xref: /freebsd/sys/net/if.c (revision 262e143bd46171a6415a5b28af260a5efa2a3db8)
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/mac.h>
43 #include <sys/malloc.h>
44 #include <sys/sbuf.h>
45 #include <sys/bus.h>
46 #include <sys/mbuf.h>
47 #include <sys/systm.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 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
86 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
87 
88 /* Log link state change events */
89 static int log_link_state_change = 1;
90 
91 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
92 	&log_link_state_change, 0,
93 	"log interface link state change events");
94 
95 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
96 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
97 
98 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
99 
100 static void	if_attachdomain(void *);
101 static void	if_attachdomain1(struct ifnet *);
102 static int	ifconf(u_long, caddr_t);
103 static void	if_grow(void);
104 static void	if_init(void *);
105 static void	if_check(void *);
106 static void	if_qflush(struct ifaltq *);
107 static void	if_route(struct ifnet *, int flag, int fam);
108 static int	if_setflag(struct ifnet *, int, int, int *, int);
109 static void	if_slowtimo(void *);
110 static void	if_unroute(struct ifnet *, int flag, int fam);
111 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
112 static int	if_rtdel(struct radix_node *, void *);
113 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
114 static void	if_start_deferred(void *context, int pending);
115 static void	do_link_state_change(void *, int);
116 #ifdef INET6
117 /*
118  * XXX: declare here to avoid to include many inet6 related files..
119  * should be more generalized?
120  */
121 extern void	nd6_setmtu(struct ifnet *);
122 #endif
123 
124 int	if_index = 0;
125 struct	ifindex_entry *ifindex_table = NULL;
126 int	ifqmaxlen = IFQ_MAXLEN;
127 struct	ifnethead ifnet;	/* depend on static init XXX */
128 struct	mtx ifnet_lock;
129 static	if_com_alloc_t *if_com_alloc[256];
130 static	if_com_free_t *if_com_free[256];
131 
132 static int	if_indexlim = 8;
133 static struct	knlist ifklist;
134 
135 static void	filt_netdetach(struct knote *kn);
136 static int	filt_netdev(struct knote *kn, long hint);
137 
138 static struct filterops netdev_filtops =
139     { 1, NULL, filt_netdetach, filt_netdev };
140 
141 /*
142  * System initialization
143  */
144 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL)
145 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_check, NULL)
146 
147 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
148 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
149 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
150 
151 static d_open_t		netopen;
152 static d_close_t	netclose;
153 static d_ioctl_t	netioctl;
154 static d_kqfilter_t	netkqfilter;
155 
156 static struct cdevsw net_cdevsw = {
157 	.d_version =	D_VERSION,
158 	.d_flags =	D_NEEDGIANT,
159 	.d_open =	netopen,
160 	.d_close =	netclose,
161 	.d_ioctl =	netioctl,
162 	.d_name =	"net",
163 	.d_kqfilter =	netkqfilter,
164 };
165 
166 static int
167 netopen(struct cdev *dev, int flag, int mode, struct thread *td)
168 {
169 	return (0);
170 }
171 
172 static int
173 netclose(struct cdev *dev, int flags, int fmt, struct thread *td)
174 {
175 	return (0);
176 }
177 
178 static int
179 netioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
180 {
181 	struct ifnet *ifp;
182 	int error, idx;
183 
184 	/* only support interface specific ioctls */
185 	if (IOCGROUP(cmd) != 'i')
186 		return (EOPNOTSUPP);
187 	idx = minor(dev);
188 	if (idx == 0) {
189 		/*
190 		 * special network device, not interface.
191 		 */
192 		if (cmd == SIOCGIFCONF)
193 			return (ifconf(cmd, data));	/* XXX remove cmd */
194 		return (EOPNOTSUPP);
195 	}
196 
197 	ifp = ifnet_byindex(idx);
198 	if (ifp == NULL)
199 		return (ENXIO);
200 
201 	error = ifhwioctl(cmd, ifp, data, td);
202 	if (error == ENOIOCTL)
203 		error = EOPNOTSUPP;
204 	return (error);
205 }
206 
207 static int
208 netkqfilter(struct cdev *dev, struct knote *kn)
209 {
210 	struct knlist *klist;
211 	struct ifnet *ifp;
212 	int idx;
213 
214 	switch (kn->kn_filter) {
215 	case EVFILT_NETDEV:
216 		kn->kn_fop = &netdev_filtops;
217 		break;
218 	default:
219 		return (EINVAL);
220 	}
221 
222 	idx = minor(dev);
223 	if (idx == 0) {
224 		klist = &ifklist;
225 	} else {
226 		ifp = ifnet_byindex(idx);
227 		if (ifp == NULL)
228 			return (1);
229 		klist = &ifp->if_klist;
230 	}
231 
232 	kn->kn_hook = (caddr_t)klist;
233 
234 	knlist_add(klist, kn, 0);
235 
236 	return (0);
237 }
238 
239 static void
240 filt_netdetach(struct knote *kn)
241 {
242 	struct knlist *klist = (struct knlist *)kn->kn_hook;
243 
244 	knlist_remove(klist, kn, 0);
245 }
246 
247 static int
248 filt_netdev(struct knote *kn, long hint)
249 {
250 	struct knlist *klist = (struct knlist *)kn->kn_hook;
251 
252 	/*
253 	 * Currently NOTE_EXIT is abused to indicate device detach.
254 	 */
255 	if (hint == NOTE_EXIT) {
256 		kn->kn_data = NOTE_LINKINV;
257 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
258 		knlist_remove_inevent(klist, kn);
259 		return (1);
260 	}
261 	if (hint != 0)
262 		kn->kn_data = hint;			/* current status */
263 	if (kn->kn_sfflags & hint)
264 		kn->kn_fflags |= hint;
265 	return (kn->kn_fflags != 0);
266 }
267 
268 /*
269  * Network interface utility routines.
270  *
271  * Routines with ifa_ifwith* names take sockaddr *'s as
272  * parameters.
273  */
274 /* ARGSUSED*/
275 static void
276 if_init(void *dummy __unused)
277 {
278 
279 	IFNET_LOCK_INIT();
280 	TAILQ_INIT(&ifnet);
281 	knlist_init(&ifklist, NULL, NULL, NULL, NULL);
282 	if_grow();				/* create initial table */
283 	ifdev_byindex(0) = make_dev(&net_cdevsw, 0,
284 	    UID_ROOT, GID_WHEEL, 0600, "network");
285 	if_clone_init();
286 }
287 
288 static void
289 if_grow(void)
290 {
291 	u_int n;
292 	struct ifindex_entry *e;
293 
294 	if_indexlim <<= 1;
295 	n = if_indexlim * sizeof(*e);
296 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
297 	if (ifindex_table != NULL) {
298 		memcpy((caddr_t)e, (caddr_t)ifindex_table, n/2);
299 		free((caddr_t)ifindex_table, M_IFNET);
300 	}
301 	ifindex_table = e;
302 }
303 
304 /* ARGSUSED*/
305 static void
306 if_check(void *dummy __unused)
307 {
308 	struct ifnet *ifp;
309 	int s;
310 
311 	s = splimp();
312 	IFNET_RLOCK();	/* could sleep on rare error; mostly okay XXX */
313 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
314 		if (ifp->if_snd.ifq_maxlen == 0) {
315 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
316 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
317 		}
318 		if (!mtx_initialized(&ifp->if_snd.ifq_mtx)) {
319 			if_printf(ifp,
320 			    "XXX: driver didn't initialize queue mtx\n");
321 			mtx_init(&ifp->if_snd.ifq_mtx, "unknown",
322 			    MTX_NETWORK_LOCK, MTX_DEF);
323 		}
324 	}
325 	IFNET_RUNLOCK();
326 	splx(s);
327 	if_slowtimo(0);
328 }
329 
330 /*
331  * Allocate a struct ifnet and in index for an interface.
332  */
333 struct ifnet*
334 if_alloc(u_char type)
335 {
336 	struct ifnet *ifp;
337 
338 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
339 
340 	/*
341 	 * Try to find an empty slot below if_index.  If we fail, take
342 	 * the next slot.
343 	 *
344 	 * XXX: should be locked!
