xref: /dragonfly/sys/net/if.c (revision 8a7bdfea)
1 /*
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)if.c	8.3 (Berkeley) 1/4/94
34  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
35  * $DragonFly: src/sys/net/if.c,v 1.63 2008/03/29 03:38:53 sephe Exp $
36  */
37 
38 #include "opt_compat.h"
39 #include "opt_inet6.h"
40 #include "opt_inet.h"
41 #include "opt_polling.h"
42 
43 #include <sys/param.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.h>
47 #include <sys/proc.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/socketops.h>
52 #include <sys/protosw.h>
53 #include <sys/kernel.h>
54 #include <sys/sockio.h>
55 #include <sys/syslog.h>
56 #include <sys/sysctl.h>
57 #include <sys/domain.h>
58 #include <sys/thread.h>
59 #include <sys/thread2.h>
60 #include <sys/serialize.h>
61 #include <sys/msgport2.h>
62 
63 #include <net/if.h>
64 #include <net/if_arp.h>
65 #include <net/if_dl.h>
66 #include <net/if_types.h>
67 #include <net/if_var.h>
68 #include <net/ifq_var.h>
69 #include <net/radix.h>
70 #include <net/route.h>
71 #include <net/if_clone.h>
72 #include <net/netmsg2.h>
73 
74 #include <machine/stdarg.h>
75 #include <machine/smp.h>
76 
77 #if defined(INET) || defined(INET6)
78 /*XXX*/
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 #include <netinet/if_ether.h>
82 #ifdef INET6
83 #include <netinet6/in6_var.h>
84 #include <netinet6/in6_ifattach.h>
85 #endif
86 #endif
87 
88 #if defined(COMPAT_43)
89 #include <emulation/43bsd/43bsd_socket.h>
90 #endif /* COMPAT_43 */
91 
92 struct netmsg_ifaddr {
93 	struct netmsg	netmsg;
94 	struct ifaddr	*ifa;
95 	struct ifnet	*ifp;
96 	int		tail;
97 };
98 
99 /*
100  * Support for non-ALTQ interfaces.
101  */
102 static int	ifq_classic_enqueue(struct ifaltq *, struct mbuf *,
103 				    struct altq_pktattr *);
104 static struct mbuf *
105 		ifq_classic_dequeue(struct ifaltq *, struct mbuf *, int);
106 static int	ifq_classic_request(struct ifaltq *, int, void *);
107 
108 /*
109  * System initialization
110  */
111 static void	if_attachdomain(void *);
112 static void	if_attachdomain1(struct ifnet *);
113 static int	ifconf(u_long, caddr_t, struct ucred *);
114 static void	ifinit(void *);
115 static void	ifaddrinit(void *);
116 static void	if_slowtimo(void *);
117 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
118 static int	if_rtdel(struct radix_node *, void *);
119 
120 #ifdef INET6
121 /*
122  * XXX: declare here to avoid to include many inet6 related files..
123  * should be more generalized?
124  */
125 extern void	nd6_setmtu(struct ifnet *);
126 #endif
127 
128 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
129 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
130 
131 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL)
132 /* Must be after netisr_init */
133 SYSINIT(ifaddr, SI_SUB_PRE_DRIVERS, SI_ORDER_SECOND, ifaddrinit, NULL)
134 
135 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
136 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
137 
138 int			ifqmaxlen = IFQ_MAXLEN;
139 struct ifnethead	ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
140 
141 struct callout		if_slowtimo_timer;
142 
143 int			if_index = 0;
144 struct ifnet		**ifindex2ifnet = NULL;
145 static struct thread	ifaddr_threads[MAXCPU];
146 
147 /*
148  * Network interface utility routines.
149  *
150  * Routines with ifa_ifwith* names take sockaddr *'s as
151  * parameters.
152  */
153 /* ARGSUSED*/
154 void
155 ifinit(void *dummy)
156 {
157 	struct ifnet *ifp;
158 
159 	callout_init(&if_slowtimo_timer);
160 
161 	crit_enter();
162 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
163 		if (ifp->if_snd.ifq_maxlen == 0) {
164 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
165 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
166 		}
167 	}
168 	crit_exit();
169 
170 	if_slowtimo(0);
171 }
172 
173 /*
174  * Attach an interface to the list of "active" interfaces.
175  *
176  * The serializer is optional.  If non-NULL access to the interface
177  * may be MPSAFE.
178  */
179 void
180 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
181 {
182 	unsigned socksize, ifasize;
183 	int namelen, masklen;
184 	struct sockaddr_dl *sdl;
185 	struct ifaddr *ifa;
186 	struct ifaltq *ifq;
187 	int i;
188 
189 	static int if_indexlim = 8;
190 
191 	/*
192 	 * The serializer can be passed in from the device, allowing the
193 	 * same serializer to be used for both the interrupt interlock and
194 	 * the device queue.  If not specified, the netif structure will
195 	 * use an embedded serializer.
196 	 */
197 	if (serializer == NULL) {
198 		serializer = &ifp->if_default_serializer;
199 		lwkt_serialize_init(serializer);
200 	}
201 	ifp->if_serializer = serializer;
202 
203 #ifdef DEVICE_POLLING
204 	/* Device is not in polling mode by default */
205 	ifp->if_poll_cpuid = -1;
206 #endif
207 
208 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
209 	ifp->if_index = ++if_index;
210 
211 	/*
212 	 * XXX -
213 	 * The old code would work if the interface passed a pre-existing
214 	 * chain of ifaddrs to this code.  We don't trust our callers to
215 	 * properly initialize the tailq, however, so we no longer allow
216 	 * this unlikely case.
217 	 */
218 	ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
219 				    M_IFADDR, M_WAITOK | M_ZERO);
220 	for (i = 0; i < ncpus; ++i)
221 		TAILQ_INIT(&ifp->if_addrheads[i]);
222 
223 	TAILQ_INIT(&ifp->if_prefixhead);
224 	LIST_INIT(&ifp->if_multiaddrs);
225 	getmicrotime(&ifp->if_lastchange);
226 	if (ifindex2ifnet == NULL || if_index >= if_indexlim) {
227 		unsigned int n;
228 		struct ifnet **q;
229 
230 		if_indexlim <<= 1;
231 
232 		/* grow ifindex2ifnet */
233 		n = if_indexlim * sizeof(*q);
234 		q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
235 		if (ifindex2ifnet) {
236 			bcopy(ifindex2ifnet, q, n/2);
237 			kfree(ifindex2ifnet, M_IFADDR);
238 		}
239 		ifindex2ifnet = q;
240 	}
241 
242 	ifindex2ifnet[if_index] = ifp;
243 
244 	/*
245 	 * create a Link Level name for this device
246 	 */
247 	namelen = strlen(ifp->if_xname);
248 #define _offsetof(t, m) ((int)((caddr_t)&((t *)0)->m))
249 	masklen = _offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
250 	socksize = masklen + ifp->if_addrlen;
251 #define ROUNDUP(a) (1 + (((a) - 1) | (sizeof(long) - 1)))
252 	if (socksize < sizeof(*sdl))
253 		socksize = sizeof(*sdl);
254 	socksize = ROUNDUP(socksize);
255 	ifasize = sizeof(struct ifaddr) + 2 * socksize;
256 	ifa = ifa_create(ifasize, M_WAITOK);
257 	sdl = (struct sockaddr_dl *)(ifa + 1);
258 	sdl->sdl_len = socksize;
259 	sdl->sdl_family = AF_LINK;
260 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
261 	sdl->sdl_nlen = namelen;
262 	sdl->sdl_index = ifp->if_index;
263 	sdl->sdl_type = ifp->if_type;
264 	ifp->if_lladdr = ifa;
265 	ifa->ifa_ifp = ifp;
266 	ifa->ifa_rtrequest = link_rtrequest;
267 	ifa->ifa_addr = (struct sockaddr *)sdl;
268 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
269 	ifa->ifa_netmask = (struct sockaddr *)sdl;
270 	sdl->sdl_len = masklen;
271 	while (namelen != 0)
272 		sdl->sdl_data[--namelen] = 0xff;
273 	ifa_iflink(ifa, ifp, 0 /* Insert head */);
274 
275 	EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
276 
277 	ifq = &ifp->if_snd;
278 	ifq->altq_type = 0;
279 	ifq->altq_disc = NULL;
280 	ifq->altq_flags &= ALTQF_CANTCHANGE;
