xref: /freebsd/sys/net/if.c (revision f0a75d274af375d15b97b830966b99a02b7db911)
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  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.5 (Berkeley) 1/9/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 #include "opt_mac.h"
37 #include "opt_carp.h"
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/conf.h>
42 #include <sys/malloc.h>
43 #include <sys/sbuf.h>
44 #include <sys/bus.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/protosw.h>
52 #include <sys/kernel.h>
53 #include <sys/sockio.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
56 #include <sys/taskqueue.h>
57 #include <sys/domain.h>
58 #include <sys/jail.h>
59 #include <machine/stdarg.h>
60 
61 #include <net/if.h>
62 #include <net/if_clone.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/if_var.h>
66 #include <net/radix.h>
67 #include <net/route.h>
68 
69 #if defined(INET) || defined(INET6)
70 /*XXX*/
71 #include <netinet/in.h>
72 #include <netinet/in_var.h>
73 #ifdef INET6
74 #include <netinet6/in6_var.h>
75 #include <netinet6/in6_ifattach.h>
76 #endif
77 #endif
78 #ifdef INET
79 #include <netinet/if_ether.h>
80 #endif
81 #ifdef DEV_CARP
82 #include <netinet/ip_carp.h>
83 #endif
84 
85 #include <security/mac/mac_framework.h>
86 
87 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
88 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
89 
90 /* Log link state change events */
91 static int log_link_state_change = 1;
92 
93 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
94 	&log_link_state_change, 0,
95 	"log interface link state change events");
96 
97 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
98 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
99 
100 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
101 
102 /*
103  * XXX: Style; these should be sorted alphabetically, and unprototyped
104  * static functions should be prototyped. Currently they are sorted by
105  * declaration order.
106  */
107 static void	if_attachdomain(void *);
108 static void	if_attachdomain1(struct ifnet *);
109 static void	if_purgemaddrs(struct ifnet *);
110 static int	ifconf(u_long, caddr_t);
111 static void	if_freemulti(struct ifmultiaddr *);
112 static void	if_grow(void);
113 static void	if_init(void *);
114 static void	if_check(void *);
115 static void	if_qflush(struct ifaltq *);
116 static void	if_route(struct ifnet *, int flag, int fam);
117 static int	if_setflag(struct ifnet *, int, int, int *, int);
118 static void	if_slowtimo(void *);
119 static void	if_unroute(struct ifnet *, int flag, int fam);
120 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
121 static int	if_rtdel(struct radix_node *, void *);
122 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
123 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
124 static void	if_start_deferred(void *context, int pending);
125 static void	do_link_state_change(void *, int);
126 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
127 static int	if_getgroupmembers(struct ifgroupreq *);
128 #ifdef INET6
129 /*
130  * XXX: declare here to avoid to include many inet6 related files..
131  * should be more generalized?
132  */
133 extern void	nd6_setmtu(struct ifnet *);
134 #endif
135 
136 int	if_index = 0;
137 struct	ifindex_entry *ifindex_table = NULL;
138 int	ifqmaxlen = IFQ_MAXLEN;
139 struct	ifnethead ifnet;	/* depend on static init XXX */
140 struct	ifgrouphead ifg_head;
141 struct	mtx ifnet_lock;
142 static	if_com_alloc_t *if_com_alloc[256];
143 static	if_com_free_t *if_com_free[256];
144 
145 static int	if_indexlim = 8;
146 static struct	knlist ifklist;
147 
148 static void	filt_netdetach(struct knote *kn);
149 static int	filt_netdev(struct knote *kn, long hint);
150 
151 static struct filterops netdev_filtops =
152     { 1, NULL, filt_netdetach, filt_netdev };
153 
154 /*
155  * System initialization
156  */
157 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL)
158 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_check, NULL)
159 
160 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
161 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
162 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
163 
164 static d_open_t		netopen;
165 static d_close_t	netclose;
166 static d_ioctl_t	netioctl;
167 static d_kqfilter_t	netkqfilter;
168 
169 static struct cdevsw net_cdevsw = {
170 	.d_version =	D_VERSION,
171 	.d_flags =	D_NEEDGIANT,
172 	.d_open =	netopen,
173 	.d_close =	netclose,
174 	.d_ioctl =	netioctl,
175 	.d_name =	"net",
176 	.d_kqfilter =	netkqfilter,
177 };
178 
179 static int
180 netopen(struct cdev *dev, int flag, int mode, struct thread *td)
181 {
182 	return (0);
183 }
184 
185 static int
186 netclose(struct cdev *dev, int flags, int fmt, struct thread *td)
187 {
188 	return (0);
189 }
190 
191 static int
192 netioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
193 {
194 	struct ifnet *ifp;
195 	int error, idx;
196 
197 	/* only support interface specific ioctls */
198 	if (IOCGROUP(cmd) != 'i')
199 		return (EOPNOTSUPP);
200 	idx = minor(dev);
201 	if (idx == 0) {
202 		/*
203 		 * special network device, not interface.
204 		 */
205 		if (cmd == SIOCGIFCONF)
206 			return (ifconf(cmd, data));	/* XXX remove cmd */
207 #ifdef __amd64__
208 		if (cmd == SIOCGIFCONF32)
209 			return (ifconf(cmd, data));	/* XXX remove cmd */
210 #endif
211 		return (EOPNOTSUPP);
212 	}
213 
214 	ifp = ifnet_byindex(idx);
215 	if (ifp == NULL)
216 		return (ENXIO);
217 
218 	error = ifhwioctl(cmd, ifp, data, td);
219 	if (error == ENOIOCTL)
220 		error = EOPNOTSUPP;
221 	return (error);
222 }
223 
224 static int
225 netkqfilter(struct cdev *dev, struct knote *kn)
226 {
227 	struct knlist *klist;
228 	struct ifnet *ifp;
229 	int idx;
230 
231 	switch (kn->kn_filter) {
232 	case EVFILT_NETDEV:
233 		kn->kn_fop = &netdev_filtops;
234 		break;
235 	default:
236 		return (EINVAL);
237 	}
238 
239 	idx = minor(dev);
240 	if (idx == 0) {
241 		klist = &ifklist;
242 	} else {
243 		ifp = ifnet_byindex(idx);
244 		if (ifp == NULL)
245 			return (1);
246 		klist = &ifp->if_klist;
247 	}
248 
249 	kn->kn_hook = (caddr_t)klist;
250 
251 	knlist_add(klist, kn, 0);
252 
253 	return (0);
254 }
255 
256 static void
257 filt_netdetach(struct knote *kn)
258 {
259 	struct knlist *klist = (struct knlist *)kn->kn_hook;
260 
261 	knlist_remove(klist, kn, 0);
262 }
263 
264 static int
265 filt_netdev(struct knote *kn, long hint)
266 {
267 	struct knlist *klist = (struct knlist *)kn->kn_hook;
268 
269 	/*
270 	 * Currently NOTE_EXIT is abused to indicate device detach.
271 	 */
272 	if (hint == NOTE_EXIT) {
273 		kn->kn_data = NOTE_LINKINV;
274 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
275 		knlist_remove_inevent(klist, kn);
276 		return (1);
277 	}
278 	if (hint != 0)
279 		kn->kn_data = hint;			/* current status */
280 	if (kn->kn_sfflags & hint)
281 		kn->kn_fflags |= hint;
282 	return (kn->kn_fflags != 0);
283 }
284 
285 /*
286  * Network interface utility routines.
287  *
288  * Routines with ifa_ifwith* names take sockaddr *'s as
289  * parameters.
290  */
291 /* ARGSUSED*/
292 static void
293 if_init(void *dummy __unused)
294 {
295 
296 	IFNET_LOCK_INIT();
297 	TAILQ_INIT(&ifnet);
298 	TAILQ_INIT(&ifg_head);
299 	knlist_init(&ifklist, NULL, NULL, NULL, NULL);
300 	if_grow();				/* create initial table */
301 	ifdev_byindex(0) = make_dev(&net_cdevsw, 0,
302 	    UID_ROOT, GID_WHEEL, 0600, "network");
303 	if_clone_init();
304 }
305 
306 static void
307 if_grow(void)
308 {
309 	u_int n;
310 	struct ifindex_entry *e;
311 
312 	if_indexlim <<= 1;
313 	n = if_indexlim * sizeof(*e);
314 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
315 	if (ifindex_table != NULL) {
316 		memcpy((caddr_t)e, (caddr_t)ifindex_table, n/2);
317 		free((caddr_t)ifindex_table, M_IFNET);
318 	}
319 	ifindex_table = e;
320 }
321 
322 /* ARGSUSED*/
323 static void
324 if_check(void *dummy __unused)
325 {
326 	struct ifnet *ifp;
327 	int s;
328 
329 	s = splimp();
330 	IFNET_RLOCK();	/* could sleep on rare error; mostly okay XXX */
331 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
332 		if (ifp->if_snd.ifq_maxlen == 0) {
333 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
334 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
335 		}
336 		if (!mtx_initialized(&ifp->if_snd.ifq_mtx)) {
337 			if_printf(ifp,
338 			    "XXX: driver didn't initialize queue mtx\n");
339 			mtx_init(&ifp->if_snd.ifq_mtx, "unknown",
340 			    MTX_NETWORK_LOCK, MTX_DEF);
341 		}
342 	}
343 	IFNET_RUNLOCK();
344 	splx(s);
345 	if_slowtimo(0);
346 }
347 
348 /*
349  * Allocate a struct ifnet and in index for an interface.
350  */
351 struct ifnet*
352 if_alloc(u_char type)
353 {
354 	struct ifnet *ifp;
355 
356 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
357 
358 	/*
359 	 * Try to find an empty slot below if_index.  If we fail, take
360 	 * the next slot.
361 	 *
362 	 * XXX: should be locked!
363 	 */
364 	for (ifp->if_index = 1; ifp->if_index <= if_index; ifp->if_index++) {
365 		if (ifnet_byindex(ifp->if_index) == NULL)
366 			break;
367 	}
368 	/* Catch if_index overflow. */
369 	if (ifp->if_index < 1) {
370 		free(ifp, M_IFNET);
371 		return (NULL);
372 	}
373 	if (ifp->if_index > if_index)
374 		if_index = ifp->if_index;
375 	if (if_index >= if_indexlim)
376 		if_grow();
377 	ifnet_byindex(ifp->if_index) = ifp;
378 
379 	ifp->if_type = type;
380 
381 	if (if_com_alloc[type] != NULL) {
382 		ifp->if_l2com = if_com_alloc[type](type, ifp);
383 		if (ifp->if_l2com == NULL) {
384 			free(ifp, M_IFNET);
385 			return (NULL);
386 		}
387 	}
388 	IF_ADDR_LOCK_INIT(ifp);
389 
390 	return (ifp);
391 }
392 
393 void
394 if_free(struct ifnet *ifp)
395 {
396 
397 	/* Do not add code to this function!  Add it to if_free_type(). */
398 	if_free_type(ifp, ifp->if_type);
399 }
400 
401 void
402 if_free_type(struct ifnet *ifp, u_char type)
403 {
404 
405 	if (ifp != ifnet_byindex(ifp->if_index)) {
406 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
407 		    __func__);
408 		return;
409 	}
410 
411 	IF_ADDR_LOCK_DESTROY(ifp);
412 
413 	ifnet_byindex(ifp->if_index) = NULL;
414 
415 	/* XXX: should be locked with if_findindex() */
416 	while (if_index > 0 && ifnet_byindex(if_index) == NULL)
417 		if_index--;
418 
419 	if (if_com_free[type] != NULL)
420 		if_com_free[type](ifp->if_l2com, type);
421 
422 	free(ifp, M_IFNET);
423 };
424 
425 /*
426  * Attach an interface to the
427  * list of "active" interfaces.
