet_head_assert_exclusive();
+
+       // Remove from list
+       TAILQ_REMOVE(&ifnet_ordered_head, ifp, if_ordered_link);
+       ifp->if_ordered_link.tqe_next = NULL;
+       ifp->if_ordered_link.tqe_prev = NULL;
+
+       // Update ordered count
+       VERIFY(if_ordered_count > 0);
+       if_ordered_count--;
+}
+
+static int
+ifnet_reset_order(u_int32_t *ordered_indices, u_int32_t count)
+{
+       struct ifnet *ifp = NULL;
+       int error = 0;
+
+       ifnet_head_lock_exclusive();
+       for (u_int32_t order_index = 0; order_index < count; order_index++) {
+               if (ordered_indices[order_index] == IFSCOPE_NONE ||
+                   ordered_indices[order_index] > (uint32_t)if_index) {
+                       error = EINVAL;
+                       ifnet_head_done();
+                       return (error);
+               }
+       }
+       // Flush current ordered list
+       for (ifp = TAILQ_FIRST(&ifnet_ordered_head); ifp != NULL;
+           ifp = TAILQ_FIRST(&ifnet_ordered_head)) {
+               ifnet_lock_exclusive(ifp);
+               ifnet_remove_from_ordered_list(ifp);
+               ifnet_lock_done(ifp);
+       }
+
+       VERIFY(if_ordered_count == 0);
+
+       for (u_int32_t order_index = 0; order_index < count; order_index++) {
+               u_int32_t interface_index = ordered_indices[order_index];
+               ifp = ifindex2ifnet[interface_index];
+               if (ifp == NULL) {
+                       continue;
+               }
+               ifnet_lock_exclusive(ifp);
+               TAILQ_INSERT_TAIL(&ifnet_ordered_head, ifp, if_ordered_link);
+               ifnet_lock_done(ifp);
+               if_ordered_count++;
+       }
+
+       ifnet_head_done();
+
+       necp_update_all_clients();
+
+       return (error);
+}
+
+int
+if_set_qosmarking_mode(struct ifnet *ifp, u_int32_t mode)
+{
+       int error = 0;
+       u_int32_t old_mode = ifp->if_qosmarking_mode;
+
+       switch (mode) {
+               case IFRTYPE_QOSMARKING_MODE_NONE:
+                       ifp->if_qosmarking_mode = IFRTYPE_QOSMARKING_MODE_NONE;
+                       ifp->if_eflags &= ~IFEF_QOSMARKING_CAPABLE;
+                       break;
+               case IFRTYPE_QOSMARKING_FASTLANE:
+                       ifp->if_qosmarking_mode = IFRTYPE_QOSMARKING_FASTLANE;
+                       ifp->if_eflags |= IFEF_QOSMARKING_CAPABLE;
+                       if (net_qos_policy_capable_enabled != 0)
+                               ifp->if_eflags |= IFEF_QOSMARKING_ENABLED;
+                       break;
+               default:
+                       error = EINVAL;
+                       break;
+       }
+       if (error == 0 && old_mode != ifp->if_qosmarking_mode) {
+               dlil_post_msg(ifp, KEV_DL_SUBCLASS, KEV_DL_QOS_MODE_CHANGED,
+                   NULL, sizeof(struct kev_dl_rrc_state));
+
+       }
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_iforder(u_long cmd, caddr_t data)
+{
+       int error = 0;
+       u_int32_t *ordered_indices = NULL;
+       if (data == NULL) {
+               return (EINVAL);
+       }
+
+       switch (cmd) {
+       case SIOCSIFORDER: {            /* struct if_order */
+               struct if_order *ifo = (struct if_order *)(void *)data;
+
+               if (ifo->ifo_count > (u_int32_t)if_index) {
+                       error = EINVAL;
+                       break;
+               }
+
+               size_t length = (ifo->ifo_count * sizeof(u_int32_t));
+               if (length > 0) {
+                       if (ifo->ifo_ordered_indices == USER_ADDR_NULL) {
+                               error = EINVAL;
+                               break;
+                       }
+                       ordered_indices = _MALLOC(length, M_NECP, M_WAITOK);
+                       if (ordered_indices == NULL) {
+                               error = ENOMEM;
+                               break;
+                       }
+
+                       error = copyin(ifo->ifo_ordered_indices,
+                           ordered_indices, length);
+                       if (error != 0) {
+                               break;
+                       }
+
+                       /* ordered_indices should not contain duplicates */
+                       bool found_duplicate = FALSE;
+                       for (uint32_t i = 0; i < (ifo->ifo_count - 1) && !found_duplicate ; i++){
+                               for (uint32_t j = i + 1; j < ifo->ifo_count && !found_duplicate ; j++){
+                                       if (ordered_indices[j] == ordered_indices[i]){
+                                               error = EINVAL;
+                                               found_duplicate = TRUE;
+                                               break;
+                                       }
+                               }
+                       }
+                       if (found_duplicate) {
+                               break;
+                       }
+               }
+
+               error = ifnet_reset_order(ordered_indices, ifo->ifo_count);
+
+               break;
+       }
+
+       case SIOCGIFORDER: {            /* struct if_order */
+               struct if_order *ifo = (struct if_order *)(void *)data;
+               u_int32_t ordered_count = *((volatile u_int32_t *)&if_ordered_count);
+
+               if (ifo->ifo_count == 0 ||
+                       ordered_count == 0) {
+                       ifo->ifo_count = 0;
+               } else if (ifo->ifo_ordered_indices != USER_ADDR_NULL) {
+                       u_int32_t count_to_copy =
+                           MIN(ordered_count, ifo->ifo_count);
+                       size_t length = (count_to_copy * sizeof(u_int32_t));
+                       struct ifnet *ifp = NULL;
+                       u_int32_t cursor = 0;
+
+                       ordered_indices = _MALLOC(length, M_NECP, M_WAITOK | M_ZERO);
+                       if (ordered_indices == NULL) {
+                               error = ENOMEM;
+                               break;
+                       }
+
+                       ifnet_head_lock_shared();
+                       TAILQ_FOREACH(ifp, &ifnet_ordered_head, if_ordered_link) {
+                               if (cursor >= count_to_copy ||
+                                   cursor >= if_ordered_count) {
+                                       break;
+                               }
+                               ordered_indices[cursor] = ifp->if_index;
+                               cursor++;
+                       }
+                       ifnet_head_done();
+
+                       /* We might have parsed less than the original length
+                        * because the list could have changed.
+                        */
+                       length = cursor * sizeof(u_int32_t);
+                       ifo->ifo_count = cursor;
+                       error = copyout(ordered_indices,
+                           ifo->ifo_ordered_indices, length);
+               } else {
+                       error = EINVAL;
+               }
+               break;
+       }
+
+       default: {
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+       }
+
+       if (ordered_indices != NULL) {
+               _FREE(ordered_indices, M_NECP);
+       }
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_netsignature(struct ifnet *ifp, u_long cmd, caddr_t data)
+{
+       struct if_nsreq *ifnsr = (struct if_nsreq *)(void *)data;
+       u_int16_t flags;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+
+       switch (cmd) {
+       case SIOCSIFNETSIGNATURE:               /* struct if_nsreq */
+               if (ifnsr->ifnsr_len > sizeof (ifnsr->ifnsr_data)) {
+                       error = EINVAL;
+                       break;
+               }
+               bcopy(&ifnsr->ifnsr_flags, &flags, sizeof (flags));
+               error = ifnet_set_netsignature(ifp, ifnsr->ifnsr_family,
+                   ifnsr->ifnsr_len, flags, ifnsr->ifnsr_data);
+               break;
+
+       case SIOCGIFNETSIGNATURE:               /* struct if_nsreq */
+               ifnsr->ifnsr_len = sizeof (ifnsr->ifnsr_data);
+               error = ifnet_get_netsignature(ifp, ifnsr->ifnsr_family,
+                   &ifnsr->ifnsr_len, &flags, ifnsr->ifnsr_data);
+               if (error == 0)
+                       bcopy(&flags, &ifnsr->ifnsr_flags, sizeof (flags));
+               else
+                       ifnsr->ifnsr_len = 0;
+               break;
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+#if INET6
+static __attribute__((noinline)) int
+ifioctl_nat64prefix(struct ifnet *ifp, u_long cmd, caddr_t data)
+{
+       struct if_nat64req *ifnat64 = (struct if_nat64req *)(void *)data;
+       int error = 0;
+
+       VERIFY(ifp != NULL);
+
+       switch (cmd) {
+       case SIOCSIFNAT64PREFIX:                /* struct if_nat64req */
+               error = ifnet_set_nat64prefix(ifp, ifnat64->ifnat64_prefixes);
+               break;
+
+       case SIOCGIFNAT64PREFIX:                /* struct if_nat64req */
+               error = ifnet_get_nat64prefix(ifp, ifnat64->ifnat64_prefixes);
+               break;
+
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+#endif
+
+
+static int
+ifioctl_get_protolist(struct ifnet *ifp, u_int32_t * ret_count,
+    user_addr_t ifpl)
+{
+       u_int32_t       actual_count;
+       u_int32_t       count;
+       int             error = 0;
+       u_int32_t       *list = NULL;
+
+       /* find out how many */
+       count = if_get_protolist(ifp, NULL, 0);
+       if (ifpl == USER_ADDR_NULL) {
+               goto done;
+       }
+
+       /* copy out how many there's space for */
+       if (*ret_count < count) {
+               count = *ret_count;
+       }
+       if (count == 0) {
+               goto done;
+       }
+       list = _MALLOC(count * sizeof(*list), M_TEMP, M_WAITOK);
+       if (list == NULL) {
+               error = ENOMEM;
+               goto done;
+       }
+       actual_count = if_get_protolist(ifp, list, count);
+       if (actual_count < count) {
+               count = actual_count;
+       }
+       if (count != 0) {
+               error = copyout((caddr_t)list, ifpl, count * sizeof(*list));
+       }
+
+ done:
+       if (list != NULL) {
+               if_free_protolist(list);
+       }
+       *ret_count = count;
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_protolist(struct ifnet *ifp, u_long cmd, caddr_t data)
+{
+       int error = 0;
+
+       switch (cmd) {
+       case SIOCGIFPROTOLIST32: {              /* struct if_protolistreq32 */
+               struct if_protolistreq32        ifpl;
+
+               bcopy(data, &ifpl, sizeof(ifpl));
+               if (ifpl.ifpl_reserved != 0) {
+                       error = EINVAL;
+                       break;
+               }
+               error = ifioctl_get_protolist(ifp, &ifpl.ifpl_count,
+                   CAST_USER_ADDR_T(ifpl.ifpl_list));
+               bcopy(&ifpl, data, sizeof(ifpl));
+               break;
+       }
+       case SIOCGIFPROTOLIST64: {              /* struct if_protolistreq64 */
+               struct if_protolistreq64        ifpl;
+
+               bcopy(data, &ifpl, sizeof(ifpl));
+               if (ifpl.ifpl_reserved != 0) {
+                       error = EINVAL;
+                       break;
+               }
+               error = ifioctl_get_protolist(ifp, &ifpl.ifpl_count,
+                   ifpl.ifpl_list);
+               bcopy(&ifpl, data, sizeof(ifpl));
+               break;
+       }
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
+       }
+
+       return (error);
+}
+
+/*
+ * List the ioctl()s we can perform on restricted INTCOPROC interfaces.
+ */
+static bool
+ifioctl_restrict_intcoproc(unsigned long cmd, const char *ifname,
+    struct ifnet *ifp, struct proc *p)
+{
+
+       if (intcoproc_unrestricted == TRUE) {
+               return (false);
+       }
+       if (proc_pid(p) == 0) {
+               return (false);
+       }
+       if (ifname) {
+               ifp = ifunit(ifname);
+       }
+       if (ifp == NULL) {
+               return (false);
+       }
+       if (!IFNET_IS_INTCOPROC(ifp)) {
+               return (false);
+       }
+       switch (cmd) {
+       case SIOCGIFBRDADDR:
