xref: /freebsd/sys/net/if.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/malloc.h>
41 #include <sys/sbuf.h>
42 #include <sys/bus.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/refcount.h>
53 #include <sys/module.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 <machine/stdarg.h>
62 #include <vm/uma.h>
63 
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_clone.h>
67 #include <net/if_dl.h>
68 #include <net/if_types.h>
69 #include <net/if_var.h>
70 #include <net/radix.h>
71 #include <net/route.h>
72 #include <net/vnet.h>
73 
74 #if defined(INET) || defined(INET6)
75 /*XXX*/
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
78 #include <netinet/ip_carp.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 #endif
87 
88 #include <security/mac/mac_framework.h>
89 
90 #ifdef COMPAT_FREEBSD32
91 #include <sys/mount.h>
92 #include <compat/freebsd32/freebsd32.h>
93 #endif
94 
95 struct ifindex_entry {
96 	struct  ifnet *ife_ifnet;
97 };
98 
99 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
100 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
101 
102 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
103 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
104     &ifqmaxlen, 0, "max send queue size");
105 
106 /* Log link state change events */
107 static int log_link_state_change = 1;
108 
109 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
110 	&log_link_state_change, 0,
111 	"log interface link state change events");
112 
113 /* Interface description */
114 static unsigned int ifdescr_maxlen = 1024;
115 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
116 	&ifdescr_maxlen, 0,
117 	"administrative maximum length for interface description");
118 
119 MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
120 
121 /* global sx for non-critical path ifdescr */
122 static struct sx ifdescr_sx;
123 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
124 
125 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
126 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
127 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
128 /* These are external hooks for CARP. */
129 void	(*carp_linkstate_p)(struct ifnet *ifp);
130 #if defined(INET) || defined(INET6)
131 struct ifnet *(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
132 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
133     struct sockaddr *sa, struct rtentry *rt);
134 #endif
135 #ifdef INET
136 int (*carp_iamatch_p)(struct ifnet *, struct in_ifaddr *, struct in_addr *,
137     u_int8_t **);
138 #endif
139 #ifdef INET6
140 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
141 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
142     const struct in6_addr *taddr);
143 #endif
144 
145 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
146 
147 /*
148  * XXX: Style; these should be sorted alphabetically, and unprototyped
149  * static functions should be prototyped. Currently they are sorted by
150  * declaration order.
151  */
152 static void	if_attachdomain(void *);
153 static void	if_attachdomain1(struct ifnet *);
154 static int	ifconf(u_long, caddr_t);
155 static void	if_freemulti(struct ifmultiaddr *);
156 static void	if_init(void *);
157 static void	if_grow(void);
158 static void	if_route(struct ifnet *, int flag, int fam);
159 static int	if_setflag(struct ifnet *, int, int, int *, int);
160 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
161 static void	if_unroute(struct ifnet *, int flag, int fam);
162 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
163 static int	if_rtdel(struct radix_node *, void *);
164 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
165 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
166 static void	do_link_state_change(void *, int);
167 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
168 static int	if_getgroupmembers(struct ifgroupreq *);
169 static void	if_delgroups(struct ifnet *);
170 static void	if_attach_internal(struct ifnet *, int);
171 static void	if_detach_internal(struct ifnet *, int);
172 
173 #ifdef INET6
174 /*
175  * XXX: declare here to avoid to include many inet6 related files..
176  * should be more generalized?
177  */
178 extern void	nd6_setmtu(struct ifnet *);
179 #endif
180 
181 VNET_DEFINE(int, if_index);
182 int	ifqmaxlen = IFQ_MAXLEN;
183 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
184 VNET_DEFINE(struct ifgrouphead, ifg_head);
185 
186 static VNET_DEFINE(int, if_indexlim) = 8;
187 
188 /* Table of ifnet by index. */
189 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
190 
191 #define	V_if_indexlim		VNET(if_indexlim)
192 #define	V_ifindex_table		VNET(ifindex_table)
193 
194 /*
195  * The global network interface list (V_ifnet) and related state (such as
196  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
197  * an rwlock.  Either may be acquired shared to stablize the list, but both
198  * must be acquired writable to modify the list.  This model allows us to
199  * both stablize the interface list during interrupt thread processing, but
200  * also to stablize it over long-running ioctls, without introducing priority
201  * inversions and deadlocks.
202  */
203 struct rwlock ifnet_rwlock;
204 struct sx ifnet_sxlock;
205 
206 /*
207  * The allocation of network interfaces is a rather non-atomic affair; we
208  * need to select an index before we are ready to expose the interface for
209  * use, so will use this pointer value to indicate reservation.
210  */
211 #define	IFNET_HOLD	(void *)(uintptr_t)(-1)
212 
213 static	if_com_alloc_t *if_com_alloc[256];
214 static	if_com_free_t *if_com_free[256];
215 
216 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
217 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
218 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
219 
220 struct ifnet *
221 ifnet_byindex_locked(u_short idx)
222 {
223 
224 	if (idx > V_if_index)
225 		return (NULL);
226 	if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
227 		return (NULL);
228 	return (V_ifindex_table[idx].ife_ifnet);
229 }
230 
231 struct ifnet *
232 ifnet_byindex(u_short idx)
233 {
234 	struct ifnet *ifp;
235 
236 	IFNET_RLOCK_NOSLEEP();
237 	ifp = ifnet_byindex_locked(idx);
238 	IFNET_RUNLOCK_NOSLEEP();
239 	return (ifp);
240 }
241 
242 struct ifnet *
243 ifnet_byindex_ref(u_short idx)
244 {
245 	struct ifnet *ifp;
246 
247 	IFNET_RLOCK_NOSLEEP();
248 	ifp = ifnet_byindex_locked(idx);
249 	if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
250 		IFNET_RUNLOCK_NOSLEEP();
251 		return (NULL);
252 	}
253 	if_ref(ifp);
254 	IFNET_RUNLOCK_NOSLEEP();
255 	return (ifp);
256 }
257 
258 /*
259  * Allocate an ifindex array entry; return 0 on success or an error on
260  * failure.
261  */
262 static int
263 ifindex_alloc_locked(u_short *idxp)
264 {
265 	u_short idx;
266 
267 	IFNET_WLOCK_ASSERT();
268 
269 retry:
270 	/*
271 	 * Try to find an empty slot below V_if_index.  If we fail, take the
272 	 * next slot.
273 	 */
274 	for (idx = 1; idx <= V_if_index; idx++) {
275 		if (V_ifindex_table[idx].ife_ifnet == NULL)
276 			break;
277 	}
278 
279 	/* Catch if_index overflow. */
280 	if (idx < 1)
281 		return (ENOSPC);
282 	if (idx >= V_if_indexlim) {
283 		if_grow();
284 		goto retry;
285 	}
286 	if (idx > V_if_index)
287 		V_if_index = idx;
288 	*idxp = idx;
289 	return (0);
290 }
291 
292 static void
293 ifindex_free_locked(u_short idx)
294 {
295 
296 	IFNET_WLOCK_ASSERT();
297 
298 	V_ifindex_table[idx].ife_ifnet = NULL;
299 	while (V_if_index > 0 &&
300 	    V_ifindex_table[V_if_index].ife_ifnet == NULL)
301 		V_if_index--;
302 }
303 
304 static void
305 ifindex_free(u_short idx)
306 {
307 
308 	IFNET_WLOCK();
309 	ifindex_free_locked(idx);
310 	IFNET_WUNLOCK();
311 }
312 
313 static void
314 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
315 {
316 
317 	IFNET_WLOCK_ASSERT();
318 
319 	V_ifindex_table[idx].ife_ifnet = ifp;
320 }
321 
322 static void
323 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
324 {
325 
326 	IFNET_WLOCK();
327 	ifnet_setbyindex_locked(idx, ifp);
328 	IFNET_WUNLOCK();
329 }
330 
331 struct ifaddr *
332 ifaddr_byindex(u_short idx)
333 {
334 	struct ifaddr *ifa;
335 
336 	IFNET_RLOCK_NOSLEEP();
337 	ifa = ifnet_byindex_locked(idx)->if_addr;
338 	if (ifa != NULL)
339 		ifa_ref(ifa);
340 	IFNET_RUNLOCK_NOSLEEP();
341 	return (ifa);
342 }
343 
344 /*
345  * Network interface utility routines.
346  *
347  * Routines with ifa_ifwith* names take sockaddr *'s as
348  * parameters.
349  */
350 
351 static void
352 vnet_if_init(const void *unused __unused)
353 {
354 
355 	TAILQ_INIT(&V_ifnet);
356 	TAILQ_INIT(&V_ifg_head);
357 	IFNET_WLOCK();
358 	if_grow();				/* create initial table */
359 	IFNET_WUNLOCK();
360 	vnet_if_clone_init();
361 }
362 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
363     NULL);
364 
365 /* ARGSUSED*/
366 static void
367 if_init(void *dummy __unused)
368 {
369 
370 	IFNET_LOCK_INIT();
371 	if_clone_init();
372 }
373 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
374 
375 
376 #ifdef VIMAGE
377 static void
378 vnet_if_uninit(const void *unused __unused)
379 {
380 
381 	VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
382 	    "not empty", __func__, __LINE__, &V_ifnet));
383 	VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
384 	    "not empty", __func__, __LINE__, &V_ifg_head));
385 
386 	free((caddr_t)V_ifindex_table, M_IFNET);
387 }
388 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
389     vnet_if_uninit, NULL);
390 #endif
391 
392 static void
393 if_grow(void)
394 {
395 	int oldlim;
396 	u_int n;
397 	struct ifindex_entry *e;
398 
399 	IFNET_WLOCK_ASSERT();
400 	oldlim = V_if_indexlim;
401 	IFNET_WUNLOCK();
402 	n = (oldlim << 1) * sizeof(*e);
403 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
404 	IFNET_WLOCK();
405 	if (V_if_indexlim != oldlim) {
406 		free(e, M_IFNET);
407 		return;
408 	}
409 	if (V_ifindex_table != NULL) {
410 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
411 		free((caddr_t)V_ifindex_table, M_IFNET);
412 	}
413 	V_if_indexlim <<= 1;
414 	V_ifindex_table = e;
415 }
416 
417 /*
418  * Allocate a struct ifnet and an index for an interface.  A layer 2
419  * common structure will also be allocated if an allocation routine is
420  * registered for the passed type.
421  */
422 struct ifnet *
423 if_alloc(u_char type)
424 {
425 	struct ifnet *ifp;
426 	u_short idx;
427 
428 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
429 	IFNET_WLOCK();
430 	if (ifindex_alloc_locked(&idx) != 0) {
431 		IFNET_WUNLOCK();
432 		free(ifp, M_IFNET);
433 		return (NULL);
434 	}
435 	ifnet_setbyindex_locked(idx, IFNET_HOLD);
436 	IFNET_WUNLOCK();
437 	ifp->if_index = idx;
438 	ifp->if_type = type;
439 	ifp->if_alloctype = type;
440 	if (if_com_alloc[type] != NULL) {
441 		ifp->if_l2com = if_com_alloc[type](type, ifp);
442 		if (ifp->if_l2com == NULL) {
443 			free(ifp, M_IFNET);
444 			ifindex_free(idx);
445 			return (NULL);
446 		}
447 	}
448 
449 	IF_ADDR_LOCK_INIT(ifp);
450 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
451 	ifp->if_afdata_initialized = 0;
452 	IF_AFDATA_LOCK_INIT(ifp);
453 	TAILQ_INIT(&ifp->if_addrhead);
454 	TAILQ_INIT(&ifp->if_prefixhead);
455 	TAILQ_INIT(&ifp->if_multiaddrs);
456 	TAILQ_INIT(&ifp->if_groups);
457 #ifdef MAC
458 	mac_ifnet_init(ifp);
459 #endif
460 	ifq_init(&ifp->if_snd, ifp);
461 
462 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
463 	ifnet_setbyindex(ifp->if_index, ifp);
464 	return (ifp);
465 }
466 
467 /*
468  * Do the actual work of freeing a struct ifnet, and layer 2 common
469  * structure.  This call is made when the last reference to an
470  * interface is released.
471  */
472 static void
473 if_free_internal(struct ifnet *ifp)
474 {
475 
476 	KASSERT((ifp->if_flags & IFF_DYING),
477 	    ("if_free_internal: interface not dying"));
478 
479 	if (if_com_free[ifp->if_alloctype] != NULL)
480 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
481 		    ifp->if_alloctype);
482 
483 #ifdef MAC
484 	mac_ifnet_destroy(ifp);
485 #endif /* MAC */
486 	if (ifp->if_description != NULL)
487 		free(ifp->if_description, M_IFDESCR);
488 	IF_AFDATA_DESTROY(ifp);
489 	IF_ADDR_LOCK_DESTROY(ifp);
490 	ifq_delete(&ifp->if_snd);
491 	free(ifp, M_IFNET);
492 }
493 
494 /*
495  * This version should only be called by intefaces that switch their type
496  * after calling if_alloc().  if_free_type() will go away again now that we
497  * have if_alloctype to cache the original allocation type.  For now, assert
498  * that they match, since we require that in practice.