345 	 */
346 	for (ifp->if_index = 1; ifp->if_index <= if_index; ifp->if_index++) {
347 		if (ifnet_byindex(ifp->if_index) == NULL)
348 			break;
349 	}
350 	/* Catch if_index overflow. */
351 	if (ifp->if_index < 1) {
352 		free(ifp, M_IFNET);
353 		return (NULL);
354 	}
355 	if (ifp->if_index > if_index)
356 		if_index = ifp->if_index;
357 	if (if_index >= if_indexlim)
358 		if_grow();
359 	ifnet_byindex(ifp->if_index) = ifp;
360 
361 	ifp->if_type = type;
362 
363 	if (if_com_alloc[type] != NULL) {
364 		ifp->if_l2com = if_com_alloc[type](type, ifp);
365 		if (ifp->if_l2com == NULL) {
366 			free(ifp, M_IFNET);
367 			return (NULL);
368 		}
369 	}
370 	IF_ADDR_LOCK_INIT(ifp);
371 
372 	return (ifp);
373 }
374 
375 void
376 if_free(struct ifnet *ifp)
377 {
378 
379 	/* Do not add code to this function!  Add it to if_free_type(). */
380 	if_free_type(ifp, ifp->if_type);
381 }
382 
383 void
384 if_free_type(struct ifnet *ifp, u_char type)
385 {
386 
387 	if (ifp != ifnet_byindex(ifp->if_index)) {
388 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
389 		    __func__);
390 		return;
391 	}
392 
393 	IF_ADDR_LOCK_DESTROY(ifp);
394 
395 	ifnet_byindex(ifp->if_index) = NULL;
396 
397 	/* XXX: should be locked with if_findindex() */
398 	while (if_index > 0 && ifnet_byindex(if_index) == NULL)
399 		if_index--;
400 
401 	if (if_com_free[type] != NULL)
402 		if_com_free[type](ifp->if_l2com, type);
403 
404 	free(ifp, M_IFNET);
405 };
406 
407 /*
408  * Attach an interface to the
409  * list of "active" interfaces.
410  */
411 void
412 if_attach(struct ifnet *ifp)
413 {
414 	unsigned socksize, ifasize;
415 	int namelen, masklen;
416 	struct sockaddr_dl *sdl;
417 	struct ifaddr *ifa;
418 
419 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
420 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
421 		    ifp->if_xname);
422 
423 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
424 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
425 	IF_AFDATA_LOCK_INIT(ifp);
426 	ifp->if_afdata_initialized = 0;
427 	IFNET_WLOCK();
428 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
429 	IFNET_WUNLOCK();
430 	/*
431 	 * XXX -
432 	 * The old code would work if the interface passed a pre-existing
433 	 * chain of ifaddrs to this code.  We don't trust our callers to
434 	 * properly initialize the tailq, however, so we no longer allow
435 	 * this unlikely case.
436 	 */
437 	TAILQ_INIT(&ifp->if_addrhead);
438 	TAILQ_INIT(&ifp->if_prefixhead);
439 	TAILQ_INIT(&ifp->if_multiaddrs);
440 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
441 	getmicrotime(&ifp->if_lastchange);
442 	ifp->if_data.ifi_epoch = time_uptime;
443 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
444 
445 #ifdef MAC
446 	mac_init_ifnet(ifp);
447 	mac_create_ifnet(ifp);
448 #endif
449 
450 	ifdev_byindex(ifp->if_index) = make_dev(&net_cdevsw,
451 	    unit2minor(ifp->if_index),
452 	    UID_ROOT, GID_WHEEL, 0600, "%s/%s",
453 	    net_cdevsw.d_name, ifp->if_xname);
454 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
455 	    net_cdevsw.d_name, ifp->if_index);
456 
457 	mtx_init(&ifp->if_snd.ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
458 
459 	/*
460 	 * create a Link Level name for this device
461 	 */
462 	namelen = strlen(ifp->if_xname);
463 	/*
464 	 * Always save enough space for any possiable name so we can do
465 	 * a rename in place later.
466 	 */
467 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
468 	socksize = masklen + ifp->if_addrlen;
469 	if (socksize < sizeof(*sdl))
470 		socksize = sizeof(*sdl);
471 	socksize = roundup2(socksize, sizeof(long));
472 	ifasize = sizeof(*ifa) + 2 * socksize;
473 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
474 	IFA_LOCK_INIT(ifa);
475 	sdl = (struct sockaddr_dl *)(ifa + 1);
476 	sdl->sdl_len = socksize;
477 	sdl->sdl_family = AF_LINK;
478 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
479 	sdl->sdl_nlen = namelen;
480 	sdl->sdl_index = ifp->if_index;
481 	sdl->sdl_type = ifp->if_type;
482 	ifp->if_addr = ifa;
483 	ifa->ifa_ifp = ifp;
484 	ifa->ifa_rtrequest = link_rtrequest;
485 	ifa->ifa_addr = (struct sockaddr *)sdl;
486 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
487 	ifa->ifa_netmask = (struct sockaddr *)sdl;
488 	sdl->sdl_len = masklen;
489 	while (namelen != 0)
490 		sdl->sdl_data[--namelen] = 0xff;
491 	ifa->ifa_refcnt = 1;
492 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
493 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
494 	ifp->if_snd.altq_type = 0;
495 	ifp->if_snd.altq_disc = NULL;
496 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
497 	ifp->if_snd.altq_tbr  = NULL;
498 	ifp->if_snd.altq_ifp  = ifp;
499 
500 	if (domain_init_status >= 2)
501 		if_attachdomain1(ifp);
502 
503 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
504 
505 	/* Announce the interface. */
506 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
507 }
508 
509 static void
510 if_attachdomain(void *dummy)
511 {
512 	struct ifnet *ifp;
513 	int s;
514 
515 	s = splnet();
516 	TAILQ_FOREACH(ifp, &ifnet, if_link)
517 		if_attachdomain1(ifp);
518 	splx(s);
519 }
520 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
521     if_attachdomain, NULL);
522 
523 static void
524 if_attachdomain1(struct ifnet *ifp)
525 {
526 	struct domain *dp;
527 	int s;
528 
529 	s = splnet();
530 
531 	/*
532 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
533 	 * cannot lock ifp->if_afdata initialization, entirely.
534 	 */
535 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
536 		splx(s);
537 		return;
538 	}
539 	if (ifp->if_afdata_initialized >= domain_init_status) {
540 		IF_AFDATA_UNLOCK(ifp);
541 		splx(s);
542 		printf("if_attachdomain called more than once on %s\n",
543 		    ifp->if_xname);
544 		return;
545 	}
546 	ifp->if_afdata_initialized = domain_init_status;
547 	IF_AFDATA_UNLOCK(ifp);
548 
549 	/* address family dependent data region */
550 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
551 	for (dp = domains; dp; dp = dp->dom_next) {
552 		if (dp->dom_ifattach)
553 			ifp->if_afdata[dp->dom_family] =
554 			    (*dp->dom_ifattach)(ifp);
555 	}
556 
557 	splx(s);
558 }
559 
560 /*
561  * Remove any network addresses from an interface.
562  */
563 
564 void
565 if_purgeaddrs(struct ifnet *ifp)
566 {
567 	struct ifaddr *ifa, *next;
568 
569 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
570 
571 		if (ifa->ifa_addr->sa_family == AF_LINK)
572 			continue;
573 #ifdef INET
574 		/* XXX: Ugly!! ad hoc just for INET */
575 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
576 			struct ifaliasreq ifr;
577 
578 			bzero(&ifr, sizeof(ifr));
579 			ifr.ifra_addr = *ifa->ifa_addr;
580 			if (ifa->ifa_dstaddr)
581 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
582 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
583 			    NULL) == 0)
584 				continue;
585 		}
586 #endif /* INET */
587 #ifdef INET6
588 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
589 			in6_purgeaddr(ifa);
590 			/* ifp_addrhead is already updated */
591 			continue;
592 		}
593 #endif /* INET6 */
594 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
595 		IFAFREE(ifa);
596 	}
597 }
598 
599 /*
600  * Detach an interface, removing it from the
601  * list of "active" interfaces.
602  *
603  * XXXRW: There are some significant questions about event ordering, and
604  * how to prevent things from starting to use the interface during detach.
605  */
606 void
607 if_detach(struct ifnet *ifp)
608 {
609 	struct ifaddr *ifa;
610 	struct radix_node_head	*rnh;
611 	int s;
612 	int i;
613 	struct domain *dp;
614  	struct ifnet *iter;
615  	int found;
616 
617 	/*
618 	 * Remove/wait for pending events.
619 	 */
620 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
621 
622 #ifdef DEV_CARP
623 	/* Maybe hook to the generalized departure handler above?!? */
624 	if (ifp->if_carp)
625 		carp_ifdetach(ifp);
626 #endif
627 
628 	/*
629 	 * Remove routes and flush queues.
630 	 */
631 	s = splnet();
632 	if_down(ifp);
633 #ifdef ALTQ
634 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
635 		altq_disable(&ifp->if_snd);
636 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
637 		altq_detach(&ifp->if_snd);
638 #endif
639 
640 	if_purgeaddrs(ifp);
641 
642 #ifdef INET
643 	in_ifdetach(ifp);
644 #endif
645 
646 #ifdef INET6
647 	/*
648 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
649 	 * before removing routing entries below, since IPv6 interface direct
650 	 * routes are expected to be removed by the IPv6-specific kernel API.