+       case SIOCGIFCONF32:
+       case SIOCGIFCONF64:
+       case SIOCGIFFLAGS:
+       case SIOCGIFEFLAGS:
+       case SIOCGIFCAP:
+       case SIOCGIFMAC:
+       case SIOCGIFMETRIC:
+       case SIOCGIFMTU:
+       case SIOCGIFPHYS:
+       case SIOCGIFTYPE:
+       case SIOCGIFFUNCTIONALTYPE:
+       case SIOCGIFPSRCADDR:
+       case SIOCGIFPDSTADDR:
+       case SIOCGIFGENERIC:
+       case SIOCGIFDEVMTU:
+       case SIOCGIFVLAN:
+       case SIOCGIFBOND:
+       case SIOCGIFWAKEFLAGS:
+       case SIOCGIFGETRTREFCNT:
+       case SIOCGIFOPPORTUNISTIC:
+       case SIOCGIFLINKQUALITYMETRIC:
+       case SIOCGIFLOG:
+       case SIOCGIFDELEGATE:
+       case SIOCGIFEXPENSIVE:
+       case SIOCGIFINTERFACESTATE:
+       case SIOCGIFPROBECONNECTIVITY:
+       case SIOCGIFTIMESTAMPENABLED:
+       case SIOCGECNMODE:
+       case SIOCGQOSMARKINGMODE:
+       case SIOCGQOSMARKINGENABLED:
+       case SIOCGIFLOWINTERNET:
+       case SIOCGIFSTATUS:
+       case SIOCGIFMEDIA32:
+       case SIOCGIFMEDIA64:
+       case SIOCGIFDESC:
+       case SIOCGIFLINKPARAMS:
+       case SIOCGIFQUEUESTATS:
+       case SIOCGIFTHROTTLE:
+       case SIOCGIFAGENTIDS32:
+       case SIOCGIFAGENTIDS64:
+       case SIOCGIFNETSIGNATURE:
+       case SIOCGIFINFO_IN6:
+       case SIOCGIFAFLAG_IN6:
+       case SIOCGNBRINFO_IN6:
+       case SIOCGIFALIFETIME_IN6:
+       case SIOCGIFNETMASK_IN6:
+       case SIOCGIFPROTOLIST32:
+       case SIOCGIFPROTOLIST64:
+               return (false);
+       default:
+#if (DEBUG || DEVELOPMENT)
+               printf("%s: cmd 0x%lx not allowed (pid %u)\n",
+                   __func__, cmd, proc_pid(p));
+#endif
+               return (true);
+       }
+       return (false);
+}
+
+/*
+ * Interface ioctls.
+ *
+ * Most of the routines called to handle the ioctls would end up being
+ * tail-call optimized, which unfortunately causes this routine to
+ * consume too much stack space; this is the reason for the "noinline"
+ * attribute used on those routines.
+ */
+int
+ifioctl(struct socket *so, u_long cmd, caddr_t data, struct proc *p)
+{
+       char ifname[IFNAMSIZ + 1];
+       struct ifnet *ifp = NULL;
+       struct ifstat *ifs = NULL;
+       int error = 0;
+
+       bzero(ifname, sizeof (ifname));
+
+       /*
+        * ioctls which don't require ifp, or ifreq ioctls
+        */
+       switch (cmd) {
+       case OSIOCGIFCONF32:                    /* struct ifconf32 */
+       case SIOCGIFCONF32:                     /* struct ifconf32 */
+       case SIOCGIFCONF64:                     /* struct ifconf64 */
+       case OSIOCGIFCONF64:                    /* struct ifconf64 */
+               error = ifioctl_ifconf(cmd, data);
+               goto done;
+
+       case SIOCIFGCLONERS32:                  /* struct if_clonereq32 */
+       case SIOCIFGCLONERS64:                  /* struct if_clonereq64 */
+               error = ifioctl_ifclone(cmd, data);
+               goto done;
+
+       case SIOCGIFAGENTDATA32:                /* struct netagent_req32 */
+       case SIOCGIFAGENTDATA64:                /* struct netagent_req64 */
+       case SIOCGIFAGENTLIST32:                /* struct netagentlist_req32 */
+       case SIOCGIFAGENTLIST64:                /* struct netagentlist_req64 */
+               error = netagent_ioctl(cmd, data);
+               goto done;
+
+       case SIOCSIFORDER:                      /* struct if_order */
+       case SIOCGIFORDER:              /* struct if_order */
+               error = ifioctl_iforder(cmd, data);
+               goto done;
+
+       case SIOCSIFDSTADDR:                    /* struct ifreq */
+       case SIOCSIFADDR:                       /* struct ifreq */
+       case SIOCSIFBRDADDR:                    /* struct ifreq */
+       case SIOCSIFNETMASK:                    /* struct ifreq */
+       case OSIOCGIFADDR:                      /* struct ifreq */
+       case OSIOCGIFDSTADDR:                   /* struct ifreq */
+       case OSIOCGIFBRDADDR:                   /* struct ifreq */
+       case OSIOCGIFNETMASK:                   /* struct ifreq */
+       case SIOCSIFKPI:                        /* struct ifreq */
+               if (so->so_proto == NULL) {
+                       error = EOPNOTSUPP;
+                       goto done;
+               }
+               /* FALLTHRU */
+       case SIOCIFCREATE:                      /* struct ifreq */
+       case SIOCIFCREATE2:                     /* struct ifreq */
+       case SIOCIFDESTROY:                     /* struct ifreq */
+       case SIOCGIFFLAGS:                      /* struct ifreq */
+       case SIOCGIFEFLAGS:                     /* struct ifreq */
+       case SIOCGIFCAP:                        /* struct ifreq */
+#if CONFIG_MACF_NET
+       case SIOCGIFMAC:                        /* struct ifreq */
+       case SIOCSIFMAC:                        /* struct ifreq */
+#endif /* CONFIG_MACF_NET */
+       case SIOCGIFMETRIC:                     /* struct ifreq */
+       case SIOCGIFMTU:                        /* struct ifreq */
+       case SIOCGIFPHYS:                       /* struct ifreq */
+       case SIOCSIFFLAGS:                      /* struct ifreq */
+       case SIOCSIFCAP:                        /* struct ifreq */
+       case SIOCSIFMETRIC:                     /* struct ifreq */
+       case SIOCSIFPHYS:                       /* struct ifreq */
+       case SIOCSIFMTU:                        /* struct ifreq */
+       case SIOCADDMULTI:                      /* struct ifreq */
+       case SIOCDELMULTI:                      /* struct ifreq */
+       case SIOCDIFPHYADDR:                    /* struct ifreq */
+       case SIOCSIFMEDIA:                      /* struct ifreq */
+       case SIOCSIFGENERIC:                    /* struct ifreq */
+       case SIOCSIFLLADDR:                     /* struct ifreq */
+       case SIOCSIFALTMTU:                     /* struct ifreq */
+       case SIOCSIFVLAN:                       /* struct ifreq */
+       case SIOCSIFBOND:                       /* struct ifreq */
+       case SIOCGIFLLADDR:                     /* struct ifreq */
+       case SIOCGIFTYPE:                       /* struct ifreq */
+       case SIOCGIFFUNCTIONALTYPE:             /* struct ifreq */
+       case SIOCGIFPSRCADDR:                   /* struct ifreq */
+       case SIOCGIFPDSTADDR:                   /* struct ifreq */
+       case SIOCGIFGENERIC:                    /* struct ifreq */
+       case SIOCGIFDEVMTU:                     /* struct ifreq */
+       case SIOCGIFVLAN:                       /* struct ifreq */
+       case SIOCGIFBOND:                       /* struct ifreq */
+       case SIOCGIFWAKEFLAGS:                  /* struct ifreq */
+       case SIOCGIFGETRTREFCNT:                /* struct ifreq */
+       case SIOCSIFOPPORTUNISTIC:              /* struct ifreq */
+       case SIOCGIFOPPORTUNISTIC:              /* struct ifreq */
+       case SIOCGIFLINKQUALITYMETRIC:          /* struct ifreq */
+       case SIOCSIFLOG:                        /* struct ifreq */
+       case SIOCGIFLOG:                        /* struct ifreq */
+       case SIOCGIFDELEGATE:                   /* struct ifreq */
+       case SIOCGIFEXPENSIVE:                  /* struct ifreq */
+       case SIOCSIFEXPENSIVE:                  /* struct ifreq */
+       case SIOCSIF2KCL:                       /* struct ifreq */
+       case SIOCGIF2KCL:                       /* struct ifreq */
+       case SIOCSIFINTERFACESTATE:             /* struct ifreq */
+       case SIOCGIFINTERFACESTATE:             /* struct ifreq */
+       case SIOCSIFPROBECONNECTIVITY:          /* struct ifreq */
+       case SIOCGIFPROBECONNECTIVITY:          /* struct ifreq */
+       case SIOCGSTARTDELAY:                   /* struct ifreq */
+       case SIOCSIFTIMESTAMPENABLE:            /* struct ifreq */
+       case SIOCSIFTIMESTAMPDISABLE:           /* struct ifreq */
+       case SIOCGIFTIMESTAMPENABLED:           /* struct ifreq */
+#if (DEBUG || DEVELOPMENT)
+       case SIOCSIFDISABLEOUTPUT:              /* struct ifreq */
+#endif /* (DEBUG || DEVELOPMENT) */
+       case SIOCGECNMODE:                      /* struct ifreq */
+       case SIOCSECNMODE:
+       case SIOCSQOSMARKINGMODE:               /* struct ifreq */
+       case SIOCSQOSMARKINGENABLED:            /* struct ifreq */
+       case SIOCGQOSMARKINGMODE:               /* struct ifreq */
+       case SIOCGQOSMARKINGENABLED:            /* struct ifreq */
+       case SIOCSIFLOWINTERNET:                /* struct ifreq */
+       case SIOCGIFLOWINTERNET:                /* struct ifreq */
+       {                       /* struct ifreq */
+               struct ifreq ifr;
+               bcopy(data, &ifr, sizeof (ifr));
+               ifr.ifr_name[IFNAMSIZ - 1] = '\0';
+               bcopy(&ifr.ifr_name, ifname, IFNAMSIZ);
+               if (ifioctl_restrict_intcoproc(cmd, ifname, NULL, p) == true) {
+                       error = EPERM;
+                       goto done;
+               }
+               error = ifioctl_ifreq(so, cmd, &ifr, p);
+               bcopy(&ifr, data, sizeof (ifr));
+               goto done;
+       }
+       }
+
+       /*
+        * 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();
+       switch (cmd) {
+       case SIOCSIFPHYADDR:                    /* struct {if,in_}aliasreq */
+               bcopy(((struct in_aliasreq *)(void *)data)->ifra_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+#if INET6
+       case SIOCSIFPHYADDR_IN6_32:             /* struct in6_aliasreq_32 */
+               bcopy(((struct in6_aliasreq_32 *)(void *)data)->ifra_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCSIFPHYADDR_IN6_64:             /* struct in6_aliasreq_64 */
+               bcopy(((struct in6_aliasreq_64 *)(void *)data)->ifra_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+#endif /* INET6 */
+
+       case SIOCGIFSTATUS:                     /* struct ifstat */
+               ifs = _MALLOC(sizeof (*ifs), M_DEVBUF, M_WAITOK);
+               if (ifs == NULL) {
+                       error = ENOMEM;
+                       dlil_if_unlock();
+                       goto done;
+               }
+               bcopy(data, ifs, sizeof (*ifs));
+               ifs->ifs_name[IFNAMSIZ - 1] = '\0';
+               bcopy(ifs->ifs_name, ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCGIFMEDIA32:                    /* struct ifmediareq32 */
+               bcopy(((struct ifmediareq32 *)(void *)data)->ifm_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCGIFMEDIA64:                    /* struct ifmediareq64 */
+               bcopy(((struct ifmediareq64 *)(void *)data)->ifm_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCSIFDESC:                       /* struct if_descreq */
+       case SIOCGIFDESC:                       /* struct if_descreq */
+               bcopy(((struct if_descreq *)(void *)data)->ifdr_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCSIFLINKPARAMS:                 /* struct if_linkparamsreq */
+       case SIOCGIFLINKPARAMS:                 /* struct if_linkparamsreq */
+               bcopy(((struct if_linkparamsreq *)(void *)data)->iflpr_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCGIFQUEUESTATS:                 /* struct if_qstatsreq */
+               bcopy(((struct if_qstatsreq *)(void *)data)->ifqr_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCSIFTHROTTLE:                   /* struct if_throttlereq */
+       case SIOCGIFTHROTTLE:                   /* struct if_throttlereq */
+               bcopy(((struct if_throttlereq *)(void *)data)->ifthr_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCAIFAGENTID:                    /* struct if_agentidreq */
+       case SIOCDIFAGENTID:                    /* struct if_agentidreq */
+       case SIOCGIFAGENTIDS32:         /* struct if_agentidsreq32 */
+       case SIOCGIFAGENTIDS64:         /* struct if_agentidsreq64 */
+               bcopy(((struct if_agentidreq *)(void *)data)->ifar_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+
+       case SIOCSIFNETSIGNATURE:               /* struct if_nsreq */
+       case SIOCGIFNETSIGNATURE:               /* struct if_nsreq */
+               bcopy(((struct if_nsreq *)(void *)data)->ifnsr_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
 
