xref: /freebsd/sys/net/if.c (revision a9148abd9da5db2f1c682fb17bed791845fc41c9)
1 /*-
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.5 (Berkeley) 1/9/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 #include "opt_mac.h"
37 #include "opt_carp.h"
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/conf.h>
42 #include <sys/malloc.h>
43 #include <sys/sbuf.h>
44 #include <sys/bus.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/protosw.h>
52 #include <sys/kernel.h>
53 #include <sys/sockio.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
56 #include <sys/taskqueue.h>
57 #include <sys/domain.h>
58 #include <sys/jail.h>
59 #include <sys/vimage.h>
60 #include <machine/stdarg.h>
61 
62 #include <net/if.h>
63 #include <net/if_clone.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/if_var.h>
67 #include <net/radix.h>
68 #include <net/route.h>
69 
70 #if defined(INET) || defined(INET6)
71 /*XXX*/
72 #include <netinet/in.h>
73 #include <netinet/in_var.h>
74 #ifdef INET6
75 #include <netinet6/in6_var.h>
76 #include <netinet6/in6_ifattach.h>
77 #endif
78 #endif
79 #ifdef INET
80 #include <netinet/if_ether.h>
81 #endif
82 #ifdef DEV_CARP
83 #include <netinet/ip_carp.h>
84 #endif
85 
86 #include <security/mac/mac_framework.h>
87 
88 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
89 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
90 
91 /* Log link state change events */
92 static int log_link_state_change = 1;
93 
94 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
95 	&log_link_state_change, 0,
96 	"log interface link state change events");
97 
98 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
99 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
100 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
101 
102 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
103 
104 /*
105  * XXX: Style; these should be sorted alphabetically, and unprototyped
106  * static functions should be prototyped. Currently they are sorted by
107  * declaration order.
108  */
109 static void	if_attachdomain(void *);
110 static void	if_attachdomain1(struct ifnet *);
111 static int	ifconf(u_long, caddr_t);
112 static void	if_freemulti(struct ifmultiaddr *);
113 static void	if_grow(void);
114 static void	if_init(void *);
115 static void	if_qflush(struct ifaltq *);
116 static void	if_route(struct ifnet *, int flag, int fam);
117 static int	if_setflag(struct ifnet *, int, int, int *, int);
118 static void	if_slowtimo(void *);
119 static void	if_unroute(struct ifnet *, int flag, int fam);
120 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
121 static int	if_rtdel(struct radix_node *, void *);
122 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
123 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
124 static void	if_start_deferred(void *context, int pending);
125 static void	do_link_state_change(void *, int);
126 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
127 static int	if_getgroupmembers(struct ifgroupreq *);
128 #ifdef INET6
129 /*
130  * XXX: declare here to avoid to include many inet6 related files..
131  * should be more generalized?
132  */
133 extern void	nd6_setmtu(struct ifnet *);
134 #endif
135 
136 int	if_index = 0;
137 int	ifqmaxlen = IFQ_MAXLEN;
138 struct	ifnethead ifnet;	/* depend on static init XXX */
139 struct	ifgrouphead ifg_head;
140 struct	mtx ifnet_lock;
141 static	if_com_alloc_t *if_com_alloc[256];
142 static	if_com_free_t *if_com_free[256];
143 
144 static int	if_indexlim = 8;
145 static struct	knlist ifklist;
146 
147 /*
148  * Table of ifnet/cdev by index.  Locked with ifnet_lock.
149  */
150 static struct ifindex_entry *ifindex_table = NULL;
151 
152 static void	filt_netdetach(struct knote *kn);
153 static int	filt_netdev(struct knote *kn, long hint);
154 
155 static struct filterops netdev_filtops =
156     { 1, NULL, filt_netdetach, filt_netdev };
157 
158 /*
159  * System initialization
160  */
161 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
162 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_slowtimo, NULL);
163 
164 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
165 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
166 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
167 
168 struct ifnet *
169 ifnet_byindex(u_short idx)
170 {
171 	INIT_VNET_NET(curvnet);
172 	struct ifnet *ifp;
173 
174 	IFNET_RLOCK();
175 	ifp = V_ifindex_table[idx].ife_ifnet;
176 	IFNET_RUNLOCK();
177 	return (ifp);
178 }
179 
180 static void
181 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
182 {
183 	INIT_VNET_NET(curvnet);
184 
185 	IFNET_WLOCK_ASSERT();
186 
187 	V_ifindex_table[idx].ife_ifnet = ifp;
188 }
189 
190 struct ifaddr *
191 ifaddr_byindex(u_short idx)
192 {
193 	INIT_VNET_NET(curvnet);
194 	struct ifaddr *ifa;
195 
196 	IFNET_RLOCK();
197 	ifa = ifnet_byindex(idx)->if_addr;
198 	IFNET_RUNLOCK();
199 	return (ifa);
200 }
201 
202 struct cdev *
203 ifdev_byindex(u_short idx)
204 {
205 	INIT_VNET_NET(curvnet);
206 	struct cdev *cdev;
207 
208 	IFNET_RLOCK();
209 	cdev = V_ifindex_table[idx].ife_dev;
210 	IFNET_RUNLOCK();
211 	return (cdev);
212 }
213 
214 static void
215 ifdev_setbyindex(u_short idx, struct cdev *cdev)
216 {
217 	INIT_VNET_NET(curvnet);
218 
219 	IFNET_WLOCK();
220 	V_ifindex_table[idx].ife_dev = cdev;
221 	IFNET_WUNLOCK();
222 }
223 
224 static d_open_t		netopen;
225 static d_close_t	netclose;
226 static d_ioctl_t	netioctl;
227 static d_kqfilter_t	netkqfilter;
228 
229 static struct cdevsw net_cdevsw = {
230 	.d_version =	D_VERSION,
231 	.d_flags =	D_NEEDGIANT,
232 	.d_open =	netopen,
233 	.d_close =	netclose,
234 	.d_ioctl =	netioctl,
235 	.d_name =	"net",
236 	.d_kqfilter =	netkqfilter,
237 };
238 
239 static int
240 netopen(struct cdev *dev, int flag, int mode, struct thread *td)
241 {
242 	return (0);
243 }
244 
245 static int
246 netclose(struct cdev *dev, int flags, int fmt, struct thread *td)
247 {
248 	return (0);
249 }
250 
251 static int
252 netioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
253 {
254 	struct ifnet *ifp;
255 	int error, idx;
256 
257 	/* only support interface specific ioctls */
258 	if (IOCGROUP(cmd) != 'i')
259 		return (EOPNOTSUPP);
260 	idx = dev2unit(dev);
261 	if (idx == 0) {
262 		/*
263 		 * special network device, not interface.
264 		 */
265 		if (cmd == SIOCGIFCONF)
266 			return (ifconf(cmd, data));	/* XXX remove cmd */
267 #ifdef __amd64__
268 		if (cmd == SIOCGIFCONF32)
269 			return (ifconf(cmd, data));	/* XXX remove cmd */
270 #endif
271 		return (EOPNOTSUPP);
272 	}
273 
274 	ifp = ifnet_byindex(idx);
275 	if (ifp == NULL)
276 		return (ENXIO);
277 
278 	error = ifhwioctl(cmd, ifp, data, td);
279 	if (error == ENOIOCTL)
280 		error = EOPNOTSUPP;
281 	return (error);
282 }
283 
284 static int
285 netkqfilter(struct cdev *dev, struct knote *kn)
286 {
287 	INIT_VNET_NET(curvnet);
288 	struct knlist *klist;
289 	struct ifnet *ifp;
290 	int idx;
291 
292 	switch (kn->kn_filter) {
293 	case EVFILT_NETDEV:
294 		kn->kn_fop = &netdev_filtops;
295 		break;
296 	default:
297 		return (EINVAL);
298 	}
299 
300 	idx = dev2unit(dev);
301 	if (idx == 0) {
302 		klist = &V_ifklist;
303 	} else {
304 		ifp = ifnet_byindex(idx);
305 		if (ifp == NULL)
306 			return (1);
307 		klist = &ifp->if_klist;
308 	}
309 
310 	kn->kn_hook = (caddr_t)klist;
311 
312 	knlist_add(klist, kn, 0);
313 
314 	return (0);
315 }
316 
317 static void
318 filt_netdetach(struct knote *kn)
319 {
320 	struct knlist *klist = (struct knlist *)kn->kn_hook;
321 
322 	knlist_remove(klist, kn, 0);
323 }
324 
325 static int
326 filt_netdev(struct knote *kn, long hint)
327 {
328 	struct knlist *klist = (struct knlist *)kn->kn_hook;
329 
330 	/*
331 	 * Currently NOTE_EXIT is abused to indicate device detach.
332 	 */
333 	if (hint == NOTE_EXIT) {
334 		kn->kn_data = NOTE_LINKINV;
335 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
336 		knlist_remove_inevent(klist, kn);
337 		return (1);
338 	}
339 	if (hint != 0)
340 		kn->kn_data = hint;			/* current status */
341 	if (kn->kn_sfflags & hint)
342 		kn->kn_fflags |= hint;
343 	return (kn->kn_fflags != 0);
344 }
345 
346 /*
347  * Network interface utility routines.
348  *
349  * Routines with ifa_ifwith* names take sockaddr *'s as
350  * parameters.
351  */
352 
353 /* ARGSUSED*/
354 static void
355 if_init(void *dummy __unused)
356 {
357 	INIT_VNET_NET(curvnet);
358 
359 	IFNET_LOCK_INIT();
360 	TAILQ_INIT(&V_ifnet);
361 	TAILQ_INIT(&V_ifg_head);
362 	knlist_init(&V_ifklist, NULL, NULL, NULL, NULL);
363 	if_grow();				/* create initial table */
364 	ifdev_setbyindex(0, make_dev(&net_cdevsw, 0, UID_ROOT, GID_WHEEL,
365 	    0600, "network"));
366 	if_clone_init();
367 }
368 
369 static void
370 if_grow(void)
371 {
372 	INIT_VNET_NET(curvnet);
373 	u_int n;
374 	struct ifindex_entry *e;
375 
376 	V_if_indexlim <<= 1;
377 	n = V_if_indexlim * sizeof(*e);
378 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
379 	if (V_ifindex_table != NULL) {
380 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
381 		free((caddr_t)V_ifindex_table, M_IFNET);
382 	}
383 	V_ifindex_table = e;
384 }
385 
386 /*
387  * Allocate a struct ifnet and an index for an interface.  A layer 2
388  * common structure will also be allocated if an allocation routine is
389  * registered for the passed type.
390  */
391 struct ifnet*
392 if_alloc(u_char type)
393 {
394 	INIT_VNET_NET(curvnet);
395 	struct ifnet *ifp;
396 
397 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
398 
399 	/*
400 	 * Try to find an empty slot below if_index.  If we fail, take
401 	 * the next slot.
402 	 *
403 	 * XXX: should be locked!
