xref: /freebsd/sys/net/if.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
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 #ifdef __amd64__
195 		if (cmd == SIOCGIFCONF32)
196 			return (ifconf(cmd, data));	/* XXX remove cmd */
197 #endif
198 		return (EOPNOTSUPP);
199 	}
200 
201 	ifp = ifnet_byindex(idx);
202 	if (ifp == NULL)
203 		return (ENXIO);
204 
205 	error = ifhwioctl(cmd, ifp, data, td);
206 	if (error == ENOIOCTL)
207 		error = EOPNOTSUPP;
208 	return (error);
209 }
210 
211 static int
212 netkqfilter(struct cdev *dev, struct knote *kn)
213 {
214 	struct knlist *klist;
215 	struct ifnet *ifp;
216 	int idx;
217 
218 	switch (kn->kn_filter) {
219 	case EVFILT_NETDEV:
220 		kn->kn_fop = &netdev_filtops;
221 		break;
222 	default:
223 		return (EINVAL);
224 	}
225 
226 	idx = minor(dev);
227 	if (idx == 0) {
228 		klist = &ifklist;
229 	} else {
230 		ifp = ifnet_byindex(idx);
231 		if (ifp == NULL)
232 			return (1);
233 		klist = &ifp->if_klist;
234 	}
235 
236 	kn->kn_hook = (caddr_t)klist;
237 
238 	knlist_add(klist, kn, 0);
239 
240 	return (0);
241 }
242 
243 static void
244 filt_netdetach(struct knote *kn)
245 {
246 	struct knlist *klist = (struct knlist *)kn->kn_hook;
247 
248 	knlist_remove(klist, kn, 0);
249 }
250 
251 static int
252 filt_netdev(struct knote *kn, long hint)
253 {
254 	struct knlist *klist = (struct knlist *)kn->kn_hook;
255 
256 	/*
257 	 * Currently NOTE_EXIT is abused to indicate device detach.
258 	 */
259 	if (hint == NOTE_EXIT) {
260 		kn->kn_data = NOTE_LINKINV;
261 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
262 		knlist_remove_inevent(klist, kn);
263 		return (1);
264 	}
265 	if (hint != 0)
266 		kn->kn_data = hint;			/* current status */
267 	if (kn->kn_sfflags & hint)
268 		kn->kn_fflags |= hint;
269 	return (kn->kn_fflags != 0);
270 }
271 
272 /*
273  * Network interface utility routines.
274  *
275  * Routines with ifa_ifwith* names take sockaddr *'s as
276  * parameters.
277  */
278 /* ARGSUSED*/
279 static void
280 if_init(void *dummy __unused)
281 {
282 
283 	IFNET_LOCK_INIT();
284 	TAILQ_INIT(&ifnet);
285 	knlist_init(&ifklist, NULL, NULL, NULL, NULL);
286 	if_grow();				/* create initial table */
287 	ifdev_byindex(0) = make_dev(&net_cdevsw, 0,
288 	    UID_ROOT, GID_WHEEL, 0600, "network");
289 	if_clone_init();
290 }
291 
292 static void
293 if_grow(void)
294 {
295 	u_int n;
296 	struct ifindex_entry *e;
297 
298 	if_indexlim <<= 1;
299 	n = if_indexlim * sizeof(*e);
300 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
301 	if (ifindex_table != NULL) {
302 		memcpy((caddr_t)e, (caddr_t)ifindex_table, n/2);
303 		free((caddr_t)ifindex_table, M_IFNET);
304 	}
305 	ifindex_table = e;
306 }
307 
308 /* ARGSUSED*/
309 static void
310 if_check(void *dummy __unused)
311 {
312 	struct ifnet *ifp;
313 	int s;
314 
315 	s = splimp();
316 	IFNET_RLOCK();	/* could sleep on rare error; mostly okay XXX */
317 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
318 		if (ifp->if_snd.ifq_maxlen == 0) {
319 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
320 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
321 		}
322 		if (!mtx_initialized(&ifp->if_snd.ifq_mtx)) {
323 			if_printf(ifp,
324 			    "XXX: driver didn't initialize queue mtx\n");
325 			mtx_init(&ifp->if_snd.ifq_mtx, "unknown",
326 			    MTX_NETWORK_LOCK, MTX_DEF);
327 		}
328 	}
329 	IFNET_RUNLOCK();
330 	splx(s);
331 	if_slowtimo(0);
332 }
333 
334 /*
335  * Allocate a struct ifnet and in index for an interface.
336  */
337 struct ifnet*
338 if_alloc(u_char type)
339 {
340 	struct ifnet *ifp;
341 
342 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
343 
344 	/*
345 	 * Try to find an empty slot below if_index.  If we fail, take
346 	 * the next slot.
347 	 *
348 	 * XXX: should be locked!
349 	 */
350 	for (ifp->if_index = 1; ifp->if_index <= if_index; ifp->if_index++) {
351 		if (ifnet_byindex(ifp->if_index) == NULL)
352 			break;
353 	}
354 	/* Catch if_index overflow. */
355 	if (ifp->if_index < 1) {
356 		free(ifp, M_IFNET);
357 		return (NULL);
358 	}
359 	if (ifp->if_index > if_index)
360 		if_index = ifp->if_index;
361 	if (if_index >= if_indexlim)
362 		if_grow();
363 	ifnet_byindex(ifp->if_index) = ifp;
364 
365 	ifp->if_type = type;
366 
367 	if (if_com_alloc[type] != NULL) {
368 		ifp->if_l2com = if_com_alloc[type](type, ifp);
369 		if (ifp->if_l2com == NULL) {
370 			free(ifp, M_IFNET);
371 			return (NULL);
372 		}
373 	}
374 	IF_ADDR_LOCK_INIT(ifp);
375 
376 	return (ifp);
377 }
378 
379 void
380 if_free(struct ifnet *ifp)
381 {
382 
383 	/* Do not add code to this function!  Add it to if_free_type(). */
384 	if_free_type(ifp, ifp->if_type);
385 }
386 
387 void
388 if_free_type(struct ifnet *ifp, u_char type)
389 {
390 
391 	if (ifp != ifnet_byindex(ifp->if_index)) {
392 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
393 		    __func__);
394 		return;
395 	}
396 
397 	IF_ADDR_LOCK_DESTROY(ifp);
398 
399 	ifnet_byindex(ifp->if_index) = NULL;
400 
401 	/* XXX: should be locked with if_findindex() */
402 	while (if_index > 0 && ifnet_byindex(if_index) == NULL)
403 		if_index--;
404 
405 	if (if_com_free[type] != NULL)
406 		if_com_free[type](ifp->if_l2com, type);
407 
408 	free(ifp, M_IFNET);
409 };
410 
411 /*
412  * Attach an interface to the
413  * list of "active" interfaces.
414  */
415 void
416 if_attach(struct ifnet *ifp)
417 {
418 	unsigned socksize, ifasize;
419 	int namelen, masklen;
420 	struct sockaddr_dl *sdl;
421 	struct ifaddr *ifa;
422 
423 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
424 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
425 		    ifp->if_xname);
426 
427 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
428 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
429 	IF_AFDATA_LOCK_INIT(ifp);
430 	ifp->if_afdata_initialized = 0;
431 	IFNET_WLOCK();
432 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
433 	IFNET_WUNLOCK();
434 	/*
435 	 * XXX -
436 	 * The old code would work if the interface passed a pre-existing
437 	 * chain of ifaddrs to this code.  We don't trust our callers to
438 	 * properly initialize the tailq, however, so we no longer allow
439 	 * this unlikely case.
440 	 */
441 	TAILQ_INIT(&ifp->if_addrhead);
442 	TAILQ_INIT(&ifp->if_prefixhead);
443 	TAILQ_INIT(&ifp->if_multiaddrs);
444 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
445 	getmicrotime(&ifp->if_lastchange);
446 	ifp->if_data.ifi_epoch = time_uptime;
447 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
448 
449 #ifdef MAC
450 	mac_init_ifnet(ifp);
451 	mac_create_ifnet(ifp);
452 #endif
453 
454 	ifdev_byindex(ifp->if_index) = make_dev(&net_cdevsw,
455 	    unit2minor(ifp->if_index),
456 	    UID_ROOT, GID_WHEEL, 0600, "%s/%s",
457 	    net_cdevsw.d_name, ifp->if_xname);
458 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
459 	    net_cdevsw.d_name, ifp->if_index);
460 
461 	mtx_init(&ifp->if_snd.ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
462 
463 	/*
464 	 * create a Link Level name for this device
465 	 */
466 	namelen = strlen(ifp->if_xname);
467 	/*
468 	 * Always save enough space for any possiable name so we can do
469 	 * a rename in place later.
