xref: /freebsd/sys/net/if.c (revision 8a166cafe0965f6bd72cd3d2f5372704f05cb5e8)
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 /*XXX*/
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip_carp.h>
81 #ifdef INET6
82 #include <netinet6/in6_var.h>
83 #include <netinet6/in6_ifattach.h>
84 #endif
85 #endif
86 #ifdef INET
87 #include <netinet/if_ether.h>
88 #endif
89 
90 #include <security/mac/mac_framework.h>
91 
92 #ifdef COMPAT_FREEBSD32
93 #include <sys/mount.h>
94 #include <compat/freebsd32/freebsd32.h>
95 #endif
96 
97 struct ifindex_entry {
98 	struct  ifnet *ife_ifnet;
99 };
100 
101 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
102 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
103 
104 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
105 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
106     &ifqmaxlen, 0, "max send queue size");
107 
108 /* Log link state change events */
109 static int log_link_state_change = 1;
110 
111 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
112 	&log_link_state_change, 0,
113 	"log interface link state change events");
114 
115 /* Interface description */
116 static unsigned int ifdescr_maxlen = 1024;
117 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
118 	&ifdescr_maxlen, 0,
119 	"administrative maximum length for interface description");
120 
121 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
122 
123 /* global sx for non-critical path ifdescr */
124 static struct sx ifdescr_sx;
125 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
126 
127 void	(*bridge_linkstate_p)(struct ifnet *ifp);
128 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
129 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
130 /* These are external hooks for CARP. */
131 void	(*carp_linkstate_p)(struct ifnet *ifp);
132 void	(*carp_demote_adj_p)(int, char *);
133 int	(*carp_master_p)(struct ifaddr *);
134 #if defined(INET) || defined(INET6)
135 int	(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
136 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
137     struct sockaddr *sa);
138 int	(*carp_ioctl_p)(struct ifreq *, u_long, struct thread *);
139 int	(*carp_attach_p)(struct ifaddr *, int);
140 void	(*carp_detach_p)(struct ifaddr *);
141 #endif
142 #ifdef INET
143 int	(*carp_iamatch_p)(struct ifaddr *, uint8_t **);
144 #endif
145 #ifdef INET6
146 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
147 caddr_t	(*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
148     const struct in6_addr *taddr);
149 #endif
150 
151 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
152 
153 /*
154  * XXX: Style; these should be sorted alphabetically, and unprototyped
155  * static functions should be prototyped. Currently they are sorted by
156  * declaration order.
157  */
158 static void	if_attachdomain(void *);
159 static void	if_attachdomain1(struct ifnet *);
160 static int	ifconf(u_long, caddr_t);
161 static void	if_freemulti(struct ifmultiaddr *);
162 static void	if_init(void *);
163 static void	if_grow(void);
164 static void	if_route(struct ifnet *, int flag, int fam);
165 static int	if_setflag(struct ifnet *, int, int, int *, int);
166 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
167 static void	if_unroute(struct ifnet *, int flag, int fam);
168 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
169 static int	if_rtdel(struct radix_node *, void *);
170 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
171 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
172 static void	do_link_state_change(void *, int);
173 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
174 static int	if_getgroupmembers(struct ifgroupreq *);
175 static void	if_delgroups(struct ifnet *);
176 static void	if_attach_internal(struct ifnet *, int);
177 static void	if_detach_internal(struct ifnet *, int);
178 
179 #ifdef INET6
180 /*
181  * XXX: declare here to avoid to include many inet6 related files..
182  * should be more generalized?
183  */
184 extern void	nd6_setmtu(struct ifnet *);
185 #endif
186 
187 VNET_DEFINE(int, if_index);
188 int	ifqmaxlen = IFQ_MAXLEN;
189 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
190 VNET_DEFINE(struct ifgrouphead, ifg_head);
191 
192 static VNET_DEFINE(int, if_indexlim) = 8;
193 
194 /* Table of ifnet by index. */
195 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
196 
197 #define	V_if_indexlim		VNET(if_indexlim)
198 #define	V_ifindex_table		VNET(ifindex_table)
199 
200 /*
201  * The global network interface list (V_ifnet) and related state (such as
202  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
203  * an rwlock.  Either may be acquired shared to stablize the list, but both
204  * must be acquired writable to modify the list.  This model allows us to
205  * both stablize the interface list during interrupt thread processing, but
206  * also to stablize it over long-running ioctls, without introducing priority
207  * inversions and deadlocks.
208  */
209 struct rwlock ifnet_rwlock;
210 struct sx ifnet_sxlock;
211 
212 /*
213  * The allocation of network interfaces is a rather non-atomic affair; we
214  * need to select an index before we are ready to expose the interface for
215  * use, so will use this pointer value to indicate reservation.
216  */
217 #define	IFNET_HOLD	(void *)(uintptr_t)(-1)
218 
219 static	if_com_alloc_t *if_com_alloc[256];
220 static	if_com_free_t *if_com_free[256];
221 
222 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
223 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
224 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
225 
226 struct ifnet *
227 ifnet_byindex_locked(u_short idx)
228 {
229 
230 	if (idx > V_if_index)
231 		return (NULL);
232 	if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
233 		return (NULL);
234 	return (V_ifindex_table[idx].ife_ifnet);
235 }
236 
237 struct ifnet *
238 ifnet_byindex(u_short idx)
239 {
240 	struct ifnet *ifp;
241 
242 	IFNET_RLOCK_NOSLEEP();
243 	ifp = ifnet_byindex_locked(idx);
244 	IFNET_RUNLOCK_NOSLEEP();
245 	return (ifp);
246 }
247 
248 struct ifnet *
249 ifnet_byindex_ref(u_short idx)
250 {
251 	struct ifnet *ifp;
252 
253 	IFNET_RLOCK_NOSLEEP();
254 	ifp = ifnet_byindex_locked(idx);
255 	if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
256 		IFNET_RUNLOCK_NOSLEEP();
257 		return (NULL);
258 	}
259 	if_ref(ifp);
260 	IFNET_RUNLOCK_NOSLEEP();
261 	return (ifp);
262 }
263 
264 /*
265  * Allocate an ifindex array entry; return 0 on success or an error on
266  * failure.
267  */
268 static int
269 ifindex_alloc_locked(u_short *idxp)
270 {
271 	u_short idx;
272 
273 	IFNET_WLOCK_ASSERT();
274 
275 retry:
276 	/*
277 	 * Try to find an empty slot below V_if_index.  If we fail, take the
278 	 * next slot.
279 	 */
280 	for (idx = 1; idx <= V_if_index; idx++) {
281 		if (V_ifindex_table[idx].ife_ifnet == NULL)
282 			break;
283 	}
284 
285 	/* Catch if_index overflow. */
286 	if (idx < 1)
287 		return (ENOSPC);
288 	if (idx >= V_if_indexlim) {
289 		if_grow();
290 		goto retry;
291 	}
292 	if (idx > V_if_index)
293 		V_if_index = idx;
294 	*idxp = idx;
295 	return (0);
296 }
297 
298 static void
299 ifindex_free_locked(u_short idx)
300 {
301 
302 	IFNET_WLOCK_ASSERT();
303 
304 	V_ifindex_table[idx].ife_ifnet = NULL;
305 	while (V_if_index > 0 &&
306 	    V_ifindex_table[V_if_index].ife_ifnet == NULL)
307 		V_if_index--;
308 }
309 
310 static void
311 ifindex_free(u_short idx)
312 {
313 
314 	IFNET_WLOCK();
315 	ifindex_free_locked(idx);
316 	IFNET_WUNLOCK();
317 }
318 
319 static void
320 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
321 {
322 
323 	IFNET_WLOCK_ASSERT();
324 
325 	V_ifindex_table[idx].ife_ifnet = ifp;
326 }
327 
328 static void
329 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
330 {
331 
332 	IFNET_WLOCK();
333 	ifnet_setbyindex_locked(idx, ifp);
334 	IFNET_WUNLOCK();
335 }
336 
337 struct ifaddr *
338 ifaddr_byindex(u_short idx)
339 {
340 	struct ifaddr *ifa;
341 
342 	IFNET_RLOCK_NOSLEEP();
343 	ifa = ifnet_byindex_locked(idx)->if_addr;
344 	if (ifa != NULL)
345 		ifa_ref(ifa);
346 	IFNET_RUNLOCK_NOSLEEP();
347 	return (ifa);
348 }
349 
350 /*
351  * Network interface utility routines.
352  *
353  * Routines with ifa_ifwith* names take sockaddr *'s as
354  * parameters.