499  */
500 void
501 if_free_type(struct ifnet *ifp, u_char type)
502 {
503 
504 	KASSERT(ifp->if_alloctype == type,
505 	    ("if_free_type: type (%d) != alloctype (%d)", type,
506 	    ifp->if_alloctype));
507 
508 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
509 
510 	IFNET_WLOCK();
511 	KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
512 	    ("%s: freeing unallocated ifnet", ifp->if_xname));
513 
514 	ifindex_free_locked(ifp->if_index);
515 	IFNET_WUNLOCK();
516 
517 	if (!refcount_release(&ifp->if_refcount))
518 		return;
519 	if_free_internal(ifp);
520 }
521 
522 /*
523  * This is the normal version of if_free(), used by device drivers to free a
524  * detached network interface.  The contents of if_free_type() will move into
525  * here when if_free_type() goes away.
526  */
527 void
528 if_free(struct ifnet *ifp)
529 {
530 
531 	if_free_type(ifp, ifp->if_alloctype);
532 }
533 
534 /*
535  * Interfaces to keep an ifnet type-stable despite the possibility of the
536  * driver calling if_free().  If there are additional references, we defer
537  * freeing the underlying data structure.
538  */
539 void
540 if_ref(struct ifnet *ifp)
541 {
542 
543 	/* We don't assert the ifnet list lock here, but arguably should. */
544 	refcount_acquire(&ifp->if_refcount);
545 }
546 
547 void
548 if_rele(struct ifnet *ifp)
549 {
550 
551 	if (!refcount_release(&ifp->if_refcount))
552 		return;
553 	if_free_internal(ifp);
554 }
555 
556 void
557 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
558 {
559 
560 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
561 
562 	if (ifq->ifq_maxlen == 0)
563 		ifq->ifq_maxlen = ifqmaxlen;
564 
565 	ifq->altq_type = 0;
566 	ifq->altq_disc = NULL;
567 	ifq->altq_flags &= ALTQF_CANTCHANGE;
568 	ifq->altq_tbr  = NULL;
569 	ifq->altq_ifp  = ifp;
570 }
571 
572 void
573 ifq_delete(struct ifaltq *ifq)
574 {
575 	mtx_destroy(&ifq->ifq_mtx);
576 }
577 
578 /*
579  * Perform generic interface initalization tasks and attach the interface
580  * to the list of "active" interfaces.  If vmove flag is set on entry
581  * to if_attach_internal(), perform only a limited subset of initialization
582  * tasks, given that we are moving from one vnet to another an ifnet which
583  * has already been fully initialized.
584  *
585  * XXX:
586  *  - The decision to return void and thus require this function to
587  *    succeed is questionable.
588  *  - We should probably do more sanity checking.  For instance we don't
589  *    do anything to insure if_xname is unique or non-empty.
590  */
591 void
592 if_attach(struct ifnet *ifp)
593 {
594 
595 	if_attach_internal(ifp, 0);
596 }
597 
598 static void
599 if_attach_internal(struct ifnet *ifp, int vmove)
600 {
601 	unsigned socksize, ifasize;
602 	int namelen, masklen;
603 	struct sockaddr_dl *sdl;
604 	struct ifaddr *ifa;
605 
606 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
607 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
608 		    ifp->if_xname);
609 
610 #ifdef VIMAGE
611 	ifp->if_vnet = curvnet;
612 	if (ifp->if_home_vnet == NULL)
613 		ifp->if_home_vnet = curvnet;
614 #endif
615 
616 	if_addgroup(ifp, IFG_ALL);
617 
618 	getmicrotime(&ifp->if_lastchange);
619 	ifp->if_data.ifi_epoch = time_uptime;
620 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
621 
622 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
623 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
624 	    ("transmit and qflush must both either be set or both be NULL"));
625 	if (ifp->if_transmit == NULL) {
626 		ifp->if_transmit = if_transmit;
627 		ifp->if_qflush = if_qflush;
628 	}
629 
630 	if (!vmove) {
631 #ifdef MAC
632 		mac_ifnet_create(ifp);
633 #endif
634 
635 		/*
636 		 * Create a Link Level name for this device.
637 		 */
638 		namelen = strlen(ifp->if_xname);
639 		/*
640 		 * Always save enough space for any possiable name so we
641 		 * can do a rename in place later.
642 		 */
643 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
644 		socksize = masklen + ifp->if_addrlen;
645 		if (socksize < sizeof(*sdl))
646 			socksize = sizeof(*sdl);
647 		socksize = roundup2(socksize, sizeof(long));
648 		ifasize = sizeof(*ifa) + 2 * socksize;
649 		ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
650 		ifa_init(ifa);
651 		sdl = (struct sockaddr_dl *)(ifa + 1);
652 		sdl->sdl_len = socksize;
653 		sdl->sdl_family = AF_LINK;
654 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
655 		sdl->sdl_nlen = namelen;
656 		sdl->sdl_index = ifp->if_index;
657 		sdl->sdl_type = ifp->if_type;
658 		ifp->if_addr = ifa;
659 		ifa->ifa_ifp = ifp;
660 		ifa->ifa_rtrequest = link_rtrequest;
661 		ifa->ifa_addr = (struct sockaddr *)sdl;
662 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
663 		ifa->ifa_netmask = (struct sockaddr *)sdl;
664 		sdl->sdl_len = masklen;
665 		while (namelen != 0)
666 			sdl->sdl_data[--namelen] = 0xff;
667 		TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
668 		/* Reliably crash if used uninitialized. */
669 		ifp->if_broadcastaddr = NULL;
670 	}
671 #ifdef VIMAGE
672 	else {
673 		/*
674 		 * Update the interface index in the link layer address
675 		 * of the interface.
676 		 */
677 		for (ifa = ifp->if_addr; ifa != NULL;
678 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
679 			if (ifa->ifa_addr->sa_family == AF_LINK) {
680 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
681 				sdl->sdl_index = ifp->if_index;
682 			}
683 		}
684 	}
685 #endif
686 
687 	IFNET_WLOCK();
688 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
689 #ifdef VIMAGE
690 	curvnet->vnet_ifcnt++;
691 #endif
692 	IFNET_WUNLOCK();
693 
694 	if (domain_init_status >= 2)
695 		if_attachdomain1(ifp);
696 
697 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
698 	if (IS_DEFAULT_VNET(curvnet))
699 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
700 
701 	/* Announce the interface. */
702 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
703 }
704 
705 static void
706 if_attachdomain(void *dummy)
707 {
708 	struct ifnet *ifp;
709 	int s;
710 
711 	s = splnet();
712 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
713 		if_attachdomain1(ifp);
714 	splx(s);
715 }
716 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
717     if_attachdomain, NULL);
718 
719 static void
720 if_attachdomain1(struct ifnet *ifp)
721 {
722 	struct domain *dp;
723 	int s;
724 
725 	s = splnet();
726 
727 	/*
728 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
729 	 * cannot lock ifp->if_afdata initialization, entirely.
730 	 */
731 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
732 		splx(s);
733 		return;
734 	}
735 	if (ifp->if_afdata_initialized >= domain_init_status) {
736 		IF_AFDATA_UNLOCK(ifp);
737 		splx(s);
738 		printf("if_attachdomain called more than once on %s\n",
739 		    ifp->if_xname);
740 		return;
741 	}
742 	ifp->if_afdata_initialized = domain_init_status;
743 	IF_AFDATA_UNLOCK(ifp);
744 
745 	/* address family dependent data region */
746 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
747 	for (dp = domains; dp; dp = dp->dom_next) {
748 		if (dp->dom_ifattach)
749 			ifp->if_afdata[dp->dom_family] =
750 			    (*dp->dom_ifattach)(ifp);
751 	}
752 
753 	splx(s);
754 }
755 
756 /*
757  * Remove any unicast or broadcast network addresses from an interface.
758  */
759 void
760 if_purgeaddrs(struct ifnet *ifp)
761 {
762 	struct ifaddr *ifa, *next;
763 
764 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
765 		if (ifa->ifa_addr->sa_family == AF_LINK)
766 			continue;
767 #ifdef INET
768 		/* XXX: Ugly!! ad hoc just for INET */
769 		if (ifa->ifa_addr->sa_family == AF_INET) {
770 			struct ifaliasreq ifr;
771 
772 			bzero(&ifr, sizeof(ifr));
773 			ifr.ifra_addr = *ifa->ifa_addr;
774 			if (ifa->ifa_dstaddr)
775 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
776 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
777 			    NULL) == 0)
778 				continue;
779 		}
780 #endif /* INET */
781 #ifdef INET6
782 		if (ifa->ifa_addr->sa_family == AF_INET6) {
783 			in6_purgeaddr(ifa);
784 			/* ifp_addrhead is already updated */
785 			continue;
786 		}
787 #endif /* INET6 */
788 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
789 		ifa_free(ifa);
790 	}
791 }
792 
793 /*
794  * Remove any multicast network addresses from an interface when an ifnet
795  * is going away.
796  */
797 static void
798 if_purgemaddrs(struct ifnet *ifp)
799 {
800 	struct ifmultiaddr *ifma;
801 	struct ifmultiaddr *next;
802 
803 	IF_ADDR_LOCK(ifp);
804 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
805 		if_delmulti_locked(ifp, ifma, 1);
806 	IF_ADDR_UNLOCK(ifp);
807 }
808 
809 /*
810  * Detach an interface, removing it from the list of "active" interfaces.
811  * If vmove flag is set on entry to if_detach_internal(), perform only a
812  * limited subset of cleanup tasks, given that we are moving an ifnet from
813  * one vnet to another, where it must be fully operational.
814  *
815  * XXXRW: There are some significant questions about event ordering, and
816  * how to prevent things from starting to use the interface during detach.
817  */
818 void
819 if_detach(struct ifnet *ifp)
820 {
821 
822 	if_detach_internal(ifp, 0);
823 }
824 
825 static void
826 if_detach_internal(struct ifnet *ifp, int vmove)
827 {
828 	struct ifaddr *ifa;
829 	struct radix_node_head	*rnh;
830 	int i, j;
831 	struct domain *dp;
832  	struct ifnet *iter;
833  	int found = 0;
834 
835 	IFNET_WLOCK();
836 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
837 		if (iter == ifp) {
838 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
839 			found = 1;
840 			break;
841 		}
842 #ifdef VIMAGE
843 	if (found)
844 		curvnet->vnet_ifcnt--;
845 #endif
846 	IFNET_WUNLOCK();
847 	if (!found) {
848 		if (vmove)
849 			panic("%s: ifp=%p not on the ifnet tailq %p",
850 			    __func__, ifp, &V_ifnet);
851 		else
852 			return; /* XXX this should panic as well? */
853 	}
854 
855 	/*
856 	 * Remove/wait for pending events.
857 	 */
858 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
859 
860 	/*
861 	 * Remove routes and flush queues.
862 	 */
863 	if_down(ifp);
864 #ifdef ALTQ
865 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
866 		altq_disable(&ifp->if_snd);
867 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
868 		altq_detach(&ifp->if_snd);
869 #endif
870 
871 	if_purgeaddrs(ifp);
872 
873 #ifdef INET
874 	in_ifdetach(ifp);
875 #endif
876 
877 #ifdef INET6
878 	/*
879 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
880 	 * before removing routing entries below, since IPv6 interface direct
881 	 * routes are expected to be removed by the IPv6-specific kernel API.
882 	 * Otherwise, the kernel will detect some inconsistency and bark it.
883 	 */
884 	in6_ifdetach(ifp);
885 #endif
886 	if_purgemaddrs(ifp);
887 
888 	if (!vmove) {
889 		/*
890 		 * Prevent further calls into the device driver via ifnet.
891 		 */
892 		if_dead(ifp);
893 
894 		/*
895 		 * Remove link ifaddr pointer and maybe decrement if_index.
896 		 * Clean up all addresses.
897 		 */
898 		ifp->if_addr = NULL;
899 
900 		/* We can now free link ifaddr. */
901 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
902 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
903 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
904 			ifa_free(ifa);
905 		}
906 	}
907 
908 	/*
909 	 * Delete all remaining routes using this interface
910 	 * Unfortuneatly the only way to do this is to slog through
911 	 * the entire routing table looking for routes which point
912 	 * to this interface...oh well...
913 	 */
914 	for (i = 1; i <= AF_MAX; i++) {
915 		for (j = 0; j < rt_numfibs; j++) {
916 			rnh = rt_tables_get_rnh(j, i);
917 			if (rnh == NULL)
918 				continue;
919 			RADIX_NODE_HEAD_LOCK(rnh);
920 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
921 			RADIX_NODE_HEAD_UNLOCK(rnh);
922 		}
923 	}
924 
925 	/* Announce that the interface is gone. */
926 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
927 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
928 	if (IS_DEFAULT_VNET(curvnet))
929 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
930 	if_delgroups(ifp);
931 
932 	/*
933 	 * We cannot hold the lock over dom_ifdetach calls as they might
934 	 * sleep, for example trying to drain a callout, thus open up the
935 	 * theoretical race with re-attaching.