281 	ifq->altq_tbr = NULL;
282 	ifq->altq_ifp = ifp;
283 	ifq_set_classic(ifq);
284 
285 	if (!SLIST_EMPTY(&domains))
286 		if_attachdomain1(ifp);
287 
288 	/* Announce the interface. */
289 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
290 }
291 
292 static void
293 if_attachdomain(void *dummy)
294 {
295 	struct ifnet *ifp;
296 
297 	crit_enter();
298 	TAILQ_FOREACH(ifp, &ifnet, if_list)
299 		if_attachdomain1(ifp);
300 	crit_exit();
301 }
302 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
303 	if_attachdomain, NULL);
304 
305 static void
306 if_attachdomain1(struct ifnet *ifp)
307 {
308 	struct domain *dp;
309 
310 	crit_enter();
311 
312 	/* address family dependent data region */
313 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
314 	SLIST_FOREACH(dp, &domains, dom_next)
315 		if (dp->dom_ifattach)
316 			ifp->if_afdata[dp->dom_family] =
317 				(*dp->dom_ifattach)(ifp);
318 	crit_exit();
319 }
320 
321 /*
322  * Purge all addresses whose type is _not_ AF_LINK
323  */
324 void
325 if_purgeaddrs_nolink(struct ifnet *ifp)
326 {
327 	struct ifaddr_container *ifac, *next;
328 
329 	TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
330 			      ifa_link, next) {
331 		struct ifaddr *ifa = ifac->ifa;
332 
333 		/* Leave link ifaddr as it is */
334 		if (ifa->ifa_addr->sa_family == AF_LINK)
335 			continue;
336 #ifdef INET
337 		/* XXX: Ugly!! ad hoc just for INET */
338 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
339 			struct ifaliasreq ifr;
340 #ifdef IFADDR_DEBUG_VERBOSE
341 			int i;
342 
343 			kprintf("purge in4 addr %p: ", ifa);
344 			for (i = 0; i < ncpus; ++i)
345 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
346 			kprintf("\n");
347 #endif
348 
349 			bzero(&ifr, sizeof ifr);
350 			ifr.ifra_addr = *ifa->ifa_addr;
351 			if (ifa->ifa_dstaddr)
352 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
353 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
354 				       NULL) == 0)
355 				continue;
356 		}
357 #endif /* INET */
358 #ifdef INET6
359 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
360 #ifdef IFADDR_DEBUG_VERBOSE
361 			int i;
362 
363 			kprintf("purge in6 addr %p: ", ifa);
364 			for (i = 0; i < ncpus; ++i)
365 				kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
366 			kprintf("\n");
367 #endif
368 
369 			in6_purgeaddr(ifa);
370 			/* ifp_addrhead is already updated */
371 			continue;
372 		}
373 #endif /* INET6 */
374 		ifa_ifunlink(ifa, ifp);
375 		ifa_destroy(ifa);
376 	}
377 }
378 
379 /*
380  * Detach an interface, removing it from the
381  * list of "active" interfaces.
382  */
383 void
384 if_detach(struct ifnet *ifp)
385 {
386 	struct radix_node_head	*rnh;
387 	int i;
388 	int cpu, origcpu;
389 	struct domain *dp;
390 
391 	EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
392 
393 	/*
394 	 * Remove routes and flush queues.
395 	 */
396 	crit_enter();
397 #ifdef DEVICE_POLLING
398 	if (ifp->if_flags & IFF_POLLING)
399 		ether_poll_deregister(ifp);
400 #endif
401 	if_down(ifp);
402 
403 	if (ifq_is_enabled(&ifp->if_snd))
404 		altq_disable(&ifp->if_snd);
405 	if (ifq_is_attached(&ifp->if_snd))
406 		altq_detach(&ifp->if_snd);
407 
408 	/*
409 	 * Clean up all addresses.
410 	 */
411 	ifp->if_lladdr = NULL;
412 
413 	if_purgeaddrs_nolink(ifp);
414 	if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
415 		struct ifaddr *ifa;
416 
417 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
418 		KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
419 			("non-link ifaddr is left on if_addrheads"));
420 
421 		ifa_ifunlink(ifa, ifp);
422 		ifa_destroy(ifa);
423 		KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
424 			("there are still ifaddrs left on if_addrheads"));
425 	}
426 
427 #ifdef INET
428 	/*
429 	 * Remove all IPv4 kernel structures related to ifp.
430 	 */
431 	in_ifdetach(ifp);
432 #endif
433 
434 #ifdef INET6
435 	/*
436 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
437 	 * before removing routing entries below, since IPv6 interface direct
438 	 * routes are expected to be removed by the IPv6-specific kernel API.
439 	 * Otherwise, the kernel will detect some inconsistency and bark it.
440 	 */
441 	in6_ifdetach(ifp);
442 #endif
443 
444 	/*
445 	 * Delete all remaining routes using this interface
446 	 * Unfortuneatly the only way to do this is to slog through
447 	 * the entire routing table looking for routes which point
448 	 * to this interface...oh well...
449 	 */
450 	origcpu = mycpuid;
451 	for (cpu = 0; cpu < ncpus2; cpu++) {
452 		lwkt_migratecpu(cpu);
453 		for (i = 1; i <= AF_MAX; i++) {
454 			if ((rnh = rt_tables[cpu][i]) == NULL)
455 				continue;
456 			rnh->rnh_walktree(rnh, if_rtdel, ifp);
457 		}
458 	}
459 	lwkt_migratecpu(origcpu);
460 
461 	/* Announce that the interface is gone. */
462 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
463 
464 	SLIST_FOREACH(dp, &domains, dom_next)
465 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
466 			(*dp->dom_ifdetach)(ifp,
467 				ifp->if_afdata[dp->dom_family]);
468 
469 	/*
470 	 * Remove interface from ifindex2ifp[] and maybe decrement if_index.
471 	 */
472 	ifindex2ifnet[ifp->if_index] = NULL;
473 	while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
474 		if_index--;
475 
476 	TAILQ_REMOVE(&ifnet, ifp, if_link);
477 	kfree(ifp->if_addrheads, M_IFADDR);
478 	crit_exit();
479 }
480 
481 /*
482  * Delete Routes for a Network Interface
483  *
484  * Called for each routing entry via the rnh->rnh_walktree() call above
485  * to delete all route entries referencing a detaching network interface.
486  *
487  * Arguments:
488  *	rn	pointer to node in the routing table
489  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
490  *
491  * Returns:
492  *	0	successful
493  *	errno	failed - reason indicated
494  *
495  */
496 static int
497 if_rtdel(struct radix_node *rn, void *arg)
498 {
499 	struct rtentry	*rt = (struct rtentry *)rn;
500 	struct ifnet	*ifp = arg;
501 	int		err;
502 
503 	if (rt->rt_ifp == ifp) {
504 
505 		/*
506 		 * Protect (sorta) against walktree recursion problems
507 		 * with cloned routes
508 		 */
509 		if (!(rt->rt_flags & RTF_UP))
510 			return (0);
511 
512 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
513 				rt_mask(rt), rt->rt_flags,
514 				(struct rtentry **) NULL);
515 		if (err) {
516 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
517 		}
518 	}
519 
520 	return (0);
521 }
522 
523 /*
524  * Locate an interface based on a complete address.
525  */
526 struct ifaddr *
527 ifa_ifwithaddr(struct sockaddr *addr)
528 {
529 	struct ifnet *ifp;
530 
531 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
532 		struct ifaddr_container *ifac;
533 
534 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
535 			struct ifaddr *ifa = ifac->ifa;
536 
537 			if (ifa->ifa_addr->sa_family != addr->sa_family)
538 				continue;
539 			if (sa_equal(addr, ifa->ifa_addr))
540 				return (ifa);
541 			if ((ifp->if_flags & IFF_BROADCAST) &&
542 			    ifa->ifa_broadaddr &&
543 			    /* IPv6 doesn't have broadcast */
544 			    ifa->ifa_broadaddr->sa_len != 0 &&
545 			    sa_equal(ifa->ifa_broadaddr, addr))
546 				return (ifa);
547 		}
548 	}
549 	return (NULL);
550 }
551 /*
552  * Locate the point to point interface with a given destination address.
