xref: /freebsd/sys/net/if.c (revision b3aaa0cc21c63d388230c7ef2a80abd631ff20d5)
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/lock.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
60 #include <sys/jail.h>
61 #include <sys/vimage.h>
62 #include <machine/stdarg.h>
63 #include <vm/uma.h>
64 
65 #include <net/if.h>
66 #include <net/if_arp.h>
67 #include <net/if_clone.h>
68 #include <net/if_dl.h>
69 #include <net/if_types.h>
70 #include <net/if_var.h>
71 #include <net/radix.h>
72 #include <net/route.h>
73 #include <net/vnet.h>
74 
75 #if defined(INET) || defined(INET6)
76 /*XXX*/
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
79 #ifdef INET6
80 #include <netinet6/in6_var.h>
81 #include <netinet6/in6_ifattach.h>
82 #endif
83 #endif
84 #ifdef INET
85 #include <netinet/if_ether.h>
86 #include <netinet/vinet.h>
87 #endif
88 #ifdef DEV_CARP
89 #include <netinet/ip_carp.h>
90 #endif
91 
92 #include <security/mac/mac_framework.h>
93 
94 #ifndef VIMAGE
95 #ifndef VIMAGE_GLOBALS
96 struct vnet_net vnet_net_0;
97 #endif
98 #endif
99 
100 static int slowtimo_started;
101 
102 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
103 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
104 
105 /* Log link state change events */
106 static int log_link_state_change = 1;
107 
108 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
109 	&log_link_state_change, 0,
110 	"log interface link state change events");
111 
112 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
113 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
114 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
115 
116 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
117 
118 /*
119  * XXX: Style; these should be sorted alphabetically, and unprototyped
120  * static functions should be prototyped. Currently they are sorted by
121  * declaration order.
122  */
123 static void	if_attachdomain(void *);
124 static void	if_attachdomain1(struct ifnet *);
125 static int	ifconf(u_long, caddr_t);
126 static void	if_freemulti(struct ifmultiaddr *);
127 static void	if_grow(void);
128 static void	if_init(void *);
129 static void	if_check(void *);
130 static void	if_qflush(struct ifnet *);
131 static void	if_route(struct ifnet *, int flag, int fam);
132 static int	if_setflag(struct ifnet *, int, int, int *, int);
133 static void	if_slowtimo(void *);
134 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
135 static void	if_unroute(struct ifnet *, int flag, int fam);
136 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
137 static int	if_rtdel(struct radix_node *, void *);
138 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
139 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
140 static void	if_start_deferred(void *context, int pending);
141 static void	do_link_state_change(void *, int);
142 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
143 static int	if_getgroupmembers(struct ifgroupreq *);
144 
145 #ifdef INET6
146 /*
147  * XXX: declare here to avoid to include many inet6 related files..
148  * should be more generalized?
149  */
150 extern void	nd6_setmtu(struct ifnet *);
151 #endif
152 
153 #ifdef VIMAGE_GLOBALS
154 struct	ifnethead ifnet;	/* depend on static init XXX */
155 struct	ifgrouphead ifg_head;
156 int	if_index;
157 static	int if_indexlim;
158 /* Table of ifnet/cdev by index.  Locked with ifnet_lock. */
159 static struct ifindex_entry *ifindex_table;
160 static struct	knlist ifklist;
161 #endif
162 
163 int	ifqmaxlen = IFQ_MAXLEN;
164 struct rwlock ifnet_lock;
165 static	if_com_alloc_t *if_com_alloc[256];
166 static	if_com_free_t *if_com_free[256];
167 
168 static void	filt_netdetach(struct knote *kn);
169 static int	filt_netdev(struct knote *kn, long hint);
170 
171 static struct filterops netdev_filtops =
172     { 1, NULL, filt_netdetach, filt_netdev };
173 
174 #ifndef VIMAGE_GLOBALS
175 static struct vnet_symmap vnet_net_symmap[] = {
176 	VNET_SYMMAP(net, ifnet),
177 	VNET_SYMMAP(net, rt_tables),
178 	VNET_SYMMAP(net, rtstat),
179 	VNET_SYMMAP(net, rttrash),
180 	VNET_SYMMAP_END
181 };
182 
183 VNET_MOD_DECLARE(NET, net, vnet_net_iattach, vnet_net_idetach,
184     NONE, vnet_net_symmap)
185 #endif
186 
187 /*
188  * System initialization
189  */
190 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
191 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_check, NULL);
192 
193 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
194 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
195 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
196 
197 static struct ifnet *
198 ifnet_byindex_locked(u_short idx)
199 {
200 	INIT_VNET_NET(curvnet);
201 	struct ifnet *ifp;
202 
203 	ifp = V_ifindex_table[idx].ife_ifnet;
204 	return (ifp);
205 }
206 
207 struct ifnet *
208 ifnet_byindex(u_short idx)
209 {
210 	struct ifnet *ifp;
211 
212 	IFNET_RLOCK();
213 	ifp = ifnet_byindex_locked(idx);
214 	IFNET_RUNLOCK();
215 	return (ifp);
216 }
217 
218 static void
219 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
220 {
221 	INIT_VNET_NET(curvnet);
222 
223 	IFNET_WLOCK_ASSERT();
224 
225 	V_ifindex_table[idx].ife_ifnet = ifp;
226 }
227 
228 struct ifaddr *
229 ifaddr_byindex(u_short idx)
230 {
231 	struct ifaddr *ifa;
232 
233 	IFNET_RLOCK();
234 	ifa = ifnet_byindex_locked(idx)->if_addr;
235 	IFNET_RUNLOCK();
236 	return (ifa);
237 }
238 
239 struct cdev *
240 ifdev_byindex(u_short idx)
241 {
242 	INIT_VNET_NET(curvnet);
243 	struct cdev *cdev;
244 
245 	IFNET_RLOCK();
246 	cdev = V_ifindex_table[idx].ife_dev;
247 	IFNET_RUNLOCK();
248 	return (cdev);
249 }
250 
251 static void
252 ifdev_setbyindex(u_short idx, struct cdev *cdev)
253 {
254 	INIT_VNET_NET(curvnet);
255 
256 	IFNET_WLOCK();
257 	V_ifindex_table[idx].ife_dev = cdev;
258 	IFNET_WUNLOCK();
259 }
260 
261 static d_open_t		netopen;
262 static d_close_t	netclose;
263 static d_ioctl_t	netioctl;
264 static d_kqfilter_t	netkqfilter;
265 
266 static struct cdevsw net_cdevsw = {
267 	.d_version =	D_VERSION,
268 	.d_flags =	D_NEEDGIANT,
269 	.d_open =	netopen,
270 	.d_close =	netclose,
271 	.d_ioctl =	netioctl,
272 	.d_name =	"net",
273 	.d_kqfilter =	netkqfilter,
274 };
275 
276 static int
277 netopen(struct cdev *dev, int flag, int mode, struct thread *td)
278 {
279 	return (0);
280 }
281 
282 static int
283 netclose(struct cdev *dev, int flags, int fmt, struct thread *td)
284 {
285 	return (0);
286 }
287 
288 static int
289 netioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
290 {
291 	struct ifnet *ifp;
292 	int error, idx;
293 
294 	/* only support interface specific ioctls */
295 	if (IOCGROUP(cmd) != 'i')
296 		return (EOPNOTSUPP);
297 	idx = dev2unit(dev);
298 	if (idx == 0) {
299 		/*
300 		 * special network device, not interface.
301 		 */
302 		if (cmd == SIOCGIFCONF)
303 			return (ifconf(cmd, data));	/* XXX remove cmd */
304 #ifdef __amd64__
305 		if (cmd == SIOCGIFCONF32)
306 			return (ifconf(cmd, data));	/* XXX remove cmd */
307 #endif
308 		return (EOPNOTSUPP);
309 	}
310 
311 	ifp = ifnet_byindex(idx);
312 	if (ifp == NULL)
313 		return (ENXIO);
314 
315 	error = ifhwioctl(cmd, ifp, data, td);
316 	if (error == ENOIOCTL)
317 		error = EOPNOTSUPP;
318 	return (error);
319 }
320 
321 static int
322 netkqfilter(struct cdev *dev, struct knote *kn)
323 {
324 	INIT_VNET_NET(curvnet);
325 	struct knlist *klist;
326 	struct ifnet *ifp;
327 	int idx;
328 
329 	switch (kn->kn_filter) {
330 	case EVFILT_NETDEV:
331 		kn->kn_fop = &netdev_filtops;
332 		break;
333 	default:
334 		return (EINVAL);
335 	}
336 
337 	idx = dev2unit(dev);
338 	if (idx == 0) {
339 		klist = &V_ifklist;
340 	} else {
341 		ifp = ifnet_byindex(idx);
342 		if (ifp == NULL)
343 			return (1);
344 		klist = &ifp->if_klist;
345 	}
346 
347 	kn->kn_hook = (caddr_t)klist;
348 
349 	knlist_add(klist, kn, 0);
350 
351 	return (0);
352 }
353 
354 static void
355 filt_netdetach(struct knote *kn)
356 {
357 	struct knlist *klist = (struct knlist *)kn->kn_hook;
358 
359 	knlist_remove(klist, kn, 0);
360 }
361 
362 static int
363 filt_netdev(struct knote *kn, long hint)
364 {
365 	struct knlist *klist = (struct knlist *)kn->kn_hook;
366 
367 	/*
368 	 * Currently NOTE_EXIT is abused to indicate device detach.
369 	 */
370 	if (hint == NOTE_EXIT) {
371 		kn->kn_data = NOTE_LINKINV;
372 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
373 		knlist_remove_inevent(klist, kn);
374 		return (1);
375 	}
376 	if (hint != 0)
377 		kn->kn_data = hint;			/* current status */
378 	if (kn->kn_sfflags & hint)
379 		kn->kn_fflags |= hint;
380 	return (kn->kn_fflags != 0);
381 }
382 
383 /*
384  * Network interface utility routines.
385  *
386  * Routines with ifa_ifwith* names take sockaddr *'s as
387  * parameters.
388  */
389 
390 /* ARGSUSED*/
391 static void
392 if_init(void *dummy __unused)
393 {
394 	INIT_VNET_NET(curvnet);
395 
396 #ifndef VIMAGE_GLOBALS
397 	vnet_mod_register(&vnet_net_modinfo);
398 #endif
399 
400 	V_if_index = 0;
401 	V_ifindex_table = NULL;
402 	V_if_indexlim = 8;
403 
404 	IFNET_LOCK_INIT();
405 	TAILQ_INIT(&V_ifnet);
406 	TAILQ_INIT(&V_ifg_head);
407 	knlist_init(&V_ifklist, NULL, NULL, NULL, NULL);
408 	if_grow();				/* create initial table */
409 	ifdev_setbyindex(0, make_dev(&net_cdevsw, 0, UID_ROOT, GID_WHEEL,
410 	    0600, "network"));
411 	if_clone_init();
412 }
413 
414 static void
415 if_grow(void)
416 {
417 	INIT_VNET_NET(curvnet);
418 	u_int n;
419 	struct ifindex_entry *e;
420 
421 	V_if_indexlim <<= 1;
422 	n = V_if_indexlim * sizeof(*e);
423 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
424 	if (V_ifindex_table != NULL) {
425 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
426 		free((caddr_t)V_ifindex_table, M_IFNET);
427 	}
428 	V_ifindex_table = e;
429 }
430 
431 static void
432 if_check(void *dummy __unused)
433 {
434 
435 	/*
436 	 * If at least one interface added during boot uses
437 	 * if_watchdog then start the timer.