404 	 */
405 	for (ifp->if_index = 1; ifp->if_index <= V_if_index; ifp->if_index++) {
406 		if (ifnet_byindex(ifp->if_index) == NULL)
407 			break;
408 	}
409 	/* Catch if_index overflow. */
410 	if (ifp->if_index < 1) {
411 		free(ifp, M_IFNET);
412 		return (NULL);
413 	}
414 	if (ifp->if_index > V_if_index)
415 		V_if_index = ifp->if_index;
416 	if (V_if_index >= V_if_indexlim)
417 		if_grow();
418 
419 	ifp->if_type = type;
420 
421 	if (if_com_alloc[type] != NULL) {
422 		ifp->if_l2com = if_com_alloc[type](type, ifp);
423 		if (ifp->if_l2com == NULL) {
424 			free(ifp, M_IFNET);
425 			return (NULL);
426 		}
427 	}
428 	IFNET_WLOCK();
429 	ifnet_setbyindex(ifp->if_index, ifp);
430 	IFNET_WUNLOCK();
431 	IF_ADDR_LOCK_INIT(ifp);
432 
433 	return (ifp);
434 }
435 
436 /*
437  * Free the struct ifnet, the associated index, and the layer 2 common
438  * structure if needed.  All the work is done in if_free_type().
439  *
440  * Do not add code to this function!  Add it to if_free_type().
441  */
442 void
443 if_free(struct ifnet *ifp)
444 {
445 
446 	if_free_type(ifp, ifp->if_type);
447 }
448 
449 /*
450  * Do the actual work of freeing a struct ifnet, associated index, and
451  * layer 2 common structure.  This version should only be called by
452  * intefaces that switch their type after calling if_alloc().
453  */
454 void
455 if_free_type(struct ifnet *ifp, u_char type)
456 {
457 	INIT_VNET_NET(curvnet); /* ifp->if_vnet can be NULL here ! */
458 
459 	if (ifp != ifnet_byindex(ifp->if_index)) {
460 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
461 		    __func__);
462 		return;
463 	}
464 
465 	IFNET_WLOCK();
466 	ifnet_setbyindex(ifp->if_index, NULL);
467 
468 	/* XXX: should be locked with if_findindex() */
469 	while (V_if_index > 0 && ifnet_byindex(V_if_index) == NULL)
470 		V_if_index--;
471 	IFNET_WUNLOCK();
472 
473 	if (if_com_free[type] != NULL)
474 		if_com_free[type](ifp->if_l2com, type);
475 
476 	IF_ADDR_LOCK_DESTROY(ifp);
477 	free(ifp, M_IFNET);
478 };
479 
480 /*
481  * Perform generic interface initalization tasks and attach the interface
482  * to the list of "active" interfaces.
483  *
484  * XXX:
485  *  - The decision to return void and thus require this function to
486  *    succeed is questionable.
487  *  - We do more initialization here then is probably a good idea.
488  *    Some of this should probably move to if_alloc().
489  *  - We should probably do more sanity checking.  For instance we don't
490  *    do anything to insure if_xname is unique or non-empty.
491  */
492 void
493 if_attach(struct ifnet *ifp)
494 {
495 	INIT_VNET_NET(curvnet);
496 	unsigned socksize, ifasize;
497 	int namelen, masklen;
498 	struct sockaddr_dl *sdl;
499 	struct ifaddr *ifa;
500 
501 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
502 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
503 		    ifp->if_xname);
504 
505 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
506 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
507 	IF_AFDATA_LOCK_INIT(ifp);
508 	ifp->if_afdata_initialized = 0;
509 
510 	TAILQ_INIT(&ifp->if_addrhead);
511 	TAILQ_INIT(&ifp->if_prefixhead);
512 	TAILQ_INIT(&ifp->if_multiaddrs);
513 	TAILQ_INIT(&ifp->if_groups);
514 
515 	if_addgroup(ifp, IFG_ALL);
516 
517 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
518 	getmicrotime(&ifp->if_lastchange);
519 	ifp->if_data.ifi_epoch = time_uptime;
520 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
521 
522 #ifdef MAC
523 	mac_ifnet_init(ifp);
524 	mac_ifnet_create(ifp);
525 #endif
526 
527 	ifdev_setbyindex(ifp->if_index, make_dev(&net_cdevsw,
528 	    ifp->if_index, UID_ROOT, GID_WHEEL, 0600, "%s/%s",
529 	    net_cdevsw.d_name, ifp->if_xname));
530 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
531 	    net_cdevsw.d_name, ifp->if_index);
532 
533 	mtx_init(&ifp->if_snd.ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
534 
535 	/*
536 	 * create a Link Level name for this device
537 	 */
538 	namelen = strlen(ifp->if_xname);
539 	/*
540 	 * Always save enough space for any possiable name so we can do
541 	 * a rename in place later.
542 	 */
543 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
544 	socksize = masklen + ifp->if_addrlen;
545 	if (socksize < sizeof(*sdl))
546 		socksize = sizeof(*sdl);
547 	socksize = roundup2(socksize, sizeof(long));
548 	ifasize = sizeof(*ifa) + 2 * socksize;
549 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
550 	IFA_LOCK_INIT(ifa);
551 	sdl = (struct sockaddr_dl *)(ifa + 1);
552 	sdl->sdl_len = socksize;
553 	sdl->sdl_family = AF_LINK;
554 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
555 	sdl->sdl_nlen = namelen;
556 	sdl->sdl_index = ifp->if_index;
557 	sdl->sdl_type = ifp->if_type;
558 	ifp->if_addr = ifa;
559 	ifa->ifa_ifp = ifp;
560 	ifa->ifa_rtrequest = link_rtrequest;
561 	ifa->ifa_addr = (struct sockaddr *)sdl;
562 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
563 	ifa->ifa_netmask = (struct sockaddr *)sdl;
564 	sdl->sdl_len = masklen;
565 	while (namelen != 0)
566 		sdl->sdl_data[--namelen] = 0xff;
567 	ifa->ifa_refcnt = 1;
568 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
569 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
570 
571 	/*
572 	 * XXX: why do we warn about this? We're correcting it and most
573 	 * drivers just set the value the way we do.
574 	 */
575 	if (ifp->if_snd.ifq_maxlen == 0) {
576 		if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
577 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
578 	}
579 	ifp->if_snd.altq_type = 0;
580 	ifp->if_snd.altq_disc = NULL;
581 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
582 	ifp->if_snd.altq_tbr  = NULL;
583 	ifp->if_snd.altq_ifp  = ifp;
584 
585 	IFNET_WLOCK();
586 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
587 	IFNET_WUNLOCK();
588 
589 	if (domain_init_status >= 2)
590 		if_attachdomain1(ifp);
591 
592 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
593 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
594 
595 	/* Announce the interface. */
596 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
597 
598 	if (ifp->if_watchdog != NULL)
599 		if_printf(ifp,
600 		    "WARNING: using obsoleted if_watchdog interface\n");
601 	if (ifp->if_flags & IFF_NEEDSGIANT)
602 		if_printf(ifp,
603 		    "WARNING: using obsoleted IFF_NEEDSGIANT flag\n");
604 }
605 
606 static void
607 if_attachdomain(void *dummy)
608 {
609 	INIT_VNET_NET(curvnet);
610 	struct ifnet *ifp;
611 	int s;
612 
613 	s = splnet();
614 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
615 		if_attachdomain1(ifp);
616 	splx(s);
617 }
618 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
619     if_attachdomain, NULL);
620 
621 static void
622 if_attachdomain1(struct ifnet *ifp)
623 {
624 	struct domain *dp;
625 	int s;
626 
627 	s = splnet();
628 
629 	/*
630 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
631 	 * cannot lock ifp->if_afdata initialization, entirely.
632 	 */
633 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
634 		splx(s);
635 		return;
636 	}
637 	if (ifp->if_afdata_initialized >= domain_init_status) {
638 		IF_AFDATA_UNLOCK(ifp);
639 		splx(s);
640 		printf("if_attachdomain called more than once on %s\n",
641 		    ifp->if_xname);
642 		return;
643 	}
644 	ifp->if_afdata_initialized = domain_init_status;
645 	IF_AFDATA_UNLOCK(ifp);
646 
647 	/* address family dependent data region */
648 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
649 	for (dp = domains; dp; dp = dp->dom_next) {
650 		if (dp->dom_ifattach)
651 			ifp->if_afdata[dp->dom_family] =
652 			    (*dp->dom_ifattach)(ifp);
653 	}
654 
655 	splx(s);
656 }
657 
658 /*
659  * Remove any unicast or broadcast network addresses from an interface.
660  */
661 void
662 if_purgeaddrs(struct ifnet *ifp)
663 {
664 	struct ifaddr *ifa, *next;
665 
666 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
667 		if (ifa->ifa_addr->sa_family == AF_LINK)
668 			continue;
669 #ifdef INET
670 		/* XXX: Ugly!! ad hoc just for INET */
671 		if (ifa->ifa_addr->sa_family == AF_INET) {
672 			struct ifaliasreq ifr;
673 
674 			bzero(&ifr, sizeof(ifr));
675 			ifr.ifra_addr = *ifa->ifa_addr;
676 			if (ifa->ifa_dstaddr)
677 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
678 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
679 			    NULL) == 0)
680 				continue;
681 		}
682 #endif /* INET */
683 #ifdef INET6
684 		if (ifa->ifa_addr->sa_family == AF_INET6) {
685 			in6_purgeaddr(ifa);
686 			/* ifp_addrhead is already updated */
687 			continue;
688 		}
689 #endif /* INET6 */
690 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
691 		IFAFREE(ifa);
692 	}
693 }
694 
695 /*
696  * Remove any multicast network addresses from an interface.
697  */
698 void
699 if_purgemaddrs(struct ifnet *ifp)
700 {
701 	struct ifmultiaddr *ifma;
702 	struct ifmultiaddr *next;
703 
704 	IF_ADDR_LOCK(ifp);
705 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
706 		if_delmulti_locked(ifp, ifma, 1);
707 	IF_ADDR_UNLOCK(ifp);
708 }
709 
710 /*
711  * Detach an interface, removing it from the
712  * list of "active" interfaces.
713  *
714  * XXXRW: There are some significant questions about event ordering, and
715  * how to prevent things from starting to use the interface during detach.
716  */
717 void
718 if_detach(struct ifnet *ifp)
719 {
720 	INIT_VNET_NET(ifp->if_vnet);
721 	struct ifaddr *ifa;
722 	struct radix_node_head	*rnh;
723 	int s;
724 	int i;
725 	struct domain *dp;
726  	struct ifnet *iter;
727  	int found = 0;
728 
729 	IFNET_WLOCK();
730 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
731 		if (iter == ifp) {
732 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
733 			found = 1;
734 			break;
735 		}
736 	IFNET_WUNLOCK();
737 	if (!found)
738 		return;
739 
740 	/*
741 	 * Remove/wait for pending events.
742 	 */
743 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
744 
745 	/*
746 	 * Remove routes and flush queues.
747 	 */
748 	s = splnet();
749 	if_down(ifp);
750 #ifdef ALTQ
751 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
752 		altq_disable(&ifp->if_snd);
753 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
754 		altq_detach(&ifp->if_snd);
755 #endif
756 
757 	if_purgeaddrs(ifp);
758 
759 #ifdef INET
760 	in_ifdetach(ifp);
761 #endif
762 
763 #ifdef INET6
764 	/*
765 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
766 	 * before removing routing entries below, since IPv6 interface direct
767 	 * routes are expected to be removed by the IPv6-specific kernel API.