553  */
554 struct ifaddr *
555 ifa_ifwithdstaddr(struct sockaddr *addr)
556 {
557 	struct ifnet *ifp;
558 
559 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
560 		struct ifaddr_container *ifac;
561 
562 		if (!(ifp->if_flags & IFF_POINTOPOINT))
563 			continue;
564 
565 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
566 			struct ifaddr *ifa = ifac->ifa;
567 
568 			if (ifa->ifa_addr->sa_family != addr->sa_family)
569 				continue;
570 			if (ifa->ifa_dstaddr &&
571 			    sa_equal(addr, ifa->ifa_dstaddr))
572 				return (ifa);
573 		}
574 	}
575 	return (NULL);
576 }
577 
578 /*
579  * Find an interface on a specific network.  If many, choice
580  * is most specific found.
581  */
582 struct ifaddr *
583 ifa_ifwithnet(struct sockaddr *addr)
584 {
585 	struct ifnet *ifp;
586 	struct ifaddr *ifa_maybe = NULL;
587 	u_int af = addr->sa_family;
588 	char *addr_data = addr->sa_data, *cplim;
589 
590 	/*
591 	 * AF_LINK addresses can be looked up directly by their index number,
592 	 * so do that if we can.
593 	 */
594 	if (af == AF_LINK) {
595 		struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
596 
597 		if (sdl->sdl_index && sdl->sdl_index <= if_index)
598 			return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
599 	}
600 
601 	/*
602 	 * Scan though each interface, looking for ones that have
603 	 * addresses in this address family.
604 	 */
605 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
606 		struct ifaddr_container *ifac;
607 
608 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
609 			struct ifaddr *ifa = ifac->ifa;
610 			char *cp, *cp2, *cp3;
611 
612 			if (ifa->ifa_addr->sa_family != af)
613 next:				continue;
614 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
615 				/*
616 				 * This is a bit broken as it doesn't
617 				 * take into account that the remote end may
618 				 * be a single node in the network we are
619 				 * looking for.
620 				 * The trouble is that we don't know the
621 				 * netmask for the remote end.
622 				 */
623 				if (ifa->ifa_dstaddr != NULL &&
624 				    sa_equal(addr, ifa->ifa_dstaddr))
625 					return (ifa);
626 			} else {
627 				/*
628 				 * if we have a special address handler,
629 				 * then use it instead of the generic one.
630 				 */
631 				if (ifa->ifa_claim_addr) {
632 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
633 						return (ifa);
634 					} else {
635 						continue;
636 					}
637 				}
638 
639 				/*
640 				 * Scan all the bits in the ifa's address.
641 				 * If a bit dissagrees with what we are
642 				 * looking for, mask it with the netmask
643 				 * to see if it really matters.
644 				 * (A byte at a time)
645 				 */
646 				if (ifa->ifa_netmask == 0)
647 					continue;
648 				cp = addr_data;
649 				cp2 = ifa->ifa_addr->sa_data;
650 				cp3 = ifa->ifa_netmask->sa_data;
651 				cplim = ifa->ifa_netmask->sa_len +
652 					(char *)ifa->ifa_netmask;
653 				while (cp3 < cplim)
654 					if ((*cp++ ^ *cp2++) & *cp3++)
655 						goto next; /* next address! */
656 				/*
657 				 * If the netmask of what we just found
658 				 * is more specific than what we had before
659 				 * (if we had one) then remember the new one
660 				 * before continuing to search
661 				 * for an even better one.
662 				 */
663 				if (ifa_maybe == 0 ||
664 				    rn_refines((char *)ifa->ifa_netmask,
665 					       (char *)ifa_maybe->ifa_netmask))
666 					ifa_maybe = ifa;
667 			}
668 		}
669 	}
670 	return (ifa_maybe);
671 }
672 
673 /*
674  * Find an interface address specific to an interface best matching
675  * a given address.
676  */
677 struct ifaddr *
678 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
679 {
680 	struct ifaddr_container *ifac;
681 	char *cp, *cp2, *cp3;
682 	char *cplim;
683 	struct ifaddr *ifa_maybe = 0;
684 	u_int af = addr->sa_family;
685 
686 	if (af >= AF_MAX)
687 		return (0);
688 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
689 		struct ifaddr *ifa = ifac->ifa;
690 
691 		if (ifa->ifa_addr->sa_family != af)
692 			continue;
693 		if (ifa_maybe == 0)
694 			ifa_maybe = ifa;
695 		if (ifa->ifa_netmask == NULL) {
696 			if (sa_equal(addr, ifa->ifa_addr) ||
697 			    (ifa->ifa_dstaddr != NULL &&
698 			     sa_equal(addr, ifa->ifa_dstaddr)))
699 				return (ifa);
700 			continue;
701 		}
702 		if (ifp->if_flags & IFF_POINTOPOINT) {
703 			if (sa_equal(addr, ifa->ifa_dstaddr))
704 				return (ifa);
705 		} else {
706 			cp = addr->sa_data;
707 			cp2 = ifa->ifa_addr->sa_data;
708 			cp3 = ifa->ifa_netmask->sa_data;
709 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
710 			for (; cp3 < cplim; cp3++)
711 				if ((*cp++ ^ *cp2++) & *cp3)
712 					break;
713 			if (cp3 == cplim)
714 				return (ifa);
715 		}
716 	}
717 	return (ifa_maybe);
718 }
719 
720 /*
721  * Default action when installing a route with a Link Level gateway.
722  * Lookup an appropriate real ifa to point to.
723  * This should be moved to /sys/net/link.c eventually.
724  */
725 static void
726 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
727 {
728 	struct ifaddr *ifa;
729 	struct sockaddr *dst;
730 	struct ifnet *ifp;
731 
732 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
733 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
734 		return;
735 	ifa = ifaof_ifpforaddr(dst, ifp);
736 	if (ifa != NULL) {
737 		IFAFREE(rt->rt_ifa);
738 		IFAREF(ifa);
739 		rt->rt_ifa = ifa;
740 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
741 			ifa->ifa_rtrequest(cmd, rt, info);
742 	}
743 }
744 
745 /*
746  * Mark an interface down and notify protocols of
747  * the transition.
748  * NOTE: must be called at splnet or eqivalent.
749  */
750 void
751 if_unroute(struct ifnet *ifp, int flag, int fam)
752 {
753 	struct ifaddr_container *ifac;
754 
755 	ifp->if_flags &= ~flag;
756 	getmicrotime(&ifp->if_lastchange);
757 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
758 		struct ifaddr *ifa = ifac->ifa;
759 
760 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
761 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
762 	}
763 	ifq_purge(&ifp->if_snd);
764 	rt_ifmsg(ifp);
765 }
766 
767 /*
768  * Mark an interface up and notify protocols of
769  * the transition.
770  * NOTE: must be called at splnet or eqivalent.
771  */
772 void
773 if_route(struct ifnet *ifp, int flag, int fam)
774 {
775 	struct ifaddr_container *ifac;
776 
777 	ifp->if_flags |= flag;
778 	getmicrotime(&ifp->if_lastchange);
779 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
780 		struct ifaddr *ifa = ifac->ifa;
781 
782 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
783 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
784 	}
785 	rt_ifmsg(ifp);
786 #ifdef INET6
787 	in6_if_up(ifp);
788 #endif
789 }
790 
791 /*
792  * Mark an interface down and notify protocols of the transition.  An
793  * interface going down is also considered to be a synchronizing event.
794  * We must ensure that all packet processing related to the interface
795  * has completed before we return so e.g. the caller can free the ifnet
796  * structure that the mbufs may be referencing.
797  *
798  * NOTE: must be called at splnet or eqivalent.
799  */
800 void
801 if_down(struct ifnet *ifp)
802 {
803 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
804 	netmsg_service_sync();
805 }
806 
807 /*
808  * Mark an interface up and notify protocols of
809  * the transition.
810  * NOTE: must be called at splnet or eqivalent.
811  */
812 void
813 if_up(struct ifnet *ifp)
814 {
815 
816 	if_route(ifp, IFF_UP, AF_UNSPEC);
817 }
818 
819 /*
820  * Process a link state change.
821  * NOTE: must be called at splsoftnet or equivalent.