936 	 */
937 	IF_AFDATA_LOCK(ifp);
938 	i = ifp->if_afdata_initialized;
939 	ifp->if_afdata_initialized = 0;
940 	IF_AFDATA_UNLOCK(ifp);
941 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
942 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
943 			(*dp->dom_ifdetach)(ifp,
944 			    ifp->if_afdata[dp->dom_family]);
945 	}
946 }
947 
948 #ifdef VIMAGE
949 /*
950  * if_vmove() performs a limited version of if_detach() in current
951  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
952  * An attempt is made to shrink if_index in current vnet, find an
953  * unused if_index in target vnet and calls if_grow() if necessary,
954  * and finally find an unused if_xname for the target vnet.
955  */
956 void
957 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
958 {
959 	u_short idx;
960 
961 	/*
962 	 * Detach from current vnet, but preserve LLADDR info, do not
963 	 * mark as dead etc. so that the ifnet can be reattached later.
964 	 */
965 	if_detach_internal(ifp, 1);
966 
967 	/*
968 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
969 	 * the if_index for that vnet if possible.
970 	 *
971 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
972 	 * or we'd lock on one vnet and unlock on another.
973 	 */
974 	IFNET_WLOCK();
975 	ifindex_free_locked(ifp->if_index);
976 	IFNET_WUNLOCK();
977 
978 	/*
979 	 * Perform interface-specific reassignment tasks, if provided by
980 	 * the driver.
981 	 */
982 	if (ifp->if_reassign != NULL)
983 		ifp->if_reassign(ifp, new_vnet, NULL);
984 
985 	/*
986 	 * Switch to the context of the target vnet.
987 	 */
988 	CURVNET_SET_QUIET(new_vnet);
989 
990 	IFNET_WLOCK();
991 	if (ifindex_alloc_locked(&idx) != 0) {
992 		IFNET_WUNLOCK();
993 		panic("if_index overflow");
994 	}
995 	ifp->if_index = idx;
996 	ifnet_setbyindex_locked(ifp->if_index, ifp);
997 	IFNET_WUNLOCK();
998 
999 	if_attach_internal(ifp, 1);
1000 
1001 	CURVNET_RESTORE();
1002 }
1003 
1004 /*
1005  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1006  */
1007 static int
1008 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1009 {
1010 	struct prison *pr;
1011 	struct ifnet *difp;
1012 
1013 	/* Try to find the prison within our visibility. */
1014 	sx_slock(&allprison_lock);
1015 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1016 	sx_sunlock(&allprison_lock);
1017 	if (pr == NULL)
1018 		return (ENXIO);
1019 	prison_hold_locked(pr);
1020 	mtx_unlock(&pr->pr_mtx);
1021 
1022 	/* Do not try to move the iface from and to the same prison. */
1023 	if (pr->pr_vnet == ifp->if_vnet) {
1024 		prison_free(pr);
1025 		return (EEXIST);
1026 	}
1027 
1028 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1029 	/* XXX Lock interfaces to avoid races. */
1030 	CURVNET_SET_QUIET(pr->pr_vnet);
1031 	difp = ifunit(ifname);
1032 	CURVNET_RESTORE();
1033 	if (difp != NULL) {
1034 		prison_free(pr);
1035 		return (EEXIST);
1036 	}
1037 
1038 	/* Move the interface into the child jail/vnet. */
1039 	if_vmove(ifp, pr->pr_vnet);
1040 
1041 	/* Report the new if_xname back to the userland. */
1042 	sprintf(ifname, "%s", ifp->if_xname);
1043 
1044 	prison_free(pr);
1045 	return (0);
1046 }
1047 
1048 static int
1049 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1050 {
1051 	struct prison *pr;
1052 	struct vnet *vnet_dst;
1053 	struct ifnet *ifp;
1054 
1055 	/* Try to find the prison within our visibility. */
1056 	sx_slock(&allprison_lock);
1057 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1058 	sx_sunlock(&allprison_lock);
1059 	if (pr == NULL)
1060 		return (ENXIO);
1061 	prison_hold_locked(pr);
1062 	mtx_unlock(&pr->pr_mtx);
1063 
1064 	/* Make sure the named iface exists in the source prison/vnet. */
1065 	CURVNET_SET(pr->pr_vnet);
1066 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1067 	if (ifp == NULL) {
1068 		CURVNET_RESTORE();
1069 		prison_free(pr);
1070 		return (ENXIO);
1071 	}
1072 
1073 	/* Do not try to move the iface from and to the same prison. */
1074 	vnet_dst = TD_TO_VNET(td);
1075 	if (vnet_dst == ifp->if_vnet) {
1076 		CURVNET_RESTORE();
1077 		prison_free(pr);
1078 		return (EEXIST);
1079 	}
1080 
1081 	/* Get interface back from child jail/vnet. */
1082 	if_vmove(ifp, vnet_dst);
1083 	CURVNET_RESTORE();
1084 
1085 	/* Report the new if_xname back to the userland. */
1086 	sprintf(ifname, "%s", ifp->if_xname);
1087 
1088 	prison_free(pr);
1089 	return (0);
1090 }
1091 #endif /* VIMAGE */
1092 
1093 /*
1094  * Add a group to an interface
1095  */
1096 int
1097 if_addgroup(struct ifnet *ifp, const char *groupname)
1098 {
1099 	struct ifg_list		*ifgl;
1100 	struct ifg_group	*ifg = NULL;
1101 	struct ifg_member	*ifgm;
1102 
1103 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1104 	    groupname[strlen(groupname) - 1] <= '9')
1105 		return (EINVAL);
1106 
1107 	IFNET_WLOCK();
1108 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1109 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1110 			IFNET_WUNLOCK();
1111 			return (EEXIST);
1112 		}
1113 
1114 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1115 	    M_NOWAIT)) == NULL) {
1116 	    	IFNET_WUNLOCK();
1117 		return (ENOMEM);
1118 	}
1119 
1120 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1121 	    M_TEMP, M_NOWAIT)) == NULL) {
1122 		free(ifgl, M_TEMP);
1123 		IFNET_WUNLOCK();
1124 		return (ENOMEM);
1125 	}
1126 
1127 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1128 		if (!strcmp(ifg->ifg_group, groupname))
1129 			break;
1130 
1131 	if (ifg == NULL) {
1132 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1133 		    M_TEMP, M_NOWAIT)) == NULL) {
1134 			free(ifgl, M_TEMP);
1135 			free(ifgm, M_TEMP);
1136 			IFNET_WUNLOCK();
1137 			return (ENOMEM);
1138 		}
1139 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1140 		ifg->ifg_refcnt = 0;
1141 		TAILQ_INIT(&ifg->ifg_members);
1142 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1143 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1144 	}
1145 
1146 	ifg->ifg_refcnt++;
1147 	ifgl->ifgl_group = ifg;
1148 	ifgm->ifgm_ifp = ifp;
1149 
1150 	IF_ADDR_LOCK(ifp);
1151 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1152 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1153 	IF_ADDR_UNLOCK(ifp);
1154 
1155 	IFNET_WUNLOCK();
1156 
1157 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1158 
1159 	return (0);
1160 }
1161 
1162 /*
1163  * Remove a group from an interface
1164  */
1165 int
1166 if_delgroup(struct ifnet *ifp, const char *groupname)
1167 {
1168 	struct ifg_list		*ifgl;
1169 	struct ifg_member	*ifgm;
1170 
1171 	IFNET_WLOCK();
1172 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1173 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1174 			break;
1175 	if (ifgl == NULL) {
1176 		IFNET_WUNLOCK();
1177 		return (ENOENT);
1178 	}
1179 
1180 	IF_ADDR_LOCK(ifp);
1181 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1182 	IF_ADDR_UNLOCK(ifp);
1183 
1184 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1185 		if (ifgm->ifgm_ifp == ifp)
1186 			break;
1187 
1188 	if (ifgm != NULL) {
1189 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1190 		free(ifgm, M_TEMP);
1191 	}
1192 
1193 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1194 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1195 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1196 		free(ifgl->ifgl_group, M_TEMP);
1197 	}
1198 	IFNET_WUNLOCK();
1199 
1200 	free(ifgl, M_TEMP);
1201 
1202 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1203 
1204 	return (0);
1205 }
1206 
1207 /*
1208  * Remove an interface from all groups
1209  */
1210 static void
1211 if_delgroups(struct ifnet *ifp)
1212 {
1213 	struct ifg_list		*ifgl;
1214 	struct ifg_member	*ifgm;
1215 	char groupname[IFNAMSIZ];
1216 
1217 	IFNET_WLOCK();
1218 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1219 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1220 
1221 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1222 
1223 		IF_ADDR_LOCK(ifp);
1224 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1225 		IF_ADDR_UNLOCK(ifp);
1226 
1227 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1228 			if (ifgm->ifgm_ifp == ifp)
1229 				break;
1230 
1231 		if (ifgm != NULL) {
1232 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1233 			    ifgm_next);
1234 			free(ifgm, M_TEMP);
1235 		}
1236 
1237 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1238 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1239 			EVENTHANDLER_INVOKE(group_detach_event,
1240 			    ifgl->ifgl_group);
1241 			free(ifgl->ifgl_group, M_TEMP);
1242 		}
1243 		IFNET_WUNLOCK();
1244 
1245 		free(ifgl, M_TEMP);
1246 
1247 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1248 
1249 		IFNET_WLOCK();
1250 	}
1251 	IFNET_WUNLOCK();
1252 }
1253 
1254 /*
1255  * Stores all groups from an interface in memory pointed
1256  * to by data
1257  */
1258 static int
1259 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1260 {
1261 	int			 len, error;
1262 	struct ifg_list		*ifgl;
1263 	struct ifg_req		 ifgrq, *ifgp;
1264 	struct ifgroupreq	*ifgr = data;
1265 
1266 	if (ifgr->ifgr_len == 0) {
1267 		IF_ADDR_LOCK(ifp);
1268 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1269 			ifgr->ifgr_len += sizeof(struct ifg_req);
1270 		IF_ADDR_UNLOCK(ifp);
1271 		return (0);
1272 	}
1273 
1274 	len = ifgr->ifgr_len;
1275 	ifgp = ifgr->ifgr_groups;
1276 	/* XXX: wire */
1277 	IF_ADDR_LOCK(ifp);
1278 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1279 		if (len < sizeof(ifgrq)) {
1280 			IF_ADDR_UNLOCK(ifp);
1281 			return (EINVAL);
1282 		}
1283 		bzero(&ifgrq, sizeof ifgrq);
1284 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1285 		    sizeof(ifgrq.ifgrq_group));
1286 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1287 		    	IF_ADDR_UNLOCK(ifp);
1288 			return (error);
1289 		}
1290 		len -= sizeof(ifgrq);
1291 		ifgp++;
1292 	}
1293 	IF_ADDR_UNLOCK(ifp);
1294 
1295 	return (0);
1296 }
1297 
1298 /*
1299  * Stores all members of a group in memory pointed to by data
1300  */
1301 static int
1302 if_getgroupmembers(struct ifgroupreq *data)
1303 {
1304 	struct ifgroupreq	*ifgr = data;
1305 	struct ifg_group	*ifg;
1306 	struct ifg_member	*ifgm;
1307 	struct ifg_req		 ifgrq, *ifgp;
1308 	int			 len, error;
1309 
1310 	IFNET_RLOCK();
1311 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1312 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1313 			break;
1314 	if (ifg == NULL) {
1315 		IFNET_RUNLOCK();
1316 		return (ENOENT);
1317 	}
1318 
1319 	if (ifgr->ifgr_len == 0) {
1320 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1321 			ifgr->ifgr_len += sizeof(ifgrq);
1322 		IFNET_RUNLOCK();
1323 		return (0);
1324 	}
1325 
1326 	len = ifgr->ifgr_len;
1327 	ifgp = ifgr->ifgr_groups;
1328 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1329 		if (len < sizeof(ifgrq)) {
1330 			IFNET_RUNLOCK();
1331 			return (EINVAL);
1332 		}
1333 		bzero(&ifgrq, sizeof ifgrq);
1334 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1335 		    sizeof(ifgrq.ifgrq_member));
1336 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1337 			IFNET_RUNLOCK();
1338 			return (error);
1339 		}
1340 		len -= sizeof(ifgrq);
1341 		ifgp++;
1342 	}
1343 	IFNET_RUNLOCK();
1344 
1345 	return (0);
1346 }
1347 
1348 /*
1349  * Delete Routes for a Network Interface
1350  *
1351  * Called for each routing entry via the rnh->rnh_walktree() call above
1352  * to delete all route entries referencing a detaching network interface.