470 	 */
471 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
472 	socksize = masklen + ifp->if_addrlen;
473 	if (socksize < sizeof(*sdl))
474 		socksize = sizeof(*sdl);
475 	socksize = roundup2(socksize, sizeof(long));
476 	ifasize = sizeof(*ifa) + 2 * socksize;
477 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
478 	IFA_LOCK_INIT(ifa);
479 	sdl = (struct sockaddr_dl *)(ifa + 1);
480 	sdl->sdl_len = socksize;
481 	sdl->sdl_family = AF_LINK;
482 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
483 	sdl->sdl_nlen = namelen;
484 	sdl->sdl_index = ifp->if_index;
485 	sdl->sdl_type = ifp->if_type;
486 	ifp->if_addr = ifa;
487 	ifa->ifa_ifp = ifp;
488 	ifa->ifa_rtrequest = link_rtrequest;
489 	ifa->ifa_addr = (struct sockaddr *)sdl;
490 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
491 	ifa->ifa_netmask = (struct sockaddr *)sdl;
492 	sdl->sdl_len = masklen;
493 	while (namelen != 0)
494 		sdl->sdl_data[--namelen] = 0xff;
495 	ifa->ifa_refcnt = 1;
496 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
497 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
498 	ifp->if_snd.altq_type = 0;
499 	ifp->if_snd.altq_disc = NULL;
500 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
501 	ifp->if_snd.altq_tbr  = NULL;
502 	ifp->if_snd.altq_ifp  = ifp;
503 
504 	if (domain_init_status >= 2)
505 		if_attachdomain1(ifp);
506 
507 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
508 
509 	/* Announce the interface. */
510 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
511 }
512 
513 static void
514 if_attachdomain(void *dummy)
515 {
516 	struct ifnet *ifp;
517 	int s;
518 
519 	s = splnet();
520 	TAILQ_FOREACH(ifp, &ifnet, if_link)
521 		if_attachdomain1(ifp);
522 	splx(s);
523 }
524 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
525     if_attachdomain, NULL);
526 
527 static void
528 if_attachdomain1(struct ifnet *ifp)
529 {
530 	struct domain *dp;
531 	int s;
532 
533 	s = splnet();
534 
535 	/*
536 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
537 	 * cannot lock ifp->if_afdata initialization, entirely.
538 	 */
539 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
540 		splx(s);
541 		return;
542 	}
543 	if (ifp->if_afdata_initialized >= domain_init_status) {
544 		IF_AFDATA_UNLOCK(ifp);
545 		splx(s);
546 		printf("if_attachdomain called more than once on %s\n",
547 		    ifp->if_xname);
548 		return;
549 	}
550 	ifp->if_afdata_initialized = domain_init_status;
551 	IF_AFDATA_UNLOCK(ifp);
552 
553 	/* address family dependent data region */
554 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
555 	for (dp = domains; dp; dp = dp->dom_next) {
556 		if (dp->dom_ifattach)
557 			ifp->if_afdata[dp->dom_family] =
558 			    (*dp->dom_ifattach)(ifp);
559 	}
560 
561 	splx(s);
562 }
563 
564 /*
565  * Remove any network addresses from an interface.
566  */
567 
568 void
569 if_purgeaddrs(struct ifnet *ifp)
570 {
571 	struct ifaddr *ifa, *next;
572 
573 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
574 
575 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_LINK)
576 			continue;
577 #ifdef INET
578 		/* XXX: Ugly!! ad hoc just for INET */
579 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
580 			struct ifaliasreq ifr;
581 
582 			bzero(&ifr, sizeof(ifr));
583 			ifr.ifra_addr = *ifa->ifa_addr;
584 			if (ifa->ifa_dstaddr)
585 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
586 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
587 			    NULL) == 0)
588 				continue;
589 		}
590 #endif /* INET */
591 #ifdef INET6
592 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
593 			in6_purgeaddr(ifa);
594 			/* ifp_addrhead is already updated */
595 			continue;
596 		}
597 #endif /* INET6 */
598 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
599 		IFAFREE(ifa);
600 	}
601 }
602 
603 /*
604  * Detach an interface, removing it from the
605  * list of "active" interfaces.
606  *
607  * XXXRW: There are some significant questions about event ordering, and
608  * how to prevent things from starting to use the interface during detach.
609  */
610 void
611 if_detach(struct ifnet *ifp)
612 {
613 	struct ifaddr *ifa;
614 	struct radix_node_head	*rnh;
615 	int s;
616 	int i;
617 	struct domain *dp;
618  	struct ifnet *iter;
619  	int found;
620 
621 	/*
622 	 * Remove/wait for pending events.
623 	 */
624 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
625 
626 #ifdef DEV_CARP
627 	/* Maybe hook to the generalized departure handler above?!? */
628 	if (ifp->if_carp)
629 		carp_ifdetach(ifp);
630 #endif
631 
632 	/*
633 	 * Remove routes and flush queues.
634 	 */
635 	s = splnet();
636 	if_down(ifp);
637 #ifdef ALTQ
638 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
639 		altq_disable(&ifp->if_snd);
640 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
641 		altq_detach(&ifp->if_snd);
642 #endif
643 
644 	if_purgeaddrs(ifp);
645 
646 #ifdef INET
647 	in_ifdetach(ifp);
648 #endif
649 
650 #ifdef INET6
651 	/*
652 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
653 	 * before removing routing entries below, since IPv6 interface direct
654 	 * routes are expected to be removed by the IPv6-specific kernel API.
655 	 * Otherwise, the kernel will detect some inconsistency and bark it.
656 	 */
657 	in6_ifdetach(ifp);
658 #endif
659 	/*
660 	 * Remove link ifaddr pointer and maybe decrement if_index.
661 	 * Clean up all addresses.
662 	 */
663 	ifp->if_addr = NULL;
664 	destroy_dev(ifdev_byindex(ifp->if_index));
665 	ifdev_byindex(ifp->if_index) = NULL;
666 
667 	/* We can now free link ifaddr. */
668 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
669 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
670 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
671 		IFAFREE(ifa);
672 	}
673 
674 	/*
675 	 * Delete all remaining routes using this interface
676 	 * Unfortuneatly the only way to do this is to slog through
677 	 * the entire routing table looking for routes which point
678 	 * to this interface...oh well...
679 	 */
680 	for (i = 1; i <= AF_MAX; i++) {
681 		if ((rnh = rt_tables[i]) == NULL)
682 			continue;
683 		RADIX_NODE_HEAD_LOCK(rnh);
684 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
685 		RADIX_NODE_HEAD_UNLOCK(rnh);
686 	}
687 
688 	/* Announce that the interface is gone. */
689 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
690 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
691 
692 	IF_AFDATA_LOCK(ifp);
693 	for (dp = domains; dp; dp = dp->dom_next) {
694 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
695 			(*dp->dom_ifdetach)(ifp,
696 			    ifp->if_afdata[dp->dom_family]);
697 	}
698 	IF_AFDATA_UNLOCK(ifp);
699 
700 #ifdef MAC
701 	mac_destroy_ifnet(ifp);
702 #endif /* MAC */
703 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
704 	knlist_clear(&ifp->if_klist, 0);
705 	knlist_destroy(&ifp->if_klist);
706 	IFNET_WLOCK();
707  	found = 0;
708  	TAILQ_FOREACH(iter, &ifnet, if_link)
709  		if (iter == ifp) {
710  			found = 1;
711  			break;
712  		}
713  	if (found)
714  		TAILQ_REMOVE(&ifnet, ifp, if_link);
715 	IFNET_WUNLOCK();
716 	mtx_destroy(&ifp->if_snd.ifq_mtx);
717 	IF_AFDATA_DESTROY(ifp);
718 	splx(s);
719 }
720 
721 /*
722  * Delete Routes for a Network Interface
723  *
724  * Called for each routing entry via the rnh->rnh_walktree() call above
725  * to delete all route entries referencing a detaching network interface.