822  */
823 void
824 if_link_state_change(struct ifnet *ifp)
825 {
826 	rt_ifmsg(ifp);
827 }
828 
829 /*
830  * Handle interface watchdog timer routines.  Called
831  * from softclock, we decrement timers (if set) and
832  * call the appropriate interface routine on expiration.
833  */
834 static void
835 if_slowtimo(void *arg)
836 {
837 	struct ifnet *ifp;
838 
839 	crit_enter();
840 
841 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
842 		if (ifp->if_timer == 0 || --ifp->if_timer)
843 			continue;
844 		if (ifp->if_watchdog) {
845 			if (lwkt_serialize_try(ifp->if_serializer)) {
846 				(*ifp->if_watchdog)(ifp);
847 				lwkt_serialize_exit(ifp->if_serializer);
848 			} else {
849 				/* try again next timeout */
850 				++ifp->if_timer;
851 			}
852 		}
853 	}
854 
855 	crit_exit();
856 
857 	callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
858 }
859 
860 /*
861  * Map interface name to
862  * interface structure pointer.
863  */
864 struct ifnet *
865 ifunit(const char *name)
866 {
867 	struct ifnet *ifp;
868 
869 	/*
870 	 * Search all the interfaces for this name/number
871 	 */
872 
873 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
874 		if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
875 			break;
876 	}
877 	return (ifp);
878 }
879 
880 
881 /*
882  * Map interface name in a sockaddr_dl to
883  * interface structure pointer.
884  */
885 struct ifnet *
886 if_withname(struct sockaddr *sa)
887 {
888 	char ifname[IFNAMSIZ+1];
889 	struct sockaddr_dl *sdl = (struct sockaddr_dl *)sa;
890 
891 	if ( (sa->sa_family != AF_LINK) || (sdl->sdl_nlen == 0) ||
892 	     (sdl->sdl_nlen > IFNAMSIZ) )
893 		return NULL;
894 
895 	/*
896 	 * ifunit wants a null-terminated name.  It may not be null-terminated
897 	 * in the sockaddr.  We don't want to change the caller's sockaddr,
898 	 * and there might not be room to put the trailing null anyway, so we
899 	 * make a local copy that we know we can null terminate safely.
900 	 */
901 
902 	bcopy(sdl->sdl_data, ifname, sdl->sdl_nlen);
903 	ifname[sdl->sdl_nlen] = '\0';
904 	return ifunit(ifname);
905 }
906 
907 
908 /*
909  * Interface ioctls.
910  */
911 int
912 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
913 {
914 	struct ifnet *ifp;
915 	struct ifreq *ifr;
916 	struct ifstat *ifs;
917 	int error;
918 	short oif_flags;
919 	int new_flags;
920 	size_t namelen, onamelen;
921 	char new_name[IFNAMSIZ];
922 	struct ifaddr *ifa;
923 	struct sockaddr_dl *sdl;
924 
925 	switch (cmd) {
926 
927 	case SIOCGIFCONF:
928 	case OSIOCGIFCONF:
929 		return (ifconf(cmd, data, cred));
930 	}
931 	ifr = (struct ifreq *)data;
932 
933 	switch (cmd) {
934 	case SIOCIFCREATE:
935 	case SIOCIFDESTROY:
936 		if ((error = suser_cred(cred, 0)) != 0)
937 			return (error);
938 		return ((cmd == SIOCIFCREATE) ?
939 			if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
940 			if_clone_destroy(ifr->ifr_name));
941 
942 	case SIOCIFGCLONERS:
943 		return (if_clone_list((struct if_clonereq *)data));
944 	}
945 
946 	ifp = ifunit(ifr->ifr_name);
947 	if (ifp == 0)
948 		return (ENXIO);
949 	switch (cmd) {
950 
951 	case SIOCGIFFLAGS:
952 		ifr->ifr_flags = ifp->if_flags;
953 		ifr->ifr_flagshigh = ifp->if_flags >> 16;
954 		break;
955 
956 	case SIOCGIFCAP:
957 		ifr->ifr_reqcap = ifp->if_capabilities;
958 		ifr->ifr_curcap = ifp->if_capenable;
959 		break;
960 
961 	case SIOCGIFMETRIC:
962 		ifr->ifr_metric = ifp->if_metric;
963 		break;
964 
965 	case SIOCGIFMTU:
966 		ifr->ifr_mtu = ifp->if_mtu;
967 		break;
968 
969 	case SIOCGIFPHYS:
970 		ifr->ifr_phys = ifp->if_physical;
971 		break;
972 
973 	case SIOCGIFPOLLCPU:
974 #ifdef DEVICE_POLLING
975 		ifr->ifr_pollcpu = ifp->if_poll_cpuid;
976 #else
977 		ifr->ifr_pollcpu = -1;
978 #endif
979 		break;
980 
981 	case SIOCSIFPOLLCPU:
982 #ifdef DEVICE_POLLING
983 		if ((ifp->if_flags & IFF_POLLING) == 0)
984 			ether_pollcpu_register(ifp, ifr->ifr_pollcpu);
985 #endif
986 		break;
987 
988 	case SIOCSIFFLAGS:
989 		error = suser_cred(cred, 0);
990 		if (error)
991 			return (error);
992 		new_flags = (ifr->ifr_flags & 0xffff) |
993 		    (ifr->ifr_flagshigh << 16);
994 		if (ifp->if_flags & IFF_SMART) {
995 			/* Smart drivers twiddle their own routes */
996 		} else if (ifp->if_flags & IFF_UP &&
997 		    (new_flags & IFF_UP) == 0) {
998 			crit_enter();
999 			if_down(ifp);
1000 			crit_exit();
1001 		} else if (new_flags & IFF_UP &&
1002 		    (ifp->if_flags & IFF_UP) == 0) {
1003 			crit_enter();
1004 			if_up(ifp);
1005 			crit_exit();
1006 		}
1007 
1008 #ifdef DEVICE_POLLING
1009 		if ((new_flags ^ ifp->if_flags) & IFF_POLLING) {
1010 			if (new_flags & IFF_POLLING) {
1011 				ether_poll_register(ifp);
1012 			} else {
1013 				ether_poll_deregister(ifp);
1014 			}
1015 		}
1016 #endif
1017 
1018 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1019 			(new_flags &~ IFF_CANTCHANGE);
1020 		if (new_flags & IFF_PPROMISC) {
1021 			/* Permanently promiscuous mode requested */
1022 			ifp->if_flags |= IFF_PROMISC;
1023 		} else if (ifp->if_pcount == 0) {
1024 			ifp->if_flags &= ~IFF_PROMISC;
1025 		}
1026 		if (ifp->if_ioctl) {
1027 			lwkt_serialize_enter(ifp->if_serializer);
1028 			ifp->if_ioctl(ifp, cmd, data, cred);
1029 			lwkt_serialize_exit(ifp->if_serializer);
1030 		}
1031 		getmicrotime(&ifp->if_lastchange);
1032 		break;
1033 
1034 	case SIOCSIFCAP:
1035 		error = suser_cred(cred, 0);
1036 		if (error)
1037 			return (error);
1038 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1039 			return (EINVAL);
1040 		lwkt_serialize_enter(ifp->if_serializer);
1041 		ifp->if_ioctl(ifp, cmd, data, cred);
1042 		lwkt_serialize_exit(ifp->if_serializer);
1043 		break;
1044 
1045 	case SIOCSIFNAME:
1046 		error = suser_cred(cred, 0);
1047 		if (error != 0)
1048 			return (error);
1049 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1050 		if (error != 0)
1051 			return (error);
1052 		if (new_name[0] == '\0')
1053 			return (EINVAL);
1054 		if (ifunit(new_name) != NULL)
1055 			return (EEXIST);
1056 
1057 		EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
1058 
1059 		/* Announce the departure of the interface. */
1060 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1061 
1062 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1063 		ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1064 		/* XXX IFA_LOCK(ifa); */
1065 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1066 		namelen = strlen(new_name);
1067 		onamelen = sdl->sdl_nlen;
1068 		/*
1069 		 * Move the address if needed.  This is safe because we
1070 		 * allocate space for a name of length IFNAMSIZ when we
1071 		 * create this in if_attach().
