xref: /freebsd/sys/net/if.c (revision 7dfd9569a2f0637fb9a48157b1c1bfe5709faee3)
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 	/*
627 	 * Remove routes and flush queues.
628 	 */
629 	s = splnet();
630 	if_down(ifp);
631 #ifdef ALTQ
632 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
633 		altq_disable(&ifp->if_snd);
634 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
635 		altq_detach(&ifp->if_snd);
636 #endif
637 
638 	if_purgeaddrs(ifp);
639 
640 #ifdef INET
641 	in_ifdetach(ifp);
642 #endif
643 
644 #ifdef INET6
645 	/*
646 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
647 	 * before removing routing entries below, since IPv6 interface direct
648 	 * routes are expected to be removed by the IPv6-specific kernel API.
649 	 * Otherwise, the kernel will detect some inconsistency and bark it.
650 	 */
651 	in6_ifdetach(ifp);
652 #endif
653 	/*
654 	 * Remove link ifaddr pointer and maybe decrement if_index.
655 	 * Clean up all addresses.
656 	 */
657 	ifp->if_addr = NULL;
658 	destroy_dev(ifdev_byindex(ifp->if_index));
659 	ifdev_byindex(ifp->if_index) = NULL;
660 
661 	/* We can now free link ifaddr. */
662 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
663 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
664 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
665 		IFAFREE(ifa);
666 	}
667 
668 	/*
669 	 * Delete all remaining routes using this interface
670 	 * Unfortuneatly the only way to do this is to slog through
671 	 * the entire routing table looking for routes which point
672 	 * to this interface...oh well...
673 	 */
674 	for (i = 1; i <= AF_MAX; i++) {
675 		if ((rnh = rt_tables[i]) == NULL)
676 			continue;
677 		RADIX_NODE_HEAD_LOCK(rnh);
678 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
679 		RADIX_NODE_HEAD_UNLOCK(rnh);
680 	}
681 
682 	/* Announce that the interface is gone. */
683 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
684 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
685 
686 	IF_AFDATA_LOCK(ifp);
687 	for (dp = domains; dp; dp = dp->dom_next) {
688 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
689 			(*dp->dom_ifdetach)(ifp,
690 			    ifp->if_afdata[dp->dom_family]);
691 	}
692 	IF_AFDATA_UNLOCK(ifp);
693 
694 #ifdef MAC
695 	mac_destroy_ifnet(ifp);
696 #endif /* MAC */
697 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
698 	knlist_clear(&ifp->if_klist, 0);
699 	knlist_destroy(&ifp->if_klist);
700 	IFNET_WLOCK();
701  	found = 0;
702  	TAILQ_FOREACH(iter, &ifnet, if_link)
703  		if (iter == ifp) {
704  			found = 1;
705  			break;
706  		}
707  	if (found)
708  		TAILQ_REMOVE(&ifnet, ifp, if_link);
709 	IFNET_WUNLOCK();
710 	mtx_destroy(&ifp->if_snd.ifq_mtx);
711 	IF_AFDATA_DESTROY(ifp);
712 	splx(s);
713 }
714 
715 /*
716  * Delete Routes for a Network Interface
717  *
718  * Called for each routing entry via the rnh->rnh_walktree() call above
719  * to delete all route entries referencing a detaching network interface.
720  *
721  * Arguments:
722  *	rn	pointer to node in the routing table
723  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
724  *
725  * Returns:
726  *	0	successful
727  *	errno	failed - reason indicated
728  *
729  */
730 static int
731 if_rtdel(struct radix_node *rn, void *arg)
732 {
733 	struct rtentry	*rt = (struct rtentry *)rn;
734 	struct ifnet	*ifp = arg;
735 	int		err;
736 
737 	if (rt->rt_ifp == ifp) {
738 
739 		/*
740 		 * Protect (sorta) against walktree recursion problems
741 		 * with cloned routes
742 		 */
743 		if ((rt->rt_flags & RTF_UP) == 0)
744 			return (0);
745 
746 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
747 				rt_mask(rt), rt->rt_flags,
748 				(struct rtentry **) NULL);
749 		if (err) {
750 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
751 		}
752 	}
753 
754 	return (0);
755 }
756 
757 #define	sa_equal(a1, a2)	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
758 
759 /*
760  * Locate an interface based on a complete address.
761  */
762 /*ARGSUSED*/
763 struct ifaddr *
764 ifa_ifwithaddr(struct sockaddr *addr)
765 {
766 	struct ifnet *ifp;
767 	struct ifaddr *ifa;
768 
769 	IFNET_RLOCK();
770 	TAILQ_FOREACH(ifp, &ifnet, if_link)
771 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
772 			if (ifa->ifa_addr->sa_family != addr->sa_family)
773 				continue;
774 			if (sa_equal(addr, ifa->ifa_addr))
775 				goto done;
776 			/* IP6 doesn't have broadcast */
777 			if ((ifp->if_flags & IFF_BROADCAST) &&
778 			    ifa->ifa_broadaddr &&
779 			    ifa->ifa_broadaddr->sa_len != 0 &&
780 			    sa_equal(ifa->ifa_broadaddr, addr))
781 				goto done;
782 		}
783 	ifa = NULL;
784 done:
785 	IFNET_RUNLOCK();
786 	return (ifa);
787 }
788 
789 /*
790  * Locate the point to point interface with a given destination address.
791  */
792 /*ARGSUSED*/
793 struct ifaddr *
794 ifa_ifwithdstaddr(struct sockaddr *addr)
795 {
796 	struct ifnet *ifp;
797 	struct ifaddr *ifa;
798 
799 	IFNET_RLOCK();
800 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
801 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
802 			continue;
803 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
804 			if (ifa->ifa_addr->sa_family != addr->sa_family)
805 				continue;
806 			if (ifa->ifa_dstaddr &&
807 			    sa_equal(addr, ifa->ifa_dstaddr))
808 				goto done;
809 		}
810 	}
811 	ifa = NULL;
812 done:
813 	IFNET_RUNLOCK();
814 	return (ifa);
815 }
816 
817 /*
818  * Find an interface on a specific network.  If many, choice
819  * is most specific found.
820  */
821 struct ifaddr *
822 ifa_ifwithnet(struct sockaddr *addr)
823 {
824 	struct ifnet *ifp;
825 	struct ifaddr *ifa;
826 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
827 	u_int af = addr->sa_family;
828 	char *addr_data = addr->sa_data, *cplim;
829 
830 	/*
831 	 * AF_LINK addresses can be looked up directly by their index number,
832 	 * so do that if we can.
833 	 */
834 	if (af == AF_LINK) {
835 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
836 	    if (sdl->sdl_index && sdl->sdl_index <= if_index)
837 		return (ifaddr_byindex(sdl->sdl_index));
838 	}
839 
840 	/*
841 	 * Scan though each interface, looking for ones that have
842 	 * addresses in this address family.