438 	 */
439 	if (slowtimo_started)
440 		if_slowtimo(0);
441 }
442 
443 /*
444  * Allocate a struct ifnet and an index for an interface.  A layer 2
445  * common structure will also be allocated if an allocation routine is
446  * registered for the passed type.
447  */
448 struct ifnet*
449 if_alloc(u_char type)
450 {
451 	INIT_VNET_NET(curvnet);
452 	struct ifnet *ifp;
453 
454 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
455 
456 	/*
457 	 * Try to find an empty slot below if_index.  If we fail, take
458 	 * the next slot.
459 	 *
460 	 * XXX: should be locked!
461 	 */
462 	for (ifp->if_index = 1; ifp->if_index <= V_if_index; ifp->if_index++) {
463 		if (ifnet_byindex(ifp->if_index) == NULL)
464 			break;
465 	}
466 	/* Catch if_index overflow. */
467 	if (ifp->if_index < 1) {
468 		free(ifp, M_IFNET);
469 		return (NULL);
470 	}
471 	if (ifp->if_index > V_if_index)
472 		V_if_index = ifp->if_index;
473 	if (V_if_index >= V_if_indexlim)
474 		if_grow();
475 
476 	ifp->if_type = type;
477 
478 	if (if_com_alloc[type] != NULL) {
479 		ifp->if_l2com = if_com_alloc[type](type, ifp);
480 		if (ifp->if_l2com == NULL) {
481 			free(ifp, M_IFNET);
482 			return (NULL);
483 		}
484 	}
485 	IFNET_WLOCK();
486 	ifnet_setbyindex(ifp->if_index, ifp);
487 	IFNET_WUNLOCK();
488 	IF_ADDR_LOCK_INIT(ifp);
489 
490 	return (ifp);
491 }
492 
493 /*
494  * Free the struct ifnet, the associated index, and the layer 2 common
495  * structure if needed.  All the work is done in if_free_type().
496  *
497  * Do not add code to this function!  Add it to if_free_type().
498  */
499 void
500 if_free(struct ifnet *ifp)
501 {
502 
503 	if_free_type(ifp, ifp->if_type);
504 }
505 
506 /*
507  * Do the actual work of freeing a struct ifnet, associated index, and
508  * layer 2 common structure.  This version should only be called by
509  * intefaces that switch their type after calling if_alloc().
510  */
511 void
512 if_free_type(struct ifnet *ifp, u_char type)
513 {
514 	INIT_VNET_NET(curvnet); /* ifp->if_vnet can be NULL here ! */
515 
516 	if (ifp != ifnet_byindex(ifp->if_index)) {
517 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
518 		    __func__);
519 		return;
520 	}
521 
522 	IFNET_WLOCK();
523 	ifnet_setbyindex(ifp->if_index, NULL);
524 
525 	/* XXX: should be locked with if_findindex() */
526 	while (V_if_index > 0 && ifnet_byindex_locked(V_if_index) == NULL)
527 		V_if_index--;
528 	IFNET_WUNLOCK();
529 
530 	if (if_com_free[type] != NULL)
531 		if_com_free[type](ifp->if_l2com, type);
532 
533 	IF_ADDR_LOCK_DESTROY(ifp);
534 	free(ifp, M_IFNET);
535 };
536 
537 void
538 ifq_attach(struct ifaltq *ifq, struct ifnet *ifp)
539 {
540 
541 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
542 
543 	if (ifq->ifq_maxlen == 0)
544 		ifq->ifq_maxlen = ifqmaxlen;
545 
546 	ifq->altq_type = 0;
547 	ifq->altq_disc = NULL;
548 	ifq->altq_flags &= ALTQF_CANTCHANGE;
549 	ifq->altq_tbr  = NULL;
550 	ifq->altq_ifp  = ifp;
551 }
552 
553 void
554 ifq_detach(struct ifaltq *ifq)
555 {
556 	mtx_destroy(&ifq->ifq_mtx);
557 }
558 
559 /*
560  * Perform generic interface initalization tasks and attach the interface
561  * to the list of "active" interfaces.
562  *
563  * XXX:
564  *  - The decision to return void and thus require this function to
565  *    succeed is questionable.
566  *  - We do more initialization here then is probably a good idea.
567  *    Some of this should probably move to if_alloc().
568  *  - We should probably do more sanity checking.  For instance we don't
569  *    do anything to insure if_xname is unique or non-empty.
570  */
571 void
572 if_attach(struct ifnet *ifp)
573 {
574 	INIT_VNET_NET(curvnet);
575 	unsigned socksize, ifasize;
576 	int namelen, masklen;
577 	struct sockaddr_dl *sdl;
578 	struct ifaddr *ifa;
579 
580 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
581 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
582 		    ifp->if_xname);
583 
584 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
585 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
586 	IF_AFDATA_LOCK_INIT(ifp);
587 	ifp->if_afdata_initialized = 0;
588 
589 	TAILQ_INIT(&ifp->if_addrhead);
590 	TAILQ_INIT(&ifp->if_prefixhead);
591 	TAILQ_INIT(&ifp->if_multiaddrs);
592 	TAILQ_INIT(&ifp->if_groups);
593 
594 	if_addgroup(ifp, IFG_ALL);
595 
596 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
597 	getmicrotime(&ifp->if_lastchange);
598 	ifp->if_data.ifi_epoch = time_uptime;
599 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
600 	ifp->if_transmit = if_transmit;
601 	ifp->if_qflush = if_qflush;
602 #ifdef MAC
603 	mac_ifnet_init(ifp);
604 	mac_ifnet_create(ifp);
605 #endif
606 
607 	ifdev_setbyindex(ifp->if_index, make_dev(&net_cdevsw,
608 	    ifp->if_index, UID_ROOT, GID_WHEEL, 0600, "%s/%s",
609 	    net_cdevsw.d_name, ifp->if_xname));
610 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
611 	    net_cdevsw.d_name, ifp->if_index);
612 
613 	ifq_attach(&ifp->if_snd, ifp);
614 
615 	/*
616 	 * create a Link Level name for this device
617 	 */
618 	namelen = strlen(ifp->if_xname);
619 	/*
620 	 * Always save enough space for any possiable name so we can do
621 	 * a rename in place later.
622 	 */
623 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
624 	socksize = masklen + ifp->if_addrlen;
625 	if (socksize < sizeof(*sdl))
626 		socksize = sizeof(*sdl);
627 	socksize = roundup2(socksize, sizeof(long));
628 	ifasize = sizeof(*ifa) + 2 * socksize;
629 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
630 	IFA_LOCK_INIT(ifa);
631 	sdl = (struct sockaddr_dl *)(ifa + 1);
632 	sdl->sdl_len = socksize;
633 	sdl->sdl_family = AF_LINK;
634 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
635 	sdl->sdl_nlen = namelen;
636 	sdl->sdl_index = ifp->if_index;
637 	sdl->sdl_type = ifp->if_type;
638 	ifp->if_addr = ifa;
639 	ifa->ifa_ifp = ifp;
640 	ifa->ifa_rtrequest = link_rtrequest;
641 	ifa->ifa_addr = (struct sockaddr *)sdl;
642 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
643 	ifa->ifa_netmask = (struct sockaddr *)sdl;
644 	sdl->sdl_len = masklen;
645 	while (namelen != 0)
646 		sdl->sdl_data[--namelen] = 0xff;
647 	ifa->ifa_refcnt = 1;
648 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
649 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
650 
651 
652 	IFNET_WLOCK();
653 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
654 	IFNET_WUNLOCK();
655 
656 	if (domain_init_status >= 2)
657 		if_attachdomain1(ifp);
658 
659 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
660 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
661 
662 	/* Announce the interface. */
663 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
664 
665 	if (ifp->if_watchdog != NULL) {
666 		if_printf(ifp,
667 		    "WARNING: using obsoleted if_watchdog interface\n");
668 
669 		/*
670 		 * Note that we need if_slowtimo().  If this happens after
671 		 * boot, then call if_slowtimo() directly.
672 		 */
673 		if (atomic_cmpset_int(&slowtimo_started, 0, 1) && !cold)
674 			if_slowtimo(0);
675 	}
676 	if (ifp->if_flags & IFF_NEEDSGIANT)
677 		if_printf(ifp,
678 		    "WARNING: using obsoleted IFF_NEEDSGIANT flag\n");
679 }
680 
681 static void
682 if_attachdomain(void *dummy)
683 {
684 	INIT_VNET_NET(curvnet);
685 	struct ifnet *ifp;
686 	int s;
687 
688 	s = splnet();
689 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
690 		if_attachdomain1(ifp);
691 	splx(s);
692 }
693 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
694     if_attachdomain, NULL);
695 
696 static void
697 if_attachdomain1(struct ifnet *ifp)
698 {
699 	struct domain *dp;
700 	int s;
701 
702 	s = splnet();
703 
704 	/*
705 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
706 	 * cannot lock ifp->if_afdata initialization, entirely.
707 	 */
708 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
709 		splx(s);
710 		return;
711 	}
712 	if (ifp->if_afdata_initialized >= domain_init_status) {
713 		IF_AFDATA_UNLOCK(ifp);
714 		splx(s);
715 		printf("if_attachdomain called more than once on %s\n",
716 		    ifp->if_xname);
717 		return;
718 	}
719 	ifp->if_afdata_initialized = domain_init_status;
720 	IF_AFDATA_UNLOCK(ifp);
721 
722 	/* address family dependent data region */
723 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
724 	for (dp = domains; dp; dp = dp->dom_next) {
725 		if (dp->dom_ifattach)
726 			ifp->if_afdata[dp->dom_family] =
727 			    (*dp->dom_ifattach)(ifp);
728 	}
729 
730 	splx(s);
731 }
732 
733 /*
734  * Remove any unicast or broadcast network addresses from an interface.
735  */
736 void
737 if_purgeaddrs(struct ifnet *ifp)
738 {
739 	struct ifaddr *ifa, *next;
740 
741 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
742 		if (ifa->ifa_addr->sa_family == AF_LINK)
743 			continue;
744 #ifdef INET
745 		/* XXX: Ugly!! ad hoc just for INET */
746 		if (ifa->ifa_addr->sa_family == AF_INET) {
747 			struct ifaliasreq ifr;
748 
749 			bzero(&ifr, sizeof(ifr));
750 			ifr.ifra_addr = *ifa->ifa_addr;
751 			if (ifa->ifa_dstaddr)
752 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
753 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
754 			    NULL) == 0)
755 				continue;
756 		}
757 #endif /* INET */
758 #ifdef INET6
759 		if (ifa->ifa_addr->sa_family == AF_INET6) {
760 			in6_purgeaddr(ifa);
761 			/* ifp_addrhead is already updated */
762 			continue;
763 		}
764 #endif /* INET6 */
765 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
766 		IFAFREE(ifa);
767 	}
768 }
769 
770 /*
771  * Remove any multicast network addresses from an interface.
772  */
773 void
774 if_purgemaddrs(struct ifnet *ifp)
775 {
776 	struct ifmultiaddr *ifma;
777 	struct ifmultiaddr *next;
778 
779 	IF_ADDR_LOCK(ifp);
780 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
781 		if_delmulti_locked(ifp, ifma, 1);
782 	IF_ADDR_UNLOCK(ifp);
783 }
784 
785 /*
786  * Detach an interface, removing it from the
787  * list of "active" interfaces.