768 	 * Otherwise, the kernel will detect some inconsistency and bark it.
769 	 */
770 	in6_ifdetach(ifp);
771 #endif
772 	if_purgemaddrs(ifp);
773 
774 	/*
775 	 * Remove link ifaddr pointer and maybe decrement if_index.
776 	 * Clean up all addresses.
777 	 */
778 	ifp->if_addr = NULL;
779 	destroy_dev(ifdev_byindex(ifp->if_index));
780 	ifdev_setbyindex(ifp->if_index, NULL);
781 
782 	/* We can now free link ifaddr. */
783 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
784 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
785 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
786 		IFAFREE(ifa);
787 	}
788 
789 	/*
790 	 * Delete all remaining routes using this interface
791 	 * Unfortuneatly the only way to do this is to slog through
792 	 * the entire routing table looking for routes which point
793 	 * to this interface...oh well...
794 	 */
795 	for (i = 1; i <= AF_MAX; i++) {
796 	    int j;
797 	    for (j = 0; j < rt_numfibs; j++) {
798 		if ((rnh = V_rt_tables[j][i]) == NULL)
799 			continue;
800 		RADIX_NODE_HEAD_LOCK(rnh);
801 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
802 		RADIX_NODE_HEAD_UNLOCK(rnh);
803 	    }
804 	}
805 
806 	/* Announce that the interface is gone. */
807 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
808 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
809 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
810 
811 	IF_AFDATA_LOCK(ifp);
812 	for (dp = domains; dp; dp = dp->dom_next) {
813 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
814 			(*dp->dom_ifdetach)(ifp,
815 			    ifp->if_afdata[dp->dom_family]);
816 	}
817 	IF_AFDATA_UNLOCK(ifp);
818 
819 #ifdef MAC
820 	mac_ifnet_destroy(ifp);
821 #endif /* MAC */
822 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
823 	knlist_clear(&ifp->if_klist, 0);
824 	knlist_destroy(&ifp->if_klist);
825 	mtx_destroy(&ifp->if_snd.ifq_mtx);
826 	IF_AFDATA_DESTROY(ifp);
827 	splx(s);
828 }
829 
830 /*
831  * Add a group to an interface
832  */
833 int
834 if_addgroup(struct ifnet *ifp, const char *groupname)
835 {
836 	INIT_VNET_NET(ifp->if_vnet);
837 	struct ifg_list		*ifgl;
838 	struct ifg_group	*ifg = NULL;
839 	struct ifg_member	*ifgm;
840 
841 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
842 	    groupname[strlen(groupname) - 1] <= '9')
843 		return (EINVAL);
844 
845 	IFNET_WLOCK();
846 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
847 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
848 			IFNET_WUNLOCK();
849 			return (EEXIST);
850 		}
851 
852 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
853 	    M_NOWAIT)) == NULL) {
854 	    	IFNET_WUNLOCK();
855 		return (ENOMEM);
856 	}
857 
858 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
859 	    M_TEMP, M_NOWAIT)) == NULL) {
860 		free(ifgl, M_TEMP);
861 		IFNET_WUNLOCK();
862 		return (ENOMEM);
863 	}
864 
865 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
866 		if (!strcmp(ifg->ifg_group, groupname))
867 			break;
868 
869 	if (ifg == NULL) {
870 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
871 		    M_TEMP, M_NOWAIT)) == NULL) {
872 			free(ifgl, M_TEMP);
873 			free(ifgm, M_TEMP);
874 			IFNET_WUNLOCK();
875 			return (ENOMEM);
876 		}
877 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
878 		ifg->ifg_refcnt = 0;
879 		TAILQ_INIT(&ifg->ifg_members);
880 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
881 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
882 	}
883 
884 	ifg->ifg_refcnt++;
885 	ifgl->ifgl_group = ifg;
886 	ifgm->ifgm_ifp = ifp;
887 
888 	IF_ADDR_LOCK(ifp);
889 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
890 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
891 	IF_ADDR_UNLOCK(ifp);
892 
893 	IFNET_WUNLOCK();
894 
895 	EVENTHANDLER_INVOKE(group_change_event, groupname);
896 
897 	return (0);
898 }
899 
900 /*
901  * Remove a group from an interface
902  */
903 int
904 if_delgroup(struct ifnet *ifp, const char *groupname)
905 {
906 	INIT_VNET_NET(ifp->if_vnet);
907 	struct ifg_list		*ifgl;
908 	struct ifg_member	*ifgm;
909 
910 	IFNET_WLOCK();
911 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
912 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
913 			break;
914 	if (ifgl == NULL) {
915 		IFNET_WUNLOCK();
916 		return (ENOENT);
917 	}
918 
919 	IF_ADDR_LOCK(ifp);
920 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
921 	IF_ADDR_UNLOCK(ifp);
922 
923 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
924 		if (ifgm->ifgm_ifp == ifp)
925 			break;
926 
927 	if (ifgm != NULL) {
928 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
929 		free(ifgm, M_TEMP);
930 	}
931 
932 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
933 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
934 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
935 		free(ifgl->ifgl_group, M_TEMP);
936 	}
937 	IFNET_WUNLOCK();
938 
939 	free(ifgl, M_TEMP);
940 
941 	EVENTHANDLER_INVOKE(group_change_event, groupname);
942 
943 	return (0);
944 }
945 
946 /*
947  * Stores all groups from an interface in memory pointed
948  * to by data
949  */
950 static int
951 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
952 {
953 	int			 len, error;
954 	struct ifg_list		*ifgl;
955 	struct ifg_req		 ifgrq, *ifgp;
956 	struct ifgroupreq	*ifgr = data;
957 
958 	if (ifgr->ifgr_len == 0) {
959 		IF_ADDR_LOCK(ifp);
960 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
961 			ifgr->ifgr_len += sizeof(struct ifg_req);
962 		IF_ADDR_UNLOCK(ifp);
963 		return (0);
964 	}
965 
966 	len = ifgr->ifgr_len;
967 	ifgp = ifgr->ifgr_groups;
968 	/* XXX: wire */
969 	IF_ADDR_LOCK(ifp);
970 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
971 		if (len < sizeof(ifgrq)) {
972 			IF_ADDR_UNLOCK(ifp);
973 			return (EINVAL);
974 		}
975 		bzero(&ifgrq, sizeof ifgrq);
976 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
977 		    sizeof(ifgrq.ifgrq_group));
978 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
979 		    	IF_ADDR_UNLOCK(ifp);
980 			return (error);
981 		}
982 		len -= sizeof(ifgrq);
983 		ifgp++;
984 	}
985 	IF_ADDR_UNLOCK(ifp);
986 
987 	return (0);
988 }
989 
990 /*
991  * Stores all members of a group in memory pointed to by data
992  */
993 static int
994 if_getgroupmembers(struct ifgroupreq *data)
995 {
996 	INIT_VNET_NET(curvnet);
997 	struct ifgroupreq	*ifgr = data;
998 	struct ifg_group	*ifg;
999 	struct ifg_member	*ifgm;
1000 	struct ifg_req		 ifgrq, *ifgp;
1001 	int			 len, error;
1002 
1003 	IFNET_RLOCK();
1004 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1005 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1006 			break;
1007 	if (ifg == NULL) {
1008 		IFNET_RUNLOCK();
1009 		return (ENOENT);
1010 	}
1011 
1012 	if (ifgr->ifgr_len == 0) {
1013 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1014 			ifgr->ifgr_len += sizeof(ifgrq);
1015 		IFNET_RUNLOCK();
1016 		return (0);
1017 	}
1018 
1019 	len = ifgr->ifgr_len;
1020 	ifgp = ifgr->ifgr_groups;
1021 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1022 		if (len < sizeof(ifgrq)) {
1023 			IFNET_RUNLOCK();
1024 			return (EINVAL);
1025 		}
1026 		bzero(&ifgrq, sizeof ifgrq);
1027 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1028 		    sizeof(ifgrq.ifgrq_member));
1029 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1030 			IFNET_RUNLOCK();
1031 			return (error);
1032 		}
1033 		len -= sizeof(ifgrq);
1034 		ifgp++;
1035 	}
1036 	IFNET_RUNLOCK();
1037 
1038 	return (0);
1039 }
1040 
1041 /*
1042  * Delete Routes for a Network Interface
1043  *
1044  * Called for each routing entry via the rnh->rnh_walktree() call above
1045  * to delete all route entries referencing a detaching network interface.
1046  *
1047  * Arguments:
1048  *	rn	pointer to node in the routing table
1049  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1050  *
1051  * Returns:
1052  *	0	successful
1053  *	errno	failed - reason indicated
1054  *
1055  */
1056 static int
1057 if_rtdel(struct radix_node *rn, void *arg)
1058 {
1059 	struct rtentry	*rt = (struct rtentry *)rn;
1060 	struct ifnet	*ifp = arg;
1061 	int		err;
1062 
1063 	if (rt->rt_ifp == ifp) {
1064 
1065 		/*
1066 		 * Protect (sorta) against walktree recursion problems
1067 		 * with cloned routes
1068 		 */
1069 		if ((rt->rt_flags & RTF_UP) == 0)
1070 			return (0);
1071 
1072 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1073 				rt_mask(rt), rt->rt_flags,
1074 				(struct rtentry **) NULL, rt->rt_fibnum);
1075 		if (err) {
1076 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1077 		}
1078 	}
1079 
1080 	return (0);
1081 }
1082 
1083 /*
1084  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1085  * structs used to represent other address families, it is necessary
1086  * to perform a different comparison.
1087  */
1088 
1089 #define	sa_equal(a1, a2)	\
1090 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1091 
1092 #define	sa_dl_equal(a1, a2)	\
1093 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1094 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1095 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1096 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1097 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1098 
1099 /*
1100  * Locate an interface based on a complete address.
1101  */
1102 /*ARGSUSED*/
1103 struct ifaddr *
1104 ifa_ifwithaddr(struct sockaddr *addr)
1105 {
1106 	INIT_VNET_NET(curvnet);
1107 	struct ifnet *ifp;
1108 	struct ifaddr *ifa;
1109 
1110 	IFNET_RLOCK();
1111 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
1112 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1113 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1114 				continue;
1115 			if (sa_equal(addr, ifa->ifa_addr))
1116 				goto done;
1117 			/* IP6 doesn't have broadcast */
1118 			if ((ifp->if_flags & IFF_BROADCAST) &&
1119 			    ifa->ifa_broadaddr &&
1120 			    ifa->ifa_broadaddr->sa_len != 0 &&
1121 			    sa_equal(ifa->ifa_broadaddr, addr))
1122 				goto done;
1123 		}
1124 	ifa = NULL;
1125 done:
1126 	IFNET_RUNLOCK();
1127 	return (ifa);
1128 }
1129 
1130 /*
1131  * Locate an interface based on the broadcast address.