355  */
356 
357 static void
358 vnet_if_init(const void *unused __unused)
359 {
360 
361 	TAILQ_INIT(&V_ifnet);
362 	TAILQ_INIT(&V_ifg_head);
363 	IFNET_WLOCK();
364 	if_grow();				/* create initial table */
365 	IFNET_WUNLOCK();
366 	vnet_if_clone_init();
367 }
368 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
369     NULL);
370 
371 /* ARGSUSED*/
372 static void
373 if_init(void *dummy __unused)
374 {
375 
376 	IFNET_LOCK_INIT();
377 	if_clone_init();
378 }
379 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
380 
381 
382 #ifdef VIMAGE
383 static void
384 vnet_if_uninit(const void *unused __unused)
385 {
386 
387 	VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
388 	    "not empty", __func__, __LINE__, &V_ifnet));
389 	VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
390 	    "not empty", __func__, __LINE__, &V_ifg_head));
391 
392 	free((caddr_t)V_ifindex_table, M_IFNET);
393 }
394 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
395     vnet_if_uninit, NULL);
396 #endif
397 
398 static void
399 if_grow(void)
400 {
401 	int oldlim;
402 	u_int n;
403 	struct ifindex_entry *e;
404 
405 	IFNET_WLOCK_ASSERT();
406 	oldlim = V_if_indexlim;
407 	IFNET_WUNLOCK();
408 	n = (oldlim << 1) * sizeof(*e);
409 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
410 	IFNET_WLOCK();
411 	if (V_if_indexlim != oldlim) {
412 		free(e, M_IFNET);
413 		return;
414 	}
415 	if (V_ifindex_table != NULL) {
416 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
417 		free((caddr_t)V_ifindex_table, M_IFNET);
418 	}
419 	V_if_indexlim <<= 1;
420 	V_ifindex_table = e;
421 }
422 
423 /*
424  * Allocate a struct ifnet and an index for an interface.  A layer 2
425  * common structure will also be allocated if an allocation routine is
426  * registered for the passed type.
427  */
428 struct ifnet *
429 if_alloc(u_char type)
430 {
431 	struct ifnet *ifp;
432 	u_short idx;
433 
434 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
435 	IFNET_WLOCK();
436 	if (ifindex_alloc_locked(&idx) != 0) {
437 		IFNET_WUNLOCK();
438 		free(ifp, M_IFNET);
439 		return (NULL);
440 	}
441 	ifnet_setbyindex_locked(idx, IFNET_HOLD);
442 	IFNET_WUNLOCK();
443 	ifp->if_index = idx;
444 	ifp->if_type = type;
445 	ifp->if_alloctype = type;
446 	if (if_com_alloc[type] != NULL) {
447 		ifp->if_l2com = if_com_alloc[type](type, ifp);
448 		if (ifp->if_l2com == NULL) {
449 			free(ifp, M_IFNET);
450 			ifindex_free(idx);
451 			return (NULL);
452 		}
453 	}
454 
455 	IF_ADDR_LOCK_INIT(ifp);
456 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
457 	ifp->if_afdata_initialized = 0;
458 	IF_AFDATA_LOCK_INIT(ifp);
459 	TAILQ_INIT(&ifp->if_addrhead);
460 	TAILQ_INIT(&ifp->if_multiaddrs);
461 	TAILQ_INIT(&ifp->if_groups);
462 #ifdef MAC
463 	mac_ifnet_init(ifp);
464 #endif
465 	ifq_init(&ifp->if_snd, ifp);
466 
467 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
468 	ifnet_setbyindex(ifp->if_index, ifp);
469 	return (ifp);
470 }
471 
472 /*
473  * Do the actual work of freeing a struct ifnet, and layer 2 common
474  * structure.  This call is made when the last reference to an
475  * interface is released.
476  */
477 static void
478 if_free_internal(struct ifnet *ifp)
479 {
480 
481 	KASSERT((ifp->if_flags & IFF_DYING),
482 	    ("if_free_internal: interface not dying"));
483 
484 	if (if_com_free[ifp->if_alloctype] != NULL)
485 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
486 		    ifp->if_alloctype);
487 
488 #ifdef MAC
489 	mac_ifnet_destroy(ifp);
490 #endif /* MAC */
491 	if (ifp->if_description != NULL)
492 		free(ifp->if_description, M_IFDESCR);
493 	IF_AFDATA_DESTROY(ifp);
494 	IF_ADDR_LOCK_DESTROY(ifp);
495 	ifq_delete(&ifp->if_snd);
496 	free(ifp, M_IFNET);
497 }
498 
499 /*
500  * Deregister an interface and free the associated storage.
501  */
502 void
503 if_free(struct ifnet *ifp)
504 {
505 
506 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
507 
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 		return;
517 	if_free_internal(ifp);
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 = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
636 		ifa_init(ifa);
637 		sdl = (struct sockaddr_dl *)(ifa + 1);
638 		sdl->sdl_len = socksize;
639 		sdl->sdl_family = AF_LINK;
640 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
641 		sdl->sdl_nlen = namelen;
642 		sdl->sdl_index = ifp->if_index;
643 		sdl->sdl_type = ifp->if_type;
644 		ifp->if_addr = ifa;
645 		ifa->ifa_ifp = ifp;
646 		ifa->ifa_rtrequest = link_rtrequest;
647 		ifa->ifa_addr = (struct sockaddr *)sdl;
648 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
649 		ifa->ifa_netmask = (struct sockaddr *)sdl;
650 		sdl->sdl_len = masklen;
651 		while (namelen != 0)
652 			sdl->sdl_data[--namelen] = 0xff;
653 		TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
654 		/* Reliably crash if used uninitialized. */
655 		ifp->if_broadcastaddr = NULL;
656 	}
657 #ifdef VIMAGE
658 	else {
659 		/*
660 		 * Update the interface index in the link layer address
661 		 * of the interface.
662 		 */
663 		for (ifa = ifp->if_addr; ifa != NULL;
664 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
665 			if (ifa->ifa_addr->sa_family == AF_LINK) {
666 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
667 				sdl->sdl_index = ifp->if_index;
668 			}
669 		}
670 	}
671 #endif
672 
673 	IFNET_WLOCK();
674 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
675 #ifdef VIMAGE
676 	curvnet->vnet_ifcnt++;
677 #endif
678 	IFNET_WUNLOCK();
679 
680 	if (domain_init_status >= 2)
681 		if_attachdomain1(ifp);
682 
683 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
684 	if (IS_DEFAULT_VNET(curvnet))
685 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
686 
687 	/* Announce the interface. */
688 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
689 }
690 
691 static void
692 if_attachdomain(void *dummy)
693 {
694 	struct ifnet *ifp;
695 
696 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
697 		if_attachdomain1(ifp);
698 }
699 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
700     if_attachdomain, NULL);
701 
702 static void
703 if_attachdomain1(struct ifnet *ifp)
704 {
705 	struct domain *dp;
706 
707 	/*
708 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
709 	 * cannot lock ifp->if_afdata initialization, entirely.
710 	 */
711 	if (IF_AFDATA_TRYLOCK(ifp) == 0)
712 		return;
713 	if (ifp->if_afdata_initialized >= domain_init_status) {
714 		IF_AFDATA_UNLOCK(ifp);
715 		log(LOG_WARNING, "%s called more than once on %s\n",
716 		    __func__, ifp->if_xname);
717 		return;
718 	}
719 	ifp->if_afdata_initialized = domain_init_status;
720 	IF_AFDATA_UNLOCK(ifp);
721 
722 	/* address family dependent data region */
723 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
724 	for (dp = domains; dp; dp = dp->dom_next) {
725 		if (dp->dom_ifattach)
726 			ifp->if_afdata[dp->dom_family] =
727 			    (*dp->dom_ifattach)(ifp);
728 	}
729 }
730 
731 /*
732  * Remove any unicast or broadcast network addresses from an interface.
733  */
734 void
735 if_purgeaddrs(struct ifnet *ifp)
736 {
737 	struct ifaddr *ifa, *next;
738 
739 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
740 		if (ifa->ifa_addr->sa_family == AF_LINK)
741 			continue;
742 #ifdef INET
743 		/* XXX: Ugly!! ad hoc just for INET */
744 		if (ifa->ifa_addr->sa_family == AF_INET) {
745 			struct ifaliasreq ifr;
746 
747 			bzero(&ifr, sizeof(ifr));
748 			ifr.ifra_addr = *ifa->ifa_addr;
749 			if (ifa->ifa_dstaddr)
750 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
751 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
752 			    NULL) == 0)
753 				continue;
754 		}
755 #endif /* INET */
756 #ifdef INET6
757 		if (ifa->ifa_addr->sa_family == AF_INET6) {
758 			in6_purgeaddr(ifa);
759 			/* ifp_addrhead is already updated */
760 			continue;
761 		}
762 #endif /* INET6 */
763 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
764 		ifa_free(ifa);
765 	}
766 }
767 
768 /*
769  * Remove any multicast network addresses from an interface when an ifnet
770  * is going away.
771  */
772 static void
773 if_purgemaddrs(struct ifnet *ifp)
774 {
775 	struct ifmultiaddr *ifma;
776 	struct ifmultiaddr *next;
777 
778 	IF_ADDR_WLOCK(ifp);
779 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
780 		if_delmulti_locked(ifp, ifma, 1);
781 	IF_ADDR_WUNLOCK(ifp);
782 }
783 
784 /*
785  * Detach an interface, removing it from the list of "active" interfaces.
786  * If vmove flag is set on entry to if_detach_internal(), perform only a
787  * limited subset of cleanup tasks, given that we are moving an ifnet from
788  * one vnet to another, where it must be fully operational.