428  */
429 void
430 if_attach(struct ifnet *ifp)
431 {
432 	unsigned socksize, ifasize;
433 	int namelen, masklen;
434 	struct sockaddr_dl *sdl;
435 	struct ifaddr *ifa;
436 
437 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
438 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
439 		    ifp->if_xname);
440 
441 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
442 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
443 	IF_AFDATA_LOCK_INIT(ifp);
444 	ifp->if_afdata_initialized = 0;
445 	/*
446 	 * XXX -
447 	 * The old code would work if the interface passed a pre-existing
448 	 * chain of ifaddrs to this code.  We don't trust our callers to
449 	 * properly initialize the tailq, however, so we no longer allow
450 	 * this unlikely case.
451 	 */
452 	TAILQ_INIT(&ifp->if_addrhead);
453 	TAILQ_INIT(&ifp->if_prefixhead);
454 	TAILQ_INIT(&ifp->if_multiaddrs);
455 	TAILQ_INIT(&ifp->if_groups);
456 
457 	if_addgroup(ifp, IFG_ALL);
458 
459 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
460 	getmicrotime(&ifp->if_lastchange);
461 	ifp->if_data.ifi_epoch = time_uptime;
462 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
463 
464 #ifdef MAC
465 	mac_init_ifnet(ifp);
466 	mac_create_ifnet(ifp);
467 #endif
468 
469 	ifdev_byindex(ifp->if_index) = make_dev(&net_cdevsw,
470 	    unit2minor(ifp->if_index),
471 	    UID_ROOT, GID_WHEEL, 0600, "%s/%s",
472 	    net_cdevsw.d_name, ifp->if_xname);
473 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
474 	    net_cdevsw.d_name, ifp->if_index);
475 
476 	mtx_init(&ifp->if_snd.ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
477 
478 	/*
479 	 * create a Link Level name for this device
480 	 */
481 	namelen = strlen(ifp->if_xname);
482 	/*
483 	 * Always save enough space for any possiable name so we can do
484 	 * a rename in place later.
485 	 */
486 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
487 	socksize = masklen + ifp->if_addrlen;
488 	if (socksize < sizeof(*sdl))
489 		socksize = sizeof(*sdl);
490 	socksize = roundup2(socksize, sizeof(long));
491 	ifasize = sizeof(*ifa) + 2 * socksize;
492 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
493 	IFA_LOCK_INIT(ifa);
494 	sdl = (struct sockaddr_dl *)(ifa + 1);
495 	sdl->sdl_len = socksize;
496 	sdl->sdl_family = AF_LINK;
497 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
498 	sdl->sdl_nlen = namelen;
499 	sdl->sdl_index = ifp->if_index;
500 	sdl->sdl_type = ifp->if_type;
501 	ifp->if_addr = ifa;
502 	ifa->ifa_ifp = ifp;
503 	ifa->ifa_rtrequest = link_rtrequest;
504 	ifa->ifa_addr = (struct sockaddr *)sdl;
505 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
506 	ifa->ifa_netmask = (struct sockaddr *)sdl;
507 	sdl->sdl_len = masklen;
508 	while (namelen != 0)
509 		sdl->sdl_data[--namelen] = 0xff;
510 	ifa->ifa_refcnt = 1;
511 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
512 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
513 	ifp->if_snd.altq_type = 0;
514 	ifp->if_snd.altq_disc = NULL;
515 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
516 	ifp->if_snd.altq_tbr  = NULL;
517 	ifp->if_snd.altq_ifp  = ifp;
518 
519 	IFNET_WLOCK();
520 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
521 	IFNET_WUNLOCK();
522 
523 	if (domain_init_status >= 2)
524 		if_attachdomain1(ifp);
525 
526 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
527 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
528 
529 	/* Announce the interface. */
530 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
531 
532 	if (ifp->if_watchdog != NULL)
533 		if_printf(ifp, "using obsoleted if_watchdog interface\n");
534 }
535 
536 static void
537 if_attachdomain(void *dummy)
538 {
539 	struct ifnet *ifp;
540 	int s;
541 
542 	s = splnet();
543 	TAILQ_FOREACH(ifp, &ifnet, if_link)
544 		if_attachdomain1(ifp);
545 	splx(s);
546 }
547 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
548     if_attachdomain, NULL);
549 
550 static void
551 if_attachdomain1(struct ifnet *ifp)
552 {
553 	struct domain *dp;
554 	int s;
555 
556 	s = splnet();
557 
558 	/*
559 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
560 	 * cannot lock ifp->if_afdata initialization, entirely.
561 	 */
562 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
563 		splx(s);
564 		return;
565 	}
566 	if (ifp->if_afdata_initialized >= domain_init_status) {
567 		IF_AFDATA_UNLOCK(ifp);
568 		splx(s);
569 		printf("if_attachdomain called more than once on %s\n",
570 		    ifp->if_xname);
571 		return;
572 	}
573 	ifp->if_afdata_initialized = domain_init_status;
574 	IF_AFDATA_UNLOCK(ifp);
575 
576 	/* address family dependent data region */
577 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
578 	for (dp = domains; dp; dp = dp->dom_next) {
579 		if (dp->dom_ifattach)
580 			ifp->if_afdata[dp->dom_family] =
581 			    (*dp->dom_ifattach)(ifp);
582 	}
583 
584 	splx(s);
585 }
586 
587 /*
588  * Remove any unicast or broadcast network addresses from an interface.
589  */
590 
591 void
592 if_purgeaddrs(struct ifnet *ifp)
593 {
594 	struct ifaddr *ifa, *next;
595 
596 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
597 		if (ifa->ifa_addr->sa_family == AF_LINK)
598 			continue;
599 #ifdef INET
600 		/* XXX: Ugly!! ad hoc just for INET */
601 		if (ifa->ifa_addr->sa_family == AF_INET) {
602 			struct ifaliasreq ifr;
603 
604 			bzero(&ifr, sizeof(ifr));
605 			ifr.ifra_addr = *ifa->ifa_addr;
606 			if (ifa->ifa_dstaddr)
607 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
608 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
609 			    NULL) == 0)
610 				continue;
611 		}
612 #endif /* INET */
613 #ifdef INET6
614 		if (ifa->ifa_addr->sa_family == AF_INET6) {
615 			in6_purgeaddr(ifa);
616 			/* ifp_addrhead is already updated */
617 			continue;
618 		}
619 #endif /* INET6 */
620 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
621 		IFAFREE(ifa);
622 	}
623 }
624 
625 /*
626  * Remove any multicast network addresses from an interface.
627  */
628 static void
629 if_purgemaddrs(struct ifnet *ifp)
630 {
631 	struct ifmultiaddr *ifma;
632 	struct ifmultiaddr *next;
633 
634 	IF_ADDR_LOCK(ifp);
635 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
636 		if_delmulti_locked(ifp, ifma, 1);
637 	IF_ADDR_UNLOCK(ifp);
638 }
639 
640 /*
641  * Detach an interface, removing it from the
642  * list of "active" interfaces.
643  *
644  * XXXRW: There are some significant questions about event ordering, and
645  * how to prevent things from starting to use the interface during detach.
646  */
647 void
648 if_detach(struct ifnet *ifp)
649 {
650 	struct ifaddr *ifa;
651 	struct radix_node_head	*rnh;
652 	int s;
653 	int i;
654 	struct domain *dp;
655  	struct ifnet *iter;
656  	int found = 0;
657 
658 	IFNET_WLOCK();
659 	TAILQ_FOREACH(iter, &ifnet, if_link)
660 		if (iter == ifp) {
661 			TAILQ_REMOVE(&ifnet, ifp, if_link);
662 			found = 1;
663 			break;
664 		}
665 	IFNET_WUNLOCK();
666 	if (!found)
667 		return;
668 
669 	/*
670 	 * Remove/wait for pending events.
671 	 */
672 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
673 
674 	/*
675 	 * Remove routes and flush queues.
676 	 */
677 	s = splnet();
678 	if_down(ifp);
679 #ifdef ALTQ
680 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
681 		altq_disable(&ifp->if_snd);
682 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
683 		altq_detach(&ifp->if_snd);
684 #endif
685 
686 	if_purgeaddrs(ifp);
687 
688 #ifdef INET
689 	in_ifdetach(ifp);
690 #endif
691 
692 #ifdef INET6
693 	/*
694 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
695 	 * before removing routing entries below, since IPv6 interface direct
696 	 * routes are expected to be removed by the IPv6-specific kernel API.
697 	 * Otherwise, the kernel will detect some inconsistency and bark it.
698 	 */
699 	in6_ifdetach(ifp);
700 #endif
701 	if_purgemaddrs(ifp);
702 
703 	/*
704 	 * Remove link ifaddr pointer and maybe decrement if_index.
705 	 * Clean up all addresses.
706 	 */
707 	ifp->if_addr = NULL;
708 	destroy_dev(ifdev_byindex(ifp->if_index));
709 	ifdev_byindex(ifp->if_index) = NULL;
710 
711 	/* We can now free link ifaddr. */
712 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
713 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
714 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
715 		IFAFREE(ifa);
716 	}
717 
718 	/*
719 	 * Delete all remaining routes using this interface
720 	 * Unfortuneatly the only way to do this is to slog through
721 	 * the entire routing table looking for routes which point
722 	 * to this interface...oh well...