1072 		 */
1073 		if (namelen != onamelen) {
1074 			bcopy(sdl->sdl_data + onamelen,
1075 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1076 		}
1077 		bcopy(new_name, sdl->sdl_data, namelen);
1078 		sdl->sdl_nlen = namelen;
1079 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1080 		bzero(sdl->sdl_data, onamelen);
1081 		while (namelen != 0)
1082 			sdl->sdl_data[--namelen] = 0xff;
1083 		/* XXX IFA_UNLOCK(ifa) */
1084 
1085 		EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
1086 
1087 		/* Announce the return of the interface. */
1088 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1089 		break;
1090 
1091 	case SIOCSIFMETRIC:
1092 		error = suser_cred(cred, 0);
1093 		if (error)
1094 			return (error);
1095 		ifp->if_metric = ifr->ifr_metric;
1096 		getmicrotime(&ifp->if_lastchange);
1097 		break;
1098 
1099 	case SIOCSIFPHYS:
1100 		error = suser_cred(cred, 0);
1101 		if (error)
1102 			return error;
1103 		if (!ifp->if_ioctl)
1104 		        return EOPNOTSUPP;
1105 		lwkt_serialize_enter(ifp->if_serializer);
1106 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1107 		lwkt_serialize_exit(ifp->if_serializer);
1108 		if (error == 0)
1109 			getmicrotime(&ifp->if_lastchange);
1110 		return (error);
1111 
1112 	case SIOCSIFMTU:
1113 	{
1114 		u_long oldmtu = ifp->if_mtu;
1115 
1116 		error = suser_cred(cred, 0);
1117 		if (error)
1118 			return (error);
1119 		if (ifp->if_ioctl == NULL)
1120 			return (EOPNOTSUPP);
1121 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1122 			return (EINVAL);
1123 		lwkt_serialize_enter(ifp->if_serializer);
1124 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1125 		lwkt_serialize_exit(ifp->if_serializer);
1126 		if (error == 0) {
1127 			getmicrotime(&ifp->if_lastchange);
1128 			rt_ifmsg(ifp);
1129 		}
1130 		/*
1131 		 * If the link MTU changed, do network layer specific procedure.
1132 		 */
1133 		if (ifp->if_mtu != oldmtu) {
1134 #ifdef INET6
1135 			nd6_setmtu(ifp);
1136 #endif
1137 		}
1138 		return (error);
1139 	}
1140 
1141 	case SIOCADDMULTI:
1142 	case SIOCDELMULTI:
1143 		error = suser_cred(cred, 0);
1144 		if (error)
1145 			return (error);
1146 
1147 		/* Don't allow group membership on non-multicast interfaces. */
1148 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1149 			return EOPNOTSUPP;
1150 
1151 		/* Don't let users screw up protocols' entries. */
1152 		if (ifr->ifr_addr.sa_family != AF_LINK)
1153 			return EINVAL;
1154 
1155 		if (cmd == SIOCADDMULTI) {
1156 			struct ifmultiaddr *ifma;
1157 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1158 		} else {
1159 			error = if_delmulti(ifp, &ifr->ifr_addr);
1160 		}
1161 		if (error == 0)
1162 			getmicrotime(&ifp->if_lastchange);
1163 		return error;
1164 
1165 	case SIOCSIFPHYADDR:
1166 	case SIOCDIFPHYADDR:
1167 #ifdef INET6
1168 	case SIOCSIFPHYADDR_IN6:
1169 #endif
1170 	case SIOCSLIFPHYADDR:
1171         case SIOCSIFMEDIA:
1172 	case SIOCSIFGENERIC:
1173 		error = suser_cred(cred, 0);
1174 		if (error)
1175 			return (error);
1176 		if (ifp->if_ioctl == 0)
1177 			return (EOPNOTSUPP);
1178 		lwkt_serialize_enter(ifp->if_serializer);
1179 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1180 		lwkt_serialize_exit(ifp->if_serializer);
1181 		if (error == 0)
1182 			getmicrotime(&ifp->if_lastchange);
1183 		return error;
1184 
1185 	case SIOCGIFSTATUS:
1186 		ifs = (struct ifstat *)data;
1187 		ifs->ascii[0] = '\0';
1188 
1189 	case SIOCGIFPSRCADDR:
1190 	case SIOCGIFPDSTADDR:
1191 	case SIOCGLIFPHYADDR:
1192 	case SIOCGIFMEDIA:
1193 	case SIOCGIFGENERIC:
1194 		if (ifp->if_ioctl == NULL)
1195 			return (EOPNOTSUPP);
1196 		lwkt_serialize_enter(ifp->if_serializer);
1197 		error = ifp->if_ioctl(ifp, cmd, data, cred);
1198 		lwkt_serialize_exit(ifp->if_serializer);
1199 		return (error);
1200 
1201 	case SIOCSIFLLADDR:
1202 		error = suser_cred(cred, 0);
1203 		if (error)
1204 			return (error);
1205 		return if_setlladdr(ifp,
1206 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1207 
1208 	default:
1209 		oif_flags = ifp->if_flags;
1210 		if (so->so_proto == 0)
1211 			return (EOPNOTSUPP);
1212 #ifndef COMPAT_43
1213 		error = so_pru_control(so, cmd, data, ifp);
1214 #else
1215 	    {
1216 		int ocmd = cmd;
1217 
1218 		switch (cmd) {
1219 
1220 		case SIOCSIFDSTADDR:
1221 		case SIOCSIFADDR:
1222 		case SIOCSIFBRDADDR:
1223 		case SIOCSIFNETMASK:
1224 #if BYTE_ORDER != BIG_ENDIAN
1225 			if (ifr->ifr_addr.sa_family == 0 &&
1226 			    ifr->ifr_addr.sa_len < 16) {
1227 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1228 				ifr->ifr_addr.sa_len = 16;
1229 			}
1230 #else
1231 			if (ifr->ifr_addr.sa_len == 0)
1232 				ifr->ifr_addr.sa_len = 16;
1233 #endif
1234 			break;
1235 
1236 		case OSIOCGIFADDR:
1237 			cmd = SIOCGIFADDR;
1238 			break;
1239 
1240 		case OSIOCGIFDSTADDR:
1241 			cmd = SIOCGIFDSTADDR;
1242 			break;
1243 
1244 		case OSIOCGIFBRDADDR:
1245 			cmd = SIOCGIFBRDADDR;
1246 			break;
1247 
1248 		case OSIOCGIFNETMASK:
1249 			cmd = SIOCGIFNETMASK;
1250 		}
1251 		error =  so_pru_control(so, cmd, data, ifp);
1252 		switch (ocmd) {
1253 
1254 		case OSIOCGIFADDR:
1255 		case OSIOCGIFDSTADDR:
1256 		case OSIOCGIFBRDADDR:
1257 		case OSIOCGIFNETMASK:
1258 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1259 
1260 		}
1261 	    }
1262 #endif /* COMPAT_43 */
1263 
1264 		if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1265 #ifdef INET6
1266 			DELAY(100);/* XXX: temporary workaround for fxp issue*/
1267 			if (ifp->if_flags & IFF_UP) {
1268 				crit_enter();
1269 				in6_if_up(ifp);
1270 				crit_exit();
1271 			}
1272 #endif
1273 		}
1274 		return (error);
1275 
1276 	}
1277 	return (0);
1278 }
1279 
1280 /*
1281  * Set/clear promiscuous mode on interface ifp based on the truth value
1282  * of pswitch.  The calls are reference counted so that only the first
1283  * "on" request actually has an effect, as does the final "off" request.
1284  * Results are undefined if the "off" and "on" requests are not matched.
1285  */
1286 int
1287 ifpromisc(struct ifnet *ifp, int pswitch)
1288 {
1289 	struct ifreq ifr;
1290 	int error;
1291 	int oldflags;
1292 
1293 	oldflags = ifp->if_flags;
1294 	if (ifp->if_flags & IFF_PPROMISC) {
1295 		/* Do nothing if device is in permanently promiscuous mode */
1296 		ifp->if_pcount += pswitch ? 1 : -1;
1297 		return (0);
1298 	}
1299 	if (pswitch) {
1300 		/*
1301 		 * If the device is not configured up, we cannot put it in
1302 		 * promiscuous mode.