788  *
789  * XXXRW: There are some significant questions about event ordering, and
790  * how to prevent things from starting to use the interface during detach.
791  */
792 void
793 if_detach(struct ifnet *ifp)
794 {
795 	INIT_VNET_NET(ifp->if_vnet);
796 	struct ifaddr *ifa;
797 	struct radix_node_head	*rnh;
798 	int s, i, j;
799 	struct domain *dp;
800  	struct ifnet *iter;
801  	int found = 0;
802 
803 	IFNET_WLOCK();
804 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
805 		if (iter == ifp) {
806 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
807 			found = 1;
808 			break;
809 		}
810 	IFNET_WUNLOCK();
811 	if (!found)
812 		return;
813 
814 	/*
815 	 * Remove/wait for pending events.
816 	 */
817 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
818 
819 	/*
820 	 * Remove routes and flush queues.
821 	 */
822 	s = splnet();
823 	if_down(ifp);
824 #ifdef ALTQ
825 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
826 		altq_disable(&ifp->if_snd);
827 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
828 		altq_detach(&ifp->if_snd);
829 #endif
830 
831 	if_purgeaddrs(ifp);
832 
833 #ifdef INET
834 	in_ifdetach(ifp);
835 #endif
836 
837 #ifdef INET6
838 	/*
839 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
840 	 * before removing routing entries below, since IPv6 interface direct
841 	 * routes are expected to be removed by the IPv6-specific kernel API.
842 	 * Otherwise, the kernel will detect some inconsistency and bark it.
843 	 */
844 	in6_ifdetach(ifp);
845 #endif
846 	if_purgemaddrs(ifp);
847 
848 	/*
849 	 * Remove link ifaddr pointer and maybe decrement if_index.
850 	 * Clean up all addresses.
851 	 */
852 	ifp->if_addr = NULL;
853 	destroy_dev(ifdev_byindex(ifp->if_index));
854 	ifdev_setbyindex(ifp->if_index, NULL);
855 
856 	/* We can now free link ifaddr. */
857 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
858 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
859 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
860 		IFAFREE(ifa);
861 	}
862 
863 	/*
864 	 * Delete all remaining routes using this interface
865 	 * Unfortuneatly the only way to do this is to slog through
866 	 * the entire routing table looking for routes which point
867 	 * to this interface...oh well...
868 	 */
869 	for (i = 1; i <= AF_MAX; i++) {
870 		for (j = 0; j < rt_numfibs; j++) {
871 			if ((rnh = V_rt_tables[j][i]) == NULL)
872 				continue;
873 			RADIX_NODE_HEAD_LOCK(rnh);
874 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
875 			RADIX_NODE_HEAD_UNLOCK(rnh);
876 		}
877 	}
878 
879 	/* Announce that the interface is gone. */
880 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
881 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
882 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
883 
884 	IF_AFDATA_LOCK(ifp);
885 	for (dp = domains; dp; dp = dp->dom_next) {
886 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
887 			(*dp->dom_ifdetach)(ifp,
888 			    ifp->if_afdata[dp->dom_family]);
889 	}
890 	IF_AFDATA_UNLOCK(ifp);
891 
892 #ifdef MAC
893 	mac_ifnet_destroy(ifp);
894 #endif /* MAC */
895 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
896 	knlist_clear(&ifp->if_klist, 0);
897 	knlist_destroy(&ifp->if_klist);
898 	ifq_detach(&ifp->if_snd);
899 	IF_AFDATA_DESTROY(ifp);
900 	splx(s);
901 }
902 
903 /*
904  * Add a group to an interface
905  */
906 int
907 if_addgroup(struct ifnet *ifp, const char *groupname)
908 {
909 	INIT_VNET_NET(ifp->if_vnet);
910 	struct ifg_list		*ifgl;
911 	struct ifg_group	*ifg = NULL;
912 	struct ifg_member	*ifgm;
913 
914 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
915 	    groupname[strlen(groupname) - 1] <= '9')
916 		return (EINVAL);
917 
918 	IFNET_WLOCK();
919 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
920 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
921 			IFNET_WUNLOCK();
922 			return (EEXIST);
923 		}
924 
925 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
926 	    M_NOWAIT)) == NULL) {
927 	    	IFNET_WUNLOCK();
928 		return (ENOMEM);
929 	}
930 
931 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
932 	    M_TEMP, M_NOWAIT)) == NULL) {
933 		free(ifgl, M_TEMP);
934 		IFNET_WUNLOCK();
935 		return (ENOMEM);
936 	}
937 
938 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
939 		if (!strcmp(ifg->ifg_group, groupname))
940 			break;
941 
942 	if (ifg == NULL) {
943 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
944 		    M_TEMP, M_NOWAIT)) == NULL) {
945 			free(ifgl, M_TEMP);
946 			free(ifgm, M_TEMP);
947 			IFNET_WUNLOCK();
948 			return (ENOMEM);
949 		}
950 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
951 		ifg->ifg_refcnt = 0;
952 		TAILQ_INIT(&ifg->ifg_members);
953 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
954 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
955 	}
956 
957 	ifg->ifg_refcnt++;
958 	ifgl->ifgl_group = ifg;
959 	ifgm->ifgm_ifp = ifp;
960 
961 	IF_ADDR_LOCK(ifp);
962 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
963 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
964 	IF_ADDR_UNLOCK(ifp);
965 
966 	IFNET_WUNLOCK();
967 
968 	EVENTHANDLER_INVOKE(group_change_event, groupname);
969 
970 	return (0);
971 }
972 
973 /*
974  * Remove a group from an interface
975  */
976 int
977 if_delgroup(struct ifnet *ifp, const char *groupname)
978 {
979 	INIT_VNET_NET(ifp->if_vnet);
980 	struct ifg_list		*ifgl;
981 	struct ifg_member	*ifgm;
982 
983 	IFNET_WLOCK();
984 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
985 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
986 			break;
987 	if (ifgl == NULL) {
988 		IFNET_WUNLOCK();
989 		return (ENOENT);
990 	}
991 
992 	IF_ADDR_LOCK(ifp);
993 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
994 	IF_ADDR_UNLOCK(ifp);
995 
996 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
997 		if (ifgm->ifgm_ifp == ifp)
998 			break;
999 
1000 	if (ifgm != NULL) {
1001 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1002 		free(ifgm, M_TEMP);
1003 	}
1004 
1005 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1006 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1007 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1008 		free(ifgl->ifgl_group, M_TEMP);
1009 	}
1010 	IFNET_WUNLOCK();
1011 
1012 	free(ifgl, M_TEMP);
1013 
1014 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1015 
1016 	return (0);
1017 }
1018 
1019 /*
1020  * Stores all groups from an interface in memory pointed
1021  * to by data
1022  */
1023 static int
1024 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1025 {
1026 	int			 len, error;
1027 	struct ifg_list		*ifgl;
1028 	struct ifg_req		 ifgrq, *ifgp;
1029 	struct ifgroupreq	*ifgr = data;
1030 
1031 	if (ifgr->ifgr_len == 0) {
1032 		IF_ADDR_LOCK(ifp);
1033 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1034 			ifgr->ifgr_len += sizeof(struct ifg_req);
1035 		IF_ADDR_UNLOCK(ifp);
1036 		return (0);
1037 	}
1038 
1039 	len = ifgr->ifgr_len;
1040 	ifgp = ifgr->ifgr_groups;
1041 	/* XXX: wire */
1042 	IF_ADDR_LOCK(ifp);
1043 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1044 		if (len < sizeof(ifgrq)) {
1045 			IF_ADDR_UNLOCK(ifp);
1046 			return (EINVAL);
1047 		}
1048 		bzero(&ifgrq, sizeof ifgrq);
1049 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1050 		    sizeof(ifgrq.ifgrq_group));
1051 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1052 		    	IF_ADDR_UNLOCK(ifp);
1053 			return (error);
1054 		}
1055 		len -= sizeof(ifgrq);
1056 		ifgp++;
1057 	}
1058 	IF_ADDR_UNLOCK(ifp);
1059 
1060 	return (0);
1061 }
1062 
1063 /*
1064  * Stores all members of a group in memory pointed to by data
1065  */
1066 static int
1067 if_getgroupmembers(struct ifgroupreq *data)
1068 {
1069 	INIT_VNET_NET(curvnet);
1070 	struct ifgroupreq	*ifgr = data;
1071 	struct ifg_group	*ifg;
1072 	struct ifg_member	*ifgm;
1073 	struct ifg_req		 ifgrq, *ifgp;
1074 	int			 len, error;
1075 
1076 	IFNET_RLOCK();
1077 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1078 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1079 			break;
1080 	if (ifg == NULL) {
1081 		IFNET_RUNLOCK();
1082 		return (ENOENT);
1083 	}
1084 
1085 	if (ifgr->ifgr_len == 0) {
1086 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1087 			ifgr->ifgr_len += sizeof(ifgrq);
1088 		IFNET_RUNLOCK();
1089 		return (0);
1090 	}
1091 
1092 	len = ifgr->ifgr_len;
1093 	ifgp = ifgr->ifgr_groups;
1094 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1095 		if (len < sizeof(ifgrq)) {
1096 			IFNET_RUNLOCK();
1097 			return (EINVAL);
1098 		}
1099 		bzero(&ifgrq, sizeof ifgrq);
1100 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1101 		    sizeof(ifgrq.ifgrq_member));
1102 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1103 			IFNET_RUNLOCK();
1104 			return (error);
1105 		}
1106 		len -= sizeof(ifgrq);
1107 		ifgp++;
1108 	}
1109 	IFNET_RUNLOCK();
1110 
1111 	return (0);
1112 }
1113 
1114 /*
1115  * Delete Routes for a Network Interface
1116  *
1117  * Called for each routing entry via the rnh->rnh_walktree() call above
1118  * to delete all route entries referencing a detaching network interface.
1119  *
1120  * Arguments:
1121  *	rn	pointer to node in the routing table
1122  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1123  *
1124  * Returns:
1125  *	0	successful
1126  *	errno	failed - reason indicated
1127  *
1128  */
1129 static int
1130 if_rtdel(struct radix_node *rn, void *arg)
1131 {
1132 	struct rtentry	*rt = (struct rtentry *)rn;
1133 	struct ifnet	*ifp = arg;
1134 	int		err;
1135 
1136 	if (rt->rt_ifp == ifp) {
1137 
1138 		/*
1139 		 * Protect (sorta) against walktree recursion problems
1140 		 * with cloned routes
1141 		 */
1142 		if ((rt->rt_flags & RTF_UP) == 0)
1143 			return (0);
1144 
1145 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1146 				rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1147 				(struct rtentry **) NULL, rt->rt_fibnum);
1148 		if (err) {
1149 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1150 		}
1151 	}
1152 
1153 	return (0);
1154 }
1155 
1156 /*
1157  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1158  * structs used to represent other address families, it is necessary
1159  * to perform a different comparison.
1160  */
1161 
1162 #define	sa_equal(a1, a2)	\
1163 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1164 
1165 #define	sa_dl_equal(a1, a2)	\
1166 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1167 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1168 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1169 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1170 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1171 
1172 /*
1173  * Locate an interface based on a complete address.