1132  */
1133 /* ARGSUSED */
1134 struct ifaddr *
1135 ifa_ifwithbroadaddr(struct sockaddr *addr)
1136 {
1137 	INIT_VNET_NET(curvnet);
1138 	struct ifnet *ifp;
1139 	struct ifaddr *ifa;
1140 
1141 	IFNET_RLOCK();
1142 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
1143 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1144 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1145 				continue;
1146 			if ((ifp->if_flags & IFF_BROADCAST) &&
1147 			    ifa->ifa_broadaddr &&
1148 			    ifa->ifa_broadaddr->sa_len != 0 &&
1149 			    sa_equal(ifa->ifa_broadaddr, addr))
1150 				goto done;
1151 		}
1152 	ifa = NULL;
1153 done:
1154 	IFNET_RUNLOCK();
1155 	return (ifa);
1156 }
1157 
1158 /*
1159  * Locate the point to point interface with a given destination address.
1160  */
1161 /*ARGSUSED*/
1162 struct ifaddr *
1163 ifa_ifwithdstaddr(struct sockaddr *addr)
1164 {
1165 	INIT_VNET_NET(curvnet);
1166 	struct ifnet *ifp;
1167 	struct ifaddr *ifa;
1168 
1169 	IFNET_RLOCK();
1170 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1171 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1172 			continue;
1173 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1174 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1175 				continue;
1176 			if (ifa->ifa_dstaddr != NULL &&
1177 			    sa_equal(addr, ifa->ifa_dstaddr))
1178 				goto done;
1179 		}
1180 	}
1181 	ifa = NULL;
1182 done:
1183 	IFNET_RUNLOCK();
1184 	return (ifa);
1185 }
1186 
1187 /*
1188  * Find an interface on a specific network.  If many, choice
1189  * is most specific found.
1190  */
1191 struct ifaddr *
1192 ifa_ifwithnet(struct sockaddr *addr)
1193 {
1194 	INIT_VNET_NET(curvnet);
1195 	struct ifnet *ifp;
1196 	struct ifaddr *ifa;
1197 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
1198 	u_int af = addr->sa_family;
1199 	char *addr_data = addr->sa_data, *cplim;
1200 
1201 	/*
1202 	 * AF_LINK addresses can be looked up directly by their index number,
1203 	 * so do that if we can.
1204 	 */
1205 	if (af == AF_LINK) {
1206 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1207 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1208 		return (ifaddr_byindex(sdl->sdl_index));
1209 	}
1210 
1211 	/*
1212 	 * Scan though each interface, looking for ones that have
1213 	 * addresses in this address family.
1214 	 */
1215 	IFNET_RLOCK();
1216 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1217 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1218 			char *cp, *cp2, *cp3;
1219 
1220 			if (ifa->ifa_addr->sa_family != af)
1221 next:				continue;
1222 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1223 				/*
1224 				 * This is a bit broken as it doesn't
1225 				 * take into account that the remote end may
1226 				 * be a single node in the network we are
1227 				 * looking for.
1228 				 * The trouble is that we don't know the
1229 				 * netmask for the remote end.
1230 				 */
1231 				if (ifa->ifa_dstaddr != NULL &&
1232 				    sa_equal(addr, ifa->ifa_dstaddr))
1233 					goto done;
1234 			} else {
1235 				/*
1236 				 * if we have a special address handler,
1237 				 * then use it instead of the generic one.
1238 				 */
1239 				if (ifa->ifa_claim_addr) {
1240 					if ((*ifa->ifa_claim_addr)(ifa, addr))
1241 						goto done;
1242 					continue;
1243 				}
1244 
1245 				/*
1246 				 * Scan all the bits in the ifa's address.
1247 				 * If a bit dissagrees with what we are
1248 				 * looking for, mask it with the netmask
1249 				 * to see if it really matters.
1250 				 * (A byte at a time)
1251 				 */
1252 				if (ifa->ifa_netmask == 0)
1253 					continue;
1254 				cp = addr_data;
1255 				cp2 = ifa->ifa_addr->sa_data;
1256 				cp3 = ifa->ifa_netmask->sa_data;
1257 				cplim = ifa->ifa_netmask->sa_len
1258 					+ (char *)ifa->ifa_netmask;
1259 				while (cp3 < cplim)
1260 					if ((*cp++ ^ *cp2++) & *cp3++)
1261 						goto next; /* next address! */
1262 				/*
1263 				 * If the netmask of what we just found
1264 				 * is more specific than what we had before
1265 				 * (if we had one) then remember the new one
1266 				 * before continuing to search
1267 				 * for an even better one.
1268 				 */
1269 				if (ifa_maybe == 0 ||
1270 				    rn_refines((caddr_t)ifa->ifa_netmask,
1271 				    (caddr_t)ifa_maybe->ifa_netmask))
1272 					ifa_maybe = ifa;
1273 			}
1274 		}
1275 	}
1276 	ifa = ifa_maybe;
1277 done:
1278 	IFNET_RUNLOCK();
1279 	return (ifa);
1280 }
1281 
1282 /*
1283  * Find an interface address specific to an interface best matching
1284  * a given address.
1285  */
1286 struct ifaddr *
1287 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1288 {
1289 	struct ifaddr *ifa;
1290 	char *cp, *cp2, *cp3;
1291 	char *cplim;
1292 	struct ifaddr *ifa_maybe = 0;
1293 	u_int af = addr->sa_family;
1294 
1295 	if (af >= AF_MAX)
1296 		return (0);
1297 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1298 		if (ifa->ifa_addr->sa_family != af)
1299 			continue;
1300 		if (ifa_maybe == 0)
1301 			ifa_maybe = ifa;
1302 		if (ifa->ifa_netmask == 0) {
1303 			if (sa_equal(addr, ifa->ifa_addr) ||
1304 			    (ifa->ifa_dstaddr &&
1305 			    sa_equal(addr, ifa->ifa_dstaddr)))
1306 				goto done;
1307 			continue;
1308 		}
1309 		if (ifp->if_flags & IFF_POINTOPOINT) {
1310 			if (sa_equal(addr, ifa->ifa_dstaddr))
1311 				goto done;
1312 		} else {
1313 			cp = addr->sa_data;
1314 			cp2 = ifa->ifa_addr->sa_data;
1315 			cp3 = ifa->ifa_netmask->sa_data;
1316 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1317 			for (; cp3 < cplim; cp3++)
1318 				if ((*cp++ ^ *cp2++) & *cp3)
1319 					break;
1320 			if (cp3 == cplim)
1321 				goto done;
1322 		}
1323 	}
1324 	ifa = ifa_maybe;
1325 done:
1326 	return (ifa);
1327 }
1328 
1329 #include <net/route.h>
1330 
1331 /*
1332  * Default action when installing a route with a Link Level gateway.
1333  * Lookup an appropriate real ifa to point to.
1334  * This should be moved to /sys/net/link.c eventually.
1335  */
1336 static void
1337 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1338 {
1339 	struct ifaddr *ifa, *oifa;
1340 	struct sockaddr *dst;
1341 	struct ifnet *ifp;
1342 
1343 	RT_LOCK_ASSERT(rt);
1344 
1345 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1346 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1347 		return;
1348 	ifa = ifaof_ifpforaddr(dst, ifp);
1349 	if (ifa) {
1350 		IFAREF(ifa);		/* XXX */
1351 		oifa = rt->rt_ifa;
1352 		rt->rt_ifa = ifa;
1353 		IFAFREE(oifa);
1354 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1355 			ifa->ifa_rtrequest(cmd, rt, info);
1356 	}
1357 }
1358 
1359 /*
1360  * Mark an interface down and notify protocols of
1361  * the transition.
1362  * NOTE: must be called at splnet or eqivalent.
1363  */
1364 static void
1365 if_unroute(struct ifnet *ifp, int flag, int fam)
1366 {
1367 	struct ifaddr *ifa;
1368 
1369 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1370 
1371 	ifp->if_flags &= ~flag;
1372 	getmicrotime(&ifp->if_lastchange);
1373 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1374 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1375 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1376 	if_qflush(&ifp->if_snd);
1377 #ifdef DEV_CARP
1378 	if (ifp->if_carp)
1379 		carp_carpdev_state(ifp->if_carp);
1380 #endif
1381 	rt_ifmsg(ifp);
1382 }
1383 
1384 /*
1385  * Mark an interface up and notify protocols of
1386  * the transition.
1387  * NOTE: must be called at splnet or eqivalent.
1388  */
1389 static void
1390 if_route(struct ifnet *ifp, int flag, int fam)
1391 {
1392 	struct ifaddr *ifa;
1393 
1394 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1395 
1396 	ifp->if_flags |= flag;
1397 	getmicrotime(&ifp->if_lastchange);
1398 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1399 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1400 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1401 #ifdef DEV_CARP
1402 	if (ifp->if_carp)
1403 		carp_carpdev_state(ifp->if_carp);
1404 #endif
1405 	rt_ifmsg(ifp);
1406 #ifdef INET6
1407 	in6_if_up(ifp);
1408 #endif
1409 }
1410 
1411 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1412 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1413 
1414 /*
1415  * Handle a change in the interface link state. To avoid LORs
1416  * between driver lock and upper layer locks, as well as possible
1417  * recursions, we post event to taskqueue, and all job
1418  * is done in static do_link_state_change().
1419  */
1420 void
1421 if_link_state_change(struct ifnet *ifp, int link_state)
1422 {
1423 	/* Return if state hasn't changed. */
1424 	if (ifp->if_link_state == link_state)
1425 		return;
1426 
1427 	ifp->if_link_state = link_state;
1428 
1429 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1430 }
1431 
1432 static void
1433 do_link_state_change(void *arg, int pending)
1434 {
1435 	struct ifnet *ifp = (struct ifnet *)arg;
1436 	int link_state = ifp->if_link_state;
1437 	int link;
1438 	CURVNET_SET(ifp->if_vnet);
1439 
1440 	/* Notify that the link state has changed. */
1441 	rt_ifmsg(ifp);
1442 	if (link_state == LINK_STATE_UP)
1443 		link = NOTE_LINKUP;
1444 	else if (link_state == LINK_STATE_DOWN)
1445 		link = NOTE_LINKDOWN;
1446 	else
1447 		link = NOTE_LINKINV;
1448 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1449 	if (ifp->if_vlantrunk != NULL)
1450 		(*vlan_link_state_p)(ifp, link);
1451 
1452 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1453 	    IFP2AC(ifp)->ac_netgraph != NULL)
1454 		(*ng_ether_link_state_p)(ifp, link_state);
1455 #ifdef DEV_CARP
1456 	if (ifp->if_carp)
1457 		carp_carpdev_state(ifp->if_carp);
1458 #endif
1459 	if (ifp->if_bridge) {
1460 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1461 		(*bstp_linkstate_p)(ifp, link_state);
1462 	}
1463 	if (ifp->if_lagg) {
1464 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1465 		(*lagg_linkstate_p)(ifp, link_state);
1466 	}
1467 
1468 	devctl_notify("IFNET", ifp->if_xname,
1469 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1470 	if (pending > 1)
1471 		if_printf(ifp, "%d link states coalesced\n", pending);
1472 	if (log_link_state_change)
1473 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1474 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1475 	CURVNET_RESTORE();
1476 }
1477 
1478 /*
1479  * Mark an interface down and notify protocols of
1480  * the transition.