726  *
727  * Arguments:
728  *	rn	pointer to node in the routing table
729  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
730  *
731  * Returns:
732  *	0	successful
733  *	errno	failed - reason indicated
734  *
735  */
736 static int
737 if_rtdel(struct radix_node *rn, void *arg)
738 {
739 	struct rtentry	*rt = (struct rtentry *)rn;
740 	struct ifnet	*ifp = arg;
741 	int		err;
742 
743 	if (rt->rt_ifp == ifp) {
744 
745 		/*
746 		 * Protect (sorta) against walktree recursion problems
747 		 * with cloned routes
748 		 */
749 		if ((rt->rt_flags & RTF_UP) == 0)
750 			return (0);
751 
752 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
753 				rt_mask(rt), rt->rt_flags,
754 				(struct rtentry **) NULL);
755 		if (err) {
756 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
757 		}
758 	}
759 
760 	return (0);
761 }
762 
763 #define	sa_equal(a1, a2)	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
764 
765 /*
766  * Locate an interface based on a complete address.
767  */
768 /*ARGSUSED*/
769 struct ifaddr *
770 ifa_ifwithaddr(struct sockaddr *addr)
771 {
772 	struct ifnet *ifp;
773 	struct ifaddr *ifa;
774 
775 	IFNET_RLOCK();
776 	TAILQ_FOREACH(ifp, &ifnet, if_link)
777 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
778 			if (ifa->ifa_addr->sa_family != addr->sa_family)
779 				continue;
780 			if (sa_equal(addr, ifa->ifa_addr))
781 				goto done;
782 			/* IP6 doesn't have broadcast */
783 			if ((ifp->if_flags & IFF_BROADCAST) &&
784 			    ifa->ifa_broadaddr &&
785 			    ifa->ifa_broadaddr->sa_len != 0 &&
786 			    sa_equal(ifa->ifa_broadaddr, addr))
787 				goto done;
788 		}
789 	ifa = NULL;
790 done:
791 	IFNET_RUNLOCK();
792 	return (ifa);
793 }
794 
795 /*
796  * Locate the point to point interface with a given destination address.
797  */
798 /*ARGSUSED*/
799 struct ifaddr *
800 ifa_ifwithdstaddr(struct sockaddr *addr)
801 {
802 	struct ifnet *ifp;
803 	struct ifaddr *ifa;
804 
805 	IFNET_RLOCK();
806 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
807 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
808 			continue;
809 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
810 			if (ifa->ifa_addr->sa_family != addr->sa_family)
811 				continue;
812 			if (ifa->ifa_dstaddr &&
813 			    sa_equal(addr, ifa->ifa_dstaddr))
814 				goto done;
815 		}
816 	}
817 	ifa = NULL;
818 done:
819 	IFNET_RUNLOCK();
820 	return (ifa);
821 }
822 
823 /*
824  * Find an interface on a specific network.  If many, choice
825  * is most specific found.
826  */
827 struct ifaddr *
828 ifa_ifwithnet(struct sockaddr *addr)
829 {
830 	struct ifnet *ifp;
831 	struct ifaddr *ifa;
832 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
833 	u_int af = addr->sa_family;
834 	char *addr_data = addr->sa_data, *cplim;
835 
836 	/*
837 	 * AF_LINK addresses can be looked up directly by their index number,
838 	 * so do that if we can.
839 	 */
840 	if (af == AF_LINK) {
841 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
842 	    if (sdl->sdl_index && sdl->sdl_index <= if_index)
843 		return (ifaddr_byindex(sdl->sdl_index));
844 	}
845 
846 	/*
847 	 * Scan though each interface, looking for ones that have
848 	 * addresses in this address family.
849 	 */
850 	IFNET_RLOCK();
851 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
852 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
853 			char *cp, *cp2, *cp3;
854 
855 			if (ifa->ifa_addr->sa_family != af)
856 next:				continue;
857 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
858 				/*
859 				 * This is a bit broken as it doesn't
860 				 * take into account that the remote end may
861 				 * be a single node in the network we are
862 				 * looking for.
863 				 * The trouble is that we don't know the
864 				 * netmask for the remote end.
865 				 */
866 				if (ifa->ifa_dstaddr != 0 &&
867 				    sa_equal(addr, ifa->ifa_dstaddr))
868 					goto done;
869 			} else {
870 				/*
871 				 * if we have a special address handler,
872 				 * then use it instead of the generic one.
873 				 */
874 				if (ifa->ifa_claim_addr) {
875 					if ((*ifa->ifa_claim_addr)(ifa, addr))
876 						goto done;
877 					continue;
878 				}
879 
880 				/*
881 				 * Scan all the bits in the ifa's address.
882 				 * If a bit dissagrees with what we are
883 				 * looking for, mask it with the netmask
884 				 * to see if it really matters.
885 				 * (A byte at a time)
886 				 */
887 				if (ifa->ifa_netmask == 0)
888 					continue;
889 				cp = addr_data;
890 				cp2 = ifa->ifa_addr->sa_data;
891 				cp3 = ifa->ifa_netmask->sa_data;
892 				cplim = ifa->ifa_netmask->sa_len
893 					+ (char *)ifa->ifa_netmask;
894 				while (cp3 < cplim)
895 					if ((*cp++ ^ *cp2++) & *cp3++)
896 						goto next; /* next address! */
897 				/*
898 				 * If the netmask of what we just found
899 				 * is more specific than what we had before
900 				 * (if we had one) then remember the new one
901 				 * before continuing to search
902 				 * for an even better one.
903 				 */
904 				if (ifa_maybe == 0 ||
905 				    rn_refines((caddr_t)ifa->ifa_netmask,
906 				    (caddr_t)ifa_maybe->ifa_netmask))
907 					ifa_maybe = ifa;
908 			}
909 		}
910 	}
911 	ifa = ifa_maybe;
912 done:
913 	IFNET_RUNLOCK();
914 	return (ifa);
915 }
916 
917 /*
918  * Find an interface address specific to an interface best matching
919  * a given address.
920  */
921 struct ifaddr *
922 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
923 {
924 	struct ifaddr *ifa;
925 	char *cp, *cp2, *cp3;
926 	char *cplim;
927 	struct ifaddr *ifa_maybe = 0;
928 	u_int af = addr->sa_family;
929 
930 	if (af >= AF_MAX)
931 		return (0);
932 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
933 		if (ifa->ifa_addr->sa_family != af)
934 			continue;
935 		if (ifa_maybe == 0)
936 			ifa_maybe = ifa;
937 		if (ifa->ifa_netmask == 0) {
938 			if (sa_equal(addr, ifa->ifa_addr) ||
939 			    (ifa->ifa_dstaddr &&
940 			    sa_equal(addr, ifa->ifa_dstaddr)))
941 				goto done;
942 			continue;
943 		}
944 		if (ifp->if_flags & IFF_POINTOPOINT) {
945 			if (sa_equal(addr, ifa->ifa_dstaddr))
946 				goto done;
947 		} else {
948 			cp = addr->sa_data;
949 			cp2 = ifa->ifa_addr->sa_data;
950 			cp3 = ifa->ifa_netmask->sa_data;
951 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
952 			for (; cp3 < cplim; cp3++)
953 				if ((*cp++ ^ *cp2++) & *cp3)
954 					break;
955 			if (cp3 == cplim)
956 				goto done;
957 		}
958 	}
959 	ifa = ifa_maybe;
960 done:
961 	return (ifa);
962 }
963 
964 #include <net/route.h>
965 
966 /*
967  * Default action when installing a route with a Link Level gateway.
968  * Lookup an appropriate real ifa to point to.
969  * This should be moved to /sys/net/link.c eventually.
970  */
971 static void
972 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
973 {
974 	struct ifaddr *ifa, *oifa;
975 	struct sockaddr *dst;
976 	struct ifnet *ifp;
977 
978 	RT_LOCK_ASSERT(rt);
979 
980 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
981 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
982 		return;
983 	ifa = ifaof_ifpforaddr(dst, ifp);
984 	if (ifa) {
985 		IFAREF(ifa);		/* XXX */
986 		oifa = rt->rt_ifa;
987 		rt->rt_ifa = ifa;
988 		IFAFREE(oifa);
989 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
990 			ifa->ifa_rtrequest(cmd, rt, info);
991 	}
992 }
993 
994 /*
995  * Mark an interface down and notify protocols of
996  * the transition.
997  * NOTE: must be called at splnet or eqivalent.