1303 		 */
1304 		if ((ifp->if_flags & IFF_UP) == 0)
1305 			return (ENETDOWN);
1306 		if (ifp->if_pcount++ != 0)
1307 			return (0);
1308 		ifp->if_flags |= IFF_PROMISC;
1309 		log(LOG_INFO, "%s: promiscuous mode enabled\n",
1310 		    ifp->if_xname);
1311 	} else {
1312 		if (--ifp->if_pcount > 0)
1313 			return (0);
1314 		ifp->if_flags &= ~IFF_PROMISC;
1315 		log(LOG_INFO, "%s: promiscuous mode disabled\n",
1316 		    ifp->if_xname);
1317 	}
1318 	ifr.ifr_flags = ifp->if_flags;
1319 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1320 	lwkt_serialize_enter(ifp->if_serializer);
1321 	error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1322 				 (struct ucred *)NULL);
1323 	lwkt_serialize_exit(ifp->if_serializer);
1324 	if (error == 0)
1325 		rt_ifmsg(ifp);
1326 	else
1327 		ifp->if_flags = oldflags;
1328 	return error;
1329 }
1330 
1331 /*
1332  * Return interface configuration
1333  * of system.  List may be used
1334  * in later ioctl's (above) to get
1335  * other information.
1336  */
1337 static int
1338 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
1339 {
1340 	struct ifconf *ifc = (struct ifconf *)data;
1341 	struct ifnet *ifp;
1342 	struct sockaddr *sa;
1343 	struct ifreq ifr, *ifrp;
1344 	int space = ifc->ifc_len, error = 0;
1345 
1346 	ifrp = ifc->ifc_req;
1347 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1348 		struct ifaddr_container *ifac;
1349 		int addrs;
1350 
1351 		if (space <= sizeof ifr)
1352 			break;
1353 
1354 		/*
1355 		 * Zero the stack declared structure first to prevent
1356 		 * memory disclosure.
1357 		 */
1358 		bzero(&ifr, sizeof(ifr));
1359 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1360 		    >= sizeof(ifr.ifr_name)) {
1361 			error = ENAMETOOLONG;
1362 			break;
1363 		}
1364 
1365 		addrs = 0;
1366 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1367 			struct ifaddr *ifa = ifac->ifa;
1368 
1369 			if (space <= sizeof ifr)
1370 				break;
1371 			sa = ifa->ifa_addr;
1372 			if (cred->cr_prison &&
1373 			    prison_if(cred, sa))
1374 				continue;
1375 			addrs++;
1376 #ifdef COMPAT_43
1377 			if (cmd == OSIOCGIFCONF) {
1378 				struct osockaddr *osa =
1379 					 (struct osockaddr *)&ifr.ifr_addr;
1380 				ifr.ifr_addr = *sa;
1381 				osa->sa_family = sa->sa_family;
1382 				error = copyout(&ifr, ifrp, sizeof ifr);
1383 				ifrp++;
1384 			} else
1385 #endif
1386 			if (sa->sa_len <= sizeof(*sa)) {
1387 				ifr.ifr_addr = *sa;
1388 				error = copyout(&ifr, ifrp, sizeof ifr);
1389 				ifrp++;
1390 			} else {
1391 				if (space < (sizeof ifr) + sa->sa_len -
1392 					    sizeof(*sa))
1393 					break;
1394 				space -= sa->sa_len - sizeof(*sa);
1395 				error = copyout(&ifr, ifrp,
1396 						sizeof ifr.ifr_name);
1397 				if (error == 0)
1398 					error = copyout(sa, &ifrp->ifr_addr,
1399 							sa->sa_len);
1400 				ifrp = (struct ifreq *)
1401 					(sa->sa_len + (caddr_t)&ifrp->ifr_addr);
1402 			}
1403 			if (error)
1404 				break;
1405 			space -= sizeof ifr;
1406 		}
1407 		if (error)
1408 			break;
1409 		if (!addrs) {
1410 			bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
1411 			error = copyout(&ifr, ifrp, sizeof ifr);
1412 			if (error)
1413 				break;
1414 			space -= sizeof ifr;
1415 			ifrp++;
1416 		}
1417 	}
1418 	ifc->ifc_len -= space;
1419 	return (error);
1420 }
1421 
1422 /*
1423  * Just like if_promisc(), but for all-multicast-reception mode.
1424  */
1425 int
1426 if_allmulti(struct ifnet *ifp, int onswitch)
1427 {
1428 	int error = 0;
1429 	struct ifreq ifr;
1430 
1431 	crit_enter();
1432 
1433 	if (onswitch) {
1434 		if (ifp->if_amcount++ == 0) {
1435 			ifp->if_flags |= IFF_ALLMULTI;
1436 			ifr.ifr_flags = ifp->if_flags;
1437 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
1438 			lwkt_serialize_enter(ifp->if_serializer);
1439 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1440 					      (struct ucred *)NULL);
1441 			lwkt_serialize_exit(ifp->if_serializer);
1442 		}
1443 	} else {
1444 		if (ifp->if_amcount > 1) {
1445 			ifp->if_amcount--;
1446 		} else {
1447 			ifp->if_amcount = 0;
1448 			ifp->if_flags &= ~IFF_ALLMULTI;
1449 			ifr.ifr_flags = ifp->if_flags;
1450 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
1451 			lwkt_serialize_enter(ifp->if_serializer);
1452 			error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1453 					      (struct ucred *)NULL);
1454 			lwkt_serialize_exit(ifp->if_serializer);
1455 		}
1456 	}
1457 
1458 	crit_exit();
1459 
1460 	if (error == 0)
1461 		rt_ifmsg(ifp);
1462 	return error;
1463 }
1464 
1465 /*
1466  * Add a multicast listenership to the interface in question.
1467  * The link layer provides a routine which converts
1468  */
1469 int
1470 if_addmulti(
1471 	struct ifnet *ifp,	/* interface to manipulate */
1472 	struct sockaddr *sa,	/* address to add */
1473 	struct ifmultiaddr **retifma)
1474 {
1475 	struct sockaddr *llsa, *dupsa;
1476 	int error;
1477 	struct ifmultiaddr *ifma;
1478 
1479 	/*
1480 	 * If the matching multicast address already exists
1481 	 * then don't add a new one, just add a reference
1482 	 */
1483 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1484 		if (sa_equal(sa, ifma->ifma_addr)) {
1485 			ifma->ifma_refcount++;
1486 			if (retifma)
1487 				*retifma = ifma;
1488 			return 0;
1489 		}
1490 	}
1491 
1492 	/*
1493 	 * Give the link layer a chance to accept/reject it, and also
1494 	 * find out which AF_LINK address this maps to, if it isn't one
1495 	 * already.
1496 	 */
1497 	if (ifp->if_resolvemulti) {
1498 		lwkt_serialize_enter(ifp->if_serializer);
1499 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
1500 		lwkt_serialize_exit(ifp->if_serializer);
1501 		if (error)
1502 			return error;
1503 	} else {
1504 		llsa = 0;
1505 	}
1506 
1507 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, M_WAITOK);
1508 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, M_WAITOK);
1509 	bcopy(sa, dupsa, sa->sa_len);
1510 
1511 	ifma->ifma_addr = dupsa;
1512 	ifma->ifma_lladdr = llsa;
1513 	ifma->ifma_ifp = ifp;
1514 	ifma->ifma_refcount = 1;
1515 	ifma->ifma_protospec = 0;
1516 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1517 
1518 	/*
1519 	 * Some network interfaces can scan the address list at
1520 	 * interrupt time; lock them out.
1521 	 */
1522 	crit_enter();
1523 	LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1524 	crit_exit();
1525 	*retifma = ifma;
1526 
1527 	if (llsa != 0) {
1528 		LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1529 			if (sa_equal(ifma->ifma_addr, llsa))
1530 				break;
1531 		}
1532 		if (ifma) {
1533 			ifma->ifma_refcount++;
1534 		} else {
1535 			MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma,
1536 			       M_IFMADDR, M_WAITOK);
1537 			MALLOC(dupsa, struct sockaddr *, llsa->sa_len,
1538 			       M_IFMADDR, M_WAITOK);
1539 			bcopy(llsa, dupsa, llsa->sa_len);
1540 			ifma->ifma_addr = dupsa;
1541 			ifma->ifma_ifp = ifp;
1542 			ifma->ifma_refcount = 1;
1543 			crit_enter();
1544 			LIST_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1545 			crit_exit();
1546 		}
1547 	}
1548 	/*
1549 	 * We are certain we have added something, so call down to the
1550 	 * interface to let them know about it.
