xref: /freebsd/sys/net/if.c (revision cc3f4b99653c34ae64f8a1fddea370abefef680e)
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 <sys/priv.h>
62 
63 #include <machine/stdarg.h>
64 #include <vm/uma.h>
65 
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_clone.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/vnet.h>
75 
76 #if defined(INET) || defined(INET6)
77 #include <net/ethernet.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #include <netinet/ip_carp.h>
82 #ifdef INET
83 #include <netinet/if_ether.h>
84 #endif /* INET */
85 #ifdef INET6
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
88 #endif /* INET6 */
89 #endif /* INET || INET6 */
90 
91 #include <security/mac/mac_framework.h>
92 
93 #ifdef COMPAT_FREEBSD32
94 #include <sys/mount.h>
95 #include <compat/freebsd32/freebsd32.h>
96 #endif
97 
98 struct ifindex_entry {
99 	struct  ifnet *ife_ifnet;
100 };
101 
102 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
103 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
104 
105 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
106 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
107     &ifqmaxlen, 0, "max send queue size");
108 
109 /* Log link state change events */
110 static int log_link_state_change = 1;
111 
112 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
113 	&log_link_state_change, 0,
114 	"log interface link state change events");
115 
116 /* Interface description */
117 static unsigned int ifdescr_maxlen = 1024;
118 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
119 	&ifdescr_maxlen, 0,
120 	"administrative maximum length for interface description");
121 
122 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
123 
124 /* global sx for non-critical path ifdescr */
125 static struct sx ifdescr_sx;
126 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
127 
128 void	(*bridge_linkstate_p)(struct ifnet *ifp);
129 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
130 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
131 /* These are external hooks for CARP. */
132 void	(*carp_linkstate_p)(struct ifnet *ifp);
133 void	(*carp_demote_adj_p)(int, char *);
134 int	(*carp_master_p)(struct ifaddr *);
135 #if defined(INET) || defined(INET6)
136 int	(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
137 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
138     const struct sockaddr *sa);
139 int	(*carp_ioctl_p)(struct ifreq *, u_long, struct thread *);
140 int	(*carp_attach_p)(struct ifaddr *, int);
141 void	(*carp_detach_p)(struct ifaddr *);
142 #endif
143 #ifdef INET
144 int	(*carp_iamatch_p)(struct ifaddr *, uint8_t **);
145 #endif
146 #ifdef INET6
147 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
148 caddr_t	(*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
149     const struct in6_addr *taddr);
150 #endif
151 
152 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
153 
154 /*
155  * XXX: Style; these should be sorted alphabetically, and unprototyped
156  * static functions should be prototyped. Currently they are sorted by
157  * declaration order.
158  */
159 static void	if_attachdomain(void *);
160 static void	if_attachdomain1(struct ifnet *);
161 static int	ifconf(u_long, caddr_t);
162 static void	if_freemulti(struct ifmultiaddr *);
163 static void	if_init(void *);
164 static void	if_grow(void);
165 static void	if_route(struct ifnet *, int flag, int fam);
166 static int	if_setflag(struct ifnet *, int, int, int *, int);
167 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
168 static void	if_unroute(struct ifnet *, int flag, int fam);
169 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
170 static int	if_rtdel(struct radix_node *, void *);
171 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
172 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
173 static void	do_link_state_change(void *, int);
174 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
175 static int	if_getgroupmembers(struct ifgroupreq *);
176 static void	if_delgroups(struct ifnet *);
177 static void	if_attach_internal(struct ifnet *, int);
178 static void	if_detach_internal(struct ifnet *, int);
179 
180 #ifdef INET6
181 /*
182  * XXX: declare here to avoid to include many inet6 related files..
183  * should be more generalized?
184  */
185 extern void	nd6_setmtu(struct ifnet *);
186 #endif
187 
188 VNET_DEFINE(int, if_index);
189 int	ifqmaxlen = IFQ_MAXLEN;
190 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
191 VNET_DEFINE(struct ifgrouphead, ifg_head);
192 
193 static VNET_DEFINE(int, if_indexlim) = 8;
194 
195 /* Table of ifnet by index. */
196 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
197 
198 #define	V_if_indexlim		VNET(if_indexlim)
199 #define	V_ifindex_table		VNET(ifindex_table)
200 
201 /*
202  * The global network interface list (V_ifnet) and related state (such as
203  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
204  * an rwlock.  Either may be acquired shared to stablize the list, but both
205  * must be acquired writable to modify the list.  This model allows us to
206  * both stablize the interface list during interrupt thread processing, but
207  * also to stablize it over long-running ioctls, without introducing priority
208  * inversions and deadlocks.
209  */
210 struct rwlock ifnet_rwlock;
211 struct sx ifnet_sxlock;
212 
213 /*
214  * The allocation of network interfaces is a rather non-atomic affair; we
215  * need to select an index before we are ready to expose the interface for
216  * use, so will use this pointer value to indicate reservation.
217  */
218 #define	IFNET_HOLD	(void *)(uintptr_t)(-1)
219 
220 static	if_com_alloc_t *if_com_alloc[256];
221 static	if_com_free_t *if_com_free[256];
222 
223 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
224 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
225 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
226 
227 struct ifnet *
228 ifnet_byindex_locked(u_short idx)
229 {
230 
231 	if (idx > V_if_index)
232 		return (NULL);
233 	if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
234 		return (NULL);
235 	return (V_ifindex_table[idx].ife_ifnet);
236 }
237 
238 struct ifnet *
239 ifnet_byindex(u_short idx)
240 {
241 	struct ifnet *ifp;
242 
243 	IFNET_RLOCK_NOSLEEP();
244 	ifp = ifnet_byindex_locked(idx);
245 	IFNET_RUNLOCK_NOSLEEP();
246 	return (ifp);
247 }
248 
249 struct ifnet *
250 ifnet_byindex_ref(u_short idx)
251 {
252 	struct ifnet *ifp;
253 
254 	IFNET_RLOCK_NOSLEEP();
255 	ifp = ifnet_byindex_locked(idx);
256 	if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
257 		IFNET_RUNLOCK_NOSLEEP();
258 		return (NULL);
259 	}
260 	if_ref(ifp);
261 	IFNET_RUNLOCK_NOSLEEP();
262 	return (ifp);
263 }
264 
265 /*
266  * Allocate an ifindex array entry; return 0 on success or an error on
267  * failure.
268  */
269 static int
270 ifindex_alloc_locked(u_short *idxp)
271 {
272 	u_short idx;
273 
274 	IFNET_WLOCK_ASSERT();
275 
276 retry:
277 	/*
278 	 * Try to find an empty slot below V_if_index.  If we fail, take the
279 	 * next slot.
280 	 */
281 	for (idx = 1; idx <= V_if_index; idx++) {
282 		if (V_ifindex_table[idx].ife_ifnet == NULL)
283 			break;
284 	}
285 
286 	/* Catch if_index overflow. */
287 	if (idx >= V_if_indexlim) {
288 		if_grow();
289 		goto retry;
290 	}
291 	if (idx > V_if_index)
292 		V_if_index = idx;
293 	*idxp = idx;
294 	return (0);
295 }
296 
297 static void
298 ifindex_free_locked(u_short idx)
299 {
300 
301 	IFNET_WLOCK_ASSERT();
302 
303 	V_ifindex_table[idx].ife_ifnet = NULL;
304 	while (V_if_index > 0 &&
305 	    V_ifindex_table[V_if_index].ife_ifnet == NULL)
306 		V_if_index--;
307 }
308 
309 static void
310 ifindex_free(u_short idx)
311 {
312 
313 	IFNET_WLOCK();
314 	ifindex_free_locked(idx);
315 	IFNET_WUNLOCK();
316 }
317 
318 static void
319 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
320 {
321 
322 	IFNET_WLOCK_ASSERT();
323 
324 	V_ifindex_table[idx].ife_ifnet = ifp;
325 }
326 
327 static void
328 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
329 {
330 
331 	IFNET_WLOCK();
332 	ifnet_setbyindex_locked(idx, ifp);
333 	IFNET_WUNLOCK();
334 }
335 
336 struct ifaddr *
337 ifaddr_byindex(u_short idx)
338 {
339 	struct ifaddr *ifa;
340 
341 	IFNET_RLOCK_NOSLEEP();
342 	ifa = ifnet_byindex_locked(idx)->if_addr;
343 	if (ifa != NULL)
344 		ifa_ref(ifa);
345 	IFNET_RUNLOCK_NOSLEEP();
346 	return (ifa);
347 }
348 
349 /*
350  * Network interface utility routines.
351  *
352  * Routines with ifa_ifwith* names take sockaddr *'s as
353  * parameters.
354  */
355 
356 static void
357 vnet_if_init(const void *unused __unused)
358 {
359 
360 	TAILQ_INIT(&V_ifnet);
361 	TAILQ_INIT(&V_ifg_head);
362 	IFNET_WLOCK();
363 	if_grow();				/* create initial table */
364 	IFNET_WUNLOCK();
365 	vnet_if_clone_init();
366 }
367 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
368     NULL);
369 
370 /* ARGSUSED*/
371 static void
372 if_init(void *dummy __unused)
373 {
374 
375 	IFNET_LOCK_INIT();
376 	if_clone_init();
377 }
378 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
379 
380 
381 #ifdef VIMAGE
382 static void
383 vnet_if_uninit(const void *unused __unused)
384 {
385 
386 	VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
387 	    "not empty", __func__, __LINE__, &V_ifnet));
388 	VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
389 	    "not empty", __func__, __LINE__, &V_ifg_head));
390 
391 	free((caddr_t)V_ifindex_table, M_IFNET);
392 }
393 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
394     vnet_if_uninit, NULL);
395 #endif
396 
397 static void
398 if_grow(void)
399 {
400 	int oldlim;
401 	u_int n;
402 	struct ifindex_entry *e;
403 
404 	IFNET_WLOCK_ASSERT();
405 	oldlim = V_if_indexlim;
406 	IFNET_WUNLOCK();
407 	n = (oldlim << 1) * sizeof(*e);
408 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
409 	IFNET_WLOCK();
410 	if (V_if_indexlim != oldlim) {
411 		free(e, M_IFNET);
412 		return;
413 	}
414 	if (V_ifindex_table != NULL) {
415 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
416 		free((caddr_t)V_ifindex_table, M_IFNET);
417 	}
418 	V_if_indexlim <<= 1;
419 	V_ifindex_table = e;
420 }
421 
422 /*
423  * Allocate a struct ifnet and an index for an interface.  A layer 2
424  * common structure will also be allocated if an allocation routine is
425  * registered for the passed type.
426  */
427 struct ifnet *
428 if_alloc(u_char type)
429 {
430 	struct ifnet *ifp;
431 	u_short idx;
432 
433 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
434 	IFNET_WLOCK();
435 	if (ifindex_alloc_locked(&idx) != 0) {
436 		IFNET_WUNLOCK();
437 		free(ifp, M_IFNET);
438 		return (NULL);
439 	}
440 	ifnet_setbyindex_locked(idx, IFNET_HOLD);
441 	IFNET_WUNLOCK();
442 	ifp->if_index = idx;
443 	ifp->if_type = type;
444 	ifp->if_alloctype = type;
445 	if (if_com_alloc[type] != NULL) {
446 		ifp->if_l2com = if_com_alloc[type](type, ifp);
447 		if (ifp->if_l2com == NULL) {
448 			free(ifp, M_IFNET);
449 			ifindex_free(idx);
450 			return (NULL);
451 		}
452 	}
453 
454 	IF_ADDR_LOCK_INIT(ifp);
455 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
456 	ifp->if_afdata_initialized = 0;
457 	IF_AFDATA_LOCK_INIT(ifp);
458 	TAILQ_INIT(&ifp->if_addrhead);
459 	TAILQ_INIT(&ifp->if_multiaddrs);
460 	TAILQ_INIT(&ifp->if_groups);
461 #ifdef MAC
462 	mac_ifnet_init(ifp);
463 #endif
464 	ifq_init(&ifp->if_snd, ifp);
465 
466 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
467 	ifnet_setbyindex(ifp->if_index, ifp);
468 	return (ifp);
469 }
470 
471 /*
472  * Do the actual work of freeing a struct ifnet, and layer 2 common
473  * structure.  This call is made when the last reference to an
474  * interface is released.
475  */
476 static void
477 if_free_internal(struct ifnet *ifp)
478 {
479 
480 	KASSERT((ifp->if_flags & IFF_DYING),
481 	    ("if_free_internal: interface not dying"));
482 
483 	if (if_com_free[ifp->if_alloctype] != NULL)
484 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
485 		    ifp->if_alloctype);
486 
487 #ifdef MAC
488 	mac_ifnet_destroy(ifp);
489 #endif /* MAC */
490 	if (ifp->if_description != NULL)
491 		free(ifp->if_description, M_IFDESCR);
492 	IF_AFDATA_DESTROY(ifp);
493 	IF_ADDR_LOCK_DESTROY(ifp);
494 	ifq_delete(&ifp->if_snd);
495 	free(ifp, M_IFNET);
496 }
497 
498 /*
499  * Deregister an interface and free the associated storage.