843 	 */
844 	IFNET_RLOCK();
845 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
846 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
847 			char *cp, *cp2, *cp3;
848 
849 			if (ifa->ifa_addr->sa_family != af)
850 next:				continue;
851 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
852 				/*
853 				 * This is a bit broken as it doesn't
854 				 * take into account that the remote end may
855 				 * be a single node in the network we are
856 				 * looking for.
857 				 * The trouble is that we don't know the
858 				 * netmask for the remote end.
859 				 */
860 				if (ifa->ifa_dstaddr != 0 &&
861 				    sa_equal(addr, ifa->ifa_dstaddr))
862 					goto done;
863 			} else {
864 				/*
865 				 * if we have a special address handler,
866 				 * then use it instead of the generic one.
867 				 */
868 				if (ifa->ifa_claim_addr) {
869 					if ((*ifa->ifa_claim_addr)(ifa, addr))
870 						goto done;
871 					continue;
872 				}
873 
874 				/*
875 				 * Scan all the bits in the ifa's address.
876 				 * If a bit dissagrees with what we are
877 				 * looking for, mask it with the netmask
878 				 * to see if it really matters.
879 				 * (A byte at a time)
880 				 */
881 				if (ifa->ifa_netmask == 0)
882 					continue;
883 				cp = addr_data;
884 				cp2 = ifa->ifa_addr->sa_data;
885 				cp3 = ifa->ifa_netmask->sa_data;
886 				cplim = ifa->ifa_netmask->sa_len
887 					+ (char *)ifa->ifa_netmask;
888 				while (cp3 < cplim)
889 					if ((*cp++ ^ *cp2++) & *cp3++)
890 						goto next; /* next address! */
891 				/*
892 				 * If the netmask of what we just found
893 				 * is more specific than what we had before
894 				 * (if we had one) then remember the new one
895 				 * before continuing to search
896 				 * for an even better one.
897 				 */
898 				if (ifa_maybe == 0 ||
899 				    rn_refines((caddr_t)ifa->ifa_netmask,
900 				    (caddr_t)ifa_maybe->ifa_netmask))
901 					ifa_maybe = ifa;
902 			}
903 		}
904 	}
905 	ifa = ifa_maybe;
906 done:
907 	IFNET_RUNLOCK();
908 	return (ifa);
909 }
910 
911 /*
912  * Find an interface address specific to an interface best matching
913  * a given address.
914  */
915 struct ifaddr *
916 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
917 {
918 	struct ifaddr *ifa;
919 	char *cp, *cp2, *cp3;
920 	char *cplim;
921 	struct ifaddr *ifa_maybe = 0;
922 	u_int af = addr->sa_family;
923 
924 	if (af >= AF_MAX)
925 		return (0);
926 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
927 		if (ifa->ifa_addr->sa_family != af)
928 			continue;
929 		if (ifa_maybe == 0)
930 			ifa_maybe = ifa;
931 		if (ifa->ifa_netmask == 0) {
932 			if (sa_equal(addr, ifa->ifa_addr) ||
933 			    (ifa->ifa_dstaddr &&
934 			    sa_equal(addr, ifa->ifa_dstaddr)))
935 				goto done;
936 			continue;
937 		}
938 		if (ifp->if_flags & IFF_POINTOPOINT) {
939 			if (sa_equal(addr, ifa->ifa_dstaddr))
940 				goto done;
941 		} else {
942 			cp = addr->sa_data;
943 			cp2 = ifa->ifa_addr->sa_data;
944 			cp3 = ifa->ifa_netmask->sa_data;
945 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
946 			for (; cp3 < cplim; cp3++)
947 				if ((*cp++ ^ *cp2++) & *cp3)
948 					break;
949 			if (cp3 == cplim)
950 				goto done;
951 		}
952 	}
953 	ifa = ifa_maybe;
954 done:
955 	return (ifa);
956 }
957 
958 #include <net/route.h>
959 
960 /*
961  * Default action when installing a route with a Link Level gateway.
962  * Lookup an appropriate real ifa to point to.
963  * This should be moved to /sys/net/link.c eventually.
964  */
965 static void
966 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
967 {
968 	struct ifaddr *ifa, *oifa;
969 	struct sockaddr *dst;
970 	struct ifnet *ifp;
971 
972 	RT_LOCK_ASSERT(rt);
973 
974 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
975 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
976 		return;
977 	ifa = ifaof_ifpforaddr(dst, ifp);
978 	if (ifa) {
979 		IFAREF(ifa);		/* XXX */
980 		oifa = rt->rt_ifa;
981 		rt->rt_ifa = ifa;
982 		IFAFREE(oifa);
983 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
984 			ifa->ifa_rtrequest(cmd, rt, info);
985 	}
986 }
987 
988 /*
989  * Mark an interface down and notify protocols of
990  * the transition.
991  * NOTE: must be called at splnet or eqivalent.
992  */
993 static void
994 if_unroute(struct ifnet *ifp, int flag, int fam)
995 {
996 	struct ifaddr *ifa;
997 
998 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
999 
1000 	ifp->if_flags &= ~flag;
1001 	getmicrotime(&ifp->if_lastchange);
1002 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1003 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1004 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1005 	if_qflush(&ifp->if_snd);
1006 #ifdef DEV_CARP
1007 	if (ifp->if_carp)
1008 		carp_carpdev_state(ifp->if_carp);
1009 #endif
1010 	rt_ifmsg(ifp);
1011 }
1012 
1013 /*
1014  * Mark an interface up and notify protocols of
1015  * the transition.
1016  * NOTE: must be called at splnet or eqivalent.
1017  */
1018 static void
1019 if_route(struct ifnet *ifp, int flag, int fam)
1020 {
1021 	struct ifaddr *ifa;
1022 
1023 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1024 
1025 	ifp->if_flags |= flag;
1026 	getmicrotime(&ifp->if_lastchange);
1027 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1028 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1029 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1030 #ifdef DEV_CARP
1031 	if (ifp->if_carp)
1032 		carp_carpdev_state(ifp->if_carp);
1033 #endif
1034 	rt_ifmsg(ifp);
1035 #ifdef INET6
1036 	in6_if_up(ifp);
1037 #endif
1038 }
1039 
1040 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1041 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1042 
1043 /*
1044  * Handle a change in the interface link state. To avoid LORs
1045  * between driver lock and upper layer locks, as well as possible
1046  * recursions, we post event to taskqueue, and all job
1047  * is done in static do_link_state_change().