651 	 * Otherwise, the kernel will detect some inconsistency and bark it.
652 	 */
653 	in6_ifdetach(ifp);
654 #endif
655 	/*
656 	 * Remove link ifaddr pointer and maybe decrement if_index.
657 	 * Clean up all addresses.
658 	 */
659 	ifp->if_addr = NULL;
660 	destroy_dev(ifdev_byindex(ifp->if_index));
661 	ifdev_byindex(ifp->if_index) = NULL;
662 
663 	/* We can now free link ifaddr. */
664 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
665 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
666 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
667 		IFAFREE(ifa);
668 	}
669 
670 	/*
671 	 * Delete all remaining routes using this interface
672 	 * Unfortuneatly the only way to do this is to slog through
673 	 * the entire routing table looking for routes which point
674 	 * to this interface...oh well...
675 	 */
676 	for (i = 1; i <= AF_MAX; i++) {
677 		if ((rnh = rt_tables[i]) == NULL)
678 			continue;
679 		RADIX_NODE_HEAD_LOCK(rnh);
680 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
681 		RADIX_NODE_HEAD_UNLOCK(rnh);
682 	}
683 
684 	/* Announce that the interface is gone. */
685 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
686 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
687 
688 	IF_AFDATA_LOCK(ifp);
689 	for (dp = domains; dp; dp = dp->dom_next) {
690 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
691 			(*dp->dom_ifdetach)(ifp,
692 			    ifp->if_afdata[dp->dom_family]);
693 	}
694 	IF_AFDATA_UNLOCK(ifp);
695 
696 #ifdef MAC
697 	mac_destroy_ifnet(ifp);
698 #endif /* MAC */
699 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
700 	knlist_clear(&ifp->if_klist, 0);
701 	knlist_destroy(&ifp->if_klist);
702 	IFNET_WLOCK();
703  	found = 0;
704  	TAILQ_FOREACH(iter, &ifnet, if_link)
705  		if (iter == ifp) {
706  			found = 1;
707  			break;
708  		}
709  	if (found)
710  		TAILQ_REMOVE(&ifnet, ifp, if_link);
711 	IFNET_WUNLOCK();
712 	mtx_destroy(&ifp->if_snd.ifq_mtx);
713 	IF_AFDATA_DESTROY(ifp);
714 	splx(s);
715 }
716 
717 /*
718  * Delete Routes for a Network Interface
719  *
720  * Called for each routing entry via the rnh->rnh_walktree() call above
721  * to delete all route entries referencing a detaching network interface.
722  *
723  * Arguments:
724  *	rn	pointer to node in the routing table
725  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
726  *
727  * Returns:
728  *	0	successful
729  *	errno	failed - reason indicated
730  *
731  */
732 static int
733 if_rtdel(struct radix_node *rn, void *arg)
734 {
735 	struct rtentry	*rt = (struct rtentry *)rn;
736 	struct ifnet	*ifp = arg;
737 	int		err;
738 
739 	if (rt->rt_ifp == ifp) {
740 
741 		/*
742 		 * Protect (sorta) against walktree recursion problems
743 		 * with cloned routes
744 		 */
745 		if ((rt->rt_flags & RTF_UP) == 0)
746 			return (0);
747 
748 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
749 				rt_mask(rt), rt->rt_flags,
750 				(struct rtentry **) NULL);
751 		if (err) {
752 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
753 		}
754 	}
755 
756 	return (0);
757 }
758 
759 #define	sa_equal(a1, a2)	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
760 
761 /*
762  * Locate an interface based on a complete address.
763  */
764 /*ARGSUSED*/
765 struct ifaddr *
766 ifa_ifwithaddr(struct sockaddr *addr)
767 {
768 	struct ifnet *ifp;
769 	struct ifaddr *ifa;
770 
771 	IFNET_RLOCK();
772 	TAILQ_FOREACH(ifp, &ifnet, if_link)
773 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
774 			if (ifa->ifa_addr->sa_family != addr->sa_family)
775 				continue;
776 			if (sa_equal(addr, ifa->ifa_addr))
777 				goto done;
778 			/* IP6 doesn't have broadcast */
779 			if ((ifp->if_flags & IFF_BROADCAST) &&
780 			    ifa->ifa_broadaddr &&
781 			    ifa->ifa_broadaddr->sa_len != 0 &&
782 			    sa_equal(ifa->ifa_broadaddr, addr))
783 				goto done;
784 		}
785 	ifa = NULL;
786 done:
787 	IFNET_RUNLOCK();
788 	return (ifa);
789 }
790 
791 /*
792  * Locate the point to point interface with a given destination address.
793  */
794 /*ARGSUSED*/
795 struct ifaddr *
796 ifa_ifwithdstaddr(struct sockaddr *addr)
797 {
798 	struct ifnet *ifp;
799 	struct ifaddr *ifa;
800 
801 	IFNET_RLOCK();
802 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
803 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
804 			continue;
805 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
806 			if (ifa->ifa_addr->sa_family != addr->sa_family)
807 				continue;
808 			if (ifa->ifa_dstaddr &&
809 			    sa_equal(addr, ifa->ifa_dstaddr))
810 				goto done;
811 		}
812 	}
813 	ifa = NULL;
814 done:
815 	IFNET_RUNLOCK();
816 	return (ifa);
817 }
818 
819 /*
820  * Find an interface on a specific network.  If many, choice
821  * is most specific found.
822  */
823 struct ifaddr *
824 ifa_ifwithnet(struct sockaddr *addr)
825 {
826 	struct ifnet *ifp;
827 	struct ifaddr *ifa;
828 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
829 	u_int af = addr->sa_family;
830 	char *addr_data = addr->sa_data, *cplim;
831 
832 	/*
833 	 * AF_LINK addresses can be looked up directly by their index number,
834 	 * so do that if we can.
835 	 */
836 	if (af == AF_LINK) {
837 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
838 	    if (sdl->sdl_index && sdl->sdl_index <= if_index)
839 		return (ifaddr_byindex(sdl->sdl_index));
840 	}
841 
842 	/*
843 	 * Scan though each interface, looking for ones that have
844 	 * addresses in this address family.
845 	 */
846 	IFNET_RLOCK();
847 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
848 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
849 			char *cp, *cp2, *cp3;
850 
851 			if (ifa->ifa_addr->sa_family != af)
852 next:				continue;
853 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
854 				/*
855 				 * This is a bit broken as it doesn't
856 				 * take into account that the remote end may
857 				 * be a single node in the network we are
858 				 * looking for.
859 				 * The trouble is that we don't know the
860 				 * netmask for the remote end.
861 				 */
862 				if (ifa->ifa_dstaddr != 0 &&
863 				    sa_equal(addr, ifa->ifa_dstaddr))
864 					goto done;
865 			} else {
866 				/*
867 				 * if we have a special address handler,
868 				 * then use it instead of the generic one.
869 				 */
870 				if (ifa->ifa_claim_addr) {
871 					if ((*ifa->ifa_claim_addr)(ifa, addr))
872 						goto done;
873 					continue;
874 				}
875 
876 				/*
877 				 * Scan all the bits in the ifa's address.
878 				 * If a bit dissagrees with what we are
879 				 * looking for, mask it with the netmask
880 				 * to see if it really matters.
881 				 * (A byte at a time)
882 				 */
883 				if (ifa->ifa_netmask == 0)
884 					continue;
885 				cp = addr_data;
886 				cp2 = ifa->ifa_addr->sa_data;
887 				cp3 = ifa->ifa_netmask->sa_data;
888 				cplim = ifa->ifa_netmask->sa_len
889 					+ (char *)ifa->ifa_netmask;
890 				while (cp3 < cplim)
891 					if ((*cp++ ^ *cp2++) & *cp3++)
892 						goto next; /* next address! */
893 				/*
894 				 * If the netmask of what we just found
895 				 * is more specific than what we had before
896 				 * (if we had one) then remember the new one
897 				 * before continuing to search
898 				 * for an even better one.
899 				 */
900 				if (ifa_maybe == 0 ||
901 				    rn_refines((caddr_t)ifa->ifa_netmask,
902 				    (caddr_t)ifa_maybe->ifa_netmask))
903 					ifa_maybe = ifa;
904 			}
905 		}
906 	}
907 	ifa = ifa_maybe;
908 done:
909 	IFNET_RUNLOCK();
910 	return (ifa);
911 }
912 
913 /*
914  * Find an interface address specific to an interface best matching
915  * a given address.