723 	 */
724 	for (i = 1; i <= AF_MAX; i++) {
725 		if ((rnh = rt_tables[i]) == NULL)
726 			continue;
727 		RADIX_NODE_HEAD_LOCK(rnh);
728 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
729 		RADIX_NODE_HEAD_UNLOCK(rnh);
730 	}
731 
732 	/* Announce that the interface is gone. */
733 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
734 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
735 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
736 
737 	IF_AFDATA_LOCK(ifp);
738 	for (dp = domains; dp; dp = dp->dom_next) {
739 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
740 			(*dp->dom_ifdetach)(ifp,
741 			    ifp->if_afdata[dp->dom_family]);
742 	}
743 	IF_AFDATA_UNLOCK(ifp);
744 
745 #ifdef MAC
746 	mac_destroy_ifnet(ifp);
747 #endif /* MAC */
748 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
749 	knlist_clear(&ifp->if_klist, 0);
750 	knlist_destroy(&ifp->if_klist);
751 	mtx_destroy(&ifp->if_snd.ifq_mtx);
752 	IF_AFDATA_DESTROY(ifp);
753 	splx(s);
754 }
755 
756 /*
757  * Add a group to an interface
758  */
759 int
760 if_addgroup(struct ifnet *ifp, const char *groupname)
761 {
762 	struct ifg_list		*ifgl;
763 	struct ifg_group	*ifg = NULL;
764 	struct ifg_member	*ifgm;
765 
766 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
767 	    groupname[strlen(groupname) - 1] <= '9')
768 		return (EINVAL);
769 
770 	IFNET_WLOCK();
771 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
772 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
773 			IFNET_WUNLOCK();
774 			return (EEXIST);
775 		}
776 
777 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
778 	    M_NOWAIT)) == NULL) {
779 	    	IFNET_WUNLOCK();
780 		return (ENOMEM);
781 	}
782 
783 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
784 	    M_TEMP, M_NOWAIT)) == NULL) {
785 		free(ifgl, M_TEMP);
786 		IFNET_WUNLOCK();
787 		return (ENOMEM);
788 	}
789 
790 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
791 		if (!strcmp(ifg->ifg_group, groupname))
792 			break;
793 
794 	if (ifg == NULL) {
795 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
796 		    M_TEMP, M_NOWAIT)) == NULL) {
797 			free(ifgl, M_TEMP);
798 			free(ifgm, M_TEMP);
799 			IFNET_WUNLOCK();
800 			return (ENOMEM);
801 		}
802 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
803 		ifg->ifg_refcnt = 0;
804 		TAILQ_INIT(&ifg->ifg_members);
805 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
806 		TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
807 	}
808 
809 	ifg->ifg_refcnt++;
810 	ifgl->ifgl_group = ifg;
811 	ifgm->ifgm_ifp = ifp;
812 
813 	IF_ADDR_LOCK(ifp);
814 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
815 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
816 	IF_ADDR_UNLOCK(ifp);
817 
818 	IFNET_WUNLOCK();
819 
820 	EVENTHANDLER_INVOKE(group_change_event, groupname);
821 
822 	return (0);
823 }
824 
825 /*
826  * Remove a group from an interface
827  */
828 int
829 if_delgroup(struct ifnet *ifp, const char *groupname)
830 {
831 	struct ifg_list		*ifgl;
832 	struct ifg_member	*ifgm;
833 
834 	IFNET_WLOCK();
835 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
836 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
837 			break;
838 	if (ifgl == NULL) {
839 		IFNET_WUNLOCK();
840 		return (ENOENT);
841 	}
842 
843 	IF_ADDR_LOCK(ifp);
844 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
845 	IF_ADDR_UNLOCK(ifp);
846 
847 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
848 		if (ifgm->ifgm_ifp == ifp)
849 			break;
850 
851 	if (ifgm != NULL) {
852 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
853 		free(ifgm, M_TEMP);
854 	}
855 
856 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
857 		TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
858 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
859 		free(ifgl->ifgl_group, M_TEMP);
860 	}
861 	IFNET_WUNLOCK();
862 
863 	free(ifgl, M_TEMP);
864 
865 	EVENTHANDLER_INVOKE(group_change_event, groupname);
866 
867 	return (0);
868 }
869 
870 /*
871  * Stores all groups from an interface in memory pointed
872  * to by data
873  */
874 static int
875 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
876 {
877 	int			 len, error;
878 	struct ifg_list		*ifgl;
879 	struct ifg_req		 ifgrq, *ifgp;
880 	struct ifgroupreq	*ifgr = data;
881 
882 	if (ifgr->ifgr_len == 0) {
883 		IF_ADDR_LOCK(ifp);
884 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
885 			ifgr->ifgr_len += sizeof(struct ifg_req);
886 		IF_ADDR_UNLOCK(ifp);
887 		return (0);
888 	}
889 
890 	len = ifgr->ifgr_len;
891 	ifgp = ifgr->ifgr_groups;
892 	/* XXX: wire */
893 	IF_ADDR_LOCK(ifp);
894 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
895 		if (len < sizeof(ifgrq)) {
896 			IF_ADDR_UNLOCK(ifp);
897 			return (EINVAL);
898 		}
899 		bzero(&ifgrq, sizeof ifgrq);
900 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
901 		    sizeof(ifgrq.ifgrq_group));
902 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
903 		    	IF_ADDR_UNLOCK(ifp);
904 			return (error);
905 		}
906 		len -= sizeof(ifgrq);
907 		ifgp++;
908 	}
909 	IF_ADDR_UNLOCK(ifp);
910 
911 	return (0);
912 }
913 
914 /*
915  * Stores all members of a group in memory pointed to by data
916  */
917 static int
918 if_getgroupmembers(struct ifgroupreq *data)
919 {
920 	struct ifgroupreq	*ifgr = data;
921 	struct ifg_group	*ifg;
922 	struct ifg_member	*ifgm;
923 	struct ifg_req		 ifgrq, *ifgp;
924 	int			 len, error;
925 
926 	IFNET_RLOCK();
927 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
928 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
929 			break;
930 	if (ifg == NULL) {
931 		IFNET_RUNLOCK();
932 		return (ENOENT);
933 	}
934 
935 	if (ifgr->ifgr_len == 0) {
936 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
937 			ifgr->ifgr_len += sizeof(ifgrq);
938 		IFNET_RUNLOCK();
939 		return (0);
940 	}
941 
942 	len = ifgr->ifgr_len;
943 	ifgp = ifgr->ifgr_groups;
944 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
945 		if (len < sizeof(ifgrq)) {
946 			IFNET_RUNLOCK();
947 			return (EINVAL);
948 		}
949 		bzero(&ifgrq, sizeof ifgrq);
950 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
951 		    sizeof(ifgrq.ifgrq_member));
952 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
953 			IFNET_RUNLOCK();
954 			return (error);
955 		}
956 		len -= sizeof(ifgrq);
957 		ifgp++;
958 	}
959 	IFNET_RUNLOCK();
960 
961 	return (0);
962 }
963 
964 /*
965  * Delete Routes for a Network Interface
966  *
967  * Called for each routing entry via the rnh->rnh_walktree() call above
968  * to delete all route entries referencing a detaching network interface.
969  *
970  * Arguments:
971  *	rn	pointer to node in the routing table
972  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
973  *
974  * Returns:
975  *	0	successful
976  *	errno	failed - reason indicated
977  *
978  */
979 static int
980 if_rtdel(struct radix_node *rn, void *arg)
981 {
982 	struct rtentry	*rt = (struct rtentry *)rn;
983 	struct ifnet	*ifp = arg;
984 	int		err;
985 
986 	if (rt->rt_ifp == ifp) {
987 
988 		/*
989 		 * Protect (sorta) against walktree recursion problems
990 		 * with cloned routes
991 		 */
992 		if ((rt->rt_flags & RTF_UP) == 0)
993 			return (0);
994 
995 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
996 				rt_mask(rt), rt->rt_flags,
997 				(struct rtentry **) NULL);
998 		if (err) {
999 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1000 		}
1001 	}
1002 
1003 	return (0);
1004 }
1005 
1006 /*
1007  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1008  * structs used to represent other address families, it is necessary
1009  * to perform a different comparison.
1010  */
1011 
1012 #define	sa_equal(a1, a2)	\
1013 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1014 
1015 #define	sa_dl_equal(a1, a2)	\
1016 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1017 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1018 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1019 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1020 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1021 
1022 /*
1023  * Locate an interface based on a complete address.
1024  */
1025 /*ARGSUSED*/
1026 struct ifaddr *
1027 ifa_ifwithaddr(struct sockaddr *addr)
1028 {
1029 	struct ifnet *ifp;
1030 	struct ifaddr *ifa;
1031 
1032 	IFNET_RLOCK();
1033 	TAILQ_FOREACH(ifp, &ifnet, if_link)
1034 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1035 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1036 				continue;
1037 			if (sa_equal(addr, ifa->ifa_addr))
1038 				goto done;
1039 			/* IP6 doesn't have broadcast */
1040 			if ((ifp->if_flags & IFF_BROADCAST) &&
1041 			    ifa->ifa_broadaddr &&
1042 			    ifa->ifa_broadaddr->sa_len != 0 &&
1043 			    sa_equal(ifa->ifa_broadaddr, addr))
1044 				goto done;
1045 		}
1046 	ifa = NULL;
1047 done:
1048 	IFNET_RUNLOCK();
1049 	return (ifa);
1050 }
1051 
1052 /*
1053  * Locate an interface based on the broadcast address.
1054  */
1055 /* ARGSUSED */
1056 struct ifaddr *
1057 ifa_ifwithbroadaddr(struct sockaddr *addr)
1058 {
1059 	struct ifnet *ifp;
1060 	struct ifaddr *ifa;
1061 
1062 	IFNET_RLOCK();
1063 	TAILQ_FOREACH(ifp, &ifnet, if_link)
1064 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1065 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1066 				continue;
1067 			if ((ifp->if_flags & IFF_BROADCAST) &&
1068 			    ifa->ifa_broadaddr &&
1069 			    ifa->ifa_broadaddr->sa_len != 0 &&
1070 			    sa_equal(ifa->ifa_broadaddr, addr))
1071 				goto done;
1072 		}
1073 	ifa = NULL;
1074 done:
1075 	IFNET_RUNLOCK();
1076 	return (ifa);
1077 }
1078 
1079 /*
1080  * Locate the point to point interface with a given destination address.
1081  */
1082 /*ARGSUSED*/
1083 struct ifaddr *
1084 ifa_ifwithdstaddr(struct sockaddr *addr)
1085 {
1086 	struct ifnet *ifp;
1087 	struct ifaddr *ifa;
1088 
1089 	IFNET_RLOCK();
1090 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1091 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1092 			continue;
1093 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1094 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1095 				continue;
1096 			if (ifa->ifa_dstaddr &&
1097 			    sa_equal(addr, ifa->ifa_dstaddr))
1098 				goto done;
1099 		}
1100 	}
1101 	ifa = NULL;
1102 done:
1103 	IFNET_RUNLOCK();
1104 	return (ifa);
1105 }
1106 
1107 /*
1108  * Find an interface on a specific network.  If many, choice
1109  * is most specific found.