789  *
790  * XXXRW: There are some significant questions about event ordering, and
791  * how to prevent things from starting to use the interface during detach.
792  */
793 void
794 if_detach(struct ifnet *ifp)
795 {
796 
797 	if_detach_internal(ifp, 0);
798 }
799 
800 static void
801 if_detach_internal(struct ifnet *ifp, int vmove)
802 {
803 	struct ifaddr *ifa;
804 	struct radix_node_head	*rnh;
805 	int i, j;
806 	struct domain *dp;
807  	struct ifnet *iter;
808  	int found = 0;
809 
810 	IFNET_WLOCK();
811 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
812 		if (iter == ifp) {
813 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
814 			found = 1;
815 			break;
816 		}
817 #ifdef VIMAGE
818 	if (found)
819 		curvnet->vnet_ifcnt--;
820 #endif
821 	IFNET_WUNLOCK();
822 	if (!found) {
823 		if (vmove)
824 			panic("%s: ifp=%p not on the ifnet tailq %p",
825 			    __func__, ifp, &V_ifnet);
826 		else
827 			return; /* XXX this should panic as well? */
828 	}
829 
830 	/*
831 	 * Remove/wait for pending events.
832 	 */
833 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
834 
835 	/*
836 	 * Remove routes and flush queues.
837 	 */
838 	if_down(ifp);
839 #ifdef ALTQ
840 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
841 		altq_disable(&ifp->if_snd);
842 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
843 		altq_detach(&ifp->if_snd);
844 #endif
845 
846 	if_purgeaddrs(ifp);
847 
848 #ifdef INET
849 	in_ifdetach(ifp);
850 #endif
851 
852 #ifdef INET6
853 	/*
854 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
855 	 * before removing routing entries below, since IPv6 interface direct
856 	 * routes are expected to be removed by the IPv6-specific kernel API.
857 	 * Otherwise, the kernel will detect some inconsistency and bark it.
858 	 */
859 	in6_ifdetach(ifp);
860 #endif
861 	if_purgemaddrs(ifp);
862 
863 	if (!vmove) {
864 		/*
865 		 * Prevent further calls into the device driver via ifnet.
866 		 */
867 		if_dead(ifp);
868 
869 		/*
870 		 * Remove link ifaddr pointer and maybe decrement if_index.
871 		 * Clean up all addresses.
872 		 */
873 		ifp->if_addr = NULL;
874 
875 		/* We can now free link ifaddr. */
876 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
877 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
878 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
879 			ifa_free(ifa);
880 		}
881 	}
882 
883 	/*
884 	 * Delete all remaining routes using this interface
885 	 * Unfortuneatly the only way to do this is to slog through
886 	 * the entire routing table looking for routes which point
887 	 * to this interface...oh well...
888 	 */
889 	for (i = 1; i <= AF_MAX; i++) {
890 		for (j = 0; j < rt_numfibs; j++) {
891 			rnh = rt_tables_get_rnh(j, i);
892 			if (rnh == NULL)
893 				continue;
894 			RADIX_NODE_HEAD_LOCK(rnh);
895 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
896 			RADIX_NODE_HEAD_UNLOCK(rnh);
897 		}
898 	}
899 
900 	/* Announce that the interface is gone. */
901 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
902 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
903 	if (IS_DEFAULT_VNET(curvnet))
904 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
905 	if_delgroups(ifp);
906 
907 	/*
908 	 * We cannot hold the lock over dom_ifdetach calls as they might
909 	 * sleep, for example trying to drain a callout, thus open up the
910 	 * theoretical race with re-attaching.
911 	 */
912 	IF_AFDATA_LOCK(ifp);
913 	i = ifp->if_afdata_initialized;
914 	ifp->if_afdata_initialized = 0;
915 	IF_AFDATA_UNLOCK(ifp);
916 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
917 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
918 			(*dp->dom_ifdetach)(ifp,
919 			    ifp->if_afdata[dp->dom_family]);
920 	}
921 }
922 
923 #ifdef VIMAGE
924 /*
925  * if_vmove() performs a limited version of if_detach() in current
926  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
927  * An attempt is made to shrink if_index in current vnet, find an
928  * unused if_index in target vnet and calls if_grow() if necessary,
929  * and finally find an unused if_xname for the target vnet.
930  */
931 void
932 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
933 {
934 	u_short idx;
935 
936 	/*
937 	 * Detach from current vnet, but preserve LLADDR info, do not
938 	 * mark as dead etc. so that the ifnet can be reattached later.
939 	 */
940 	if_detach_internal(ifp, 1);
941 
942 	/*
943 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
944 	 * the if_index for that vnet if possible.
945 	 *
946 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
947 	 * or we'd lock on one vnet and unlock on another.
948 	 */
949 	IFNET_WLOCK();
950 	ifindex_free_locked(ifp->if_index);
951 	IFNET_WUNLOCK();
952 
953 	/*
954 	 * Perform interface-specific reassignment tasks, if provided by
955 	 * the driver.
956 	 */
957 	if (ifp->if_reassign != NULL)
958 		ifp->if_reassign(ifp, new_vnet, NULL);
959 
960 	/*
961 	 * Switch to the context of the target vnet.
962 	 */
963 	CURVNET_SET_QUIET(new_vnet);
964 
965 	IFNET_WLOCK();
966 	if (ifindex_alloc_locked(&idx) != 0) {
967 		IFNET_WUNLOCK();
968 		panic("if_index overflow");
969 	}
970 	ifp->if_index = idx;
971 	ifnet_setbyindex_locked(ifp->if_index, ifp);
972 	IFNET_WUNLOCK();
973 
974 	if_attach_internal(ifp, 1);
975 
976 	CURVNET_RESTORE();
977 }
978 
979 /*
980  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
981  */
982 static int
983 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
984 {
985 	struct prison *pr;
986 	struct ifnet *difp;
987 
988 	/* Try to find the prison within our visibility. */
989 	sx_slock(&allprison_lock);
990 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
991 	sx_sunlock(&allprison_lock);
992 	if (pr == NULL)
993 		return (ENXIO);
994 	prison_hold_locked(pr);
995 	mtx_unlock(&pr->pr_mtx);
996 
997 	/* Do not try to move the iface from and to the same prison. */
998 	if (pr->pr_vnet == ifp->if_vnet) {
999 		prison_free(pr);
1000 		return (EEXIST);
1001 	}
1002 
1003 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1004 	/* XXX Lock interfaces to avoid races. */
1005 	CURVNET_SET_QUIET(pr->pr_vnet);
1006 	difp = ifunit(ifname);
1007 	CURVNET_RESTORE();
1008 	if (difp != NULL) {
1009 		prison_free(pr);
1010 		return (EEXIST);
1011 	}
1012 
1013 	/* Move the interface into the child jail/vnet. */
1014 	if_vmove(ifp, pr->pr_vnet);
1015 
1016 	/* Report the new if_xname back to the userland. */
1017 	sprintf(ifname, "%s", ifp->if_xname);
1018 
1019 	prison_free(pr);
1020 	return (0);
1021 }
1022 
1023 static int
1024 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1025 {
1026 	struct prison *pr;
1027 	struct vnet *vnet_dst;
1028 	struct ifnet *ifp;
1029 
1030 	/* Try to find the prison within our visibility. */
1031 	sx_slock(&allprison_lock);
1032 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1033 	sx_sunlock(&allprison_lock);
1034 	if (pr == NULL)
1035 		return (ENXIO);
1036 	prison_hold_locked(pr);
1037 	mtx_unlock(&pr->pr_mtx);
1038 
1039 	/* Make sure the named iface exists in the source prison/vnet. */
1040 	CURVNET_SET(pr->pr_vnet);
1041 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1042 	if (ifp == NULL) {
1043 		CURVNET_RESTORE();
1044 		prison_free(pr);
1045 		return (ENXIO);
1046 	}
1047 
1048 	/* Do not try to move the iface from and to the same prison. */
1049 	vnet_dst = TD_TO_VNET(td);
1050 	if (vnet_dst == ifp->if_vnet) {
1051 		CURVNET_RESTORE();
1052 		prison_free(pr);
1053 		return (EEXIST);
1054 	}
1055 
1056 	/* Get interface back from child jail/vnet. */
1057 	if_vmove(ifp, vnet_dst);
1058 	CURVNET_RESTORE();
1059 
1060 	/* Report the new if_xname back to the userland. */
1061 	sprintf(ifname, "%s", ifp->if_xname);
1062 
1063 	prison_free(pr);
1064 	return (0);
1065 }
1066 #endif /* VIMAGE */
1067 
1068 /*
1069  * Add a group to an interface
1070  */
1071 int
1072 if_addgroup(struct ifnet *ifp, const char *groupname)
1073 {
1074 	struct ifg_list		*ifgl;
1075 	struct ifg_group	*ifg = NULL;
1076 	struct ifg_member	*ifgm;
1077 	int 			 new = 0;
1078 
1079 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1080 	    groupname[strlen(groupname) - 1] <= '9')
1081 		return (EINVAL);
1082 
1083 	IFNET_WLOCK();
1084 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1085 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1086 			IFNET_WUNLOCK();