500  */
501 void
502 if_free(struct ifnet *ifp)
503 {
504 
505 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
506 
507 	CURVNET_SET_QUIET(ifp->if_vnet);
508 	IFNET_WLOCK();
509 	KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
510 	    ("%s: freeing unallocated ifnet", ifp->if_xname));
511 
512 	ifindex_free_locked(ifp->if_index);
513 	IFNET_WUNLOCK();
514 
515 	if (refcount_release(&ifp->if_refcount))
516 		if_free_internal(ifp);
517 	CURVNET_RESTORE();
518 }
519 
520 /*
521  * Interfaces to keep an ifnet type-stable despite the possibility of the
522  * driver calling if_free().  If there are additional references, we defer
523  * freeing the underlying data structure.
524  */
525 void
526 if_ref(struct ifnet *ifp)
527 {
528 
529 	/* We don't assert the ifnet list lock here, but arguably should. */
530 	refcount_acquire(&ifp->if_refcount);
531 }
532 
533 void
534 if_rele(struct ifnet *ifp)
535 {
536 
537 	if (!refcount_release(&ifp->if_refcount))
538 		return;
539 	if_free_internal(ifp);
540 }
541 
542 void
543 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
544 {
545 
546 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
547 
548 	if (ifq->ifq_maxlen == 0)
549 		ifq->ifq_maxlen = ifqmaxlen;
550 
551 	ifq->altq_type = 0;
552 	ifq->altq_disc = NULL;
553 	ifq->altq_flags &= ALTQF_CANTCHANGE;
554 	ifq->altq_tbr  = NULL;
555 	ifq->altq_ifp  = ifp;
556 }
557 
558 void
559 ifq_delete(struct ifaltq *ifq)
560 {
561 	mtx_destroy(&ifq->ifq_mtx);
562 }
563 
564 /*
565  * Perform generic interface initalization tasks and attach the interface
566  * to the list of "active" interfaces.  If vmove flag is set on entry
567  * to if_attach_internal(), perform only a limited subset of initialization
568  * tasks, given that we are moving from one vnet to another an ifnet which
569  * has already been fully initialized.
570  *
571  * XXX:
572  *  - The decision to return void and thus require this function to
573  *    succeed is questionable.
574  *  - We should probably do more sanity checking.  For instance we don't
575  *    do anything to insure if_xname is unique or non-empty.
576  */
577 void
578 if_attach(struct ifnet *ifp)
579 {
580 
581 	if_attach_internal(ifp, 0);
582 }
583 
584 static void
585 if_attach_internal(struct ifnet *ifp, int vmove)
586 {
587 	unsigned socksize, ifasize;
588 	int namelen, masklen;
589 	struct sockaddr_dl *sdl;
590 	struct ifaddr *ifa;
591 
592 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
593 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
594 		    ifp->if_xname);
595 
596 #ifdef VIMAGE
597 	ifp->if_vnet = curvnet;
598 	if (ifp->if_home_vnet == NULL)
599 		ifp->if_home_vnet = curvnet;
600 #endif
601 
602 	if_addgroup(ifp, IFG_ALL);
603 
604 	getmicrotime(&ifp->if_lastchange);
605 	ifp->if_data.ifi_epoch = time_uptime;
606 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
607 
608 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
609 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
610 	    ("transmit and qflush must both either be set or both be NULL"));
611 	if (ifp->if_transmit == NULL) {
612 		ifp->if_transmit = if_transmit;
613 		ifp->if_qflush = if_qflush;
614 	}
615 
616 	if (!vmove) {
617 #ifdef MAC
618 		mac_ifnet_create(ifp);
619 #endif
620 
621 		/*
622 		 * Create a Link Level name for this device.
623 		 */
624 		namelen = strlen(ifp->if_xname);
625 		/*
626 		 * Always save enough space for any possiable name so we
627 		 * can do a rename in place later.
628 		 */
629 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
630 		socksize = masklen + ifp->if_addrlen;
631 		if (socksize < sizeof(*sdl))
632 			socksize = sizeof(*sdl);
633 		socksize = roundup2(socksize, sizeof(long));
634 		ifasize = sizeof(*ifa) + 2 * socksize;
635 		ifa = ifa_alloc(ifasize, M_WAITOK);
636 		sdl = (struct sockaddr_dl *)(ifa + 1);
637 		sdl->sdl_len = socksize;
638 		sdl->sdl_family = AF_LINK;
639 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
640 		sdl->sdl_nlen = namelen;
641 		sdl->sdl_index = ifp->if_index;
642 		sdl->sdl_type = ifp->if_type;
643 		ifp->if_addr = ifa;
644 		ifa->ifa_ifp = ifp;
645 		ifa->ifa_rtrequest = link_rtrequest;
646 		ifa->ifa_addr = (struct sockaddr *)sdl;
647 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
648 		ifa->ifa_netmask = (struct sockaddr *)sdl;
649 		sdl->sdl_len = masklen;
650 		while (namelen != 0)
651 			sdl->sdl_data[--namelen] = 0xff;
652 		TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
653 		/* Reliably crash if used uninitialized. */
654 		ifp->if_broadcastaddr = NULL;
655 
656 #if defined(INET) || defined(INET6)
657 		/* Initialize to max value. */
658 		if (ifp->if_hw_tsomax == 0)
659 			ifp->if_hw_tsomax = min(IP_MAXPACKET, 32 * MCLBYTES -
660 			    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
661 		KASSERT(ifp->if_hw_tsomax <= IP_MAXPACKET &&
662 		    ifp->if_hw_tsomax >= IP_MAXPACKET / 8,
663 		    ("%s: tsomax outside of range", __func__));
664 #endif
665 	}
666 #ifdef VIMAGE
667 	else {
668 		/*
669 		 * Update the interface index in the link layer address
670 		 * of the interface.
671 		 */
672 		for (ifa = ifp->if_addr; ifa != NULL;
673 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
674 			if (ifa->ifa_addr->sa_family == AF_LINK) {
675 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
676 				sdl->sdl_index = ifp->if_index;
677 			}
678 		}
679 	}
680 #endif
681 
682 	IFNET_WLOCK();
683 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
684 #ifdef VIMAGE
685 	curvnet->vnet_ifcnt++;
686 #endif
687 	IFNET_WUNLOCK();
688 
689 	if (domain_init_status >= 2)
690 		if_attachdomain1(ifp);
691 
692 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
693 	if (IS_DEFAULT_VNET(curvnet))
694 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
695 
696 	/* Announce the interface. */
697 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
698 }
699 
700 static void
701 if_attachdomain(void *dummy)
702 {
703 	struct ifnet *ifp;
704 
705 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
706 		if_attachdomain1(ifp);
707 }
708 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
709     if_attachdomain, NULL);
710 
711 static void
712 if_attachdomain1(struct ifnet *ifp)
713 {
714 	struct domain *dp;
715 
716 	/*
717 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
718 	 * cannot lock ifp->if_afdata initialization, entirely.
719 	 */
720 	if (IF_AFDATA_TRYLOCK(ifp) == 0)
721 		return;
722 	if (ifp->if_afdata_initialized >= domain_init_status) {
723 		IF_AFDATA_UNLOCK(ifp);
724 		log(LOG_WARNING, "%s called more than once on %s\n",
725 		    __func__, ifp->if_xname);
726 		return;
727 	}
728 	ifp->if_afdata_initialized = domain_init_status;
729 	IF_AFDATA_UNLOCK(ifp);
730 
731 	/* address family dependent data region */
732 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
733 	for (dp = domains; dp; dp = dp->dom_next) {
734 		if (dp->dom_ifattach)
735 			ifp->if_afdata[dp->dom_family] =
736 			    (*dp->dom_ifattach)(ifp);
737 	}
738 }
739 
740 /*
741  * Remove any unicast or broadcast network addresses from an interface.
742  */
743 void
744 if_purgeaddrs(struct ifnet *ifp)
745 {
746 	struct ifaddr *ifa, *next;
747 
748 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
749 		if (ifa->ifa_addr->sa_family == AF_LINK)
750 			continue;
751 #ifdef INET
752 		/* XXX: Ugly!! ad hoc just for INET */
753 		if (ifa->ifa_addr->sa_family == AF_INET) {
754 			struct ifaliasreq ifr;
755 
756 			bzero(&ifr, sizeof(ifr));
757 			ifr.ifra_addr = *ifa->ifa_addr;
758 			if (ifa->ifa_dstaddr)
759 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
760 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
761 			    NULL) == 0)
762 				continue;
763 		}
764 #endif /* INET */
765 #ifdef INET6
766 		if (ifa->ifa_addr->sa_family == AF_INET6) {
767 			in6_purgeaddr(ifa);
768 			/* ifp_addrhead is already updated */
769 			continue;
770 		}
771 #endif /* INET6 */
772 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
773 		ifa_free(ifa);
774 	}
775 }
776 
777 /*
778  * Remove any multicast network addresses from an interface when an ifnet
779  * is going away.
780  */
781 static void
782 if_purgemaddrs(struct ifnet *ifp)
783 {
784 	struct ifmultiaddr *ifma;
785 	struct ifmultiaddr *next;
786 
787 	IF_ADDR_WLOCK(ifp);
788 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
789 		if_delmulti_locked(ifp, ifma, 1);
790 	IF_ADDR_WUNLOCK(ifp);
791 }
792 
793 /*
794  * Detach an interface, removing it from the list of "active" interfaces.
795  * If vmove flag is set on entry to if_detach_internal(), perform only a
796  * limited subset of cleanup tasks, given that we are moving an ifnet from
797  * one vnet to another, where it must be fully operational.
798  *
799  * XXXRW: There are some significant questions about event ordering, and
800  * how to prevent things from starting to use the interface during detach.
801  */
802 void
803 if_detach(struct ifnet *ifp)
804 {
805 
806 	CURVNET_SET_QUIET(ifp->if_vnet);
807 	if_detach_internal(ifp, 0);
808 	CURVNET_RESTORE();
809 }
810 
811 static void
812 if_detach_internal(struct ifnet *ifp, int vmove)
813 {
814 	struct ifaddr *ifa;
815 	struct radix_node_head	*rnh;
816 	int i, j;
817 	struct domain *dp;
818  	struct ifnet *iter;
819  	int found = 0;
820 
821 	IFNET_WLOCK();
822 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
823 		if (iter == ifp) {
824 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
825 			found = 1;
826 			break;
827 		}
828 #ifdef VIMAGE
829 	if (found)
830 		curvnet->vnet_ifcnt--;
831 #endif
832 	IFNET_WUNLOCK();
833 	if (!found) {
834 		if (vmove)
835 			panic("%s: ifp=%p not on the ifnet tailq %p",
836 			    __func__, ifp, &V_ifnet);
837 		else
838 			return; /* XXX this should panic as well? */
839 	}
840 
841 	/*
842 	 * Remove/wait for pending events.
843 	 */
844 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
845 
846 	/*
847 	 * Remove routes and flush queues.
848 	 */
849 	if_down(ifp);
850 #ifdef ALTQ
851 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
852 		altq_disable(&ifp->if_snd);
853 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
854 		altq_detach(&ifp->if_snd);
855 #endif
856 
857 	if_purgeaddrs(ifp);
858 
859 #ifdef INET
860 	in_ifdetach(ifp);
861 #endif
862 
863 #ifdef INET6
864 	/*
865 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
866 	 * before removing routing entries below, since IPv6 interface direct
867 	 * routes are expected to be removed by the IPv6-specific kernel API.
868 	 * Otherwise, the kernel will detect some inconsistency and bark it.
869 	 */
870 	in6_ifdetach(ifp);
871 #endif
872 	if_purgemaddrs(ifp);
873 
874 	if (!vmove) {
875 		/*
876 		 * Prevent further calls into the device driver via ifnet.
877 		 */
878 		if_dead(ifp);
879 
880 		/*
881 		 * Remove link ifaddr pointer and maybe decrement if_index.
882 		 * Clean up all addresses.
883 		 */
884 		ifp->if_addr = NULL;
885 
886 		/* We can now free link ifaddr. */
887 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
888 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
889 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
890 			ifa_free(ifa);
891 		}
892 	}
893 
894 	/*
895 	 * Delete all remaining routes using this interface
896 	 * Unfortuneatly the only way to do this is to slog through
897 	 * the entire routing table looking for routes which point
898 	 * to this interface...oh well...
899 	 */
900 	for (i = 1; i <= AF_MAX; i++) {
901 		for (j = 0; j < rt_numfibs; j++) {
902 			rnh = rt_tables_get_rnh(j, i);
903 			if (rnh == NULL)
904 				continue;
905 			RADIX_NODE_HEAD_LOCK(rnh);
906 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
907 			RADIX_NODE_HEAD_UNLOCK(rnh);
908 		}
909 	}
910 
911 	/* Announce that the interface is gone. */
912 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
913 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
914 	if (IS_DEFAULT_VNET(curvnet))
915 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
916 	if_delgroups(ifp);
917 
918 	/*
919 	 * We cannot hold the lock over dom_ifdetach calls as they might
920 	 * sleep, for example trying to drain a callout, thus open up the
921 	 * theoretical race with re-attaching.