1353  *
1354  * Arguments:
1355  *	rn	pointer to node in the routing table
1356  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1357  *
1358  * Returns:
1359  *	0	successful
1360  *	errno	failed - reason indicated
1361  *
1362  */
1363 static int
1364 if_rtdel(struct radix_node *rn, void *arg)
1365 {
1366 	struct rtentry	*rt = (struct rtentry *)rn;
1367 	struct ifnet	*ifp = arg;
1368 	int		err;
1369 
1370 	if (rt->rt_ifp == ifp) {
1371 
1372 		/*
1373 		 * Protect (sorta) against walktree recursion problems
1374 		 * with cloned routes
1375 		 */
1376 		if ((rt->rt_flags & RTF_UP) == 0)
1377 			return (0);
1378 
1379 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1380 				rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1381 				(struct rtentry **) NULL, rt->rt_fibnum);
1382 		if (err) {
1383 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1384 		}
1385 	}
1386 
1387 	return (0);
1388 }
1389 
1390 /*
1391  * Wrapper functions for struct ifnet address list locking macros.  These are
1392  * used by kernel modules to avoid encoding programming interface or binary
1393  * interface assumptions that may be violated when kernel-internal locking
1394  * approaches change.
1395  */
1396 void
1397 if_addr_rlock(struct ifnet *ifp)
1398 {
1399 
1400 	IF_ADDR_LOCK(ifp);
1401 }
1402 
1403 void
1404 if_addr_runlock(struct ifnet *ifp)
1405 {
1406 
1407 	IF_ADDR_UNLOCK(ifp);
1408 }
1409 
1410 void
1411 if_maddr_rlock(struct ifnet *ifp)
1412 {
1413 
1414 	IF_ADDR_LOCK(ifp);
1415 }
1416 
1417 void
1418 if_maddr_runlock(struct ifnet *ifp)
1419 {
1420 
1421 	IF_ADDR_UNLOCK(ifp);
1422 }
1423 
1424 /*
1425  * Reference count functions for ifaddrs.
1426  */
1427 void
1428 ifa_init(struct ifaddr *ifa)
1429 {
1430 
1431 	mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1432 	refcount_init(&ifa->ifa_refcnt, 1);
1433 }
1434 
1435 void
1436 ifa_ref(struct ifaddr *ifa)
1437 {
1438 
1439 	refcount_acquire(&ifa->ifa_refcnt);
1440 }
1441 
1442 void
1443 ifa_free(struct ifaddr *ifa)
1444 {
1445 
1446 	if (refcount_release(&ifa->ifa_refcnt)) {
1447 		mtx_destroy(&ifa->ifa_mtx);
1448 		free(ifa, M_IFADDR);
1449 	}
1450 }
1451 
1452 int
1453 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1454 {
1455 	int error = 0;
1456 	struct rtentry *rt = NULL;
1457 	struct rt_addrinfo info;
1458 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1459 
1460 	bzero(&info, sizeof(info));
1461 	info.rti_ifp = V_loif;
1462 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1463 	info.rti_info[RTAX_DST] = ia;
1464 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1465 	error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1466 
1467 	if (error == 0 && rt != NULL) {
1468 		RT_LOCK(rt);
1469 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1470 			ifa->ifa_ifp->if_type;
1471 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1472 			ifa->ifa_ifp->if_index;
1473 		RT_REMREF(rt);
1474 		RT_UNLOCK(rt);
1475 	} else if (error != 0)
1476 		log(LOG_INFO, "ifa_add_loopback_route: insertion failed\n");
1477 
1478 	return (error);
1479 }
1480 
1481 int
1482 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1483 {
1484 	int error = 0;
1485 	struct rt_addrinfo info;
1486 	struct sockaddr_dl null_sdl;
1487 
1488 	bzero(&null_sdl, sizeof(null_sdl));
1489 	null_sdl.sdl_len = sizeof(null_sdl);
1490 	null_sdl.sdl_family = AF_LINK;
1491 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1492 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1493 	bzero(&info, sizeof(info));
1494 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1495 	info.rti_info[RTAX_DST] = ia;
1496 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1497 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1498 
1499 	if (error != 0)
1500 		log(LOG_INFO, "ifa_del_loopback_route: deletion failed\n");
1501 
1502 	return (error);
1503 }
1504 
1505 /*
1506  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1507  * structs used to represent other address families, it is necessary
1508  * to perform a different comparison.
1509  */
1510 
1511 #define	sa_equal(a1, a2)	\
1512 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1513 
1514 #define	sa_dl_equal(a1, a2)	\
1515 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1516 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1517 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1518 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1519 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1520 
1521 /*
1522  * Locate an interface based on a complete address.
1523  */
1524 /*ARGSUSED*/
1525 static struct ifaddr *
1526 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1527 {
1528 	struct ifnet *ifp;
1529 	struct ifaddr *ifa;
1530 
1531 	IFNET_RLOCK_NOSLEEP();
1532 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1533 		IF_ADDR_LOCK(ifp);
1534 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1535 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1536 				continue;
1537 			if (sa_equal(addr, ifa->ifa_addr)) {
1538 				if (getref)
1539 					ifa_ref(ifa);
1540 				IF_ADDR_UNLOCK(ifp);
1541 				goto done;
1542 			}
1543 			/* IP6 doesn't have broadcast */
1544 			if ((ifp->if_flags & IFF_BROADCAST) &&
1545 			    ifa->ifa_broadaddr &&
1546 			    ifa->ifa_broadaddr->sa_len != 0 &&
1547 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1548 				if (getref)
1549 					ifa_ref(ifa);
1550 				IF_ADDR_UNLOCK(ifp);
1551 				goto done;
1552 			}
1553 		}
1554 		IF_ADDR_UNLOCK(ifp);
1555 	}
1556 	ifa = NULL;
1557 done:
1558 	IFNET_RUNLOCK_NOSLEEP();
1559 	return (ifa);
1560 }
1561 
1562 struct ifaddr *
1563 ifa_ifwithaddr(struct sockaddr *addr)
1564 {
1565 
1566 	return (ifa_ifwithaddr_internal(addr, 1));
1567 }
1568 
1569 int
1570 ifa_ifwithaddr_check(struct sockaddr *addr)
1571 {
1572 
1573 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1574 }
1575 
1576 /*
1577  * Locate an interface based on the broadcast address.
1578  */
1579 /* ARGSUSED */
1580 struct ifaddr *
1581 ifa_ifwithbroadaddr(struct sockaddr *addr)
1582 {
1583 	struct ifnet *ifp;
1584 	struct ifaddr *ifa;
1585 
1586 	IFNET_RLOCK_NOSLEEP();
1587 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1588 		IF_ADDR_LOCK(ifp);
1589 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1590 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1591 				continue;
1592 			if ((ifp->if_flags & IFF_BROADCAST) &&
1593 			    ifa->ifa_broadaddr &&
1594 			    ifa->ifa_broadaddr->sa_len != 0 &&
1595 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1596 				ifa_ref(ifa);
1597 				IF_ADDR_UNLOCK(ifp);
1598 				goto done;
1599 			}
1600 		}
1601 		IF_ADDR_UNLOCK(ifp);
1602 	}
1603 	ifa = NULL;
1604 done:
1605 	IFNET_RUNLOCK_NOSLEEP();
1606 	return (ifa);
1607 }
1608 
1609 /*
1610  * Locate the point to point interface with a given destination address.
1611  */
1612 /*ARGSUSED*/
1613 struct ifaddr *
1614 ifa_ifwithdstaddr(struct sockaddr *addr)
1615 {
1616 	struct ifnet *ifp;
1617 	struct ifaddr *ifa;
1618 
1619 	IFNET_RLOCK_NOSLEEP();
1620 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1621 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1622 			continue;
1623 		IF_ADDR_LOCK(ifp);
1624 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1625 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1626 				continue;
1627 			if (ifa->ifa_dstaddr != NULL &&
1628 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1629 				ifa_ref(ifa);
1630 				IF_ADDR_UNLOCK(ifp);
1631 				goto done;
1632 			}
1633 		}
1634 		IF_ADDR_UNLOCK(ifp);
1635 	}
1636 	ifa = NULL;
1637 done:
1638 	IFNET_RUNLOCK_NOSLEEP();
1639 	return (ifa);
1640 }
1641 
1642 /*
1643  * Find an interface on a specific network.  If many, choice
1644  * is most specific found.
1645  */
1646 struct ifaddr *
1647 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1648 {
1649 	struct ifnet *ifp;
1650 	struct ifaddr *ifa;
1651 	struct ifaddr *ifa_maybe = NULL;
1652 	u_int af = addr->sa_family;
1653 	char *addr_data = addr->sa_data, *cplim;
1654 
1655 	/*
1656 	 * AF_LINK addresses can be looked up directly by their index number,
1657 	 * so do that if we can.
1658 	 */
1659 	if (af == AF_LINK) {
1660 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1661 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1662 		return (ifaddr_byindex(sdl->sdl_index));
1663 	}
1664 
1665 	/*
1666 	 * Scan though each interface, looking for ones that have addresses
1667 	 * in this address family.  Maintain a reference on ifa_maybe once
1668 	 * we find one, as we release the IF_ADDR_LOCK() that kept it stable
1669 	 * when we move onto the next interface.
1670 	 */
1671 	IFNET_RLOCK_NOSLEEP();
1672 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1673 		IF_ADDR_LOCK(ifp);
1674 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1675 			char *cp, *cp2, *cp3;
1676 
1677 			if (ifa->ifa_addr->sa_family != af)
1678 next:				continue;
1679 			if (af == AF_INET &&
1680 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1681 				/*
1682 				 * This is a bit broken as it doesn't
1683 				 * take into account that the remote end may
1684 				 * be a single node in the network we are
1685 				 * looking for.
1686 				 * The trouble is that we don't know the
1687 				 * netmask for the remote end.
1688 				 */
1689 				if (ifa->ifa_dstaddr != NULL &&
1690 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1691 					ifa_ref(ifa);
1692 					IF_ADDR_UNLOCK(ifp);
1693 					goto done;
1694 				}
1695 			} else {
1696 				/*
1697 				 * if we have a special address handler,
1698 				 * then use it instead of the generic one.
1699 				 */
1700 				if (ifa->ifa_claim_addr) {
1701 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1702 						ifa_ref(ifa);
1703 						IF_ADDR_UNLOCK(ifp);
1704 						goto done;
1705 					}
1706 					continue;
1707 				}
1708 
1709 				/*
1710 				 * Scan all the bits in the ifa's address.
1711 				 * If a bit dissagrees with what we are
1712 				 * looking for, mask it with the netmask
1713 				 * to see if it really matters.
1714 				 * (A byte at a time)
1715 				 */
1716 				if (ifa->ifa_netmask == 0)
1717 					continue;
1718 				cp = addr_data;
1719 				cp2 = ifa->ifa_addr->sa_data;
1720 				cp3 = ifa->ifa_netmask->sa_data;
1721 				cplim = ifa->ifa_netmask->sa_len
1722 					+ (char *)ifa->ifa_netmask;
1723 				while (cp3 < cplim)
1724 					if ((*cp++ ^ *cp2++) & *cp3++)
1725 						goto next; /* next address! */
1726 				/*
1727 				 * If the netmask of what we just found
1728 				 * is more specific than what we had before
1729 				 * (if we had one) then remember the new one
1730 				 * before continuing to search
1731 				 * for an even better one.
1732 				 */
1733 				if (ifa_maybe == NULL ||
1734 				    rn_refines((caddr_t)ifa->ifa_netmask,
1735 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1736 					if (ifa_maybe != NULL)
1737 						ifa_free(ifa_maybe);
1738 					ifa_maybe = ifa;
1739 					ifa_ref(ifa_maybe);
1740 				}
1741 			}
1742 		}
1743 		IF_ADDR_UNLOCK(ifp);
1744 	}
1745 	ifa = ifa_maybe;
1746 	ifa_maybe = NULL;
1747 done:
1748 	IFNET_RUNLOCK_NOSLEEP();
1749 	if (ifa_maybe != NULL)
1750 		ifa_free(ifa_maybe);
1751 	return (ifa);
1752 }
1753 
1754 /*
1755  * Find an interface address specific to an interface best matching
1756  * a given address.