916  */
917 struct ifaddr *
918 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
919 {
920 	struct ifaddr *ifa;
921 	char *cp, *cp2, *cp3;
922 	char *cplim;
923 	struct ifaddr *ifa_maybe = 0;
924 	u_int af = addr->sa_family;
925 
926 	if (af >= AF_MAX)
927 		return (0);
928 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
929 		if (ifa->ifa_addr->sa_family != af)
930 			continue;
931 		if (ifa_maybe == 0)
932 			ifa_maybe = ifa;
933 		if (ifa->ifa_netmask == 0) {
934 			if (sa_equal(addr, ifa->ifa_addr) ||
935 			    (ifa->ifa_dstaddr &&
936 			    sa_equal(addr, ifa->ifa_dstaddr)))
937 				goto done;
938 			continue;
939 		}
940 		if (ifp->if_flags & IFF_POINTOPOINT) {
941 			if (sa_equal(addr, ifa->ifa_dstaddr))
942 				goto done;
943 		} else {
944 			cp = addr->sa_data;
945 			cp2 = ifa->ifa_addr->sa_data;
946 			cp3 = ifa->ifa_netmask->sa_data;
947 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
948 			for (; cp3 < cplim; cp3++)
949 				if ((*cp++ ^ *cp2++) & *cp3)
950 					break;
951 			if (cp3 == cplim)
952 				goto done;
953 		}
954 	}
955 	ifa = ifa_maybe;
956 done:
957 	return (ifa);
958 }
959 
960 #include <net/route.h>
961 
962 /*
963  * Default action when installing a route with a Link Level gateway.
964  * Lookup an appropriate real ifa to point to.
965  * This should be moved to /sys/net/link.c eventually.
966  */
967 static void
968 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
969 {
970 	struct ifaddr *ifa, *oifa;
971 	struct sockaddr *dst;
972 	struct ifnet *ifp;
973 
974 	RT_LOCK_ASSERT(rt);
975 
976 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
977 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
978 		return;
979 	ifa = ifaof_ifpforaddr(dst, ifp);
980 	if (ifa) {
981 		IFAREF(ifa);		/* XXX */
982 		oifa = rt->rt_ifa;
983 		rt->rt_ifa = ifa;
984 		IFAFREE(oifa);
985 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
986 			ifa->ifa_rtrequest(cmd, rt, info);
987 	}
988 }
989 
990 /*
991  * Mark an interface down and notify protocols of
992  * the transition.
993  * NOTE: must be called at splnet or eqivalent.
994  */
995 static void
996 if_unroute(struct ifnet *ifp, int flag, int fam)
997 {
998 	struct ifaddr *ifa;
999 
1000 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1001 
1002 	ifp->if_flags &= ~flag;
1003 	getmicrotime(&ifp->if_lastchange);
1004 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1005 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1006 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1007 	if_qflush(&ifp->if_snd);
1008 #ifdef DEV_CARP
1009 	if (ifp->if_carp)
1010 		carp_carpdev_state(ifp->if_carp);
1011 #endif
1012 	rt_ifmsg(ifp);
1013 }
1014 
1015 /*
1016  * Mark an interface up and notify protocols of
1017  * the transition.
1018  * NOTE: must be called at splnet or eqivalent.
1019  */
1020 static void
1021 if_route(struct ifnet *ifp, int flag, int fam)
1022 {
1023 	struct ifaddr *ifa;
1024 
1025 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1026 
1027 	ifp->if_flags |= flag;
1028 	getmicrotime(&ifp->if_lastchange);
1029 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1030 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1031 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1032 #ifdef DEV_CARP
1033 	if (ifp->if_carp)
1034 		carp_carpdev_state(ifp->if_carp);
1035 #endif
1036 	rt_ifmsg(ifp);
1037 #ifdef INET6
1038 	in6_if_up(ifp);
1039 #endif
1040 }
1041 
1042 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1043 
1044 /*
1045  * Handle a change in the interface link state. To avoid LORs
1046  * between driver lock and upper layer locks, as well as possible
1047  * recursions, we post event to taskqueue, and all job
1048  * is done in static do_link_state_change().
1049  */
1050 void
1051 if_link_state_change(struct ifnet *ifp, int link_state)
1052 {
1053 	/* Return if state hasn't changed. */
1054 	if (ifp->if_link_state == link_state)
1055 		return;
1056 
1057 	ifp->if_link_state = link_state;
1058 
1059 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1060 }
1061 
1062 static void
1063 do_link_state_change(void *arg, int pending)
1064 {
1065 	struct ifnet *ifp = (struct ifnet *)arg;
1066 	int link_state = ifp->if_link_state;
1067 	int link;
1068 
1069 	/* Notify that the link state has changed. */
1070 	rt_ifmsg(ifp);
1071 	if (link_state == LINK_STATE_UP)
1072 		link = NOTE_LINKUP;
1073 	else if (link_state == LINK_STATE_DOWN)
1074 		link = NOTE_LINKDOWN;
1075 	else
1076 		link = NOTE_LINKINV;
1077 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1078 	if (ifp->if_nvlans != 0)
1079 		(*vlan_link_state_p)(ifp, link);
1080 
1081 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1082 	    IFP2AC(ifp)->ac_netgraph != NULL)
1083 		(*ng_ether_link_state_p)(ifp, link_state);
1084 #ifdef DEV_CARP
1085 	if (ifp->if_carp)
1086 		carp_carpdev_state(ifp->if_carp);
1087 #endif
1088 	if (ifp->if_bridge) {
1089 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1090 		(*bstp_linkstate_p)(ifp, link_state);
1091 	}
1092 
1093 	devctl_notify("IFNET", ifp->if_xname,
1094 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1095 	if (pending > 1)
1096 		if_printf(ifp, "%d link states coalesced\n", pending);
1097 	if (log_link_state_change)
1098 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1099 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1100 }
1101 
1102 /*
1103  * Mark an interface down and notify protocols of
1104  * the transition.
1105  * NOTE: must be called at splnet or eqivalent.
1106  */
1107 void
1108 if_down(struct ifnet *ifp)
1109 {
1110 
1111 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1112 }
1113 
1114 /*
1115  * Mark an interface up and notify protocols of
1116  * the transition.
1117  * NOTE: must be called at splnet or eqivalent.
1118  */
1119 void
1120 if_up(struct ifnet *ifp)
1121 {
1122 
1123 	if_route(ifp, IFF_UP, AF_UNSPEC);
1124 }
1125 
1126 /*
1127  * Flush an interface queue.
1128  */
1129 static void
1130 if_qflush(struct ifaltq *ifq)
1131 {
1132 	struct mbuf *m, *n;
1133 
1134 	IFQ_LOCK(ifq);
1135 #ifdef ALTQ
1136 	if (ALTQ_IS_ENABLED(ifq))
1137 		ALTQ_PURGE(ifq);
1138 #endif
1139 	n = ifq->ifq_head;
1140 	while ((m = n) != 0) {
1141 		n = m->m_act;
1142 		m_freem(m);
1143 	}
1144 	ifq->ifq_head = 0;
1145 	ifq->ifq_tail = 0;
1146 	ifq->ifq_len = 0;
1147 	IFQ_UNLOCK(ifq);
1148 }
1149 
1150 /*
1151  * Handle interface watchdog timer routines.  Called
1152  * from softclock, we decrement timers (if set) and
1153  * call the appropriate interface routine on expiration.
1154  *
1155  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1156  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1157  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1158  */
1159 static void
1160 if_slowtimo(void *arg)
1161 {
1162 	struct ifnet *ifp;
1163 	int s = splimp();
1164 
1165 	IFNET_RLOCK();
1166 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1167 		if (ifp->if_timer == 0 || --ifp->if_timer)
1168 			continue;
1169 		if (ifp->if_watchdog)
1170 			(*ifp->if_watchdog)(ifp);
1171 	}
1172 	IFNET_RUNLOCK();
1173 	splx(s);
1174 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1175 }
1176 
1177 /*
1178  * Map interface name to
1179  * interface structure pointer.
1180  */
1181 struct ifnet *
1182 ifunit(const char *name)
1183 {
1184 	struct ifnet *ifp;
1185 
1186 	IFNET_RLOCK();
1187 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1188 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1189 			break;
1190 	}
1191 	IFNET_RUNLOCK();
1192 	return (ifp);
1193 }
1194 
1195 /*
1196  * Hardware specific interface ioctls.