1087 			return (EEXIST);
1088 		}
1089 
1090 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1091 	    M_NOWAIT)) == NULL) {
1092 	    	IFNET_WUNLOCK();
1093 		return (ENOMEM);
1094 	}
1095 
1096 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1097 	    M_TEMP, M_NOWAIT)) == NULL) {
1098 		free(ifgl, M_TEMP);
1099 		IFNET_WUNLOCK();
1100 		return (ENOMEM);
1101 	}
1102 
1103 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1104 		if (!strcmp(ifg->ifg_group, groupname))
1105 			break;
1106 
1107 	if (ifg == NULL) {
1108 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1109 		    M_TEMP, M_NOWAIT)) == NULL) {
1110 			free(ifgl, M_TEMP);
1111 			free(ifgm, M_TEMP);
1112 			IFNET_WUNLOCK();
1113 			return (ENOMEM);
1114 		}
1115 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1116 		ifg->ifg_refcnt = 0;
1117 		TAILQ_INIT(&ifg->ifg_members);
1118 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1119 		new = 1;
1120 	}
1121 
1122 	ifg->ifg_refcnt++;
1123 	ifgl->ifgl_group = ifg;
1124 	ifgm->ifgm_ifp = ifp;
1125 
1126 	IF_ADDR_WLOCK(ifp);
1127 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1128 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1129 	IF_ADDR_WUNLOCK(ifp);
1130 
1131 	IFNET_WUNLOCK();
1132 
1133 	if (new)
1134 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1135 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1136 
1137 	return (0);
1138 }
1139 
1140 /*
1141  * Remove a group from an interface
1142  */
1143 int
1144 if_delgroup(struct ifnet *ifp, const char *groupname)
1145 {
1146 	struct ifg_list		*ifgl;
1147 	struct ifg_member	*ifgm;
1148 
1149 	IFNET_WLOCK();
1150 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1151 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1152 			break;
1153 	if (ifgl == NULL) {
1154 		IFNET_WUNLOCK();
1155 		return (ENOENT);
1156 	}
1157 
1158 	IF_ADDR_WLOCK(ifp);
1159 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1160 	IF_ADDR_WUNLOCK(ifp);
1161 
1162 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1163 		if (ifgm->ifgm_ifp == ifp)
1164 			break;
1165 
1166 	if (ifgm != NULL) {
1167 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1168 		free(ifgm, M_TEMP);
1169 	}
1170 
1171 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1172 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1173 		IFNET_WUNLOCK();
1174 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1175 		free(ifgl->ifgl_group, M_TEMP);
1176 	} else
1177 		IFNET_WUNLOCK();
1178 
1179 	free(ifgl, M_TEMP);
1180 
1181 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1182 
1183 	return (0);
1184 }
1185 
1186 /*
1187  * Remove an interface from all groups
1188  */
1189 static void
1190 if_delgroups(struct ifnet *ifp)
1191 {
1192 	struct ifg_list		*ifgl;
1193 	struct ifg_member	*ifgm;
1194 	char groupname[IFNAMSIZ];
1195 
1196 	IFNET_WLOCK();
1197 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1198 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1199 
1200 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1201 
1202 		IF_ADDR_WLOCK(ifp);
1203 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1204 		IF_ADDR_WUNLOCK(ifp);
1205 
1206 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1207 			if (ifgm->ifgm_ifp == ifp)
1208 				break;
1209 
1210 		if (ifgm != NULL) {
1211 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1212 			    ifgm_next);
1213 			free(ifgm, M_TEMP);
1214 		}
1215 
1216 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1217 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1218 			IFNET_WUNLOCK();
1219 			EVENTHANDLER_INVOKE(group_detach_event,
1220 			    ifgl->ifgl_group);
1221 			free(ifgl->ifgl_group, M_TEMP);
1222 		} else
1223 			IFNET_WUNLOCK();
1224 
1225 		free(ifgl, M_TEMP);
1226 
1227 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1228 
1229 		IFNET_WLOCK();
1230 	}
1231 	IFNET_WUNLOCK();
1232 }
1233 
1234 /*
1235  * Stores all groups from an interface in memory pointed
1236  * to by data
1237  */
1238 static int
1239 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1240 {
1241 	int			 len, error;
1242 	struct ifg_list		*ifgl;
1243 	struct ifg_req		 ifgrq, *ifgp;
1244 	struct ifgroupreq	*ifgr = data;
1245 
1246 	if (ifgr->ifgr_len == 0) {
1247 		IF_ADDR_RLOCK(ifp);
1248 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1249 			ifgr->ifgr_len += sizeof(struct ifg_req);
1250 		IF_ADDR_RUNLOCK(ifp);
1251 		return (0);
1252 	}
1253 
1254 	len = ifgr->ifgr_len;
1255 	ifgp = ifgr->ifgr_groups;
1256 	/* XXX: wire */
1257 	IF_ADDR_RLOCK(ifp);
1258 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1259 		if (len < sizeof(ifgrq)) {
1260 			IF_ADDR_RUNLOCK(ifp);
1261 			return (EINVAL);
1262 		}
1263 		bzero(&ifgrq, sizeof ifgrq);
1264 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1265 		    sizeof(ifgrq.ifgrq_group));
1266 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1267 		    	IF_ADDR_RUNLOCK(ifp);
1268 			return (error);
1269 		}
1270 		len -= sizeof(ifgrq);
1271 		ifgp++;
1272 	}
1273 	IF_ADDR_RUNLOCK(ifp);
1274 
1275 	return (0);
1276 }
1277 
1278 /*
1279  * Stores all members of a group in memory pointed to by data
1280  */
1281 static int
1282 if_getgroupmembers(struct ifgroupreq *data)
1283 {
1284 	struct ifgroupreq	*ifgr = data;
1285 	struct ifg_group	*ifg;
1286 	struct ifg_member	*ifgm;
1287 	struct ifg_req		 ifgrq, *ifgp;
1288 	int			 len, error;
1289 
1290 	IFNET_RLOCK();
1291 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1292 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1293 			break;
1294 	if (ifg == NULL) {
1295 		IFNET_RUNLOCK();
1296 		return (ENOENT);
1297 	}
1298 
1299 	if (ifgr->ifgr_len == 0) {
1300 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1301 			ifgr->ifgr_len += sizeof(ifgrq);
1302 		IFNET_RUNLOCK();
1303 		return (0);
1304 	}
1305 
1306 	len = ifgr->ifgr_len;
1307 	ifgp = ifgr->ifgr_groups;
1308 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1309 		if (len < sizeof(ifgrq)) {
1310 			IFNET_RUNLOCK();
1311 			return (EINVAL);
1312 		}
1313 		bzero(&ifgrq, sizeof ifgrq);
1314 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1315 		    sizeof(ifgrq.ifgrq_member));
1316 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1317 			IFNET_RUNLOCK();
1318 			return (error);
1319 		}
1320 		len -= sizeof(ifgrq);
1321 		ifgp++;
1322 	}
1323 	IFNET_RUNLOCK();
1324 
1325 	return (0);
1326 }
1327 
1328 /*
1329  * Delete Routes for a Network Interface
1330  *
1331  * Called for each routing entry via the rnh->rnh_walktree() call above
1332  * to delete all route entries referencing a detaching network interface.
1333  *
1334  * Arguments:
1335  *	rn	pointer to node in the routing table
1336  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1337  *
1338  * Returns:
1339  *	0	successful
1340  *	errno	failed - reason indicated
1341  *
1342  */
1343 static int
1344 if_rtdel(struct radix_node *rn, void *arg)
1345 {
1346 	struct rtentry	*rt = (struct rtentry *)rn;
1347 	struct ifnet	*ifp = arg;
1348 	int		err;
1349 
1350 	if (rt->rt_ifp == ifp) {
1351 
1352 		/*
1353 		 * Protect (sorta) against walktree recursion problems
1354 		 * with cloned routes
1355 		 */
1356 		if ((rt->rt_flags & RTF_UP) == 0)
1357 			return (0);
1358 
1359 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1360 				rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1361 				(struct rtentry **) NULL, rt->rt_fibnum);
1362 		if (err) {
1363 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1364 		}
1365 	}
1366 
1367 	return (0);
1368 }
1369 
1370 /*
1371  * Wrapper functions for struct ifnet address list locking macros.  These are
1372  * used by kernel modules to avoid encoding programming interface or binary
1373  * interface assumptions that may be violated when kernel-internal locking
1374  * approaches change.
1375  */
1376 void
1377 if_addr_rlock(struct ifnet *ifp)
1378 {
1379 
1380 	IF_ADDR_RLOCK(ifp);
1381 }
1382 
1383 void
1384 if_addr_runlock(struct ifnet *ifp)
1385 {
1386 
1387 	IF_ADDR_RUNLOCK(ifp);
1388 }
1389 
1390 void
1391 if_maddr_rlock(struct ifnet *ifp)
1392 {
1393 
1394 	IF_ADDR_RLOCK(ifp);
1395 }
1396 
1397 void
1398 if_maddr_runlock(struct ifnet *ifp)
1399 {
1400 
1401 	IF_ADDR_RUNLOCK(ifp);
1402 }
1403 
1404 /*
1405  * Initialization, destruction and refcounting functions for ifaddrs.