1048  */
1049 void
1050 if_link_state_change(struct ifnet *ifp, int link_state)
1051 {
1052 	/* Return if state hasn't changed. */
1053 	if (ifp->if_link_state == link_state)
1054 		return;
1055 
1056 	ifp->if_link_state = link_state;
1057 
1058 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1059 }
1060 
1061 static void
1062 do_link_state_change(void *arg, int pending)
1063 {
1064 	struct ifnet *ifp = (struct ifnet *)arg;
1065 	int link_state = ifp->if_link_state;
1066 	int link;
1067 
1068 	/* Notify that the link state has changed. */
1069 	rt_ifmsg(ifp);
1070 	if (link_state == LINK_STATE_UP)
1071 		link = NOTE_LINKUP;
1072 	else if (link_state == LINK_STATE_DOWN)
1073 		link = NOTE_LINKDOWN;
1074 	else
1075 		link = NOTE_LINKINV;
1076 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1077 	if (ifp->if_vlantrunk != NULL)
1078 		(*vlan_link_state_p)(ifp, link);
1079 
1080 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1081 	    IFP2AC(ifp)->ac_netgraph != NULL)
1082 		(*ng_ether_link_state_p)(ifp, link_state);
1083 #ifdef DEV_CARP
1084 	if (ifp->if_carp)
1085 		carp_carpdev_state(ifp->if_carp);
1086 #endif
1087 	if (ifp->if_bridge) {
1088 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1089 		(*bstp_linkstate_p)(ifp, link_state);
1090 	}
1091 
1092 	devctl_notify("IFNET", ifp->if_xname,
1093 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1094 	if (pending > 1)
1095 		if_printf(ifp, "%d link states coalesced\n", pending);
1096 	if (log_link_state_change)
1097 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1098 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1099 }
1100 
1101 /*
1102  * Mark an interface down and notify protocols of
1103  * the transition.
1104  * NOTE: must be called at splnet or eqivalent.
1105  */
1106 void
1107 if_down(struct ifnet *ifp)
1108 {
1109 
1110 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1111 }
1112 
1113 /*
1114  * Mark an interface up and notify protocols of
1115  * the transition.
1116  * NOTE: must be called at splnet or eqivalent.
1117  */
1118 void
1119 if_up(struct ifnet *ifp)
1120 {
1121 
1122 	if_route(ifp, IFF_UP, AF_UNSPEC);
1123 }
1124 
1125 /*
1126  * Flush an interface queue.
1127  */
1128 static void
1129 if_qflush(struct ifaltq *ifq)
1130 {
1131 	struct mbuf *m, *n;
1132 
1133 	IFQ_LOCK(ifq);
1134 #ifdef ALTQ
1135 	if (ALTQ_IS_ENABLED(ifq))
1136 		ALTQ_PURGE(ifq);
1137 #endif
1138 	n = ifq->ifq_head;
1139 	while ((m = n) != 0) {
1140 		n = m->m_act;
1141 		m_freem(m);
1142 	}
1143 	ifq->ifq_head = 0;
1144 	ifq->ifq_tail = 0;
1145 	ifq->ifq_len = 0;
1146 	IFQ_UNLOCK(ifq);
1147 }
1148 
1149 /*
1150  * Handle interface watchdog timer routines.  Called
1151  * from softclock, we decrement timers (if set) and
1152  * call the appropriate interface routine on expiration.
1153  *
1154  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1155  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1156  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1157  */
1158 static void
1159 if_slowtimo(void *arg)
1160 {
1161 	struct ifnet *ifp;
1162 	int s = splimp();
1163 
1164 	IFNET_RLOCK();
1165 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1166 		if (ifp->if_timer == 0 || --ifp->if_timer)
1167 			continue;
1168 		if (ifp->if_watchdog)
1169 			(*ifp->if_watchdog)(ifp);
1170 	}
1171 	IFNET_RUNLOCK();
1172 	splx(s);
1173 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1174 }
1175 
1176 /*
1177  * Map interface name to
1178  * interface structure pointer.
1179  */
1180 struct ifnet *
1181 ifunit(const char *name)
1182 {
1183 	struct ifnet *ifp;
1184 
1185 	IFNET_RLOCK();
1186 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1187 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1188 			break;
1189 	}
1190 	IFNET_RUNLOCK();
1191 	return (ifp);
1192 }
1193 
1194 /*
1195  * Hardware specific interface ioctls.
1196  */
1197 static int
1198 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1199 {
1200 	struct ifreq *ifr;
1201 	struct ifstat *ifs;
1202 	int error = 0;
1203 	int new_flags, temp_flags;
1204 	size_t namelen, onamelen;
1205 	char new_name[IFNAMSIZ];
1206 	struct ifaddr *ifa;
1207 	struct sockaddr_dl *sdl;
1208 
1209 	ifr = (struct ifreq *)data;
1210 	switch (cmd) {
1211 	case SIOCGIFINDEX:
1212 		ifr->ifr_index = ifp->if_index;
1213 		break;
1214 
1215 	case SIOCGIFFLAGS:
1216 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1217 		ifr->ifr_flags = temp_flags & 0xffff;
1218 		ifr->ifr_flagshigh = temp_flags >> 16;
1219 		break;
1220 
1221 	case SIOCGIFCAP:
1222 		ifr->ifr_reqcap = ifp->if_capabilities;
1223 		ifr->ifr_curcap = ifp->if_capenable;
1224 		break;
1225 
1226 #ifdef MAC
1227 	case SIOCGIFMAC:
1228 		error = mac_ioctl_ifnet_get(td->td_ucred, ifr, ifp);
1229 		break;
1230 #endif
1231 
1232 	case SIOCGIFMETRIC:
1233 		ifr->ifr_metric = ifp->if_metric;
1234 		break;
1235 
1236 	case SIOCGIFMTU:
1237 		ifr->ifr_mtu = ifp->if_mtu;
1238 		break;
1239 
1240 	case SIOCGIFPHYS:
1241 		ifr->ifr_phys = ifp->if_physical;
1242 		break;
1243 
1244 	case SIOCSIFFLAGS:
1245 		error = suser(td);
1246 		if (error)
1247 			return (error);
1248 		/*
1249 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1250 		 * check, so we don't need special handling here yet.