922 	 */
923 	IF_AFDATA_LOCK(ifp);
924 	i = ifp->if_afdata_initialized;
925 	ifp->if_afdata_initialized = 0;
926 	IF_AFDATA_UNLOCK(ifp);
927 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
928 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
929 			(*dp->dom_ifdetach)(ifp,
930 			    ifp->if_afdata[dp->dom_family]);
931 	}
932 }
933 
934 #ifdef VIMAGE
935 /*
936  * if_vmove() performs a limited version of if_detach() in current
937  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
938  * An attempt is made to shrink if_index in current vnet, find an
939  * unused if_index in target vnet and calls if_grow() if necessary,
940  * and finally find an unused if_xname for the target vnet.
941  */
942 void
943 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
944 {
945 	u_short idx;
946 
947 	/*
948 	 * Detach from current vnet, but preserve LLADDR info, do not
949 	 * mark as dead etc. so that the ifnet can be reattached later.
950 	 */
951 	if_detach_internal(ifp, 1);
952 
953 	/*
954 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
955 	 * the if_index for that vnet if possible.
956 	 *
957 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
958 	 * or we'd lock on one vnet and unlock on another.
959 	 */
960 	IFNET_WLOCK();
961 	ifindex_free_locked(ifp->if_index);
962 	IFNET_WUNLOCK();
963 
964 	/*
965 	 * Perform interface-specific reassignment tasks, if provided by
966 	 * the driver.
967 	 */
968 	if (ifp->if_reassign != NULL)
969 		ifp->if_reassign(ifp, new_vnet, NULL);
970 
971 	/*
972 	 * Switch to the context of the target vnet.
973 	 */
974 	CURVNET_SET_QUIET(new_vnet);
975 
976 	IFNET_WLOCK();
977 	if (ifindex_alloc_locked(&idx) != 0) {
978 		IFNET_WUNLOCK();
979 		panic("if_index overflow");
980 	}
981 	ifp->if_index = idx;
982 	ifnet_setbyindex_locked(ifp->if_index, ifp);
983 	IFNET_WUNLOCK();
984 
985 	if_attach_internal(ifp, 1);
986 
987 	CURVNET_RESTORE();
988 }
989 
990 /*
991  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
992  */
993 static int
994 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
995 {
996 	struct prison *pr;
997 	struct ifnet *difp;
998 
999 	/* Try to find the prison within our visibility. */
1000 	sx_slock(&allprison_lock);
1001 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1002 	sx_sunlock(&allprison_lock);
1003 	if (pr == NULL)
1004 		return (ENXIO);
1005 	prison_hold_locked(pr);
1006 	mtx_unlock(&pr->pr_mtx);
1007 
1008 	/* Do not try to move the iface from and to the same prison. */
1009 	if (pr->pr_vnet == ifp->if_vnet) {
1010 		prison_free(pr);
1011 		return (EEXIST);
1012 	}
1013 
1014 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1015 	/* XXX Lock interfaces to avoid races. */
1016 	CURVNET_SET_QUIET(pr->pr_vnet);
1017 	difp = ifunit(ifname);
1018 	CURVNET_RESTORE();
1019 	if (difp != NULL) {
1020 		prison_free(pr);
1021 		return (EEXIST);
1022 	}
1023 
1024 	/* Move the interface into the child jail/vnet. */
1025 	if_vmove(ifp, pr->pr_vnet);
1026 
1027 	/* Report the new if_xname back to the userland. */
1028 	sprintf(ifname, "%s", ifp->if_xname);
1029 
1030 	prison_free(pr);
1031 	return (0);
1032 }
1033 
1034 static int
1035 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1036 {
1037 	struct prison *pr;
1038 	struct vnet *vnet_dst;
1039 	struct ifnet *ifp;
1040 
1041 	/* Try to find the prison within our visibility. */
1042 	sx_slock(&allprison_lock);
1043 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1044 	sx_sunlock(&allprison_lock);
1045 	if (pr == NULL)
1046 		return (ENXIO);
1047 	prison_hold_locked(pr);
1048 	mtx_unlock(&pr->pr_mtx);
1049 
1050 	/* Make sure the named iface exists in the source prison/vnet. */
1051 	CURVNET_SET(pr->pr_vnet);
1052 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1053 	if (ifp == NULL) {
1054 		CURVNET_RESTORE();
1055 		prison_free(pr);
1056 		return (ENXIO);
1057 	}
1058 
1059 	/* Do not try to move the iface from and to the same prison. */
1060 	vnet_dst = TD_TO_VNET(td);
1061 	if (vnet_dst == ifp->if_vnet) {
1062 		CURVNET_RESTORE();
1063 		prison_free(pr);
1064 		return (EEXIST);
1065 	}
1066 
1067 	/* Get interface back from child jail/vnet. */
1068 	if_vmove(ifp, vnet_dst);
1069 	CURVNET_RESTORE();
1070 
1071 	/* Report the new if_xname back to the userland. */
1072 	sprintf(ifname, "%s", ifp->if_xname);
1073 
1074 	prison_free(pr);
1075 	return (0);
1076 }
1077 #endif /* VIMAGE */
1078 
1079 /*
1080  * Add a group to an interface
1081  */
1082 int
1083 if_addgroup(struct ifnet *ifp, const char *groupname)
1084 {
1085 	struct ifg_list		*ifgl;
1086 	struct ifg_group	*ifg = NULL;
1087 	struct ifg_member	*ifgm;
1088 	int 			 new = 0;
1089 
1090 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1091 	    groupname[strlen(groupname) - 1] <= '9')
1092 		return (EINVAL);
1093 
1094 	IFNET_WLOCK();
1095 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1096 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1097 			IFNET_WUNLOCK();
1098 			return (EEXIST);
1099 		}
1100 
1101 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1102 	    M_NOWAIT)) == NULL) {
1103 	    	IFNET_WUNLOCK();
1104 		return (ENOMEM);
1105 	}
1106 
1107 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1108 	    M_TEMP, M_NOWAIT)) == NULL) {
1109 		free(ifgl, M_TEMP);
1110 		IFNET_WUNLOCK();
1111 		return (ENOMEM);
1112 	}
1113 
1114 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1115 		if (!strcmp(ifg->ifg_group, groupname))
1116 			break;
1117 
1118 	if (ifg == NULL) {
1119 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1120 		    M_TEMP, M_NOWAIT)) == NULL) {
1121 			free(ifgl, M_TEMP);
1122 			free(ifgm, M_TEMP);
1123 			IFNET_WUNLOCK();
1124 			return (ENOMEM);
1125 		}
1126 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1127 		ifg->ifg_refcnt = 0;
1128 		TAILQ_INIT(&ifg->ifg_members);
1129 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1130 		new = 1;
1131 	}
1132 
1133 	ifg->ifg_refcnt++;
1134 	ifgl->ifgl_group = ifg;
1135 	ifgm->ifgm_ifp = ifp;
1136 
1137 	IF_ADDR_WLOCK(ifp);
1138 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1139 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1140 	IF_ADDR_WUNLOCK(ifp);
1141 
1142 	IFNET_WUNLOCK();
1143 
1144 	if (new)
1145 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1146 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Remove a group from an interface
1153  */
1154 int
1155 if_delgroup(struct ifnet *ifp, const char *groupname)
1156 {
1157 	struct ifg_list		*ifgl;
1158 	struct ifg_member	*ifgm;
1159 
1160 	IFNET_WLOCK();
1161 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1162 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1163 			break;
1164 	if (ifgl == NULL) {
1165 		IFNET_WUNLOCK();
1166 		return (ENOENT);
1167 	}
1168 
1169 	IF_ADDR_WLOCK(ifp);
1170 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1171 	IF_ADDR_WUNLOCK(ifp);
1172 
1173 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1174 		if (ifgm->ifgm_ifp == ifp)
1175 			break;
1176 
1177 	if (ifgm != NULL) {
1178 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1179 		free(ifgm, M_TEMP);
1180 	}
1181 
1182 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1183 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1184 		IFNET_WUNLOCK();
1185 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1186 		free(ifgl->ifgl_group, M_TEMP);
1187 	} else
1188 		IFNET_WUNLOCK();
1189 
1190 	free(ifgl, M_TEMP);
1191 
1192 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1193 
1194 	return (0);
1195 }
1196 
1197 /*
1198  * Remove an interface from all groups
1199  */
1200 static void
1201 if_delgroups(struct ifnet *ifp)
1202 {
1203 	struct ifg_list		*ifgl;
1204 	struct ifg_member	*ifgm;
1205 	char groupname[IFNAMSIZ];
1206 
1207 	IFNET_WLOCK();
1208 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1209 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1210 
1211 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1212 
1213 		IF_ADDR_WLOCK(ifp);
1214 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1215 		IF_ADDR_WUNLOCK(ifp);
1216 
1217 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1218 			if (ifgm->ifgm_ifp == ifp)
1219 				break;
1220 
1221 		if (ifgm != NULL) {
1222 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1223 			    ifgm_next);
1224 			free(ifgm, M_TEMP);
1225 		}
1226 
1227 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1228 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1229 			IFNET_WUNLOCK();
1230 			EVENTHANDLER_INVOKE(group_detach_event,
1231 			    ifgl->ifgl_group);
1232 			free(ifgl->ifgl_group, M_TEMP);
1233 		} else
1234 			IFNET_WUNLOCK();
1235 
1236 		free(ifgl, M_TEMP);
1237 
1238 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1239 
1240 		IFNET_WLOCK();
1241 	}
1242 	IFNET_WUNLOCK();
1243 }
1244 
1245 /*
1246  * Stores all groups from an interface in memory pointed
1247  * to by data
1248  */
1249 static int
1250 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1251 {
1252 	int			 len, error;
1253 	struct ifg_list		*ifgl;
1254 	struct ifg_req		 ifgrq, *ifgp;
1255 	struct ifgroupreq	*ifgr = data;
1256 
1257 	if (ifgr->ifgr_len == 0) {
1258 		IF_ADDR_RLOCK(ifp);
1259 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1260 			ifgr->ifgr_len += sizeof(struct ifg_req);
1261 		IF_ADDR_RUNLOCK(ifp);
1262 		return (0);
1263 	}
1264 
1265 	len = ifgr->ifgr_len;
1266 	ifgp = ifgr->ifgr_groups;
1267 	/* XXX: wire */
1268 	IF_ADDR_RLOCK(ifp);
1269 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1270 		if (len < sizeof(ifgrq)) {
1271 			IF_ADDR_RUNLOCK(ifp);
1272 			return (EINVAL);
1273 		}
1274 		bzero(&ifgrq, sizeof ifgrq);
1275 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1276 		    sizeof(ifgrq.ifgrq_group));
1277 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1278 		    	IF_ADDR_RUNLOCK(ifp);
1279 			return (error);
1280 		}
1281 		len -= sizeof(ifgrq);
1282 		ifgp++;
1283 	}
1284 	IF_ADDR_RUNLOCK(ifp);
1285 
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Stores all members of a group in memory pointed to by data
1291  */
1292 static int
1293 if_getgroupmembers(struct ifgroupreq *data)
1294 {
1295 	struct ifgroupreq	*ifgr = data;
1296 	struct ifg_group	*ifg;
1297 	struct ifg_member	*ifgm;
1298 	struct ifg_req		 ifgrq, *ifgp;
1299 	int			 len, error;
1300 
1301 	IFNET_RLOCK();
1302 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1303 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1304 			break;
1305 	if (ifg == NULL) {
1306 		IFNET_RUNLOCK();
1307 		return (ENOENT);
1308 	}
1309 
1310 	if (ifgr->ifgr_len == 0) {
1311 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1312 			ifgr->ifgr_len += sizeof(ifgrq);
1313 		IFNET_RUNLOCK();
1314 		return (0);
1315 	}
1316 
1317 	len = ifgr->ifgr_len;
1318 	ifgp = ifgr->ifgr_groups;
1319 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1320 		if (len < sizeof(ifgrq)) {
1321 			IFNET_RUNLOCK();
1322 			return (EINVAL);
1323 		}
1324 		bzero(&ifgrq, sizeof ifgrq);
1325 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1326 		    sizeof(ifgrq.ifgrq_member));
1327 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1328 			IFNET_RUNLOCK();
1329 			return (error);
1330 		}
1331 		len -= sizeof(ifgrq);
1332 		ifgp++;
1333 	}
1334 	IFNET_RUNLOCK();
1335 
1336 	return (0);
1337 }
1338 
1339 /*
1340  * Delete Routes for a Network Interface
1341  *
1342  * Called for each routing entry via the rnh->rnh_walktree() call above
1343  * to delete all route entries referencing a detaching network interface.
1344  *
1345  * Arguments:
1346  *	rn	pointer to node in the routing table
1347  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1348  *
1349  * Returns:
1350  *	0	successful
1351  *	errno	failed - reason indicated
1352  *
1353  */
1354 static int
1355 if_rtdel(struct radix_node *rn, void *arg)
1356 {
1357 	struct rtentry	*rt = (struct rtentry *)rn;
1358 	struct ifnet	*ifp = arg;
1359 	int		err;
1360 
1361 	if (rt->rt_ifp == ifp) {
1362 
1363 		/*
1364 		 * Protect (sorta) against walktree recursion problems
1365 		 * with cloned routes
1366 		 */
1367 		if ((rt->rt_flags & RTF_UP) == 0)
1368 			return (0);
1369 
1370 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1371 				rt_mask(rt),
1372 				rt->rt_flags|RTF_RNH_LOCKED|RTF_PINNED,
1373 				(struct rtentry **) NULL, rt->rt_fibnum);
1374 		if (err) {
1375 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1376 		}
1377 	}
1378 
1379 	return (0);
1380 }
1381 
1382 /*
1383  * Wrapper functions for struct ifnet address list locking macros.  These are
1384  * used by kernel modules to avoid encoding programming interface or binary
1385  * interface assumptions that may be violated when kernel-internal locking
1386  * approaches change.