1406  */
1407 void
1408 ifa_init(struct ifaddr *ifa)
1409 {
1410 
1411 	mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1412 	refcount_init(&ifa->ifa_refcnt, 1);
1413 	ifa->if_data.ifi_datalen = sizeof(ifa->if_data);
1414 }
1415 
1416 void
1417 ifa_ref(struct ifaddr *ifa)
1418 {
1419 
1420 	refcount_acquire(&ifa->ifa_refcnt);
1421 }
1422 
1423 void
1424 ifa_free(struct ifaddr *ifa)
1425 {
1426 
1427 	if (refcount_release(&ifa->ifa_refcnt)) {
1428 		mtx_destroy(&ifa->ifa_mtx);
1429 		free(ifa, M_IFADDR);
1430 	}
1431 }
1432 
1433 int
1434 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1435 {
1436 	int error = 0;
1437 	struct rtentry *rt = NULL;
1438 	struct rt_addrinfo info;
1439 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1440 
1441 	bzero(&info, sizeof(info));
1442 	info.rti_ifp = V_loif;
1443 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1444 	info.rti_info[RTAX_DST] = ia;
1445 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1446 	error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1447 
1448 	if (error == 0 && rt != NULL) {
1449 		RT_LOCK(rt);
1450 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1451 			ifa->ifa_ifp->if_type;
1452 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1453 			ifa->ifa_ifp->if_index;
1454 		RT_REMREF(rt);
1455 		RT_UNLOCK(rt);
1456 	} else if (error != 0)
1457 		log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1458 
1459 	return (error);
1460 }
1461 
1462 int
1463 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1464 {
1465 	int error = 0;
1466 	struct rt_addrinfo info;
1467 	struct sockaddr_dl null_sdl;
1468 
1469 	bzero(&null_sdl, sizeof(null_sdl));
1470 	null_sdl.sdl_len = sizeof(null_sdl);
1471 	null_sdl.sdl_family = AF_LINK;
1472 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1473 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1474 	bzero(&info, sizeof(info));
1475 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1476 	info.rti_info[RTAX_DST] = ia;
1477 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1478 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1479 
1480 	if (error != 0)
1481 		log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1482 
1483 	return (error);
1484 }
1485 
1486 /*
1487  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1488  * structs used to represent other address families, it is necessary
1489  * to perform a different comparison.
1490  */
1491 
1492 #define	sa_equal(a1, a2)	\
1493 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1494 
1495 #define	sa_dl_equal(a1, a2)	\
1496 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1497 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1498 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1499 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1500 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1501 
1502 /*
1503  * Locate an interface based on a complete address.
1504  */
1505 /*ARGSUSED*/
1506 static struct ifaddr *
1507 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1508 {
1509 	struct ifnet *ifp;
1510 	struct ifaddr *ifa;
1511 
1512 	IFNET_RLOCK_NOSLEEP();
1513 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1514 		IF_ADDR_RLOCK(ifp);
1515 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1516 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1517 				continue;
1518 			if (sa_equal(addr, ifa->ifa_addr)) {
1519 				if (getref)
1520 					ifa_ref(ifa);
1521 				IF_ADDR_RUNLOCK(ifp);
1522 				goto done;
1523 			}
1524 			/* IP6 doesn't have broadcast */
1525 			if ((ifp->if_flags & IFF_BROADCAST) &&
1526 			    ifa->ifa_broadaddr &&
1527 			    ifa->ifa_broadaddr->sa_len != 0 &&
1528 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1529 				if (getref)
1530 					ifa_ref(ifa);
1531 				IF_ADDR_RUNLOCK(ifp);
1532 				goto done;
1533 			}
1534 		}
1535 		IF_ADDR_RUNLOCK(ifp);
1536 	}
1537 	ifa = NULL;
1538 done:
1539 	IFNET_RUNLOCK_NOSLEEP();
1540 	return (ifa);
1541 }
1542 
1543 struct ifaddr *
1544 ifa_ifwithaddr(struct sockaddr *addr)
1545 {
1546 
1547 	return (ifa_ifwithaddr_internal(addr, 1));
1548 }
1549 
1550 int
1551 ifa_ifwithaddr_check(struct sockaddr *addr)
1552 {
1553 
1554 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1555 }
1556 
1557 /*
1558  * Locate an interface based on the broadcast address.
1559  */
1560 /* ARGSUSED */
1561 struct ifaddr *
1562 ifa_ifwithbroadaddr(struct sockaddr *addr)
1563 {
1564 	struct ifnet *ifp;
1565 	struct ifaddr *ifa;
1566 
1567 	IFNET_RLOCK_NOSLEEP();
1568 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1569 		IF_ADDR_RLOCK(ifp);
1570 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1571 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1572 				continue;
1573 			if ((ifp->if_flags & IFF_BROADCAST) &&
1574 			    ifa->ifa_broadaddr &&
1575 			    ifa->ifa_broadaddr->sa_len != 0 &&
1576 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1577 				ifa_ref(ifa);
1578 				IF_ADDR_RUNLOCK(ifp);
1579 				goto done;
1580 			}
1581 		}
1582 		IF_ADDR_RUNLOCK(ifp);
1583 	}
1584 	ifa = NULL;
1585 done:
1586 	IFNET_RUNLOCK_NOSLEEP();
1587 	return (ifa);
1588 }
1589 
1590 /*
1591  * Locate the point to point interface with a given destination address.
1592  */
1593 /*ARGSUSED*/
1594 struct ifaddr *
1595 ifa_ifwithdstaddr(struct sockaddr *addr)
1596 {
1597 	struct ifnet *ifp;
1598 	struct ifaddr *ifa;
1599 
1600 	IFNET_RLOCK_NOSLEEP();
1601 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1602 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1603 			continue;
1604 		IF_ADDR_RLOCK(ifp);
1605 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1606 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1607 				continue;
1608 			if (ifa->ifa_dstaddr != NULL &&
1609 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1610 				ifa_ref(ifa);
1611 				IF_ADDR_RUNLOCK(ifp);
1612 				goto done;
1613 			}
1614 		}
1615 		IF_ADDR_RUNLOCK(ifp);
1616 	}
1617 	ifa = NULL;
1618 done:
1619 	IFNET_RUNLOCK_NOSLEEP();
1620 	return (ifa);
1621 }
1622 
1623 /*
1624  * Find an interface on a specific network.  If many, choice
1625  * is most specific found.
1626  */
1627 struct ifaddr *
1628 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1629 {
1630 	struct ifnet *ifp;
1631 	struct ifaddr *ifa;
1632 	struct ifaddr *ifa_maybe = NULL;
1633 	u_int af = addr->sa_family;
1634 	char *addr_data = addr->sa_data, *cplim;
1635 
1636 	/*
1637 	 * AF_LINK addresses can be looked up directly by their index number,
1638 	 * so do that if we can.
1639 	 */
1640 	if (af == AF_LINK) {
1641 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1642 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1643 		return (ifaddr_byindex(sdl->sdl_index));
1644 	}
1645 
1646 	/*
1647 	 * Scan though each interface, looking for ones that have addresses
1648 	 * in this address family.  Maintain a reference on ifa_maybe once
1649 	 * we find one, as we release the IF_ADDR_RLOCK() that kept it stable
1650 	 * when we move onto the next interface.
1651 	 */
1652 	IFNET_RLOCK_NOSLEEP();
1653 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1654 		IF_ADDR_RLOCK(ifp);
1655 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1656 			char *cp, *cp2, *cp3;
1657 
1658 			if (ifa->ifa_addr->sa_family != af)
1659 next:				continue;
1660 			if (af == AF_INET &&
1661 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1662 				/*
1663 				 * This is a bit broken as it doesn't
1664 				 * take into account that the remote end may
1665 				 * be a single node in the network we are
1666 				 * looking for.
1667 				 * The trouble is that we don't know the
1668 				 * netmask for the remote end.
1669 				 */
1670 				if (ifa->ifa_dstaddr != NULL &&
1671 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1672 					ifa_ref(ifa);
1673 					IF_ADDR_RUNLOCK(ifp);
1674 					goto done;
1675 				}
1676 			} else {
1677 				/*
1678 				 * if we have a special address handler,
1679 				 * then use it instead of the generic one.
1680 				 */
1681 				if (ifa->ifa_claim_addr) {
1682 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1683 						ifa_ref(ifa);
1684 						IF_ADDR_RUNLOCK(ifp);
1685 						goto done;
1686 					}
1687 					continue;
1688 				}
1689 
1690 				/*
1691 				 * Scan all the bits in the ifa's address.
1692 				 * If a bit dissagrees with what we are
1693 				 * looking for, mask it with the netmask
1694 				 * to see if it really matters.