998  */
999 static void
1000 if_unroute(struct ifnet *ifp, int flag, int fam)
1001 {
1002 	struct ifaddr *ifa;
1003 
1004 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1005 
1006 	ifp->if_flags &= ~flag;
1007 	getmicrotime(&ifp->if_lastchange);
1008 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1009 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1010 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1011 	if_qflush(&ifp->if_snd);
1012 #ifdef DEV_CARP
1013 	if (ifp->if_carp)
1014 		carp_carpdev_state(ifp->if_carp);
1015 #endif
1016 	rt_ifmsg(ifp);
1017 }
1018 
1019 /*
1020  * Mark an interface up and notify protocols of
1021  * the transition.
1022  * NOTE: must be called at splnet or eqivalent.
1023  */
1024 static void
1025 if_route(struct ifnet *ifp, int flag, int fam)
1026 {
1027 	struct ifaddr *ifa;
1028 
1029 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1030 
1031 	ifp->if_flags |= flag;
1032 	getmicrotime(&ifp->if_lastchange);
1033 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1034 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1035 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1036 #ifdef DEV_CARP
1037 	if (ifp->if_carp)
1038 		carp_carpdev_state(ifp->if_carp);
1039 #endif
1040 	rt_ifmsg(ifp);
1041 #ifdef INET6
1042 	in6_if_up(ifp);
1043 #endif
1044 }
1045 
1046 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1047 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1048 
1049 /*
1050  * Handle a change in the interface link state. To avoid LORs
1051  * between driver lock and upper layer locks, as well as possible
1052  * recursions, we post event to taskqueue, and all job
1053  * is done in static do_link_state_change().
1054  */
1055 void
1056 if_link_state_change(struct ifnet *ifp, int link_state)
1057 {
1058 	/* Return if state hasn't changed. */
1059 	if (ifp->if_link_state == link_state)
1060 		return;
1061 
1062 	ifp->if_link_state = link_state;
1063 
1064 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1065 }
1066 
1067 static void
1068 do_link_state_change(void *arg, int pending)
1069 {
1070 	struct ifnet *ifp = (struct ifnet *)arg;
1071 	int link_state = ifp->if_link_state;
1072 	int link;
1073 
1074 	/* Notify that the link state has changed. */
1075 	rt_ifmsg(ifp);
1076 	if (link_state == LINK_STATE_UP)
1077 		link = NOTE_LINKUP;
1078 	else if (link_state == LINK_STATE_DOWN)
1079 		link = NOTE_LINKDOWN;
1080 	else
1081 		link = NOTE_LINKINV;
1082 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1083 	if (ifp->if_vlantrunk != NULL)
1084 		(*vlan_link_state_p)(ifp, link);
1085 
1086 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1087 	    IFP2AC(ifp)->ac_netgraph != NULL)
1088 		(*ng_ether_link_state_p)(ifp, link_state);
1089 #ifdef DEV_CARP
1090 	if (ifp->if_carp)
1091 		carp_carpdev_state(ifp->if_carp);
1092 #endif
1093 	if (ifp->if_bridge) {
1094 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1095 		(*bstp_linkstate_p)(ifp, link_state);
1096 	}
1097 
1098 	devctl_notify("IFNET", ifp->if_xname,
1099 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1100 	if (pending > 1)
1101 		if_printf(ifp, "%d link states coalesced\n", pending);
1102 	if (log_link_state_change)
1103 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1104 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1105 }
1106 
1107 /*
1108  * Mark an interface down and notify protocols of
1109  * the transition.
1110  * NOTE: must be called at splnet or eqivalent.
1111  */
1112 void
1113 if_down(struct ifnet *ifp)
1114 {
1115 
1116 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1117 }
1118 
1119 /*
1120  * Mark an interface up and notify protocols of
1121  * the transition.
1122  * NOTE: must be called at splnet or eqivalent.
1123  */
1124 void
1125 if_up(struct ifnet *ifp)
1126 {
1127 
1128 	if_route(ifp, IFF_UP, AF_UNSPEC);
1129 }
1130 
1131 /*
1132  * Flush an interface queue.
1133  */
1134 static void
1135 if_qflush(struct ifaltq *ifq)
1136 {
1137 	struct mbuf *m, *n;
1138 
1139 	IFQ_LOCK(ifq);
1140 #ifdef ALTQ
1141 	if (ALTQ_IS_ENABLED(ifq))
1142 		ALTQ_PURGE(ifq);
1143 #endif
1144 	n = ifq->ifq_head;
1145 	while ((m = n) != 0) {
1146 		n = m->m_act;
1147 		m_freem(m);
1148 	}
1149 	ifq->ifq_head = 0;
1150 	ifq->ifq_tail = 0;
1151 	ifq->ifq_len = 0;
1152 	IFQ_UNLOCK(ifq);
1153 }
1154 
1155 /*
1156  * Handle interface watchdog timer routines.  Called
1157  * from softclock, we decrement timers (if set) and
1158  * call the appropriate interface routine on expiration.
1159  *
1160  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1161  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1162  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1163  */
1164 static void
1165 if_slowtimo(void *arg)
1166 {
1167 	struct ifnet *ifp;
1168 	int s = splimp();
1169 
1170 	IFNET_RLOCK();
1171 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1172 		if (ifp->if_timer == 0 || --ifp->if_timer)
1173 			continue;
1174 		if (ifp->if_watchdog)
1175 			(*ifp->if_watchdog)(ifp);
1176 	}
1177 	IFNET_RUNLOCK();
1178 	splx(s);
1179 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1180 }
1181 
1182 /*
1183  * Map interface name to
1184  * interface structure pointer.
1185  */
1186 struct ifnet *
1187 ifunit(const char *name)
1188 {
1189 	struct ifnet *ifp;
1190 
1191 	IFNET_RLOCK();
1192 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1193 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1194 			break;
1195 	}
1196 	IFNET_RUNLOCK();
1197 	return (ifp);
1198 }
1199 
1200 /*
1201  * Hardware specific interface ioctls.
1202  */
1203 static int
1204 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1205 {
1206 	struct ifreq *ifr;
1207 	struct ifstat *ifs;
1208 	int error = 0;
1209 	int new_flags, temp_flags;
1210 	size_t namelen, onamelen;
1211 	char new_name[IFNAMSIZ];
1212 	struct ifaddr *ifa;
1213 	struct sockaddr_dl *sdl;
1214 
1215 	ifr = (struct ifreq *)data;
1216 	switch (cmd) {
1217 	case SIOCGIFINDEX:
1218 		ifr->ifr_index = ifp->if_index;
1219 		break;
1220 
1221 	case SIOCGIFFLAGS:
1222 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1223 		ifr->ifr_flags = temp_flags & 0xffff;
1224 		ifr->ifr_flagshigh = temp_flags >> 16;
1225 		break;
1226 
1227 	case SIOCGIFCAP:
1228 		ifr->ifr_reqcap = ifp->if_capabilities;
1229 		ifr->ifr_curcap = ifp->if_capenable;
1230 		break;
1231 
1232 #ifdef MAC
1233 	case SIOCGIFMAC:
1234 		error = mac_ioctl_ifnet_get(td->td_ucred, ifr, ifp);
1235 		break;
1236 #endif
1237 
1238 	case SIOCGIFMETRIC:
1239 		ifr->ifr_metric = ifp->if_metric;
1240 		break;
1241 
1242 	case SIOCGIFMTU:
1243 		ifr->ifr_mtu = ifp->if_mtu;
1244 		break;
1245 
1246 	case SIOCGIFPHYS:
1247 		ifr->ifr_phys = ifp->if_physical;
1248 		break;
1249 
1250 	case SIOCSIFFLAGS:
1251 		error = suser(td);
1252 		if (error)
1253 			return (error);
1254 		/*
1255 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1256 		 * check, so we don't need special handling here yet.