1387  */
1388 void
1389 if_addr_rlock(struct ifnet *ifp)
1390 {
1391 
1392 	IF_ADDR_RLOCK(ifp);
1393 }
1394 
1395 void
1396 if_addr_runlock(struct ifnet *ifp)
1397 {
1398 
1399 	IF_ADDR_RUNLOCK(ifp);
1400 }
1401 
1402 void
1403 if_maddr_rlock(struct ifnet *ifp)
1404 {
1405 
1406 	IF_ADDR_RLOCK(ifp);
1407 }
1408 
1409 void
1410 if_maddr_runlock(struct ifnet *ifp)
1411 {
1412 
1413 	IF_ADDR_RUNLOCK(ifp);
1414 }
1415 
1416 /*
1417  * Initialization, destruction and refcounting functions for ifaddrs.
1418  */
1419 struct ifaddr *
1420 ifa_alloc(size_t size, int flags)
1421 {
1422 	struct ifaddr *ifa;
1423 
1424 	KASSERT(size >= sizeof(struct ifaddr),
1425 	    ("%s: invalid size %zu", __func__, size));
1426 
1427 	ifa = malloc(size, M_IFADDR, M_ZERO | flags);
1428 	if (ifa == NULL)
1429 		return (NULL);
1430 
1431 	if ((ifa->ifa_opackets = counter_u64_alloc(flags)) == NULL)
1432 		goto fail;
1433 	if ((ifa->ifa_ipackets = counter_u64_alloc(flags)) == NULL)
1434 		goto fail;
1435 	if ((ifa->ifa_obytes = counter_u64_alloc(flags)) == NULL)
1436 		goto fail;
1437 	if ((ifa->ifa_ibytes = counter_u64_alloc(flags)) == NULL)
1438 		goto fail;
1439 
1440 	refcount_init(&ifa->ifa_refcnt, 1);
1441 
1442 	return (ifa);
1443 
1444 fail:
1445 	/* free(NULL) is okay */
1446 	counter_u64_free(ifa->ifa_opackets);
1447 	counter_u64_free(ifa->ifa_ipackets);
1448 	counter_u64_free(ifa->ifa_obytes);
1449 	counter_u64_free(ifa->ifa_ibytes);
1450 	free(ifa, M_IFADDR);
1451 
1452 	return (NULL);
1453 }
1454 
1455 void
1456 ifa_ref(struct ifaddr *ifa)
1457 {
1458 
1459 	refcount_acquire(&ifa->ifa_refcnt);
1460 }
1461 
1462 void
1463 ifa_free(struct ifaddr *ifa)
1464 {
1465 
1466 	if (refcount_release(&ifa->ifa_refcnt)) {
1467 		counter_u64_free(ifa->ifa_opackets);
1468 		counter_u64_free(ifa->ifa_ipackets);
1469 		counter_u64_free(ifa->ifa_obytes);
1470 		counter_u64_free(ifa->ifa_ibytes);
1471 		free(ifa, M_IFADDR);
1472 	}
1473 }
1474 
1475 int
1476 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1477 {
1478 	int error = 0;
1479 	struct rtentry *rt = NULL;
1480 	struct rt_addrinfo info;
1481 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1482 
1483 	bzero(&info, sizeof(info));
1484 	info.rti_ifp = V_loif;
1485 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1486 	info.rti_info[RTAX_DST] = ia;
1487 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1488 	error = rtrequest1_fib(RTM_ADD, &info, &rt, ifa->ifa_ifp->if_fib);
1489 
1490 	if (error == 0 && rt != NULL) {
1491 		RT_LOCK(rt);
1492 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1493 			ifa->ifa_ifp->if_type;
1494 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1495 			ifa->ifa_ifp->if_index;
1496 		RT_REMREF(rt);
1497 		RT_UNLOCK(rt);
1498 	} else if (error != 0)
1499 		log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1500 
1501 	return (error);
1502 }
1503 
1504 int
1505 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1506 {
1507 	int error = 0;
1508 	struct rt_addrinfo info;
1509 	struct sockaddr_dl null_sdl;
1510 
1511 	bzero(&null_sdl, sizeof(null_sdl));
1512 	null_sdl.sdl_len = sizeof(null_sdl);
1513 	null_sdl.sdl_family = AF_LINK;
1514 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1515 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1516 	bzero(&info, sizeof(info));
1517 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1518 	info.rti_info[RTAX_DST] = ia;
1519 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1520 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, ifa->ifa_ifp->if_fib);
1521 
1522 	if (error != 0)
1523 		log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1524 
1525 	return (error);
1526 }
1527 
1528 int
1529 ifa_switch_loopback_route(struct ifaddr *ifa, struct sockaddr *sa, int fib)
1530 {
1531 	struct rtentry *rt;
1532 
1533 	rt = rtalloc1_fib(sa, 0, 0, fib);
1534 	if (rt == NULL) {
1535 		log(LOG_DEBUG, "%s: fail", __func__);
1536 		return (EHOSTUNREACH);
1537 	}
1538 	((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1539 	    ifa->ifa_ifp->if_type;
1540 	((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1541 	    ifa->ifa_ifp->if_index;
1542 	RTFREE_LOCKED(rt);
1543 
1544 	return (0);
1545 }
1546 
1547 /*
1548  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1549  * structs used to represent other address families, it is necessary
1550  * to perform a different comparison.
1551  */
1552 
1553 #define	sa_dl_equal(a1, a2)	\
1554 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1555 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1556 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1557 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1558 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1559 
1560 /*
1561  * Locate an interface based on a complete address.
1562  */
1563 /*ARGSUSED*/
1564 static struct ifaddr *
1565 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1566 {
1567 	struct ifnet *ifp;
1568 	struct ifaddr *ifa;
1569 
1570 	IFNET_RLOCK_NOSLEEP();
1571 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1572 		IF_ADDR_RLOCK(ifp);
1573 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1574 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1575 				continue;
1576 			if (sa_equal(addr, ifa->ifa_addr)) {
1577 				if (getref)
1578 					ifa_ref(ifa);
1579 				IF_ADDR_RUNLOCK(ifp);
1580 				goto done;
1581 			}
1582 			/* IP6 doesn't have broadcast */
1583 			if ((ifp->if_flags & IFF_BROADCAST) &&
1584 			    ifa->ifa_broadaddr &&
1585 			    ifa->ifa_broadaddr->sa_len != 0 &&
1586 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1587 				if (getref)
1588 					ifa_ref(ifa);
1589 				IF_ADDR_RUNLOCK(ifp);
1590 				goto done;
1591 			}
1592 		}
1593 		IF_ADDR_RUNLOCK(ifp);
1594 	}
1595 	ifa = NULL;
1596 done:
1597 	IFNET_RUNLOCK_NOSLEEP();
1598 	return (ifa);
1599 }
1600 
1601 struct ifaddr *
1602 ifa_ifwithaddr(struct sockaddr *addr)
1603 {
1604 
1605 	return (ifa_ifwithaddr_internal(addr, 1));
1606 }
1607 
1608 int
1609 ifa_ifwithaddr_check(struct sockaddr *addr)
1610 {
1611 
1612 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1613 }
1614 
1615 /*
1616  * Locate an interface based on the broadcast address.
1617  */
1618 /* ARGSUSED */
1619 struct ifaddr *
1620 ifa_ifwithbroadaddr(struct sockaddr *addr)
1621 {
1622 	struct ifnet *ifp;
1623 	struct ifaddr *ifa;
1624 
1625 	IFNET_RLOCK_NOSLEEP();
1626 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1627 		IF_ADDR_RLOCK(ifp);
1628 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1629 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1630 				continue;
1631 			if ((ifp->if_flags & IFF_BROADCAST) &&
1632 			    ifa->ifa_broadaddr &&
1633 			    ifa->ifa_broadaddr->sa_len != 0 &&
1634 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1635 				ifa_ref(ifa);
1636 				IF_ADDR_RUNLOCK(ifp);
1637 				goto done;
1638 			}
1639 		}
1640 		IF_ADDR_RUNLOCK(ifp);
1641 	}
1642 	ifa = NULL;
1643 done:
1644 	IFNET_RUNLOCK_NOSLEEP();
1645 	return (ifa);
1646 }
1647 
1648 /*
1649  * Locate the point to point interface with a given destination address.
1650  */
1651 /*ARGSUSED*/
1652 struct ifaddr *
1653 ifa_ifwithdstaddr(struct sockaddr *addr, int fibnum)
1654 {
1655 	struct ifnet *ifp;
1656 	struct ifaddr *ifa;
1657 
1658 	IFNET_RLOCK_NOSLEEP();
1659 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1660 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1661 			continue;
1662 		if ((ifp->if_fib != fibnum))
1663 			continue;
1664 		IF_ADDR_RLOCK(ifp);
1665 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1666 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1667 				continue;
1668 			if (ifa->ifa_dstaddr != NULL &&
1669 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1670 				ifa_ref(ifa);
1671 				IF_ADDR_RUNLOCK(ifp);
1672 				goto done;
1673 			}
1674 		}
1675 		IF_ADDR_RUNLOCK(ifp);
1676 	}
1677 	ifa = NULL;
1678 done:
1679 	IFNET_RUNLOCK_NOSLEEP();
1680 	return (ifa);
1681 }
1682 
1683 /*
1684  * Find an interface on a specific network.  If many, choice
1685  * is most specific found.
1686  */
1687 struct ifaddr *
1688 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp, int fibnum)
1689 {
1690 	struct ifnet *ifp;
1691 	struct ifaddr *ifa;
1692 	struct ifaddr *ifa_maybe = NULL;
1693 	u_int af = addr->sa_family;
1694 	char *addr_data = addr->sa_data, *cplim;
1695 
1696 	/*
1697 	 * AF_LINK addresses can be looked up directly by their index number,
1698 	 * so do that if we can.
1699 	 */
1700 	if (af == AF_LINK) {
1701 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1702 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1703 		return (ifaddr_byindex(sdl->sdl_index));
1704 	}
1705 
1706 	/*
1707 	 * Scan though each interface, looking for ones that have addresses
1708 	 * in this address family and the requested fib.  Maintain a reference
1709 	 * on ifa_maybe once we find one, as we release the IF_ADDR_RLOCK() that
1710 	 * kept it stable when we move onto the next interface.
1711 	 */
1712 	IFNET_RLOCK_NOSLEEP();
1713 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1714 		if (ifp->if_fib != fibnum)
1715 			continue;
1716 		IF_ADDR_RLOCK(ifp);
1717 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1718 			char *cp, *cp2, *cp3;
1719 
1720 			if (ifa->ifa_addr->sa_family != af)
1721 next:				continue;
1722 			if (af == AF_INET &&
1723 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1724 				/*
1725 				 * This is a bit broken as it doesn't
1726 				 * take into account that the remote end may
1727 				 * be a single node in the network we are
1728 				 * looking for.
1729 				 * The trouble is that we don't know the
1730 				 * netmask for the remote end.
1731 				 */
1732 				if (ifa->ifa_dstaddr != NULL &&
1733 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1734 					ifa_ref(ifa);
1735 					IF_ADDR_RUNLOCK(ifp);
1736 					goto done;
1737 				}
1738 			} else {
1739 				/*
1740 				 * if we have a special address handler,
1741 				 * then use it instead of the generic one.
1742 				 */
1743 				if (ifa->ifa_claim_addr) {
1744 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1745 						ifa_ref(ifa);
1746 						IF_ADDR_RUNLOCK(ifp);
1747 						goto done;
1748 					}
1749 					continue;
1750 				}
1751 
1752 				/*
1753 				 * Scan all the bits in the ifa's address.
1754 				 * If a bit dissagrees with what we are
1755 				 * looking for, mask it with the netmask
1756 				 * to see if it really matters.
1757 				 * (A byte at a time)
1758 				 */
1759 				if (ifa->ifa_netmask == 0)
1760 					continue;
1761 				cp = addr_data;
1762 				cp2 = ifa->ifa_addr->sa_data;
1763 				cp3 = ifa->ifa_netmask->sa_data;
1764 				cplim = ifa->ifa_netmask->sa_len
1765 					+ (char *)ifa->ifa_netmask;
1766 				while (cp3 < cplim)
1767 					if ((*cp++ ^ *cp2++) & *cp3++)
1768 						goto next; /* next address! */
1769 				/*
1770 				 * If the netmask of what we just found
1771 				 * is more specific than what we had before
1772 				 * (if we had one), or if the virtual status
1773 				 * of new prefix is better than of the old one,
1774 				 * then remember the new one before continuing
1775 				 * to search for an even better one.