1481  * NOTE: must be called at splnet or eqivalent.
1482  */
1483 void
1484 if_down(struct ifnet *ifp)
1485 {
1486 
1487 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1488 }
1489 
1490 /*
1491  * Mark an interface up and notify protocols of
1492  * the transition.
1493  * NOTE: must be called at splnet or eqivalent.
1494  */
1495 void
1496 if_up(struct ifnet *ifp)
1497 {
1498 
1499 	if_route(ifp, IFF_UP, AF_UNSPEC);
1500 }
1501 
1502 /*
1503  * Flush an interface queue.
1504  */
1505 static void
1506 if_qflush(struct ifaltq *ifq)
1507 {
1508 	struct mbuf *m, *n;
1509 
1510 	IFQ_LOCK(ifq);
1511 #ifdef ALTQ
1512 	if (ALTQ_IS_ENABLED(ifq))
1513 		ALTQ_PURGE(ifq);
1514 #endif
1515 	n = ifq->ifq_head;
1516 	while ((m = n) != 0) {
1517 		n = m->m_act;
1518 		m_freem(m);
1519 	}
1520 	ifq->ifq_head = 0;
1521 	ifq->ifq_tail = 0;
1522 	ifq->ifq_len = 0;
1523 	IFQ_UNLOCK(ifq);
1524 }
1525 
1526 /*
1527  * Handle interface watchdog timer routines.  Called
1528  * from softclock, we decrement timers (if set) and
1529  * call the appropriate interface routine on expiration.
1530  *
1531  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1532  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1533  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1534  */
1535 static void
1536 if_slowtimo(void *arg)
1537 {
1538 	VNET_ITERATOR_DECL(vnet_iter);
1539 	struct ifnet *ifp;
1540 	int s = splimp();
1541 
1542 	IFNET_RLOCK();
1543 	VNET_LIST_RLOCK();
1544 	VNET_FOREACH(vnet_iter) {
1545 		CURVNET_SET(vnet_iter);
1546 		INIT_VNET_NET(vnet_iter);
1547 		TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1548 			if (ifp->if_timer == 0 || --ifp->if_timer)
1549 				continue;
1550 			if (ifp->if_watchdog)
1551 				(*ifp->if_watchdog)(ifp);
1552 		}
1553 		CURVNET_RESTORE();
1554 	}
1555 	VNET_LIST_RUNLOCK();
1556 	IFNET_RUNLOCK();
1557 	splx(s);
1558 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1559 }
1560 
1561 /*
1562  * Map interface name to
1563  * interface structure pointer.
1564  */
1565 struct ifnet *
1566 ifunit(const char *name)
1567 {
1568 	INIT_VNET_NET(curvnet);
1569 	struct ifnet *ifp;
1570 
1571 	IFNET_RLOCK();
1572 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1573 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1574 			break;
1575 	}
1576 	IFNET_RUNLOCK();
1577 	return (ifp);
1578 }
1579 
1580 /*
1581  * Hardware specific interface ioctls.
1582  */
1583 static int
1584 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1585 {
1586 	struct ifreq *ifr;
1587 	struct ifstat *ifs;
1588 	int error = 0;
1589 	int new_flags, temp_flags;
1590 	size_t namelen, onamelen;
1591 	char new_name[IFNAMSIZ];
1592 	struct ifaddr *ifa;
1593 	struct sockaddr_dl *sdl;
1594 
1595 	ifr = (struct ifreq *)data;
1596 	switch (cmd) {
1597 	case SIOCGIFINDEX:
1598 		ifr->ifr_index = ifp->if_index;
1599 		break;
1600 
1601 	case SIOCGIFFLAGS:
1602 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1603 		ifr->ifr_flags = temp_flags & 0xffff;
1604 		ifr->ifr_flagshigh = temp_flags >> 16;
1605 		break;
1606 
1607 	case SIOCGIFCAP:
1608 		ifr->ifr_reqcap = ifp->if_capabilities;
1609 		ifr->ifr_curcap = ifp->if_capenable;
1610 		break;
1611 
1612 #ifdef MAC
1613 	case SIOCGIFMAC:
1614 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
1615 		break;
1616 #endif
1617 
1618 	case SIOCGIFMETRIC:
1619 		ifr->ifr_metric = ifp->if_metric;
1620 		break;
1621 
1622 	case SIOCGIFMTU:
1623 		ifr->ifr_mtu = ifp->if_mtu;
1624 		break;
1625 
1626 	case SIOCGIFPHYS:
1627 		ifr->ifr_phys = ifp->if_physical;
1628 		break;
1629 
1630 	case SIOCSIFFLAGS:
1631 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
1632 		if (error)
1633 			return (error);
1634 		/*
1635 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1636 		 * check, so we don't need special handling here yet.
1637 		 */
1638 		new_flags = (ifr->ifr_flags & 0xffff) |
1639 		    (ifr->ifr_flagshigh << 16);
1640 		if (ifp->if_flags & IFF_SMART) {
1641 			/* Smart drivers twiddle their own routes */
1642 		} else if (ifp->if_flags & IFF_UP &&
1643 		    (new_flags & IFF_UP) == 0) {
1644 			int s = splimp();
1645 			if_down(ifp);
1646 			splx(s);
1647 		} else if (new_flags & IFF_UP &&
1648 		    (ifp->if_flags & IFF_UP) == 0) {
1649 			int s = splimp();
1650 			if_up(ifp);
1651 			splx(s);
1652 		}
1653 		/* See if permanently promiscuous mode bit is about to flip */
1654 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1655 			if (new_flags & IFF_PPROMISC)
1656 				ifp->if_flags |= IFF_PROMISC;
1657 			else if (ifp->if_pcount == 0)
1658 				ifp->if_flags &= ~IFF_PROMISC;
1659 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1660 			    ifp->if_xname,
1661 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1662 		}
1663 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1664 			(new_flags &~ IFF_CANTCHANGE);
1665 		if (ifp->if_ioctl) {
1666 			IFF_LOCKGIANT(ifp);
1667 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1668 			IFF_UNLOCKGIANT(ifp);
1669 		}
1670 		getmicrotime(&ifp->if_lastchange);
1671 		break;
1672 
1673 	case SIOCSIFCAP:
1674 		error = priv_check(td, PRIV_NET_SETIFCAP);
1675 		if (error)
1676 			return (error);
1677 		if (ifp->if_ioctl == NULL)
1678 			return (EOPNOTSUPP);
1679 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1680 			return (EINVAL);
1681 		IFF_LOCKGIANT(ifp);
1682 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1683 		IFF_UNLOCKGIANT(ifp);
1684 		if (error == 0)
1685 			getmicrotime(&ifp->if_lastchange);
1686 		break;
1687 
1688 #ifdef MAC
1689 	case SIOCSIFMAC:
1690 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
1691 		break;
1692 #endif
1693 
1694 	case SIOCSIFNAME:
1695 		error = priv_check(td, PRIV_NET_SETIFNAME);
1696 		if (error)
1697 			return (error);
1698 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1699 		if (error != 0)
1700 			return (error);
1701 		if (new_name[0] == '\0')
1702 			return (EINVAL);
1703 		if (ifunit(new_name) != NULL)
1704 			return (EEXIST);
1705 
1706 		/* Announce the departure of the interface. */
1707 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1708 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1709 
1710 		log(LOG_INFO, "%s: changing name to '%s'\n",
1711 		    ifp->if_xname, new_name);
1712 
1713 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1714 		ifa = ifp->if_addr;
1715 		IFA_LOCK(ifa);
1716 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1717 		namelen = strlen(new_name);
1718 		onamelen = sdl->sdl_nlen;
1719 		/*
1720 		 * Move the address if needed.  This is safe because we
1721 		 * allocate space for a name of length IFNAMSIZ when we
1722 		 * create this in if_attach().
1723 		 */
1724 		if (namelen != onamelen) {
1725 			bcopy(sdl->sdl_data + onamelen,
1726 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1727 		}
1728 		bcopy(new_name, sdl->sdl_data, namelen);
1729 		sdl->sdl_nlen = namelen;
1730 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1731 		bzero(sdl->sdl_data, onamelen);
1732 		while (namelen != 0)
1733 			sdl->sdl_data[--namelen] = 0xff;
1734 		IFA_UNLOCK(ifa);
1735 
1736 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1737 		/* Announce the return of the interface. */
1738 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1739 		break;
1740 
1741 	case SIOCSIFMETRIC:
1742 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
1743 		if (error)
1744 			return (error);
1745 		ifp->if_metric = ifr->ifr_metric;
1746 		getmicrotime(&ifp->if_lastchange);
1747 		break;
1748 
1749 	case SIOCSIFPHYS:
1750 		error = priv_check(td, PRIV_NET_SETIFPHYS);
1751 		if (error)
1752 			return (error);
1753 		if (ifp->if_ioctl == NULL)
1754 			return (EOPNOTSUPP);
1755 		IFF_LOCKGIANT(ifp);
1756 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1757 		IFF_UNLOCKGIANT(ifp);
1758 		if (error == 0)
1759 			getmicrotime(&ifp->if_lastchange);
1760 		break;
1761 
1762 	case SIOCSIFMTU:
1763 	{
1764 		u_long oldmtu = ifp->if_mtu;
1765 
1766 		error = priv_check(td, PRIV_NET_SETIFMTU);
1767 		if (error)
1768 			return (error);
1769 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1770 			return (EINVAL);
1771 		if (ifp->if_ioctl == NULL)
1772 			return (EOPNOTSUPP);
1773 		IFF_LOCKGIANT(ifp);
1774 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1775 		IFF_UNLOCKGIANT(ifp);
1776 		if (error == 0) {
1777 			getmicrotime(&ifp->if_lastchange);
1778 			rt_ifmsg(ifp);
1779 		}
1780 		/*
1781 		 * If the link MTU changed, do network layer specific procedure.
1782 		 */
1783 		if (ifp->if_mtu != oldmtu) {
1784 #ifdef INET6
1785 			nd6_setmtu(ifp);
1786 #endif
1787 		}
1788 		break;
1789 	}
1790 
1791 	case SIOCADDMULTI:
1792 	case SIOCDELMULTI:
1793 		if (cmd == SIOCADDMULTI)
1794 			error = priv_check(td, PRIV_NET_ADDMULTI);
1795 		else
1796 			error = priv_check(td, PRIV_NET_DELMULTI);
1797 		if (error)
1798 			return (error);
1799 
1800 		/* Don't allow group membership on non-multicast interfaces. */
1801 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1802 			return (EOPNOTSUPP);
1803 
1804 		/* Don't let users screw up protocols' entries. */
1805 		if (ifr->ifr_addr.sa_family != AF_LINK)
1806 			return (EINVAL);
1807 
1808 		if (cmd == SIOCADDMULTI) {
1809 			struct ifmultiaddr *ifma;
1810 
1811 			/*
1812 			 * Userland is only permitted to join groups once
1813 			 * via the if_addmulti() KPI, because it cannot hold
1814 			 * struct ifmultiaddr * between calls. It may also
1815 			 * lose a race while we check if the membership
1816 			 * already exists.