1257 		 */
1258 		new_flags = (ifr->ifr_flags & 0xffff) |
1259 		    (ifr->ifr_flagshigh << 16);
1260 		if (ifp->if_flags & IFF_SMART) {
1261 			/* Smart drivers twiddle their own routes */
1262 		} else if (ifp->if_flags & IFF_UP &&
1263 		    (new_flags & IFF_UP) == 0) {
1264 			int s = splimp();
1265 			if_down(ifp);
1266 			splx(s);
1267 		} else if (new_flags & IFF_UP &&
1268 		    (ifp->if_flags & IFF_UP) == 0) {
1269 			int s = splimp();
1270 			if_up(ifp);
1271 			splx(s);
1272 		}
1273 		/* See if permanently promiscuous mode bit is about to flip */
1274 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1275 			if (new_flags & IFF_PPROMISC)
1276 				ifp->if_flags |= IFF_PROMISC;
1277 			else if (ifp->if_pcount == 0)
1278 				ifp->if_flags &= ~IFF_PROMISC;
1279 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1280 			    ifp->if_xname,
1281 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1282 		}
1283 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1284 			(new_flags &~ IFF_CANTCHANGE);
1285 		if (ifp->if_ioctl) {
1286 			IFF_LOCKGIANT(ifp);
1287 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1288 			IFF_UNLOCKGIANT(ifp);
1289 		}
1290 		getmicrotime(&ifp->if_lastchange);
1291 		break;
1292 
1293 	case SIOCSIFCAP:
1294 		error = suser(td);
1295 		if (error)
1296 			return (error);
1297 		if (ifp->if_ioctl == NULL)
1298 			return (EOPNOTSUPP);
1299 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1300 			return (EINVAL);
1301 		IFF_LOCKGIANT(ifp);
1302 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1303 		IFF_UNLOCKGIANT(ifp);
1304 		if (error == 0)
1305 			getmicrotime(&ifp->if_lastchange);
1306 		break;
1307 
1308 #ifdef MAC
1309 	case SIOCSIFMAC:
1310 		error = mac_ioctl_ifnet_set(td->td_ucred, ifr, ifp);
1311 		break;
1312 #endif
1313 
1314 	case SIOCSIFNAME:
1315 		error = suser(td);
1316 		if (error != 0)
1317 			return (error);
1318 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1319 		if (error != 0)
1320 			return (error);
1321 		if (new_name[0] == '\0')
1322 			return (EINVAL);
1323 		if (ifunit(new_name) != NULL)
1324 			return (EEXIST);
1325 
1326 		/* Announce the departure of the interface. */
1327 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1328 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1329 
1330 		log(LOG_INFO, "%s: changing name to '%s'\n",
1331 		    ifp->if_xname, new_name);
1332 
1333 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1334 		ifa = ifp->if_addr;
1335 		IFA_LOCK(ifa);
1336 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1337 		namelen = strlen(new_name);
1338 		onamelen = sdl->sdl_nlen;
1339 		/*
1340 		 * Move the address if needed.  This is safe because we
1341 		 * allocate space for a name of length IFNAMSIZ when we
1342 		 * create this in if_attach().
1343 		 */
1344 		if (namelen != onamelen) {
1345 			bcopy(sdl->sdl_data + onamelen,
1346 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1347 		}
1348 		bcopy(new_name, sdl->sdl_data, namelen);
1349 		sdl->sdl_nlen = namelen;
1350 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1351 		bzero(sdl->sdl_data, onamelen);
1352 		while (namelen != 0)
1353 			sdl->sdl_data[--namelen] = 0xff;
1354 		IFA_UNLOCK(ifa);
1355 
1356 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1357 		/* Announce the return of the interface. */
1358 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1359 		break;
1360 
1361 	case SIOCSIFMETRIC:
1362 		error = suser(td);
1363 		if (error)
1364 			return (error);
1365 		ifp->if_metric = ifr->ifr_metric;
1366 		getmicrotime(&ifp->if_lastchange);
1367 		break;
1368 
1369 	case SIOCSIFPHYS:
1370 		error = suser(td);
1371 		if (error)
1372 			return (error);
1373 		if (ifp->if_ioctl == NULL)
1374 			return (EOPNOTSUPP);
1375 		IFF_LOCKGIANT(ifp);
1376 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1377 		IFF_UNLOCKGIANT(ifp);
1378 		if (error == 0)
1379 			getmicrotime(&ifp->if_lastchange);
1380 		break;
1381 
1382 	case SIOCSIFMTU:
1383 	{
1384 		u_long oldmtu = ifp->if_mtu;
1385 
1386 		error = suser(td);
1387 		if (error)
1388 			return (error);
1389 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1390 			return (EINVAL);
1391 		if (ifp->if_ioctl == NULL)
1392 			return (EOPNOTSUPP);
1393 		IFF_LOCKGIANT(ifp);
1394 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1395 		IFF_UNLOCKGIANT(ifp);
1396 		if (error == 0) {
1397 			getmicrotime(&ifp->if_lastchange);
1398 			rt_ifmsg(ifp);
1399 		}
1400 		/*
1401 		 * If the link MTU changed, do network layer specific procedure.
1402 		 */
1403 		if (ifp->if_mtu != oldmtu) {
1404 #ifdef INET6
1405 			nd6_setmtu(ifp);
1406 #endif
1407 		}
1408 		break;
1409 	}
1410 
1411 	case SIOCADDMULTI:
1412 	case SIOCDELMULTI:
1413 		error = suser(td);
1414 		if (error)
1415 			return (error);
1416 
1417 		/* Don't allow group membership on non-multicast interfaces. */
1418 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1419 			return (EOPNOTSUPP);
1420 
1421 		/* Don't let users screw up protocols' entries. */
1422 		if (ifr->ifr_addr.sa_family != AF_LINK)
1423 			return (EINVAL);
1424 
1425 		if (cmd == SIOCADDMULTI) {
1426 			struct ifmultiaddr *ifma;
1427 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1428 		} else {
1429 			error = if_delmulti(ifp, &ifr->ifr_addr);
1430 		}
1431 		if (error == 0)
1432 			getmicrotime(&ifp->if_lastchange);
1433 		break;
1434 
1435 	case SIOCSIFPHYADDR:
1436 	case SIOCDIFPHYADDR:
1437 #ifdef INET6
1438 	case SIOCSIFPHYADDR_IN6:
1439 #endif
1440 	case SIOCSLIFPHYADDR:
1441 	case SIOCSIFMEDIA:
1442 	case SIOCSIFGENERIC:
1443 		error = suser(td);
1444 		if (error)
1445 			return (error);
1446 		if (ifp->if_ioctl == NULL)
1447 			return (EOPNOTSUPP);
1448 		IFF_LOCKGIANT(ifp);
1449 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1450 		IFF_UNLOCKGIANT(ifp);
1451 		if (error == 0)
1452 			getmicrotime(&ifp->if_lastchange);
1453 		break;
1454 
1455 	case SIOCGIFSTATUS:
1456 		ifs = (struct ifstat *)data;
1457 		ifs->ascii[0] = '\0';
1458 
1459 	case SIOCGIFPSRCADDR:
1460 	case SIOCGIFPDSTADDR:
1461 	case SIOCGLIFPHYADDR:
1462 	case SIOCGIFMEDIA:
1463 	case SIOCGIFGENERIC:
1464 		if (ifp->if_ioctl == NULL)
1465 			return (EOPNOTSUPP);
1466 		IFF_LOCKGIANT(ifp);
1467 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1468 		IFF_UNLOCKGIANT(ifp);
1469 		break;
1470 
1471 	case SIOCSIFLLADDR:
1472 		error = suser(td);
1473 		if (error)
1474 			return (error);
1475 		error = if_setlladdr(ifp,
1476 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1477 		break;
1478 
1479 	default:
1480 		error = ENOIOCTL;
1481 		break;
1482 	}
1483 	return (error);
1484 }
1485 
1486 /*
1487  * Interface ioctls.
1488  */
1489 int
1490 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1491 {
1492 	struct ifnet *ifp;
1493 	struct ifreq *ifr;
1494 	int error;
1495 	int oif_flags;
1496 
1497 	switch (cmd) {
1498 	case SIOCGIFCONF:
1499 	case OSIOCGIFCONF:
1500 #ifdef __amd64__
1501 	case SIOCGIFCONF32:
1502 #endif
1503 		return (ifconf(cmd, data));
1504 	}
1505 	ifr = (struct ifreq *)data;
1506 
1507 	switch (cmd) {
1508 	case SIOCIFCREATE:
1509 	case SIOCIFDESTROY:
1510 		if ((error = suser(td)) != 0)
1511 			return (error);
1512 		return ((cmd == SIOCIFCREATE) ?