1251 		 */
1252 		new_flags = (ifr->ifr_flags & 0xffff) |
1253 		    (ifr->ifr_flagshigh << 16);
1254 		if (ifp->if_flags & IFF_SMART) {
1255 			/* Smart drivers twiddle their own routes */
1256 		} else if (ifp->if_flags & IFF_UP &&
1257 		    (new_flags & IFF_UP) == 0) {
1258 			int s = splimp();
1259 			if_down(ifp);
1260 			splx(s);
1261 		} else if (new_flags & IFF_UP &&
1262 		    (ifp->if_flags & IFF_UP) == 0) {
1263 			int s = splimp();
1264 			if_up(ifp);
1265 			splx(s);
1266 		}
1267 		/* See if permanently promiscuous mode bit is about to flip */
1268 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1269 			if (new_flags & IFF_PPROMISC)
1270 				ifp->if_flags |= IFF_PROMISC;
1271 			else if (ifp->if_pcount == 0)
1272 				ifp->if_flags &= ~IFF_PROMISC;
1273 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1274 			    ifp->if_xname,
1275 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1276 		}
1277 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1278 			(new_flags &~ IFF_CANTCHANGE);
1279 		if (ifp->if_ioctl) {
1280 			IFF_LOCKGIANT(ifp);
1281 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1282 			IFF_UNLOCKGIANT(ifp);
1283 		}
1284 		getmicrotime(&ifp->if_lastchange);
1285 		break;
1286 
1287 	case SIOCSIFCAP:
1288 		error = suser(td);
1289 		if (error)
1290 			return (error);
1291 		if (ifp->if_ioctl == NULL)
1292 			return (EOPNOTSUPP);
1293 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1294 			return (EINVAL);
1295 		IFF_LOCKGIANT(ifp);
1296 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1297 		IFF_UNLOCKGIANT(ifp);
1298 		if (error == 0)
1299 			getmicrotime(&ifp->if_lastchange);
1300 		break;
1301 
1302 #ifdef MAC
1303 	case SIOCSIFMAC:
1304 		error = mac_ioctl_ifnet_set(td->td_ucred, ifr, ifp);
1305 		break;
1306 #endif
1307 
1308 	case SIOCSIFNAME:
1309 		error = suser(td);
1310 		if (error != 0)
1311 			return (error);
1312 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1313 		if (error != 0)
1314 			return (error);
1315 		if (new_name[0] == '\0')
1316 			return (EINVAL);
1317 		if (ifunit(new_name) != NULL)
1318 			return (EEXIST);
1319 
1320 		/* Announce the departure of the interface. */
1321 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1322 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1323 
1324 		log(LOG_INFO, "%s: changing name to '%s'\n",
1325 		    ifp->if_xname, new_name);
1326 
1327 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1328 		ifa = ifp->if_addr;
1329 		IFA_LOCK(ifa);
1330 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1331 		namelen = strlen(new_name);
1332 		onamelen = sdl->sdl_nlen;
1333 		/*
1334 		 * Move the address if needed.  This is safe because we
1335 		 * allocate space for a name of length IFNAMSIZ when we
1336 		 * create this in if_attach().
1337 		 */
1338 		if (namelen != onamelen) {
1339 			bcopy(sdl->sdl_data + onamelen,
1340 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1341 		}
1342 		bcopy(new_name, sdl->sdl_data, namelen);
1343 		sdl->sdl_nlen = namelen;
1344 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1345 		bzero(sdl->sdl_data, onamelen);
1346 		while (namelen != 0)
1347 			sdl->sdl_data[--namelen] = 0xff;
1348 		IFA_UNLOCK(ifa);
1349 
1350 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1351 		/* Announce the return of the interface. */
1352 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1353 		break;
1354 
1355 	case SIOCSIFMETRIC:
1356 		error = suser(td);
1357 		if (error)
1358 			return (error);
1359 		ifp->if_metric = ifr->ifr_metric;
1360 		getmicrotime(&ifp->if_lastchange);
1361 		break;
1362 
1363 	case SIOCSIFPHYS:
1364 		error = suser(td);
1365 		if (error)
1366 			return (error);
1367 		if (ifp->if_ioctl == NULL)
1368 			return (EOPNOTSUPP);
1369 		IFF_LOCKGIANT(ifp);
1370 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1371 		IFF_UNLOCKGIANT(ifp);
1372 		if (error == 0)
1373 			getmicrotime(&ifp->if_lastchange);
1374 		break;
1375 
1376 	case SIOCSIFMTU:
1377 	{
1378 		u_long oldmtu = ifp->if_mtu;
1379 
1380 		error = suser(td);
1381 		if (error)
1382 			return (error);
1383 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1384 			return (EINVAL);
1385 		if (ifp->if_ioctl == NULL)
1386 			return (EOPNOTSUPP);
1387 		IFF_LOCKGIANT(ifp);
1388 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1389 		IFF_UNLOCKGIANT(ifp);
1390 		if (error == 0) {
1391 			getmicrotime(&ifp->if_lastchange);
1392 			rt_ifmsg(ifp);
1393 		}
1394 		/*
1395 		 * If the link MTU changed, do network layer specific procedure.
1396 		 */
1397 		if (ifp->if_mtu != oldmtu) {
1398 #ifdef INET6
1399 			nd6_setmtu(ifp);
1400 #endif
1401 		}
1402 		break;
1403 	}
1404 
1405 	case SIOCADDMULTI:
1406 	case SIOCDELMULTI:
1407 		error = suser(td);
1408 		if (error)
1409 			return (error);
1410 
1411 		/* Don't allow group membership on non-multicast interfaces. */
1412 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1413 			return (EOPNOTSUPP);
1414 
1415 		/* Don't let users screw up protocols' entries. */
1416 		if (ifr->ifr_addr.sa_family != AF_LINK)
1417 			return (EINVAL);
1418 
1419 		if (cmd == SIOCADDMULTI) {
1420 			struct ifmultiaddr *ifma;
1421 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1422 		} else {
1423 			error = if_delmulti(ifp, &ifr->ifr_addr);
1424 		}
1425 		if (error == 0)
1426 			getmicrotime(&ifp->if_lastchange);
1427 		break;
1428 
1429 	case SIOCSIFPHYADDR:
1430 	case SIOCDIFPHYADDR:
1431 #ifdef INET6
1432 	case SIOCSIFPHYADDR_IN6:
1433 #endif
1434 	case SIOCSLIFPHYADDR:
1435 	case SIOCSIFMEDIA:
1436 	case SIOCSIFGENERIC:
1437 		error = suser(td);
1438 		if (error)
1439 			return (error);
1440 		if (ifp->if_ioctl == NULL)
1441 			return (EOPNOTSUPP);
1442 		IFF_LOCKGIANT(ifp);
1443 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1444 		IFF_UNLOCKGIANT(ifp);
1445 		if (error == 0)
1446 			getmicrotime(&ifp->if_lastchange);
1447 		break;
1448 
1449 	case SIOCGIFSTATUS:
1450 		ifs = (struct ifstat *)data;
1451 		ifs->ascii[0] = '\0';
1452 
1453 	case SIOCGIFPSRCADDR:
1454 	case SIOCGIFPDSTADDR:
1455 	case SIOCGLIFPHYADDR:
1456 	case SIOCGIFMEDIA:
1457 	case SIOCGIFGENERIC:
1458 		if (ifp->if_ioctl == NULL)
1459 			return (EOPNOTSUPP);
1460 		IFF_LOCKGIANT(ifp);
1461 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1462 		IFF_UNLOCKGIANT(ifp);
1463 		break;
1464 
1465 	case SIOCSIFLLADDR:
1466 		error = suser(td);
1467 		if (error)
1468 			return (error);
1469 		error = if_setlladdr(ifp,
1470 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1471 		break;
1472 
1473 	default:
1474 		error = ENOIOCTL;
1475 		break;
1476 	}
1477 	return (error);
1478 }
1479 
1480 /*
1481  * Interface ioctls.