1817 			 */
1818 			IF_ADDR_LOCK(ifp);
1819 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
1820 			IF_ADDR_UNLOCK(ifp);
1821 			if (ifma != NULL)
1822 				error = EADDRINUSE;
1823 			else
1824 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1825 		} else {
1826 			error = if_delmulti(ifp, &ifr->ifr_addr);
1827 		}
1828 		if (error == 0)
1829 			getmicrotime(&ifp->if_lastchange);
1830 		break;
1831 
1832 	case SIOCSIFPHYADDR:
1833 	case SIOCDIFPHYADDR:
1834 #ifdef INET6
1835 	case SIOCSIFPHYADDR_IN6:
1836 #endif
1837 	case SIOCSLIFPHYADDR:
1838 	case SIOCSIFMEDIA:
1839 	case SIOCSIFGENERIC:
1840 		error = priv_check(td, PRIV_NET_HWIOCTL);
1841 		if (error)
1842 			return (error);
1843 		if (ifp->if_ioctl == NULL)
1844 			return (EOPNOTSUPP);
1845 		IFF_LOCKGIANT(ifp);
1846 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1847 		IFF_UNLOCKGIANT(ifp);
1848 		if (error == 0)
1849 			getmicrotime(&ifp->if_lastchange);
1850 		break;
1851 
1852 	case SIOCGIFSTATUS:
1853 		ifs = (struct ifstat *)data;
1854 		ifs->ascii[0] = '\0';
1855 
1856 	case SIOCGIFPSRCADDR:
1857 	case SIOCGIFPDSTADDR:
1858 	case SIOCGLIFPHYADDR:
1859 	case SIOCGIFMEDIA:
1860 	case SIOCGIFGENERIC:
1861 		if (ifp->if_ioctl == NULL)
1862 			return (EOPNOTSUPP);
1863 		IFF_LOCKGIANT(ifp);
1864 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1865 		IFF_UNLOCKGIANT(ifp);
1866 		break;
1867 
1868 	case SIOCSIFLLADDR:
1869 		error = priv_check(td, PRIV_NET_SETLLADDR);
1870 		if (error)
1871 			return (error);
1872 		error = if_setlladdr(ifp,
1873 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1874 		break;
1875 
1876 	case SIOCAIFGROUP:
1877 	{
1878 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1879 
1880 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
1881 		if (error)
1882 			return (error);
1883 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
1884 			return (error);
1885 		break;
1886 	}
1887 
1888 	case SIOCGIFGROUP:
1889 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
1890 			return (error);
1891 		break;
1892 
1893 	case SIOCDIFGROUP:
1894 	{
1895 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1896 
1897 		error = priv_check(td, PRIV_NET_DELIFGROUP);
1898 		if (error)
1899 			return (error);
1900 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
1901 			return (error);
1902 		break;
1903 	}
1904 
1905 	default:
1906 		error = ENOIOCTL;
1907 		break;
1908 	}
1909 	return (error);
1910 }
1911 
1912 /*
1913  * Interface ioctls.
1914  */
1915 int
1916 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1917 {
1918 	struct ifnet *ifp;
1919 	struct ifreq *ifr;
1920 	int error;
1921 	int oif_flags;
1922 
1923 	switch (cmd) {
1924 	case SIOCGIFCONF:
1925 	case OSIOCGIFCONF:
1926 #ifdef __amd64__
1927 	case SIOCGIFCONF32:
1928 #endif
1929 		return (ifconf(cmd, data));
1930 	}
1931 	ifr = (struct ifreq *)data;
1932 
1933 	switch (cmd) {
1934 	case SIOCIFCREATE:
1935 	case SIOCIFCREATE2:
1936 		error = priv_check(td, PRIV_NET_IFCREATE);
1937 		if (error)
1938 			return (error);
1939 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1940 			cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1941 	case SIOCIFDESTROY:
1942 		error = priv_check(td, PRIV_NET_IFDESTROY);
1943 		if (error)
1944 			return (error);
1945 		return if_clone_destroy(ifr->ifr_name);
1946 
1947 	case SIOCIFGCLONERS:
1948 		return (if_clone_list((struct if_clonereq *)data));
1949 	case SIOCGIFGMEMB:
1950 		return (if_getgroupmembers((struct ifgroupreq *)data));
1951 	}
1952 
1953 	ifp = ifunit(ifr->ifr_name);
1954 	if (ifp == 0)
1955 		return (ENXIO);
1956 
1957 	error = ifhwioctl(cmd, ifp, data, td);
1958 	if (error != ENOIOCTL)
1959 		return (error);
1960 
1961 	oif_flags = ifp->if_flags;
1962 	if (so->so_proto == 0)
1963 		return (EOPNOTSUPP);
1964 #ifndef COMPAT_43
1965 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1966 								 data,
1967 								 ifp, td));
1968 #else
1969 	{
1970 		int ocmd = cmd;
1971 
1972 		switch (cmd) {
1973 
1974 		case SIOCSIFDSTADDR:
1975 		case SIOCSIFADDR:
1976 		case SIOCSIFBRDADDR:
1977 		case SIOCSIFNETMASK:
1978 #if BYTE_ORDER != BIG_ENDIAN
1979 			if (ifr->ifr_addr.sa_family == 0 &&
1980 			    ifr->ifr_addr.sa_len < 16) {
1981 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1982 				ifr->ifr_addr.sa_len = 16;
1983 			}
1984 #else
1985 			if (ifr->ifr_addr.sa_len == 0)
1986 				ifr->ifr_addr.sa_len = 16;
1987 #endif
1988 			break;
1989 
1990 		case OSIOCGIFADDR:
1991 			cmd = SIOCGIFADDR;
1992 			break;
1993 
1994 		case OSIOCGIFDSTADDR:
1995 			cmd = SIOCGIFDSTADDR;
1996 			break;
1997 
1998 		case OSIOCGIFBRDADDR:
1999 			cmd = SIOCGIFBRDADDR;
2000 			break;
2001 
2002 		case OSIOCGIFNETMASK:
2003 			cmd = SIOCGIFNETMASK;
2004 		}
2005 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2006 								   cmd,
2007 								   data,
2008 								   ifp, td));
2009 		switch (ocmd) {
2010 
2011 		case OSIOCGIFADDR:
2012 		case OSIOCGIFDSTADDR:
2013 		case OSIOCGIFBRDADDR:
2014 		case OSIOCGIFNETMASK:
2015 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2016 
2017 		}
2018 	}
2019 #endif /* COMPAT_43 */
2020 
2021 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2022 #ifdef INET6
2023 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
2024 		if (ifp->if_flags & IFF_UP) {
2025 			int s = splimp();
2026 			in6_if_up(ifp);
2027 			splx(s);
2028 		}
2029 #endif
2030 	}
2031 	return (error);
2032 }
2033 
2034 /*
2035  * The code common to handling reference counted flags,
2036  * e.g., in ifpromisc() and if_allmulti().
2037  * The "pflag" argument can specify a permanent mode flag to check,
2038  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2039  *
2040  * Only to be used on stack-owned flags, not driver-owned flags.
2041  */
2042 static int
2043 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2044 {
2045 	struct ifreq ifr;
2046 	int error;
2047 	int oldflags, oldcount;
2048 
2049 	/* Sanity checks to catch programming errors */
2050 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2051 	    ("%s: setting driver-owned flag %d", __func__, flag));
2052 
2053 	if (onswitch)
2054 		KASSERT(*refcount >= 0,
2055 		    ("%s: increment negative refcount %d for flag %d",
2056 		    __func__, *refcount, flag));
2057 	else
2058 		KASSERT(*refcount > 0,
2059 		    ("%s: decrement non-positive refcount %d for flag %d",
2060 		    __func__, *refcount, flag));
2061 
2062 	/* In case this mode is permanent, just touch refcount */
2063 	if (ifp->if_flags & pflag) {
2064 		*refcount += onswitch ? 1 : -1;
2065 		return (0);
2066 	}
2067 
2068 	/* Save ifnet parameters for if_ioctl() may fail */
2069 	oldcount = *refcount;
2070 	oldflags = ifp->if_flags;
2071 
2072 	/*
2073 	 * See if we aren't the only and touching refcount is enough.
2074 	 * Actually toggle interface flag if we are the first or last.
2075 	 */
2076 	if (onswitch) {
2077 		if ((*refcount)++)
2078 			return (0);
2079 		ifp->if_flags |= flag;
2080 	} else {
2081 		if (--(*refcount))
2082 			return (0);
2083 		ifp->if_flags &= ~flag;
2084 	}
2085 
2086 	/* Call down the driver since we've changed interface flags */
2087 	if (ifp->if_ioctl == NULL) {
2088 		error = EOPNOTSUPP;
2089 		goto recover;
2090 	}
2091 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2092 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2093 	IFF_LOCKGIANT(ifp);
2094 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2095 	IFF_UNLOCKGIANT(ifp);
2096 	if (error)
2097 		goto recover;
2098 	/* Notify userland that interface flags have changed */
2099 	rt_ifmsg(ifp);
2100 	return (0);
2101 
2102 recover:
2103 	/* Recover after driver error */
2104 	*refcount = oldcount;
2105 	ifp->if_flags = oldflags;
2106 	return (error);
2107 }
2108 
2109 /*
2110  * Set/clear promiscuous mode on interface ifp based on the truth value
2111  * of pswitch.  The calls are reference counted so that only the first
2112  * "on" request actually has an effect, as does the final "off" request.
2113  * Results are undefined if the "off" and "on" requests are not matched.
2114  */
2115 int
2116 ifpromisc(struct ifnet *ifp, int pswitch)
2117 {
2118 	int error;
2119 	int oldflags = ifp->if_flags;
2120 
2121 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2122 			   &ifp->if_pcount, pswitch);
2123 	/* If promiscuous mode status has changed, log a message */
2124 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2125 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2126 		    ifp->if_xname,
2127 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2128 	return (error);
2129 }
2130 
2131 /*
2132  * Return interface configuration
2133  * of system.  List may be used
2134  * in later ioctl's (above) to get
2135  * other information.
2136  */
2137 /*ARGSUSED*/
2138 static int
2139 ifconf(u_long cmd, caddr_t data)
2140 {
2141 	INIT_VNET_NET(curvnet);
2142 	struct ifconf *ifc = (struct ifconf *)data;
2143 #ifdef __amd64__
2144 	struct ifconf32 *ifc32 = (struct ifconf32 *)data;
2145 	struct ifconf ifc_swab;
2146 #endif
2147 	struct ifnet *ifp;
2148 	struct ifaddr *ifa;
2149 	struct ifreq ifr;
2150 	struct sbuf *sb;
2151 	int error, full = 0, valid_len, max_len;
2152 
2153 #ifdef __amd64__
2154 	if (cmd == SIOCGIFCONF32) {
2155 		ifc_swab.ifc_len = ifc32->ifc_len;
2156 		ifc_swab.ifc_buf = (caddr_t)(uintptr_t)ifc32->ifc_buf;
2157 		ifc = &ifc_swab;
2158 	}
2159 #endif
2160 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2161 	max_len = MAXPHYS - 1;
2162 
2163 	/* Prevent hostile input from being able to crash the system */
2164 	if (ifc->ifc_len <= 0)
2165 		return (EINVAL);
2166 
2167 again:
2168 	if (ifc->ifc_len <= max_len) {
2169 		max_len = ifc->ifc_len;
2170 		full = 1;
2171 	}
2172 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2173 	max_len = 0;
2174 	valid_len = 0;
2175 
2176 	IFNET_RLOCK();		/* could sleep XXX */
2177 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2178 		int addrs;
2179 
2180 		/*
2181 		 * Zero the ifr_name buffer to make sure we don't
2182 		 * disclose the contents of the stack.