1776 				 */
1777 				if (ifa_maybe == NULL ||
1778 				    ifa_preferred(ifa_maybe, ifa) ||
1779 				    rn_refines((caddr_t)ifa->ifa_netmask,
1780 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1781 					if (ifa_maybe != NULL)
1782 						ifa_free(ifa_maybe);
1783 					ifa_maybe = ifa;
1784 					ifa_ref(ifa_maybe);
1785 				}
1786 			}
1787 		}
1788 		IF_ADDR_RUNLOCK(ifp);
1789 	}
1790 	ifa = ifa_maybe;
1791 	ifa_maybe = NULL;
1792 done:
1793 	IFNET_RUNLOCK_NOSLEEP();
1794 	if (ifa_maybe != NULL)
1795 		ifa_free(ifa_maybe);
1796 	return (ifa);
1797 }
1798 
1799 /*
1800  * Find an interface address specific to an interface best matching
1801  * a given address.
1802  */
1803 struct ifaddr *
1804 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1805 {
1806 	struct ifaddr *ifa;
1807 	char *cp, *cp2, *cp3;
1808 	char *cplim;
1809 	struct ifaddr *ifa_maybe = NULL;
1810 	u_int af = addr->sa_family;
1811 
1812 	if (af >= AF_MAX)
1813 		return (NULL);
1814 	IF_ADDR_RLOCK(ifp);
1815 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1816 		if (ifa->ifa_addr->sa_family != af)
1817 			continue;
1818 		if (ifa_maybe == NULL)
1819 			ifa_maybe = ifa;
1820 		if (ifa->ifa_netmask == 0) {
1821 			if (sa_equal(addr, ifa->ifa_addr) ||
1822 			    (ifa->ifa_dstaddr &&
1823 			    sa_equal(addr, ifa->ifa_dstaddr)))
1824 				goto done;
1825 			continue;
1826 		}
1827 		if (ifp->if_flags & IFF_POINTOPOINT) {
1828 			if (sa_equal(addr, ifa->ifa_dstaddr))
1829 				goto done;
1830 		} else {
1831 			cp = addr->sa_data;
1832 			cp2 = ifa->ifa_addr->sa_data;
1833 			cp3 = ifa->ifa_netmask->sa_data;
1834 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1835 			for (; cp3 < cplim; cp3++)
1836 				if ((*cp++ ^ *cp2++) & *cp3)
1837 					break;
1838 			if (cp3 == cplim)
1839 				goto done;
1840 		}
1841 	}
1842 	ifa = ifa_maybe;
1843 done:
1844 	if (ifa != NULL)
1845 		ifa_ref(ifa);
1846 	IF_ADDR_RUNLOCK(ifp);
1847 	return (ifa);
1848 }
1849 
1850 /*
1851  * See whether new ifa is better than current one:
1852  * 1) A non-virtual one is preferred over virtual.
1853  * 2) A virtual in master state preferred over any other state.
1854  *
1855  * Used in several address selecting functions.
1856  */
1857 int
1858 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
1859 {
1860 
1861 	return (cur->ifa_carp && (!next->ifa_carp ||
1862 	    ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
1863 }
1864 
1865 #include <net/if_llatbl.h>
1866 
1867 /*
1868  * Default action when installing a route with a Link Level gateway.
1869  * Lookup an appropriate real ifa to point to.
1870  * This should be moved to /sys/net/link.c eventually.
1871  */
1872 static void
1873 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1874 {
1875 	struct ifaddr *ifa, *oifa;
1876 	struct sockaddr *dst;
1877 	struct ifnet *ifp;
1878 
1879 	RT_LOCK_ASSERT(rt);
1880 
1881 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1882 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1883 		return;
1884 	ifa = ifaof_ifpforaddr(dst, ifp);
1885 	if (ifa) {
1886 		oifa = rt->rt_ifa;
1887 		rt->rt_ifa = ifa;
1888 		ifa_free(oifa);
1889 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1890 			ifa->ifa_rtrequest(cmd, rt, info);
1891 	}
1892 }
1893 
1894 struct sockaddr_dl *
1895 link_alloc_sdl(size_t size, int flags)
1896 {
1897 
1898 	return (malloc(size, M_TEMP, flags));
1899 }
1900 
1901 void
1902 link_free_sdl(struct sockaddr *sa)
1903 {
1904 	free(sa, M_TEMP);
1905 }
1906 
1907 /*
1908  * Fills in given sdl with interface basic info.
1909  * Returns pointer to filled sdl.
1910  */
1911 struct sockaddr_dl *
1912 link_init_sdl(struct ifnet *ifp, struct sockaddr *paddr, u_char iftype)
1913 {
1914 	struct sockaddr_dl *sdl;
1915 
1916 	sdl = (struct sockaddr_dl *)paddr;
1917 	memset(sdl, 0, sizeof(struct sockaddr_dl));
1918 	sdl->sdl_len = sizeof(struct sockaddr_dl);
1919 	sdl->sdl_family = AF_LINK;
1920 	sdl->sdl_index = ifp->if_index;
1921 	sdl->sdl_type = iftype;
1922 
1923 	return (sdl);
1924 }
1925 
1926 /*
1927  * Mark an interface down and notify protocols of
1928  * the transition.
1929  */
1930 static void
1931 if_unroute(struct ifnet *ifp, int flag, int fam)
1932 {
1933 	struct ifaddr *ifa;
1934 
1935 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1936 
1937 	ifp->if_flags &= ~flag;
1938 	getmicrotime(&ifp->if_lastchange);
1939 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1940 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1941 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1942 	ifp->if_qflush(ifp);
1943 
1944 	if (ifp->if_carp)
1945 		(*carp_linkstate_p)(ifp);
1946 	rt_ifmsg(ifp);
1947 }
1948 
1949 /*
1950  * Mark an interface up and notify protocols of
1951  * the transition.
1952  */
1953 static void
1954 if_route(struct ifnet *ifp, int flag, int fam)
1955 {
1956 	struct ifaddr *ifa;
1957 
1958 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1959 
1960 	ifp->if_flags |= flag;
1961 	getmicrotime(&ifp->if_lastchange);
1962 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1963 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1964 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1965 	if (ifp->if_carp)
1966 		(*carp_linkstate_p)(ifp);
1967 	rt_ifmsg(ifp);
1968 #ifdef INET6
1969 	in6_if_up(ifp);
1970 #endif
1971 }
1972 
1973 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1974 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1975 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1976 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1977 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
1978 int	(*vlan_setcookie_p)(struct ifnet *, void *);
1979 void	*(*vlan_cookie_p)(struct ifnet *);
1980 
1981 /*
1982  * Handle a change in the interface link state. To avoid LORs
1983  * between driver lock and upper layer locks, as well as possible
1984  * recursions, we post event to taskqueue, and all job
1985  * is done in static do_link_state_change().
1986  */
1987 void
1988 if_link_state_change(struct ifnet *ifp, int link_state)
1989 {
1990 	/* Return if state hasn't changed. */
1991 	if (ifp->if_link_state == link_state)
1992 		return;
1993 
1994 	ifp->if_link_state = link_state;
1995 
1996 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1997 }
1998 
1999 static void
2000 do_link_state_change(void *arg, int pending)
2001 {
2002 	struct ifnet *ifp = (struct ifnet *)arg;
2003 	int link_state = ifp->if_link_state;
2004 	CURVNET_SET(ifp->if_vnet);
2005 
2006 	/* Notify that the link state has changed. */
2007 	rt_ifmsg(ifp);
2008 	if (ifp->if_vlantrunk != NULL)
2009 		(*vlan_link_state_p)(ifp);
2010 
2011 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
2012 	    IFP2AC(ifp)->ac_netgraph != NULL)
2013 		(*ng_ether_link_state_p)(ifp, link_state);
2014 	if (ifp->if_carp)
2015 		(*carp_linkstate_p)(ifp);
2016 	if (ifp->if_bridge)
2017 		(*bridge_linkstate_p)(ifp);
2018 	if (ifp->if_lagg)
2019 		(*lagg_linkstate_p)(ifp, link_state);
2020 
2021 	if (IS_DEFAULT_VNET(curvnet))
2022 		devctl_notify("IFNET", ifp->if_xname,
2023 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
2024 		    NULL);
2025 	if (pending > 1)
2026 		if_printf(ifp, "%d link states coalesced\n", pending);
2027 	if (log_link_state_change)
2028 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
2029 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
2030 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
2031 	CURVNET_RESTORE();
2032 }
2033 
2034 /*
2035  * Mark an interface down and notify protocols of
2036  * the transition.
2037  */
2038 void
2039 if_down(struct ifnet *ifp)
2040 {
2041 
2042 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
2043 }
2044 
2045 /*
2046  * Mark an interface up and notify protocols of
2047  * the transition.
2048  */
2049 void
2050 if_up(struct ifnet *ifp)
2051 {
2052 
2053 	if_route(ifp, IFF_UP, AF_UNSPEC);
2054 }
2055 
2056 /*
2057  * Flush an interface queue.
2058  */
2059 void
2060 if_qflush(struct ifnet *ifp)
2061 {
2062 	struct mbuf *m, *n;
2063 	struct ifaltq *ifq;
2064 
2065 	ifq = &ifp->if_snd;
2066 	IFQ_LOCK(ifq);
2067 #ifdef ALTQ
2068 	if (ALTQ_IS_ENABLED(ifq))
2069 		ALTQ_PURGE(ifq);
2070 #endif
2071 	n = ifq->ifq_head;
2072 	while ((m = n) != 0) {
2073 		n = m->m_act;
2074 		m_freem(m);
2075 	}
2076 	ifq->ifq_head = 0;
2077 	ifq->ifq_tail = 0;
2078 	ifq->ifq_len = 0;
2079 	IFQ_UNLOCK(ifq);
2080 }
2081 
2082 /*
2083  * Map interface name to interface structure pointer, with or without
2084  * returning a reference.
2085  */
2086 struct ifnet *
2087 ifunit_ref(const char *name)
2088 {
2089 	struct ifnet *ifp;
2090 
2091 	IFNET_RLOCK_NOSLEEP();
2092 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2093 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2094 		    !(ifp->if_flags & IFF_DYING))
2095 			break;
2096 	}
2097 	if (ifp != NULL)
2098 		if_ref(ifp);
2099 	IFNET_RUNLOCK_NOSLEEP();
2100 	return (ifp);
2101 }
2102 
2103 struct ifnet *
2104 ifunit(const char *name)
2105 {
2106 	struct ifnet *ifp;
2107 
2108 	IFNET_RLOCK_NOSLEEP();
2109 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2110 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2111 			break;
2112 	}
2113 	IFNET_RUNLOCK_NOSLEEP();
2114 	return (ifp);
2115 }
2116 
2117 /*
2118  * Hardware specific interface ioctls.
2119  */
2120 static int
2121 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2122 {
2123 	struct ifreq *ifr;
2124 	int error = 0;
2125 	int new_flags, temp_flags;
2126 	size_t namelen, onamelen;
2127 	size_t descrlen;
2128 	char *descrbuf, *odescrbuf;
2129 	char new_name[IFNAMSIZ];
2130 	struct ifaddr *ifa;
2131 	struct sockaddr_dl *sdl;
2132 
2133 	ifr = (struct ifreq *)data;
2134 	switch (cmd) {
2135 	case SIOCGIFINDEX:
2136 		ifr->ifr_index = ifp->if_index;
2137 		break;
2138 
2139 	case SIOCGIFFLAGS:
2140 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2141 		ifr->ifr_flags = temp_flags & 0xffff;
2142 		ifr->ifr_flagshigh = temp_flags >> 16;
2143 		break;
2144 
2145 	case SIOCGIFCAP:
2146 		ifr->ifr_reqcap = ifp->if_capabilities;
2147 		ifr->ifr_curcap = ifp->if_capenable;
2148 		break;
2149 
2150 #ifdef MAC
2151 	case SIOCGIFMAC:
2152 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2153 		break;
2154 #endif
2155 
2156 	case SIOCGIFMETRIC:
2157 		ifr->ifr_metric = ifp->if_metric;
2158 		break;
2159 
2160 	case SIOCGIFMTU:
2161 		ifr->ifr_mtu = ifp->if_mtu;
2162 		break;
2163 
2164 	case SIOCGIFPHYS:
2165 		ifr->ifr_phys = ifp->if_physical;
2166 		break;
2167 
2168 	case SIOCGIFDESCR:
2169 		error = 0;
2170 		sx_slock(&ifdescr_sx);
2171 		if (ifp->if_description == NULL)
2172 			error = ENOMSG;
2173 		else {
2174 			/* space for terminating nul */
2175 			descrlen = strlen(ifp->if_description) + 1;
2176 			if (ifr->ifr_buffer.length < descrlen)
2177 				ifr->ifr_buffer.buffer = NULL;
2178 			else
2179 				error = copyout(ifp->if_description,
2180 				    ifr->ifr_buffer.buffer, descrlen);
2181 			ifr->ifr_buffer.length = descrlen;
2182 		}
2183 		sx_sunlock(&ifdescr_sx);
2184 		break;
2185 
2186 	case SIOCSIFDESCR:
2187 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2188 		if (error)
2189 			return (error);
2190 
2191 		/*
2192 		 * Copy only (length-1) bytes to make sure that
2193 		 * if_description is always nul terminated.  The
2194 		 * length parameter is supposed to count the
2195 		 * terminating nul in.