1197  */
1198 static int
1199 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1200 {
1201 	struct ifreq *ifr;
1202 	struct ifstat *ifs;
1203 	int error = 0;
1204 	int new_flags, temp_flags;
1205 	size_t namelen, onamelen;
1206 	char new_name[IFNAMSIZ];
1207 	struct ifaddr *ifa;
1208 	struct sockaddr_dl *sdl;
1209 
1210 	ifr = (struct ifreq *)data;
1211 	switch (cmd) {
1212 	case SIOCGIFINDEX:
1213 		ifr->ifr_index = ifp->if_index;
1214 		break;
1215 
1216 	case SIOCGIFFLAGS:
1217 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1218 		ifr->ifr_flags = temp_flags & 0xffff;
1219 		ifr->ifr_flagshigh = temp_flags >> 16;
1220 		break;
1221 
1222 	case SIOCGIFCAP:
1223 		ifr->ifr_reqcap = ifp->if_capabilities;
1224 		ifr->ifr_curcap = ifp->if_capenable;
1225 		break;
1226 
1227 #ifdef MAC
1228 	case SIOCGIFMAC:
1229 		error = mac_ioctl_ifnet_get(td->td_ucred, ifr, ifp);
1230 		break;
1231 #endif
1232 
1233 	case SIOCGIFMETRIC:
1234 		ifr->ifr_metric = ifp->if_metric;
1235 		break;
1236 
1237 	case SIOCGIFMTU:
1238 		ifr->ifr_mtu = ifp->if_mtu;
1239 		break;
1240 
1241 	case SIOCGIFPHYS:
1242 		ifr->ifr_phys = ifp->if_physical;
1243 		break;
1244 
1245 	case SIOCSIFFLAGS:
1246 		error = suser(td);
1247 		if (error)
1248 			return (error);
1249 		/*
1250 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1251 		 * check, so we don't need special handling here yet.
1252 		 */
1253 		new_flags = (ifr->ifr_flags & 0xffff) |
1254 		    (ifr->ifr_flagshigh << 16);
1255 		if (ifp->if_flags & IFF_SMART) {
1256 			/* Smart drivers twiddle their own routes */
1257 		} else if (ifp->if_flags & IFF_UP &&
1258 		    (new_flags & IFF_UP) == 0) {
1259 			int s = splimp();
1260 			if_down(ifp);
1261 			splx(s);
1262 		} else if (new_flags & IFF_UP &&
1263 		    (ifp->if_flags & IFF_UP) == 0) {
1264 			int s = splimp();
1265 			if_up(ifp);
1266 			splx(s);
1267 		}
1268 		/* See if permanently promiscuous mode bit is about to flip */
1269 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1270 			if (new_flags & IFF_PPROMISC)
1271 				ifp->if_flags |= IFF_PROMISC;
1272 			else if (ifp->if_pcount == 0)
1273 				ifp->if_flags &= ~IFF_PROMISC;
1274 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1275 			    ifp->if_xname,
1276 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1277 		}
1278 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1279 			(new_flags &~ IFF_CANTCHANGE);
1280 		if (ifp->if_ioctl) {
1281 			IFF_LOCKGIANT(ifp);
1282 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1283 			IFF_UNLOCKGIANT(ifp);
1284 		}
1285 		getmicrotime(&ifp->if_lastchange);
1286 		break;
1287 
1288 	case SIOCSIFCAP:
1289 		error = suser(td);
1290 		if (error)
1291 			return (error);
1292 		if (ifp->if_ioctl == NULL)
1293 			return (EOPNOTSUPP);
1294 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1295 			return (EINVAL);
1296 		IFF_LOCKGIANT(ifp);
1297 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1298 		IFF_UNLOCKGIANT(ifp);
1299 		if (error == 0)
1300 			getmicrotime(&ifp->if_lastchange);
1301 		break;
1302 
1303 #ifdef MAC
1304 	case SIOCSIFMAC:
1305 		error = mac_ioctl_ifnet_set(td->td_ucred, ifr, ifp);
1306 		break;
1307 #endif
1308 
1309 	case SIOCSIFNAME:
1310 		error = suser(td);
1311 		if (error != 0)
1312 			return (error);
1313 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1314 		if (error != 0)
1315 			return (error);
1316 		if (new_name[0] == '\0')
1317 			return (EINVAL);
1318 		if (ifunit(new_name) != NULL)
1319 			return (EEXIST);
1320 
1321 		/* Announce the departure of the interface. */
1322 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1323 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1324 
1325 		log(LOG_INFO, "%s: changing name to '%s'\n",
1326 		    ifp->if_xname, new_name);
1327 
1328 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1329 		ifa = ifp->if_addr;
1330 		IFA_LOCK(ifa);
1331 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1332 		namelen = strlen(new_name);
1333 		onamelen = sdl->sdl_nlen;
1334 		/*
1335 		 * Move the address if needed.  This is safe because we
1336 		 * allocate space for a name of length IFNAMSIZ when we
1337 		 * create this in if_attach().
1338 		 */
1339 		if (namelen != onamelen) {
1340 			bcopy(sdl->sdl_data + onamelen,
1341 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1342 		}
1343 		bcopy(new_name, sdl->sdl_data, namelen);
1344 		sdl->sdl_nlen = namelen;
1345 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1346 		bzero(sdl->sdl_data, onamelen);
1347 		while (namelen != 0)
1348 			sdl->sdl_data[--namelen] = 0xff;
1349 		IFA_UNLOCK(ifa);
1350 
1351 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1352 		/* Announce the return of the interface. */
1353 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1354 		break;
1355 
1356 	case SIOCSIFMETRIC:
1357 		error = suser(td);
1358 		if (error)
1359 			return (error);
1360 		ifp->if_metric = ifr->ifr_metric;
1361 		getmicrotime(&ifp->if_lastchange);
1362 		break;
1363 
1364 	case SIOCSIFPHYS:
1365 		error = suser(td);
1366 		if (error)
1367 			return (error);
1368 		if (ifp->if_ioctl == NULL)
1369 			return (EOPNOTSUPP);
1370 		IFF_LOCKGIANT(ifp);
1371 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1372 		IFF_UNLOCKGIANT(ifp);
1373 		if (error == 0)
1374 			getmicrotime(&ifp->if_lastchange);
1375 		break;
1376 
1377 	case SIOCSIFMTU:
1378 	{
1379 		u_long oldmtu = ifp->if_mtu;
1380 
1381 		error = suser(td);
1382 		if (error)
1383 			return (error);
1384 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1385 			return (EINVAL);
1386 		if (ifp->if_ioctl == NULL)
1387 			return (EOPNOTSUPP);
1388 		IFF_LOCKGIANT(ifp);
1389 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1390 		IFF_UNLOCKGIANT(ifp);
1391 		if (error == 0) {
1392 			getmicrotime(&ifp->if_lastchange);
1393 			rt_ifmsg(ifp);
1394 		}
1395 		/*
1396 		 * If the link MTU changed, do network layer specific procedure.
1397 		 */
1398 		if (ifp->if_mtu != oldmtu) {
1399 #ifdef INET6
1400 			nd6_setmtu(ifp);
1401 #endif
1402 		}
1403 		break;
1404 	}
1405 
1406 	case SIOCADDMULTI:
1407 	case SIOCDELMULTI:
1408 		error = suser(td);
1409 		if (error)
1410 			return (error);
1411 
1412 		/* Don't allow group membership on non-multicast interfaces. */
1413 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1414 			return (EOPNOTSUPP);
1415 
1416 		/* Don't let users screw up protocols' entries. */
1417 		if (ifr->ifr_addr.sa_family != AF_LINK)
1418 			return (EINVAL);
1419 
1420 		if (cmd == SIOCADDMULTI) {
1421 			struct ifmultiaddr *ifma;
1422 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1423 		} else {
1424 			error = if_delmulti(ifp, &ifr->ifr_addr);
1425 		}
1426 		if (error == 0)
1427 			getmicrotime(&ifp->if_lastchange);
1428 		break;
1429 
1430 	case SIOCSIFPHYADDR:
1431 	case SIOCDIFPHYADDR:
1432 #ifdef INET6
1433 	case SIOCSIFPHYADDR_IN6:
1434 #endif
1435 	case SIOCSLIFPHYADDR:
1436 	case SIOCSIFMEDIA:
1437 	case SIOCSIFGENERIC:
1438 		error = suser(td);
1439 		if (error)
1440 			return (error);
1441 		if (ifp->if_ioctl == NULL)
1442 			return (EOPNOTSUPP);
1443 		IFF_LOCKGIANT(ifp);
1444 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1445 		IFF_UNLOCKGIANT(ifp);
1446 		if (error == 0)
1447 			getmicrotime(&ifp->if_lastchange);
1448 		break;
1449 
1450 	case SIOCGIFSTATUS:
1451 		ifs = (struct ifstat *)data;
1452 		ifs->ascii[0] = '\0';
1453 
1454 	case SIOCGIFPSRCADDR:
1455 	case SIOCGIFPDSTADDR:
1456 	case SIOCGLIFPHYADDR:
1457 	case SIOCGIFMEDIA:
1458 	case SIOCGIFGENERIC:
1459 		if (ifp->if_ioctl == NULL)
1460 			return (EOPNOTSUPP);
1461 		IFF_LOCKGIANT(ifp);
1462 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1463 		IFF_UNLOCKGIANT(ifp);
1464 		break;
1465 
1466 	case SIOCSIFLLADDR:
1467 		error = suser(td);
1468 		if (error)
1469 			return (error);
1470 		error = if_setlladdr(ifp,
1471 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1472 		break;
1473 
1474 	default:
1475 		error = ENOIOCTL;
1476 		break;
1477 	}
1478 	return (error);
1479 }
1480 
1481 /*
1482  * Interface ioctls.