1482  */
1483 int
1484 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1485 {
1486 	struct ifnet *ifp;
1487 	struct ifreq *ifr;
1488 	int error;
1489 	int oif_flags;
1490 
1491 	switch (cmd) {
1492 	case SIOCGIFCONF:
1493 	case OSIOCGIFCONF:
1494 #ifdef __amd64__
1495 	case SIOCGIFCONF32:
1496 #endif
1497 		return (ifconf(cmd, data));
1498 	}
1499 	ifr = (struct ifreq *)data;
1500 
1501 	switch (cmd) {
1502 	case SIOCIFCREATE:
1503 	case SIOCIFDESTROY:
1504 		if ((error = suser(td)) != 0)
1505 			return (error);
1506 		return ((cmd == SIOCIFCREATE) ?
1507 			if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1508 			if_clone_destroy(ifr->ifr_name));
1509 
1510 	case SIOCIFGCLONERS:
1511 		return (if_clone_list((struct if_clonereq *)data));
1512 	}
1513 
1514 	ifp = ifunit(ifr->ifr_name);
1515 	if (ifp == 0)
1516 		return (ENXIO);
1517 
1518 	error = ifhwioctl(cmd, ifp, data, td);
1519 	if (error != ENOIOCTL)
1520 		return (error);
1521 
1522 	oif_flags = ifp->if_flags;
1523 	if (so->so_proto == 0)
1524 		return (EOPNOTSUPP);
1525 #ifndef COMPAT_43
1526 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1527 								 data,
1528 								 ifp, td));
1529 #else
1530 	{
1531 		int ocmd = cmd;
1532 
1533 		switch (cmd) {
1534 
1535 		case SIOCSIFDSTADDR:
1536 		case SIOCSIFADDR:
1537 		case SIOCSIFBRDADDR:
1538 		case SIOCSIFNETMASK:
1539 #if BYTE_ORDER != BIG_ENDIAN
1540 			if (ifr->ifr_addr.sa_family == 0 &&
1541 			    ifr->ifr_addr.sa_len < 16) {
1542 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1543 				ifr->ifr_addr.sa_len = 16;
1544 			}
1545 #else
1546 			if (ifr->ifr_addr.sa_len == 0)
1547 				ifr->ifr_addr.sa_len = 16;
1548 #endif
1549 			break;
1550 
1551 		case OSIOCGIFADDR:
1552 			cmd = SIOCGIFADDR;
1553 			break;
1554 
1555 		case OSIOCGIFDSTADDR:
1556 			cmd = SIOCGIFDSTADDR;
1557 			break;
1558 
1559 		case OSIOCGIFBRDADDR:
1560 			cmd = SIOCGIFBRDADDR;
1561 			break;
1562 
1563 		case OSIOCGIFNETMASK:
1564 			cmd = SIOCGIFNETMASK;
1565 		}
1566 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
1567 								   cmd,
1568 								   data,
1569 								   ifp, td));
1570 		switch (ocmd) {
1571 
1572 		case OSIOCGIFADDR:
1573 		case OSIOCGIFDSTADDR:
1574 		case OSIOCGIFBRDADDR:
1575 		case OSIOCGIFNETMASK:
1576 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1577 
1578 		}
1579 	}
1580 #endif /* COMPAT_43 */
1581 
1582 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1583 #ifdef INET6
1584 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
1585 		if (ifp->if_flags & IFF_UP) {
1586 			int s = splimp();
1587 			in6_if_up(ifp);
1588 			splx(s);
1589 		}
1590 #endif
1591 	}
1592 	return (error);
1593 }
1594 
1595 /*
1596  * The code common to handling reference counted flags,
1597  * e.g., in ifpromisc() and if_allmulti().
1598  * The "pflag" argument can specify a permanent mode flag to check,
1599  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
1600  *
1601  * Only to be used on stack-owned flags, not driver-owned flags.
1602  */
1603 static int
1604 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
1605 {
1606 	struct ifreq ifr;
1607 	int error;
1608 	int oldflags, oldcount;
1609 
1610 	/* Sanity checks to catch programming errors */
1611 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
1612 	    ("%s: setting driver-owned flag %d", __func__, flag));
1613 
1614 	if (onswitch)
1615 		KASSERT(*refcount >= 0,
1616 		    ("%s: increment negative refcount %d for flag %d",
1617 		    __func__, *refcount, flag));
1618 	else
1619 		KASSERT(*refcount > 0,
1620 		    ("%s: decrement non-positive refcount %d for flag %d",
1621 		    __func__, *refcount, flag));
1622 
1623 	/* In case this mode is permanent, just touch refcount */
1624 	if (ifp->if_flags & pflag) {
1625 		*refcount += onswitch ? 1 : -1;
1626 		return (0);
1627 	}
1628 
1629 	/* Save ifnet parameters for if_ioctl() may fail */
1630 	oldcount = *refcount;
1631 	oldflags = ifp->if_flags;
1632 
1633 	/*
1634 	 * See if we aren't the only and touching refcount is enough.
1635 	 * Actually toggle interface flag if we are the first or last.
1636 	 */
1637 	if (onswitch) {
1638 		if ((*refcount)++)
1639 			return (0);
1640 		ifp->if_flags |= flag;
1641 	} else {
1642 		if (--(*refcount))
1643 			return (0);
1644 		ifp->if_flags &= ~flag;
1645 	}
1646 
1647 	/* Call down the driver since we've changed interface flags */
1648 	if (ifp->if_ioctl == NULL) {
1649 		error = EOPNOTSUPP;
1650 		goto recover;
1651 	}
1652 	ifr.ifr_flags = ifp->if_flags & 0xffff;
1653 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1654 	IFF_LOCKGIANT(ifp);
1655 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1656 	IFF_UNLOCKGIANT(ifp);
1657 	if (error)
1658 		goto recover;
1659 	/* Notify userland that interface flags have changed */
1660 	rt_ifmsg(ifp);
1661 	return (0);
1662 
1663 recover:
1664 	/* Recover after driver error */
1665 	*refcount = oldcount;
1666 	ifp->if_flags = oldflags;
1667 	return (error);
1668 }
1669 
1670 /*
1671  * Set/clear promiscuous mode on interface ifp based on the truth value
1672  * of pswitch.  The calls are reference counted so that only the first
1673  * "on" request actually has an effect, as does the final "off" request.
1674  * Results are undefined if the "off" and "on" requests are not matched.