2196 		 */
2197 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2198 			return (ENAMETOOLONG);
2199 		else if (ifr->ifr_buffer.length == 0)
2200 			descrbuf = NULL;
2201 		else {
2202 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2203 			    M_WAITOK | M_ZERO);
2204 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2205 			    ifr->ifr_buffer.length - 1);
2206 			if (error) {
2207 				free(descrbuf, M_IFDESCR);
2208 				break;
2209 			}
2210 		}
2211 
2212 		sx_xlock(&ifdescr_sx);
2213 		odescrbuf = ifp->if_description;
2214 		ifp->if_description = descrbuf;
2215 		sx_xunlock(&ifdescr_sx);
2216 
2217 		getmicrotime(&ifp->if_lastchange);
2218 		free(odescrbuf, M_IFDESCR);
2219 		break;
2220 
2221 	case SIOCGIFFIB:
2222 		ifr->ifr_fib = ifp->if_fib;
2223 		break;
2224 
2225 	case SIOCSIFFIB:
2226 		error = priv_check(td, PRIV_NET_SETIFFIB);
2227 		if (error)
2228 			return (error);
2229 		if (ifr->ifr_fib >= rt_numfibs)
2230 			return (EINVAL);
2231 
2232 		ifp->if_fib = ifr->ifr_fib;
2233 		break;
2234 
2235 	case SIOCSIFFLAGS:
2236 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2237 		if (error)
2238 			return (error);
2239 		/*
2240 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2241 		 * check, so we don't need special handling here yet.
2242 		 */
2243 		new_flags = (ifr->ifr_flags & 0xffff) |
2244 		    (ifr->ifr_flagshigh << 16);
2245 		if (ifp->if_flags & IFF_UP &&
2246 		    (new_flags & IFF_UP) == 0) {
2247 			if_down(ifp);
2248 		} else if (new_flags & IFF_UP &&
2249 		    (ifp->if_flags & IFF_UP) == 0) {
2250 			if_up(ifp);
2251 		}
2252 		/* See if permanently promiscuous mode bit is about to flip */
2253 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2254 			if (new_flags & IFF_PPROMISC)
2255 				ifp->if_flags |= IFF_PROMISC;
2256 			else if (ifp->if_pcount == 0)
2257 				ifp->if_flags &= ~IFF_PROMISC;
2258 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2259 			    ifp->if_xname,
2260 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2261 		}
2262 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2263 			(new_flags &~ IFF_CANTCHANGE);
2264 		if (ifp->if_ioctl) {
2265 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2266 		}
2267 		getmicrotime(&ifp->if_lastchange);
2268 		break;
2269 
2270 	case SIOCSIFCAP:
2271 		error = priv_check(td, PRIV_NET_SETIFCAP);
2272 		if (error)
2273 			return (error);
2274 		if (ifp->if_ioctl == NULL)
2275 			return (EOPNOTSUPP);
2276 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2277 			return (EINVAL);
2278 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2279 		if (error == 0)
2280 			getmicrotime(&ifp->if_lastchange);
2281 		break;
2282 
2283 #ifdef MAC
2284 	case SIOCSIFMAC:
2285 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2286 		break;
2287 #endif
2288 
2289 	case SIOCSIFNAME:
2290 		error = priv_check(td, PRIV_NET_SETIFNAME);
2291 		if (error)
2292 			return (error);
2293 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2294 		if (error != 0)
2295 			return (error);
2296 		if (new_name[0] == '\0')
2297 			return (EINVAL);
2298 		if (ifunit(new_name) != NULL)
2299 			return (EEXIST);
2300 
2301 		/*
2302 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2303 		 * and there are numerous other races with the
2304 		 * ifunit() checks not being atomic with namespace
2305 		 * changes (renames, vmoves, if_attach, etc).
2306 		 */
2307 		ifp->if_flags |= IFF_RENAMING;
2308 
2309 		/* Announce the departure of the interface. */
2310 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2311 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2312 
2313 		log(LOG_INFO, "%s: changing name to '%s'\n",
2314 		    ifp->if_xname, new_name);
2315 
2316 		IF_ADDR_WLOCK(ifp);
2317 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2318 		ifa = ifp->if_addr;
2319 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2320 		namelen = strlen(new_name);
2321 		onamelen = sdl->sdl_nlen;
2322 		/*
2323 		 * Move the address if needed.  This is safe because we
2324 		 * allocate space for a name of length IFNAMSIZ when we
2325 		 * create this in if_attach().
2326 		 */
2327 		if (namelen != onamelen) {
2328 			bcopy(sdl->sdl_data + onamelen,
2329 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2330 		}
2331 		bcopy(new_name, sdl->sdl_data, namelen);
2332 		sdl->sdl_nlen = namelen;
2333 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2334 		bzero(sdl->sdl_data, onamelen);
2335 		while (namelen != 0)
2336 			sdl->sdl_data[--namelen] = 0xff;
2337 		IF_ADDR_WUNLOCK(ifp);
2338 
2339 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2340 		/* Announce the return of the interface. */
2341 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2342 
2343 		ifp->if_flags &= ~IFF_RENAMING;
2344 		break;
2345 
2346 #ifdef VIMAGE
2347 	case SIOCSIFVNET:
2348 		error = priv_check(td, PRIV_NET_SETIFVNET);
2349 		if (error)
2350 			return (error);
2351 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2352 		break;
2353 #endif
2354 
2355 	case SIOCSIFMETRIC:
2356 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2357 		if (error)
2358 			return (error);
2359 		ifp->if_metric = ifr->ifr_metric;
2360 		getmicrotime(&ifp->if_lastchange);
2361 		break;
2362 
2363 	case SIOCSIFPHYS:
2364 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2365 		if (error)
2366 			return (error);
2367 		if (ifp->if_ioctl == NULL)
2368 			return (EOPNOTSUPP);
2369 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2370 		if (error == 0)
2371 			getmicrotime(&ifp->if_lastchange);
2372 		break;
2373 
2374 	case SIOCSIFMTU:
2375 	{
2376 		u_long oldmtu = ifp->if_mtu;
2377 
2378 		error = priv_check(td, PRIV_NET_SETIFMTU);
2379 		if (error)
2380 			return (error);
2381 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2382 			return (EINVAL);
2383 		if (ifp->if_ioctl == NULL)
2384 			return (EOPNOTSUPP);
2385 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2386 		if (error == 0) {
2387 			getmicrotime(&ifp->if_lastchange);
2388 			rt_ifmsg(ifp);
2389 		}
2390 		/*
2391 		 * If the link MTU changed, do network layer specific procedure.
2392 		 */
2393 		if (ifp->if_mtu != oldmtu) {
2394 #ifdef INET6
2395 			nd6_setmtu(ifp);
2396 #endif
2397 		}
2398 		break;
2399 	}
2400 
2401 	case SIOCADDMULTI:
2402 	case SIOCDELMULTI:
2403 		if (cmd == SIOCADDMULTI)
2404 			error = priv_check(td, PRIV_NET_ADDMULTI);
2405 		else
2406 			error = priv_check(td, PRIV_NET_DELMULTI);
2407 		if (error)
2408 			return (error);
2409 
2410 		/* Don't allow group membership on non-multicast interfaces. */
2411 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2412 			return (EOPNOTSUPP);
2413 
2414 		/* Don't let users screw up protocols' entries. */
2415 		if (ifr->ifr_addr.sa_family != AF_LINK)
2416 			return (EINVAL);
2417 
2418 		if (cmd == SIOCADDMULTI) {
2419 			struct ifmultiaddr *ifma;
2420 
2421 			/*
2422 			 * Userland is only permitted to join groups once
2423 			 * via the if_addmulti() KPI, because it cannot hold
2424 			 * struct ifmultiaddr * between calls. It may also
2425 			 * lose a race while we check if the membership
2426 			 * already exists.
2427 			 */
2428 			IF_ADDR_RLOCK(ifp);
2429 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2430 			IF_ADDR_RUNLOCK(ifp);
2431 			if (ifma != NULL)
2432 				error = EADDRINUSE;
2433 			else
2434 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2435 		} else {
2436 			error = if_delmulti(ifp, &ifr->ifr_addr);
2437 		}
2438 		if (error == 0)
2439 			getmicrotime(&ifp->if_lastchange);
2440 		break;
2441 
2442 	case SIOCSIFPHYADDR:
2443 	case SIOCDIFPHYADDR:
2444 #ifdef INET6
2445 	case SIOCSIFPHYADDR_IN6:
2446 #endif
2447 	case SIOCSIFMEDIA:
2448 	case SIOCSIFGENERIC:
2449 		error = priv_check(td, PRIV_NET_HWIOCTL);
2450 		if (error)
2451 			return (error);
2452 		if (ifp->if_ioctl == NULL)
2453 			return (EOPNOTSUPP);
2454 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2455 		if (error == 0)
2456 			getmicrotime(&ifp->if_lastchange);
2457 		break;
2458 
2459 	case SIOCGIFSTATUS:
2460 	case SIOCGIFPSRCADDR:
2461 	case SIOCGIFPDSTADDR:
2462 	case SIOCGIFMEDIA:
2463 	case SIOCGIFGENERIC:
2464 		if (ifp->if_ioctl == NULL)
2465 			return (EOPNOTSUPP);
2466 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2467 		break;
2468 
2469 	case SIOCSIFLLADDR:
2470 		error = priv_check(td, PRIV_NET_SETLLADDR);
2471 		if (error)
2472 			return (error);
2473 		error = if_setlladdr(ifp,
2474 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2475 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2476 		break;
2477 
2478 	case SIOCAIFGROUP:
2479 	{
2480 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2481 
2482 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2483 		if (error)
2484 			return (error);
2485 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2486 			return (error);
2487 		break;
2488 	}
2489 
2490 	case SIOCGIFGROUP:
2491 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2492 			return (error);
2493 		break;
2494 
2495 	case SIOCDIFGROUP:
2496 	{
2497 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2498 
2499 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2500 		if (error)
2501 			return (error);
2502 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2503 			return (error);
2504 		break;
2505 	}
2506 
2507 	default:
2508 		error = ENOIOCTL;
2509 		break;
2510 	}
2511 	return (error);
2512 }
2513 
2514 #ifdef COMPAT_FREEBSD32
2515 struct ifconf32 {
2516 	int32_t	ifc_len;
2517 	union {
2518 		uint32_t	ifcu_buf;
2519 		uint32_t	ifcu_req;
2520 	} ifc_ifcu;
2521 };
2522 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2523 #endif
2524 
2525 /*
2526  * Interface ioctls.
2527  */
2528 int
2529 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2530 {
2531 	struct ifnet *ifp;
2532 	struct ifreq *ifr;
2533 	int error;
2534 	int oif_flags;
2535 
2536 	CURVNET_SET(so->so_vnet);
2537 	switch (cmd) {
2538 	case SIOCGIFCONF:
2539 		error = ifconf(cmd, data);
2540 		CURVNET_RESTORE();
2541 		return (error);
2542 
2543 #ifdef COMPAT_FREEBSD32
2544 	case SIOCGIFCONF32:
2545 		{
2546 			struct ifconf32 *ifc32;
2547 			struct ifconf ifc;
2548 
2549 			ifc32 = (struct ifconf32 *)data;
2550 			ifc.ifc_len = ifc32->ifc_len;
2551 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2552 
2553 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2554 			CURVNET_RESTORE();
2555 			if (error == 0)
2556 				ifc32->ifc_len = ifc.ifc_len;
2557 			return (error);
2558 		}
2559 #endif
2560 	}
2561 	ifr = (struct ifreq *)data;
2562 
2563 	switch (cmd) {
2564 #ifdef VIMAGE
2565 	case SIOCSIFRVNET:
2566 		error = priv_check(td, PRIV_NET_SETIFVNET);
2567 		if (error == 0)
2568 			error = if_vmove_reclaim(td, ifr->ifr_name,
2569 			    ifr->ifr_jid);
2570 		CURVNET_RESTORE();
2571 		return (error);
2572 #endif
2573 	case SIOCIFCREATE:
2574 	case SIOCIFCREATE2:
2575 		error = priv_check(td, PRIV_NET_IFCREATE);
2576 		if (error == 0)
2577 			error = if_clone_create(ifr->ifr_name,
2578 			    sizeof(ifr->ifr_name),
2579 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2580 		CURVNET_RESTORE();
2581 		return (error);
2582 	case SIOCIFDESTROY:
2583 		error = priv_check(td, PRIV_NET_IFDESTROY);
2584 		if (error == 0)
2585 			error = if_clone_destroy(ifr->ifr_name);
2586 		CURVNET_RESTORE();
2587 		return (error);
2588 
2589 	case SIOCIFGCLONERS:
2590 		error = if_clone_list((struct if_clonereq *)data);
2591 		CURVNET_RESTORE();
2592 		return (error);
2593 	case SIOCGIFGMEMB:
2594 		error = if_getgroupmembers((struct ifgroupreq *)data);
2595 		CURVNET_RESTORE();
2596 		return (error);
2597 #if defined(INET) || defined(INET6)
2598 	case SIOCSVH:
2599 	case SIOCGVH:
2600 		if (carp_ioctl_p == NULL)
2601 			error = EPROTONOSUPPORT;
2602 		else
2603 			error = (*carp_ioctl_p)(ifr, cmd, td);
2604 		CURVNET_RESTORE();
2605 		return (error);
2606 #endif
2607 	}
2608 
2609 	ifp = ifunit_ref(ifr->ifr_name);
2610 	if (ifp == NULL) {
2611 		CURVNET_RESTORE();
2612 		return (ENXIO);
2613 	}
2614 
2615 	error = ifhwioctl(cmd, ifp, data, td);
2616 	if (error != ENOIOCTL) {
2617 		if_rele(ifp);
2618 		CURVNET_RESTORE();
2619 		return (error);
2620 	}
2621 
2622 	oif_flags = ifp->if_flags;
2623 	if (so->so_proto == NULL) {
2624 		if_rele(ifp);
2625 		CURVNET_RESTORE();
2626 		return (EOPNOTSUPP);
2627 	}
2628 
2629 	/*
2630 	 * Pass the request on to the socket control method, and if the
2631 	 * latter returns EOPNOTSUPP, directly to the interface.