1757  */
1758 struct ifaddr *
1759 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1760 {
1761 	struct ifaddr *ifa;
1762 	char *cp, *cp2, *cp3;
1763 	char *cplim;
1764 	struct ifaddr *ifa_maybe = NULL;
1765 	u_int af = addr->sa_family;
1766 
1767 	if (af >= AF_MAX)
1768 		return (NULL);
1769 	IF_ADDR_LOCK(ifp);
1770 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1771 		if (ifa->ifa_addr->sa_family != af)
1772 			continue;
1773 		if (ifa_maybe == NULL)
1774 			ifa_maybe = ifa;
1775 		if (ifa->ifa_netmask == 0) {
1776 			if (sa_equal(addr, ifa->ifa_addr) ||
1777 			    (ifa->ifa_dstaddr &&
1778 			    sa_equal(addr, ifa->ifa_dstaddr)))
1779 				goto done;
1780 			continue;
1781 		}
1782 		if (ifp->if_flags & IFF_POINTOPOINT) {
1783 			if (sa_equal(addr, ifa->ifa_dstaddr))
1784 				goto done;
1785 		} else {
1786 			cp = addr->sa_data;
1787 			cp2 = ifa->ifa_addr->sa_data;
1788 			cp3 = ifa->ifa_netmask->sa_data;
1789 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1790 			for (; cp3 < cplim; cp3++)
1791 				if ((*cp++ ^ *cp2++) & *cp3)
1792 					break;
1793 			if (cp3 == cplim)
1794 				goto done;
1795 		}
1796 	}
1797 	ifa = ifa_maybe;
1798 done:
1799 	if (ifa != NULL)
1800 		ifa_ref(ifa);
1801 	IF_ADDR_UNLOCK(ifp);
1802 	return (ifa);
1803 }
1804 
1805 #include <net/if_llatbl.h>
1806 
1807 /*
1808  * Default action when installing a route with a Link Level gateway.
1809  * Lookup an appropriate real ifa to point to.
1810  * This should be moved to /sys/net/link.c eventually.
1811  */
1812 static void
1813 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1814 {
1815 	struct ifaddr *ifa, *oifa;
1816 	struct sockaddr *dst;
1817 	struct ifnet *ifp;
1818 
1819 	RT_LOCK_ASSERT(rt);
1820 
1821 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1822 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1823 		return;
1824 	ifa = ifaof_ifpforaddr(dst, ifp);
1825 	if (ifa) {
1826 		oifa = rt->rt_ifa;
1827 		rt->rt_ifa = ifa;
1828 		ifa_free(oifa);
1829 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1830 			ifa->ifa_rtrequest(cmd, rt, info);
1831 	}
1832 }
1833 
1834 /*
1835  * Mark an interface down and notify protocols of
1836  * the transition.
1837  * NOTE: must be called at splnet or eqivalent.
1838  */
1839 static void
1840 if_unroute(struct ifnet *ifp, int flag, int fam)
1841 {
1842 	struct ifaddr *ifa;
1843 
1844 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1845 
1846 	ifp->if_flags &= ~flag;
1847 	getmicrotime(&ifp->if_lastchange);
1848 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1849 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1850 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1851 	ifp->if_qflush(ifp);
1852 
1853 	if (ifp->if_carp)
1854 		(*carp_linkstate_p)(ifp);
1855 	rt_ifmsg(ifp);
1856 }
1857 
1858 /*
1859  * Mark an interface up and notify protocols of
1860  * the transition.
1861  * NOTE: must be called at splnet or eqivalent.
1862  */
1863 static void
1864 if_route(struct ifnet *ifp, int flag, int fam)
1865 {
1866 	struct ifaddr *ifa;
1867 
1868 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1869 
1870 	ifp->if_flags |= flag;
1871 	getmicrotime(&ifp->if_lastchange);
1872 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1873 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1874 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1875 	if (ifp->if_carp)
1876 		(*carp_linkstate_p)(ifp);
1877 	rt_ifmsg(ifp);
1878 #ifdef INET6
1879 	in6_if_up(ifp);
1880 #endif
1881 }
1882 
1883 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1884 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1885 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1886 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1887 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
1888 int	(*vlan_setcookie_p)(struct ifnet *, void *);
1889 void	*(*vlan_cookie_p)(struct ifnet *);
1890 
1891 /*
1892  * Handle a change in the interface link state. To avoid LORs
1893  * between driver lock and upper layer locks, as well as possible
1894  * recursions, we post event to taskqueue, and all job
1895  * is done in static do_link_state_change().
1896  */
1897 void
1898 if_link_state_change(struct ifnet *ifp, int link_state)
1899 {
1900 	/* Return if state hasn't changed. */
1901 	if (ifp->if_link_state == link_state)
1902 		return;
1903 
1904 	ifp->if_link_state = link_state;
1905 
1906 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1907 }
1908 
1909 static void
1910 do_link_state_change(void *arg, int pending)
1911 {
1912 	struct ifnet *ifp = (struct ifnet *)arg;
1913 	int link_state = ifp->if_link_state;
1914 	CURVNET_SET(ifp->if_vnet);
1915 
1916 	/* Notify that the link state has changed. */
1917 	rt_ifmsg(ifp);
1918 	if (ifp->if_vlantrunk != NULL)
1919 		(*vlan_link_state_p)(ifp);
1920 
1921 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1922 	    IFP2AC(ifp)->ac_netgraph != NULL)
1923 		(*ng_ether_link_state_p)(ifp, link_state);
1924 	if (ifp->if_carp)
1925 		(*carp_linkstate_p)(ifp);
1926 	if (ifp->if_bridge) {
1927 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1928 		(*bstp_linkstate_p)(ifp, link_state);
1929 	}
1930 	if (ifp->if_lagg) {
1931 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1932 		(*lagg_linkstate_p)(ifp, link_state);
1933 	}
1934 
1935 	if (IS_DEFAULT_VNET(curvnet))
1936 		devctl_notify("IFNET", ifp->if_xname,
1937 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1938 		    NULL);
1939 	if (pending > 1)
1940 		if_printf(ifp, "%d link states coalesced\n", pending);
1941 	if (log_link_state_change)
1942 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1943 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1944 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
1945 	CURVNET_RESTORE();
1946 }
1947 
1948 /*
1949  * Mark an interface down and notify protocols of
1950  * the transition.
1951  * NOTE: must be called at splnet or eqivalent.
1952  */
1953 void
1954 if_down(struct ifnet *ifp)
1955 {
1956 
1957 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1958 }
1959 
1960 /*
1961  * Mark an interface up and notify protocols of
1962  * the transition.
1963  * NOTE: must be called at splnet or eqivalent.
1964  */
1965 void
1966 if_up(struct ifnet *ifp)
1967 {
1968 
1969 	if_route(ifp, IFF_UP, AF_UNSPEC);
1970 }
1971 
1972 /*
1973  * Flush an interface queue.
1974  */
1975 void
1976 if_qflush(struct ifnet *ifp)
1977 {
1978 	struct mbuf *m, *n;
1979 	struct ifaltq *ifq;
1980 
1981 	ifq = &ifp->if_snd;
1982 	IFQ_LOCK(ifq);
1983 #ifdef ALTQ
1984 	if (ALTQ_IS_ENABLED(ifq))
1985 		ALTQ_PURGE(ifq);
1986 #endif
1987 	n = ifq->ifq_head;
1988 	while ((m = n) != 0) {
1989 		n = m->m_act;
1990 		m_freem(m);
1991 	}
1992 	ifq->ifq_head = 0;
1993 	ifq->ifq_tail = 0;
1994 	ifq->ifq_len = 0;
1995 	IFQ_UNLOCK(ifq);
1996 }
1997 
1998 /*
1999  * Map interface name to interface structure pointer, with or without
2000  * returning a reference.
2001  */
2002 struct ifnet *
2003 ifunit_ref(const char *name)
2004 {
2005 	struct ifnet *ifp;
2006 
2007 	IFNET_RLOCK_NOSLEEP();
2008 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2009 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2010 		    !(ifp->if_flags & IFF_DYING))
2011 			break;
2012 	}
2013 	if (ifp != NULL)
2014 		if_ref(ifp);
2015 	IFNET_RUNLOCK_NOSLEEP();
2016 	return (ifp);
2017 }
2018 
2019 struct ifnet *
2020 ifunit(const char *name)
2021 {
2022 	struct ifnet *ifp;
2023 
2024 	IFNET_RLOCK_NOSLEEP();
2025 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2026 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2027 			break;
2028 	}
2029 	IFNET_RUNLOCK_NOSLEEP();
2030 	return (ifp);
2031 }
2032 
2033 /*
2034  * Hardware specific interface ioctls.
2035  */
2036 static int
2037 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2038 {
2039 	struct ifreq *ifr;
2040 	struct ifstat *ifs;
2041 	int error = 0;
2042 	int new_flags, temp_flags;
2043 	size_t namelen, onamelen;
2044 	size_t descrlen;
2045 	char *descrbuf, *odescrbuf;
2046 	char new_name[IFNAMSIZ];
2047 	struct ifaddr *ifa;
2048 	struct sockaddr_dl *sdl;
2049 
2050 	ifr = (struct ifreq *)data;
2051 	switch (cmd) {
2052 	case SIOCGIFINDEX:
2053 		ifr->ifr_index = ifp->if_index;
2054 		break;
2055 
2056 	case SIOCGIFFLAGS:
2057 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2058 		ifr->ifr_flags = temp_flags & 0xffff;
2059 		ifr->ifr_flagshigh = temp_flags >> 16;
2060 		break;
2061 
2062 	case SIOCGIFCAP:
2063 		ifr->ifr_reqcap = ifp->if_capabilities;
2064 		ifr->ifr_curcap = ifp->if_capenable;
2065 		break;
2066 
2067 #ifdef MAC
2068 	case SIOCGIFMAC:
2069 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2070 		break;
2071 #endif
2072 
2073 	case SIOCGIFMETRIC:
2074 		ifr->ifr_metric = ifp->if_metric;
2075 		break;
2076 
2077 	case SIOCGIFMTU:
2078 		ifr->ifr_mtu = ifp->if_mtu;
2079 		break;
2080 
2081 	case SIOCGIFPHYS:
2082 		ifr->ifr_phys = ifp->if_physical;
2083 		break;
2084 
2085 	case SIOCGIFDESCR:
2086 		error = 0;
2087 		sx_slock(&ifdescr_sx);
2088 		if (ifp->if_description == NULL)
2089 			error = ENOMSG;
2090 		else {
2091 			/* space for terminating nul */
2092 			descrlen = strlen(ifp->if_description) + 1;
2093 			if (ifr->ifr_buffer.length < descrlen)
2094 				ifr->ifr_buffer.buffer = NULL;
2095 			else
2096 				error = copyout(ifp->if_description,
2097 				    ifr->ifr_buffer.buffer, descrlen);
2098 			ifr->ifr_buffer.length = descrlen;
2099 		}
2100 		sx_sunlock(&ifdescr_sx);
2101 		break;
2102 
2103 	case SIOCSIFDESCR:
2104 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2105 		if (error)
2106 			return (error);
2107 
2108 		/*
2109 		 * Copy only (length-1) bytes to make sure that
2110 		 * if_description is always nul terminated.  The
2111 		 * length parameter is supposed to count the
2112 		 * terminating nul in.
2113 		 */
2114 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2115 			return (ENAMETOOLONG);
2116 		else if (ifr->ifr_buffer.length == 0)
2117 			descrbuf = NULL;
2118 		else {
2119 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2120 			    M_WAITOK | M_ZERO);
2121 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2122 			    ifr->ifr_buffer.length - 1);
2123 			if (error) {
2124 				free(descrbuf, M_IFDESCR);
2125 				break;
2126 			}
2127 		}
2128 
2129 		sx_xlock(&ifdescr_sx);
2130 		odescrbuf = ifp->if_description;
2131 		ifp->if_description = descrbuf;
2132 		sx_xunlock(&ifdescr_sx);
2133 
2134 		getmicrotime(&ifp->if_lastchange);
2135 		free(odescrbuf, M_IFDESCR);
2136 		break;
2137 
2138 	case SIOCSIFFLAGS:
2139 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2140 		if (error)
2141 			return (error);
2142 		/*
2143 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2144 		 * check, so we don't need special handling here yet.
2145 		 */
2146 		new_flags = (ifr->ifr_flags & 0xffff) |
2147 		    (ifr->ifr_flagshigh << 16);
2148 		if (ifp->if_flags & IFF_SMART) {
2149 			/* Smart drivers twiddle their own routes */
2150 		} else if (ifp->if_flags & IFF_UP &&
2151 		    (new_flags & IFF_UP) == 0) {
2152 			int s = splimp();
2153 			if_down(ifp);
2154 			splx(s);
2155 		} else if (new_flags & IFF_UP &&
2156 		    (ifp->if_flags & IFF_UP) == 0) {
2157 			int s = splimp();
2158 			if_up(ifp);
2159 			splx(s);
2160 		}
2161 		/* See if permanently promiscuous mode bit is about to flip */
2162 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2163 			if (new_flags & IFF_PPROMISC)
2164 				ifp->if_flags |= IFF_PROMISC;
2165 			else if (ifp->if_pcount == 0)
2166 				ifp->if_flags &= ~IFF_PROMISC;
2167 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2168 			    ifp->if_xname,
2169 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2170 		}
2171 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2172 			(new_flags &~ IFF_CANTCHANGE);
2173 		if (ifp->if_ioctl) {
2174 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2175 		}
2176 		getmicrotime(&ifp->if_lastchange);
2177 		break;
2178 
2179 	case SIOCSIFCAP:
2180 		error = priv_check(td, PRIV_NET_SETIFCAP);
2181 		if (error)
2182 			return (error);
2183 		if (ifp->if_ioctl == NULL)
2184 			return (EOPNOTSUPP);
2185 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2186 			return (EINVAL);
2187 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2188 		if (error == 0)
2189 			getmicrotime(&ifp->if_lastchange);
2190 		break;
2191 
2192 #ifdef MAC
2193 	case SIOCSIFMAC:
2194 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2195 		break;
2196 #endif
2197 
2198 	case SIOCSIFNAME:
2199 		error = priv_check(td, PRIV_NET_SETIFNAME);
2200 		if (error)
2201 			return (error);
2202 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2203 		if (error != 0)
2204 			return (error);
2205 		if (new_name[0] == '\0')
2206 			return (EINVAL);
2207 		if (ifunit(new_name) != NULL)
2208 			return (EEXIST);
2209 
2210 		/*
2211 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2212 		 * and there are numerous other races with the
2213 		 * ifunit() checks not being atomic with namespace
2214 		 * changes (renames, vmoves, if_attach, etc).