1110  */
1111 struct ifaddr *
1112 ifa_ifwithnet(struct sockaddr *addr)
1113 {
1114 	struct ifnet *ifp;
1115 	struct ifaddr *ifa;
1116 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
1117 	u_int af = addr->sa_family;
1118 	char *addr_data = addr->sa_data, *cplim;
1119 
1120 	/*
1121 	 * AF_LINK addresses can be looked up directly by their index number,
1122 	 * so do that if we can.
1123 	 */
1124 	if (af == AF_LINK) {
1125 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1126 	    if (sdl->sdl_index && sdl->sdl_index <= if_index)
1127 		return (ifaddr_byindex(sdl->sdl_index));
1128 	}
1129 
1130 	/*
1131 	 * Scan though each interface, looking for ones that have
1132 	 * addresses in this address family.
1133 	 */
1134 	IFNET_RLOCK();
1135 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1136 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1137 			char *cp, *cp2, *cp3;
1138 
1139 			if (ifa->ifa_addr->sa_family != af)
1140 next:				continue;
1141 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1142 				/*
1143 				 * This is a bit broken as it doesn't
1144 				 * take into account that the remote end may
1145 				 * be a single node in the network we are
1146 				 * looking for.
1147 				 * The trouble is that we don't know the
1148 				 * netmask for the remote end.
1149 				 */
1150 				if (ifa->ifa_dstaddr != 0 &&
1151 				    sa_equal(addr, ifa->ifa_dstaddr))
1152 					goto done;
1153 			} else {
1154 				/*
1155 				 * if we have a special address handler,
1156 				 * then use it instead of the generic one.
1157 				 */
1158 				if (ifa->ifa_claim_addr) {
1159 					if ((*ifa->ifa_claim_addr)(ifa, addr))
1160 						goto done;
1161 					continue;
1162 				}
1163 
1164 				/*
1165 				 * Scan all the bits in the ifa's address.
1166 				 * If a bit dissagrees with what we are
1167 				 * looking for, mask it with the netmask
1168 				 * to see if it really matters.
1169 				 * (A byte at a time)
1170 				 */
1171 				if (ifa->ifa_netmask == 0)
1172 					continue;
1173 				cp = addr_data;
1174 				cp2 = ifa->ifa_addr->sa_data;
1175 				cp3 = ifa->ifa_netmask->sa_data;
1176 				cplim = ifa->ifa_netmask->sa_len
1177 					+ (char *)ifa->ifa_netmask;
1178 				while (cp3 < cplim)
1179 					if ((*cp++ ^ *cp2++) & *cp3++)
1180 						goto next; /* next address! */
1181 				/*
1182 				 * If the netmask of what we just found
1183 				 * is more specific than what we had before
1184 				 * (if we had one) then remember the new one
1185 				 * before continuing to search
1186 				 * for an even better one.
1187 				 */
1188 				if (ifa_maybe == 0 ||
1189 				    rn_refines((caddr_t)ifa->ifa_netmask,
1190 				    (caddr_t)ifa_maybe->ifa_netmask))
1191 					ifa_maybe = ifa;
1192 			}
1193 		}
1194 	}
1195 	ifa = ifa_maybe;
1196 done:
1197 	IFNET_RUNLOCK();
1198 	return (ifa);
1199 }
1200 
1201 /*
1202  * Find an interface address specific to an interface best matching
1203  * a given address.
1204  */
1205 struct ifaddr *
1206 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1207 {
1208 	struct ifaddr *ifa;
1209 	char *cp, *cp2, *cp3;
1210 	char *cplim;
1211 	struct ifaddr *ifa_maybe = 0;
1212 	u_int af = addr->sa_family;
1213 
1214 	if (af >= AF_MAX)
1215 		return (0);
1216 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1217 		if (ifa->ifa_addr->sa_family != af)
1218 			continue;
1219 		if (ifa_maybe == 0)
1220 			ifa_maybe = ifa;
1221 		if (ifa->ifa_netmask == 0) {
1222 			if (sa_equal(addr, ifa->ifa_addr) ||
1223 			    (ifa->ifa_dstaddr &&
1224 			    sa_equal(addr, ifa->ifa_dstaddr)))
1225 				goto done;
1226 			continue;
1227 		}
1228 		if (ifp->if_flags & IFF_POINTOPOINT) {
1229 			if (sa_equal(addr, ifa->ifa_dstaddr))
1230 				goto done;
1231 		} else {
1232 			cp = addr->sa_data;
1233 			cp2 = ifa->ifa_addr->sa_data;
1234 			cp3 = ifa->ifa_netmask->sa_data;
1235 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1236 			for (; cp3 < cplim; cp3++)
1237 				if ((*cp++ ^ *cp2++) & *cp3)
1238 					break;
1239 			if (cp3 == cplim)
1240 				goto done;
1241 		}
1242 	}
1243 	ifa = ifa_maybe;
1244 done:
1245 	return (ifa);
1246 }
1247 
1248 #include <net/route.h>
1249 
1250 /*
1251  * Default action when installing a route with a Link Level gateway.
1252  * Lookup an appropriate real ifa to point to.
1253  * This should be moved to /sys/net/link.c eventually.
1254  */
1255 static void
1256 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1257 {
1258 	struct ifaddr *ifa, *oifa;
1259 	struct sockaddr *dst;
1260 	struct ifnet *ifp;
1261 
1262 	RT_LOCK_ASSERT(rt);
1263 
1264 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1265 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1266 		return;
1267 	ifa = ifaof_ifpforaddr(dst, ifp);
1268 	if (ifa) {
1269 		IFAREF(ifa);		/* XXX */
1270 		oifa = rt->rt_ifa;
1271 		rt->rt_ifa = ifa;
1272 		IFAFREE(oifa);
1273 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1274 			ifa->ifa_rtrequest(cmd, rt, info);
1275 	}
1276 }
1277 
1278 /*
1279  * Mark an interface down and notify protocols of
1280  * the transition.
1281  * NOTE: must be called at splnet or eqivalent.
1282  */
1283 static void
1284 if_unroute(struct ifnet *ifp, int flag, int fam)
1285 {
1286 	struct ifaddr *ifa;
1287 
1288 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1289 
1290 	ifp->if_flags &= ~flag;
1291 	getmicrotime(&ifp->if_lastchange);
1292 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1293 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1294 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1295 	if_qflush(&ifp->if_snd);
1296 #ifdef DEV_CARP
1297 	if (ifp->if_carp)
1298 		carp_carpdev_state(ifp->if_carp);
1299 #endif
1300 	rt_ifmsg(ifp);
1301 }
1302 
1303 /*
1304  * Mark an interface up and notify protocols of
1305  * the transition.
1306  * NOTE: must be called at splnet or eqivalent.
1307  */
1308 static void
1309 if_route(struct ifnet *ifp, int flag, int fam)
1310 {
1311 	struct ifaddr *ifa;
1312 
1313 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1314 
1315 	ifp->if_flags |= flag;
1316 	getmicrotime(&ifp->if_lastchange);
1317 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1318 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1319 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1320 #ifdef DEV_CARP
1321 	if (ifp->if_carp)
1322 		carp_carpdev_state(ifp->if_carp);
1323 #endif
1324 	rt_ifmsg(ifp);
1325 #ifdef INET6
1326 	in6_if_up(ifp);
1327 #endif
1328 }
1329 
1330 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1331 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1332 
1333 /*
1334  * Handle a change in the interface link state. To avoid LORs
1335  * between driver lock and upper layer locks, as well as possible
1336  * recursions, we post event to taskqueue, and all job
1337  * is done in static do_link_state_change().
1338  */
1339 void
1340 if_link_state_change(struct ifnet *ifp, int link_state)
1341 {
1342 	/* Return if state hasn't changed. */
1343 	if (ifp->if_link_state == link_state)
1344 		return;
1345 
1346 	ifp->if_link_state = link_state;
1347 
1348 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1349 }
1350 
1351 static void
1352 do_link_state_change(void *arg, int pending)
1353 {
1354 	struct ifnet *ifp = (struct ifnet *)arg;
1355 	int link_state = ifp->if_link_state;
1356 	int link;
1357 
1358 	/* Notify that the link state has changed. */
1359 	rt_ifmsg(ifp);
1360 	if (link_state == LINK_STATE_UP)
1361 		link = NOTE_LINKUP;
1362 	else if (link_state == LINK_STATE_DOWN)
1363 		link = NOTE_LINKDOWN;
1364 	else
1365 		link = NOTE_LINKINV;
1366 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1367 	if (ifp->if_vlantrunk != NULL)
1368 		(*vlan_link_state_p)(ifp, link);
1369 
1370 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1371 	    IFP2AC(ifp)->ac_netgraph != NULL)
1372 		(*ng_ether_link_state_p)(ifp, link_state);
1373 #ifdef DEV_CARP
1374 	if (ifp->if_carp)
1375 		carp_carpdev_state(ifp->if_carp);
1376 #endif
1377 	if (ifp->if_bridge) {
1378 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1379 		(*bstp_linkstate_p)(ifp, link_state);
1380 	}
1381 
1382 	devctl_notify("IFNET", ifp->if_xname,
1383 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1384 	if (pending > 1)
1385 		if_printf(ifp, "%d link states coalesced\n", pending);
1386 	if (log_link_state_change)
1387 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1388 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1389 }
1390 
1391 /*
1392  * Mark an interface down and notify protocols of
1393  * the transition.
1394  * NOTE: must be called at splnet or eqivalent.
1395  */
1396 void
1397 if_down(struct ifnet *ifp)
1398 {
1399 
1400 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1401 }
1402 
1403 /*
1404  * Mark an interface up and notify protocols of
1405  * the transition.
1406  * NOTE: must be called at splnet or eqivalent.
1407  */
1408 void
1409 if_up(struct ifnet *ifp)
1410 {
1411 
1412 	if_route(ifp, IFF_UP, AF_UNSPEC);
1413 }
1414 
1415 /*
1416  * Flush an interface queue.
1417  */
1418 static void
1419 if_qflush(struct ifaltq *ifq)
1420 {
1421 	struct mbuf *m, *n;
1422 
1423 	IFQ_LOCK(ifq);
1424 #ifdef ALTQ
1425 	if (ALTQ_IS_ENABLED(ifq))
1426 		ALTQ_PURGE(ifq);
1427 #endif
1428 	n = ifq->ifq_head;
1429 	while ((m = n) != 0) {
1430 		n = m->m_act;
1431 		m_freem(m);
1432 	}
1433 	ifq->ifq_head = 0;
1434 	ifq->ifq_tail = 0;
1435 	ifq->ifq_len = 0;
1436 	IFQ_UNLOCK(ifq);
1437 }
1438 
1439 /*
1440  * Handle interface watchdog timer routines.  Called
1441  * from softclock, we decrement timers (if set) and
1442  * call the appropriate interface routine on expiration.