1551 	 */
1552 	crit_enter();
1553 	lwkt_serialize_enter(ifp->if_serializer);
1554 	ifp->if_ioctl(ifp, SIOCADDMULTI, 0, (struct ucred *)NULL);
1555 	lwkt_serialize_exit(ifp->if_serializer);
1556 	crit_exit();
1557 
1558 	return 0;
1559 }
1560 
1561 /*
1562  * Remove a reference to a multicast address on this interface.  Yell
1563  * if the request does not match an existing membership.
1564  */
1565 int
1566 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
1567 {
1568 	struct ifmultiaddr *ifma;
1569 
1570 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1571 		if (sa_equal(sa, ifma->ifma_addr))
1572 			break;
1573 	if (ifma == 0)
1574 		return ENOENT;
1575 
1576 	if (ifma->ifma_refcount > 1) {
1577 		ifma->ifma_refcount--;
1578 		return 0;
1579 	}
1580 
1581 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
1582 	sa = ifma->ifma_lladdr;
1583 	crit_enter();
1584 	LIST_REMOVE(ifma, ifma_link);
1585 	/*
1586 	 * Make sure the interface driver is notified
1587 	 * in the case of a link layer mcast group being left.
1588 	 */
1589 	if (ifma->ifma_addr->sa_family == AF_LINK && sa == 0) {
1590 		lwkt_serialize_enter(ifp->if_serializer);
1591 		ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1592 		lwkt_serialize_exit(ifp->if_serializer);
1593 	}
1594 	crit_exit();
1595 	kfree(ifma->ifma_addr, M_IFMADDR);
1596 	kfree(ifma, M_IFMADDR);
1597 	if (sa == 0)
1598 		return 0;
1599 
1600 	/*
1601 	 * Now look for the link-layer address which corresponds to
1602 	 * this network address.  It had been squirreled away in
1603 	 * ifma->ifma_lladdr for this purpose (so we don't have
1604 	 * to call ifp->if_resolvemulti() again), and we saved that
1605 	 * value in sa above.  If some nasty deleted the
1606 	 * link-layer address out from underneath us, we can deal because
1607 	 * the address we stored was is not the same as the one which was
1608 	 * in the record for the link-layer address.  (So we don't complain
1609 	 * in that case.)
1610 	 */
1611 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1612 		if (sa_equal(sa, ifma->ifma_addr))
1613 			break;
1614 	if (ifma == 0)
1615 		return 0;
1616 
1617 	if (ifma->ifma_refcount > 1) {
1618 		ifma->ifma_refcount--;
1619 		return 0;
1620 	}
1621 
1622 	crit_enter();
1623 	lwkt_serialize_enter(ifp->if_serializer);
1624 	LIST_REMOVE(ifma, ifma_link);
1625 	ifp->if_ioctl(ifp, SIOCDELMULTI, 0, (struct ucred *)NULL);
1626 	lwkt_serialize_exit(ifp->if_serializer);
1627 	crit_exit();
1628 	kfree(ifma->ifma_addr, M_IFMADDR);
1629 	kfree(sa, M_IFMADDR);
1630 	kfree(ifma, M_IFMADDR);
1631 
1632 	return 0;
1633 }
1634 
1635 /*
1636  * Set the link layer address on an interface.
1637  *
1638  * At this time we only support certain types of interfaces,
1639  * and we don't allow the length of the address to change.
1640  */
1641 int
1642 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
1643 {
1644 	struct sockaddr_dl *sdl;
1645 	struct ifreq ifr;
1646 
1647 	sdl = IF_LLSOCKADDR(ifp);
1648 	if (sdl == NULL)
1649 		return (EINVAL);
1650 	if (len != sdl->sdl_alen)	/* don't allow length to change */
1651 		return (EINVAL);
1652 	switch (ifp->if_type) {
1653 	case IFT_ETHER:			/* these types use struct arpcom */
1654 	case IFT_XETHER:
1655 	case IFT_L2VLAN:
1656 		bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
1657 		bcopy(lladdr, LLADDR(sdl), len);
1658 		break;
1659 	default:
1660 		return (ENODEV);
1661 	}
1662 	/*
1663 	 * If the interface is already up, we need
1664 	 * to re-init it in order to reprogram its
1665 	 * address filter.
1666 	 */
1667 	lwkt_serialize_enter(ifp->if_serializer);
1668 	if ((ifp->if_flags & IFF_UP) != 0) {
1669 		struct ifaddr_container *ifac;
1670 
1671 		ifp->if_flags &= ~IFF_UP;
1672 		ifr.ifr_flags = ifp->if_flags;
1673 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
1674 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1675 			      (struct ucred *)NULL);
1676 		ifp->if_flags |= IFF_UP;
1677 		ifr.ifr_flags = ifp->if_flags;
1678 		ifr.ifr_flagshigh = ifp->if_flags >> 16;
1679 		ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
1680 				 (struct ucred *)NULL);
1681 #ifdef INET
1682 		/*
1683 		 * Also send gratuitous ARPs to notify other nodes about
1684 		 * the address change.
1685 		 */
1686 		TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1687 			struct ifaddr *ifa = ifac->ifa;
1688 
1689 			if (ifa->ifa_addr != NULL &&
1690 			    ifa->ifa_addr->sa_family == AF_INET)
1691 				arp_ifinit(ifp, ifa);
1692 		}
1693 #endif
1694 	}
1695 	lwkt_serialize_exit(ifp->if_serializer);
1696 	return (0);
1697 }
1698 
1699 struct ifmultiaddr *
1700 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
1701 {
1702 	struct ifmultiaddr *ifma;
1703 
1704 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
1705 		if (sa_equal(ifma->ifma_addr, sa))
1706 			break;
1707 
1708 	return ifma;
1709 }
1710 
1711 /*
1712  * This function locates the first real ethernet MAC from a network
1713  * card and loads it into node, returning 0 on success or ENOENT if
1714  * no suitable interfaces were found.  It is used by the uuid code to
1715  * generate a unique 6-byte number.
1716  */
1717 int
1718 if_getanyethermac(uint16_t *node, int minlen)
1719 {
1720 	struct ifnet *ifp;
1721 	struct sockaddr_dl *sdl;
1722 
1723 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1724 		if (ifp->if_type != IFT_ETHER)
1725 			continue;
1726 		sdl = IF_LLSOCKADDR(ifp);
1727 		if (sdl->sdl_alen < minlen)
1728 			continue;
1729 		bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
1730 		      minlen);
1731 		return(0);
1732 	}
1733 	return (ENOENT);
1734 }
1735 
1736 /*
1737  * The name argument must be a pointer to storage which will last as
1738  * long as the interface does.  For physical devices, the result of
1739  * device_get_name(dev) is a good choice and for pseudo-devices a
1740  * static string works well.
1741  */
1742 void
1743 if_initname(struct ifnet *ifp, const char *name, int unit)
1744 {
1745 	ifp->if_dname = name;
1746 	ifp->if_dunit = unit;
1747 	if (unit != IF_DUNIT_NONE)
1748 		ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
1749 	else
1750 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
1751 }
1752 
1753 int
1754 if_printf(struct ifnet *ifp, const char *fmt, ...)