1513 			if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1514 			if_clone_destroy(ifr->ifr_name));
1515 
1516 	case SIOCIFGCLONERS:
1517 		return (if_clone_list((struct if_clonereq *)data));
1518 	}
1519 
1520 	ifp = ifunit(ifr->ifr_name);
1521 	if (ifp == 0)
1522 		return (ENXIO);
1523 
1524 	error = ifhwioctl(cmd, ifp, data, td);
1525 	if (error != ENOIOCTL)
1526 		return (error);
1527 
1528 	oif_flags = ifp->if_flags;
1529 	if (so->so_proto == 0)
1530 		return (EOPNOTSUPP);
1531 #ifndef COMPAT_43
1532 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1533 								 data,
1534 								 ifp, td));
1535 #else
1536 	{
1537 		int ocmd = cmd;
1538 
1539 		switch (cmd) {
1540 
1541 		case SIOCSIFDSTADDR:
1542 		case SIOCSIFADDR:
1543 		case SIOCSIFBRDADDR:
1544 		case SIOCSIFNETMASK:
1545 #if BYTE_ORDER != BIG_ENDIAN
1546 			if (ifr->ifr_addr.sa_family == 0 &&
1547 			    ifr->ifr_addr.sa_len < 16) {
1548 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1549 				ifr->ifr_addr.sa_len = 16;
1550 			}
1551 #else
1552 			if (ifr->ifr_addr.sa_len == 0)
1553 				ifr->ifr_addr.sa_len = 16;
1554 #endif
1555 			break;
1556 
1557 		case OSIOCGIFADDR:
1558 			cmd = SIOCGIFADDR;
1559 			break;
1560 
1561 		case OSIOCGIFDSTADDR:
1562 			cmd = SIOCGIFDSTADDR;
1563 			break;
1564 
1565 		case OSIOCGIFBRDADDR:
1566 			cmd = SIOCGIFBRDADDR;
1567 			break;
1568 
1569 		case OSIOCGIFNETMASK:
1570 			cmd = SIOCGIFNETMASK;
1571 		}
1572 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
1573 								   cmd,
1574 								   data,
1575 								   ifp, td));
1576 		switch (ocmd) {
1577 
1578 		case OSIOCGIFADDR:
1579 		case OSIOCGIFDSTADDR:
1580 		case OSIOCGIFBRDADDR:
1581 		case OSIOCGIFNETMASK:
1582 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1583 
1584 		}
1585 	}
1586 #endif /* COMPAT_43 */
1587 
1588 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1589 #ifdef INET6
1590 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
1591 		if (ifp->if_flags & IFF_UP) {
1592 			int s = splimp();
1593 			in6_if_up(ifp);
1594 			splx(s);
1595 		}
1596 #endif
1597 	}
1598 	return (error);
1599 }
1600 
1601 /*
1602  * The code common to handling reference counted flags,
1603  * e.g., in ifpromisc() and if_allmulti().
1604  * The "pflag" argument can specify a permanent mode flag to check,
1605  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
1606  *
1607  * Only to be used on stack-owned flags, not driver-owned flags.
1608  */
1609 static int
1610 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
1611 {
1612 	struct ifreq ifr;
1613 	int error;
1614 	int oldflags, oldcount;
1615 
1616 	/* Sanity checks to catch programming errors */
1617 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
1618 	    ("%s: setting driver-owned flag %d", __func__, flag));
1619 
1620 	if (onswitch)
1621 		KASSERT(*refcount >= 0,
1622 		    ("%s: increment negative refcount %d for flag %d",
1623 		    __func__, *refcount, flag));
1624 	else
1625 		KASSERT(*refcount > 0,
1626 		    ("%s: decrement non-positive refcount %d for flag %d",
1627 		    __func__, *refcount, flag));
1628 
1629 	/* In case this mode is permanent, just touch refcount */
1630 	if (ifp->if_flags & pflag) {
1631 		*refcount += onswitch ? 1 : -1;
1632 		return (0);
1633 	}
1634 
1635 	/* Save ifnet parameters for if_ioctl() may fail */
1636 	oldcount = *refcount;
1637 	oldflags = ifp->if_flags;
1638 
1639 	/*
1640 	 * See if we aren't the only and touching refcount is enough.
1641 	 * Actually toggle interface flag if we are the first or last.
1642 	 */
1643 	if (onswitch) {
1644 		if ((*refcount)++)
1645 			return (0);
1646 		ifp->if_flags |= flag;
1647 	} else {
1648 		if (--(*refcount))
1649 			return (0);
1650 		ifp->if_flags &= ~flag;
1651 	}
1652 
1653 	/* Call down the driver since we've changed interface flags */
1654 	if (ifp->if_ioctl == NULL) {
1655 		error = EOPNOTSUPP;
1656 		goto recover;
1657 	}
1658 	ifr.ifr_flags = ifp->if_flags & 0xffff;
1659 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1660 	IFF_LOCKGIANT(ifp);
1661 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1662 	IFF_UNLOCKGIANT(ifp);
1663 	if (error)
1664 		goto recover;
1665 	/* Notify userland that interface flags have changed */
1666 	rt_ifmsg(ifp);
1667 	return (0);
1668 
1669 recover:
1670 	/* Recover after driver error */
1671 	*refcount = oldcount;
1672 	ifp->if_flags = oldflags;
1673 	return (error);
1674 }
1675 
1676 /*
1677  * Set/clear promiscuous mode on interface ifp based on the truth value
1678  * of pswitch.  The calls are reference counted so that only the first
1679  * "on" request actually has an effect, as does the final "off" request.
1680  * Results are undefined if the "off" and "on" requests are not matched.
1681  */
1682 int
1683 ifpromisc(struct ifnet *ifp, int pswitch)
1684 {
1685 	int error;
1686 	int oldflags = ifp->if_flags;
1687 
1688 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
1689 			   &ifp->if_pcount, pswitch);
1690 	/* If promiscuous mode status has changed, log a message */
1691 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
1692 		log(LOG_INFO, "%s: promiscuous mode %s\n",
1693 		    ifp->if_xname,
1694 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
1695 	return (error);
1696 }
1697 
1698 /*
1699  * Return interface configuration
1700  * of system.  List may be used
1701  * in later ioctl's (above) to get
1702  * other information.
1703  */
1704 /*ARGSUSED*/
1705 static int
1706 ifconf(u_long cmd, caddr_t data)
1707 {
1708 	struct ifconf *ifc = (struct ifconf *)data;
1709 #ifdef __amd64__
1710 	struct ifconf32 *ifc32 = (struct ifconf32 *)data;
1711 	struct ifconf ifc_swab;
1712 #endif
1713 	struct ifnet *ifp;
1714 	struct ifaddr *ifa;
1715 	struct ifreq ifr;
1716 	struct sbuf *sb;
1717 	int error, full = 0, valid_len, max_len;
1718 
1719 #ifdef __amd64__
1720 	if (cmd == SIOCGIFCONF32) {
1721 		ifc_swab.ifc_len = ifc32->ifc_len;
1722 		ifc_swab.ifc_buf = (caddr_t)(uintptr_t)ifc32->ifc_buf;
1723 		ifc = &ifc_swab;
1724 	}
1725 #endif
1726 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
1727 	max_len = MAXPHYS - 1;
1728 
1729 	/* Prevent hostile input from being able to crash the system */
1730 	if (ifc->ifc_len <= 0)
1731 		return (EINVAL);
1732 
1733 again:
1734 	if (ifc->ifc_len <= max_len) {
1735 		max_len = ifc->ifc_len;
1736 		full = 1;
1737 	}
1738 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
1739 	max_len = 0;
1740 	valid_len = 0;
1741 
1742 	IFNET_RLOCK();		/* could sleep XXX */
1743 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1744 		int addrs;
1745 
1746 		/*
1747 		 * Zero the ifr_name buffer to make sure we don't
1748 		 * disclose the contents of the stack.
1749 		 */
1750 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
1751 
1752 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1753 		    >= sizeof(ifr.ifr_name)) {
1754 			sbuf_delete(sb);
1755 			IFNET_RUNLOCK();
1756 			return (ENAMETOOLONG);
1757 		}
1758 
1759 		addrs = 0;
1760 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1761 			struct sockaddr *sa = ifa->ifa_addr;
1762 
1763 			if (jailed(curthread->td_ucred) &&
1764 			    prison_if(curthread->td_ucred, sa))
1765 				continue;
1766 			addrs++;
1767 #ifdef COMPAT_43
1768 			if (cmd == OSIOCGIFCONF) {
1769 				struct osockaddr *osa =
1770 					 (struct osockaddr *)&ifr.ifr_addr;
1771 				ifr.ifr_addr = *sa;
1772 				osa->sa_family = sa->sa_family;
1773 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1774 				max_len += sizeof(ifr);
1775 			} else
1776 #endif
1777 			if (sa->sa_len <= sizeof(*sa)) {
1778 				ifr.ifr_addr = *sa;
1779 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1780 				max_len += sizeof(ifr);
1781 			} else {
1782 				sbuf_bcat(sb, &ifr,
1783 				    offsetof(struct ifreq, ifr_addr));
1784 				max_len += offsetof(struct ifreq, ifr_addr);
1785 				sbuf_bcat(sb, sa, sa->sa_len);
1786 				max_len += sa->sa_len;
1787 			}
1788 
1789 			if (!sbuf_overflowed(sb))
1790 				valid_len = sbuf_len(sb);
1791 		}
1792 		if (addrs == 0) {
1793 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
1794 			sbuf_bcat(sb, &ifr, sizeof(ifr));
1795 			max_len += sizeof(ifr);
1796 
1797 			if (!sbuf_overflowed(sb))
1798 				valid_len = sbuf_len(sb);
1799 		}
1800 	}
1801 	IFNET_RUNLOCK();
1802 
1803 	/*
1804 	 * If we didn't allocate enough space (uncommon), try again.  If
1805 	 * we have already allocated as much space as we are allowed,
1806 	 * return what we've got.