-       n = ifq->ifq_head;
-       while ((m = n) != 0) {
-               n = m->m_act;
-               m_freem(m);
+       case SIOCGIFPROTOLIST32:                /* struct if_protolistreq32 */
+       case SIOCGIFPROTOLIST64:                /* struct if_protolistreq64 */
+               bcopy(((struct if_protolistreq *)(void *)data)->ifpl_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
+       default:
+               /*
+                * This is a bad assumption, but the code seems to
+                * have been doing this in the past; caveat emptor.
+                */
+               bcopy(((struct ifreq *)(void *)data)->ifr_name,
+                   ifname, IFNAMSIZ);
+               ifp = ifunit(ifname);
+               break;
        }
-       ifq->ifq_head = 0;
-       ifq->ifq_tail = 0;
-       ifq->ifq_len = 0;
-}
+       dlil_if_unlock();
 
-/*
- * Map interface name to
- * interface structure pointer.
- */
-struct ifnet *
-ifunit(const char *name)
-{
-       char namebuf[IFNAMSIZ + 1];
-       const char *cp;
-       struct ifnet *ifp;
-       int unit;
-       unsigned len, m;
-       char c;
+       if (ifp == NULL) {
+               error = ENXIO;
+               goto done;
+       }
 
-       len = strlen(name);
-       if (len < 2 || len > IFNAMSIZ)
-               return NULL;
-       cp = name + len - 1;
-       c = *cp;
-       if (c < '0' || c > '9')
-               return NULL;            /* trailing garbage */
-       unit = 0;
-       m = 1;
-       do {
-               if (cp == name)
-                       return NULL;    /* no interface name */
-               unit += (c - '0') * m;
-               if (unit > 1000000)
-                       return NULL;    /* number is unreasonable */
-               m *= 10;
-               c = *--cp;
-       } while (c >= '0' && c <= '9');
-       len = cp - name + 1;
-       bcopy(name, namebuf, len);
-       namebuf[len] = '\0';
-       /*
-        * 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))
-                       continue;
-               if (unit == ifp->if_unit)
-                       break;
+       if (ifioctl_restrict_intcoproc(cmd, NULL, ifp, p) == true) {
+               error = EPERM;
+               goto done;
        }
-       return (ifp);
-}
+       switch (cmd) {
+       case SIOCSIFPHYADDR:                    /* struct {if,in_}aliasreq */
+#if INET6
+       case SIOCSIFPHYADDR_IN6_32:             /* struct in6_aliasreq_32 */
+       case SIOCSIFPHYADDR_IN6_64:             /* struct in6_aliasreq_64 */
+#endif /* INET6 */
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, data);
+               if (error != 0)
+                       break;
 
-/*
- * Map interface name in a sockaddr_dl to
- * interface structure pointer.
- */
-struct ifnet *
-if_withname(sa)
-       struct sockaddr *sa;
-{
-       char ifname[IFNAMSIZ+1];
-       struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
+               ifnet_touch_lastchange(ifp);
+               break;
 
-       if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
-            (sdl->sdl_nlen > IFNAMSIZ) )
-               return NULL;
+       case SIOCGIFSTATUS:                     /* struct ifstat */
+               VERIFY(ifs != NULL);
+               ifs->ascii[0] = '\0';
 
-       /*
-        * ifunit wants a null-terminated name.  It may not be null-terminated
-        * in the sockaddr.  We don't want to change the caller's sockaddr,
-        * and there might not be room to put the trailing null anyway, so we
-        * make a local copy that we know we can null terminate safely.
-        */
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, (caddr_t)ifs);
 
-       bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
-       ifname[sdl->sdl_nlen] = '\0';
-       return ifunit(ifname);
-}
+               bcopy(ifs, data, sizeof (*ifs));
+               break;
 
+       case SIOCGIFMEDIA32:                    /* struct ifmediareq32 */
+       case SIOCGIFMEDIA64:                    /* struct ifmediareq64 */
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, data);
+               break;
 
-/*
- * 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;
-       struct ifstat *ifs;
-       int error = 0;
-       short oif_flags;
-       struct kev_msg        ev_msg;
-       struct net_event_data ev_data;
+       case SIOCSIFDESC:                       /* struct if_descreq */
+       case SIOCGIFDESC:                       /* struct if_descreq */
+               error = ifioctl_ifdesc(ifp, cmd, data, p);
+               break;
 
-       switch (cmd) {
+       case SIOCSIFLINKPARAMS:                 /* struct if_linkparamsreq */
+       case SIOCGIFLINKPARAMS:                 /* struct if_linkparamsreq */
+               error = ifioctl_linkparams(ifp, cmd, data, p);
+               break;
+
+       case SIOCGIFQUEUESTATS:                 /* struct if_qstatsreq */
+               error = ifioctl_qstats(ifp, cmd, data);
+               break;
+
+       case SIOCSIFTHROTTLE:                   /* struct if_throttlereq */
+       case SIOCGIFTHROTTLE:                   /* struct if_throttlereq */
+               error = ifioctl_throttle(ifp, cmd, data, p);
+               break;
+
+       case SIOCAIFAGENTID:                    /* struct if_agentidreq */
+       case SIOCDIFAGENTID:                    /* struct if_agentidreq */
+       case SIOCGIFAGENTIDS32:         /* struct if_agentidsreq32 */
+       case SIOCGIFAGENTIDS64:         /* struct if_agentidsreq64 */
+               error = ifioctl_netagent(ifp, cmd, data, p);
+               break;
+
+       case SIOCSIFNETSIGNATURE:               /* struct if_nsreq */
+       case SIOCGIFNETSIGNATURE:               /* struct if_nsreq */
+               error = ifioctl_netsignature(ifp, cmd, data);
+               break;
+
+#if INET6
+       case SIOCSIFNAT64PREFIX:                /* struct if_nsreq */
+       case SIOCGIFNAT64PREFIX:                /* struct if_nsreq */
+               error = ifioctl_nat64prefix(ifp, cmd, data);
+               break;
+#endif
 