2183 		 */
2184 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2185 
2186 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2187 		    >= sizeof(ifr.ifr_name)) {
2188 			sbuf_delete(sb);
2189 			IFNET_RUNLOCK();
2190 			return (ENAMETOOLONG);
2191 		}
2192 
2193 		addrs = 0;
2194 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2195 			struct sockaddr *sa = ifa->ifa_addr;
2196 
2197 			if (jailed(curthread->td_ucred) &&
2198 			    prison_if(curthread->td_ucred, sa))
2199 				continue;
2200 			addrs++;
2201 #ifdef COMPAT_43
2202 			if (cmd == OSIOCGIFCONF) {
2203 				struct osockaddr *osa =
2204 					 (struct osockaddr *)&ifr.ifr_addr;
2205 				ifr.ifr_addr = *sa;
2206 				osa->sa_family = sa->sa_family;
2207 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2208 				max_len += sizeof(ifr);
2209 			} else
2210 #endif
2211 			if (sa->sa_len <= sizeof(*sa)) {
2212 				ifr.ifr_addr = *sa;
2213 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2214 				max_len += sizeof(ifr);
2215 			} else {
2216 				sbuf_bcat(sb, &ifr,
2217 				    offsetof(struct ifreq, ifr_addr));
2218 				max_len += offsetof(struct ifreq, ifr_addr);
2219 				sbuf_bcat(sb, sa, sa->sa_len);
2220 				max_len += sa->sa_len;
2221 			}
2222 
2223 			if (!sbuf_overflowed(sb))
2224 				valid_len = sbuf_len(sb);
2225 		}
2226 		if (addrs == 0) {
2227 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2228 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2229 			max_len += sizeof(ifr);
2230 
2231 			if (!sbuf_overflowed(sb))
2232 				valid_len = sbuf_len(sb);
2233 		}
2234 	}
2235 	IFNET_RUNLOCK();
2236 
2237 	/*
2238 	 * If we didn't allocate enough space (uncommon), try again.  If
2239 	 * we have already allocated as much space as we are allowed,
2240 	 * return what we've got.
2241 	 */
2242 	if (valid_len != max_len && !full) {
2243 		sbuf_delete(sb);
2244 		goto again;
2245 	}
2246 
2247 	ifc->ifc_len = valid_len;
2248 #ifdef __amd64__
2249 	if (cmd == SIOCGIFCONF32)
2250 		ifc32->ifc_len = valid_len;
2251 #endif
2252 	sbuf_finish(sb);
2253 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2254 	sbuf_delete(sb);
2255 	return (error);
2256 }
2257 
2258 /*
2259  * Just like ifpromisc(), but for all-multicast-reception mode.
2260  */
2261 int
2262 if_allmulti(struct ifnet *ifp, int onswitch)
2263 {
2264 
2265 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2266 }
2267 
2268 struct ifmultiaddr *
2269 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2270 {
2271 	struct ifmultiaddr *ifma;
2272 
2273 	IF_ADDR_LOCK_ASSERT(ifp);
2274 
2275 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2276 		if (sa->sa_family == AF_LINK) {
2277 			if (sa_dl_equal(ifma->ifma_addr, sa))
2278 				break;
2279 		} else {
2280 			if (sa_equal(ifma->ifma_addr, sa))
2281 				break;
2282 		}
2283 	}
2284 
2285 	return ifma;
2286 }
2287 
2288 /*
2289  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2290  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2291  * the ifnet multicast address list here, so the caller must do that and
2292  * other setup work (such as notifying the device driver).  The reference
2293  * count is initialized to 1.
2294  */
2295 static struct ifmultiaddr *
2296 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2297     int mflags)
2298 {
2299 	struct ifmultiaddr *ifma;
2300 	struct sockaddr *dupsa;
2301 
2302 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2303 	    M_ZERO);
2304 	if (ifma == NULL)
2305 		return (NULL);
2306 
2307 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2308 	if (dupsa == NULL) {
2309 		free(ifma, M_IFMADDR);
2310 		return (NULL);
2311 	}
2312 	bcopy(sa, dupsa, sa->sa_len);
2313 	ifma->ifma_addr = dupsa;
2314 
2315 	ifma->ifma_ifp = ifp;
2316 	ifma->ifma_refcount = 1;
2317 	ifma->ifma_protospec = NULL;
2318 
2319 	if (llsa == NULL) {
2320 		ifma->ifma_lladdr = NULL;
2321 		return (ifma);
2322 	}
2323 
2324 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2325 	if (dupsa == NULL) {
2326 		free(ifma->ifma_addr, M_IFMADDR);
2327 		free(ifma, M_IFMADDR);
2328 		return (NULL);
2329 	}
2330 	bcopy(llsa, dupsa, llsa->sa_len);
2331 	ifma->ifma_lladdr = dupsa;
2332 
2333 	return (ifma);
2334 }
2335 
2336 /*
2337  * if_freemulti: free ifmultiaddr structure and possibly attached related
2338  * addresses.  The caller is responsible for implementing reference
2339  * counting, notifying the driver, handling routing messages, and releasing
2340  * any dependent link layer state.
2341  */
2342 static void
2343 if_freemulti(struct ifmultiaddr *ifma)
2344 {
2345 
2346 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2347 	    ifma->ifma_refcount));
2348 	KASSERT(ifma->ifma_protospec == NULL,
2349 	    ("if_freemulti: protospec not NULL"));
2350 
2351 	if (ifma->ifma_lladdr != NULL)
2352 		free(ifma->ifma_lladdr, M_IFMADDR);
2353 	free(ifma->ifma_addr, M_IFMADDR);
2354 	free(ifma, M_IFMADDR);
2355 }
2356 
2357 /*
2358  * Register an additional multicast address with a network interface.
2359  *
2360  * - If the address is already present, bump the reference count on the
2361  *   address and return.
2362  * - If the address is not link-layer, look up a link layer address.
2363  * - Allocate address structures for one or both addresses, and attach to the
2364  *   multicast address list on the interface.  If automatically adding a link
2365  *   layer address, the protocol address will own a reference to the link
2366  *   layer address, to be freed when it is freed.
2367  * - Notify the network device driver of an addition to the multicast address
2368  *   list.
2369  *
2370  * 'sa' points to caller-owned memory with the desired multicast address.
2371  *
2372  * 'retifma' will be used to return a pointer to the resulting multicast
2373  * address reference, if desired.
2374  */
2375 int
2376 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2377     struct ifmultiaddr **retifma)
2378 {
2379 	struct ifmultiaddr *ifma, *ll_ifma;
2380 	struct sockaddr *llsa;
2381 	int error;
2382 
2383 	/*
2384 	 * If the address is already present, return a new reference to it;
2385 	 * otherwise, allocate storage and set up a new address.
2386 	 */
2387 	IF_ADDR_LOCK(ifp);
2388 	ifma = if_findmulti(ifp, sa);
2389 	if (ifma != NULL) {
2390 		ifma->ifma_refcount++;
2391 		if (retifma != NULL)
2392 			*retifma = ifma;
2393 		IF_ADDR_UNLOCK(ifp);
2394 		return (0);
2395 	}
2396 
2397 	/*
2398 	 * The address isn't already present; resolve the protocol address
2399 	 * into a link layer address, and then look that up, bump its
2400 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2401 	 * returned, we will need to free it later.
2402 	 */
2403 	llsa = NULL;
2404 	ll_ifma = NULL;
2405 	if (ifp->if_resolvemulti != NULL) {
2406 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2407 		if (error)
2408 			goto unlock_out;
2409 	}
2410 
2411 	/*
2412 	 * Allocate the new address.  Don't hook it up yet, as we may also
2413 	 * need to allocate a link layer multicast address.
2414 	 */
2415 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2416 	if (ifma == NULL) {
2417 		error = ENOMEM;
2418 		goto free_llsa_out;
2419 	}
2420 
2421 	/*
2422 	 * If a link layer address is found, we'll need to see if it's
2423 	 * already present in the address list, or allocate is as well.
2424 	 * When this block finishes, the link layer address will be on the
2425 	 * list.
2426 	 */
2427 	if (llsa != NULL) {
2428 		ll_ifma = if_findmulti(ifp, llsa);
2429 		if (ll_ifma == NULL) {
2430 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2431 			if (ll_ifma == NULL) {
2432 				--ifma->ifma_refcount;
2433 				if_freemulti(ifma);
2434 				error = ENOMEM;
2435 				goto free_llsa_out;
2436 			}
2437 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
2438 			    ifma_link);
2439 		} else
2440 			ll_ifma->ifma_refcount++;
2441 		ifma->ifma_llifma = ll_ifma;
2442 	}
2443 
2444 	/*
2445 	 * We now have a new multicast address, ifma, and possibly a new or
2446 	 * referenced link layer address.  Add the primary address to the
2447 	 * ifnet address list.
2448 	 */
2449 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2450 
2451 	if (retifma != NULL)
2452 		*retifma = ifma;
2453 
2454 	/*
2455 	 * Must generate the message while holding the lock so that 'ifma'
2456 	 * pointer is still valid.
2457 	 */
2458 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2459 	IF_ADDR_UNLOCK(ifp);
2460 
2461 	/*
2462 	 * We are certain we have added something, so call down to the
2463 	 * interface to let them know about it.
2464 	 */
2465 	if (ifp->if_ioctl != NULL) {
2466 		IFF_LOCKGIANT(ifp);
2467 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2468 		IFF_UNLOCKGIANT(ifp);
2469 	}
2470 
2471 	if (llsa != NULL)
2472 		free(llsa, M_IFMADDR);
2473 
2474 	return (0);
2475 
2476 free_llsa_out:
2477 	if (llsa != NULL)
2478 		free(llsa, M_IFMADDR);
2479 
2480 unlock_out:
2481 	IF_ADDR_UNLOCK(ifp);
2482 	return (error);
2483 }
2484 
2485 /*
2486  * Delete a multicast group membership by network-layer group address.
2487  *
2488  * Returns ENOENT if the entry could not be found. If ifp no longer
2489  * exists, results are undefined. This entry point should only be used
2490  * from subsystems which do appropriate locking to hold ifp for the
2491  * duration of the call.
2492  * Network-layer protocol domains must use if_delmulti_ifma().