1174  */
1175 /*ARGSUSED*/
1176 struct ifaddr *
1177 ifa_ifwithaddr(struct sockaddr *addr)
1178 {
1179 	INIT_VNET_NET(curvnet);
1180 	struct ifnet *ifp;
1181 	struct ifaddr *ifa;
1182 
1183 	IFNET_RLOCK();
1184 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
1185 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1186 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1187 				continue;
1188 			if (sa_equal(addr, ifa->ifa_addr))
1189 				goto done;
1190 			/* IP6 doesn't have broadcast */
1191 			if ((ifp->if_flags & IFF_BROADCAST) &&
1192 			    ifa->ifa_broadaddr &&
1193 			    ifa->ifa_broadaddr->sa_len != 0 &&
1194 			    sa_equal(ifa->ifa_broadaddr, addr))
1195 				goto done;
1196 		}
1197 	ifa = NULL;
1198 done:
1199 	IFNET_RUNLOCK();
1200 	return (ifa);
1201 }
1202 
1203 /*
1204  * Locate an interface based on the broadcast address.
1205  */
1206 /* ARGSUSED */
1207 struct ifaddr *
1208 ifa_ifwithbroadaddr(struct sockaddr *addr)
1209 {
1210 	INIT_VNET_NET(curvnet);
1211 	struct ifnet *ifp;
1212 	struct ifaddr *ifa;
1213 
1214 	IFNET_RLOCK();
1215 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
1216 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1217 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1218 				continue;
1219 			if ((ifp->if_flags & IFF_BROADCAST) &&
1220 			    ifa->ifa_broadaddr &&
1221 			    ifa->ifa_broadaddr->sa_len != 0 &&
1222 			    sa_equal(ifa->ifa_broadaddr, addr))
1223 				goto done;
1224 		}
1225 	ifa = NULL;
1226 done:
1227 	IFNET_RUNLOCK();
1228 	return (ifa);
1229 }
1230 
1231 /*
1232  * Locate the point to point interface with a given destination address.
1233  */
1234 /*ARGSUSED*/
1235 struct ifaddr *
1236 ifa_ifwithdstaddr(struct sockaddr *addr)
1237 {
1238 	INIT_VNET_NET(curvnet);
1239 	struct ifnet *ifp;
1240 	struct ifaddr *ifa;
1241 
1242 	IFNET_RLOCK();
1243 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1244 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1245 			continue;
1246 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1247 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1248 				continue;
1249 			if (ifa->ifa_dstaddr != NULL &&
1250 			    sa_equal(addr, ifa->ifa_dstaddr))
1251 				goto done;
1252 		}
1253 	}
1254 	ifa = NULL;
1255 done:
1256 	IFNET_RUNLOCK();
1257 	return (ifa);
1258 }
1259 
1260 /*
1261  * Find an interface on a specific network.  If many, choice
1262  * is most specific found.
1263  */
1264 struct ifaddr *
1265 ifa_ifwithnet(struct sockaddr *addr)
1266 {
1267 	INIT_VNET_NET(curvnet);
1268 	struct ifnet *ifp;
1269 	struct ifaddr *ifa;
1270 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
1271 	u_int af = addr->sa_family;
1272 	char *addr_data = addr->sa_data, *cplim;
1273 
1274 	/*
1275 	 * AF_LINK addresses can be looked up directly by their index number,
1276 	 * so do that if we can.
1277 	 */
1278 	if (af == AF_LINK) {
1279 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1280 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1281 		return (ifaddr_byindex(sdl->sdl_index));
1282 	}
1283 
1284 	/*
1285 	 * Scan though each interface, looking for ones that have
1286 	 * addresses in this address family.
1287 	 */
1288 	IFNET_RLOCK();
1289 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1290 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1291 			char *cp, *cp2, *cp3;
1292 
1293 			if (ifa->ifa_addr->sa_family != af)
1294 next:				continue;
1295 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1296 				/*
1297 				 * This is a bit broken as it doesn't
1298 				 * take into account that the remote end may
1299 				 * be a single node in the network we are
1300 				 * looking for.
1301 				 * The trouble is that we don't know the
1302 				 * netmask for the remote end.
1303 				 */
1304 				if (ifa->ifa_dstaddr != NULL &&
1305 				    sa_equal(addr, ifa->ifa_dstaddr))
1306 					goto done;
1307 			} else {
1308 				/*
1309 				 * if we have a special address handler,
1310 				 * then use it instead of the generic one.
1311 				 */
1312 				if (ifa->ifa_claim_addr) {
1313 					if ((*ifa->ifa_claim_addr)(ifa, addr))
1314 						goto done;
1315 					continue;
1316 				}
1317 
1318 				/*
1319 				 * Scan all the bits in the ifa's address.
1320 				 * If a bit dissagrees with what we are
1321 				 * looking for, mask it with the netmask
1322 				 * to see if it really matters.
1323 				 * (A byte at a time)
1324 				 */
1325 				if (ifa->ifa_netmask == 0)
1326 					continue;
1327 				cp = addr_data;
1328 				cp2 = ifa->ifa_addr->sa_data;
1329 				cp3 = ifa->ifa_netmask->sa_data;
1330 				cplim = ifa->ifa_netmask->sa_len
1331 					+ (char *)ifa->ifa_netmask;
1332 				while (cp3 < cplim)
1333 					if ((*cp++ ^ *cp2++) & *cp3++)
1334 						goto next; /* next address! */
1335 				/*
1336 				 * If the netmask of what we just found
1337 				 * is more specific than what we had before
1338 				 * (if we had one) then remember the new one
1339 				 * before continuing to search
1340 				 * for an even better one.
1341 				 */
1342 				if (ifa_maybe == 0 ||
1343 				    rn_refines((caddr_t)ifa->ifa_netmask,
1344 				    (caddr_t)ifa_maybe->ifa_netmask))
1345 					ifa_maybe = ifa;
1346 			}
1347 		}
1348 	}
1349 	ifa = ifa_maybe;
1350 done:
1351 	IFNET_RUNLOCK();
1352 	return (ifa);
1353 }
1354 
1355 /*
1356  * Find an interface address specific to an interface best matching
1357  * a given address.
1358  */
1359 struct ifaddr *
1360 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1361 {
1362 	struct ifaddr *ifa;
1363 	char *cp, *cp2, *cp3;
1364 	char *cplim;
1365 	struct ifaddr *ifa_maybe = 0;
1366 	u_int af = addr->sa_family;
1367 
1368 	if (af >= AF_MAX)
1369 		return (0);
1370 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1371 		if (ifa->ifa_addr->sa_family != af)
1372 			continue;
1373 		if (ifa_maybe == 0)
1374 			ifa_maybe = ifa;
1375 		if (ifa->ifa_netmask == 0) {
1376 			if (sa_equal(addr, ifa->ifa_addr) ||
1377 			    (ifa->ifa_dstaddr &&
1378 			    sa_equal(addr, ifa->ifa_dstaddr)))
1379 				goto done;
1380 			continue;
1381 		}
1382 		if (ifp->if_flags & IFF_POINTOPOINT) {
1383 			if (sa_equal(addr, ifa->ifa_dstaddr))
1384 				goto done;
1385 		} else {
1386 			cp = addr->sa_data;
1387 			cp2 = ifa->ifa_addr->sa_data;
1388 			cp3 = ifa->ifa_netmask->sa_data;
1389 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1390 			for (; cp3 < cplim; cp3++)
1391 				if ((*cp++ ^ *cp2++) & *cp3)
1392 					break;
1393 			if (cp3 == cplim)
1394 				goto done;
1395 		}
1396 	}
1397 	ifa = ifa_maybe;
1398 done:
1399 	return (ifa);
1400 }
1401 
1402 #include <net/route.h>
1403 #include <net/if_llatbl.h>
1404 
1405 /*
1406  * Default action when installing a route with a Link Level gateway.
1407  * Lookup an appropriate real ifa to point to.
1408  * This should be moved to /sys/net/link.c eventually.
1409  */
1410 static void
1411 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1412 {
1413 	struct ifaddr *ifa, *oifa;
1414 	struct sockaddr *dst;
1415 	struct ifnet *ifp;
1416 
1417 	RT_LOCK_ASSERT(rt);
1418 
1419 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1420 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1421 		return;
1422 	ifa = ifaof_ifpforaddr(dst, ifp);
1423 	if (ifa) {
1424 		IFAREF(ifa);		/* XXX */
1425 		oifa = rt->rt_ifa;
1426 		rt->rt_ifa = ifa;
1427 		IFAFREE(oifa);
1428 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1429 			ifa->ifa_rtrequest(cmd, rt, info);
1430 	}
1431 }
1432 
1433 /*
1434  * Mark an interface down and notify protocols of
1435  * the transition.
1436  * NOTE: must be called at splnet or eqivalent.
1437  */
1438 static void
1439 if_unroute(struct ifnet *ifp, int flag, int fam)
1440 {
1441 	struct ifaddr *ifa;
1442 
1443 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1444 
1445 	ifp->if_flags &= ~flag;
1446 	getmicrotime(&ifp->if_lastchange);
1447 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1448 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1449 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1450 	ifp->if_qflush(ifp);
1451 
1452 #ifdef DEV_CARP
1453 	if (ifp->if_carp)
1454 		carp_carpdev_state(ifp->if_carp);
1455 #endif
1456 	rt_ifmsg(ifp);
1457 }
1458 
1459 /*
1460  * Mark an interface up and notify protocols of
1461  * the transition.
1462  * NOTE: must be called at splnet or eqivalent.
1463  */
1464 static void
1465 if_route(struct ifnet *ifp, int flag, int fam)
1466 {
1467 	struct ifaddr *ifa;
1468 
1469 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1470 
1471 	ifp->if_flags |= flag;
1472 	getmicrotime(&ifp->if_lastchange);
1473 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1474 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1475 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1476 #ifdef DEV_CARP
1477 	if (ifp->if_carp)
1478 		carp_carpdev_state(ifp->if_carp);
1479 #endif
1480 	rt_ifmsg(ifp);
1481 #ifdef INET6
1482 	in6_if_up(ifp);
1483 #endif
1484 }
1485 
1486 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1487 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1488 
1489 /*
1490  * Handle a change in the interface link state. To avoid LORs
1491  * between driver lock and upper layer locks, as well as possible
1492  * recursions, we post event to taskqueue, and all job
1493  * is done in static do_link_state_change().
1494  */
1495 void
1496 if_link_state_change(struct ifnet *ifp, int link_state)
1497 {
1498 	/* Return if state hasn't changed. */
1499 	if (ifp->if_link_state == link_state)
1500 		return;
1501 
1502 	ifp->if_link_state = link_state;
1503 
1504 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1505 }
1506 
1507 static void
1508 do_link_state_change(void *arg, int pending)
1509 {
1510 	struct ifnet *ifp = (struct ifnet *)arg;
1511 	int link_state = ifp->if_link_state;
1512 	int link;
1513 	CURVNET_SET(ifp->if_vnet);
1514 
1515 	/* Notify that the link state has changed. */
1516 	rt_ifmsg(ifp);
1517 	if (link_state == LINK_STATE_UP)
1518 		link = NOTE_LINKUP;
1519 	else if (link_state == LINK_STATE_DOWN)
1520 		link = NOTE_LINKDOWN;
1521 	else
1522 		link = NOTE_LINKINV;
1523 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1524 	if (ifp->if_vlantrunk != NULL)
1525 		(*vlan_link_state_p)(ifp, link);
1526 
1527 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1528 	    IFP2AC(ifp)->ac_netgraph != NULL)
1529 		(*ng_ether_link_state_p)(ifp, link_state);
1530 #ifdef DEV_CARP
1531 	if (ifp->if_carp)
1532 		carp_carpdev_state(ifp->if_carp);
1533 #endif
1534 	if (ifp->if_bridge) {
1535 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1536 		(*bstp_linkstate_p)(ifp, link_state);
1537 	}
1538 	if (ifp->if_lagg) {
1539 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1540 		(*lagg_linkstate_p)(ifp, link_state);
1541 	}
1542 
1543 	devctl_notify("IFNET", ifp->if_xname,
1544 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1545 	if (pending > 1)
1546 		if_printf(ifp, "%d link states coalesced\n", pending);
1547 	if (log_link_state_change)
1548 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1549 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1550 	CURVNET_RESTORE();
1551 }
1552 
1553 /*
1554  * Mark an interface down and notify protocols of
1555  * the transition.