1807 	 */
1808 	if (valid_len != max_len && !full) {
1809 		sbuf_delete(sb);
1810 		goto again;
1811 	}
1812 
1813 	ifc->ifc_len = valid_len;
1814 #ifdef __amd64__
1815 	if (cmd == SIOCGIFCONF32)
1816 		ifc32->ifc_len = valid_len;
1817 #endif
1818 	sbuf_finish(sb);
1819 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
1820 	sbuf_delete(sb);
1821 	return (error);
1822 }
1823 
1824 /*
1825  * Just like ifpromisc(), but for all-multicast-reception mode.
1826  */
1827 int
1828 if_allmulti(struct ifnet *ifp, int onswitch)
1829 {
1830 
1831 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
1832 }
1833 
1834 static struct ifmultiaddr *
1835 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
1836 {
1837 	struct ifmultiaddr *ifma;
1838 
1839 	IF_ADDR_LOCK_ASSERT(ifp);
1840 
1841 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1842 		if (sa_equal(ifma->ifma_addr, sa))
1843 			break;
1844 	}
1845 
1846 	return ifma;
1847 }
1848 
1849 /*
1850  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
1851  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
1852  * the ifnet multicast address list here, so the caller must do that and
1853  * other setup work (such as notifying the device driver).  The reference
1854  * count is initialized to 1.
1855  */
1856 static struct ifmultiaddr *
1857 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
1858     int mflags)
1859 {
1860 	struct ifmultiaddr *ifma;
1861 	struct sockaddr *dupsa;
1862 
1863 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, mflags |
1864 	    M_ZERO);
1865 	if (ifma == NULL)
1866 		return (NULL);
1867 
1868 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, mflags);
1869 	if (dupsa == NULL) {
1870 		FREE(ifma, M_IFMADDR);
1871 		return (NULL);
1872 	}
1873 	bcopy(sa, dupsa, sa->sa_len);
1874 	ifma->ifma_addr = dupsa;
1875 
1876 	ifma->ifma_ifp = ifp;
1877 	ifma->ifma_refcount = 1;
1878 	ifma->ifma_protospec = NULL;
1879 
1880 	if (llsa == NULL) {
1881 		ifma->ifma_lladdr = NULL;
1882 		return (ifma);
1883 	}
1884 
1885 	MALLOC(dupsa, struct sockaddr *, llsa->sa_len, M_IFMADDR, mflags);
1886 	if (dupsa == NULL) {
1887 		FREE(ifma->ifma_addr, M_IFMADDR);
1888 		FREE(ifma, M_IFMADDR);
1889 		return (NULL);
1890 	}
1891 	bcopy(llsa, dupsa, llsa->sa_len);
1892 	ifma->ifma_lladdr = dupsa;
1893 
1894 	return (ifma);
1895 }
1896 
1897 /*
1898  * if_freemulti: free ifmultiaddr structure and possibly attached related
1899  * addresses.  The caller is responsible for implementing reference
1900  * counting, notifying the driver, handling routing messages, and releasing
1901  * any dependent link layer state.
1902  */
1903 static void
1904 if_freemulti(struct ifmultiaddr *ifma)
1905 {
1906 
1907 	KASSERT(ifma->ifma_refcount == 1, ("if_freemulti: refcount %d",
1908 	    ifma->ifma_refcount));
1909 	KASSERT(ifma->ifma_protospec == NULL,
1910 	    ("if_freemulti: protospec not NULL"));
1911 
1912 	if (ifma->ifma_lladdr != NULL)
1913 		FREE(ifma->ifma_lladdr, M_IFMADDR);
1914 	FREE(ifma->ifma_addr, M_IFMADDR);
1915 	FREE(ifma, M_IFMADDR);
1916 }
1917 
1918 /*
1919  * Register an additional multicast address with a network interface.
1920  *
1921  * - If the address is already present, bump the reference count on the
1922  *   address and return.
1923  * - If the address is not link-layer, look up a link layer address.
1924  * - Allocate address structures for one or both addresses, and attach to the
1925  *   multicast address list on the interface.  If automatically adding a link
1926  *   layer address, the protocol address will own a reference to the link
1927  *   layer address, to be freed when it is freed.
1928  * - Notify the network device driver of an addition to the multicast address
1929  *   list.
1930  *
1931  * 'sa' points to caller-owned memory with the desired multicast address.
1932  *
1933  * 'retifma' will be used to return a pointer to the resulting multicast
1934  * address reference, if desired.
1935  */
1936 int
1937 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
1938     struct ifmultiaddr **retifma)
1939 {
1940 	struct ifmultiaddr *ifma, *ll_ifma;
1941 	struct sockaddr *llsa;
1942 	int error;
1943 
1944 	/*
1945 	 * If the address is already present, return a new reference to it;
1946 	 * otherwise, allocate storage and set up a new address.
1947 	 */
1948 	IF_ADDR_LOCK(ifp);
1949 	ifma = if_findmulti(ifp, sa);
1950 	if (ifma != NULL) {
1951 		ifma->ifma_refcount++;
1952 		if (retifma != NULL)
1953 			*retifma = ifma;
1954 		IF_ADDR_UNLOCK(ifp);
1955 		return (0);
1956 	}
1957 
1958 	/*
1959 	 * The address isn't already present; resolve the protocol address
1960 	 * into a link layer address, and then look that up, bump its
1961 	 * refcount or allocate an ifma for that also.  If 'llsa' was
1962 	 * returned, we will need to free it later.
1963 	 */
1964 	llsa = NULL;
1965 	ll_ifma = NULL;
1966 	if (ifp->if_resolvemulti != NULL) {
1967 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
1968 		if (error)
1969 			goto unlock_out;
1970 	}
1971 
1972 	/*
1973 	 * Allocate the new address.  Don't hook it up yet, as we may also
1974 	 * need to allocate a link layer multicast address.
1975 	 */
1976 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
1977 	if (ifma == NULL) {
1978 		error = ENOMEM;
1979 		goto free_llsa_out;
1980 	}
1981 
1982 	/*
1983 	 * If a link layer address is found, we'll need to see if it's
1984 	 * already present in the address list, or allocate is as well.
1985 	 * When this block finishes, the link layer address will be on the
1986 	 * list.
1987 	 */
1988 	if (llsa != NULL) {
1989 		ll_ifma = if_findmulti(ifp, llsa);
1990 		if (ll_ifma == NULL) {
1991 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
1992 			if (ll_ifma == NULL) {
1993 				if_freemulti(ifma);
1994 				error = ENOMEM;
1995 				goto free_llsa_out;
1996 			}
1997 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
1998 			    ifma_link);
1999 		} else
2000 			ll_ifma->ifma_refcount++;
2001 	}
2002 
2003 	/*
2004 	 * We now have a new multicast address, ifma, and possibly a new or
2005 	 * referenced link layer address.  Add the primary address to the
2006 	 * ifnet address list.
2007 	 */
2008 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2009 
2010 	if (retifma != NULL)
2011 		*retifma = ifma;
2012 
2013 	/*
2014 	 * Must generate the message while holding the lock so that 'ifma'
2015 	 * pointer is still valid.
2016 	 *
2017 	 * XXXRW: How come we don't announce ll_ifma?
2018 	 */
2019 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2020 	IF_ADDR_UNLOCK(ifp);
2021 
2022 	/*
2023 	 * We are certain we have added something, so call down to the
2024 	 * interface to let them know about it.