1695 				 * (A byte at a time)
1696 				 */
1697 				if (ifa->ifa_netmask == 0)
1698 					continue;
1699 				cp = addr_data;
1700 				cp2 = ifa->ifa_addr->sa_data;
1701 				cp3 = ifa->ifa_netmask->sa_data;
1702 				cplim = ifa->ifa_netmask->sa_len
1703 					+ (char *)ifa->ifa_netmask;
1704 				while (cp3 < cplim)
1705 					if ((*cp++ ^ *cp2++) & *cp3++)
1706 						goto next; /* next address! */
1707 				/*
1708 				 * If the netmask of what we just found
1709 				 * is more specific than what we had before
1710 				 * (if we had one), or if the virtual status
1711 				 * of new prefix is better than of the old one,
1712 				 * then remember the new one before continuing
1713 				 * to search for an even better one.
1714 				 */
1715 				if (ifa_maybe == NULL ||
1716 				    ifa_preferred(ifa_maybe, ifa) ||
1717 				    rn_refines((caddr_t)ifa->ifa_netmask,
1718 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1719 					if (ifa_maybe != NULL)
1720 						ifa_free(ifa_maybe);
1721 					ifa_maybe = ifa;
1722 					ifa_ref(ifa_maybe);
1723 				}
1724 			}
1725 		}
1726 		IF_ADDR_RUNLOCK(ifp);
1727 	}
1728 	ifa = ifa_maybe;
1729 	ifa_maybe = NULL;
1730 done:
1731 	IFNET_RUNLOCK_NOSLEEP();
1732 	if (ifa_maybe != NULL)
1733 		ifa_free(ifa_maybe);
1734 	return (ifa);
1735 }
1736 
1737 /*
1738  * Find an interface address specific to an interface best matching
1739  * a given address.
1740  */
1741 struct ifaddr *
1742 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1743 {
1744 	struct ifaddr *ifa;
1745 	char *cp, *cp2, *cp3;
1746 	char *cplim;
1747 	struct ifaddr *ifa_maybe = NULL;
1748 	u_int af = addr->sa_family;
1749 
1750 	if (af >= AF_MAX)
1751 		return (NULL);
1752 	IF_ADDR_RLOCK(ifp);
1753 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1754 		if (ifa->ifa_addr->sa_family != af)
1755 			continue;
1756 		if (ifa_maybe == NULL)
1757 			ifa_maybe = ifa;
1758 		if (ifa->ifa_netmask == 0) {
1759 			if (sa_equal(addr, ifa->ifa_addr) ||
1760 			    (ifa->ifa_dstaddr &&
1761 			    sa_equal(addr, ifa->ifa_dstaddr)))
1762 				goto done;
1763 			continue;
1764 		}
1765 		if (ifp->if_flags & IFF_POINTOPOINT) {
1766 			if (sa_equal(addr, ifa->ifa_dstaddr))
1767 				goto done;
1768 		} else {
1769 			cp = addr->sa_data;
1770 			cp2 = ifa->ifa_addr->sa_data;
1771 			cp3 = ifa->ifa_netmask->sa_data;
1772 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1773 			for (; cp3 < cplim; cp3++)
1774 				if ((*cp++ ^ *cp2++) & *cp3)
1775 					break;
1776 			if (cp3 == cplim)
1777 				goto done;
1778 		}
1779 	}
1780 	ifa = ifa_maybe;
1781 done:
1782 	if (ifa != NULL)
1783 		ifa_ref(ifa);
1784 	IF_ADDR_RUNLOCK(ifp);
1785 	return (ifa);
1786 }
1787 
1788 /*
1789  * See whether new ifa is better than current one:
1790  * 1) A non-virtual one is preferred over virtual.
1791  * 2) A virtual in master state preferred over any other state.
1792  *
1793  * Used in several address selecting functions.
1794  */
1795 int
1796 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
1797 {
1798 
1799 	return (cur->ifa_carp && (!next->ifa_carp ||
1800 	    ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
1801 }
1802 
1803 #include <net/if_llatbl.h>
1804 
1805 /*
1806  * Default action when installing a route with a Link Level gateway.
1807  * Lookup an appropriate real ifa to point to.
1808  * This should be moved to /sys/net/link.c eventually.
1809  */
1810 static void
1811 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1812 {
1813 	struct ifaddr *ifa, *oifa;
1814 	struct sockaddr *dst;
1815 	struct ifnet *ifp;
1816 
1817 	RT_LOCK_ASSERT(rt);
1818 
1819 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1820 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1821 		return;
1822 	ifa = ifaof_ifpforaddr(dst, ifp);
1823 	if (ifa) {
1824 		oifa = rt->rt_ifa;
1825 		rt->rt_ifa = ifa;
1826 		ifa_free(oifa);
1827 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1828 			ifa->ifa_rtrequest(cmd, rt, info);
1829 	}
1830 }
1831 
1832 /*
1833  * Mark an interface down and notify protocols of
1834  * the transition.
1835  */
1836 static void
1837 if_unroute(struct ifnet *ifp, int flag, int fam)
1838 {
1839 	struct ifaddr *ifa;
1840 
1841 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1842 
1843 	ifp->if_flags &= ~flag;
1844 	getmicrotime(&ifp->if_lastchange);
1845 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1846 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1847 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1848 	ifp->if_qflush(ifp);
1849 
1850 	if (ifp->if_carp)
1851 		(*carp_linkstate_p)(ifp);
1852 	rt_ifmsg(ifp);
1853 }
1854 
1855 /*
1856  * Mark an interface up and notify protocols of
1857  * the transition.
1858  */
1859 static void
1860 if_route(struct ifnet *ifp, int flag, int fam)
1861 {
1862 	struct ifaddr *ifa;
1863 
1864 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1865 
1866 	ifp->if_flags |= flag;
1867 	getmicrotime(&ifp->if_lastchange);
1868 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1869 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1870 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1871 	if (ifp->if_carp)
1872 		(*carp_linkstate_p)(ifp);
1873 	rt_ifmsg(ifp);
1874 #ifdef INET6
1875 	in6_if_up(ifp);
1876 #endif
1877 }
1878 
1879 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1880 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1881 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1882 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1883 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
1884 int	(*vlan_setcookie_p)(struct ifnet *, void *);
1885 void	*(*vlan_cookie_p)(struct ifnet *);
1886 
1887 /*
1888  * Handle a change in the interface link state. To avoid LORs
1889  * between driver lock and upper layer locks, as well as possible
1890  * recursions, we post event to taskqueue, and all job
1891  * is done in static do_link_state_change().
1892  */
1893 void
1894 if_link_state_change(struct ifnet *ifp, int link_state)
1895 {
1896 	/* Return if state hasn't changed. */
1897 	if (ifp->if_link_state == link_state)
1898 		return;
1899 
1900 	ifp->if_link_state = link_state;
1901 
1902 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1903 }
1904 
1905 static void
1906 do_link_state_change(void *arg, int pending)
1907 {
1908 	struct ifnet *ifp = (struct ifnet *)arg;
1909 	int link_state = ifp->if_link_state;
1910 	CURVNET_SET(ifp->if_vnet);
1911 
1912 	/* Notify that the link state has changed. */
1913 	rt_ifmsg(ifp);
1914 	if (ifp->if_vlantrunk != NULL)
1915 		(*vlan_link_state_p)(ifp);
1916 
1917 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1918 	    IFP2AC(ifp)->ac_netgraph != NULL)
1919 		(*ng_ether_link_state_p)(ifp, link_state);
1920 	if (ifp->if_carp)
1921 		(*carp_linkstate_p)(ifp);
1922 	if (ifp->if_bridge)
1923 		(*bridge_linkstate_p)(ifp);
1924 	if (ifp->if_lagg)
1925 		(*lagg_linkstate_p)(ifp, link_state);
1926 
1927 	if (IS_DEFAULT_VNET(curvnet))
1928 		devctl_notify("IFNET", ifp->if_xname,
1929 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1930 		    NULL);
1931 	if (pending > 1)
1932 		if_printf(ifp, "%d link states coalesced\n", pending);
1933 	if (log_link_state_change)
1934 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1935 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1936 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
1937 	CURVNET_RESTORE();
1938 }
1939 
1940 /*
1941  * Mark an interface down and notify protocols of
1942  * the transition.
1943  */
1944 void
1945 if_down(struct ifnet *ifp)
1946 {
1947 
1948 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1949 }
1950 
1951 /*
1952  * Mark an interface up and notify protocols of
1953  * the transition.
1954  */
1955 void
1956 if_up(struct ifnet *ifp)
1957 {
1958 
1959 	if_route(ifp, IFF_UP, AF_UNSPEC);
1960 }
1961 
1962 /*
1963  * Flush an interface queue.
1964  */
1965 void
1966 if_qflush(struct ifnet *ifp)
1967 {
1968 	struct mbuf *m, *n;
1969 	struct ifaltq *ifq;
1970 
1971 	ifq = &ifp->if_snd;
1972 	IFQ_LOCK(ifq);
1973 #ifdef ALTQ
1974 	if (ALTQ_IS_ENABLED(ifq))
1975 		ALTQ_PURGE(ifq);
1976 #endif
1977 	n = ifq->ifq_head;
1978 	while ((m = n) != 0) {
1979 		n = m->m_act;
1980 		m_freem(m);
1981 	}
1982 	ifq->ifq_head = 0;
1983 	ifq->ifq_tail = 0;
1984 	ifq->ifq_len = 0;
1985 	IFQ_UNLOCK(ifq);
1986 }
1987 
1988 /*
1989  * Map interface name to interface structure pointer, with or without
1990  * returning a reference.