1556  * NOTE: must be called at splnet or eqivalent.
1557  */
1558 void
1559 if_down(struct ifnet *ifp)
1560 {
1561 
1562 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1563 }
1564 
1565 /*
1566  * Mark an interface up and notify protocols of
1567  * the transition.
1568  * NOTE: must be called at splnet or eqivalent.
1569  */
1570 void
1571 if_up(struct ifnet *ifp)
1572 {
1573 
1574 	if_route(ifp, IFF_UP, AF_UNSPEC);
1575 }
1576 
1577 /*
1578  * Flush an interface queue.
1579  */
1580 static void
1581 if_qflush(struct ifnet *ifp)
1582 {
1583 	struct mbuf *m, *n;
1584 	struct ifaltq *ifq;
1585 
1586 	ifq = &ifp->if_snd;
1587 	IFQ_LOCK(ifq);
1588 #ifdef ALTQ
1589 	if (ALTQ_IS_ENABLED(ifq))
1590 		ALTQ_PURGE(ifq);
1591 #endif
1592 	n = ifq->ifq_head;
1593 	while ((m = n) != 0) {
1594 		n = m->m_act;
1595 		m_freem(m);
1596 	}
1597 	ifq->ifq_head = 0;
1598 	ifq->ifq_tail = 0;
1599 	ifq->ifq_len = 0;
1600 	IFQ_UNLOCK(ifq);
1601 }
1602 
1603 /*
1604  * Handle interface watchdog timer routines.  Called
1605  * from softclock, we decrement timers (if set) and
1606  * call the appropriate interface routine on expiration.
1607  *
1608  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1609  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1610  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1611  */
1612 static void
1613 if_slowtimo(void *arg)
1614 {
1615 	VNET_ITERATOR_DECL(vnet_iter);
1616 	struct ifnet *ifp;
1617 	int s = splimp();
1618 
1619 	IFNET_RLOCK();
1620 	VNET_LIST_RLOCK();
1621 	VNET_FOREACH(vnet_iter) {
1622 		CURVNET_SET(vnet_iter);
1623 		INIT_VNET_NET(vnet_iter);
1624 		TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1625 			if (ifp->if_timer == 0 || --ifp->if_timer)
1626 				continue;
1627 			if (ifp->if_watchdog)
1628 				(*ifp->if_watchdog)(ifp);
1629 		}
1630 		CURVNET_RESTORE();
1631 	}
1632 	VNET_LIST_RUNLOCK();
1633 	IFNET_RUNLOCK();
1634 	splx(s);
1635 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1636 }
1637 
1638 /*
1639  * Map interface name to
1640  * interface structure pointer.
1641  */
1642 struct ifnet *
1643 ifunit(const char *name)
1644 {
1645 	INIT_VNET_NET(curvnet);
1646 	struct ifnet *ifp;
1647 
1648 	IFNET_RLOCK();
1649 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1650 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1651 			break;
1652 	}
1653 	IFNET_RUNLOCK();
1654 	return (ifp);
1655 }
1656 
1657 /*
1658  * Hardware specific interface ioctls.
1659  */
1660 static int
1661 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1662 {
1663 	struct ifreq *ifr;
1664 	struct ifstat *ifs;
1665 	int error = 0;
1666 	int new_flags, temp_flags;
1667 	size_t namelen, onamelen;
1668 	char new_name[IFNAMSIZ];
1669 	struct ifaddr *ifa;
1670 	struct sockaddr_dl *sdl;
1671 
1672 	ifr = (struct ifreq *)data;
1673 	switch (cmd) {
1674 	case SIOCGIFINDEX:
1675 		ifr->ifr_index = ifp->if_index;
1676 		break;
1677 
1678 	case SIOCGIFFLAGS:
1679 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1680 		ifr->ifr_flags = temp_flags & 0xffff;
1681 		ifr->ifr_flagshigh = temp_flags >> 16;
1682 		break;
1683 
1684 	case SIOCGIFCAP:
1685 		ifr->ifr_reqcap = ifp->if_capabilities;
1686 		ifr->ifr_curcap = ifp->if_capenable;
1687 		break;
1688 
1689 #ifdef MAC
1690 	case SIOCGIFMAC:
1691 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
1692 		break;
1693 #endif
1694 
1695 	case SIOCGIFMETRIC:
1696 		ifr->ifr_metric = ifp->if_metric;
1697 		break;
1698 
1699 	case SIOCGIFMTU:
1700 		ifr->ifr_mtu = ifp->if_mtu;
1701 		break;
1702 
1703 	case SIOCGIFPHYS:
1704 		ifr->ifr_phys = ifp->if_physical;
1705 		break;
1706 
1707 	case SIOCSIFFLAGS:
1708 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
1709 		if (error)
1710 			return (error);
1711 		/*
1712 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1713 		 * check, so we don't need special handling here yet.
1714 		 */
1715 		new_flags = (ifr->ifr_flags & 0xffff) |
1716 		    (ifr->ifr_flagshigh << 16);
1717 		if (ifp->if_flags & IFF_SMART) {
1718 			/* Smart drivers twiddle their own routes */
1719 		} else if (ifp->if_flags & IFF_UP &&
1720 		    (new_flags & IFF_UP) == 0) {
1721 			int s = splimp();
1722 			if_down(ifp);
1723 			splx(s);
1724 		} else if (new_flags & IFF_UP &&
1725 		    (ifp->if_flags & IFF_UP) == 0) {
1726 			int s = splimp();
1727 			if_up(ifp);
1728 			splx(s);
1729 		}
1730 		/* See if permanently promiscuous mode bit is about to flip */
1731 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1732 			if (new_flags & IFF_PPROMISC)
1733 				ifp->if_flags |= IFF_PROMISC;
1734 			else if (ifp->if_pcount == 0)
1735 				ifp->if_flags &= ~IFF_PROMISC;
1736 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1737 			    ifp->if_xname,
1738 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1739 		}
1740 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1741 			(new_flags &~ IFF_CANTCHANGE);
1742 		if (ifp->if_ioctl) {
1743 			IFF_LOCKGIANT(ifp);
1744 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1745 			IFF_UNLOCKGIANT(ifp);
1746 		}
1747 		getmicrotime(&ifp->if_lastchange);
1748 		break;
1749 
1750 	case SIOCSIFCAP:
1751 		error = priv_check(td, PRIV_NET_SETIFCAP);
1752 		if (error)
1753 			return (error);
1754 		if (ifp->if_ioctl == NULL)
1755 			return (EOPNOTSUPP);
1756 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1757 			return (EINVAL);
1758 		IFF_LOCKGIANT(ifp);
1759 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1760 		IFF_UNLOCKGIANT(ifp);
1761 		if (error == 0)
1762 			getmicrotime(&ifp->if_lastchange);
1763 		break;
1764 
1765 #ifdef MAC
1766 	case SIOCSIFMAC:
1767 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
1768 		break;
1769 #endif
1770 
1771 	case SIOCSIFNAME:
1772 		error = priv_check(td, PRIV_NET_SETIFNAME);
1773 		if (error)
1774 			return (error);
1775 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1776 		if (error != 0)
1777 			return (error);
1778 		if (new_name[0] == '\0')
1779 			return (EINVAL);
1780 		if (ifunit(new_name) != NULL)
1781 			return (EEXIST);
1782 
1783 		/* Announce the departure of the interface. */
1784 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1785 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1786 
1787 		log(LOG_INFO, "%s: changing name to '%s'\n",
1788 		    ifp->if_xname, new_name);
1789 
1790 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1791 		ifa = ifp->if_addr;
1792 		IFA_LOCK(ifa);
1793 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1794 		namelen = strlen(new_name);
1795 		onamelen = sdl->sdl_nlen;
1796 		/*
1797 		 * Move the address if needed.  This is safe because we
1798 		 * allocate space for a name of length IFNAMSIZ when we
1799 		 * create this in if_attach().
1800 		 */
1801 		if (namelen != onamelen) {
1802 			bcopy(sdl->sdl_data + onamelen,
1803 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1804 		}
1805 		bcopy(new_name, sdl->sdl_data, namelen);
1806 		sdl->sdl_nlen = namelen;
1807 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1808 		bzero(sdl->sdl_data, onamelen);
1809 		while (namelen != 0)
1810 			sdl->sdl_data[--namelen] = 0xff;
1811 		IFA_UNLOCK(ifa);
1812 
1813 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1814 		/* Announce the return of the interface. */
1815 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1816 		break;
1817 
1818 	case SIOCSIFMETRIC:
1819 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
1820 		if (error)
1821 			return (error);
1822 		ifp->if_metric = ifr->ifr_metric;
1823 		getmicrotime(&ifp->if_lastchange);
1824 		break;
1825 
1826 	case SIOCSIFPHYS:
1827 		error = priv_check(td, PRIV_NET_SETIFPHYS);
1828 		if (error)
1829 			return (error);
1830 		if (ifp->if_ioctl == NULL)
1831 			return (EOPNOTSUPP);
1832 		IFF_LOCKGIANT(ifp);
1833 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1834 		IFF_UNLOCKGIANT(ifp);
1835 		if (error == 0)
1836 			getmicrotime(&ifp->if_lastchange);
1837 		break;
1838 
1839 	case SIOCSIFMTU:
1840 	{
1841 		u_long oldmtu = ifp->if_mtu;
1842 
1843 		error = priv_check(td, PRIV_NET_SETIFMTU);
1844 		if (error)
1845 			return (error);
1846 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1847 			return (EINVAL);
1848 		if (ifp->if_ioctl == NULL)
1849 			return (EOPNOTSUPP);
1850 		IFF_LOCKGIANT(ifp);
1851 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1852 		IFF_UNLOCKGIANT(ifp);
1853 		if (error == 0) {
1854 			getmicrotime(&ifp->if_lastchange);
1855 			rt_ifmsg(ifp);
1856 		}
1857 		/*
1858 		 * If the link MTU changed, do network layer specific procedure.
1859 		 */
1860 		if (ifp->if_mtu != oldmtu) {
1861 #ifdef INET6
1862 			nd6_setmtu(ifp);
1863 #endif
1864 		}
1865 		break;
1866 	}
1867 
1868 	case SIOCADDMULTI:
1869 	case SIOCDELMULTI:
1870 		if (cmd == SIOCADDMULTI)
1871 			error = priv_check(td, PRIV_NET_ADDMULTI);
1872 		else
1873 			error = priv_check(td, PRIV_NET_DELMULTI);
1874 		if (error)
1875 			return (error);
1876 
1877 		/* Don't allow group membership on non-multicast interfaces. */
1878 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1879 			return (EOPNOTSUPP);
1880 
1881 		/* Don't let users screw up protocols' entries. */
1882 		if (ifr->ifr_addr.sa_family != AF_LINK)
1883 			return (EINVAL);
1884 
1885 		if (cmd == SIOCADDMULTI) {
1886 			struct ifmultiaddr *ifma;
1887 
1888 			/*
1889 			 * Userland is only permitted to join groups once
1890 			 * via the if_addmulti() KPI, because it cannot hold
1891 			 * struct ifmultiaddr * between calls. It may also
1892 			 * lose a race while we check if the membership
1893 			 * already exists.