-       case SIOCGIFCONF:
-       case OSIOCGIFCONF:
-               return (ifconf(cmd, data));
+       case SIOCGIFPROTOLIST32:                /* struct if_protolistreq32 */
+       case SIOCGIFPROTOLIST64:                /* struct if_protolistreq64 */
+               error = ifioctl_protolist(ifp, cmd, data);
+               break;
+
+       default:
+               if (so->so_proto == NULL) {
+                       error = EOPNOTSUPP;
+                       break;
+               }
+
+               socket_lock(so, 1);
+               error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
+                   data, ifp, p));
+               socket_unlock(so, 1);
+
+               if (error == EOPNOTSUPP || error == ENOTSUP) {
+                       error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, data);
+               }
+               break;
+       }
+
+done:
+       if (ifs != NULL)
+               _FREE(ifs, M_DEVBUF);
+
+       if (if_verbose) {
+               if (ifname[0] == '\0')
+                       (void) snprintf(ifname, sizeof (ifname), "%s",
+                           "NULL");
+               else if (ifp != NULL)
+                       (void) snprintf(ifname, sizeof (ifname), "%s",
+                           if_name(ifp));
+
+               if (error != 0) {
+                       printf("%s[%s,%d]: ifp %s cmd 0x%08lx (%c%c [%lu] "
+                           "%c %lu) error %d\n", __func__,
+                           proc_name_address(p), proc_pid(p),
+                           ifname, cmd, (cmd & IOC_IN) ? 'I' : ' ',
+                           (cmd & IOC_OUT) ? 'O' : ' ', IOCPARM_LEN(cmd),
+                           (char)IOCGROUP(cmd), cmd & 0xff, error);
+               } else if (if_verbose > 1) {
+                       printf("%s[%s,%d]: ifp %s cmd 0x%08lx (%c%c [%lu] "
+                           "%c %lu) OK\n", __func__,
+                           proc_name_address(p), proc_pid(p),
+                           ifname, cmd, (cmd & IOC_IN) ? 'I' : ' ',
+                           (cmd & IOC_OUT) ? 'O' : ' ', IOCPARM_LEN(cmd),
+                           (char)IOCGROUP(cmd), cmd & 0xff);
+               }
        }
-       ifr = (struct ifreq *)data;
+
+       return (error);
+}
+
+static __attribute__((noinline)) int
+ifioctl_ifreq(struct socket *so, u_long cmd, struct ifreq *ifr, struct proc *p)
+{
+       struct ifnet *ifp;
+       u_long ocmd = cmd;
+       int error = 0;
+       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 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 = suser(p->p_ucred, &p->p_acflag);
+               error = proc_suser(p);
                if (error)
                        return (error);
-               return ((cmd == SIOCIFCREATE) ?
-                       if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
-                       if_clone_destroy(ifr->ifr_name));
-#if 0
-       case SIOCIFGCLONERS:
-               return (if_clone_list((struct if_clonereq *)data));
-#endif 0
+               return (if_clone_destroy(ifr->ifr_name));
        }
 
+       /*
+        * 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 SIOCGIFEFLAGS:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_eflags = ifp->if_eflags;
+               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;
+
+       case SIOCSIFMAC:
+               error = mac_ifnet_label_set(kauth_cred_get(), ifr, ifp);
                break;
+#endif /* CONFIG_MACF_NET */
 
        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 = 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;
+               (void) ifnet_set_flags(ifp, ifr->ifr_flags,
+                   (u_int16_t)~IFF_CANTCHANGE);
+
+               /*
+                * Note that we intentionally ignore any error from below
+                * for the SIOCSIFFLAGS case.
+                */
+               (void) ifnet_ioctl(ifp, SOCK_DOM(so), cmd, (caddr_t)ifr);
+
+               /*
+                * Send the event even upon error from the driver because
+                * we changed the flags.
+                */
+               dlil_post_sifflags_msg(ifp);
+
+               ifnet_touch_lastchange(ifp);
+               break;
+
+       case SIOCSIFCAP:
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
-                        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);
+               if ((ifr->ifr_reqcap & ~ifp->if_capabilities)) {
+                       error = EINVAL;
+                       break;
                }
-               getmicrotime(&ifp->if_lastchange);
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, (caddr_t)ifr);
+
+               ifnet_touch_lastchange(ifp);
                break;
 
        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   = (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);
+               dlil_post_complete_msg(ifp, &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;
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
+
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, (caddr_t)ifr);
+               if (error != 0)
+                       break;
 
-               error = dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                       ifp, cmd, (caddr_t) data);
+               ev_msg.vendor_code    = KEV_VENDOR_APPLE;
+               ev_msg.kev_class      = KEV_NETWORK_CLASS;
+               ev_msg.kev_subclass   = KEV_DL_SUBCLASS;
 
-               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;
+               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;
+               dlil_post_complete_msg(ifp, &ev_msg);
+
+               ifnet_touch_lastchange(ifp);
+               break;
+
+       case SIOCSIFMTU: {
+               u_int32_t oldmtu = ifp->if_mtu;
+               struct ifclassq *ifq = &ifp->if_snd;
 
-                       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);
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
-                       getmicrotime(&ifp->if_lastchange);
+               if (ifp->if_ioctl == NULL) {
+                       error = EOPNOTSUPP;
+                       break;
+               }
+               if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
+                       error = EINVAL;
+                       break;
                }
-               return(error);
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, (caddr_t)ifr);
+               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;
+               dlil_post_complete_msg(ifp, &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
+#endif /* INET6 */
+                       /* Inform all transmit queues about the new MTU */
+                       IFCQ_LOCK(ifq);
+                       ifnet_update_sndq(ifq, CLASSQ_EV_LINK_MTU);
+                       IFCQ_UNLOCK(ifq);
                }
-               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;
-
-       case SIOCSETVLAN:
-               if (ifp->if_type != IFT_L2VLAN) {
-                       return (EOPNOTSUPP);
-               }
-       case SIOCSIFPHYADDR:
+               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;
+               dlil_post_complete_msg(ifp, &ev_msg);
+
+               ifnet_touch_lastchange(ifp);
+               break;
+
        case SIOCDIFPHYADDR:
-#ifdef INET6
-       case SIOCSIFPHYADDR_IN6:
-#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, SOCK_DOM(so), cmd, (caddr_t)ifr);
+               if (error != 0)
+                       break;
 
-               if (error == 0)
-                       getmicrotime(&ifp->if_lastchange);
-               return error;
+               ifnet_touch_lastchange(ifp);
+               break;
+
+       case SIOCGIFLLADDR: {
+               struct sockaddr_dl *sdl = SDL(ifp->if_lladdr->ifa_addr);
+
+               if (sdl->sdl_alen == 0) {
+                       error = EADDRNOTAVAIL;
+                       break;
+               }
+               /* If larger than 14-bytes we'll need another mechanism */
+               if (sdl->sdl_alen > sizeof (ifr->ifr_addr.sa_data)) {
+                       error = EMSGSIZE;
+                       break;
+               }
+               /* Follow the same convention used by SIOCSIFLLADDR */
+               bzero(&ifr->ifr_addr, sizeof (ifr->ifr_addr));
+               ifr->ifr_addr.sa_family = AF_LINK;
+               ifr->ifr_addr.sa_len = sdl->sdl_alen;
+               error = ifnet_guarded_lladdr_copy_bytes(ifp,
+                   &ifr->ifr_addr.sa_data, sdl->sdl_alen);
+               break;
+       }
+
+       case SIOCGIFTYPE:
+               ifr->ifr_type.ift_type = ifp->if_type;
+               ifr->ifr_type.ift_family = ifp->if_family;
+               ifr->ifr_type.ift_subfamily = ifp->if_subfamily;
+               break;
+
+       case SIOCGIFFUNCTIONALTYPE:
+               ifr->ifr_functional_type = if_functional_type(ifp, FALSE);
+               break;
 
-       case SIOCGIFSTATUS:
-               ifs = (struct ifstat *)data;
-               ifs->ascii[0] = '\0';
-               
        case SIOCGIFPSRCADDR:
        case SIOCGIFPDSTADDR:
-       case SIOCGLIFPHYADDR:
-       case SIOCGIFMEDIA:
        case SIOCGIFGENERIC:
+       case SIOCGIFDEVMTU:
+       case SIOCGIFVLAN:
+       case SIOCGIFBOND:
+               error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd, (caddr_t)ifr);
+               break;
+
+       case SIOCGIFWAKEFLAGS:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_wake_flags = ifnet_get_wake_flags(ifp);
+               ifnet_lock_done(ifp);
+               break;
+
+       case SIOCGIFGETRTREFCNT:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_route_refcnt = ifp->if_route_refcnt;
+               ifnet_lock_done(ifp);
+               break;
+
+       case SIOCSIFOPPORTUNISTIC:
+       case SIOCGIFOPPORTUNISTIC:
+               error = ifnet_getset_opportunistic(ifp, cmd, ifr, p);
+               break;
 