2215 		 */
2216 		ifp->if_flags |= IFF_RENAMING;
2217 
2218 		/* Announce the departure of the interface. */
2219 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2220 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2221 
2222 		log(LOG_INFO, "%s: changing name to '%s'\n",
2223 		    ifp->if_xname, new_name);
2224 
2225 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2226 		ifa = ifp->if_addr;
2227 		IFA_LOCK(ifa);
2228 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2229 		namelen = strlen(new_name);
2230 		onamelen = sdl->sdl_nlen;
2231 		/*
2232 		 * Move the address if needed.  This is safe because we
2233 		 * allocate space for a name of length IFNAMSIZ when we
2234 		 * create this in if_attach().
2235 		 */
2236 		if (namelen != onamelen) {
2237 			bcopy(sdl->sdl_data + onamelen,
2238 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2239 		}
2240 		bcopy(new_name, sdl->sdl_data, namelen);
2241 		sdl->sdl_nlen = namelen;
2242 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2243 		bzero(sdl->sdl_data, onamelen);
2244 		while (namelen != 0)
2245 			sdl->sdl_data[--namelen] = 0xff;
2246 		IFA_UNLOCK(ifa);
2247 
2248 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2249 		/* Announce the return of the interface. */
2250 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2251 
2252 		ifp->if_flags &= ~IFF_RENAMING;
2253 		break;
2254 
2255 #ifdef VIMAGE
2256 	case SIOCSIFVNET:
2257 		error = priv_check(td, PRIV_NET_SETIFVNET);
2258 		if (error)
2259 			return (error);
2260 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2261 		break;
2262 #endif
2263 
2264 	case SIOCSIFMETRIC:
2265 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2266 		if (error)
2267 			return (error);
2268 		ifp->if_metric = ifr->ifr_metric;
2269 		getmicrotime(&ifp->if_lastchange);
2270 		break;
2271 
2272 	case SIOCSIFPHYS:
2273 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2274 		if (error)
2275 			return (error);
2276 		if (ifp->if_ioctl == NULL)
2277 			return (EOPNOTSUPP);
2278 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2279 		if (error == 0)
2280 			getmicrotime(&ifp->if_lastchange);
2281 		break;
2282 
2283 	case SIOCSIFMTU:
2284 	{
2285 		u_long oldmtu = ifp->if_mtu;
2286 
2287 		error = priv_check(td, PRIV_NET_SETIFMTU);
2288 		if (error)
2289 			return (error);
2290 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2291 			return (EINVAL);
2292 		if (ifp->if_ioctl == NULL)
2293 			return (EOPNOTSUPP);
2294 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2295 		if (error == 0) {
2296 			getmicrotime(&ifp->if_lastchange);
2297 			rt_ifmsg(ifp);
2298 		}
2299 		/*
2300 		 * If the link MTU changed, do network layer specific procedure.
2301 		 */
2302 		if (ifp->if_mtu != oldmtu) {
2303 #ifdef INET6
2304 			nd6_setmtu(ifp);
2305 #endif
2306 		}
2307 		break;
2308 	}
2309 
2310 	case SIOCADDMULTI:
2311 	case SIOCDELMULTI:
2312 		if (cmd == SIOCADDMULTI)
2313 			error = priv_check(td, PRIV_NET_ADDMULTI);
2314 		else
2315 			error = priv_check(td, PRIV_NET_DELMULTI);
2316 		if (error)
2317 			return (error);
2318 
2319 		/* Don't allow group membership on non-multicast interfaces. */
2320 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2321 			return (EOPNOTSUPP);
2322 
2323 		/* Don't let users screw up protocols' entries. */
2324 		if (ifr->ifr_addr.sa_family != AF_LINK)
2325 			return (EINVAL);
2326 
2327 		if (cmd == SIOCADDMULTI) {
2328 			struct ifmultiaddr *ifma;
2329 
2330 			/*
2331 			 * Userland is only permitted to join groups once
2332 			 * via the if_addmulti() KPI, because it cannot hold
2333 			 * struct ifmultiaddr * between calls. It may also
2334 			 * lose a race while we check if the membership
2335 			 * already exists.
2336 			 */
2337 			IF_ADDR_LOCK(ifp);
2338 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2339 			IF_ADDR_UNLOCK(ifp);
2340 			if (ifma != NULL)
2341 				error = EADDRINUSE;
2342 			else
2343 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2344 		} else {
2345 			error = if_delmulti(ifp, &ifr->ifr_addr);
2346 		}
2347 		if (error == 0)
2348 			getmicrotime(&ifp->if_lastchange);
2349 		break;
2350 
2351 	case SIOCSIFPHYADDR:
2352 	case SIOCDIFPHYADDR:
2353 #ifdef INET6
2354 	case SIOCSIFPHYADDR_IN6:
2355 #endif
2356 	case SIOCSLIFPHYADDR:
2357 	case SIOCSIFMEDIA:
2358 	case SIOCSIFGENERIC:
2359 		error = priv_check(td, PRIV_NET_HWIOCTL);
2360 		if (error)
2361 			return (error);
2362 		if (ifp->if_ioctl == NULL)
2363 			return (EOPNOTSUPP);
2364 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2365 		if (error == 0)
2366 			getmicrotime(&ifp->if_lastchange);
2367 		break;
2368 
2369 	case SIOCGIFSTATUS:
2370 		ifs = (struct ifstat *)data;
2371 		ifs->ascii[0] = '\0';
2372 
2373 	case SIOCGIFPSRCADDR:
2374 	case SIOCGIFPDSTADDR:
2375 	case SIOCGLIFPHYADDR:
2376 	case SIOCGIFMEDIA:
2377 	case SIOCGIFGENERIC:
2378 		if (ifp->if_ioctl == NULL)
2379 			return (EOPNOTSUPP);
2380 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2381 		break;
2382 
2383 	case SIOCSIFLLADDR:
2384 		error = priv_check(td, PRIV_NET_SETLLADDR);
2385 		if (error)
2386 			return (error);
2387 		error = if_setlladdr(ifp,
2388 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2389 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2390 		break;
2391 
2392 	case SIOCAIFGROUP:
2393 	{
2394 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2395 
2396 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2397 		if (error)
2398 			return (error);
2399 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2400 			return (error);
2401 		break;
2402 	}
2403 
2404 	case SIOCGIFGROUP:
2405 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2406 			return (error);
2407 		break;
2408 
2409 	case SIOCDIFGROUP:
2410 	{
2411 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2412 
2413 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2414 		if (error)
2415 			return (error);
2416 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2417 			return (error);
2418 		break;
2419 	}
2420 
2421 	default:
2422 		error = ENOIOCTL;
2423 		break;
2424 	}
2425 	return (error);
2426 }
2427 
2428 #ifdef COMPAT_FREEBSD32
2429 struct ifconf32 {
2430 	int32_t	ifc_len;
2431 	union {
2432 		uint32_t	ifcu_buf;
2433 		uint32_t	ifcu_req;
2434 	} ifc_ifcu;
2435 };
2436 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2437 #endif
2438 
2439 /*
2440  * Interface ioctls.
2441  */
2442 int
2443 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2444 {
2445 	struct ifnet *ifp;
2446 	struct ifreq *ifr;
2447 	int error;
2448 	int oif_flags;
2449 
2450 	CURVNET_SET(so->so_vnet);
2451 	switch (cmd) {
2452 	case SIOCGIFCONF:
2453 	case OSIOCGIFCONF:
2454 		error = ifconf(cmd, data);
2455 		CURVNET_RESTORE();
2456 		return (error);
2457 
2458 #ifdef COMPAT_FREEBSD32
2459 	case SIOCGIFCONF32:
2460 		{
2461 			struct ifconf32 *ifc32;
2462 			struct ifconf ifc;
2463 
2464 			ifc32 = (struct ifconf32 *)data;
2465 			ifc.ifc_len = ifc32->ifc_len;
2466 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2467 
2468 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2469 			CURVNET_RESTORE();
2470 			return (error);
2471 		}
2472 #endif
2473 	}
2474 	ifr = (struct ifreq *)data;
2475 
2476 	switch (cmd) {
2477 #ifdef VIMAGE
2478 	case SIOCSIFRVNET:
2479 		error = priv_check(td, PRIV_NET_SETIFVNET);
2480 		if (error == 0)
2481 			error = if_vmove_reclaim(td, ifr->ifr_name,
2482 			    ifr->ifr_jid);
2483 		CURVNET_RESTORE();
2484 		return (error);
2485 #endif
2486 	case SIOCIFCREATE:
2487 	case SIOCIFCREATE2:
2488 		error = priv_check(td, PRIV_NET_IFCREATE);
2489 		if (error == 0)
2490 			error = if_clone_create(ifr->ifr_name,
2491 			    sizeof(ifr->ifr_name),
2492 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2493 		CURVNET_RESTORE();
2494 		return (error);
2495 	case SIOCIFDESTROY:
2496 		error = priv_check(td, PRIV_NET_IFDESTROY);
2497 		if (error == 0)
2498 			error = if_clone_destroy(ifr->ifr_name);
2499 		CURVNET_RESTORE();
2500 		return (error);
2501 
2502 	case SIOCIFGCLONERS:
2503 		error = if_clone_list((struct if_clonereq *)data);
2504 		CURVNET_RESTORE();
2505 		return (error);
2506 	case SIOCGIFGMEMB:
2507 		error = if_getgroupmembers((struct ifgroupreq *)data);
2508 		CURVNET_RESTORE();
2509 		return (error);
2510 	}
2511 
2512 	ifp = ifunit_ref(ifr->ifr_name);
2513 	if (ifp == NULL) {
2514 		CURVNET_RESTORE();
2515 		return (ENXIO);
2516 	}
2517 
2518 	error = ifhwioctl(cmd, ifp, data, td);
2519 	if (error != ENOIOCTL) {
2520 		if_rele(ifp);
2521 		CURVNET_RESTORE();
2522 		return (error);
2523 	}
2524 
2525 	oif_flags = ifp->if_flags;
2526 	if (so->so_proto == NULL) {
2527 		if_rele(ifp);
2528 		CURVNET_RESTORE();
2529 		return (EOPNOTSUPP);
2530 	}
2531 #ifndef COMPAT_43
2532 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2533 								 data,
2534 								 ifp, td));
2535 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2536 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2537 #else
2538 	{
2539 		u_long ocmd = cmd;
2540 
2541 		switch (cmd) {
2542 
2543 		case SIOCSIFDSTADDR:
2544 		case SIOCSIFADDR:
2545 		case SIOCSIFBRDADDR:
2546 		case SIOCSIFNETMASK:
2547 #if BYTE_ORDER != BIG_ENDIAN
2548 			if (ifr->ifr_addr.sa_family == 0 &&
2549 			    ifr->ifr_addr.sa_len < 16) {
2550 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2551 				ifr->ifr_addr.sa_len = 16;
2552 			}
2553 #else
2554 			if (ifr->ifr_addr.sa_len == 0)
2555 				ifr->ifr_addr.sa_len = 16;
2556 #endif
2557 			break;
2558 
2559 		case OSIOCGIFADDR:
2560 			cmd = SIOCGIFADDR;
2561 			break;
2562 
2563 		case OSIOCGIFDSTADDR:
2564 			cmd = SIOCGIFDSTADDR;
2565 			break;
2566 
2567 		case OSIOCGIFBRDADDR:
2568 			cmd = SIOCGIFBRDADDR;
2569 			break;
2570 
2571 		case OSIOCGIFNETMASK:
2572 			cmd = SIOCGIFNETMASK;
2573 		}
2574 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2575 								   cmd,
2576 								   data,
2577 								   ifp, td));
2578 		if (error == EOPNOTSUPP && ifp != NULL &&
2579 		    ifp->if_ioctl != NULL)
2580 			error = (*ifp->if_ioctl)(ifp, cmd, data);
2581 		switch (ocmd) {
2582 
2583 		case OSIOCGIFADDR:
2584 		case OSIOCGIFDSTADDR:
2585 		case OSIOCGIFBRDADDR:
2586 		case OSIOCGIFNETMASK:
2587 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2588 
2589 		}
2590 	}
2591 #endif /* COMPAT_43 */
2592 
2593 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2594 #ifdef INET6
2595 		if (ifp->if_flags & IFF_UP) {
2596 			int s = splimp();
2597 			in6_if_up(ifp);
2598 			splx(s);
2599 		}
2600 #endif
2601 	}
2602 	if_rele(ifp);
2603 	CURVNET_RESTORE();
2604 	return (error);
2605 }
2606 
2607 /*
2608  * The code common to handling reference counted flags,
2609  * e.g., in ifpromisc() and if_allmulti().