1443  *
1444  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1445  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1446  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1447  */
1448 static void
1449 if_slowtimo(void *arg)
1450 {
1451 	struct ifnet *ifp;
1452 	int s = splimp();
1453 
1454 	IFNET_RLOCK();
1455 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1456 		if (ifp->if_timer == 0 || --ifp->if_timer)
1457 			continue;
1458 		if (ifp->if_watchdog)
1459 			(*ifp->if_watchdog)(ifp);
1460 	}
1461 	IFNET_RUNLOCK();
1462 	splx(s);
1463 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1464 }
1465 
1466 /*
1467  * Map interface name to
1468  * interface structure pointer.
1469  */
1470 struct ifnet *
1471 ifunit(const char *name)
1472 {
1473 	struct ifnet *ifp;
1474 
1475 	IFNET_RLOCK();
1476 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1477 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1478 			break;
1479 	}
1480 	IFNET_RUNLOCK();
1481 	return (ifp);
1482 }
1483 
1484 /*
1485  * Hardware specific interface ioctls.
1486  */
1487 static int
1488 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1489 {
1490 	struct ifreq *ifr;
1491 	struct ifstat *ifs;
1492 	int error = 0;
1493 	int new_flags, temp_flags;
1494 	size_t namelen, onamelen;
1495 	char new_name[IFNAMSIZ];
1496 	struct ifaddr *ifa;
1497 	struct sockaddr_dl *sdl;
1498 
1499 	ifr = (struct ifreq *)data;
1500 	switch (cmd) {
1501 	case SIOCGIFINDEX:
1502 		ifr->ifr_index = ifp->if_index;
1503 		break;
1504 
1505 	case SIOCGIFFLAGS:
1506 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1507 		ifr->ifr_flags = temp_flags & 0xffff;
1508 		ifr->ifr_flagshigh = temp_flags >> 16;
1509 		break;
1510 
1511 	case SIOCGIFCAP:
1512 		ifr->ifr_reqcap = ifp->if_capabilities;
1513 		ifr->ifr_curcap = ifp->if_capenable;
1514 		break;
1515 
1516 #ifdef MAC
1517 	case SIOCGIFMAC:
1518 		error = mac_ioctl_ifnet_get(td->td_ucred, ifr, ifp);
1519 		break;
1520 #endif
1521 
1522 	case SIOCGIFMETRIC:
1523 		ifr->ifr_metric = ifp->if_metric;
1524 		break;
1525 
1526 	case SIOCGIFMTU:
1527 		ifr->ifr_mtu = ifp->if_mtu;
1528 		break;
1529 
1530 	case SIOCGIFPHYS:
1531 		ifr->ifr_phys = ifp->if_physical;
1532 		break;
1533 
1534 	case SIOCSIFFLAGS:
1535 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
1536 		if (error)
1537 			return (error);
1538 		/*
1539 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1540 		 * check, so we don't need special handling here yet.
1541 		 */
1542 		new_flags = (ifr->ifr_flags & 0xffff) |
1543 		    (ifr->ifr_flagshigh << 16);
1544 		if (ifp->if_flags & IFF_SMART) {
1545 			/* Smart drivers twiddle their own routes */
1546 		} else if (ifp->if_flags & IFF_UP &&
1547 		    (new_flags & IFF_UP) == 0) {
1548 			int s = splimp();
1549 			if_down(ifp);
1550 			splx(s);
1551 		} else if (new_flags & IFF_UP &&
1552 		    (ifp->if_flags & IFF_UP) == 0) {
1553 			int s = splimp();
1554 			if_up(ifp);
1555 			splx(s);
1556 		}
1557 		/* See if permanently promiscuous mode bit is about to flip */
1558 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1559 			if (new_flags & IFF_PPROMISC)
1560 				ifp->if_flags |= IFF_PROMISC;
1561 			else if (ifp->if_pcount == 0)
1562 				ifp->if_flags &= ~IFF_PROMISC;
1563 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1564 			    ifp->if_xname,
1565 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1566 		}
1567 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1568 			(new_flags &~ IFF_CANTCHANGE);
1569 		if (ifp->if_ioctl) {
1570 			IFF_LOCKGIANT(ifp);
1571 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1572 			IFF_UNLOCKGIANT(ifp);
1573 		}
1574 		getmicrotime(&ifp->if_lastchange);
1575 		break;
1576 
1577 	case SIOCSIFCAP:
1578 		error = priv_check(td, PRIV_NET_SETIFCAP);
1579 		if (error)
1580 			return (error);
1581 		if (ifp->if_ioctl == NULL)
1582 			return (EOPNOTSUPP);
1583 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1584 			return (EINVAL);
1585 		IFF_LOCKGIANT(ifp);
1586 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1587 		IFF_UNLOCKGIANT(ifp);
1588 		if (error == 0)
1589 			getmicrotime(&ifp->if_lastchange);
1590 		break;
1591 
1592 #ifdef MAC
1593 	case SIOCSIFMAC:
1594 		error = mac_ioctl_ifnet_set(td->td_ucred, ifr, ifp);
1595 		break;
1596 #endif
1597 
1598 	case SIOCSIFNAME:
1599 		error = priv_check(td, PRIV_NET_SETIFNAME);
1600 		if (error)
1601 			return (error);
1602 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1603 		if (error != 0)
1604 			return (error);
1605 		if (new_name[0] == '\0')
1606 			return (EINVAL);
1607 		if (ifunit(new_name) != NULL)
1608 			return (EEXIST);
1609 
1610 		/* Announce the departure of the interface. */
1611 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1612 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1613 
1614 		log(LOG_INFO, "%s: changing name to '%s'\n",
1615 		    ifp->if_xname, new_name);
1616 
1617 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1618 		ifa = ifp->if_addr;
1619 		IFA_LOCK(ifa);
1620 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1621 		namelen = strlen(new_name);
1622 		onamelen = sdl->sdl_nlen;
1623 		/*
1624 		 * Move the address if needed.  This is safe because we
1625 		 * allocate space for a name of length IFNAMSIZ when we
1626 		 * create this in if_attach().
1627 		 */
1628 		if (namelen != onamelen) {
1629 			bcopy(sdl->sdl_data + onamelen,
1630 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1631 		}
1632 		bcopy(new_name, sdl->sdl_data, namelen);
1633 		sdl->sdl_nlen = namelen;
1634 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1635 		bzero(sdl->sdl_data, onamelen);
1636 		while (namelen != 0)
1637 			sdl->sdl_data[--namelen] = 0xff;
1638 		IFA_UNLOCK(ifa);
1639 
1640 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1641 		/* Announce the return of the interface. */
1642 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1643 		break;
1644 
1645 	case SIOCSIFMETRIC:
1646 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
1647 		if (error)
1648 			return (error);
1649 		ifp->if_metric = ifr->ifr_metric;
1650 		getmicrotime(&ifp->if_lastchange);
1651 		break;
1652 
1653 	case SIOCSIFPHYS:
1654 		error = priv_check(td, PRIV_NET_SETIFPHYS);
1655 		if (error)
1656 			return (error);
1657 		if (ifp->if_ioctl == NULL)
1658 			return (EOPNOTSUPP);
1659 		IFF_LOCKGIANT(ifp);
1660 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1661 		IFF_UNLOCKGIANT(ifp);
1662 		if (error == 0)
1663 			getmicrotime(&ifp->if_lastchange);
1664 		break;
1665 
1666 	case SIOCSIFMTU:
1667 	{
1668 		u_long oldmtu = ifp->if_mtu;
1669 
1670 		error = priv_check(td, PRIV_NET_SETIFMTU);
1671 		if (error)
1672 			return (error);
1673 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1674 			return (EINVAL);
1675 		if (ifp->if_ioctl == NULL)
1676 			return (EOPNOTSUPP);
1677 		IFF_LOCKGIANT(ifp);
1678 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1679 		IFF_UNLOCKGIANT(ifp);
1680 		if (error == 0) {
1681 			getmicrotime(&ifp->if_lastchange);
1682 			rt_ifmsg(ifp);
1683 		}
1684 		/*
1685 		 * If the link MTU changed, do network layer specific procedure.
1686 		 */
1687 		if (ifp->if_mtu != oldmtu) {
1688 #ifdef INET6
1689 			nd6_setmtu(ifp);
1690 #endif
1691 		}
1692 		break;
1693 	}
1694 
1695 	case SIOCADDMULTI:
1696 	case SIOCDELMULTI:
1697 		if (cmd == SIOCADDMULTI)
1698 			error = priv_check(td, PRIV_NET_ADDMULTI);
1699 		else
1700 			error = priv_check(td, PRIV_NET_DELMULTI);
1701 		if (error)
1702 			return (error);
1703 
1704 		/* Don't allow group membership on non-multicast interfaces. */
1705 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1706 			return (EOPNOTSUPP);
1707 
1708 		/* Don't let users screw up protocols' entries. */
1709 		if (ifr->ifr_addr.sa_family != AF_LINK)
1710 			return (EINVAL);
1711 
1712 		if (cmd == SIOCADDMULTI) {
1713 			struct ifmultiaddr *ifma;
1714 
1715 			/*
1716 			 * Userland is only permitted to join groups once
1717 			 * via the if_addmulti() KPI, because it cannot hold
1718 			 * struct ifmultiaddr * between calls. It may also
1719 			 * lose a race while we check if the membership
1720 			 * already exists.
1721 			 */
1722 			IF_ADDR_LOCK(ifp);
1723 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
1724 			IF_ADDR_UNLOCK(ifp);
1725 			if (ifma != NULL)
1726 				error = EADDRINUSE;
1727 			else
1728 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1729 		} else {
1730 			error = if_delmulti(ifp, &ifr->ifr_addr);
1731 		}
1732 		if (error == 0)
1733 			getmicrotime(&ifp->if_lastchange);
1734 		break;
1735 
1736 	case SIOCSIFPHYADDR:
1737 	case SIOCDIFPHYADDR:
1738 #ifdef INET6
1739 	case SIOCSIFPHYADDR_IN6:
1740 #endif
1741 	case SIOCSLIFPHYADDR:
1742 	case SIOCSIFMEDIA:
1743 	case SIOCSIFGENERIC:
1744 		error = priv_check(td, PRIV_NET_HWIOCTL);
1745 		if (error)
1746 			return (error);
1747 		if (ifp->if_ioctl == NULL)
1748 			return (EOPNOTSUPP);
1749 		IFF_LOCKGIANT(ifp);
1750 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1751 		IFF_UNLOCKGIANT(ifp);
1752 		if (error == 0)
1753 			getmicrotime(&ifp->if_lastchange);
1754 		break;
1755 
1756 	case SIOCGIFSTATUS:
1757 		ifs = (struct ifstat *)data;
1758 		ifs->ascii[0] = '\0';
1759 
1760 	case SIOCGIFPSRCADDR:
1761 	case SIOCGIFPDSTADDR:
1762 	case SIOCGLIFPHYADDR:
1763 	case SIOCGIFMEDIA:
1764 	case SIOCGIFGENERIC:
1765 		if (ifp->if_ioctl == NULL)
1766 			return (EOPNOTSUPP);
1767 		IFF_LOCKGIANT(ifp);
1768 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1769 		IFF_UNLOCKGIANT(ifp);
1770 		break;
1771 
1772 	case SIOCSIFLLADDR:
1773 		error = priv_check(td, PRIV_NET_SETLLADDR);
1774 		if (error)
1775 			return (error);
1776 		error = if_setlladdr(ifp,
1777 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1778 		break;
1779 
1780 	case SIOCAIFGROUP:
1781 	{
1782 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1783 
1784 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
1785 		if (error)
1786 			return (error);
1787 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
1788 			return (error);
1789 		break;
1790 	}
1791 
1792 	case SIOCGIFGROUP:
1793 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
1794 			return (error);
1795 		break;
1796 
1797 	case SIOCDIFGROUP:
1798 	{
1799 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1800 
1801 		error = priv_check(td, PRIV_NET_DELIFGROUP);
1802 		if (error)
1803 			return (error);
1804 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
1805 			return (error);
1806 		break;
1807 	}
1808 
1809 	default:
1810 		error = ENOIOCTL;
1811 		break;
1812 	}
1813 	return (error);
1814 }
1815 
1816 /*
1817  * Interface ioctls.