-               return dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                  ifp, cmd, (caddr_t) data);
-       case SIOCGETVLAN:
-               if (ifp->if_type != IFT_L2VLAN) {
-                       return (EOPNOTSUPP);
+       case SIOCGIFLINKQUALITYMETRIC:
+               ifnet_lock_shared(ifp);
+               if ((ifp->if_interface_state.valid_bitmask &
+                   IF_INTERFACE_STATE_LQM_STATE_VALID)) {
+                       ifr->ifr_link_quality_metric =
+                           ifp->if_interface_state.lqm_state;
+               } else if (IF_FULLY_ATTACHED(ifp)) {
+                       ifr->ifr_link_quality_metric =
+                           IFNET_LQM_THRESH_UNKNOWN;
+               } else {
+                       ifr->ifr_link_quality_metric =
+                           IFNET_LQM_THRESH_OFF;
                }
-               return dlil_ioctl(so->so_proto->pr_domain->dom_family, 
-                                  ifp, cmd, (caddr_t) data);
+               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
-           {
-               int ocmd = cmd;
+       case SIOCSIFLOG:
+       case SIOCGIFLOG:
+               error = ifnet_getset_log(ifp, cmd, ifr, p);
+               break;
+
+       case SIOCGIFDELEGATE:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_delegated = ((ifp->if_delegated.ifp != NULL) ?
+                   ifp->if_delegated.ifp->if_index : 0);
+               ifnet_lock_done(ifp);
+               break;
+
+       case SIOCGIFEXPENSIVE:
+               ifnet_lock_shared(ifp);
+               if (ifp->if_eflags & IFEF_EXPENSIVE)
+                       ifr->ifr_expensive = 1;
+               else
+                       ifr->ifr_expensive = 0;
+               ifnet_lock_done(ifp);
+               break;
+
+       case SIOCSIFEXPENSIVE:
+       {
+               struct ifnet *difp;
+
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               ifnet_lock_exclusive(ifp);
+               if (ifr->ifr_expensive)
+                       ifp->if_eflags |= IFEF_EXPENSIVE;
+               else
+                       ifp->if_eflags &= ~IFEF_EXPENSIVE;
+               ifnet_lock_done(ifp);
+               /*
+                * Update the expensive bit in the delegated interface
+                * structure.
+                */
+               ifnet_head_lock_shared();
+               TAILQ_FOREACH(difp, &ifnet_head, if_link) {
+                       ifnet_lock_exclusive(difp);
+                       if (difp->if_delegated.ifp == ifp) {
+                               difp->if_delegated.expensive =
+                                   ifp->if_eflags & IFEF_EXPENSIVE ? 1 : 0;
+
+                       }
+                       ifnet_lock_done(difp);
+               }
+               ifnet_head_done();
+               break;
+       }
 
-               switch (cmd) {
+       case SIOCGIF2KCL:
+               ifnet_lock_shared(ifp);
+               if (ifp->if_eflags & IFEF_2KCL)
+                       ifr->ifr_2kcl = 1;
+               else
+                       ifr->ifr_2kcl = 0;
+               ifnet_lock_done(ifp);
+               break;
 
-               case SIOCSIFDSTADDR:
-               case SIOCSIFADDR:
-               case SIOCSIFBRDADDR:
-               case SIOCSIFNETMASK:
+       case SIOCSIF2KCL:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               ifnet_lock_exclusive(ifp);
+               if (ifr->ifr_2kcl)
+                       ifp->if_eflags |= IFEF_2KCL;
+               else
+                       ifp->if_eflags &= ~IFEF_2KCL;
+               ifnet_lock_done(ifp);
+               break;
+       case SIOCGSTARTDELAY:
+               ifnet_lock_shared(ifp);
+               if (ifp->if_eflags & IFEF_ENQUEUE_MULTI) {
+                       ifr->ifr_start_delay_qlen =
+                           ifp->if_start_delay_qlen;
+                       ifr->ifr_start_delay_timeout =
+                           ifp->if_start_delay_timeout;
+               } else {
+                       ifr->ifr_start_delay_qlen = 0;
+                       ifr->ifr_start_delay_timeout = 0;
+               }
+               ifnet_lock_done(ifp);
+               break;
+       case SIOCSIFDSTADDR:
+       case SIOCSIFADDR:
+       case SIOCSIFBRDADDR:
+       case SIOCSIFNETMASK:
+       case OSIOCGIFADDR:
+       case OSIOCGIFDSTADDR:
+       case OSIOCGIFBRDADDR:
+       case OSIOCGIFNETMASK:
+       case SIOCSIFKPI:
+               VERIFY(so->so_proto != NULL);
+
+               if (cmd == SIOCSIFDSTADDR || cmd == SIOCSIFADDR ||
+                   cmd == SIOCSIFBRDADDR || cmd == SIOCSIFNETMASK) {
 #if BYTE_ORDER != BIG_ENDIAN
                        if (ifr->ifr_addr.sa_family == 0 &&
                            ifr->ifr_addr.sa_len < 16) {
@@ -1255,46 +3391,214 @@ ifioctl(so, cmd, data, p)
                        if (ifr->ifr_addr.sa_len == 0)
                                ifr->ifr_addr.sa_len = 16;
 #endif
-                       break;
-
-               case OSIOCGIFADDR:
-                       cmd = SIOCGIFADDR;
-                       break;
-
-               case OSIOCGIFDSTADDR:
-                       cmd = SIOCGIFDSTADDR;
-                       break;
+               } else if (cmd == OSIOCGIFADDR) {
+                       cmd = SIOCGIFADDR;      /* struct ifreq */
+               } else if (cmd == OSIOCGIFDSTADDR) {
+                       cmd = SIOCGIFDSTADDR;   /* struct ifreq */
+               } else if (cmd == OSIOCGIFBRDADDR) {
+                       cmd = SIOCGIFBRDADDR;   /* struct ifreq */
+               } else if (cmd == OSIOCGIFNETMASK) {
+                       cmd = SIOCGIFNETMASK;   /* struct ifreq */
+               }
 
-               case OSIOCGIFBRDADDR:
-                       cmd = SIOCGIFBRDADDR;
-                       break;
+               socket_lock(so, 1);
+               error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
+                   (caddr_t)ifr, ifp, p));
+               socket_unlock(so, 1);
 
-               case OSIOCGIFNETMASK:
-                       cmd = SIOCGIFNETMASK;
-               }
-               error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
-                                                                  cmd,
-                                                                  data,
-                                                                  ifp, p));
                switch (ocmd) {
-
                case OSIOCGIFADDR:
                case OSIOCGIFDSTADDR:
                case OSIOCGIFBRDADDR:
                case OSIOCGIFNETMASK:
-                       *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
+                       bcopy(&ifr->ifr_addr.sa_family, &ifr->ifr_addr,
+                           sizeof (u_short));
+               }
+
+               if (cmd == SIOCSIFKPI) {
+                       int temperr = proc_suser(p);
+                       if (temperr != 0)
+                               error = temperr;
+               }
+
+               if (error == EOPNOTSUPP || error == ENOTSUP) {
+                       error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd,
+                           (caddr_t)ifr);
+               }
+               break;
+
+       case SIOCGIFINTERFACESTATE:
+               if_get_state(ifp, &ifr->ifr_interface_state);
+
+               break;
+       case SIOCSIFINTERFACESTATE:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+
+               error = if_state_update(ifp, &ifr->ifr_interface_state);
+
+               break;
+       case SIOCSIFPROBECONNECTIVITY:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               error = if_probe_connectivity(ifp,
+                   ifr->ifr_probe_connectivity);
+               break;
+       case SIOCGIFPROBECONNECTIVITY:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               if (ifp->if_eflags & IFEF_PROBE_CONNECTIVITY)
+                       ifr->ifr_probe_connectivity = 1;
+               else
+                       ifr->ifr_probe_connectivity = 0;
+               break;
+       case SIOCGECNMODE:
+               if ((ifp->if_eflags & (IFEF_ECN_ENABLE|IFEF_ECN_DISABLE)) ==
+                   IFEF_ECN_ENABLE)
+                       ifr->ifr_ecn_mode = IFRTYPE_ECN_ENABLE;
+               else if ((ifp->if_eflags & (IFEF_ECN_ENABLE|IFEF_ECN_DISABLE)) ==
+                   IFEF_ECN_DISABLE)
+                       ifr->ifr_ecn_mode = IFRTYPE_ECN_DISABLE;
+               else
+                       ifr->ifr_ecn_mode = IFRTYPE_ECN_DEFAULT;
+               break;
+       case SIOCSECNMODE:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               if (ifr->ifr_ecn_mode == IFRTYPE_ECN_DEFAULT) {
+                       ifp->if_eflags &= ~(IFEF_ECN_ENABLE|IFEF_ECN_DISABLE);
+               } else if (ifr->ifr_ecn_mode == IFRTYPE_ECN_ENABLE) {
+                       ifp->if_eflags |= IFEF_ECN_ENABLE;
+                       ifp->if_eflags &= ~IFEF_ECN_DISABLE;
+               } else if (ifr->ifr_ecn_mode == IFRTYPE_ECN_DISABLE) {
+                       ifp->if_eflags |= IFEF_ECN_DISABLE;
+                       ifp->if_eflags &= ~IFEF_ECN_ENABLE;
+               } else
+                       error = EINVAL;
+               break;
+       case SIOCSIFTIMESTAMPENABLE:
+       case SIOCSIFTIMESTAMPDISABLE:
+               error = proc_suser(p);
+               if (error != 0)
+                       break;
 
+               ifnet_lock_exclusive(ifp);
+               if ((cmd == SIOCSIFTIMESTAMPENABLE &&
+                   (ifp->if_xflags & IFXF_TIMESTAMP_ENABLED) != 0) ||
+                   (cmd == SIOCSIFTIMESTAMPDISABLE &&
+                   (ifp->if_xflags & IFXF_TIMESTAMP_ENABLED) == 0)) {
+                       ifnet_lock_done(ifp);
+                       break;
+               }
+               if (cmd == SIOCSIFTIMESTAMPENABLE)
+                       ifp->if_xflags |= IFXF_TIMESTAMP_ENABLED;
+               else
+                       ifp->if_xflags &= ~IFXF_TIMESTAMP_ENABLED;
+               ifnet_lock_done(ifp);
+               /*
+                * Pass the setting to the interface if it supports either
+                * software or hardware time stamping
+                */
+               if (ifp->if_capabilities & (IFCAP_HW_TIMESTAMP |
+                   IFCAP_SW_TIMESTAMP)) {
+                       error = ifnet_ioctl(ifp, SOCK_DOM(so), cmd,
+                           (caddr_t)ifr);
                }
-           }
-#endif /* COMPAT_43 */
+               break;
+       case SIOCGIFTIMESTAMPENABLED: {
+               if ((ifp->if_xflags & IFXF_TIMESTAMP_ENABLED) != 0)
+                       ifr->ifr_intval = 1;
+               else
+                       ifr->ifr_intval = 0;
+               break;
+       }
+       case SIOCSQOSMARKINGMODE:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               error = if_set_qosmarking_mode(ifp, ifr->ifr_qosmarking_mode);
+               break;
+
+       case SIOCGQOSMARKINGMODE:
+               ifr->ifr_qosmarking_mode = ifp->if_qosmarking_mode;
+               break;
 