1991  */
1992 struct ifnet *
1993 ifunit_ref(const char *name)
1994 {
1995 	struct ifnet *ifp;
1996 
1997 	IFNET_RLOCK_NOSLEEP();
1998 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1999 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2000 		    !(ifp->if_flags & IFF_DYING))
2001 			break;
2002 	}
2003 	if (ifp != NULL)
2004 		if_ref(ifp);
2005 	IFNET_RUNLOCK_NOSLEEP();
2006 	return (ifp);
2007 }
2008 
2009 struct ifnet *
2010 ifunit(const char *name)
2011 {
2012 	struct ifnet *ifp;
2013 
2014 	IFNET_RLOCK_NOSLEEP();
2015 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2016 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2017 			break;
2018 	}
2019 	IFNET_RUNLOCK_NOSLEEP();
2020 	return (ifp);
2021 }
2022 
2023 /*
2024  * Hardware specific interface ioctls.
2025  */
2026 static int
2027 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2028 {
2029 	struct ifreq *ifr;
2030 	struct ifstat *ifs;
2031 	int error = 0;
2032 	int new_flags, temp_flags;
2033 	size_t namelen, onamelen;
2034 	size_t descrlen;
2035 	char *descrbuf, *odescrbuf;
2036 	char new_name[IFNAMSIZ];
2037 	struct ifaddr *ifa;
2038 	struct sockaddr_dl *sdl;
2039 
2040 	ifr = (struct ifreq *)data;
2041 	switch (cmd) {
2042 	case SIOCGIFINDEX:
2043 		ifr->ifr_index = ifp->if_index;
2044 		break;
2045 
2046 	case SIOCGIFFLAGS:
2047 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2048 		ifr->ifr_flags = temp_flags & 0xffff;
2049 		ifr->ifr_flagshigh = temp_flags >> 16;
2050 		break;
2051 
2052 	case SIOCGIFCAP:
2053 		ifr->ifr_reqcap = ifp->if_capabilities;
2054 		ifr->ifr_curcap = ifp->if_capenable;
2055 		break;
2056 
2057 #ifdef MAC
2058 	case SIOCGIFMAC:
2059 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2060 		break;
2061 #endif
2062 
2063 	case SIOCGIFMETRIC:
2064 		ifr->ifr_metric = ifp->if_metric;
2065 		break;
2066 
2067 	case SIOCGIFMTU:
2068 		ifr->ifr_mtu = ifp->if_mtu;
2069 		break;
2070 
2071 	case SIOCGIFPHYS:
2072 		ifr->ifr_phys = ifp->if_physical;
2073 		break;
2074 
2075 	case SIOCGIFDESCR:
2076 		error = 0;
2077 		sx_slock(&ifdescr_sx);
2078 		if (ifp->if_description == NULL)
2079 			error = ENOMSG;
2080 		else {
2081 			/* space for terminating nul */
2082 			descrlen = strlen(ifp->if_description) + 1;
2083 			if (ifr->ifr_buffer.length < descrlen)
2084 				ifr->ifr_buffer.buffer = NULL;
2085 			else
2086 				error = copyout(ifp->if_description,
2087 				    ifr->ifr_buffer.buffer, descrlen);
2088 			ifr->ifr_buffer.length = descrlen;
2089 		}
2090 		sx_sunlock(&ifdescr_sx);
2091 		break;
2092 
2093 	case SIOCSIFDESCR:
2094 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2095 		if (error)
2096 			return (error);
2097 
2098 		/*
2099 		 * Copy only (length-1) bytes to make sure that
2100 		 * if_description is always nul terminated.  The
2101 		 * length parameter is supposed to count the
2102 		 * terminating nul in.
2103 		 */
2104 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2105 			return (ENAMETOOLONG);
2106 		else if (ifr->ifr_buffer.length == 0)
2107 			descrbuf = NULL;
2108 		else {
2109 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2110 			    M_WAITOK | M_ZERO);
2111 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2112 			    ifr->ifr_buffer.length - 1);
2113 			if (error) {
2114 				free(descrbuf, M_IFDESCR);
2115 				break;
2116 			}
2117 		}
2118 
2119 		sx_xlock(&ifdescr_sx);
2120 		odescrbuf = ifp->if_description;
2121 		ifp->if_description = descrbuf;
2122 		sx_xunlock(&ifdescr_sx);
2123 
2124 		getmicrotime(&ifp->if_lastchange);
2125 		free(odescrbuf, M_IFDESCR);
2126 		break;
2127 
2128 	case SIOCGIFFIB:
2129 		ifr->ifr_fib = ifp->if_fib;
2130 		break;
2131 
2132 	case SIOCSIFFIB:
2133 		error = priv_check(td, PRIV_NET_SETIFFIB);
2134 		if (error)
2135 			return (error);
2136 		if (ifr->ifr_fib >= rt_numfibs)
2137 			return (EINVAL);
2138 
2139 		ifp->if_fib = ifr->ifr_fib;
2140 		break;
2141 
2142 	case SIOCSIFFLAGS:
2143 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2144 		if (error)
2145 			return (error);
2146 		/*
2147 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2148 		 * check, so we don't need special handling here yet.
2149 		 */
2150 		new_flags = (ifr->ifr_flags & 0xffff) |
2151 		    (ifr->ifr_flagshigh << 16);
2152 		if (ifp->if_flags & IFF_SMART) {
2153 			/* Smart drivers twiddle their own routes */
2154 		} else if (ifp->if_flags & IFF_UP &&
2155 		    (new_flags & IFF_UP) == 0) {
2156 			if_down(ifp);
2157 		} else if (new_flags & IFF_UP &&
2158 		    (ifp->if_flags & IFF_UP) == 0) {
2159 			if_up(ifp);
2160 		}
2161 		/* See if permanently promiscuous mode bit is about to flip */
2162 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2163 			if (new_flags & IFF_PPROMISC)
2164 				ifp->if_flags |= IFF_PROMISC;
2165 			else if (ifp->if_pcount == 0)
2166 				ifp->if_flags &= ~IFF_PROMISC;
2167 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2168 			    ifp->if_xname,
2169 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2170 		}
2171 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2172 			(new_flags &~ IFF_CANTCHANGE);
2173 		if (ifp->if_ioctl) {
2174 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2175 		}
2176 		getmicrotime(&ifp->if_lastchange);
2177 		break;
2178 
2179 	case SIOCSIFCAP:
2180 		error = priv_check(td, PRIV_NET_SETIFCAP);
2181 		if (error)
2182 			return (error);
2183 		if (ifp->if_ioctl == NULL)
2184 			return (EOPNOTSUPP);
2185 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2186 			return (EINVAL);
2187 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2188 		if (error == 0)
2189 			getmicrotime(&ifp->if_lastchange);
2190 		break;
2191 
2192 #ifdef MAC
2193 	case SIOCSIFMAC:
2194 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2195 		break;
2196 #endif
2197 
2198 	case SIOCSIFNAME:
2199 		error = priv_check(td, PRIV_NET_SETIFNAME);
2200 		if (error)
2201 			return (error);
2202 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2203 		if (error != 0)
2204 			return (error);
2205 		if (new_name[0] == '\0')
2206 			return (EINVAL);
2207 		if (ifunit(new_name) != NULL)
2208 			return (EEXIST);
2209 
2210 		/*
2211 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2212 		 * and there are numerous other races with the
2213 		 * ifunit() checks not being atomic with namespace
2214 		 * changes (renames, vmoves, if_attach, etc).
2215 		 */
2216 		ifp->if_flags |= IFF_RENAMING;
2217 
2218 		/* Announce the departure of the interface. */
2219 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2220 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2221 
2222 		log(LOG_INFO, "%s: changing name to '%s'\n",
2223 		    ifp->if_xname, new_name);
2224 
2225 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2226 		ifa = ifp->if_addr;
2227 		IFA_LOCK(ifa);
2228 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2229 		namelen = strlen(new_name);
2230 		onamelen = sdl->sdl_nlen;
2231 		/*
2232 		 * Move the address if needed.  This is safe because we
2233 		 * allocate space for a name of length IFNAMSIZ when we
2234 		 * create this in if_attach().