1818  */
1819 int
1820 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1821 {
1822 	struct ifnet *ifp;
1823 	struct ifreq *ifr;
1824 	int error;
1825 	int oif_flags;
1826 
1827 	switch (cmd) {
1828 	case SIOCGIFCONF:
1829 	case OSIOCGIFCONF:
1830 #ifdef __amd64__
1831 	case SIOCGIFCONF32:
1832 #endif
1833 		return (ifconf(cmd, data));
1834 	}
1835 	ifr = (struct ifreq *)data;
1836 
1837 	switch (cmd) {
1838 	case SIOCIFCREATE:
1839 	case SIOCIFCREATE2:
1840 		error = priv_check(td, PRIV_NET_IFCREATE);
1841 		if (error)
1842 			return (error);
1843 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1844 			cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1845 	case SIOCIFDESTROY:
1846 		error = priv_check(td, PRIV_NET_IFDESTROY);
1847 		if (error)
1848 			return (error);
1849 		return if_clone_destroy(ifr->ifr_name);
1850 
1851 	case SIOCIFGCLONERS:
1852 		return (if_clone_list((struct if_clonereq *)data));
1853 	case SIOCGIFGMEMB:
1854 		return (if_getgroupmembers((struct ifgroupreq *)data));
1855 	}
1856 
1857 	ifp = ifunit(ifr->ifr_name);
1858 	if (ifp == 0)
1859 		return (ENXIO);
1860 
1861 	error = ifhwioctl(cmd, ifp, data, td);
1862 	if (error != ENOIOCTL)
1863 		return (error);
1864 
1865 	oif_flags = ifp->if_flags;
1866 	if (so->so_proto == 0)
1867 		return (EOPNOTSUPP);
1868 #ifndef COMPAT_43
1869 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1870 								 data,
1871 								 ifp, td));
1872 #else
1873 	{
1874 		int ocmd = cmd;
1875 
1876 		switch (cmd) {
1877 
1878 		case SIOCSIFDSTADDR:
1879 		case SIOCSIFADDR:
1880 		case SIOCSIFBRDADDR:
1881 		case SIOCSIFNETMASK:
1882 #if BYTE_ORDER != BIG_ENDIAN
1883 			if (ifr->ifr_addr.sa_family == 0 &&
1884 			    ifr->ifr_addr.sa_len < 16) {
1885 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1886 				ifr->ifr_addr.sa_len = 16;
1887 			}
1888 #else
1889 			if (ifr->ifr_addr.sa_len == 0)
1890 				ifr->ifr_addr.sa_len = 16;
1891 #endif
1892 			break;
1893 
1894 		case OSIOCGIFADDR:
1895 			cmd = SIOCGIFADDR;
1896 			break;
1897 
1898 		case OSIOCGIFDSTADDR:
1899 			cmd = SIOCGIFDSTADDR;
1900 			break;
1901 
1902 		case OSIOCGIFBRDADDR:
1903 			cmd = SIOCGIFBRDADDR;
1904 			break;
1905 
1906 		case OSIOCGIFNETMASK:
1907 			cmd = SIOCGIFNETMASK;
1908 		}
1909 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
1910 								   cmd,
1911 								   data,
1912 								   ifp, td));
1913 		switch (ocmd) {
1914 
1915 		case OSIOCGIFADDR:
1916 		case OSIOCGIFDSTADDR:
1917 		case OSIOCGIFBRDADDR:
1918 		case OSIOCGIFNETMASK:
1919 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1920 
1921 		}
1922 	}
1923 #endif /* COMPAT_43 */
1924 
1925 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1926 #ifdef INET6
1927 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
1928 		if (ifp->if_flags & IFF_UP) {
1929 			int s = splimp();
1930 			in6_if_up(ifp);
1931 			splx(s);
1932 		}
1933 #endif
1934 	}
1935 	return (error);
1936 }
1937 
1938 /*
1939  * The code common to handling reference counted flags,
1940  * e.g., in ifpromisc() and if_allmulti().
1941  * The "pflag" argument can specify a permanent mode flag to check,
1942  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
1943  *
1944  * Only to be used on stack-owned flags, not driver-owned flags.
1945  */
1946 static int
1947 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
1948 {
1949 	struct ifreq ifr;
1950 	int error;
1951 	int oldflags, oldcount;
1952 
1953 	/* Sanity checks to catch programming errors */
1954 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
1955 	    ("%s: setting driver-owned flag %d", __func__, flag));
1956 
1957 	if (onswitch)
1958 		KASSERT(*refcount >= 0,
1959 		    ("%s: increment negative refcount %d for flag %d",
1960 		    __func__, *refcount, flag));
1961 	else
1962 		KASSERT(*refcount > 0,
1963 		    ("%s: decrement non-positive refcount %d for flag %d",
1964 		    __func__, *refcount, flag));
1965 
1966 	/* In case this mode is permanent, just touch refcount */
1967 	if (ifp->if_flags & pflag) {
1968 		*refcount += onswitch ? 1 : -1;
1969 		return (0);
1970 	}
1971 
1972 	/* Save ifnet parameters for if_ioctl() may fail */
1973 	oldcount = *refcount;
1974 	oldflags = ifp->if_flags;
1975 
1976 	/*
1977 	 * See if we aren't the only and touching refcount is enough.
1978 	 * Actually toggle interface flag if we are the first or last.
1979 	 */
1980 	if (onswitch) {
1981 		if ((*refcount)++)
1982 			return (0);
1983 		ifp->if_flags |= flag;
1984 	} else {
1985 		if (--(*refcount))
1986 			return (0);
1987 		ifp->if_flags &= ~flag;
1988 	}
1989 
1990 	/* Call down the driver since we've changed interface flags */
1991 	if (ifp->if_ioctl == NULL) {
1992 		error = EOPNOTSUPP;
1993 		goto recover;
1994 	}
1995 	ifr.ifr_flags = ifp->if_flags & 0xffff;
1996 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1997 	IFF_LOCKGIANT(ifp);
1998 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1999 	IFF_UNLOCKGIANT(ifp);
2000 	if (error)
2001 		goto recover;
2002 	/* Notify userland that interface flags have changed */
2003 	rt_ifmsg(ifp);
2004 	return (0);
2005 
2006 recover:
2007 	/* Recover after driver error */
2008 	*refcount = oldcount;
2009 	ifp->if_flags = oldflags;
2010 	return (error);
2011 }
2012 
2013 /*
2014  * Set/clear promiscuous mode on interface ifp based on the truth value
2015  * of pswitch.  The calls are reference counted so that only the first
2016  * "on" request actually has an effect, as does the final "off" request.
2017  * Results are undefined if the "off" and "on" requests are not matched.
2018  */
2019 int
2020 ifpromisc(struct ifnet *ifp, int pswitch)
2021 {
2022 	int error;
2023 	int oldflags = ifp->if_flags;
2024 
2025 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2026 			   &ifp->if_pcount, pswitch);
2027 	/* If promiscuous mode status has changed, log a message */
2028 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2029 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2030 		    ifp->if_xname,
2031 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2032 	return (error);
2033 }
2034 
2035 /*
2036  * Return interface configuration
2037  * of system.  List may be used
2038  * in later ioctl's (above) to get
2039  * other information.
2040  */
2041 /*ARGSUSED*/
2042 static int
2043 ifconf(u_long cmd, caddr_t data)
2044 {
2045 	struct ifconf *ifc = (struct ifconf *)data;
2046 #ifdef __amd64__
2047 	struct ifconf32 *ifc32 = (struct ifconf32 *)data;
2048 	struct ifconf ifc_swab;
2049 #endif
2050 	struct ifnet *ifp;
2051 	struct ifaddr *ifa;
2052 	struct ifreq ifr;
2053 	struct sbuf *sb;
2054 	int error, full = 0, valid_len, max_len;
2055 
2056 #ifdef __amd64__
2057 	if (cmd == SIOCGIFCONF32) {
2058 		ifc_swab.ifc_len = ifc32->ifc_len;
2059 		ifc_swab.ifc_buf = (caddr_t)(uintptr_t)ifc32->ifc_buf;
2060 		ifc = &ifc_swab;
2061 	}
2062 #endif
2063 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2064 	max_len = MAXPHYS - 1;
2065 
2066 	/* Prevent hostile input from being able to crash the system */
2067 	if (ifc->ifc_len <= 0)
2068 		return (EINVAL);
2069 
2070 again:
2071 	if (ifc->ifc_len <= max_len) {
2072 		max_len = ifc->ifc_len;
2073 		full = 1;
2074 	}
2075 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2076 	max_len = 0;
2077 	valid_len = 0;
2078 
2079 	IFNET_RLOCK();		/* could sleep XXX */
2080 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
2081 		int addrs;
2082 
2083 		/*
2084 		 * Zero the ifr_name buffer to make sure we don't
2085 		 * disclose the contents of the stack.