-               if (error == EOPNOTSUPP)
-                       error = dlil_ioctl(so->so_proto->pr_domain->dom_family,
-                                                               ifp, cmd, (caddr_t) data);
+       case SIOCSQOSMARKINGENABLED:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
+               if (ifr->ifr_qosmarking_enabled != 0)
+                       ifp->if_eflags |= IFEF_QOSMARKING_ENABLED;
+               else
+                       ifp->if_eflags &= ~IFEF_QOSMARKING_ENABLED;
+               break;
+
+       case SIOCGQOSMARKINGENABLED:
+               ifr->ifr_qosmarking_enabled =
+                       (ifp->if_eflags & IFEF_QOSMARKING_ENABLED) ? 1 : 0;
+               break;
+
+       case SIOCSIFDISABLEOUTPUT:
+#if (DEBUG || DEVELOPMENT)
+               if (ifr->ifr_disable_output == 1) {
+                       error = ifnet_disable_output(ifp);
+               } else if (ifr->ifr_disable_output == 0) {
+                       error = ifnet_enable_output(ifp);
+               } else {
+                       error = EINVAL;
+               }
+#else
+               error = EINVAL;
+#endif /* (DEBUG || DEVELOPMENT) */
+               break;
+       case SIOCSIFLOWINTERNET:
+               if ((error = priv_check_cred(kauth_cred_get(),
+                   PRIV_NET_INTERFACE_CONTROL, 0)) != 0)
+                       return (error);
 
-               return (error);
+               ifnet_lock_exclusive(ifp);
+               if (ifr->ifr_low_internet & IFRTYPE_LOW_INTERNET_ENABLE_UL)
+                       ifp->if_xflags |= IFXF_LOW_INTERNET_UL;
+               else
+                       ifp->if_xflags &= ~(IFXF_LOW_INTERNET_UL);
+               if (ifr->ifr_low_internet & IFRTYPE_LOW_INTERNET_ENABLE_DL)
+                       ifp->if_xflags |= IFXF_LOW_INTERNET_DL;
+               else
+                       ifp->if_xflags &= ~(IFXF_LOW_INTERNET_DL);
+               ifnet_lock_done(ifp);
+               break;
+       case SIOCGIFLOWINTERNET:
+               ifnet_lock_shared(ifp);
+               ifr->ifr_low_internet = 0;
+               if (ifp->if_xflags & IFXF_LOW_INTERNET_UL)
+                       ifr->ifr_low_internet |=
+                           IFRTYPE_LOW_INTERNET_ENABLE_UL;
+               if (ifp->if_xflags & IFXF_LOW_INTERNET_DL)
+                       ifr->ifr_low_internet |=
+                           IFRTYPE_LOW_INTERNET_ENABLE_DL;
+               ifnet_lock_done(ifp);
+               break;
+       default:
+               VERIFY(0);
+               /* NOTREACHED */
        }
-       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);
 }
 
 /*
@@ -1303,42 +3607,50 @@ ifioctl(so, cmd, data, p)
  * "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);
+               }
        }
-       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;
+
+       if (newflags != oldflags) {
+               log(LOG_INFO, "%s: promiscuous mode %s%s\n",
+                   if_name(ifp),
+                   (newflags & IFF_PROMISC) != 0 ? "enable" : "disable",
+                   error != 0 ? " failed" : " succeeded");
+       }
+       return (error);
 }
 
 /*
@@ -1349,86 +3661,123 @@ 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;
+       net_thread_marks_t marks;
+
+       marks = net_thread_marks_push(NET_THREAD_CKREQ_LLADDR);
+
+       /*
+        * 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);
-               if(ifnlen + 1 > sizeof ifr.ifr_name) {
+               ifnlen = snprintf(workbuf, sizeof (workbuf),
+                   "%s", if_name(ifp));
+               if (ifnlen + 1 > sizeof (ifr.ifr_name)) {
                        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;
+               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
+                       struct sockaddr *sa;
+                       union {
+                               struct sockaddr sa;
+                               struct sockaddr_dl sdl;
+                               uint8_t buf[SOCK_MAXADDRLEN + 1];
+                       } u;
+
+                       /*
+                        * Make sure to accomodate the largest possible
+                        * size of SA(if_lladdr)->sa_len.
+                        */
+                       _CASSERT(sizeof (u) == (SOCK_MAXADDRLEN + 1));
+
+                       IFA_LOCK(ifa);
+                       sa = ifa->ifa_addr;
                        addrs++;
-#ifdef COMPAT_43
-                       if (cmd == OSIOCGIFCONF) {
+
+                       if (ifa == ifp->if_lladdr) {
+                               VERIFY(sa->sa_family == AF_LINK);
+                               bcopy(sa, &u, sa->sa_len);
+                               IFA_UNLOCK(ifa);
+                               ifnet_guarded_lladdr_copy_bytes(ifp,
+                                   LLADDR(&u.sdl), u.sdl.sdl_alen);
+                               IFA_LOCK(ifa);
+                               sa = &u.sa;
+                       }
+
+                       if (cmd == OSIOCGIFCONF32 || cmd == OSIOCGIFCONF64) {
                                struct osockaddr *osa =
-                                        (struct osockaddr *)&ifr.ifr_addr;
+                                   (struct osockaddr *)(void *)&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)) {
+                               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, (caddr_t)ifrp,
-                                               sizeof (ifr));
-                               ifrp++;
+                               error = copyout((caddr_t)&ifr, ifrp,
+                                   sizeof (ifr));
+                               ifrp += sizeof (struct ifreq);
                        } else {
-                               if (space < sizeof (ifr) + sa->sa_len -
-                                           sizeof(*sa))
+                               if (space <
+                                   sizeof (ifr) + sa->sa_len - sizeof (*sa)) {
+                                       IFA_UNLOCK(ifa);
                                        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);
+                               }
+                               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, (caddr_t)ifrp,
-                           sizeof (ifr));
+                       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++;
+                       ifrp += sizeof (struct ifreq);
                }
        }
-       ifc->ifc_len -= space;
+       ifnet_head_done();
+       *ret_space -= space;
+       net_thread_marks_pop(marks);
        return (error);
 }
 
@@ -1436,17 +3785,17 @@ ifconf(cmd, data)
  * Just like if_promisc(), but for all-multicast-reception mode.
  */
 int
-if_allmulti(ifp, onswitch)
-       struct ifnet *ifp;
-       int onswitch;
+if_allmulti(struct ifnet *ifp, int onswitch)
 {
        int error = 0;
-       int s = splimp();
+       int     modified = 0;
+
+       ifnet_lock_exclusive(ifp);
 
        if (onswitch) {
                if (ifp->if_amcount++ == 0) {
                        ifp->if_flags |= IFF_ALLMULTI;
-                       error = dlil_ioctl(0, ifp, SIOCSIFFLAGS, (caddr_t) 0);
+                       modified = 1;
                }
        } else {
                if (ifp->if_amcount > 1) {
@@ -1454,282 +3803,732 @@ if_allmulti(ifp, onswitch)
                } else {
                        ifp->if_amcount = 0;
                        ifp->if_flags &= ~IFF_ALLMULTI;
-                       error = dlil_ioctl(0, ifp, SIOCSIFFLAGS, (caddr_t) 0);
+                       modified = 1;
                }
        }
-       splx(s);
+       ifnet_lock_done(ifp);
+
+       if (modified)
+               error = ifnet_ioctl(ifp, 0, SIOCSIFFLAGS, NULL);
 
        if (error == 0)
                rt_ifmsg(ifp);
-       return error;
+       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;
+
+       for (ifma = LIST_FIRST(&ifp->if_multiaddrs); ifma != NULL;
+           ifma = LIST_NEXT(ifma, ifma_link)) {
+               IFMA_LOCK_SPIN(ifma);
+               if (!ifa_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 (!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);
+       }
+       return (ENOENT);
+}
+
+/*
+ * Radar 3642395, make sure all multicasts are in a standard format.
+ */
+static struct sockaddr *
+copy_and_normalize(const struct sockaddr *original)
+{
+       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 =
+                           (struct sockaddr_dl *)(uintptr_t)(size_t)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);
+               }
+               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);
+       }
+
+       return ((struct sockaddr *)sdl_new);
+}
+
+/*
+ * 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));
+}
+
+/*
+ * 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(), if_bond.
+ */
+int
+if_addmulti_anon(struct ifnet *ifp, const struct sockaddr *sa,
+    struct ifmultiaddr **retifma)
+{
+       return (if_addmulti_common(ifp, sa, retifma, 1));
 }
 
-/*
- * Add a multicast listenership to the interface in question.
- * The link layer provides a routine which converts
- */
-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;
+/*
+ * 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;
 
        /*
-        * If the matching multicast address already exists
-        * then don't add a new one, just add a reference
+        * 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.
         */
-       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;
+       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;
                }
-       }
 
-       /*
-        * 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;
+               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;
+       }
 
-       if (error) 
-            return error;
-
-       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);
-
-       ifma->ifma_addr = dupsa;
-       ifma->ifma_lladdr = llsa;
-       ifma->ifma_ifp = ifp;
-       ifma->ifma_refcount = 1;
-       ifma->ifma_protospec = 0;
-       rt_newmaddrmsg(RTM_NEWMADDR, ifma);
-
+       ifnet_lock_exclusive(ifp);
        /*
-        * Some network interfaces can scan the address list at
-        * interrupt time; lock them out.
+        * Check again for the matching multicast.
         */
-       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++;
+       error = if_addmulti_doesexist(ifp, sa, retifma, anon);
+       if (error == 0) {
+               ifnet_lock_done(ifp);
+               goto cleanup;
+       }
+
+       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 {
-                       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);
+                       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;
                }
        }
-       /*
-        * We are certain we have added something, so call down to the
-        * interface to let them know about it.
-        */
-       s = splimp();
-       dlil_ioctl(0, ifp, SIOCADDMULTI, (caddr_t) 0);
-       splx(s);
 
-       return 0;
-}
+       /* "anonymous" request should not result in network address */
+       VERIFY(!anon || ifma->ifma_ll == NULL);
 