2025 	 */
2026 	if (ifp->if_ioctl != NULL) {
2027 		IFF_LOCKGIANT(ifp);
2028 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2029 		IFF_UNLOCKGIANT(ifp);
2030 	}
2031 
2032 	if (llsa != NULL)
2033 		FREE(llsa, M_IFMADDR);
2034 
2035 	return (0);
2036 
2037 free_llsa_out:
2038 	if (llsa != NULL)
2039 		FREE(llsa, M_IFMADDR);
2040 
2041 unlock_out:
2042 	IF_ADDR_UNLOCK(ifp);
2043 	return (error);
2044 }
2045 
2046 /*
2047  * Remove a reference to a multicast address on this interface.  Yell
2048  * if the request does not match an existing membership.
2049  */
2050 int
2051 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2052 {
2053 	struct ifmultiaddr *ifma, *ll_ifma;
2054 
2055 	IF_ADDR_LOCK(ifp);
2056 	ifma = if_findmulti(ifp, sa);
2057 	if (ifma == NULL) {
2058 		IF_ADDR_UNLOCK(ifp);
2059 		return ENOENT;
2060 	}
2061 
2062 	if (ifma->ifma_refcount > 1) {
2063 		ifma->ifma_refcount--;
2064 		IF_ADDR_UNLOCK(ifp);
2065 		return 0;
2066 	}
2067 
2068 	sa = ifma->ifma_lladdr;
2069 	if (sa != NULL)
2070 		ll_ifma = if_findmulti(ifp, sa);
2071 	else
2072 		ll_ifma = NULL;
2073 
2074 	/*
2075 	 * XXXRW: How come we don't announce ll_ifma?
2076 	 */
2077 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2078 
2079 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2080 	if_freemulti(ifma);
2081 
2082 	if (ll_ifma != NULL) {
2083 		if (ll_ifma->ifma_refcount == 1) {
2084 			TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifma_link);
2085 			if_freemulti(ll_ifma);
2086 		} else
2087 			ll_ifma->ifma_refcount--;
2088 	}
2089 	IF_ADDR_UNLOCK(ifp);
2090 
2091 	/*
2092 	 * Make sure the interface driver is notified
2093 	 * in the case of a link layer mcast group being left.
2094 	 */
2095 	if (ifp->if_ioctl) {
2096 		IFF_LOCKGIANT(ifp);
2097 		(void) (*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2098 		IFF_UNLOCKGIANT(ifp);
2099 	}
2100 
2101 	return 0;
2102 }
2103 
2104 /*
2105  * Set the link layer address on an interface.
2106  *
2107  * At this time we only support certain types of interfaces,
2108  * and we don't allow the length of the address to change.
2109  */
2110 int
2111 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2112 {
2113 	struct sockaddr_dl *sdl;
2114 	struct ifaddr *ifa;
2115 	struct ifreq ifr;
2116 
2117 	ifa = ifp->if_addr;
2118 	if (ifa == NULL)
2119 		return (EINVAL);
2120 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2121 	if (sdl == NULL)
2122 		return (EINVAL);
2123 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2124 		return (EINVAL);
2125 	switch (ifp->if_type) {
2126 	case IFT_ETHER:
2127 	case IFT_FDDI:
2128 	case IFT_XETHER:
2129 	case IFT_ISO88025:
2130 	case IFT_L2VLAN:
2131 	case IFT_BRIDGE:
2132 	case IFT_ARCNET:
2133 		bcopy(lladdr, LLADDR(sdl), len);
2134 		break;
2135 	default:
2136 		return (ENODEV);
2137 	}
2138 	/*
2139 	 * If the interface is already up, we need
2140 	 * to re-init it in order to reprogram its
2141 	 * address filter.
2142 	 */
2143 	if ((ifp->if_flags & IFF_UP) != 0) {
2144 		if (ifp->if_ioctl) {
2145 			IFF_LOCKGIANT(ifp);
2146 			ifp->if_flags &= ~IFF_UP;
2147 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2148 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2149 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2150 			ifp->if_flags |= IFF_UP;
2151 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2152 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2153 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2154 			IFF_UNLOCKGIANT(ifp);
2155 		}
2156 #ifdef INET
2157 		/*
2158 		 * Also send gratuitous ARPs to notify other nodes about
2159 		 * the address change.
2160 		 */
2161 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2162 			if (ifa->ifa_addr != NULL &&
2163 			    ifa->ifa_addr->sa_family == AF_INET)
2164 				arp_ifinit(ifp, ifa);
2165 		}
2166 #endif
2167 	}
2168 	return (0);
2169 }
2170 
2171 /*
2172  * The name argument must be a pointer to storage which will last as
2173  * long as the interface does.  For physical devices, the result of
2174  * device_get_name(dev) is a good choice and for pseudo-devices a
2175  * static string works well.
2176  */
2177 void
2178 if_initname(struct ifnet *ifp, const char *name, int unit)
2179 {
2180 	ifp->if_dname = name;
2181 	ifp->if_dunit = unit;
2182 	if (unit != IF_DUNIT_NONE)
2183 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2184 	else
2185 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2186 }
2187 
2188 int
2189 if_printf(struct ifnet *ifp, const char * fmt, ...)
2190 {
2191 	va_list ap;
2192 	int retval;
2193 
2194 	retval = printf("%s: ", ifp->if_xname);
2195 	va_start(ap, fmt);
2196 	retval += vprintf(fmt, ap);
2197 	va_end(ap);
2198 	return (retval);
2199 }
2200 
2201 /*
2202  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2203  * be called without Giant.  However, we often can't acquire the Giant lock
2204  * at those points; instead, we run it via a task queue that holds Giant via
2205  * if_start_deferred.
2206  *
2207  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2208  * outstanding if_start_deferred() tasks that will run after the free.  This
2209  * probably means waiting in if_detach().
2210  */
2211 void
2212 if_start(struct ifnet *ifp)
2213 {
2214 
2215 	NET_ASSERT_GIANT();
2216 
2217 	if ((ifp->if_flags & IFF_NEEDSGIANT) != 0 && debug_mpsafenet != 0) {
2218 		if (mtx_owned(&Giant))
2219 			(*(ifp)->if_start)(ifp);
2220 		else
2221 			taskqueue_enqueue(taskqueue_swi_giant,
2222 			    &ifp->if_starttask);
2223 	} else
2224 		(*(ifp)->if_start)(ifp);
2225 }
2226 
2227 static void
2228 if_start_deferred(void *context, int pending)
2229 {
2230 	struct ifnet *ifp;
2231 
2232 	/*
2233 	 * This code must be entered with Giant, and should never run if
2234 	 * we're not running with debug.mpsafenet.
2235 	 */
2236 	KASSERT(debug_mpsafenet != 0, ("if_start_deferred: debug.mpsafenet"));
2237 	GIANT_REQUIRED;
2238 
2239 	ifp = context;
2240 	(ifp->if_start)(ifp);
2241 }
2242 
2243 int
2244 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2245 {
2246 	int active = 0;
2247 
2248 	IF_LOCK(ifq);
2249 	if (_IF_QFULL(ifq)) {
2250 		_IF_DROP(ifq);
2251 		IF_UNLOCK(ifq);
2252 		m_freem(m);
2253 		return (0);
2254 	}
2255 	if (ifp != NULL) {
2256 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2257 		if (m->m_flags & (M_BCAST|M_MCAST))
2258 			ifp->if_omcasts++;
2259 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2260 	}
2261 	_IF_ENQUEUE(ifq, m);
2262 	IF_UNLOCK(ifq);
2263 	if (ifp != NULL && !active)
2264 		if_start(ifp);
2265 	return (1);
2266 }
2267 
2268 void
2269 if_register_com_alloc(u_char type,
2270     if_com_alloc_t *a, if_com_free_t *f)
2271 {
2272 
2273 	KASSERT(if_com_alloc[type] == NULL,
2274 	    ("if_register_com_alloc: %d already registered", type));
2275 	KASSERT(if_com_free[type] == NULL,
2276 	    ("if_register_com_alloc: %d free already registered", type));
2277 
2278 	if_com_alloc[type] = a;
2279 	if_com_free[type] = f;
2280 }
2281 
2282 void
2283 if_deregister_com_alloc(u_char type)
2284 {
2285 
2286 	KASSERT(if_com_alloc[type] == NULL,
2287 	    ("if_deregister_com_alloc: %d not registered", type));
2288 	KASSERT(if_com_free[type] == NULL,
2289 	    ("if_deregister_com_alloc: %d free not registered", type));
2290 	if_com_alloc[type] = NULL;
2291 	if_com_free[type] = NULL;
2292 }
2293