2610  * The "pflag" argument can specify a permanent mode flag to check,
2611  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2612  *
2613  * Only to be used on stack-owned flags, not driver-owned flags.
2614  */
2615 static int
2616 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2617 {
2618 	struct ifreq ifr;
2619 	int error;
2620 	int oldflags, oldcount;
2621 
2622 	/* Sanity checks to catch programming errors */
2623 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2624 	    ("%s: setting driver-owned flag %d", __func__, flag));
2625 
2626 	if (onswitch)
2627 		KASSERT(*refcount >= 0,
2628 		    ("%s: increment negative refcount %d for flag %d",
2629 		    __func__, *refcount, flag));
2630 	else
2631 		KASSERT(*refcount > 0,
2632 		    ("%s: decrement non-positive refcount %d for flag %d",
2633 		    __func__, *refcount, flag));
2634 
2635 	/* In case this mode is permanent, just touch refcount */
2636 	if (ifp->if_flags & pflag) {
2637 		*refcount += onswitch ? 1 : -1;
2638 		return (0);
2639 	}
2640 
2641 	/* Save ifnet parameters for if_ioctl() may fail */
2642 	oldcount = *refcount;
2643 	oldflags = ifp->if_flags;
2644 
2645 	/*
2646 	 * See if we aren't the only and touching refcount is enough.
2647 	 * Actually toggle interface flag if we are the first or last.
2648 	 */
2649 	if (onswitch) {
2650 		if ((*refcount)++)
2651 			return (0);
2652 		ifp->if_flags |= flag;
2653 	} else {
2654 		if (--(*refcount))
2655 			return (0);
2656 		ifp->if_flags &= ~flag;
2657 	}
2658 
2659 	/* Call down the driver since we've changed interface flags */
2660 	if (ifp->if_ioctl == NULL) {
2661 		error = EOPNOTSUPP;
2662 		goto recover;
2663 	}
2664 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2665 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2666 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2667 	if (error)
2668 		goto recover;
2669 	/* Notify userland that interface flags have changed */
2670 	rt_ifmsg(ifp);
2671 	return (0);
2672 
2673 recover:
2674 	/* Recover after driver error */
2675 	*refcount = oldcount;
2676 	ifp->if_flags = oldflags;
2677 	return (error);
2678 }
2679 
2680 /*
2681  * Set/clear promiscuous mode on interface ifp based on the truth value
2682  * of pswitch.  The calls are reference counted so that only the first
2683  * "on" request actually has an effect, as does the final "off" request.
2684  * Results are undefined if the "off" and "on" requests are not matched.
2685  */
2686 int
2687 ifpromisc(struct ifnet *ifp, int pswitch)
2688 {
2689 	int error;
2690 	int oldflags = ifp->if_flags;
2691 
2692 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2693 			   &ifp->if_pcount, pswitch);
2694 	/* If promiscuous mode status has changed, log a message */
2695 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2696 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2697 		    ifp->if_xname,
2698 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2699 	return (error);
2700 }
2701 
2702 /*
2703  * Return interface configuration
2704  * of system.  List may be used
2705  * in later ioctl's (above) to get
2706  * other information.
2707  */
2708 /*ARGSUSED*/
2709 static int
2710 ifconf(u_long cmd, caddr_t data)
2711 {
2712 	struct ifconf *ifc = (struct ifconf *)data;
2713 	struct ifnet *ifp;
2714 	struct ifaddr *ifa;
2715 	struct ifreq ifr;
2716 	struct sbuf *sb;
2717 	int error, full = 0, valid_len, max_len;
2718 
2719 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2720 	max_len = MAXPHYS - 1;
2721 
2722 	/* Prevent hostile input from being able to crash the system */
2723 	if (ifc->ifc_len <= 0)
2724 		return (EINVAL);
2725 
2726 again:
2727 	if (ifc->ifc_len <= max_len) {
2728 		max_len = ifc->ifc_len;
2729 		full = 1;
2730 	}
2731 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2732 	max_len = 0;
2733 	valid_len = 0;
2734 
2735 	IFNET_RLOCK();
2736 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2737 		int addrs;
2738 
2739 		/*
2740 		 * Zero the ifr_name buffer to make sure we don't
2741 		 * disclose the contents of the stack.
2742 		 */
2743 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2744 
2745 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2746 		    >= sizeof(ifr.ifr_name)) {
2747 			sbuf_delete(sb);
2748 			IFNET_RUNLOCK();
2749 			return (ENAMETOOLONG);
2750 		}
2751 
2752 		addrs = 0;
2753 		IF_ADDR_LOCK(ifp);
2754 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2755 			struct sockaddr *sa = ifa->ifa_addr;
2756 
2757 			if (prison_if(curthread->td_ucred, sa) != 0)
2758 				continue;
2759 			addrs++;
2760 #ifdef COMPAT_43
2761 			if (cmd == OSIOCGIFCONF) {
2762 				struct osockaddr *osa =
2763 					 (struct osockaddr *)&ifr.ifr_addr;
2764 				ifr.ifr_addr = *sa;
2765 				osa->sa_family = sa->sa_family;
2766 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2767 				max_len += sizeof(ifr);
2768 			} else
2769 #endif
2770 			if (sa->sa_len <= sizeof(*sa)) {
2771 				ifr.ifr_addr = *sa;
2772 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2773 				max_len += sizeof(ifr);
2774 			} else {
2775 				sbuf_bcat(sb, &ifr,
2776 				    offsetof(struct ifreq, ifr_addr));
2777 				max_len += offsetof(struct ifreq, ifr_addr);
2778 				sbuf_bcat(sb, sa, sa->sa_len);
2779 				max_len += sa->sa_len;
2780 			}
2781 
2782 			if (sbuf_error(sb) == 0)
2783 				valid_len = sbuf_len(sb);
2784 		}
2785 		IF_ADDR_UNLOCK(ifp);
2786 		if (addrs == 0) {
2787 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2788 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2789 			max_len += sizeof(ifr);
2790 
2791 			if (sbuf_error(sb) == 0)
2792 				valid_len = sbuf_len(sb);
2793 		}
2794 	}
2795 	IFNET_RUNLOCK();
2796 
2797 	/*
2798 	 * If we didn't allocate enough space (uncommon), try again.  If
2799 	 * we have already allocated as much space as we are allowed,
2800 	 * return what we've got.
2801 	 */
2802 	if (valid_len != max_len && !full) {
2803 		sbuf_delete(sb);
2804 		goto again;
2805 	}
2806 
2807 	ifc->ifc_len = valid_len;
2808 	sbuf_finish(sb);
2809 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2810 	sbuf_delete(sb);
2811 	return (error);
2812 }
2813 
2814 /*
2815  * Just like ifpromisc(), but for all-multicast-reception mode.
2816  */
2817 int
2818 if_allmulti(struct ifnet *ifp, int onswitch)
2819 {
2820 
2821 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2822 }
2823 
2824 struct ifmultiaddr *
2825 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2826 {
2827 	struct ifmultiaddr *ifma;
2828 
2829 	IF_ADDR_LOCK_ASSERT(ifp);
2830 
2831 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2832 		if (sa->sa_family == AF_LINK) {
2833 			if (sa_dl_equal(ifma->ifma_addr, sa))
2834 				break;
2835 		} else {
2836 			if (sa_equal(ifma->ifma_addr, sa))
2837 				break;
2838 		}
2839 	}
2840 
2841 	return ifma;
2842 }
2843 
2844 /*
2845  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2846  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2847  * the ifnet multicast address list here, so the caller must do that and
2848  * other setup work (such as notifying the device driver).  The reference
2849  * count is initialized to 1.
2850  */
2851 static struct ifmultiaddr *
2852 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2853     int mflags)
2854 {
2855 	struct ifmultiaddr *ifma;
2856 	struct sockaddr *dupsa;
2857 
2858 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2859 	    M_ZERO);
2860 	if (ifma == NULL)
2861 		return (NULL);
2862 
2863 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2864 	if (dupsa == NULL) {
2865 		free(ifma, M_IFMADDR);
2866 		return (NULL);
2867 	}
2868 	bcopy(sa, dupsa, sa->sa_len);
2869 	ifma->ifma_addr = dupsa;
2870 
2871 	ifma->ifma_ifp = ifp;
2872 	ifma->ifma_refcount = 1;
2873 	ifma->ifma_protospec = NULL;
2874 
2875 	if (llsa == NULL) {
2876 		ifma->ifma_lladdr = NULL;
2877 		return (ifma);
2878 	}
2879 
2880 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2881 	if (dupsa == NULL) {
2882 		free(ifma->ifma_addr, M_IFMADDR);
2883 		free(ifma, M_IFMADDR);
2884 		return (NULL);
2885 	}
2886 	bcopy(llsa, dupsa, llsa->sa_len);
2887 	ifma->ifma_lladdr = dupsa;
2888 
2889 	return (ifma);
2890 }
2891 
2892 /*
2893  * if_freemulti: free ifmultiaddr structure and possibly attached related
2894  * addresses.  The caller is responsible for implementing reference
2895  * counting, notifying the driver, handling routing messages, and releasing
2896  * any dependent link layer state.
2897  */
2898 static void
2899 if_freemulti(struct ifmultiaddr *ifma)
2900 {
2901 
2902 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2903 	    ifma->ifma_refcount));
2904 	KASSERT(ifma->ifma_protospec == NULL,
2905 	    ("if_freemulti: protospec not NULL"));
2906 
2907 	if (ifma->ifma_lladdr != NULL)
2908 		free(ifma->ifma_lladdr, M_IFMADDR);
2909 	free(ifma->ifma_addr, M_IFMADDR);
2910 	free(ifma, M_IFMADDR);
2911 }
2912 
2913 /*
2914  * Register an additional multicast address with a network interface.
2915  *
2916  * - If the address is already present, bump the reference count on the
2917  *   address and return.
2918  * - If the address is not link-layer, look up a link layer address.
2919  * - Allocate address structures for one or both addresses, and attach to the
2920  *   multicast address list on the interface.  If automatically adding a link
2921  *   layer address, the protocol address will own a reference to the link
2922  *   layer address, to be freed when it is freed.
2923  * - Notify the network device driver of an addition to the multicast address
2924  *   list.
2925  *
2926  * 'sa' points to caller-owned memory with the desired multicast address.
2927  *
2928  * 'retifma' will be used to return a pointer to the resulting multicast
2929  * address reference, if desired.
2930  */
2931 int
2932 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2933     struct ifmultiaddr **retifma)
2934 {
2935 	struct ifmultiaddr *ifma, *ll_ifma;
2936 	struct sockaddr *llsa;
2937 	int error;
2938 
2939 	/*
2940 	 * If the address is already present, return a new reference to it;
2941 	 * otherwise, allocate storage and set up a new address.
2942 	 */
2943 	IF_ADDR_LOCK(ifp);
2944 	ifma = if_findmulti(ifp, sa);
2945 	if (ifma != NULL) {
2946 		ifma->ifma_refcount++;
2947 		if (retifma != NULL)
2948 			*retifma = ifma;
2949 		IF_ADDR_UNLOCK(ifp);
2950 		return (0);
2951 	}
2952 
2953 	/*
2954 	 * The address isn't already present; resolve the protocol address
2955 	 * into a link layer address, and then look that up, bump its
2956 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2957 	 * returned, we will need to free it later.
2958 	 */
2959 	llsa = NULL;
2960 	ll_ifma = NULL;
2961 	if (ifp->if_resolvemulti != NULL) {
2962 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2963 		if (error)
2964 			goto unlock_out;
2965 	}
2966 
2967 	/*
2968 	 * Allocate the new address.  Don't hook it up yet, as we may also
2969 	 * need to allocate a link layer multicast address.
2970 	 */
2971 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2972 	if (ifma == NULL) {
2973 		error = ENOMEM;
2974 		goto free_llsa_out;
2975 	}
2976 
2977 	/*
2978 	 * If a link layer address is found, we'll need to see if it's
2979 	 * already present in the address list, or allocate is as well.
2980 	 * When this block finishes, the link layer address will be on the
2981 	 * list.