-int
-if_delmultiaddr(struct ifmultiaddr *ifma)
-{
-       struct sockaddr *sa;
-       struct ifnet *ifp;
-       
-       /* Verify ifma is valid */
-       {
-               struct ifmultiaddr *match = NULL;
-               for (ifp = ifnet.tqh_first; ifp; ifp = ifp->if_link.tqe_next) {
-                       for (match = ifp->if_multiaddrs.lh_first; match; match = match->ifma_link.le_next) {
-                               if (match->ifma_ifp != ifp) {
-                                       printf("if_delmultiaddr: ifma (%x) on ifp i(%s) is stale\n",
-                                                       match, if_name(ifp));
-                                       return (0) ; /* swallow error ? */
-                               }
-                               if (match == ifma)
-                                       break;
-                       }
-                       if (match == ifma)
-                               break;
-               }
-               if (match != ifma) {
-                       for (match = ifma_lostlist.lh_first; match; match = match->ifma_link.le_next) {
-                               if (match->ifma_ifp != NULL) {
-                                       printf("if_delmultiaddr: item on lost list (%x) contains non-null ifp=%s\n",
-                                                       match, if_name(match->ifma_ifp));
-                                       return (0) ; /* swallow error ? */
-                               }
-                               if (match == ifma)
-                                       break;
-                       }
-               }
-               
-               if (match != ifma) {
-                       printf("if_delmultiaddr: ifma 0x%X is invalid\n", ifma);
-                       return 0;
-               }
-       }
-       
-       if (ifma->ifma_refcount > 1) {
-               ifma->ifma_refcount--;
-               return 0;
+       ifma->ifma_addr = dupsa;
+       ifma->ifma_ifp = ifp;
+       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);
 
-       sa = ifma->ifma_lladdr;
-
-       if (sa) /* send a routing msg for network addresses only */
-               rt_newmaddrmsg(RTM_DELMADDR, ifma);
+       rt_newmaddrmsg(RTM_NEWMADDR, ifma);
+       IFMA_REMREF(ifma);                      /* for this routine */
 
-       ifp = ifma->ifma_ifp;
-       
-       LIST_REMOVE(ifma, ifma_link);
        /*
-        * Make sure the interface driver is notified
-        * in the case of a link layer mcast group being left.
+        * We are certain we have added something, so call down to the
+        * interface to let them know about it.  Do this only for newly-
+        * added AF_LINK/AF_UNSPEC address in the if_multiaddrs set.
         */
-       if (ifp && ifma->ifma_addr->sa_family == AF_LINK && sa == 0)
-               dlil_ioctl(0, ifp, SIOCDELMULTI, 0);
-       FREE(ifma->ifma_addr, M_IFMADDR);
-       FREE(ifma, M_IFMADDR);
-       if (sa == 0)
-               return 0;
+       if (lladdr || ll_firstref)
+               (void) ifnet_ioctl(ifp, 0, SIOCADDMULTI, NULL);
 
-       /*
-        * 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.)
-        */
-       if (ifp)
-               ifma = ifp->if_multiaddrs.lh_first;
-       else
-               ifma = ifma_lostlist.lh_first;
-       for (; ifma; ifma = ifma->ifma_link.le_next)
-               if (equal(sa, ifma->ifma_addr))
-                       break;
-       
-       FREE(sa, M_IFMADDR);
-       if (ifma == 0) {
-               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 if_delmultiaddr(ifma);
+       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;
-
-       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;
-       
-       return if_delmultiaddr(ifma);
+       return (if_delmulti_common(NULL, ifp, sa, 0));
 }
 
+/*
+ * 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));
+}
 
 /*
- * We don't use if_setlladdr, our interfaces are responsible for
- * handling the SIOCSIFLLADDR ioctl.
+ * 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(), if_bond.
  */
-#ifndef __APPLE__
 int
-if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
+if_delmulti_anon(struct ifnet *ifp, const struct sockaddr *sa)
 {
-       ...
+       return (if_delmulti_common(NULL, ifp, sa, 1));
 }
-#endif
 
-struct ifmultiaddr *
-ifmaof_ifpforaddr(sa, ifp)
-       struct sockaddr *sa;
-       struct ifnet *ifp;
+/*
+ * 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 ifmultiaddr *ifma;
-       
-       for (ifma = ifp->if_multiaddrs.lh_first; ifma;
-            ifma = ifma->ifma_link.le_next)
-               if (equal(ifma->ifma_addr, sa))
+       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;
+       }
+
+       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 (!ifa_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 == 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);
 
-       return ifma;
-}
+       if (lastref)
+               rt_newmaddrmsg(RTM_DELMADDR, ifma);
+
+       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);
+       }
 
-SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
-SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
+       if (lastref)
+               IFMA_REMREF(ifma);      /* for if_multiaddrs list */
+       if (ll_lastref)
+               IFMA_REMREF(ll);        /* for if_multiaddrs list */
 
+       IFMA_REMREF(ifma);              /* for this routine */
+       if (dupsa != NULL)
+               FREE(dupsa, M_IFADDR);
+
+       return (0);
+}
 
 /*
  * 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);
 }
 
 /*
  * Delete Routes for a Network Interface
- * 
+ *
  * Called for each routing entry via the rnh->rnh_walktree() call above
  * to delete all route entries referencing a detaching network interface.
  *
@@ -1743,44 +4542,673 @@ 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)
+void
+if_rtproto_del(struct ifnet *ifp, int protocol)
 {
-       
-        struct radix_node_head  *rnh;
+       struct radix_node_head  *rnh;
 