1755 {
1756 	__va_list ap;
1757 	int retval;
1758 
1759 	retval = kprintf("%s: ", ifp->if_xname);
1760 	__va_start(ap, fmt);
1761 	retval += kvprintf(fmt, ap);
1762 	__va_end(ap);
1763 	return (retval);
1764 }
1765 
1766 void
1767 ifq_set_classic(struct ifaltq *ifq)
1768 {
1769 	ifq->altq_enqueue = ifq_classic_enqueue;
1770 	ifq->altq_dequeue = ifq_classic_dequeue;
1771 	ifq->altq_request = ifq_classic_request;
1772 }
1773 
1774 static int
1775 ifq_classic_enqueue(struct ifaltq *ifq, struct mbuf *m,
1776 		    struct altq_pktattr *pa __unused)
1777 {
1778 	crit_enter();
1779 	if (IF_QFULL(ifq)) {
1780 		m_freem(m);
1781 		crit_exit();
1782 		return(ENOBUFS);
1783 	} else {
1784 		IF_ENQUEUE(ifq, m);
1785 		crit_exit();
1786 		return(0);
1787 	}
1788 }
1789 
1790 static struct mbuf *
1791 ifq_classic_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
1792 {
1793 	struct mbuf *m;
1794 
1795 	crit_enter();
1796 	switch (op) {
1797 	case ALTDQ_POLL:
1798 		IF_POLL(ifq, m);
1799 		break;
1800 	case ALTDQ_REMOVE:
1801 		IF_DEQUEUE(ifq, m);
1802 		break;
1803 	default:
1804 		panic("unsupported ALTQ dequeue op: %d", op);
1805 	}
1806 	crit_exit();
1807 	KKASSERT(mpolled == NULL || mpolled == m);
1808 	return(m);
1809 }
1810 
1811 static int
1812 ifq_classic_request(struct ifaltq *ifq, int req, void *arg)
1813 {
1814 	crit_enter();
1815 	switch (req) {
1816 	case ALTRQ_PURGE:
1817 		IF_DRAIN(ifq);
1818 		break;
1819 	default:
1820 		panic("unsupported ALTQ request: %d", req);
1821 	}
1822 	crit_exit();
1823 	return(0);
1824 }
1825 
1826 void *
1827 ifa_create(int size, int flags)
1828 {
1829 	struct ifaddr *ifa;
1830 	int i;
1831 
1832 	KASSERT(size >= sizeof(*ifa), ("ifaddr size too small\n"));
1833 
1834 	ifa = kmalloc(size, M_IFADDR, flags | M_ZERO);
1835 	if (ifa == NULL)
1836 		return NULL;
1837 
1838 	ifa->ifa_containers = kmalloc(ncpus * sizeof(struct ifaddr_container),
1839 				      M_IFADDR, M_WAITOK | M_ZERO);
1840 	ifa->ifa_cpumask = smp_active_mask;
1841 	for (i = 0; i < ncpus; ++i) {
1842 		struct ifaddr_container *ifac = &ifa->ifa_containers[i];
1843 
1844 		ifac->ifa_magic = IFA_CONTAINER_MAGIC;
1845 		ifac->ifa = ifa;
1846 		ifac->ifa_refcnt = 1;
1847 	}
1848 #ifdef IFADDR_DEBUG
1849 	kprintf("alloc ifa %p %d\n", ifa, size);
1850 #endif
1851 	return ifa;
1852 }
1853 
1854 struct ifac_free_arg {
1855 	struct ifaddr	*ifa;
1856 	int		cpuid;
1857 };
1858 
1859 static void
1860 ifac_free_dispatch(struct netmsg *nmsg)
1861 {
1862 	struct lwkt_msg *msg = &nmsg->nm_lmsg;
1863 	struct ifac_free_arg *arg = msg->u.ms_resultp;
1864 	struct ifaddr *ifa = arg->ifa;
1865 
1866 	ifa->ifa_cpumask &= ~(1 << arg->cpuid);
1867 	if (ifa->ifa_cpumask == 0) {
1868 #ifdef IFADDR_DEBUG
1869 		kprintf("free ifa %p\n", ifa);
1870 #endif
1871 		kfree(ifa->ifa_containers, M_IFADDR);
1872 		kfree(ifa, M_IFADDR);
1873 	}
1874 	lwkt_replymsg(msg, 0);
1875 }
1876 
1877 void
1878 ifac_free(struct ifaddr_container *ifac, int cpu_id)
1879 {
1880 	struct ifac_free_arg arg;
1881 	struct netmsg nmsg;
1882 	struct lwkt_msg *msg;
1883 
1884 	KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
1885 	KKASSERT(ifac->ifa_refcnt == 0);
1886 
1887 	ifac->ifa_magic = IFA_CONTAINER_DEAD;
1888 
1889 	bzero(&arg, sizeof(arg));
1890 	arg.ifa = ifac->ifa;
1891 	arg.cpuid = cpu_id;
1892 #ifdef IFADDR_DEBUG_VERBOSE
1893 	kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, arg.cpuid);
1894 #endif
1895 
1896 	netmsg_init(&nmsg, &curthread->td_msgport, 0, ifac_free_dispatch);
1897 	msg = &nmsg.nm_lmsg;
1898 	msg->u.ms_resultp = &arg;
1899 
1900 	lwkt_domsg(ifa_portfn(0), msg, 0);
1901 }
1902 
1903 static __inline void
1904 ifa_forwardmsg(struct lwkt_msg *lmsg, int next_cpu)
1905 {
1906 	if (next_cpu < ncpus)
1907 		lwkt_forwardmsg(ifa_portfn(next_cpu), lmsg);
1908 	else
1909 		lwkt_replymsg(lmsg, 0);
1910 }
1911 
1912 static void
1913 ifa_iflink_dispatch(struct netmsg *nmsg)
1914 {
1915 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
1916 	struct ifaddr *ifa = msg->ifa;
1917 	struct ifnet *ifp = msg->ifp;
1918 	int cpu = mycpuid;
1919 
1920 	crit_enter();
1921 	if (msg->tail) {
1922 		TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu],
1923 				  &ifa->ifa_containers[cpu], ifa_link);
1924 	} else {
1925 		TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu],
1926 				  &ifa->ifa_containers[cpu], ifa_link);
1927 	}
1928 	crit_exit();
1929 
1930 	ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
1931 }
1932 
1933 void
1934 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
1935 {
1936 	struct netmsg_ifaddr msg;
1937 
1938 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
1939 		    ifa_iflink_dispatch);
1940 	msg.ifa = ifa;
1941 	msg.ifp = ifp;
1942 	msg.tail = tail;
1943 
1944 	lwkt_domsg(ifa_portfn(0), &msg.netmsg.nm_lmsg, 0);
1945 }
1946 
1947 static void
1948 ifa_ifunlink_dispatch(struct netmsg *nmsg)
1949 {
1950 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
1951 	struct ifaddr *ifa = msg->ifa;
1952 	struct ifnet *ifp = msg->ifp;
1953 	int cpu = mycpuid;
1954 
1955 	crit_enter();
1956 	TAILQ_REMOVE(&ifp->if_addrheads[cpu],
1957 		     &ifa->ifa_containers[cpu], ifa_link);
1958 	crit_exit();
1959 
1960 	ifa_forwardmsg(&nmsg->nm_lmsg, cpu + 1);
1961 }
1962 
1963 void
1964 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
1965 {
1966 	struct netmsg_ifaddr msg;
1967 
1968 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
1969 		    ifa_ifunlink_dispatch);
1970 	msg.ifa = ifa;
1971 	msg.ifp = ifp;
1972 
1973 	lwkt_domsg(ifa_portfn(0), &msg.netmsg.nm_lmsg, 0);
1974 }
1975 
1976 static void
1977 ifa_destroy_dispatch(struct netmsg *nmsg)
1978 {
1979 	struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
1980 
1981 	IFAFREE(msg->ifa);
1982 	ifa_forwardmsg(&nmsg->nm_lmsg, mycpuid + 1);
1983 }
1984 
1985 void
1986 ifa_destroy(struct ifaddr *ifa)
1987 {
1988 	struct netmsg_ifaddr msg;
1989 
1990 	netmsg_init(&msg.netmsg, &curthread->td_msgport, 0,
1991 		    ifa_destroy_dispatch);
1992 	msg.ifa = ifa;
1993 
1994 	lwkt_domsg(ifa_portfn(0), &msg.netmsg.nm_lmsg, 0);
1995 }
1996 
1997 struct lwkt_port *
1998 ifa_portfn(int cpu)
1999 {
2000 	return &ifaddr_threads[cpu].td_msgport;
2001 }
2002 
2003 static void
2004 ifaddrinit(void *dummy __unused)
2005 {
2006 	int i;
2007 
2008 	for (i = 0; i < ncpus; ++i) {
2009 		struct thread *thr = &ifaddr_threads[i];
2010 
2011 		lwkt_create(netmsg_service_loop_mpsafe, NULL, NULL, thr, 0, i,
2012 			    "ifaddr %d", i);
2013 		netmsg_service_port_init(&thr->td_msgport);
2014 	}
2015 }
2016