2632 	 *
2633 	 * Make an exception for the legacy SIOCSIF* requests.  Drivers
2634 	 * trust SIOCSIFADDR et al to come from an already privileged
2635 	 * layer, and do not perform any credentials checks or input
2636 	 * validation.
2637 	 */
2638 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd, data,
2639 	    ifp, td));
2640 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
2641 	    cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2642 	    cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2643 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2644 
2645 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2646 #ifdef INET6
2647 		if (ifp->if_flags & IFF_UP)
2648 			in6_if_up(ifp);
2649 #endif
2650 	}
2651 	if_rele(ifp);
2652 	CURVNET_RESTORE();
2653 	return (error);
2654 }
2655 
2656 /*
2657  * The code common to handling reference counted flags,
2658  * e.g., in ifpromisc() and if_allmulti().
2659  * The "pflag" argument can specify a permanent mode flag to check,
2660  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2661  *
2662  * Only to be used on stack-owned flags, not driver-owned flags.
2663  */
2664 static int
2665 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2666 {
2667 	struct ifreq ifr;
2668 	int error;
2669 	int oldflags, oldcount;
2670 
2671 	/* Sanity checks to catch programming errors */
2672 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2673 	    ("%s: setting driver-owned flag %d", __func__, flag));
2674 
2675 	if (onswitch)
2676 		KASSERT(*refcount >= 0,
2677 		    ("%s: increment negative refcount %d for flag %d",
2678 		    __func__, *refcount, flag));
2679 	else
2680 		KASSERT(*refcount > 0,
2681 		    ("%s: decrement non-positive refcount %d for flag %d",
2682 		    __func__, *refcount, flag));
2683 
2684 	/* In case this mode is permanent, just touch refcount */
2685 	if (ifp->if_flags & pflag) {
2686 		*refcount += onswitch ? 1 : -1;
2687 		return (0);
2688 	}
2689 
2690 	/* Save ifnet parameters for if_ioctl() may fail */
2691 	oldcount = *refcount;
2692 	oldflags = ifp->if_flags;
2693 
2694 	/*
2695 	 * See if we aren't the only and touching refcount is enough.
2696 	 * Actually toggle interface flag if we are the first or last.
2697 	 */
2698 	if (onswitch) {
2699 		if ((*refcount)++)
2700 			return (0);
2701 		ifp->if_flags |= flag;
2702 	} else {
2703 		if (--(*refcount))
2704 			return (0);
2705 		ifp->if_flags &= ~flag;
2706 	}
2707 
2708 	/* Call down the driver since we've changed interface flags */
2709 	if (ifp->if_ioctl == NULL) {
2710 		error = EOPNOTSUPP;
2711 		goto recover;
2712 	}
2713 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2714 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2715 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2716 	if (error)
2717 		goto recover;
2718 	/* Notify userland that interface flags have changed */
2719 	rt_ifmsg(ifp);
2720 	return (0);
2721 
2722 recover:
2723 	/* Recover after driver error */
2724 	*refcount = oldcount;
2725 	ifp->if_flags = oldflags;
2726 	return (error);
2727 }
2728 
2729 /*
2730  * Set/clear promiscuous mode on interface ifp based on the truth value
2731  * of pswitch.  The calls are reference counted so that only the first
2732  * "on" request actually has an effect, as does the final "off" request.
2733  * Results are undefined if the "off" and "on" requests are not matched.
2734  */
2735 int
2736 ifpromisc(struct ifnet *ifp, int pswitch)
2737 {
2738 	int error;
2739 	int oldflags = ifp->if_flags;
2740 
2741 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2742 			   &ifp->if_pcount, pswitch);
2743 	/* If promiscuous mode status has changed, log a message */
2744 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2745 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2746 		    ifp->if_xname,
2747 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2748 	return (error);
2749 }
2750 
2751 /*
2752  * Return interface configuration
2753  * of system.  List may be used
2754  * in later ioctl's (above) to get
2755  * other information.
2756  */
2757 /*ARGSUSED*/
2758 static int
2759 ifconf(u_long cmd, caddr_t data)
2760 {
2761 	struct ifconf *ifc = (struct ifconf *)data;
2762 	struct ifnet *ifp;
2763 	struct ifaddr *ifa;
2764 	struct ifreq ifr;
2765 	struct sbuf *sb;
2766 	int error, full = 0, valid_len, max_len;
2767 
2768 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2769 	max_len = MAXPHYS - 1;
2770 
2771 	/* Prevent hostile input from being able to crash the system */
2772 	if (ifc->ifc_len <= 0)
2773 		return (EINVAL);
2774 
2775 again:
2776 	if (ifc->ifc_len <= max_len) {
2777 		max_len = ifc->ifc_len;
2778 		full = 1;
2779 	}
2780 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2781 	max_len = 0;
2782 	valid_len = 0;
2783 
2784 	IFNET_RLOCK();
2785 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2786 		int addrs;
2787 
2788 		/*
2789 		 * Zero the ifr_name buffer to make sure we don't
2790 		 * disclose the contents of the stack.
2791 		 */
2792 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2793 
2794 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2795 		    >= sizeof(ifr.ifr_name)) {
2796 			sbuf_delete(sb);
2797 			IFNET_RUNLOCK();
2798 			return (ENAMETOOLONG);
2799 		}
2800 
2801 		addrs = 0;
2802 		IF_ADDR_RLOCK(ifp);
2803 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2804 			struct sockaddr *sa = ifa->ifa_addr;
2805 
2806 			if (prison_if(curthread->td_ucred, sa) != 0)
2807 				continue;
2808 			addrs++;
2809 			if (sa->sa_len <= sizeof(*sa)) {
2810 				ifr.ifr_addr = *sa;
2811 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2812 				max_len += sizeof(ifr);
2813 			} else {
2814 				sbuf_bcat(sb, &ifr,
2815 				    offsetof(struct ifreq, ifr_addr));
2816 				max_len += offsetof(struct ifreq, ifr_addr);
2817 				sbuf_bcat(sb, sa, sa->sa_len);
2818 				max_len += sa->sa_len;
2819 			}
2820 
2821 			if (sbuf_error(sb) == 0)
2822 				valid_len = sbuf_len(sb);
2823 		}
2824 		IF_ADDR_RUNLOCK(ifp);
2825 		if (addrs == 0) {
2826 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2827 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2828 			max_len += sizeof(ifr);
2829 
2830 			if (sbuf_error(sb) == 0)
2831 				valid_len = sbuf_len(sb);
2832 		}
2833 	}
2834 	IFNET_RUNLOCK();
2835 
2836 	/*
2837 	 * If we didn't allocate enough space (uncommon), try again.  If
2838 	 * we have already allocated as much space as we are allowed,
2839 	 * return what we've got.
2840 	 */
2841 	if (valid_len != max_len && !full) {
2842 		sbuf_delete(sb);
2843 		goto again;
2844 	}
2845 
2846 	ifc->ifc_len = valid_len;
2847 	sbuf_finish(sb);
2848 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2849 	sbuf_delete(sb);
2850 	return (error);
2851 }
2852 
2853 /*
2854  * Just like ifpromisc(), but for all-multicast-reception mode.
2855  */
2856 int
2857 if_allmulti(struct ifnet *ifp, int onswitch)
2858 {
2859 
2860 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2861 }
2862 
2863 struct ifmultiaddr *
2864 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2865 {
2866 	struct ifmultiaddr *ifma;
2867 
2868 	IF_ADDR_LOCK_ASSERT(ifp);
2869 
2870 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2871 		if (sa->sa_family == AF_LINK) {
2872 			if (sa_dl_equal(ifma->ifma_addr, sa))
2873 				break;
2874 		} else {
2875 			if (sa_equal(ifma->ifma_addr, sa))
2876 				break;
2877 		}
2878 	}
2879 
2880 	return ifma;
2881 }
2882 
2883 /*
2884  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2885  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2886  * the ifnet multicast address list here, so the caller must do that and
2887  * other setup work (such as notifying the device driver).  The reference
2888  * count is initialized to 1.
2889  */
2890 static struct ifmultiaddr *
2891 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2892     int mflags)
2893 {
2894 	struct ifmultiaddr *ifma;
2895 	struct sockaddr *dupsa;
2896 
2897 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2898 	    M_ZERO);
2899 	if (ifma == NULL)
2900 		return (NULL);
2901 
2902 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2903 	if (dupsa == NULL) {
2904 		free(ifma, M_IFMADDR);
2905 		return (NULL);
2906 	}
2907 	bcopy(sa, dupsa, sa->sa_len);
2908 	ifma->ifma_addr = dupsa;
2909 
2910 	ifma->ifma_ifp = ifp;
2911 	ifma->ifma_refcount = 1;
2912 	ifma->ifma_protospec = NULL;
2913 
2914 	if (llsa == NULL) {
2915 		ifma->ifma_lladdr = NULL;
2916 		return (ifma);
2917 	}
2918 
2919 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2920 	if (dupsa == NULL) {
2921 		free(ifma->ifma_addr, M_IFMADDR);
2922 		free(ifma, M_IFMADDR);
2923 		return (NULL);
2924 	}
2925 	bcopy(llsa, dupsa, llsa->sa_len);
2926 	ifma->ifma_lladdr = dupsa;
2927 
2928 	return (ifma);
2929 }
2930 
2931 /*
2932  * if_freemulti: free ifmultiaddr structure and possibly attached related
2933  * addresses.  The caller is responsible for implementing reference
2934  * counting, notifying the driver, handling routing messages, and releasing
2935  * any dependent link layer state.
2936  */
2937 static void
2938 if_freemulti(struct ifmultiaddr *ifma)
2939 {
2940 
2941 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2942 	    ifma->ifma_refcount));
2943 
2944 	if (ifma->ifma_lladdr != NULL)
2945 		free(ifma->ifma_lladdr, M_IFMADDR);
2946 	free(ifma->ifma_addr, M_IFMADDR);
2947 	free(ifma, M_IFMADDR);
2948 }
2949 
2950 /*
2951  * Register an additional multicast address with a network interface.
2952  *
2953  * - If the address is already present, bump the reference count on the
2954  *   address and return.
2955  * - If the address is not link-layer, look up a link layer address.
2956  * - Allocate address structures for one or both addresses, and attach to the
2957  *   multicast address list on the interface.  If automatically adding a link
2958  *   layer address, the protocol address will own a reference to the link
2959  *   layer address, to be freed when it is freed.
2960  * - Notify the network device driver of an addition to the multicast address
2961  *   list.
2962  *
2963  * 'sa' points to caller-owned memory with the desired multicast address.
2964  *
2965  * 'retifma' will be used to return a pointer to the resulting multicast
2966  * address reference, if desired.
2967  */
2968 int
2969 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2970     struct ifmultiaddr **retifma)
2971 {
2972 	struct ifmultiaddr *ifma, *ll_ifma;
2973 	struct sockaddr *llsa;
2974 	struct sockaddr_dl sdl;
2975 	int error;
2976 
2977 	/*
2978 	 * If the address is already present, return a new reference to it;
2979 	 * otherwise, allocate storage and set up a new address.
2980 	 */
2981 	IF_ADDR_WLOCK(ifp);
2982 	ifma = if_findmulti(ifp, sa);
2983 	if (ifma != NULL) {
2984 		ifma->ifma_refcount++;
2985 		if (retifma != NULL)
2986 			*retifma = ifma;
2987 		IF_ADDR_WUNLOCK(ifp);
2988 		return (0);
2989 	}
2990 
2991 	/*
2992 	 * The address isn't already present; resolve the protocol address
2993 	 * into a link layer address, and then look that up, bump its
2994 	 * refcount or allocate an ifma for that also.
2995 	 * Most link layer resolving functions returns address data which
2996 	 * fits inside default sockaddr_dl structure. However callback
2997 	 * can allocate another sockaddr structure, in that case we need to
2998 	 * free it later.
2999 	 */
3000 	llsa = NULL;
3001 	ll_ifma = NULL;
3002 	if (ifp->if_resolvemulti != NULL) {
3003 		/* Provide called function with buffer size information */
3004 		sdl.sdl_len = sizeof(sdl);
3005 		llsa = (struct sockaddr *)&sdl;
3006 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
3007 		if (error)
3008 			goto unlock_out;
3009 	}
3010 
3011 	/*
3012 	 * Allocate the new address.  Don't hook it up yet, as we may also
3013 	 * need to allocate a link layer multicast address.
3014 	 */
3015 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3016 	if (ifma == NULL) {
3017 		error = ENOMEM;
3018 		goto free_llsa_out;
3019 	}
3020 
3021 	/*
3022 	 * If a link layer address is found, we'll need to see if it's
3023 	 * already present in the address list, or allocate is as well.