1675  */
1676 int
1677 ifpromisc(struct ifnet *ifp, int pswitch)
1678 {
1679 	int error;
1680 	int oldflags = ifp->if_flags;
1681 
1682 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
1683 			   &ifp->if_pcount, pswitch);
1684 	/* If promiscuous mode status has changed, log a message */
1685 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
1686 		log(LOG_INFO, "%s: promiscuous mode %s\n",
1687 		    ifp->if_xname,
1688 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
1689 	return (error);
1690 }
1691 
1692 /*
1693  * Return interface configuration
1694  * of system.  List may be used
1695  * in later ioctl's (above) to get
1696  * other information.
1697  */
1698 /*ARGSUSED*/
1699 static int
1700 ifconf(u_long cmd, caddr_t data)
1701 {
1702 	struct ifconf *ifc = (struct ifconf *)data;
1703 #ifdef __amd64__
1704 	struct ifconf32 *ifc32 = (struct ifconf32 *)data;
1705 	struct ifconf ifc_swab;
1706 #endif
1707 	struct ifnet *ifp;
1708 	struct ifaddr *ifa;
1709 	struct ifreq ifr;
1710 	struct sbuf *sb;
1711 	int error, full = 0, valid_len, max_len;
1712 
1713 #ifdef __amd64__
1714 	if (cmd == SIOCGIFCONF32) {
1715 		ifc_swab.ifc_len = ifc32->ifc_len;
1716 		ifc_swab.ifc_buf = (caddr_t)(uintptr_t)ifc32->ifc_buf;
1717 		ifc = &ifc_swab;
1718 	}
1719 #endif
1720 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
1721 	max_len = MAXPHYS - 1;
1722 
1723 	/* Prevent hostile input from being able to crash the system */
1724 	if (ifc->ifc_len <= 0)
1725 		return (EINVAL);
1726 
1727 again:
1728 	if (ifc->ifc_len <= max_len) {
1729 		max_len = ifc->ifc_len;
1730 		full = 1;
1731 	}
1732 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
1733 	max_len = 0;
1734 	valid_len = 0;
1735 
1736 	IFNET_RLOCK();		/* could sleep XXX */
1737 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1738 		int addrs;
1739 
1740 		/*
1741 		 * Zero the ifr_name buffer to make sure we don't
1742 		 * disclose the contents of the stack.
1743 		 */
1744 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
1745 
1746 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1747 		    >= sizeof(ifr.ifr_name)) {
1748 			sbuf_delete(sb);
1749 			IFNET_RUNLOCK();
1750 			return (ENAMETOOLONG);
1751 		}
1752 
1753 		addrs = 0;
1754 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1755 			struct sockaddr *sa = ifa->ifa_addr;
1756 
1757 			if (jailed(curthread->td_ucred) &&
1758 			    prison_if(curthread->td_ucred, sa))
1759 				continue;
1760 			addrs++;
1761 #ifdef COMPAT_43
1762 			if (cmd == OSIOCGIFCONF) {
1763 				struct osockaddr *osa =
1764 					 (struct osockaddr *)&ifr.ifr_addr;
1765 				ifr.ifr_addr = *sa;
1766 				osa->sa_family = sa->sa_family;
1767 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1768 				max_len += sizeof(ifr);
1769 			} else
1770 #endif
1771 			if (sa->sa_len <= sizeof(*sa)) {
1772 				ifr.ifr_addr = *sa;
1773 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1774 				max_len += sizeof(ifr);
1775 			} else {
1776 				sbuf_bcat(sb, &ifr,
1777 				    offsetof(struct ifreq, ifr_addr));
1778 				max_len += offsetof(struct ifreq, ifr_addr);
1779 				sbuf_bcat(sb, sa, sa->sa_len);
1780 				max_len += sa->sa_len;
1781 			}
1782 
1783 			if (!sbuf_overflowed(sb))
1784 				valid_len = sbuf_len(sb);
1785 		}
1786 		if (addrs == 0) {
1787 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
1788 			sbuf_bcat(sb, &ifr, sizeof(ifr));
1789 			max_len += sizeof(ifr);
1790 
1791 			if (!sbuf_overflowed(sb))
1792 				valid_len = sbuf_len(sb);
1793 		}
1794 	}
1795 	IFNET_RUNLOCK();
1796 
1797 	/*
1798 	 * If we didn't allocate enough space (uncommon), try again.  If
1799 	 * we have already allocated as much space as we are allowed,
1800 	 * return what we've got.
1801 	 */
1802 	if (valid_len != max_len && !full) {
1803 		sbuf_delete(sb);
1804 		goto again;
1805 	}
1806 
1807 	ifc->ifc_len = valid_len;
1808 #ifdef __amd64__
1809 	if (cmd == SIOCGIFCONF32)
1810 		ifc32->ifc_len = valid_len;
1811 #endif
1812 	sbuf_finish(sb);
1813 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
1814 	sbuf_delete(sb);
1815 	return (error);
1816 }
1817 
1818 /*
1819  * Just like ifpromisc(), but for all-multicast-reception mode.
1820  */
1821 int
1822 if_allmulti(struct ifnet *ifp, int onswitch)
1823 {
1824 
1825 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
1826 }
1827 
1828 static struct ifmultiaddr *
1829 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
1830 {
1831 	struct ifmultiaddr *ifma;
1832 
1833 	IF_ADDR_LOCK_ASSERT(ifp);
1834 
1835 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1836 		if (sa_equal(ifma->ifma_addr, sa))
1837 			break;
1838 	}
1839 
1840 	return ifma;
1841 }
1842 
1843 /*
1844  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
1845  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
1846  * the ifnet multicast address list here, so the caller must do that and
1847  * other setup work (such as notifying the device driver).  The reference
1848  * count is initialized to 1.
1849  */
1850 static struct ifmultiaddr *
1851 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
1852     int mflags)
1853 {
1854 	struct ifmultiaddr *ifma;
1855 	struct sockaddr *dupsa;
1856 
1857 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, mflags |
1858 	    M_ZERO);
1859 	if (ifma == NULL)
1860 		return (NULL);
1861 
1862 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, mflags);
1863 	if (dupsa == NULL) {
1864 		FREE(ifma, M_IFMADDR);
1865 		return (NULL);
1866 	}
1867 	bcopy(sa, dupsa, sa->sa_len);
1868 	ifma->ifma_addr = dupsa;
1869 
1870 	ifma->ifma_ifp = ifp;
1871 	ifma->ifma_refcount = 1;
1872 	ifma->ifma_protospec = NULL;
1873 
1874 	if (llsa == NULL) {
1875 		ifma->ifma_lladdr = NULL;
1876 		return (ifma);
1877 	}
1878 
1879 	MALLOC(dupsa, struct sockaddr *, llsa->sa_len, M_IFMADDR, mflags);
1880 	if (dupsa == NULL) {
1881 		FREE(ifma->ifma_addr, M_IFMADDR);
1882 		FREE(ifma, M_IFMADDR);
1883 		return (NULL);
1884 	}
1885 	bcopy(llsa, dupsa, llsa->sa_len);
1886 	ifma->ifma_lladdr = dupsa;
1887 
1888 	return (ifma);
1889 }
1890 
1891 /*
1892  * if_freemulti: free ifmultiaddr structure and possibly attached related
1893  * addresses.  The caller is responsible for implementing reference
1894  * counting, notifying the driver, handling routing messages, and releasing
1895  * any dependent link layer state.