2086 		 */
2087 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2088 
2089 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2090 		    >= sizeof(ifr.ifr_name)) {
2091 			sbuf_delete(sb);
2092 			IFNET_RUNLOCK();
2093 			return (ENAMETOOLONG);
2094 		}
2095 
2096 		addrs = 0;
2097 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2098 			struct sockaddr *sa = ifa->ifa_addr;
2099 
2100 			if (jailed(curthread->td_ucred) &&
2101 			    prison_if(curthread->td_ucred, sa))
2102 				continue;
2103 			addrs++;
2104 #ifdef COMPAT_43
2105 			if (cmd == OSIOCGIFCONF) {
2106 				struct osockaddr *osa =
2107 					 (struct osockaddr *)&ifr.ifr_addr;
2108 				ifr.ifr_addr = *sa;
2109 				osa->sa_family = sa->sa_family;
2110 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2111 				max_len += sizeof(ifr);
2112 			} else
2113 #endif
2114 			if (sa->sa_len <= sizeof(*sa)) {
2115 				ifr.ifr_addr = *sa;
2116 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2117 				max_len += sizeof(ifr);
2118 			} else {
2119 				sbuf_bcat(sb, &ifr,
2120 				    offsetof(struct ifreq, ifr_addr));
2121 				max_len += offsetof(struct ifreq, ifr_addr);
2122 				sbuf_bcat(sb, sa, sa->sa_len);
2123 				max_len += sa->sa_len;
2124 			}
2125 
2126 			if (!sbuf_overflowed(sb))
2127 				valid_len = sbuf_len(sb);
2128 		}
2129 		if (addrs == 0) {
2130 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2131 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2132 			max_len += sizeof(ifr);
2133 
2134 			if (!sbuf_overflowed(sb))
2135 				valid_len = sbuf_len(sb);
2136 		}
2137 	}
2138 	IFNET_RUNLOCK();
2139 
2140 	/*
2141 	 * If we didn't allocate enough space (uncommon), try again.  If
2142 	 * we have already allocated as much space as we are allowed,
2143 	 * return what we've got.
2144 	 */
2145 	if (valid_len != max_len && !full) {
2146 		sbuf_delete(sb);
2147 		goto again;
2148 	}
2149 
2150 	ifc->ifc_len = valid_len;
2151 #ifdef __amd64__
2152 	if (cmd == SIOCGIFCONF32)
2153 		ifc32->ifc_len = valid_len;
2154 #endif
2155 	sbuf_finish(sb);
2156 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2157 	sbuf_delete(sb);
2158 	return (error);
2159 }
2160 
2161 /*
2162  * Just like ifpromisc(), but for all-multicast-reception mode.
2163  */
2164 int
2165 if_allmulti(struct ifnet *ifp, int onswitch)
2166 {
2167 
2168 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2169 }
2170 
2171 struct ifmultiaddr *
2172 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2173 {
2174 	struct ifmultiaddr *ifma;
2175 
2176 	IF_ADDR_LOCK_ASSERT(ifp);
2177 
2178 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2179 		if (sa->sa_family == AF_LINK) {
2180 			if (sa_dl_equal(ifma->ifma_addr, sa))
2181 				break;
2182 		} else {
2183 			if (sa_equal(ifma->ifma_addr, sa))
2184 				break;
2185 		}
2186 	}
2187 
2188 	return ifma;
2189 }
2190 
2191 /*
2192  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2193  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2194  * the ifnet multicast address list here, so the caller must do that and
2195  * other setup work (such as notifying the device driver).  The reference
2196  * count is initialized to 1.
2197  */
2198 static struct ifmultiaddr *
2199 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2200     int mflags)
2201 {
2202 	struct ifmultiaddr *ifma;
2203 	struct sockaddr *dupsa;
2204 
2205 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, mflags |
2206 	    M_ZERO);
2207 	if (ifma == NULL)
2208 		return (NULL);
2209 
2210 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, mflags);
2211 	if (dupsa == NULL) {
2212 		FREE(ifma, M_IFMADDR);
2213 		return (NULL);
2214 	}
2215 	bcopy(sa, dupsa, sa->sa_len);
2216 	ifma->ifma_addr = dupsa;
2217 
2218 	ifma->ifma_ifp = ifp;
2219 	ifma->ifma_refcount = 1;
2220 	ifma->ifma_protospec = NULL;
2221 
2222 	if (llsa == NULL) {
2223 		ifma->ifma_lladdr = NULL;
2224 		return (ifma);
2225 	}
2226 
2227 	MALLOC(dupsa, struct sockaddr *, llsa->sa_len, M_IFMADDR, mflags);
2228 	if (dupsa == NULL) {
2229 		FREE(ifma->ifma_addr, M_IFMADDR);
2230 		FREE(ifma, M_IFMADDR);
2231 		return (NULL);
2232 	}
2233 	bcopy(llsa, dupsa, llsa->sa_len);
2234 	ifma->ifma_lladdr = dupsa;
2235 
2236 	return (ifma);
2237 }
2238 
2239 /*
2240  * if_freemulti: free ifmultiaddr structure and possibly attached related
2241  * addresses.  The caller is responsible for implementing reference
2242  * counting, notifying the driver, handling routing messages, and releasing
2243  * any dependent link layer state.
2244  */
2245 static void
2246 if_freemulti(struct ifmultiaddr *ifma)
2247 {
2248 
2249 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2250 	    ifma->ifma_refcount));
2251 	KASSERT(ifma->ifma_protospec == NULL,
2252 	    ("if_freemulti: protospec not NULL"));
2253 
2254 	if (ifma->ifma_lladdr != NULL)
2255 		FREE(ifma->ifma_lladdr, M_IFMADDR);
2256 	FREE(ifma->ifma_addr, M_IFMADDR);
2257 	FREE(ifma, M_IFMADDR);
2258 }
2259 
2260 /*
2261  * Register an additional multicast address with a network interface.
2262  *
2263  * - If the address is already present, bump the reference count on the
2264  *   address and return.
2265  * - If the address is not link-layer, look up a link layer address.
2266  * - Allocate address structures for one or both addresses, and attach to the
2267  *   multicast address list on the interface.  If automatically adding a link
2268  *   layer address, the protocol address will own a reference to the link
2269  *   layer address, to be freed when it is freed.
2270  * - Notify the network device driver of an addition to the multicast address
2271  *   list.
2272  *
2273  * 'sa' points to caller-owned memory with the desired multicast address.
2274  *
2275  * 'retifma' will be used to return a pointer to the resulting multicast
2276  * address reference, if desired.
2277  */
2278 int
2279 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2280     struct ifmultiaddr **retifma)
2281 {
2282 	struct ifmultiaddr *ifma, *ll_ifma;
2283 	struct sockaddr *llsa;
2284 	int error;
2285 
2286 	/*
2287 	 * If the address is already present, return a new reference to it;
2288 	 * otherwise, allocate storage and set up a new address.
2289 	 */
2290 	IF_ADDR_LOCK(ifp);
2291 	ifma = if_findmulti(ifp, sa);
2292 	if (ifma != NULL) {
2293 		ifma->ifma_refcount++;
2294 		if (retifma != NULL)
2295 			*retifma = ifma;
2296 		IF_ADDR_UNLOCK(ifp);
2297 		return (0);
2298 	}
2299 
2300 	/*
2301 	 * The address isn't already present; resolve the protocol address
2302 	 * into a link layer address, and then look that up, bump its
2303 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2304 	 * returned, we will need to free it later.
2305 	 */
2306 	llsa = NULL;
2307 	ll_ifma = NULL;
2308 	if (ifp->if_resolvemulti != NULL) {
2309 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2310 		if (error)
2311 			goto unlock_out;
2312 	}
2313 
2314 	/*
2315 	 * Allocate the new address.  Don't hook it up yet, as we may also
2316 	 * need to allocate a link layer multicast address.
2317 	 */
2318 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2319 	if (ifma == NULL) {
2320 		error = ENOMEM;
2321 		goto free_llsa_out;
2322 	}
2323 
2324 	/*
2325 	 * If a link layer address is found, we'll need to see if it's
2326 	 * already present in the address list, or allocate is as well.
2327 	 * When this block finishes, the link layer address will be on the
2328 	 * list.
2329 	 */
2330 	if (llsa != NULL) {
2331 		ll_ifma = if_findmulti(ifp, llsa);
2332 		if (ll_ifma == NULL) {
2333 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2334 			if (ll_ifma == NULL) {
2335 				--ifma->ifma_refcount;
2336 				if_freemulti(ifma);
2337 				error = ENOMEM;
2338 				goto free_llsa_out;
2339 			}
2340 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
2341 			    ifma_link);
2342 		} else
2343 			ll_ifma->ifma_refcount++;
2344 		ifma->ifma_llifma = ll_ifma;
2345 	}
2346 
2347 	/*
2348 	 * We now have a new multicast address, ifma, and possibly a new or
2349 	 * referenced link layer address.  Add the primary address to the
2350 	 * ifnet address list.
2351 	 */
2352 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2353 
2354 	if (retifma != NULL)
2355 		*retifma = ifma;
2356 
2357 	/*
2358 	 * Must generate the message while holding the lock so that 'ifma'
2359 	 * pointer is still valid.
2360 	 */
2361 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2362 	IF_ADDR_UNLOCK(ifp);
2363 
2364 	/*
2365 	 * We are certain we have added something, so call down to the
2366 	 * interface to let them know about it.
2367 	 */
2368 	if (ifp->if_ioctl != NULL) {
2369 		IFF_LOCKGIANT(ifp);
2370 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2371 		IFF_UNLOCKGIANT(ifp);
2372 	}
2373 
2374 	if (llsa != NULL)
2375 		FREE(llsa, M_IFMADDR);
2376 
2377 	return (0);
2378 
2379 free_llsa_out:
2380 	if (llsa != NULL)
2381 		FREE(llsa, M_IFMADDR);
2382 
2383 unlock_out:
2384 	IF_ADDR_UNLOCK(ifp);
2385 	return (error);
2386 }
2387 
2388 /*
2389  * Delete a multicast group membership by network-layer group address.
2390  *
2391  * Returns ENOENT if the entry could not be found. If ifp no longer
2392  * exists, results are undefined. This entry point should only be used
2393  * from subsystems which do appropriate locking to hold ifp for the
2394  * duration of the call.
2395  * Network-layer protocol domains must use if_delmulti_ifma().
2396  */
2397 int
2398 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2399 {
2400 	struct ifmultiaddr *ifma;
2401 	int lastref;
2402 #ifdef INVARIANTS
2403 	struct ifnet *oifp;
2404 
2405 	IFNET_RLOCK();
2406 	TAILQ_FOREACH(oifp, &ifnet, if_link)
2407 		if (ifp == oifp)
2408 			break;
2409 	if (ifp != oifp)
2410 		ifp = NULL;
2411 	IFNET_RUNLOCK();
2412 
2413 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
2414 #endif
2415 	if (ifp == NULL)
2416 		return (ENOENT);
2417 
2418 	IF_ADDR_LOCK(ifp);
2419 	lastref = 0;
2420 	ifma = if_findmulti(ifp, sa);
2421 	if (ifma != NULL)
2422 		lastref = if_delmulti_locked(ifp, ifma, 0);
2423 	IF_ADDR_UNLOCK(ifp);
2424 
2425 	if (ifma == NULL)
2426 		return (ENOENT);
2427 
2428 	if (lastref && ifp->if_ioctl != NULL) {
2429 		IFF_LOCKGIANT(ifp);
2430 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2431 		IFF_UNLOCKGIANT(ifp);
2432 	}
2433 
2434 	return (0);
2435 }
2436 
2437 /*
2438  * Delete a multicast group membership by group membership pointer.