3024 	 * When this block finishes, the link layer address will be on the
3025 	 * list.
3026 	 */
3027 	if (llsa != NULL) {
3028 		ll_ifma = if_findmulti(ifp, llsa);
3029 		if (ll_ifma == NULL) {
3030 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3031 			if (ll_ifma == NULL) {
3032 				--ifma->ifma_refcount;
3033 				if_freemulti(ifma);
3034 				error = ENOMEM;
3035 				goto free_llsa_out;
3036 			}
3037 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3038 			    ifma_link);
3039 		} else
3040 			ll_ifma->ifma_refcount++;
3041 		ifma->ifma_llifma = ll_ifma;
3042 	}
3043 
3044 	/*
3045 	 * We now have a new multicast address, ifma, and possibly a new or
3046 	 * referenced link layer address.  Add the primary address to the
3047 	 * ifnet address list.
3048 	 */
3049 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3050 
3051 	if (retifma != NULL)
3052 		*retifma = ifma;
3053 
3054 	/*
3055 	 * Must generate the message while holding the lock so that 'ifma'
3056 	 * pointer is still valid.
3057 	 */
3058 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3059 	IF_ADDR_WUNLOCK(ifp);
3060 
3061 	/*
3062 	 * We are certain we have added something, so call down to the
3063 	 * interface to let them know about it.
3064 	 */
3065 	if (ifp->if_ioctl != NULL) {
3066 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3067 	}
3068 
3069 	if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl))
3070 		link_free_sdl(llsa);
3071 
3072 	return (0);
3073 
3074 free_llsa_out:
3075 	if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl))
3076 		link_free_sdl(llsa);
3077 
3078 unlock_out:
3079 	IF_ADDR_WUNLOCK(ifp);
3080 	return (error);
3081 }
3082 
3083 /*
3084  * Delete a multicast group membership by network-layer group address.
3085  *
3086  * Returns ENOENT if the entry could not be found. If ifp no longer
3087  * exists, results are undefined. This entry point should only be used
3088  * from subsystems which do appropriate locking to hold ifp for the
3089  * duration of the call.
3090  * Network-layer protocol domains must use if_delmulti_ifma().
3091  */
3092 int
3093 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3094 {
3095 	struct ifmultiaddr *ifma;
3096 	int lastref;
3097 #ifdef INVARIANTS
3098 	struct ifnet *oifp;
3099 
3100 	IFNET_RLOCK_NOSLEEP();
3101 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3102 		if (ifp == oifp)
3103 			break;
3104 	if (ifp != oifp)
3105 		ifp = NULL;
3106 	IFNET_RUNLOCK_NOSLEEP();
3107 
3108 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3109 #endif
3110 	if (ifp == NULL)
3111 		return (ENOENT);
3112 
3113 	IF_ADDR_WLOCK(ifp);
3114 	lastref = 0;
3115 	ifma = if_findmulti(ifp, sa);
3116 	if (ifma != NULL)
3117 		lastref = if_delmulti_locked(ifp, ifma, 0);
3118 	IF_ADDR_WUNLOCK(ifp);
3119 
3120 	if (ifma == NULL)
3121 		return (ENOENT);
3122 
3123 	if (lastref && ifp->if_ioctl != NULL) {
3124 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3125 	}
3126 
3127 	return (0);
3128 }
3129 
3130 /*
3131  * Delete all multicast group membership for an interface.
3132  * Should be used to quickly flush all multicast filters.
3133  */
3134 void
3135 if_delallmulti(struct ifnet *ifp)
3136 {
3137 	struct ifmultiaddr *ifma;
3138 	struct ifmultiaddr *next;
3139 
3140 	IF_ADDR_WLOCK(ifp);
3141 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3142 		if_delmulti_locked(ifp, ifma, 0);
3143 	IF_ADDR_WUNLOCK(ifp);
3144 }
3145 
3146 /*
3147  * Delete a multicast group membership by group membership pointer.
3148  * Network-layer protocol domains must use this routine.
3149  *
3150  * It is safe to call this routine if the ifp disappeared.
3151  */
3152 void
3153 if_delmulti_ifma(struct ifmultiaddr *ifma)
3154 {
3155 	struct ifnet *ifp;
3156 	int lastref;
3157 
3158 	ifp = ifma->ifma_ifp;
3159 #ifdef DIAGNOSTIC
3160 	if (ifp == NULL) {
3161 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3162 	} else {
3163 		struct ifnet *oifp;
3164 
3165 		IFNET_RLOCK_NOSLEEP();
3166 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3167 			if (ifp == oifp)
3168 				break;
3169 		if (ifp != oifp) {
3170 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3171 			ifp = NULL;
3172 		}
3173 		IFNET_RUNLOCK_NOSLEEP();
3174 	}
3175 #endif
3176 	/*
3177 	 * If and only if the ifnet instance exists: Acquire the address lock.
3178 	 */
3179 	if (ifp != NULL)
3180 		IF_ADDR_WLOCK(ifp);
3181 
3182 	lastref = if_delmulti_locked(ifp, ifma, 0);
3183 
3184 	if (ifp != NULL) {
3185 		/*
3186 		 * If and only if the ifnet instance exists:
3187 		 *  Release the address lock.
3188 		 *  If the group was left: update the hardware hash filter.
3189 		 */
3190 		IF_ADDR_WUNLOCK(ifp);
3191 		if (lastref && ifp->if_ioctl != NULL) {
3192 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3193 		}
3194 	}
3195 }
3196 
3197 /*
3198  * Perform deletion of network-layer and/or link-layer multicast address.
3199  *
3200  * Return 0 if the reference count was decremented.
3201  * Return 1 if the final reference was released, indicating that the
3202  * hardware hash filter should be reprogrammed.
3203  */
3204 static int
3205 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3206 {
3207 	struct ifmultiaddr *ll_ifma;
3208 
3209 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3210 		KASSERT(ifma->ifma_ifp == ifp,
3211 		    ("%s: inconsistent ifp %p", __func__, ifp));
3212 		IF_ADDR_WLOCK_ASSERT(ifp);
3213 	}
3214 
3215 	ifp = ifma->ifma_ifp;
3216 
3217 	/*
3218 	 * If the ifnet is detaching, null out references to ifnet,
3219 	 * so that upper protocol layers will notice, and not attempt
3220 	 * to obtain locks for an ifnet which no longer exists. The
3221 	 * routing socket announcement must happen before the ifnet
3222 	 * instance is detached from the system.
3223 	 */
3224 	if (detaching) {
3225 #ifdef DIAGNOSTIC
3226 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3227 #endif
3228 		/*
3229 		 * ifp may already be nulled out if we are being reentered
3230 		 * to delete the ll_ifma.
3231 		 */
3232 		if (ifp != NULL) {
3233 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3234 			ifma->ifma_ifp = NULL;
3235 		}
3236 	}
3237 
3238 	if (--ifma->ifma_refcount > 0)
3239 		return 0;
3240 
3241 	/*
3242 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3243 	 * have been associated with it. Release it first if so.
3244 	 */
3245 	ll_ifma = ifma->ifma_llifma;
3246 	if (ll_ifma != NULL) {
3247 		KASSERT(ifma->ifma_lladdr != NULL,
3248 		    ("%s: llifma w/o lladdr", __func__));
3249 		if (detaching)
3250 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3251 		if (--ll_ifma->ifma_refcount == 0) {
3252 			if (ifp != NULL) {
3253 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3254 				    ifma_link);
3255 			}
3256 			if_freemulti(ll_ifma);
3257 		}
3258 	}
3259 
3260 	if (ifp != NULL)
3261 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3262 
3263 	if_freemulti(ifma);
3264 
3265 	/*
3266 	 * The last reference to this instance of struct ifmultiaddr
3267 	 * was released; the hardware should be notified of this change.
3268 	 */
3269 	return 1;
3270 }
3271 
3272 /*
3273  * Set the link layer address on an interface.
3274  *
3275  * At this time we only support certain types of interfaces,
3276  * and we don't allow the length of the address to change.
3277  */
3278 int
3279 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3280 {
3281 	struct sockaddr_dl *sdl;
3282 	struct ifaddr *ifa;
3283 	struct ifreq ifr;
3284 
3285 	IF_ADDR_RLOCK(ifp);
3286 	ifa = ifp->if_addr;
3287 	if (ifa == NULL) {
3288 		IF_ADDR_RUNLOCK(ifp);
3289 		return (EINVAL);
3290 	}
3291 	ifa_ref(ifa);
3292 	IF_ADDR_RUNLOCK(ifp);
3293 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3294 	if (sdl == NULL) {
3295 		ifa_free(ifa);
3296 		return (EINVAL);
3297 	}
3298 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3299 		ifa_free(ifa);
3300 		return (EINVAL);
3301 	}
3302 	switch (ifp->if_type) {
3303 	case IFT_ETHER:
3304 	case IFT_FDDI:
3305 	case IFT_XETHER:
3306 	case IFT_ISO88025:
3307 	case IFT_L2VLAN:
3308 	case IFT_BRIDGE:
3309 	case IFT_ARCNET:
3310 	case IFT_IEEE8023ADLAG:
3311 	case IFT_IEEE80211:
3312 		bcopy(lladdr, LLADDR(sdl), len);
3313 		ifa_free(ifa);
3314 		break;
3315 	default:
3316 		ifa_free(ifa);
3317 		return (ENODEV);
3318 	}
3319 
3320 	/*
3321 	 * If the interface is already up, we need
3322 	 * to re-init it in order to reprogram its
3323 	 * address filter.
3324 	 */
3325 	if ((ifp->if_flags & IFF_UP) != 0) {
3326 		if (ifp->if_ioctl) {
3327 			ifp->if_flags &= ~IFF_UP;
3328 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3329 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3330 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3331 			ifp->if_flags |= IFF_UP;
3332 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3333 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3334 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3335 		}
3336 #ifdef INET
3337 		/*
3338 		 * Also send gratuitous ARPs to notify other nodes about
3339 		 * the address change.
3340 		 */
3341 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3342 			if (ifa->ifa_addr->sa_family == AF_INET)
3343 				arp_ifinit(ifp, ifa);
3344 		}
3345 #endif
3346 	}
3347 	return (0);
3348 }
3349 
3350 /*
3351  * The name argument must be a pointer to storage which will last as
3352  * long as the interface does.  For physical devices, the result of
3353  * device_get_name(dev) is a good choice and for pseudo-devices a
3354  * static string works well.
3355  */
3356 void
3357 if_initname(struct ifnet *ifp, const char *name, int unit)
3358 {
3359 	ifp->if_dname = name;
3360 	ifp->if_dunit = unit;
3361 	if (unit != IF_DUNIT_NONE)
3362 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3363 	else
3364 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3365 }
3366 
3367 int
3368 if_printf(struct ifnet *ifp, const char * fmt, ...)
3369 {
3370 	va_list ap;
3371 	int retval;
3372 
3373 	retval = printf("%s: ", ifp->if_xname);
3374 	va_start(ap, fmt);
3375 	retval += vprintf(fmt, ap);
3376 	va_end(ap);
3377 	return (retval);
3378 }
3379 
3380 void
3381 if_start(struct ifnet *ifp)
3382 {
3383 
3384 	(*(ifp)->if_start)(ifp);
3385 }
3386 
3387 /*
3388  * Backwards compatibility interface for drivers
3389  * that have not implemented it
3390  */
3391 static int
3392 if_transmit(struct ifnet *ifp, struct mbuf *m)
3393 {
3394 	int error;
3395 
3396 	IFQ_HANDOFF(ifp, m, error);
3397 	return (error);
3398 }
3399 
3400 int
3401 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3402 {
3403 	int active = 0;
3404 
3405 	IF_LOCK(ifq);
3406 	if (_IF_QFULL(ifq)) {
3407 		_IF_DROP(ifq);
3408 		IF_UNLOCK(ifq);
3409 		m_freem(m);
3410 		return (0);
3411 	}
3412 	if (ifp != NULL) {
3413 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3414 		if (m->m_flags & (M_BCAST|M_MCAST))
3415 			ifp->if_omcasts++;
3416 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3417 	}
3418 	_IF_ENQUEUE(ifq, m);
3419 	IF_UNLOCK(ifq);
3420 	if (ifp != NULL && !active)
3421 		(*(ifp)->if_start)(ifp);
3422 	return (1);
3423 }
3424 
3425 void
3426 if_register_com_alloc(u_char type,
3427     if_com_alloc_t *a, if_com_free_t *f)
3428 {
3429 
3430 	KASSERT(if_com_alloc[type] == NULL,
3431 	    ("if_register_com_alloc: %d already registered", type));
3432 	KASSERT(if_com_free[type] == NULL,
3433 	    ("if_register_com_alloc: %d free already registered", type));
3434 
3435 	if_com_alloc[type] = a;
3436 	if_com_free[type] = f;
3437 }
3438 
3439 void
3440 if_deregister_com_alloc(u_char type)
3441 {
3442 
3443 	KASSERT(if_com_alloc[type] != NULL,
3444 	    ("if_deregister_com_alloc: %d not registered", type));
3445 	KASSERT(if_com_free[type] != NULL,
3446 	    ("if_deregister_com_alloc: %d free not registered", type));
3447 	if_com_alloc[type] = NULL;
3448 	if_com_free[type] = NULL;
3449 }
3450