1896  */
1897 static void
1898 if_freemulti(struct ifmultiaddr *ifma)
1899 {
1900 
1901 	KASSERT(ifma->ifma_refcount == 1, ("if_freemulti: refcount %d",
1902 	    ifma->ifma_refcount));
1903 	KASSERT(ifma->ifma_protospec == NULL,
1904 	    ("if_freemulti: protospec not NULL"));
1905 
1906 	if (ifma->ifma_lladdr != NULL)
1907 		FREE(ifma->ifma_lladdr, M_IFMADDR);
1908 	FREE(ifma->ifma_addr, M_IFMADDR);
1909 	FREE(ifma, M_IFMADDR);
1910 }
1911 
1912 /*
1913  * Register an additional multicast address with a network interface.
1914  *
1915  * - If the address is already present, bump the reference count on the
1916  *   address and return.
1917  * - If the address is not link-layer, look up a link layer address.
1918  * - Allocate address structures for one or both addresses, and attach to the
1919  *   multicast address list on the interface.  If automatically adding a link
1920  *   layer address, the protocol address will own a reference to the link
1921  *   layer address, to be freed when it is freed.
1922  * - Notify the network device driver of an addition to the multicast address
1923  *   list.
1924  *
1925  * 'sa' points to caller-owned memory with the desired multicast address.
1926  *
1927  * 'retifma' will be used to return a pointer to the resulting multicast
1928  * address reference, if desired.
1929  */
1930 int
1931 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
1932     struct ifmultiaddr **retifma)
1933 {
1934 	struct ifmultiaddr *ifma, *ll_ifma;
1935 	struct sockaddr *llsa;
1936 	int error;
1937 
1938 	/*
1939 	 * If the address is already present, return a new reference to it;
1940 	 * otherwise, allocate storage and set up a new address.
1941 	 */
1942 	IF_ADDR_LOCK(ifp);
1943 	ifma = if_findmulti(ifp, sa);
1944 	if (ifma != NULL) {
1945 		ifma->ifma_refcount++;
1946 		if (retifma != NULL)
1947 			*retifma = ifma;
1948 		IF_ADDR_UNLOCK(ifp);
1949 		return (0);
1950 	}
1951 
1952 	/*
1953 	 * The address isn't already present; resolve the protocol address
1954 	 * into a link layer address, and then look that up, bump its
1955 	 * refcount or allocate an ifma for that also.  If 'llsa' was
1956 	 * returned, we will need to free it later.
1957 	 */
1958 	llsa = NULL;
1959 	ll_ifma = NULL;
1960 	if (ifp->if_resolvemulti != NULL) {
1961 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
1962 		if (error)
1963 			goto unlock_out;
1964 	}
1965 
1966 	/*
1967 	 * Allocate the new address.  Don't hook it up yet, as we may also
1968 	 * need to allocate a link layer multicast address.
1969 	 */
1970 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
1971 	if (ifma == NULL) {
1972 		error = ENOMEM;
1973 		goto free_llsa_out;
1974 	}
1975 
1976 	/*
1977 	 * If a link layer address is found, we'll need to see if it's
1978 	 * already present in the address list, or allocate is as well.
1979 	 * When this block finishes, the link layer address will be on the
1980 	 * list.
1981 	 */
1982 	if (llsa != NULL) {
1983 		ll_ifma = if_findmulti(ifp, llsa);
1984 		if (ll_ifma == NULL) {
1985 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
1986 			if (ll_ifma == NULL) {
1987 				if_freemulti(ifma);
1988 				error = ENOMEM;
1989 				goto free_llsa_out;
1990 			}
1991 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
1992 			    ifma_link);
1993 		} else
1994 			ll_ifma->ifma_refcount++;
1995 	}
1996 
1997 	/*
1998 	 * We now have a new multicast address, ifma, and possibly a new or
1999 	 * referenced link layer address.  Add the primary address to the
2000 	 * ifnet address list.
2001 	 */
2002 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2003 
2004 	if (retifma != NULL)
2005 		*retifma = ifma;
2006 
2007 	/*
2008 	 * Must generate the message while holding the lock so that 'ifma'
2009 	 * pointer is still valid.
2010 	 *
2011 	 * XXXRW: How come we don't announce ll_ifma?
2012 	 */
2013 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2014 	IF_ADDR_UNLOCK(ifp);
2015 
2016 	/*
2017 	 * We are certain we have added something, so call down to the
2018 	 * interface to let them know about it.
2019 	 */
2020 	if (ifp->if_ioctl != NULL) {
2021 		IFF_LOCKGIANT(ifp);
2022 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2023 		IFF_UNLOCKGIANT(ifp);
2024 	}
2025 
2026 	if (llsa != NULL)
2027 		FREE(llsa, M_IFMADDR);
2028 
2029 	return (0);
2030 
2031 free_llsa_out:
2032 	if (llsa != NULL)
2033 		FREE(llsa, M_IFMADDR);
2034 
2035 unlock_out:
2036 	IF_ADDR_UNLOCK(ifp);
2037 	return (error);
2038 }
2039 
2040 /*
2041  * Remove a reference to a multicast address on this interface.  Yell
2042  * if the request does not match an existing membership.
2043  */
2044 int
2045 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2046 {
2047 	struct ifmultiaddr *ifma, *ll_ifma;
2048 
2049 	IF_ADDR_LOCK(ifp);
2050 	ifma = if_findmulti(ifp, sa);
2051 	if (ifma == NULL) {
2052 		IF_ADDR_UNLOCK(ifp);
2053 		return ENOENT;
2054 	}
2055 
2056 	if (ifma->ifma_refcount > 1) {
2057 		ifma->ifma_refcount--;
2058 		IF_ADDR_UNLOCK(ifp);
2059 		return 0;
2060 	}
2061 
2062 	sa = ifma->ifma_lladdr;
2063 	if (sa != NULL)
2064 		ll_ifma = if_findmulti(ifp, sa);
2065 	else
2066 		ll_ifma = NULL;
2067 
2068 	/*
2069 	 * XXXRW: How come we don't announce ll_ifma?