2439  * Network-layer protocol domains must use this routine.
2440  *
2441  * It is safe to call this routine if the ifp disappeared. Callers should
2442  * hold IFF_LOCKGIANT() to avoid a LOR in case the hardware needs to be
2443  * reconfigured.
2444  */
2445 void
2446 if_delmulti_ifma(struct ifmultiaddr *ifma)
2447 {
2448 	struct ifnet *ifp;
2449 	int lastref;
2450 
2451 	ifp = ifma->ifma_ifp;
2452 #ifdef DIAGNOSTIC
2453 	if (ifp == NULL) {
2454 		printf("%s: ifma_ifp seems to be detached\n", __func__);
2455 	} else {
2456 		struct ifnet *oifp;
2457 
2458 		IFNET_RLOCK();
2459 		TAILQ_FOREACH(oifp, &ifnet, if_link)
2460 			if (ifp == oifp)
2461 				break;
2462 		if (ifp != oifp) {
2463 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
2464 			ifp = NULL;
2465 		}
2466 		IFNET_RUNLOCK();
2467 	}
2468 #endif
2469 	/*
2470 	 * If and only if the ifnet instance exists: Acquire the address lock.
2471 	 */
2472 	if (ifp != NULL)
2473 		IF_ADDR_LOCK(ifp);
2474 
2475 	lastref = if_delmulti_locked(ifp, ifma, 0);
2476 
2477 	if (ifp != NULL) {
2478 		/*
2479 		 * If and only if the ifnet instance exists:
2480 		 *  Release the address lock.
2481 		 *  If the group was left: update the hardware hash filter.
2482 		 */
2483 		IF_ADDR_UNLOCK(ifp);
2484 		if (lastref && ifp->if_ioctl != NULL) {
2485 			IFF_LOCKGIANT(ifp);
2486 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2487 			IFF_UNLOCKGIANT(ifp);
2488 		}
2489 	}
2490 }
2491 
2492 /*
2493  * Perform deletion of network-layer and/or link-layer multicast address.
2494  *
2495  * Return 0 if the reference count was decremented.
2496  * Return 1 if the final reference was released, indicating that the
2497  * hardware hash filter should be reprogrammed.
2498  */
2499 static int
2500 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
2501 {
2502 	struct ifmultiaddr *ll_ifma;
2503 
2504 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
2505 		KASSERT(ifma->ifma_ifp == ifp,
2506 		    ("%s: inconsistent ifp %p", __func__, ifp));
2507 		IF_ADDR_LOCK_ASSERT(ifp);
2508 	}
2509 
2510 	ifp = ifma->ifma_ifp;
2511 
2512 	/*
2513 	 * If the ifnet is detaching, null out references to ifnet,
2514 	 * so that upper protocol layers will notice, and not attempt
2515 	 * to obtain locks for an ifnet which no longer exists. The
2516 	 * routing socket announcement must happen before the ifnet
2517 	 * instance is detached from the system.
2518 	 */
2519 	if (detaching) {
2520 #ifdef DIAGNOSTIC
2521 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
2522 #endif
2523 		/*
2524 		 * ifp may already be nulled out if we are being reentered
2525 		 * to delete the ll_ifma.
2526 		 */
2527 		if (ifp != NULL) {
2528 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
2529 			ifma->ifma_ifp = NULL;
2530 		}
2531 	}
2532 
2533 	if (--ifma->ifma_refcount > 0)
2534 		return 0;
2535 
2536 	/*
2537 	 * If this ifma is a network-layer ifma, a link-layer ifma may
2538 	 * have been associated with it. Release it first if so.
2539 	 */
2540 	ll_ifma = ifma->ifma_llifma;
2541 	if (ll_ifma != NULL) {
2542 		KASSERT(ifma->ifma_lladdr != NULL,
2543 		    ("%s: llifma w/o lladdr", __func__));
2544 		if (detaching)
2545 			ll_ifma->ifma_ifp = NULL;	/* XXX */
2546 		if (--ll_ifma->ifma_refcount == 0) {
2547 			if (ifp != NULL) {
2548 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
2549 				    ifma_link);
2550 			}
2551 			if_freemulti(ll_ifma);
2552 		}
2553 	}
2554 
2555 	if (ifp != NULL)
2556 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2557 
2558 	if_freemulti(ifma);
2559 
2560 	/*
2561 	 * The last reference to this instance of struct ifmultiaddr
2562 	 * was released; the hardware should be notified of this change.
2563 	 */
2564 	return 1;
2565 }
2566 
2567 /*
2568  * Set the link layer address on an interface.
2569  *
2570  * At this time we only support certain types of interfaces,
2571  * and we don't allow the length of the address to change.
2572  */
2573 int
2574 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2575 {
2576 	struct sockaddr_dl *sdl;
2577 	struct ifaddr *ifa;
2578 	struct ifreq ifr;
2579 
2580 	ifa = ifp->if_addr;
2581 	if (ifa == NULL)
2582 		return (EINVAL);
2583 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2584 	if (sdl == NULL)
2585 		return (EINVAL);
2586 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2587 		return (EINVAL);
2588 	switch (ifp->if_type) {
2589 	case IFT_ETHER:
2590 	case IFT_FDDI:
2591 	case IFT_XETHER:
2592 	case IFT_ISO88025:
2593 	case IFT_L2VLAN:
2594 	case IFT_BRIDGE:
2595 	case IFT_ARCNET:
2596 		bcopy(lladdr, LLADDR(sdl), len);
2597 		break;
2598 	default:
2599 		return (ENODEV);
2600 	}
2601 	/*
2602 	 * If the interface is already up, we need
2603 	 * to re-init it in order to reprogram its
2604 	 * address filter.
2605 	 */
2606 	if ((ifp->if_flags & IFF_UP) != 0) {
2607 		if (ifp->if_ioctl) {
2608 			IFF_LOCKGIANT(ifp);
2609 			ifp->if_flags &= ~IFF_UP;
2610 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2611 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2612 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2613 			ifp->if_flags |= IFF_UP;
2614 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2615 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2616 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2617 			IFF_UNLOCKGIANT(ifp);
2618 		}
2619 #ifdef INET
2620 		/*
2621 		 * Also send gratuitous ARPs to notify other nodes about
2622 		 * the address change.
2623 		 */
2624 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2625 			if (ifa->ifa_addr->sa_family == AF_INET)
2626 				arp_ifinit(ifp, ifa);
2627 		}
2628 #endif
2629 	}
2630 	return (0);
2631 }
2632 
2633 /*
2634  * The name argument must be a pointer to storage which will last as
2635  * long as the interface does.  For physical devices, the result of
2636  * device_get_name(dev) is a good choice and for pseudo-devices a
2637  * static string works well.
2638  */
2639 void
2640 if_initname(struct ifnet *ifp, const char *name, int unit)
2641 {
2642 	ifp->if_dname = name;
2643 	ifp->if_dunit = unit;
2644 	if (unit != IF_DUNIT_NONE)
2645 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2646 	else
2647 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2648 }
2649 
2650 int
2651 if_printf(struct ifnet *ifp, const char * fmt, ...)
2652 {
2653 	va_list ap;
2654 	int retval;
2655 
2656 	retval = printf("%s: ", ifp->if_xname);
2657 	va_start(ap, fmt);
2658 	retval += vprintf(fmt, ap);
2659 	va_end(ap);
2660 	return (retval);
2661 }
2662 
2663 /*
2664  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2665  * be called without Giant.  However, we often can't acquire the Giant lock
2666  * at those points; instead, we run it via a task queue that holds Giant via
2667  * if_start_deferred.
2668  *
2669  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2670  * outstanding if_start_deferred() tasks that will run after the free.  This
2671  * probably means waiting in if_detach().
2672  */
2673 void
2674 if_start(struct ifnet *ifp)
2675 {
2676 
2677 	NET_ASSERT_GIANT();
2678 
2679 	if ((ifp->if_flags & IFF_NEEDSGIANT) != 0 && debug_mpsafenet != 0) {
2680 		if (mtx_owned(&Giant))
2681 			(*(ifp)->if_start)(ifp);
2682 		else
2683 			taskqueue_enqueue(taskqueue_swi_giant,
2684 			    &ifp->if_starttask);
2685 	} else
2686 		(*(ifp)->if_start)(ifp);
2687 }
2688 
2689 static void
2690 if_start_deferred(void *context, int pending)
2691 {
2692 	struct ifnet *ifp;
2693 
2694 	/*
2695 	 * This code must be entered with Giant, and should never run if
2696 	 * we're not running with debug.mpsafenet.
2697 	 */
2698 	KASSERT(debug_mpsafenet != 0, ("if_start_deferred: debug.mpsafenet"));
2699 	GIANT_REQUIRED;
2700 
2701 	ifp = context;
2702 	(ifp->if_start)(ifp);
2703 }
2704 
2705 int
2706 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2707 {
2708 	int active = 0;
2709 
2710 	IF_LOCK(ifq);
2711 	if (_IF_QFULL(ifq)) {
2712 		_IF_DROP(ifq);
2713 		IF_UNLOCK(ifq);
2714 		m_freem(m);
2715 		return (0);
2716 	}
2717 	if (ifp != NULL) {
2718 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2719 		if (m->m_flags & (M_BCAST|M_MCAST))
2720 			ifp->if_omcasts++;
2721 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2722 	}
2723 	_IF_ENQUEUE(ifq, m);
2724 	IF_UNLOCK(ifq);
2725 	if (ifp != NULL && !active)
2726 		if_start(ifp);
2727 	return (1);
2728 }
2729 
2730 void
2731 if_register_com_alloc(u_char type,
2732     if_com_alloc_t *a, if_com_free_t *f)
2733 {
2734 
2735 	KASSERT(if_com_alloc[type] == NULL,
2736 	    ("if_register_com_alloc: %d already registered", type));
2737 	KASSERT(if_com_free[type] == NULL,
2738 	    ("if_register_com_alloc: %d free already registered", type));
2739 
2740 	if_com_alloc[type] = a;
2741 	if_com_free[type] = f;
2742 }
2743 
2744 void
2745 if_deregister_com_alloc(u_char type)
2746 {
2747 
2748 	KASSERT(if_com_alloc[type] != NULL,
2749 	    ("if_deregister_com_alloc: %d not registered", type));
2750 	KASSERT(if_com_free[type] != NULL,
2751 	    ("if_deregister_com_alloc: %d free not registered", type));
2752 	if_com_alloc[type] = NULL;
2753 	if_com_free[type] = NULL;
2754 }
2755