1483  */
1484 int
1485 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1486 {
1487 	struct ifnet *ifp;
1488 	struct ifreq *ifr;
1489 	int error;
1490 	int oif_flags;
1491 
1492 	switch (cmd) {
1493 	case SIOCGIFCONF:
1494 	case OSIOCGIFCONF:
1495 		return (ifconf(cmd, data));
1496 	}
1497 	ifr = (struct ifreq *)data;
1498 
1499 	switch (cmd) {
1500 	case SIOCIFCREATE:
1501 	case SIOCIFDESTROY:
1502 		if ((error = suser(td)) != 0)
1503 			return (error);
1504 		return ((cmd == SIOCIFCREATE) ?
1505 			if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1506 			if_clone_destroy(ifr->ifr_name));
1507 
1508 	case SIOCIFGCLONERS:
1509 		return (if_clone_list((struct if_clonereq *)data));
1510 	}
1511 
1512 	ifp = ifunit(ifr->ifr_name);
1513 	if (ifp == 0)
1514 		return (ENXIO);
1515 
1516 	error = ifhwioctl(cmd, ifp, data, td);
1517 	if (error != ENOIOCTL)
1518 		return (error);
1519 
1520 	oif_flags = ifp->if_flags;
1521 	if (so->so_proto == 0)
1522 		return (EOPNOTSUPP);
1523 #ifndef COMPAT_43
1524 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1525 								 data,
1526 								 ifp, td));
1527 #else
1528 	{
1529 		int ocmd = cmd;
1530 
1531 		switch (cmd) {
1532 
1533 		case SIOCSIFDSTADDR:
1534 		case SIOCSIFADDR:
1535 		case SIOCSIFBRDADDR:
1536 		case SIOCSIFNETMASK:
1537 #if BYTE_ORDER != BIG_ENDIAN
1538 			if (ifr->ifr_addr.sa_family == 0 &&
1539 			    ifr->ifr_addr.sa_len < 16) {
1540 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1541 				ifr->ifr_addr.sa_len = 16;
1542 			}
1543 #else
1544 			if (ifr->ifr_addr.sa_len == 0)
1545 				ifr->ifr_addr.sa_len = 16;
1546 #endif
1547 			break;
1548 
1549 		case OSIOCGIFADDR:
1550 			cmd = SIOCGIFADDR;
1551 			break;
1552 
1553 		case OSIOCGIFDSTADDR:
1554 			cmd = SIOCGIFDSTADDR;
1555 			break;
1556 
1557 		case OSIOCGIFBRDADDR:
1558 			cmd = SIOCGIFBRDADDR;
1559 			break;
1560 
1561 		case OSIOCGIFNETMASK:
1562 			cmd = SIOCGIFNETMASK;
1563 		}
1564 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
1565 								   cmd,
1566 								   data,
1567 								   ifp, td));
1568 		switch (ocmd) {
1569 
1570 		case OSIOCGIFADDR:
1571 		case OSIOCGIFDSTADDR:
1572 		case OSIOCGIFBRDADDR:
1573 		case OSIOCGIFNETMASK:
1574 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1575 
1576 		}
1577 	}
1578 #endif /* COMPAT_43 */
1579 
1580 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1581 #ifdef INET6
1582 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
1583 		if (ifp->if_flags & IFF_UP) {
1584 			int s = splimp();
1585 			in6_if_up(ifp);
1586 			splx(s);
1587 		}
1588 #endif
1589 	}
1590 	return (error);
1591 }
1592 
1593 /*
1594  * The code common to handling reference counted flags,
1595  * e.g., in ifpromisc() and if_allmulti().
1596  * The "pflag" argument can specify a permanent mode flag to check,
1597  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
1598  *
1599  * Only to be used on stack-owned flags, not driver-owned flags.
1600  */
1601 static int
1602 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
1603 {
1604 	struct ifreq ifr;
1605 	int error;
1606 	int oldflags, oldcount;
1607 
1608 	/* Sanity checks to catch programming errors */
1609 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
1610 	    ("%s: setting driver-owned flag %d", __func__, flag));
1611 
1612 	if (onswitch)
1613 		KASSERT(*refcount >= 0,
1614 		    ("%s: increment negative refcount %d for flag %d",
1615 		    __func__, *refcount, flag));
1616 	else
1617 		KASSERT(*refcount > 0,
1618 		    ("%s: decrement non-positive refcount %d for flag %d",
1619 		    __func__, *refcount, flag));
1620 
1621 	/* In case this mode is permanent, just touch refcount */
1622 	if (ifp->if_flags & pflag) {
1623 		*refcount += onswitch ? 1 : -1;
1624 		return (0);
1625 	}
1626 
1627 	/* Save ifnet parameters for if_ioctl() may fail */
1628 	oldcount = *refcount;
1629 	oldflags = ifp->if_flags;
1630 
1631 	/*
1632 	 * See if we aren't the only and touching refcount is enough.
1633 	 * Actually toggle interface flag if we are the first or last.
1634 	 */
1635 	if (onswitch) {
1636 		if ((*refcount)++)
1637 			return (0);
1638 		ifp->if_flags |= flag;
1639 	} else {
1640 		if (--(*refcount))
1641 			return (0);
1642 		ifp->if_flags &= ~flag;
1643 	}
1644 
1645 	/* Call down the driver since we've changed interface flags */
1646 	if (ifp->if_ioctl == NULL) {
1647 		error = EOPNOTSUPP;
1648 		goto recover;
1649 	}
1650 	ifr.ifr_flags = ifp->if_flags & 0xffff;
1651 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1652 	IFF_LOCKGIANT(ifp);
1653 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1654 	IFF_UNLOCKGIANT(ifp);
1655 	if (error)
1656 		goto recover;
1657 	/* Notify userland that interface flags have changed */
1658 	rt_ifmsg(ifp);
1659 	return (0);
1660 
1661 recover:
1662 	/* Recover after driver error */
1663 	*refcount = oldcount;
1664 	ifp->if_flags = oldflags;
1665 	return (error);
1666 }
1667 
1668 /*
1669  * Set/clear promiscuous mode on interface ifp based on the truth value
1670  * of pswitch.  The calls are reference counted so that only the first
1671  * "on" request actually has an effect, as does the final "off" request.
1672  * Results are undefined if the "off" and "on" requests are not matched.
1673  */
1674 int
1675 ifpromisc(struct ifnet *ifp, int pswitch)
1676 {
1677 	int error;
1678 	int oldflags = ifp->if_flags;
1679 
1680 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
1681 			   &ifp->if_pcount, pswitch);
1682 	/* If promiscuous mode status has changed, log a message */
1683 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
1684 		log(LOG_INFO, "%s: promiscuous mode %s\n",
1685 		    ifp->if_xname,
1686 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
1687 	return (error);
1688 }
1689 
1690 /*
1691  * Return interface configuration
1692  * of system.  List may be used
1693  * in later ioctl's (above) to get
1694  * other information.
1695  */
1696 /*ARGSUSED*/
1697 static int
1698 ifconf(u_long cmd, caddr_t data)
1699 {
1700 	struct ifconf *ifc = (struct ifconf *)data;
1701 	struct ifnet *ifp;
1702 	struct ifaddr *ifa;
1703 	struct ifreq ifr;
1704 	struct sbuf *sb;
1705 	int error, full = 0, valid_len, max_len;
1706 
1707 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
1708 	max_len = MAXPHYS - 1;
1709 
1710 	/* Prevent hostile input from being able to crash the system */
1711 	if (ifc->ifc_len <= 0)
1712 		return (EINVAL);
1713 
1714 again:
1715 	if (ifc->ifc_len <= max_len) {
1716 		max_len = ifc->ifc_len;
1717 		full = 1;
1718 	}
1719 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
1720 	max_len = 0;
1721 	valid_len = 0;
1722 
1723 	IFNET_RLOCK();		/* could sleep XXX */
1724 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1725 		int addrs;
1726 
1727 		/*
1728 		 * Zero the ifr_name buffer to make sure we don't
1729 		 * disclose the contents of the stack.