2982 	 */
2983 	if (llsa != NULL) {
2984 		ll_ifma = if_findmulti(ifp, llsa);
2985 		if (ll_ifma == NULL) {
2986 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2987 			if (ll_ifma == NULL) {
2988 				--ifma->ifma_refcount;
2989 				if_freemulti(ifma);
2990 				error = ENOMEM;
2991 				goto free_llsa_out;
2992 			}
2993 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
2994 			    ifma_link);
2995 		} else
2996 			ll_ifma->ifma_refcount++;
2997 		ifma->ifma_llifma = ll_ifma;
2998 	}
2999 
3000 	/*
3001 	 * We now have a new multicast address, ifma, and possibly a new or
3002 	 * referenced link layer address.  Add the primary address to the
3003 	 * ifnet address list.
3004 	 */
3005 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3006 
3007 	if (retifma != NULL)
3008 		*retifma = ifma;
3009 
3010 	/*
3011 	 * Must generate the message while holding the lock so that 'ifma'
3012 	 * pointer is still valid.
3013 	 */
3014 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3015 	IF_ADDR_UNLOCK(ifp);
3016 
3017 	/*
3018 	 * We are certain we have added something, so call down to the
3019 	 * interface to let them know about it.
3020 	 */
3021 	if (ifp->if_ioctl != NULL) {
3022 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3023 	}
3024 
3025 	if (llsa != NULL)
3026 		free(llsa, M_IFMADDR);
3027 
3028 	return (0);
3029 
3030 free_llsa_out:
3031 	if (llsa != NULL)
3032 		free(llsa, M_IFMADDR);
3033 
3034 unlock_out:
3035 	IF_ADDR_UNLOCK(ifp);
3036 	return (error);
3037 }
3038 
3039 /*
3040  * Delete a multicast group membership by network-layer group address.
3041  *
3042  * Returns ENOENT if the entry could not be found. If ifp no longer
3043  * exists, results are undefined. This entry point should only be used
3044  * from subsystems which do appropriate locking to hold ifp for the
3045  * duration of the call.
3046  * Network-layer protocol domains must use if_delmulti_ifma().
3047  */
3048 int
3049 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3050 {
3051 	struct ifmultiaddr *ifma;
3052 	int lastref;
3053 #ifdef INVARIANTS
3054 	struct ifnet *oifp;
3055 
3056 	IFNET_RLOCK_NOSLEEP();
3057 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3058 		if (ifp == oifp)
3059 			break;
3060 	if (ifp != oifp)
3061 		ifp = NULL;
3062 	IFNET_RUNLOCK_NOSLEEP();
3063 
3064 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3065 #endif
3066 	if (ifp == NULL)
3067 		return (ENOENT);
3068 
3069 	IF_ADDR_LOCK(ifp);
3070 	lastref = 0;
3071 	ifma = if_findmulti(ifp, sa);
3072 	if (ifma != NULL)
3073 		lastref = if_delmulti_locked(ifp, ifma, 0);
3074 	IF_ADDR_UNLOCK(ifp);
3075 
3076 	if (ifma == NULL)
3077 		return (ENOENT);
3078 
3079 	if (lastref && ifp->if_ioctl != NULL) {
3080 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3081 	}
3082 
3083 	return (0);
3084 }
3085 
3086 /*
3087  * Delete all multicast group membership for an interface.
3088  * Should be used to quickly flush all multicast filters.
3089  */
3090 void
3091 if_delallmulti(struct ifnet *ifp)
3092 {
3093 	struct ifmultiaddr *ifma;
3094 	struct ifmultiaddr *next;
3095 
3096 	IF_ADDR_LOCK(ifp);
3097 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3098 		if_delmulti_locked(ifp, ifma, 0);
3099 	IF_ADDR_UNLOCK(ifp);
3100 }
3101 
3102 /*
3103  * Delete a multicast group membership by group membership pointer.
3104  * Network-layer protocol domains must use this routine.
3105  *
3106  * It is safe to call this routine if the ifp disappeared.
3107  */
3108 void
3109 if_delmulti_ifma(struct ifmultiaddr *ifma)
3110 {
3111 	struct ifnet *ifp;
3112 	int lastref;
3113 
3114 	ifp = ifma->ifma_ifp;
3115 #ifdef DIAGNOSTIC
3116 	if (ifp == NULL) {
3117 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3118 	} else {
3119 		struct ifnet *oifp;
3120 
3121 		IFNET_RLOCK_NOSLEEP();
3122 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3123 			if (ifp == oifp)
3124 				break;
3125 		if (ifp != oifp) {
3126 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3127 			ifp = NULL;
3128 		}
3129 		IFNET_RUNLOCK_NOSLEEP();
3130 	}
3131 #endif
3132 	/*
3133 	 * If and only if the ifnet instance exists: Acquire the address lock.
3134 	 */
3135 	if (ifp != NULL)
3136 		IF_ADDR_LOCK(ifp);
3137 
3138 	lastref = if_delmulti_locked(ifp, ifma, 0);
3139 
3140 	if (ifp != NULL) {
3141 		/*
3142 		 * If and only if the ifnet instance exists:
3143 		 *  Release the address lock.
3144 		 *  If the group was left: update the hardware hash filter.
3145 		 */
3146 		IF_ADDR_UNLOCK(ifp);
3147 		if (lastref && ifp->if_ioctl != NULL) {
3148 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3149 		}
3150 	}
3151 }
3152 
3153 /*
3154  * Perform deletion of network-layer and/or link-layer multicast address.
3155  *
3156  * Return 0 if the reference count was decremented.
3157  * Return 1 if the final reference was released, indicating that the
3158  * hardware hash filter should be reprogrammed.
3159  */
3160 static int
3161 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3162 {
3163 	struct ifmultiaddr *ll_ifma;
3164 
3165 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3166 		KASSERT(ifma->ifma_ifp == ifp,
3167 		    ("%s: inconsistent ifp %p", __func__, ifp));
3168 		IF_ADDR_LOCK_ASSERT(ifp);
3169 	}
3170 
3171 	ifp = ifma->ifma_ifp;
3172 
3173 	/*
3174 	 * If the ifnet is detaching, null out references to ifnet,
3175 	 * so that upper protocol layers will notice, and not attempt
3176 	 * to obtain locks for an ifnet which no longer exists. The
3177 	 * routing socket announcement must happen before the ifnet
3178 	 * instance is detached from the system.
3179 	 */
3180 	if (detaching) {
3181 #ifdef DIAGNOSTIC
3182 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3183 #endif
3184 		/*
3185 		 * ifp may already be nulled out if we are being reentered
3186 		 * to delete the ll_ifma.
3187 		 */
3188 		if (ifp != NULL) {
3189 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3190 			ifma->ifma_ifp = NULL;
3191 		}
3192 	}
3193 
3194 	if (--ifma->ifma_refcount > 0)
3195 		return 0;
3196 
3197 	/*
3198 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3199 	 * have been associated with it. Release it first if so.
3200 	 */
3201 	ll_ifma = ifma->ifma_llifma;
3202 	if (ll_ifma != NULL) {
3203 		KASSERT(ifma->ifma_lladdr != NULL,
3204 		    ("%s: llifma w/o lladdr", __func__));
3205 		if (detaching)
3206 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3207 		if (--ll_ifma->ifma_refcount == 0) {
3208 			if (ifp != NULL) {
3209 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3210 				    ifma_link);
3211 			}
3212 			if_freemulti(ll_ifma);
3213 		}
3214 	}
3215 
3216 	if (ifp != NULL)
3217 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3218 
3219 	if_freemulti(ifma);
3220 
3221 	/*
3222 	 * The last reference to this instance of struct ifmultiaddr
3223 	 * was released; the hardware should be notified of this change.
3224 	 */
3225 	return 1;
3226 }
3227 
3228 /*
3229  * Set the link layer address on an interface.
3230  *
3231  * At this time we only support certain types of interfaces,
3232  * and we don't allow the length of the address to change.
3233  */
3234 int
3235 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3236 {
3237 	struct sockaddr_dl *sdl;
3238 	struct ifaddr *ifa;
3239 	struct ifreq ifr;
3240 
3241 	IF_ADDR_LOCK(ifp);
3242 	ifa = ifp->if_addr;
3243 	if (ifa == NULL) {
3244 		IF_ADDR_UNLOCK(ifp);
3245 		return (EINVAL);
3246 	}
3247 	ifa_ref(ifa);
3248 	IF_ADDR_UNLOCK(ifp);
3249 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3250 	if (sdl == NULL) {
3251 		ifa_free(ifa);
3252 		return (EINVAL);
3253 	}
3254 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3255 		ifa_free(ifa);
3256 		return (EINVAL);
3257 	}
3258 	switch (ifp->if_type) {
3259 	case IFT_ETHER:
3260 	case IFT_FDDI:
3261 	case IFT_XETHER:
3262 	case IFT_ISO88025:
3263 	case IFT_L2VLAN:
3264 	case IFT_BRIDGE:
3265 	case IFT_ARCNET:
3266 	case IFT_IEEE8023ADLAG:
3267 	case IFT_IEEE80211:
3268 		bcopy(lladdr, LLADDR(sdl), len);
3269 		ifa_free(ifa);
3270 		break;
3271 	default:
3272 		ifa_free(ifa);
3273 		return (ENODEV);
3274 	}
3275 
3276 	/*
3277 	 * If the interface is already up, we need
3278 	 * to re-init it in order to reprogram its
3279 	 * address filter.
3280 	 */
3281 	if ((ifp->if_flags & IFF_UP) != 0) {
3282 		if (ifp->if_ioctl) {
3283 			ifp->if_flags &= ~IFF_UP;
3284 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3285 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3286 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3287 			ifp->if_flags |= IFF_UP;
3288 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3289 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3290 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3291 		}
3292 #ifdef INET
3293 		/*
3294 		 * Also send gratuitous ARPs to notify other nodes about
3295 		 * the address change.
3296 		 */
3297 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3298 			if (ifa->ifa_addr->sa_family == AF_INET)
3299 				arp_ifinit(ifp, ifa);
3300 		}
3301 #endif
3302 	}
3303 	return (0);
3304 }
3305 
3306 /*
3307  * The name argument must be a pointer to storage which will last as
3308  * long as the interface does.  For physical devices, the result of
3309  * device_get_name(dev) is a good choice and for pseudo-devices a
3310  * static string works well.
3311  */
3312 void
3313 if_initname(struct ifnet *ifp, const char *name, int unit)
3314 {
3315 	ifp->if_dname = name;
3316 	ifp->if_dunit = unit;
3317 	if (unit != IF_DUNIT_NONE)
3318 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3319 	else
3320 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3321 }
3322 
3323 int
3324 if_printf(struct ifnet *ifp, const char * fmt, ...)
3325 {
3326 	va_list ap;
3327 	int retval;
3328 
3329 	retval = printf("%s: ", ifp->if_xname);
3330 	va_start(ap, fmt);
3331 	retval += vprintf(fmt, ap);
3332 	va_end(ap);
3333 	return (retval);
3334 }
3335 
3336 void
3337 if_start(struct ifnet *ifp)
3338 {
3339 
3340 	(*(ifp)->if_start)(ifp);
3341 }
3342 
3343 /*
3344  * Backwards compatibility interface for drivers
3345  * that have not implemented it
3346  */
3347 static int
3348 if_transmit(struct ifnet *ifp, struct mbuf *m)
3349 {
3350 	int error;
3351 
3352 	IFQ_HANDOFF(ifp, m, error);
3353 	return (error);
3354 }
3355 
3356 int
3357 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3358 {
3359 	int active = 0;
3360 
3361 	IF_LOCK(ifq);
3362 	if (_IF_QFULL(ifq)) {
3363 		_IF_DROP(ifq);
3364 		IF_UNLOCK(ifq);
3365 		m_freem(m);
3366 		return (0);
3367 	}
3368 	if (ifp != NULL) {
3369 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3370 		if (m->m_flags & (M_BCAST|M_MCAST))
3371 			ifp->if_omcasts++;
3372 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3373 	}
3374 	_IF_ENQUEUE(ifq, m);
3375 	IF_UNLOCK(ifq);
3376 	if (ifp != NULL && !active)
3377 		(*(ifp)->if_start)(ifp);
3378 	return (1);
3379 }
3380 
3381 void
3382 if_register_com_alloc(u_char type,
3383     if_com_alloc_t *a, if_com_free_t *f)
3384 {
3385 
3386 	KASSERT(if_com_alloc[type] == NULL,
3387 	    ("if_register_com_alloc: %d already registered", type));
3388 	KASSERT(if_com_free[type] == NULL,
3389 	    ("if_register_com_alloc: %d free already registered", type));
3390 
3391 	if_com_alloc[type] = a;
3392 	if_com_free[type] = f;
3393 }
3394 
3395 void
3396 if_deregister_com_alloc(u_char type)
3397 {
3398 
3399 	KASSERT(if_com_alloc[type] != NULL,
3400 	    ("if_deregister_com_alloc: %d not registered", type));
3401 	KASSERT(if_com_free[type] != NULL,
3402 	    ("if_deregister_com_alloc: %d free not registered", type));
3403 	if_com_alloc[type] = NULL;
3404 	if_com_free[type] = NULL;
3405 }
3406