2493  */
2494 int
2495 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2496 {
2497 	struct ifmultiaddr *ifma;
2498 	int lastref;
2499 #ifdef INVARIANTS
2500 	struct ifnet *oifp;
2501 	INIT_VNET_NET(ifp->if_vnet);
2502 
2503 	IFNET_RLOCK();
2504 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
2505 		if (ifp == oifp)
2506 			break;
2507 	if (ifp != oifp)
2508 		ifp = NULL;
2509 	IFNET_RUNLOCK();
2510 
2511 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
2512 #endif
2513 	if (ifp == NULL)
2514 		return (ENOENT);
2515 
2516 	IF_ADDR_LOCK(ifp);
2517 	lastref = 0;
2518 	ifma = if_findmulti(ifp, sa);
2519 	if (ifma != NULL)
2520 		lastref = if_delmulti_locked(ifp, ifma, 0);
2521 	IF_ADDR_UNLOCK(ifp);
2522 
2523 	if (ifma == NULL)
2524 		return (ENOENT);
2525 
2526 	if (lastref && ifp->if_ioctl != NULL) {
2527 		IFF_LOCKGIANT(ifp);
2528 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2529 		IFF_UNLOCKGIANT(ifp);
2530 	}
2531 
2532 	return (0);
2533 }
2534 
2535 /*
2536  * Delete a multicast group membership by group membership pointer.
2537  * Network-layer protocol domains must use this routine.
2538  *
2539  * It is safe to call this routine if the ifp disappeared. Callers should
2540  * hold IFF_LOCKGIANT() to avoid a LOR in case the hardware needs to be
2541  * reconfigured.
2542  */
2543 void
2544 if_delmulti_ifma(struct ifmultiaddr *ifma)
2545 {
2546 #ifdef DIAGNOSTIC
2547 	INIT_VNET_NET(curvnet);
2548 #endif
2549 	struct ifnet *ifp;
2550 	int lastref;
2551 
2552 	ifp = ifma->ifma_ifp;
2553 #ifdef DIAGNOSTIC
2554 	if (ifp == NULL) {
2555 		printf("%s: ifma_ifp seems to be detached\n", __func__);
2556 	} else {
2557 		struct ifnet *oifp;
2558 
2559 		IFNET_RLOCK();
2560 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
2561 			if (ifp == oifp)
2562 				break;
2563 		if (ifp != oifp) {
2564 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
2565 			ifp = NULL;
2566 		}
2567 		IFNET_RUNLOCK();
2568 	}
2569 #endif
2570 	/*
2571 	 * If and only if the ifnet instance exists: Acquire the address lock.
2572 	 */
2573 	if (ifp != NULL)
2574 		IF_ADDR_LOCK(ifp);
2575 
2576 	lastref = if_delmulti_locked(ifp, ifma, 0);
2577 
2578 	if (ifp != NULL) {
2579 		/*
2580 		 * If and only if the ifnet instance exists:
2581 		 *  Release the address lock.
2582 		 *  If the group was left: update the hardware hash filter.
2583 		 */
2584 		IF_ADDR_UNLOCK(ifp);
2585 		if (lastref && ifp->if_ioctl != NULL) {
2586 			IFF_LOCKGIANT(ifp);
2587 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2588 			IFF_UNLOCKGIANT(ifp);
2589 		}
2590 	}
2591 }
2592 
2593 /*
2594  * Perform deletion of network-layer and/or link-layer multicast address.
2595  *
2596  * Return 0 if the reference count was decremented.
2597  * Return 1 if the final reference was released, indicating that the
2598  * hardware hash filter should be reprogrammed.
2599  */
2600 static int
2601 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
2602 {
2603 	struct ifmultiaddr *ll_ifma;
2604 
2605 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
2606 		KASSERT(ifma->ifma_ifp == ifp,
2607 		    ("%s: inconsistent ifp %p", __func__, ifp));
2608 		IF_ADDR_LOCK_ASSERT(ifp);
2609 	}
2610 
2611 	ifp = ifma->ifma_ifp;
2612 
2613 	/*
2614 	 * If the ifnet is detaching, null out references to ifnet,
2615 	 * so that upper protocol layers will notice, and not attempt
2616 	 * to obtain locks for an ifnet which no longer exists. The
2617 	 * routing socket announcement must happen before the ifnet
2618 	 * instance is detached from the system.
2619 	 */
2620 	if (detaching) {
2621 #ifdef DIAGNOSTIC
2622 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
2623 #endif
2624 		/*
2625 		 * ifp may already be nulled out if we are being reentered
2626 		 * to delete the ll_ifma.
2627 		 */
2628 		if (ifp != NULL) {
2629 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
2630 			ifma->ifma_ifp = NULL;
2631 		}
2632 	}
2633 
2634 	if (--ifma->ifma_refcount > 0)
2635 		return 0;
2636 
2637 	/*
2638 	 * If this ifma is a network-layer ifma, a link-layer ifma may
2639 	 * have been associated with it. Release it first if so.
2640 	 */
2641 	ll_ifma = ifma->ifma_llifma;
2642 	if (ll_ifma != NULL) {
2643 		KASSERT(ifma->ifma_lladdr != NULL,
2644 		    ("%s: llifma w/o lladdr", __func__));
2645 		if (detaching)
2646 			ll_ifma->ifma_ifp = NULL;	/* XXX */
2647 		if (--ll_ifma->ifma_refcount == 0) {
2648 			if (ifp != NULL) {
2649 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
2650 				    ifma_link);
2651 			}
2652 			if_freemulti(ll_ifma);
2653 		}
2654 	}
2655 
2656 	if (ifp != NULL)
2657 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2658 
2659 	if_freemulti(ifma);
2660 
2661 	/*
2662 	 * The last reference to this instance of struct ifmultiaddr
2663 	 * was released; the hardware should be notified of this change.
2664 	 */
2665 	return 1;
2666 }
2667 
2668 /*
2669  * Set the link layer address on an interface.
2670  *
2671  * At this time we only support certain types of interfaces,
2672  * and we don't allow the length of the address to change.
2673  */
2674 int
2675 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2676 {
2677 	struct sockaddr_dl *sdl;
2678 	struct ifaddr *ifa;
2679 	struct ifreq ifr;
2680 
2681 	ifa = ifp->if_addr;
2682 	if (ifa == NULL)
2683 		return (EINVAL);
2684 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2685 	if (sdl == NULL)
2686 		return (EINVAL);
2687 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2688 		return (EINVAL);
2689 	switch (ifp->if_type) {
2690 	case IFT_ETHER:
2691 	case IFT_FDDI:
2692 	case IFT_XETHER:
2693 	case IFT_ISO88025:
2694 	case IFT_L2VLAN:
2695 	case IFT_BRIDGE:
2696 	case IFT_ARCNET:
2697 	case IFT_IEEE8023ADLAG:
2698 		bcopy(lladdr, LLADDR(sdl), len);
2699 		break;
2700 	default:
2701 		return (ENODEV);
2702 	}
2703 	/*
2704 	 * If the interface is already up, we need
2705 	 * to re-init it in order to reprogram its
2706 	 * address filter.
2707 	 */
2708 	if ((ifp->if_flags & IFF_UP) != 0) {
2709 		if (ifp->if_ioctl) {
2710 			IFF_LOCKGIANT(ifp);
2711 			ifp->if_flags &= ~IFF_UP;
2712 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2713 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2714 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2715 			ifp->if_flags |= IFF_UP;
2716 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2717 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2718 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2719 			IFF_UNLOCKGIANT(ifp);
2720 		}
2721 #ifdef INET
2722 		/*
2723 		 * Also send gratuitous ARPs to notify other nodes about
2724 		 * the address change.
2725 		 */
2726 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2727 			if (ifa->ifa_addr->sa_family == AF_INET)
2728 				arp_ifinit(ifp, ifa);
2729 		}
2730 #endif
2731 	}
2732 	return (0);
2733 }
2734 
2735 /*
2736  * The name argument must be a pointer to storage which will last as
2737  * long as the interface does.  For physical devices, the result of
2738  * device_get_name(dev) is a good choice and for pseudo-devices a
2739  * static string works well.
2740  */
2741 void
2742 if_initname(struct ifnet *ifp, const char *name, int unit)
2743 {
2744 	ifp->if_dname = name;
2745 	ifp->if_dunit = unit;
2746 	if (unit != IF_DUNIT_NONE)
2747 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2748 	else
2749 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2750 }
2751 
2752 int
2753 if_printf(struct ifnet *ifp, const char * fmt, ...)
2754 {
2755 	va_list ap;
2756 	int retval;
2757 
2758 	retval = printf("%s: ", ifp->if_xname);
2759 	va_start(ap, fmt);
2760 	retval += vprintf(fmt, ap);
2761 	va_end(ap);
2762 	return (retval);
2763 }
2764 
2765 /*
2766  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2767  * be called without Giant.  However, we often can't acquire the Giant lock
2768  * at those points; instead, we run it via a task queue that holds Giant via
2769  * if_start_deferred.
2770  *
2771  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2772  * outstanding if_start_deferred() tasks that will run after the free.  This
2773  * probably means waiting in if_detach().
2774  */
2775 void
2776 if_start(struct ifnet *ifp)
2777 {
2778 
2779 	if (ifp->if_flags & IFF_NEEDSGIANT) {
2780 		if (mtx_owned(&Giant))
2781 			(*(ifp)->if_start)(ifp);
2782 		else
2783 			taskqueue_enqueue(taskqueue_swi_giant,
2784 			    &ifp->if_starttask);
2785 	} else
2786 		(*(ifp)->if_start)(ifp);
2787 }
2788 
2789 static void
2790 if_start_deferred(void *context, int pending)
2791 {
2792 	struct ifnet *ifp;
2793 
2794 	GIANT_REQUIRED;
2795 
2796 	ifp = context;
2797 	(ifp->if_start)(ifp);
2798 }
2799 
2800 int
2801 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2802 {
2803 	int active = 0;
2804 
2805 	IF_LOCK(ifq);
2806 	if (_IF_QFULL(ifq)) {
2807 		_IF_DROP(ifq);
2808 		IF_UNLOCK(ifq);
2809 		m_freem(m);
2810 		return (0);
2811 	}
2812 	if (ifp != NULL) {
2813 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2814 		if (m->m_flags & (M_BCAST|M_MCAST))
2815 			ifp->if_omcasts++;
2816 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2817 	}
2818 	_IF_ENQUEUE(ifq, m);
2819 	IF_UNLOCK(ifq);
2820 	if (ifp != NULL && !active)
2821 		if_start(ifp);
2822 	return (1);
2823 }
2824 
2825 void
2826 if_register_com_alloc(u_char type,
2827     if_com_alloc_t *a, if_com_free_t *f)
2828 {
2829 
2830 	KASSERT(if_com_alloc[type] == NULL,
2831 	    ("if_register_com_alloc: %d already registered", type));
2832 	KASSERT(if_com_free[type] == NULL,
2833 	    ("if_register_com_alloc: %d free already registered", type));
2834 
2835 	if_com_alloc[type] = a;
2836 	if_com_free[type] = f;
2837 }
2838 
2839 void
2840 if_deregister_com_alloc(u_char type)
2841 {
2842 
2843 	KASSERT(if_com_alloc[type] != NULL,
2844 	    ("if_deregister_com_alloc: %d not registered", type));
2845 	KASSERT(if_com_free[type] != NULL,
2846 	    ("if_deregister_com_alloc: %d free not registered", type));
2847 	if_com_alloc[type] = NULL;
2848 	if_com_free[type] = NULL;
2849 }
2850