1730 		 */
1731 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
1732 
1733 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1734 		    >= sizeof(ifr.ifr_name)) {
1735 			sbuf_delete(sb);
1736 			IFNET_RUNLOCK();
1737 			return (ENAMETOOLONG);
1738 		}
1739 
1740 		addrs = 0;
1741 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1742 			struct sockaddr *sa = ifa->ifa_addr;
1743 
1744 			if (jailed(curthread->td_ucred) &&
1745 			    prison_if(curthread->td_ucred, sa))
1746 				continue;
1747 			addrs++;
1748 #ifdef COMPAT_43
1749 			if (cmd == OSIOCGIFCONF) {
1750 				struct osockaddr *osa =
1751 					 (struct osockaddr *)&ifr.ifr_addr;
1752 				ifr.ifr_addr = *sa;
1753 				osa->sa_family = sa->sa_family;
1754 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1755 				max_len += sizeof(ifr);
1756 			} else
1757 #endif
1758 			if (sa->sa_len <= sizeof(*sa)) {
1759 				ifr.ifr_addr = *sa;
1760 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1761 				max_len += sizeof(ifr);
1762 			} else {
1763 				sbuf_bcat(sb, &ifr,
1764 				    offsetof(struct ifreq, ifr_addr));
1765 				max_len += offsetof(struct ifreq, ifr_addr);
1766 				sbuf_bcat(sb, sa, sa->sa_len);
1767 				max_len += sa->sa_len;
1768 			}
1769 
1770 			if (!sbuf_overflowed(sb))
1771 				valid_len = sbuf_len(sb);
1772 		}
1773 		if (addrs == 0) {
1774 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
1775 			sbuf_bcat(sb, &ifr, sizeof(ifr));
1776 			max_len += sizeof(ifr);
1777 
1778 			if (!sbuf_overflowed(sb))
1779 				valid_len = sbuf_len(sb);
1780 		}
1781 	}
1782 	IFNET_RUNLOCK();
1783 
1784 	/*
1785 	 * If we didn't allocate enough space (uncommon), try again.  If
1786 	 * we have already allocated as much space as we are allowed,
1787 	 * return what we've got.
1788 	 */
1789 	if (valid_len != max_len && !full) {
1790 		sbuf_delete(sb);
1791 		goto again;
1792 	}
1793 
1794 	ifc->ifc_len = valid_len;
1795 	sbuf_finish(sb);
1796 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
1797 	sbuf_delete(sb);
1798 	return (error);
1799 }
1800 
1801 /*
1802  * Just like ifpromisc(), but for all-multicast-reception mode.
1803  */
1804 int
1805 if_allmulti(struct ifnet *ifp, int onswitch)
1806 {
1807 
1808 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
1809 }
1810 
1811 static struct ifmultiaddr *
1812 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
1813 {
1814 	struct ifmultiaddr *ifma;
1815 
1816 	IF_ADDR_LOCK_ASSERT(ifp);
1817 
1818 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1819 		if (sa_equal(ifma->ifma_addr, sa))
1820 			break;
1821 	}
1822 
1823 	return ifma;
1824 }
1825 
1826 /*
1827  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
1828  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
1829  * the ifnet multicast address list here, so the caller must do that and
1830  * other setup work (such as notifying the device driver).  The reference
1831  * count is initialized to 1.
1832  */
1833 static struct ifmultiaddr *
1834 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
1835     int mflags)
1836 {
1837 	struct ifmultiaddr *ifma;
1838 	struct sockaddr *dupsa;
1839 
1840 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, mflags |
1841 	    M_ZERO);
1842 	if (ifma == NULL)
1843 		return (NULL);
1844 
1845 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, mflags);
1846 	if (dupsa == NULL) {
1847 		FREE(ifma, M_IFMADDR);
1848 		return (NULL);
1849 	}
1850 	bcopy(sa, dupsa, sa->sa_len);
1851 	ifma->ifma_addr = dupsa;
1852 
1853 	ifma->ifma_ifp = ifp;
1854 	ifma->ifma_refcount = 1;
1855 	ifma->ifma_protospec = NULL;
1856 
1857 	if (llsa == NULL) {
1858 		ifma->ifma_lladdr = NULL;
1859 		return (ifma);
1860 	}
1861 
1862 	MALLOC(dupsa, struct sockaddr *, llsa->sa_len, M_IFMADDR, mflags);
1863 	if (dupsa == NULL) {
1864 		FREE(ifma->ifma_addr, M_IFMADDR);
1865 		FREE(ifma, M_IFMADDR);
1866 		return (NULL);
1867 	}
1868 	bcopy(llsa, dupsa, llsa->sa_len);
1869 	ifma->ifma_lladdr = dupsa;
1870 
1871 	return (ifma);
1872 }
1873 
1874 /*
1875  * if_freemulti: free ifmultiaddr structure and possibly attached related
1876  * addresses.  The caller is responsible for implementing reference
1877  * counting, notifying the driver, handling routing messages, and releasing
1878  * any dependent link layer state.
1879  */
1880 static void
1881 if_freemulti(struct ifmultiaddr *ifma)
1882 {
1883 
1884 	KASSERT(ifma->ifma_refcount == 1, ("if_freemulti: refcount %d",
1885 	    ifma->ifma_refcount));
1886 	KASSERT(ifma->ifma_protospec == NULL,
1887 	    ("if_freemulti: protospec not NULL"));
1888 
1889 	if (ifma->ifma_lladdr != NULL)
1890 		FREE(ifma->ifma_lladdr, M_IFMADDR);
1891 	FREE(ifma->ifma_addr, M_IFMADDR);
1892 	FREE(ifma, M_IFMADDR);
1893 }
1894 
1895 /*
1896  * Register an additional multicast address with a network interface.
1897  *
1898  * - If the address is already present, bump the reference count on the
1899  *   address and return.
1900  * - If the address is not link-layer, look up a link layer address.
1901  * - Allocate address structures for one or both addresses, and attach to the
1902  *   multicast address list on the interface.  If automatically adding a link
1903  *   layer address, the protocol address will own a reference to the link
1904  *   layer address, to be freed when it is freed.
1905  * - Notify the network device driver of an addition to the multicast address
1906  *   list.
1907  *
1908  * 'sa' points to caller-owned memory with the desired multicast address.
1909  *
1910  * 'retifma' will be used to return a pointer to the resulting multicast
1911  * address reference, if desired.
1912  */
1913 int
1914 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
1915     struct ifmultiaddr **retifma)
1916 {
1917 	struct ifmultiaddr *ifma, *ll_ifma;
1918 	struct sockaddr *llsa;
1919 	int error;
1920 
1921 	/*
1922 	 * If the address is already present, return a new reference to it;
1923 	 * otherwise, allocate storage and set up a new address.
1924 	 */
1925 	IF_ADDR_LOCK(ifp);
1926 	ifma = if_findmulti(ifp, sa);
1927 	if (ifma != NULL) {
1928 		ifma->ifma_refcount++;
1929 		if (retifma != NULL)
1930 			*retifma = ifma;
1931 		IF_ADDR_UNLOCK(ifp);
1932 		return (0);
1933 	}
1934 
1935 	/*
1936 	 * The address isn't already present; resolve the protocol address
1937 	 * into a link layer address, and then look that up, bump its
1938 	 * refcount or allocate an ifma for that also.  If 'llsa' was
1939 	 * returned, we will need to free it later.
1940 	 */
1941 	llsa = NULL;
1942 	ll_ifma = NULL;
1943 	if (ifp->if_resolvemulti != NULL) {
1944 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
1945 		if (error)
1946 			goto unlock_out;
1947 	}
1948 
1949 	/*
1950 	 * Allocate the new address.  Don't hook it up yet, as we may also
1951 	 * need to allocate a link layer multicast address.
1952 	 */
1953 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
1954 	if (ifma == NULL) {
1955 		error = ENOMEM;
1956 		goto free_llsa_out;
1957 	}
1958 
1959 	/*
1960 	 * If a link layer address is found, we'll need to see if it's
1961 	 * already present in the address list, or allocate is as well.
1962 	 * When this block finishes, the link layer address will be on the
1963 	 * list.
1964 	 */
1965 	if (llsa != NULL) {
1966 		ll_ifma = if_findmulti(ifp, llsa);
1967 		if (ll_ifma == NULL) {
1968 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
1969 			if (ll_ifma == NULL) {
1970 				if_freemulti(ifma);
1971 				error = ENOMEM;
1972 				goto free_llsa_out;
1973 			}
1974 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
1975 			    ifma_link);
1976 		} else
1977 			ll_ifma->ifma_refcount++;
1978 	}
1979 
1980 	/*
1981 	 * We now have a new multicast address, ifma, and possibly a new or
1982 	 * referenced link layer address.  Add the primary address to the
1983 	 * ifnet address list.
1984 	 */
1985 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1986 
1987 	if (retifma != NULL)
1988 		*retifma = ifma;
1989 
1990 	/*
1991 	 * Must generate the message while holding the lock so that 'ifma'
1992 	 * pointer is still valid.
1993 	 *
1994 	 * XXXRW: How come we don't announce ll_ifma?