-       if ((protocol <= AF_MAX) && ((rnh = rt_tables[protocol]) != NULL) && (ifp != NULL))
+       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);
+       }
+       routegenid_update();
+}
+
+__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);
+
+       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_data->ifi_lastchange.tv_sec += boottime_sec();
+
+       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_data64->ifi_lastchange.tv_sec += boottime_sec();
+
+#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)
+
+       bzero(if_tc, sizeof (*if_tc));
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ibepackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ibebytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_obepackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_obebytes);
+       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);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ipvpackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_ipvbytes);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_opvpackets);
+       COPY_IF_TC_FIELD64_ATOMIC(ifi_opvbytes);
+
+#undef COPY_IF_TC_FIELD64_ATOMIC
+}
+
+void
+if_copy_data_extended(struct ifnet *ifp, struct if_data_extended *if_de)
+{
+#define        COPY_IF_DE_FIELD64_ATOMIC(fld) do {                     \
+       atomic_get_64(if_de->fld,                               \
+           (u_int64_t *)(void *)(uintptr_t)&ifp->if_data.fld); \
+} while (0)
+
+       bzero(if_de, sizeof (*if_de));
+       COPY_IF_DE_FIELD64_ATOMIC(ifi_alignerrs);
+       COPY_IF_DE_FIELD64_ATOMIC(ifi_dt_bytes);
+       COPY_IF_DE_FIELD64_ATOMIC(ifi_fpackets);
+       COPY_IF_DE_FIELD64_ATOMIC(ifi_fbytes);
+
+#undef COPY_IF_DE_FIELD64_ATOMIC
+}
+
+void
+if_copy_packet_stats(struct ifnet *ifp, struct if_packet_stats *if_ps)
+{
+#define        COPY_IF_PS_TCP_FIELD64_ATOMIC(fld) do {                         \
+       atomic_get_64(if_ps->ifi_tcp_##fld,                             \
+           (u_int64_t *)(void *)(uintptr_t)&ifp->if_tcp_stat->fld);    \
+} while (0)
+
+#define        COPY_IF_PS_UDP_FIELD64_ATOMIC(fld) do {                         \
+       atomic_get_64(if_ps->ifi_udp_##fld,                             \
+           (u_int64_t *)(void *)(uintptr_t)&ifp->if_udp_stat->fld);    \
+} while (0)
+
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(badformat);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(unspecv6);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(synfin);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(badformatipsec);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(noconnnolist);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(noconnlist);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(listbadsyn);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(icmp6unreach);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(deprecate6);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(ooopacket);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(rstinsynrcv);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(dospacket);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(cleanup);
+       COPY_IF_PS_TCP_FIELD64_ATOMIC(synwindow);
+
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(port_unreach);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(faithprefix);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(port0);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(badlength);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(badchksum);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(badmcast);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(cleanup);
+       COPY_IF_PS_UDP_FIELD64_ATOMIC(badipsec);
+
+#undef COPY_IF_PS_TCP_FIELD64_ATOMIC
+#undef COPY_IF_PS_UDP_FIELD64_ATOMIC
+}
+
+void
+if_copy_rxpoll_stats(struct ifnet *ifp, struct if_rxpoll_stats *if_rs)
+{
+       bzero(if_rs, sizeof (*if_rs));
+       if (!(ifp->if_eflags & IFEF_RXPOLL) || !ifnet_is_attached(ifp, 1))
+               return;
+
+       /* by now, ifnet will stay attached so if_inp must be valid */
+       VERIFY(ifp->if_inp != NULL);
+       bcopy(&ifp->if_inp->pstats, if_rs, sizeof (*if_rs));
+
+       /* Release the IO refcnt */
+       ifnet_decr_iorefcnt(ifp);
+}
+
+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.  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);
+}
+
+/*
+ * 'i' group ioctls.
+ *
+ * The switch statement below does nothing at runtime, as it serves as a
+ * compile time check to ensure that all of the socket 'i' ioctls (those
+ * in the 'i' group going thru soo_ioctl) that are made available by the
+ * networking stack is unique.  This works as long as this routine gets
+ * updated each time a new interface ioctl gets added.
+ *
+ * Any failures at compile time indicates duplicated ioctl values.
+ */
+static __attribute__((unused)) void
+ifioctl_cassert(void)
+{
+       /*
+        * This is equivalent to _CASSERT() and the compiler wouldn't
+        * generate any instructions, thus for compile time only.
+        */
+       switch ((u_long)0) {
+       case 0:
+
+       /* bsd/net/if_ppp.h */
+       case SIOCGPPPSTATS:
+       case SIOCGPPPCSTATS:
+
+#if INET6
+       /* bsd/netinet6/in6_var.h */
+       case SIOCSIFADDR_IN6:
+       case SIOCGIFADDR_IN6:
+       case SIOCSIFDSTADDR_IN6:
+       case SIOCSIFNETMASK_IN6:
+       case SIOCGIFDSTADDR_IN6:
+       case SIOCGIFNETMASK_IN6:
+       case SIOCDIFADDR_IN6:
+       case SIOCAIFADDR_IN6_32:
+       case SIOCAIFADDR_IN6_64:
+       case SIOCSIFPHYADDR_IN6_32:
+       case SIOCSIFPHYADDR_IN6_64:
+       case SIOCGIFPSRCADDR_IN6:
+       case SIOCGIFPDSTADDR_IN6:
+       case SIOCGIFAFLAG_IN6:
+       case SIOCGDRLST_IN6_32:
+       case SIOCGDRLST_IN6_64:
+       case SIOCGPRLST_IN6_32:
+       case SIOCGPRLST_IN6_64:
+       case OSIOCGIFINFO_IN6:
+       case SIOCGIFINFO_IN6:
+       case SIOCSNDFLUSH_IN6:
+       case SIOCGNBRINFO_IN6_32:
+       case SIOCGNBRINFO_IN6_64:
+       case SIOCSPFXFLUSH_IN6:
+       case SIOCSRTRFLUSH_IN6:
+       case SIOCGIFALIFETIME_IN6:
+       case SIOCSIFALIFETIME_IN6:
+       case SIOCGIFSTAT_IN6:
+       case SIOCGIFSTAT_ICMP6:
+       case SIOCSDEFIFACE_IN6_32:
+       case SIOCSDEFIFACE_IN6_64:
+       case SIOCGDEFIFACE_IN6_32:
+       case SIOCGDEFIFACE_IN6_64:
+       case SIOCSIFINFO_FLAGS:
+       case SIOCSSCOPE6:
+       case SIOCGSCOPE6:
+       case SIOCGSCOPE6DEF:
+       case SIOCSIFPREFIX_IN6:
+       case SIOCGIFPREFIX_IN6:
+       case SIOCDIFPREFIX_IN6:
+       case SIOCAIFPREFIX_IN6:
+       case SIOCCIFPREFIX_IN6:
+       case SIOCSGIFPREFIX_IN6:
+       case SIOCPROTOATTACH_IN6_32:
+       case SIOCPROTOATTACH_IN6_64:
+       case SIOCPROTODETACH_IN6:
+       case SIOCLL_START_32:
+       case SIOCLL_START_64:
+       case SIOCLL_STOP:
+       case SIOCAUTOCONF_START:
+       case SIOCAUTOCONF_STOP:
+       case SIOCSETROUTERMODE_IN6:
+       case SIOCLL_CGASTART_32:
+       case SIOCLL_CGASTART_64:
+       case SIOCGIFCGAPREP_IN6:
+       case SIOCSIFCGAPREP_IN6:
+#endif /* INET6 */
+
+       /* bsd/sys/sockio.h */
+       case SIOCSIFADDR:
+       case OSIOCGIFADDR:
+       case SIOCSIFDSTADDR:
+       case OSIOCGIFDSTADDR:
+       case SIOCSIFFLAGS:
+       case SIOCGIFFLAGS:
+       case OSIOCGIFBRDADDR:
+       case SIOCSIFBRDADDR:
+       case OSIOCGIFCONF32:
+       case OSIOCGIFCONF64:
+       case OSIOCGIFNETMASK:
+       case SIOCSIFNETMASK:
+       case SIOCGIFMETRIC:
+       case SIOCSIFMETRIC:
+       case SIOCDIFADDR:
+       case SIOCAIFADDR:
+
+       case SIOCGIFADDR:
+       case SIOCGIFDSTADDR:
+       case SIOCGIFBRDADDR:
+       case SIOCGIFCONF32:
+       case SIOCGIFCONF64:
+       case SIOCGIFNETMASK:
+       case SIOCAUTOADDR:
+       case SIOCAUTONETMASK:
+       case SIOCARPIPLL:
+
+       case SIOCADDMULTI:
+       case SIOCDELMULTI:
+       case SIOCGIFMTU:
+       case SIOCSIFMTU:
+       case SIOCGIFPHYS:
+       case SIOCSIFPHYS:
+       case SIOCSIFMEDIA:
+       case SIOCGIFMEDIA32:
+       case SIOCGIFMEDIA64:
+       case SIOCSIFGENERIC:
+       case SIOCGIFGENERIC:
+       case SIOCRSLVMULTI:
+
+       case SIOCSIFLLADDR:
+       case SIOCGIFSTATUS:
+       case SIOCSIFPHYADDR:
+       case SIOCGIFPSRCADDR:
+       case SIOCGIFPDSTADDR:
+       case SIOCDIFPHYADDR:
+
+       case SIOCGIFDEVMTU:
+       case SIOCSIFALTMTU:
+       case SIOCGIFALTMTU:
+       case SIOCSIFBOND:
+       case SIOCGIFBOND:
+
+       case SIOCPROTOATTACH:
+       case SIOCPROTODETACH:
+
+       case SIOCSIFCAP:
+       case SIOCGIFCAP:
+
+       case SIOCIFCREATE:
+       case SIOCIFDESTROY:
+       case SIOCIFCREATE2:
+
+       case SIOCSDRVSPEC32:
+       case SIOCGDRVSPEC32:
+       case SIOCSDRVSPEC64:
+       case SIOCGDRVSPEC64:
+
+       case SIOCSIFVLAN:
+       case SIOCGIFVLAN:
+
+       case SIOCIFGCLONERS32:
+       case SIOCIFGCLONERS64:
+
+       case SIOCGIFASYNCMAP:
+       case SIOCSIFASYNCMAP:
+#if CONFIG_MACF_NET
+       case SIOCGIFMAC:
+       case SIOCSIFMAC:
+#endif /* CONFIG_MACF_NET */
+       case SIOCSIFKPI:
+       case SIOCGIFKPI:
+
+       case SIOCGIFWAKEFLAGS:
+
+       case SIOCGIFGETRTREFCNT:
+       case SIOCGIFLINKQUALITYMETRIC:
+       case SIOCSIFOPPORTUNISTIC:
+       case SIOCGIFOPPORTUNISTIC:
+       case SIOCSETROUTERMODE:
+       case SIOCGIFEFLAGS:
+       case SIOCSIFDESC:
+       case SIOCGIFDESC:
+       case SIOCSIFLINKPARAMS:
+       case SIOCGIFLINKPARAMS:
+       case SIOCGIFQUEUESTATS:
+       case SIOCSIFTHROTTLE:
+       case SIOCGIFTHROTTLE:
+
+       case SIOCGASSOCIDS32:
+       case SIOCGASSOCIDS64:
+       case SIOCGCONNIDS32:
+       case SIOCGCONNIDS64:
+       case SIOCGCONNINFO32:
+       case SIOCGCONNINFO64:
+       case SIOCSCONNORDER:
+       case SIOCGCONNORDER:
+
+       case SIOCSIFLOG:
+       case SIOCGIFLOG:
+       case SIOCGIFDELEGATE:
+       case SIOCGIFLLADDR:
+       case SIOCGIFTYPE:
+       case SIOCGIFEXPENSIVE:
+       case SIOCSIFEXPENSIVE:
+       case SIOCGIF2KCL:
+       case SIOCSIF2KCL:
+       case SIOCGSTARTDELAY:
+
+       case SIOCAIFAGENTID:
+       case SIOCDIFAGENTID:
+       case SIOCGIFAGENTIDS32:
+       case SIOCGIFAGENTIDS64:
+       case SIOCGIFAGENTDATA32:
+       case SIOCGIFAGENTDATA64:
+       case SIOCGIFAGENTLIST32:
+       case SIOCGIFAGENTLIST64:
+
+
+       case SIOCSIFINTERFACESTATE:
+       case SIOCGIFINTERFACESTATE:
+       case SIOCSIFPROBECONNECTIVITY:
+       case SIOCGIFPROBECONNECTIVITY:
+
+       case SIOCGIFFUNCTIONALTYPE:
+       case SIOCSIFNETSIGNATURE:
+       case SIOCGIFNETSIGNATURE:
+
+       case SIOCGECNMODE:
+       case SIOCSECNMODE:
+
+       case SIOCSQOSMARKINGMODE:
+       case SIOCSQOSMARKINGENABLED:
+       case SIOCGQOSMARKINGMODE:
+       case SIOCGQOSMARKINGENABLED:
+
+       case SIOCGIFPROTOLIST32:
+       case SIOCGIFPROTOLIST64:
+               ;
+       }
+}
+
+uint32_t
+ifnet_mbuf_packetpreamblelen(struct ifnet *ifp)
+{
+#pragma unused(ifp)
+       return (0);
+}
+
+/* The following is used to enqueue work items for interface events */
+struct intf_event {
+       struct ifnet *ifp;
+       union sockaddr_in_4_6 addr;
+       uint32_t intf_event_code;
+};
+
+static void
+intf_event_callback(void *arg)
+{
+       struct intf_event *p_intf_ev = (struct intf_event *)arg;
+
+       /* Call this before we walk the tree */
+       EVENTHANDLER_INVOKE(&ifnet_evhdlr_ctxt, ifnet_event, p_intf_ev->ifp,
+           (struct sockaddr *)&(p_intf_ev->addr), p_intf_ev->intf_event_code);
+}
+
+struct intf_event_nwk_wq_entry {
+       struct nwk_wq_entry nwk_wqe;
+       struct intf_event intf_ev_arg;
+};
+
+void
+intf_event_enqueue_nwk_wq_entry(struct ifnet *ifp, struct sockaddr *addrp,
+    uint32_t intf_event_code)
+{
+#pragma unused(addrp)
+       struct intf_event_nwk_wq_entry *p_intf_ev = NULL;
+
+       MALLOC(p_intf_ev, struct intf_event_nwk_wq_entry *,
+           sizeof(struct intf_event_nwk_wq_entry),
+           M_NWKWQ, M_WAITOK | M_ZERO);
+
+       p_intf_ev->intf_ev_arg.ifp = ifp;
+       /*
+        * XXX Not using addr in the arg. This will be used
+        * once we need IP address add/delete events
+        */
+       p_intf_ev->intf_ev_arg.intf_event_code = intf_event_code;
+       p_intf_ev->nwk_wqe.func = intf_event_callback;
+       p_intf_ev->nwk_wqe.is_arg_managed = TRUE;
+       p_intf_ev->nwk_wqe.arg = &p_intf_ev->intf_ev_arg;
+       nwk_wq_enqueue((struct nwk_wq_entry*)p_intf_ev);
 }