1894 			 */
1895 			IF_ADDR_LOCK(ifp);
1896 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
1897 			IF_ADDR_UNLOCK(ifp);
1898 			if (ifma != NULL)
1899 				error = EADDRINUSE;
1900 			else
1901 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1902 		} else {
1903 			error = if_delmulti(ifp, &ifr->ifr_addr);
1904 		}
1905 		if (error == 0)
1906 			getmicrotime(&ifp->if_lastchange);
1907 		break;
1908 
1909 	case SIOCSIFPHYADDR:
1910 	case SIOCDIFPHYADDR:
1911 #ifdef INET6
1912 	case SIOCSIFPHYADDR_IN6:
1913 #endif
1914 	case SIOCSLIFPHYADDR:
1915 	case SIOCSIFMEDIA:
1916 	case SIOCSIFGENERIC:
1917 		error = priv_check(td, PRIV_NET_HWIOCTL);
1918 		if (error)
1919 			return (error);
1920 		if (ifp->if_ioctl == NULL)
1921 			return (EOPNOTSUPP);
1922 		IFF_LOCKGIANT(ifp);
1923 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1924 		IFF_UNLOCKGIANT(ifp);
1925 		if (error == 0)
1926 			getmicrotime(&ifp->if_lastchange);
1927 		break;
1928 
1929 	case SIOCGIFSTATUS:
1930 		ifs = (struct ifstat *)data;
1931 		ifs->ascii[0] = '\0';
1932 
1933 	case SIOCGIFPSRCADDR:
1934 	case SIOCGIFPDSTADDR:
1935 	case SIOCGLIFPHYADDR:
1936 	case SIOCGIFMEDIA:
1937 	case SIOCGIFGENERIC:
1938 		if (ifp->if_ioctl == NULL)
1939 			return (EOPNOTSUPP);
1940 		IFF_LOCKGIANT(ifp);
1941 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1942 		IFF_UNLOCKGIANT(ifp);
1943 		break;
1944 
1945 	case SIOCSIFLLADDR:
1946 		error = priv_check(td, PRIV_NET_SETLLADDR);
1947 		if (error)
1948 			return (error);
1949 		error = if_setlladdr(ifp,
1950 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1951 		break;
1952 
1953 	case SIOCAIFGROUP:
1954 	{
1955 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1956 
1957 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
1958 		if (error)
1959 			return (error);
1960 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
1961 			return (error);
1962 		break;
1963 	}
1964 
1965 	case SIOCGIFGROUP:
1966 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
1967 			return (error);
1968 		break;
1969 
1970 	case SIOCDIFGROUP:
1971 	{
1972 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1973 
1974 		error = priv_check(td, PRIV_NET_DELIFGROUP);
1975 		if (error)
1976 			return (error);
1977 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
1978 			return (error);
1979 		break;
1980 	}
1981 
1982 	default:
1983 		error = ENOIOCTL;
1984 		break;
1985 	}
1986 	return (error);
1987 }
1988 
1989 /*
1990  * Interface ioctls.
1991  */
1992 int
1993 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1994 {
1995 	struct ifnet *ifp;
1996 	struct ifreq *ifr;
1997 	int error;
1998 	int oif_flags;
1999 
2000 	switch (cmd) {
2001 	case SIOCGIFCONF:
2002 	case OSIOCGIFCONF:
2003 #ifdef __amd64__
2004 	case SIOCGIFCONF32:
2005 #endif
2006 		return (ifconf(cmd, data));
2007 	}
2008 	ifr = (struct ifreq *)data;
2009 
2010 	switch (cmd) {
2011 	case SIOCIFCREATE:
2012 	case SIOCIFCREATE2:
2013 		error = priv_check(td, PRIV_NET_IFCREATE);
2014 		if (error)
2015 			return (error);
2016 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
2017 			cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
2018 	case SIOCIFDESTROY:
2019 		error = priv_check(td, PRIV_NET_IFDESTROY);
2020 		if (error)
2021 			return (error);
2022 		return if_clone_destroy(ifr->ifr_name);
2023 
2024 	case SIOCIFGCLONERS:
2025 		return (if_clone_list((struct if_clonereq *)data));
2026 	case SIOCGIFGMEMB:
2027 		return (if_getgroupmembers((struct ifgroupreq *)data));
2028 	}
2029 
2030 	ifp = ifunit(ifr->ifr_name);
2031 	if (ifp == 0)
2032 		return (ENXIO);
2033 
2034 	error = ifhwioctl(cmd, ifp, data, td);
2035 	if (error != ENOIOCTL)
2036 		return (error);
2037 
2038 	oif_flags = ifp->if_flags;
2039 	if (so->so_proto == 0)
2040 		return (EOPNOTSUPP);
2041 #ifndef COMPAT_43
2042 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2043 								 data,
2044 								 ifp, td));
2045 #else
2046 	{
2047 		int ocmd = cmd;
2048 
2049 		switch (cmd) {
2050 
2051 		case SIOCSIFDSTADDR:
2052 		case SIOCSIFADDR:
2053 		case SIOCSIFBRDADDR:
2054 		case SIOCSIFNETMASK:
2055 #if BYTE_ORDER != BIG_ENDIAN
2056 			if (ifr->ifr_addr.sa_family == 0 &&
2057 			    ifr->ifr_addr.sa_len < 16) {
2058 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2059 				ifr->ifr_addr.sa_len = 16;
2060 			}
2061 #else
2062 			if (ifr->ifr_addr.sa_len == 0)
2063 				ifr->ifr_addr.sa_len = 16;
2064 #endif
2065 			break;
2066 
2067 		case OSIOCGIFADDR:
2068 			cmd = SIOCGIFADDR;
2069 			break;
2070 
2071 		case OSIOCGIFDSTADDR:
2072 			cmd = SIOCGIFDSTADDR;
2073 			break;
2074 
2075 		case OSIOCGIFBRDADDR:
2076 			cmd = SIOCGIFBRDADDR;
2077 			break;
2078 
2079 		case OSIOCGIFNETMASK:
2080 			cmd = SIOCGIFNETMASK;
2081 		}
2082 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2083 								   cmd,
2084 								   data,
2085 								   ifp, td));
2086 		switch (ocmd) {
2087 
2088 		case OSIOCGIFADDR:
2089 		case OSIOCGIFDSTADDR:
2090 		case OSIOCGIFBRDADDR:
2091 		case OSIOCGIFNETMASK:
2092 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2093 
2094 		}
2095 	}
2096 #endif /* COMPAT_43 */
2097 
2098 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2099 #ifdef INET6
2100 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
2101 		if (ifp->if_flags & IFF_UP) {
2102 			int s = splimp();
2103 			in6_if_up(ifp);
2104 			splx(s);
2105 		}
2106 #endif
2107 	}
2108 	return (error);
2109 }
2110 
2111 /*
2112  * The code common to handling reference counted flags,
2113  * e.g., in ifpromisc() and if_allmulti().
2114  * The "pflag" argument can specify a permanent mode flag to check,
2115  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2116  *
2117  * Only to be used on stack-owned flags, not driver-owned flags.
2118  */
2119 static int
2120 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2121 {
2122 	struct ifreq ifr;
2123 	int error;
2124 	int oldflags, oldcount;
2125 
2126 	/* Sanity checks to catch programming errors */
2127 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2128 	    ("%s: setting driver-owned flag %d", __func__, flag));
2129 
2130 	if (onswitch)
2131 		KASSERT(*refcount >= 0,
2132 		    ("%s: increment negative refcount %d for flag %d",
2133 		    __func__, *refcount, flag));
2134 	else
2135 		KASSERT(*refcount > 0,
2136 		    ("%s: decrement non-positive refcount %d for flag %d",
2137 		    __func__, *refcount, flag));
2138 
2139 	/* In case this mode is permanent, just touch refcount */
2140 	if (ifp->if_flags & pflag) {
2141 		*refcount += onswitch ? 1 : -1;
2142 		return (0);
2143 	}
2144 
2145 	/* Save ifnet parameters for if_ioctl() may fail */
2146 	oldcount = *refcount;
2147 	oldflags = ifp->if_flags;
2148 
2149 	/*
2150 	 * See if we aren't the only and touching refcount is enough.
2151 	 * Actually toggle interface flag if we are the first or last.
2152 	 */
2153 	if (onswitch) {
2154 		if ((*refcount)++)
2155 			return (0);
2156 		ifp->if_flags |= flag;
2157 	} else {
2158 		if (--(*refcount))
2159 			return (0);
2160 		ifp->if_flags &= ~flag;
2161 	}
2162 
2163 	/* Call down the driver since we've changed interface flags */
2164 	if (ifp->if_ioctl == NULL) {
2165 		error = EOPNOTSUPP;
2166 		goto recover;
2167 	}
2168 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2169 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2170 	IFF_LOCKGIANT(ifp);
2171 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2172 	IFF_UNLOCKGIANT(ifp);
2173 	if (error)
2174 		goto recover;
2175 	/* Notify userland that interface flags have changed */
2176 	rt_ifmsg(ifp);
2177 	return (0);
2178 
2179 recover:
2180 	/* Recover after driver error */
2181 	*refcount = oldcount;
2182 	ifp->if_flags = oldflags;
2183 	return (error);
2184 }
2185 
2186 /*
2187  * Set/clear promiscuous mode on interface ifp based on the truth value
2188  * of pswitch.  The calls are reference counted so that only the first
2189  * "on" request actually has an effect, as does the final "off" request.
2190  * Results are undefined if the "off" and "on" requests are not matched.
2191  */
2192 int
2193 ifpromisc(struct ifnet *ifp, int pswitch)
2194 {
2195 	int error;
2196 	int oldflags = ifp->if_flags;
2197 
2198 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2199 			   &ifp->if_pcount, pswitch);
2200 	/* If promiscuous mode status has changed, log a message */
2201 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2202 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2203 		    ifp->if_xname,
2204 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2205 	return (error);
2206 }
2207 
2208 /*
2209  * Return interface configuration
2210  * of system.  List may be used
2211  * in later ioctl's (above) to get
2212  * other information.
2213  */
2214 /*ARGSUSED*/
2215 static int
2216 ifconf(u_long cmd, caddr_t data)
2217 {
2218 	INIT_VNET_NET(curvnet);
2219 	struct ifconf *ifc = (struct ifconf *)data;
2220 #ifdef __amd64__
2221 	struct ifconf32 *ifc32 = (struct ifconf32 *)data;
2222 	struct ifconf ifc_swab;
2223 #endif
2224 	struct ifnet *ifp;
2225 	struct ifaddr *ifa;
2226 	struct ifreq ifr;
2227 	struct sbuf *sb;
2228 	int error, full = 0, valid_len, max_len;
2229 
2230 #ifdef __amd64__
2231 	if (cmd == SIOCGIFCONF32) {
2232 		ifc_swab.ifc_len = ifc32->ifc_len;
2233 		ifc_swab.ifc_buf = (caddr_t)(uintptr_t)ifc32->ifc_buf;
2234 		ifc = &ifc_swab;
2235 	}
2236 #endif
2237 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2238 	max_len = MAXPHYS - 1;
2239 
2240 	/* Prevent hostile input from being able to crash the system */
2241 	if (ifc->ifc_len <= 0)
2242 		return (EINVAL);
2243 
2244 again:
2245 	if (ifc->ifc_len <= max_len) {
2246 		max_len = ifc->ifc_len;
2247 		full = 1;
2248 	}
2249 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2250 	max_len = 0;
2251 	valid_len = 0;
2252 
2253 	IFNET_RLOCK();		/* could sleep XXX */
2254 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2255 		int addrs;
2256 
2257 		/*
2258 		 * Zero the ifr_name buffer to make sure we don't
2259 		 * disclose the contents of the stack.