1995 	 */
1996 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
1997 	IF_ADDR_UNLOCK(ifp);
1998 
1999 	/*
2000 	 * We are certain we have added something, so call down to the
2001 	 * interface to let them know about it.
2002 	 */
2003 	if (ifp->if_ioctl != NULL) {
2004 		IFF_LOCKGIANT(ifp);
2005 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2006 		IFF_UNLOCKGIANT(ifp);
2007 	}
2008 
2009 	if (llsa != NULL)
2010 		FREE(llsa, M_IFMADDR);
2011 
2012 	return (0);
2013 
2014 free_llsa_out:
2015 	if (llsa != NULL)
2016 		FREE(llsa, M_IFMADDR);
2017 
2018 unlock_out:
2019 	IF_ADDR_UNLOCK(ifp);
2020 	return (error);
2021 }
2022 
2023 /*
2024  * Remove a reference to a multicast address on this interface.  Yell
2025  * if the request does not match an existing membership.
2026  */
2027 int
2028 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2029 {
2030 	struct ifmultiaddr *ifma, *ll_ifma;
2031 
2032 	IF_ADDR_LOCK(ifp);
2033 	ifma = if_findmulti(ifp, sa);
2034 	if (ifma == NULL) {
2035 		IF_ADDR_UNLOCK(ifp);
2036 		return ENOENT;
2037 	}
2038 
2039 	if (ifma->ifma_refcount > 1) {
2040 		ifma->ifma_refcount--;
2041 		IF_ADDR_UNLOCK(ifp);
2042 		return 0;
2043 	}
2044 
2045 	sa = ifma->ifma_lladdr;
2046 	if (sa != NULL)
2047 		ll_ifma = if_findmulti(ifp, sa);
2048 	else
2049 		ll_ifma = NULL;
2050 
2051 	/*
2052 	 * XXXRW: How come we don't announce ll_ifma?
2053 	 */
2054 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2055 
2056 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2057 	if_freemulti(ifma);
2058 
2059 	if (ll_ifma != NULL) {
2060 		if (ll_ifma->ifma_refcount == 1) {
2061 			TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifma_link);
2062 			if_freemulti(ll_ifma);
2063 		} else
2064 			ll_ifma->ifma_refcount--;
2065 	}
2066 	IF_ADDR_UNLOCK(ifp);
2067 
2068 	/*
2069 	 * Make sure the interface driver is notified
2070 	 * in the case of a link layer mcast group being left.
2071 	 */
2072 	if (ifp->if_ioctl) {
2073 		IFF_LOCKGIANT(ifp);
2074 		(void) (*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2075 		IFF_UNLOCKGIANT(ifp);
2076 	}
2077 
2078 	return 0;
2079 }
2080 
2081 /*
2082  * Set the link layer address on an interface.
2083  *
2084  * At this time we only support certain types of interfaces,
2085  * and we don't allow the length of the address to change.
2086  */
2087 int
2088 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2089 {
2090 	struct sockaddr_dl *sdl;
2091 	struct ifaddr *ifa;
2092 	struct ifreq ifr;
2093 
2094 	ifa = ifp->if_addr;
2095 	if (ifa == NULL)
2096 		return (EINVAL);
2097 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2098 	if (sdl == NULL)
2099 		return (EINVAL);
2100 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2101 		return (EINVAL);
2102 	switch (ifp->if_type) {
2103 	case IFT_ETHER:
2104 	case IFT_FDDI:
2105 	case IFT_XETHER:
2106 	case IFT_ISO88025:
2107 	case IFT_L2VLAN:
2108 	case IFT_BRIDGE:
2109 	case IFT_ARCNET:
2110 		bcopy(lladdr, LLADDR(sdl), len);
2111 		break;
2112 	default:
2113 		return (ENODEV);
2114 	}
2115 	/*
2116 	 * If the interface is already up, we need
2117 	 * to re-init it in order to reprogram its
2118 	 * address filter.
2119 	 */
2120 	if ((ifp->if_flags & IFF_UP) != 0) {
2121 		if (ifp->if_ioctl) {
2122 			IFF_LOCKGIANT(ifp);
2123 			ifp->if_flags &= ~IFF_UP;
2124 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2125 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2126 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2127 			ifp->if_flags |= IFF_UP;
2128 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2129 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2130 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2131 			IFF_UNLOCKGIANT(ifp);
2132 		}
2133 #ifdef INET
2134 		/*
2135 		 * Also send gratuitous ARPs to notify other nodes about
2136 		 * the address change.
2137 		 */
2138 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2139 			if (ifa->ifa_addr != NULL &&
2140 			    ifa->ifa_addr->sa_family == AF_INET)
2141 				arp_ifinit(ifp, ifa);
2142 		}
2143 #endif
2144 	}
2145 	return (0);
2146 }
2147 
2148 /*
2149  * The name argument must be a pointer to storage which will last as
2150  * long as the interface does.  For physical devices, the result of
2151  * device_get_name(dev) is a good choice and for pseudo-devices a
2152  * static string works well.
2153  */
2154 void
2155 if_initname(struct ifnet *ifp, const char *name, int unit)
2156 {
2157 	ifp->if_dname = name;
2158 	ifp->if_dunit = unit;
2159 	if (unit != IF_DUNIT_NONE)
2160 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2161 	else
2162 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2163 }
2164 
2165 int
2166 if_printf(struct ifnet *ifp, const char * fmt, ...)
2167 {
2168 	va_list ap;
2169 	int retval;
2170 
2171 	retval = printf("%s: ", ifp->if_xname);
2172 	va_start(ap, fmt);
2173 	retval += vprintf(fmt, ap);
2174 	va_end(ap);
2175 	return (retval);
2176 }
2177 
2178 /*
2179  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2180  * be called without Giant.  However, we often can't acquire the Giant lock
2181  * at those points; instead, we run it via a task queue that holds Giant via
2182  * if_start_deferred.
2183  *
2184  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2185  * outstanding if_start_deferred() tasks that will run after the free.  This
2186  * probably means waiting in if_detach().
2187  */
2188 void
2189 if_start(struct ifnet *ifp)
2190 {
2191 
2192 	NET_ASSERT_GIANT();
2193 
2194 	if ((ifp->if_flags & IFF_NEEDSGIANT) != 0 && debug_mpsafenet != 0) {
2195 		if (mtx_owned(&Giant))
2196 			(*(ifp)->if_start)(ifp);
2197 		else
2198 			taskqueue_enqueue(taskqueue_swi_giant,
2199 			    &ifp->if_starttask);
2200 	} else
2201 		(*(ifp)->if_start)(ifp);
2202 }
2203 
2204 static void
2205 if_start_deferred(void *context, int pending)
2206 {
2207 	struct ifnet *ifp;
2208 
2209 	/*
2210 	 * This code must be entered with Giant, and should never run if
2211 	 * we're not running with debug.mpsafenet.
2212 	 */
2213 	KASSERT(debug_mpsafenet != 0, ("if_start_deferred: debug.mpsafenet"));
2214 	GIANT_REQUIRED;
2215 
2216 	ifp = context;
2217 	(ifp->if_start)(ifp);
2218 }
2219 
2220 int
2221 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2222 {
2223 	int active = 0;
2224 
2225 	IF_LOCK(ifq);
2226 	if (_IF_QFULL(ifq)) {
2227 		_IF_DROP(ifq);
2228 		IF_UNLOCK(ifq);
2229 		m_freem(m);
2230 		return (0);
2231 	}
2232 	if (ifp != NULL) {
2233 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2234 		if (m->m_flags & (M_BCAST|M_MCAST))
2235 			ifp->if_omcasts++;
2236 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2237 	}
2238 	_IF_ENQUEUE(ifq, m);
2239 	IF_UNLOCK(ifq);
2240 	if (ifp != NULL && !active)
2241 		if_start(ifp);
2242 	return (1);
2243 }
2244 
2245 void
2246 if_register_com_alloc(u_char type,
2247     if_com_alloc_t *a, if_com_free_t *f)
2248 {
2249 
2250 	KASSERT(if_com_alloc[type] == NULL,
2251 	    ("if_register_com_alloc: %d already registered", type));
2252 	KASSERT(if_com_free[type] == NULL,
2253 	    ("if_register_com_alloc: %d free already registered", type));
2254 
2255 	if_com_alloc[type] = a;
2256 	if_com_free[type] = f;
2257 }
2258 
2259 void
2260 if_deregister_com_alloc(u_char type)
2261 {
2262 
2263 	KASSERT(if_com_alloc[type] == NULL,
2264 	    ("if_deregister_com_alloc: %d not registered", type));
2265 	KASSERT(if_com_free[type] == NULL,
2266 	    ("if_deregister_com_alloc: %d free not registered", type));
2267 	if_com_alloc[type] = NULL;
2268 	if_com_free[type] = NULL;
2269 }
2270