2070 	 */
2071 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2072 
2073 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2074 	if_freemulti(ifma);
2075 
2076 	if (ll_ifma != NULL) {
2077 		if (ll_ifma->ifma_refcount == 1) {
2078 			TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifma_link);
2079 			if_freemulti(ll_ifma);
2080 		} else
2081 			ll_ifma->ifma_refcount--;
2082 	}
2083 	IF_ADDR_UNLOCK(ifp);
2084 
2085 	/*
2086 	 * Make sure the interface driver is notified
2087 	 * in the case of a link layer mcast group being left.
2088 	 */
2089 	if (ifp->if_ioctl) {
2090 		IFF_LOCKGIANT(ifp);
2091 		(void) (*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2092 		IFF_UNLOCKGIANT(ifp);
2093 	}
2094 
2095 	return 0;
2096 }
2097 
2098 /*
2099  * Set the link layer address on an interface.
2100  *
2101  * At this time we only support certain types of interfaces,
2102  * and we don't allow the length of the address to change.
2103  */
2104 int
2105 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2106 {
2107 	struct sockaddr_dl *sdl;
2108 	struct ifaddr *ifa;
2109 	struct ifreq ifr;
2110 
2111 	ifa = ifp->if_addr;
2112 	if (ifa == NULL)
2113 		return (EINVAL);
2114 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2115 	if (sdl == NULL)
2116 		return (EINVAL);
2117 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2118 		return (EINVAL);
2119 	switch (ifp->if_type) {
2120 	case IFT_ETHER:
2121 	case IFT_FDDI:
2122 	case IFT_XETHER:
2123 	case IFT_ISO88025:
2124 	case IFT_L2VLAN:
2125 	case IFT_BRIDGE:
2126 	case IFT_ARCNET:
2127 		bcopy(lladdr, LLADDR(sdl), len);
2128 		break;
2129 	default:
2130 		return (ENODEV);
2131 	}
2132 	/*
2133 	 * If the interface is already up, we need
2134 	 * to re-init it in order to reprogram its
2135 	 * address filter.
2136 	 */
2137 	if ((ifp->if_flags & IFF_UP) != 0) {
2138 		if (ifp->if_ioctl) {
2139 			IFF_LOCKGIANT(ifp);
2140 			ifp->if_flags &= ~IFF_UP;
2141 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2142 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2143 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2144 			ifp->if_flags |= IFF_UP;
2145 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2146 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2147 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2148 			IFF_UNLOCKGIANT(ifp);
2149 		}
2150 #ifdef INET
2151 		/*
2152 		 * Also send gratuitous ARPs to notify other nodes about
2153 		 * the address change.
2154 		 */
2155 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2156 			if (ifa->ifa_addr != NULL &&
2157 			    ifa->ifa_addr->sa_family == AF_INET)
2158 				arp_ifinit(ifp, ifa);
2159 		}
2160 #endif
2161 	}
2162 	return (0);
2163 }
2164 
2165 /*
2166  * The name argument must be a pointer to storage which will last as
2167  * long as the interface does.  For physical devices, the result of
2168  * device_get_name(dev) is a good choice and for pseudo-devices a
2169  * static string works well.
2170  */
2171 void
2172 if_initname(struct ifnet *ifp, const char *name, int unit)
2173 {
2174 	ifp->if_dname = name;
2175 	ifp->if_dunit = unit;
2176 	if (unit != IF_DUNIT_NONE)
2177 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2178 	else
2179 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2180 }
2181 
2182 int
2183 if_printf(struct ifnet *ifp, const char * fmt, ...)
2184 {
2185 	va_list ap;
2186 	int retval;
2187 
2188 	retval = printf("%s: ", ifp->if_xname);
2189 	va_start(ap, fmt);
2190 	retval += vprintf(fmt, ap);
2191 	va_end(ap);
2192 	return (retval);
2193 }
2194 
2195 /*
2196  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2197  * be called without Giant.  However, we often can't acquire the Giant lock
2198  * at those points; instead, we run it via a task queue that holds Giant via
2199  * if_start_deferred.
2200  *
2201  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2202  * outstanding if_start_deferred() tasks that will run after the free.  This
2203  * probably means waiting in if_detach().
2204  */
2205 void
2206 if_start(struct ifnet *ifp)
2207 {
2208 
2209 	NET_ASSERT_GIANT();
2210 
2211 	if ((ifp->if_flags & IFF_NEEDSGIANT) != 0 && debug_mpsafenet != 0) {
2212 		if (mtx_owned(&Giant))
2213 			(*(ifp)->if_start)(ifp);
2214 		else
2215 			taskqueue_enqueue(taskqueue_swi_giant,
2216 			    &ifp->if_starttask);
2217 	} else
2218 		(*(ifp)->if_start)(ifp);
2219 }
2220 
2221 static void
2222 if_start_deferred(void *context, int pending)
2223 {
2224 	struct ifnet *ifp;
2225 
2226 	/*
2227 	 * This code must be entered with Giant, and should never run if
2228 	 * we're not running with debug.mpsafenet.
2229 	 */
2230 	KASSERT(debug_mpsafenet != 0, ("if_start_deferred: debug.mpsafenet"));
2231 	GIANT_REQUIRED;
2232 
2233 	ifp = context;
2234 	(ifp->if_start)(ifp);
2235 }
2236 
2237 int
2238 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2239 {
2240 	int active = 0;
2241 
2242 	IF_LOCK(ifq);
2243 	if (_IF_QFULL(ifq)) {
2244 		_IF_DROP(ifq);
2245 		IF_UNLOCK(ifq);
2246 		m_freem(m);
2247 		return (0);
2248 	}
2249 	if (ifp != NULL) {
2250 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2251 		if (m->m_flags & (M_BCAST|M_MCAST))
2252 			ifp->if_omcasts++;
2253 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2254 	}
2255 	_IF_ENQUEUE(ifq, m);
2256 	IF_UNLOCK(ifq);
2257 	if (ifp != NULL && !active)
2258 		if_start(ifp);
2259 	return (1);
2260 }
2261 
2262 void
2263 if_register_com_alloc(u_char type,
2264     if_com_alloc_t *a, if_com_free_t *f)
2265 {
2266 
2267 	KASSERT(if_com_alloc[type] == NULL,
2268 	    ("if_register_com_alloc: %d already registered", type));
2269 	KASSERT(if_com_free[type] == NULL,
2270 	    ("if_register_com_alloc: %d free already registered", type));
2271 
2272 	if_com_alloc[type] = a;
2273 	if_com_free[type] = f;
2274 }
2275 
2276 void
2277 if_deregister_com_alloc(u_char type)
2278 {
2279 
2280 	KASSERT(if_com_alloc[type] == NULL,
2281 	    ("if_deregister_com_alloc: %d not registered", type));
2282 	KASSERT(if_com_free[type] == NULL,
2283 	    ("if_deregister_com_alloc: %d free not registered", type));
2284 	if_com_alloc[type] = NULL;
2285 	if_com_free[type] = NULL;
2286 }
2287