2260 		 */
2261 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2262 
2263 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2264 		    >= sizeof(ifr.ifr_name)) {
2265 			sbuf_delete(sb);
2266 			IFNET_RUNLOCK();
2267 			return (ENAMETOOLONG);
2268 		}
2269 
2270 		addrs = 0;
2271 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2272 			struct sockaddr *sa = ifa->ifa_addr;
2273 
2274 			if (prison_if(curthread->td_ucred, sa) != 0)
2275 				continue;
2276 			addrs++;
2277 #ifdef COMPAT_43
2278 			if (cmd == OSIOCGIFCONF) {
2279 				struct osockaddr *osa =
2280 					 (struct osockaddr *)&ifr.ifr_addr;
2281 				ifr.ifr_addr = *sa;
2282 				osa->sa_family = sa->sa_family;
2283 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2284 				max_len += sizeof(ifr);
2285 			} else
2286 #endif
2287 			if (sa->sa_len <= sizeof(*sa)) {
2288 				ifr.ifr_addr = *sa;
2289 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2290 				max_len += sizeof(ifr);
2291 			} else {
2292 				sbuf_bcat(sb, &ifr,
2293 				    offsetof(struct ifreq, ifr_addr));
2294 				max_len += offsetof(struct ifreq, ifr_addr);
2295 				sbuf_bcat(sb, sa, sa->sa_len);
2296 				max_len += sa->sa_len;
2297 			}
2298 
2299 			if (!sbuf_overflowed(sb))
2300 				valid_len = sbuf_len(sb);
2301 		}
2302 		if (addrs == 0) {
2303 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2304 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2305 			max_len += sizeof(ifr);
2306 
2307 			if (!sbuf_overflowed(sb))
2308 				valid_len = sbuf_len(sb);
2309 		}
2310 	}
2311 	IFNET_RUNLOCK();
2312 
2313 	/*
2314 	 * If we didn't allocate enough space (uncommon), try again.  If
2315 	 * we have already allocated as much space as we are allowed,
2316 	 * return what we've got.
2317 	 */
2318 	if (valid_len != max_len && !full) {
2319 		sbuf_delete(sb);
2320 		goto again;
2321 	}
2322 
2323 	ifc->ifc_len = valid_len;
2324 #ifdef __amd64__
2325 	if (cmd == SIOCGIFCONF32)
2326 		ifc32->ifc_len = valid_len;
2327 #endif
2328 	sbuf_finish(sb);
2329 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2330 	sbuf_delete(sb);
2331 	return (error);
2332 }
2333 
2334 /*
2335  * Just like ifpromisc(), but for all-multicast-reception mode.
2336  */
2337 int
2338 if_allmulti(struct ifnet *ifp, int onswitch)
2339 {
2340 
2341 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2342 }
2343 
2344 struct ifmultiaddr *
2345 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2346 {
2347 	struct ifmultiaddr *ifma;
2348 
2349 	IF_ADDR_LOCK_ASSERT(ifp);
2350 
2351 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2352 		if (sa->sa_family == AF_LINK) {
2353 			if (sa_dl_equal(ifma->ifma_addr, sa))
2354 				break;
2355 		} else {
2356 			if (sa_equal(ifma->ifma_addr, sa))
2357 				break;
2358 		}
2359 	}
2360 
2361 	return ifma;
2362 }
2363 
2364 /*
2365  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2366  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2367  * the ifnet multicast address list here, so the caller must do that and
2368  * other setup work (such as notifying the device driver).  The reference
2369  * count is initialized to 1.
2370  */
2371 static struct ifmultiaddr *
2372 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2373     int mflags)
2374 {
2375 	struct ifmultiaddr *ifma;
2376 	struct sockaddr *dupsa;
2377 
2378 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2379 	    M_ZERO);
2380 	if (ifma == NULL)
2381 		return (NULL);
2382 
2383 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2384 	if (dupsa == NULL) {
2385 		free(ifma, M_IFMADDR);
2386 		return (NULL);
2387 	}
2388 	bcopy(sa, dupsa, sa->sa_len);
2389 	ifma->ifma_addr = dupsa;
2390 
2391 	ifma->ifma_ifp = ifp;
2392 	ifma->ifma_refcount = 1;
2393 	ifma->ifma_protospec = NULL;
2394 
2395 	if (llsa == NULL) {
2396 		ifma->ifma_lladdr = NULL;
2397 		return (ifma);
2398 	}
2399 
2400 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2401 	if (dupsa == NULL) {
2402 		free(ifma->ifma_addr, M_IFMADDR);
2403 		free(ifma, M_IFMADDR);
2404 		return (NULL);
2405 	}
2406 	bcopy(llsa, dupsa, llsa->sa_len);
2407 	ifma->ifma_lladdr = dupsa;
2408 
2409 	return (ifma);
2410 }
2411 
2412 /*
2413  * if_freemulti: free ifmultiaddr structure and possibly attached related
2414  * addresses.  The caller is responsible for implementing reference
2415  * counting, notifying the driver, handling routing messages, and releasing
2416  * any dependent link layer state.
2417  */
2418 static void
2419 if_freemulti(struct ifmultiaddr *ifma)
2420 {
2421 
2422 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2423 	    ifma->ifma_refcount));
2424 	KASSERT(ifma->ifma_protospec == NULL,
2425 	    ("if_freemulti: protospec not NULL"));
2426 
2427 	if (ifma->ifma_lladdr != NULL)
2428 		free(ifma->ifma_lladdr, M_IFMADDR);
2429 	free(ifma->ifma_addr, M_IFMADDR);
2430 	free(ifma, M_IFMADDR);
2431 }
2432 
2433 /*
2434  * Register an additional multicast address with a network interface.
2435  *
2436  * - If the address is already present, bump the reference count on the
2437  *   address and return.
2438  * - If the address is not link-layer, look up a link layer address.
2439  * - Allocate address structures for one or both addresses, and attach to the
2440  *   multicast address list on the interface.  If automatically adding a link
2441  *   layer address, the protocol address will own a reference to the link
2442  *   layer address, to be freed when it is freed.
2443  * - Notify the network device driver of an addition to the multicast address
2444  *   list.
2445  *
2446  * 'sa' points to caller-owned memory with the desired multicast address.
2447  *
2448  * 'retifma' will be used to return a pointer to the resulting multicast
2449  * address reference, if desired.
2450  */
2451 int
2452 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2453     struct ifmultiaddr **retifma)
2454 {
2455 	struct ifmultiaddr *ifma, *ll_ifma;
2456 	struct sockaddr *llsa;
2457 	int error;
2458 
2459 	/*
2460 	 * If the address is already present, return a new reference to it;
2461 	 * otherwise, allocate storage and set up a new address.
2462 	 */
2463 	IF_ADDR_LOCK(ifp);
2464 	ifma = if_findmulti(ifp, sa);
2465 	if (ifma != NULL) {
2466 		ifma->ifma_refcount++;
2467 		if (retifma != NULL)
2468 			*retifma = ifma;
2469 		IF_ADDR_UNLOCK(ifp);
2470 		return (0);
2471 	}
2472 
2473 	/*
2474 	 * The address isn't already present; resolve the protocol address
2475 	 * into a link layer address, and then look that up, bump its
2476 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2477 	 * returned, we will need to free it later.
2478 	 */
2479 	llsa = NULL;
2480 	ll_ifma = NULL;
2481 	if (ifp->if_resolvemulti != NULL) {
2482 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2483 		if (error)
2484 			goto unlock_out;
2485 	}
2486 
2487 	/*
2488 	 * Allocate the new address.  Don't hook it up yet, as we may also
2489 	 * need to allocate a link layer multicast address.
2490 	 */
2491 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2492 	if (ifma == NULL) {
2493 		error = ENOMEM;
2494 		goto free_llsa_out;
2495 	}
2496 
2497 	/*
2498 	 * If a link layer address is found, we'll need to see if it's
2499 	 * already present in the address list, or allocate is as well.
2500 	 * When this block finishes, the link layer address will be on the
2501 	 * list.
2502 	 */
2503 	if (llsa != NULL) {
2504 		ll_ifma = if_findmulti(ifp, llsa);
2505 		if (ll_ifma == NULL) {
2506 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2507 			if (ll_ifma == NULL) {
2508 				--ifma->ifma_refcount;
2509 				if_freemulti(ifma);
2510 				error = ENOMEM;
2511 				goto free_llsa_out;
2512 			}
2513 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
2514 			    ifma_link);
2515 		} else
2516 			ll_ifma->ifma_refcount++;
2517 		ifma->ifma_llifma = ll_ifma;
2518 	}
2519 
2520 	/*
2521 	 * We now have a new multicast address, ifma, and possibly a new or
2522 	 * referenced link layer address.  Add the primary address to the
2523 	 * ifnet address list.
2524 	 */
2525 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2526 
2527 	if (retifma != NULL)
2528 		*retifma = ifma;
2529 
2530 	/*
2531 	 * Must generate the message while holding the lock so that 'ifma'
2532 	 * pointer is still valid.
2533 	 */
2534 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2535 	IF_ADDR_UNLOCK(ifp);
2536 
2537 	/*
2538 	 * We are certain we have added something, so call down to the
2539 	 * interface to let them know about it.
2540 	 */
2541 	if (ifp->if_ioctl != NULL) {
2542 		IFF_LOCKGIANT(ifp);
2543 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2544 		IFF_UNLOCKGIANT(ifp);
2545 	}
2546 
2547 	if (llsa != NULL)
2548 		free(llsa, M_IFMADDR);
2549 
2550 	return (0);
2551 
2552 free_llsa_out:
2553 	if (llsa != NULL)
2554 		free(llsa, M_IFMADDR);
2555 
2556 unlock_out:
2557 	IF_ADDR_UNLOCK(ifp);
2558 	return (error);
2559 }
2560 
2561 /*
2562  * Delete a multicast group membership by network-layer group address.
2563  *
2564  * Returns ENOENT if the entry could not be found. If ifp no longer
2565  * exists, results are undefined. This entry point should only be used
2566  * from subsystems which do appropriate locking to hold ifp for the
2567  * duration of the call.
2568  * Network-layer protocol domains must use if_delmulti_ifma().
2569  */
2570 int
2571 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2572 {
2573 	struct ifmultiaddr *ifma;
2574 	int lastref;
2575 #ifdef INVARIANTS
2576 	struct ifnet *oifp;
2577 	INIT_VNET_NET(ifp->if_vnet);
2578 
2579 	IFNET_RLOCK();
2580 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
2581 		if (ifp == oifp)
2582 			break;
2583 	if (ifp != oifp)
2584 		ifp = NULL;
2585 	IFNET_RUNLOCK();
2586 
2587 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
2588 #endif
2589 	if (ifp == NULL)
2590 		return (ENOENT);
2591 
2592 	IF_ADDR_LOCK(ifp);
2593 	lastref = 0;
2594 	ifma = if_findmulti(ifp, sa);
2595 	if (ifma != NULL)
2596 		lastref = if_delmulti_locked(ifp, ifma, 0);
2597 	IF_ADDR_UNLOCK(ifp);
2598 
2599 	if (ifma == NULL)
2600 		return (ENOENT);
2601 
2602 	if (lastref && ifp->if_ioctl != NULL) {
2603 		IFF_LOCKGIANT(ifp);
2604 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2605 		IFF_UNLOCKGIANT(ifp);
2606 	}
2607 
2608 	return (0);
2609 }
2610 
2611 /*
2612  * Delete a multicast group membership by group membership pointer.
2613  * Network-layer protocol domains must use this routine.
2614  *
2615  * It is safe to call this routine if the ifp disappeared. Callers should
2616  * hold IFF_LOCKGIANT() to avoid a LOR in case the hardware needs to be
2617  * reconfigured.
2618  */
2619 void
2620 if_delmulti_ifma(struct ifmultiaddr *ifma)
2621 {
2622 #ifdef DIAGNOSTIC
2623 	INIT_VNET_NET(curvnet);
2624 #endif
2625 	struct ifnet *ifp;
2626 	int lastref;
2627 
2628 	ifp = ifma->ifma_ifp;
2629 #ifdef DIAGNOSTIC
2630 	if (ifp == NULL) {
2631 		printf("%s: ifma_ifp seems to be detached\n", __func__);
2632 	} else {
2633 		struct ifnet *oifp;
2634 
2635 		IFNET_RLOCK();
2636 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
2637 			if (ifp == oifp)
2638 				break;
2639 		if (ifp != oifp) {
2640 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
2641 			ifp = NULL;
2642 		}
2643 		IFNET_RUNLOCK();
2644 	}
2645 #endif
2646 	/*
2647 	 * If and only if the ifnet instance exists: Acquire the address lock.
2648 	 */
2649 	if (ifp != NULL)
2650 		IF_ADDR_LOCK(ifp);
2651 
2652 	lastref = if_delmulti_locked(ifp, ifma, 0);
2653 
2654 	if (ifp != NULL) {
2655 		/*
2656 		 * If and only if the ifnet instance exists:
2657 		 *  Release the address lock.
2658 		 *  If the group was left: update the hardware hash filter.
2659 		 */
2660 		IF_ADDR_UNLOCK(ifp);
2661 		if (lastref && ifp->if_ioctl != NULL) {
2662 			IFF_LOCKGIANT(ifp);
2663 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2664 			IFF_UNLOCKGIANT(ifp);
2665 		}
2666 	}
2667 }
2668 
2669 /*
2670  * Perform deletion of network-layer and/or link-layer multicast address.
2671  *
2672  * Return 0 if the reference count was decremented.
2673  * Return 1 if the final reference was released, indicating that the
2674  * hardware hash filter should be reprogrammed.
2675  */
2676 static int
2677 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
2678 {
2679 	struct ifmultiaddr *ll_ifma;
2680 
2681 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
2682 		KASSERT(ifma->ifma_ifp == ifp,
2683 		    ("%s: inconsistent ifp %p", __func__, ifp));
2684 		IF_ADDR_LOCK_ASSERT(ifp);
2685 	}
2686 
2687 	ifp = ifma->ifma_ifp;
2688 
2689 	/*
2690 	 * If the ifnet is detaching, null out references to ifnet,
2691 	 * so that upper protocol layers will notice, and not attempt
2692 	 * to obtain locks for an ifnet which no longer exists. The
2693 	 * routing socket announcement must happen before the ifnet
2694 	 * instance is detached from the system.
2695 	 */
2696 	if (detaching) {
2697 #ifdef DIAGNOSTIC
2698 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
2699 #endif
2700 		/*
2701 		 * ifp may already be nulled out if we are being reentered
2702 		 * to delete the ll_ifma.
2703 		 */
2704 		if (ifp != NULL) {
2705 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
2706 			ifma->ifma_ifp = NULL;
2707 		}
2708 	}
2709 
2710 	if (--ifma->ifma_refcount > 0)
2711 		return 0;
2712 
2713 	/*
2714 	 * If this ifma is a network-layer ifma, a link-layer ifma may
2715 	 * have been associated with it. Release it first if so.
2716 	 */
2717 	ll_ifma = ifma->ifma_llifma;
2718 	if (ll_ifma != NULL) {
2719 		KASSERT(ifma->ifma_lladdr != NULL,
2720 		    ("%s: llifma w/o lladdr", __func__));
2721 		if (detaching)
2722 			ll_ifma->ifma_ifp = NULL;	/* XXX */
2723 		if (--ll_ifma->ifma_refcount == 0) {
2724 			if (ifp != NULL) {
2725 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
2726 				    ifma_link);
2727 			}
2728 			if_freemulti(ll_ifma);
2729 		}
2730 	}
2731 
2732 	if (ifp != NULL)
2733 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2734 
2735 	if_freemulti(ifma);
2736 
2737 	/*
2738 	 * The last reference to this instance of struct ifmultiaddr
2739 	 * was released; the hardware should be notified of this change.
2740 	 */
2741 	return 1;
2742 }
2743 
2744 /*
2745  * Set the link layer address on an interface.
2746  *
2747  * At this time we only support certain types of interfaces,
2748  * and we don't allow the length of the address to change.
2749  */
2750 int
2751 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2752 {
2753 	struct sockaddr_dl *sdl;
2754 	struct ifaddr *ifa;
2755 	struct ifreq ifr;
2756 
2757 	ifa = ifp->if_addr;
2758 	if (ifa == NULL)
2759 		return (EINVAL);
2760 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2761 	if (sdl == NULL)
2762 		return (EINVAL);
2763 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2764 		return (EINVAL);
2765 	switch (ifp->if_type) {
2766 	case IFT_ETHER:
2767 	case IFT_FDDI:
2768 	case IFT_XETHER:
2769 	case IFT_ISO88025:
2770 	case IFT_L2VLAN:
2771 	case IFT_BRIDGE:
2772 	case IFT_ARCNET:
2773 	case IFT_IEEE8023ADLAG:
2774 		bcopy(lladdr, LLADDR(sdl), len);
2775 		break;
2776 	default:
2777 		return (ENODEV);
2778 	}
2779 	/*
2780 	 * If the interface is already up, we need
2781 	 * to re-init it in order to reprogram its
2782 	 * address filter.
2783 	 */
2784 	if ((ifp->if_flags & IFF_UP) != 0) {
2785 		if (ifp->if_ioctl) {
2786 			IFF_LOCKGIANT(ifp);
2787 			ifp->if_flags &= ~IFF_UP;
2788 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2789 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2790 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2791 			ifp->if_flags |= IFF_UP;
2792 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2793 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2794 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2795 			IFF_UNLOCKGIANT(ifp);
2796 		}
2797 #ifdef INET
2798 		/*
2799 		 * Also send gratuitous ARPs to notify other nodes about
2800 		 * the address change.
2801 		 */
2802 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2803 			if (ifa->ifa_addr->sa_family == AF_INET)
2804 				arp_ifinit(ifp, ifa);
2805 		}
2806 #endif
2807 	}
2808 	return (0);
2809 }
2810 
2811 /*
2812  * The name argument must be a pointer to storage which will last as
2813  * long as the interface does.  For physical devices, the result of
2814  * device_get_name(dev) is a good choice and for pseudo-devices a
2815  * static string works well.
2816  */
2817 void
2818 if_initname(struct ifnet *ifp, const char *name, int unit)
2819 {
2820 	ifp->if_dname = name;
2821 	ifp->if_dunit = unit;
2822 	if (unit != IF_DUNIT_NONE)
2823 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2824 	else
2825 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2826 }
2827 
2828 int
2829 if_printf(struct ifnet *ifp, const char * fmt, ...)
2830 {
2831 	va_list ap;
2832 	int retval;
2833 
2834 	retval = printf("%s: ", ifp->if_xname);
2835 	va_start(ap, fmt);
2836 	retval += vprintf(fmt, ap);
2837 	va_end(ap);
2838 	return (retval);
2839 }
2840 
2841 /*
2842  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2843  * be called without Giant.  However, we often can't acquire the Giant lock
2844  * at those points; instead, we run it via a task queue that holds Giant via
2845  * if_start_deferred.
2846  *
2847  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2848  * outstanding if_start_deferred() tasks that will run after the free.  This
2849  * probably means waiting in if_detach().
2850  */
2851 void
2852 if_start(struct ifnet *ifp)
2853 {
2854 
2855 	if (ifp->if_flags & IFF_NEEDSGIANT) {
2856 		if (mtx_owned(&Giant))
2857 			(*(ifp)->if_start)(ifp);
2858 		else
2859 			taskqueue_enqueue(taskqueue_swi_giant,
2860 			    &ifp->if_starttask);
2861 	} else
2862 		(*(ifp)->if_start)(ifp);
2863 }
2864 
2865 static void
2866 if_start_deferred(void *context, int pending)
2867 {
2868 	struct ifnet *ifp;
2869 
2870 	GIANT_REQUIRED;
2871 
2872 	ifp = context;
2873 	(ifp->if_start)(ifp);
2874 }
2875 
2876 /*
2877  * Backwards compatibility interface for drivers
2878  * that have not implemented it
2879  */
2880 static int
2881 if_transmit(struct ifnet *ifp, struct mbuf *m)
2882 {
2883 	int error;
2884 
2885 	IFQ_HANDOFF(ifp, m, error);
2886 	return (error);
2887 }
2888 
2889 int
2890 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2891 {
2892 	int active = 0;
2893 
2894 	IF_LOCK(ifq);
2895 	if (_IF_QFULL(ifq)) {
2896 		_IF_DROP(ifq);
2897 		IF_UNLOCK(ifq);
2898 		m_freem(m);
2899 		return (0);
2900 	}
2901 	if (ifp != NULL) {
2902 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2903 		if (m->m_flags & (M_BCAST|M_MCAST))
2904 			ifp->if_omcasts++;
2905 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2906 	}
2907 	_IF_ENQUEUE(ifq, m);
2908 	IF_UNLOCK(ifq);
2909 	if (ifp != NULL && !active)
2910 		if_start(ifp);
2911 	return (1);
2912 }
2913 
2914 void
2915 if_register_com_alloc(u_char type,
2916     if_com_alloc_t *a, if_com_free_t *f)
2917 {
2918 
2919 	KASSERT(if_com_alloc[type] == NULL,
2920 	    ("if_register_com_alloc: %d already registered", type));
2921 	KASSERT(if_com_free[type] == NULL,
2922 	    ("if_register_com_alloc: %d free already registered", type));
2923 
2924 	if_com_alloc[type] = a;
2925 	if_com_free[type] = f;
2926 }
2927 
2928 void
2929 if_deregister_com_alloc(u_char type)
2930 {
2931 
2932 	KASSERT(if_com_alloc[type] != NULL,
2933 	    ("if_deregister_com_alloc: %d not registered", type));
2934 	KASSERT(if_com_free[type] != NULL,
2935 	    ("if_deregister_com_alloc: %d free not registered", type));
2936 	if_com_alloc[type] = NULL;
2937 	if_com_free[type] = NULL;
2938 }
2939