2235 		 */
2236 		if (namelen != onamelen) {
2237 			bcopy(sdl->sdl_data + onamelen,
2238 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2239 		}
2240 		bcopy(new_name, sdl->sdl_data, namelen);
2241 		sdl->sdl_nlen = namelen;
2242 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2243 		bzero(sdl->sdl_data, onamelen);
2244 		while (namelen != 0)
2245 			sdl->sdl_data[--namelen] = 0xff;
2246 		IFA_UNLOCK(ifa);
2247 
2248 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2249 		/* Announce the return of the interface. */
2250 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2251 
2252 		ifp->if_flags &= ~IFF_RENAMING;
2253 		break;
2254 
2255 #ifdef VIMAGE
2256 	case SIOCSIFVNET:
2257 		error = priv_check(td, PRIV_NET_SETIFVNET);
2258 		if (error)
2259 			return (error);
2260 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2261 		break;
2262 #endif
2263 
2264 	case SIOCSIFMETRIC:
2265 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2266 		if (error)
2267 			return (error);
2268 		ifp->if_metric = ifr->ifr_metric;
2269 		getmicrotime(&ifp->if_lastchange);
2270 		break;
2271 
2272 	case SIOCSIFPHYS:
2273 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2274 		if (error)
2275 			return (error);
2276 		if (ifp->if_ioctl == NULL)
2277 			return (EOPNOTSUPP);
2278 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2279 		if (error == 0)
2280 			getmicrotime(&ifp->if_lastchange);
2281 		break;
2282 
2283 	case SIOCSIFMTU:
2284 	{
2285 		u_long oldmtu = ifp->if_mtu;
2286 
2287 		error = priv_check(td, PRIV_NET_SETIFMTU);
2288 		if (error)
2289 			return (error);
2290 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2291 			return (EINVAL);
2292 		if (ifp->if_ioctl == NULL)
2293 			return (EOPNOTSUPP);
2294 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2295 		if (error == 0) {
2296 			getmicrotime(&ifp->if_lastchange);
2297 			rt_ifmsg(ifp);
2298 		}
2299 		/*
2300 		 * If the link MTU changed, do network layer specific procedure.
2301 		 */
2302 		if (ifp->if_mtu != oldmtu) {
2303 #ifdef INET6
2304 			nd6_setmtu(ifp);
2305 #endif
2306 		}
2307 		break;
2308 	}
2309 
2310 	case SIOCADDMULTI:
2311 	case SIOCDELMULTI:
2312 		if (cmd == SIOCADDMULTI)
2313 			error = priv_check(td, PRIV_NET_ADDMULTI);
2314 		else
2315 			error = priv_check(td, PRIV_NET_DELMULTI);
2316 		if (error)
2317 			return (error);
2318 
2319 		/* Don't allow group membership on non-multicast interfaces. */
2320 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2321 			return (EOPNOTSUPP);
2322 
2323 		/* Don't let users screw up protocols' entries. */
2324 		if (ifr->ifr_addr.sa_family != AF_LINK)
2325 			return (EINVAL);
2326 
2327 		if (cmd == SIOCADDMULTI) {
2328 			struct ifmultiaddr *ifma;
2329 
2330 			/*
2331 			 * Userland is only permitted to join groups once
2332 			 * via the if_addmulti() KPI, because it cannot hold
2333 			 * struct ifmultiaddr * between calls. It may also
2334 			 * lose a race while we check if the membership
2335 			 * already exists.
2336 			 */
2337 			IF_ADDR_RLOCK(ifp);
2338 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2339 			IF_ADDR_RUNLOCK(ifp);
2340 			if (ifma != NULL)
2341 				error = EADDRINUSE;
2342 			else
2343 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2344 		} else {
2345 			error = if_delmulti(ifp, &ifr->ifr_addr);
2346 		}
2347 		if (error == 0)
2348 			getmicrotime(&ifp->if_lastchange);
2349 		break;
2350 
2351 	case SIOCSIFPHYADDR:
2352 	case SIOCDIFPHYADDR:
2353 #ifdef INET6
2354 	case SIOCSIFPHYADDR_IN6:
2355 #endif
2356 	case SIOCSLIFPHYADDR:
2357 	case SIOCSIFMEDIA:
2358 	case SIOCSIFGENERIC:
2359 		error = priv_check(td, PRIV_NET_HWIOCTL);
2360 		if (error)
2361 			return (error);
2362 		if (ifp->if_ioctl == NULL)
2363 			return (EOPNOTSUPP);
2364 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2365 		if (error == 0)
2366 			getmicrotime(&ifp->if_lastchange);
2367 		break;
2368 
2369 	case SIOCGIFSTATUS:
2370 		ifs = (struct ifstat *)data;
2371 		ifs->ascii[0] = '\0';
2372 
2373 	case SIOCGIFPSRCADDR:
2374 	case SIOCGIFPDSTADDR:
2375 	case SIOCGLIFPHYADDR:
2376 	case SIOCGIFMEDIA:
2377 	case SIOCGIFGENERIC:
2378 		if (ifp->if_ioctl == NULL)
2379 			return (EOPNOTSUPP);
2380 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2381 		break;
2382 
2383 	case SIOCSIFLLADDR:
2384 		error = priv_check(td, PRIV_NET_SETLLADDR);
2385 		if (error)
2386 			return (error);
2387 		error = if_setlladdr(ifp,
2388 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2389 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2390 		break;
2391 
2392 	case SIOCAIFGROUP:
2393 	{
2394 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2395 
2396 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2397 		if (error)
2398 			return (error);
2399 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2400 			return (error);
2401 		break;
2402 	}
2403 
2404 	case SIOCGIFGROUP:
2405 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2406 			return (error);
2407 		break;
2408 
2409 	case SIOCDIFGROUP:
2410 	{
2411 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2412 
2413 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2414 		if (error)
2415 			return (error);
2416 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2417 			return (error);
2418 		break;
2419 	}
2420 
2421 	default:
2422 		error = ENOIOCTL;
2423 		break;
2424 	}
2425 	return (error);
2426 }
2427 
2428 #ifdef COMPAT_FREEBSD32
2429 struct ifconf32 {
2430 	int32_t	ifc_len;
2431 	union {
2432 		uint32_t	ifcu_buf;
2433 		uint32_t	ifcu_req;
2434 	} ifc_ifcu;
2435 };
2436 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2437 #endif
2438 
2439 /*
2440  * Interface ioctls.
2441  */
2442 int
2443 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2444 {
2445 	struct ifnet *ifp;
2446 	struct ifreq *ifr;
2447 	int error;
2448 	int oif_flags;
2449 
2450 	CURVNET_SET(so->so_vnet);
2451 	switch (cmd) {
2452 	case SIOCGIFCONF:
2453 	case OSIOCGIFCONF:
2454 		error = ifconf(cmd, data);
2455 		CURVNET_RESTORE();
2456 		return (error);
2457 
2458 #ifdef COMPAT_FREEBSD32
2459 	case SIOCGIFCONF32:
2460 		{
2461 			struct ifconf32 *ifc32;
2462 			struct ifconf ifc;
2463 
2464 			ifc32 = (struct ifconf32 *)data;
2465 			ifc.ifc_len = ifc32->ifc_len;
2466 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2467 
2468 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2469 			CURVNET_RESTORE();
2470 			if (error == 0)
2471 				ifc32->ifc_len = ifc.ifc_len;
2472 			return (error);
2473 		}
2474 #endif
2475 	}
2476 	ifr = (struct ifreq *)data;
2477 
2478 	switch (cmd) {
2479 #ifdef VIMAGE
2480 	case SIOCSIFRVNET:
2481 		error = priv_check(td, PRIV_NET_SETIFVNET);
2482 		if (error == 0)
2483 			error = if_vmove_reclaim(td, ifr->ifr_name,
2484 			    ifr->ifr_jid);
2485 		CURVNET_RESTORE();
2486 		return (error);
2487 #endif
2488 	case SIOCIFCREATE:
2489 	case SIOCIFCREATE2:
2490 		error = priv_check(td, PRIV_NET_IFCREATE);
2491 		if (error == 0)
2492 			error = if_clone_create(ifr->ifr_name,
2493 			    sizeof(ifr->ifr_name),
2494 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2495 		CURVNET_RESTORE();
2496 		return (error);
2497 	case SIOCIFDESTROY:
2498 		error = priv_check(td, PRIV_NET_IFDESTROY);
2499 		if (error == 0)
2500 			error = if_clone_destroy(ifr->ifr_name);
2501 		CURVNET_RESTORE();
2502 		return (error);
2503 
2504 	case SIOCIFGCLONERS:
2505 		error = if_clone_list((struct if_clonereq *)data);
2506 		CURVNET_RESTORE();
2507 		return (error);
2508 	case SIOCGIFGMEMB:
2509 		error = if_getgroupmembers((struct ifgroupreq *)data);
2510 		CURVNET_RESTORE();
2511 		return (error);
2512 #if defined(INET) || defined(INET6)
2513 	case SIOCSVH:
2514 	case SIOCGVH:
2515 		if (carp_ioctl_p == NULL)
2516 			error = EPROTONOSUPPORT;
2517 		else
2518 			error = (*carp_ioctl_p)(ifr, cmd, td);
2519 		CURVNET_RESTORE();
2520 		return (error);
2521 #endif
2522 	}
2523 
2524 	ifp = ifunit_ref(ifr->ifr_name);
2525 	if (ifp == NULL) {
2526 		CURVNET_RESTORE();
2527 		return (ENXIO);
2528 	}
2529 
2530 	error = ifhwioctl(cmd, ifp, data, td);
2531 	if (error != ENOIOCTL) {
2532 		if_rele(ifp);
2533 		CURVNET_RESTORE();
2534 		return (error);
2535 	}
2536 
2537 	oif_flags = ifp->if_flags;
2538 	if (so->so_proto == NULL) {
2539 		if_rele(ifp);
2540 		CURVNET_RESTORE();
2541 		return (EOPNOTSUPP);
2542 	}
2543 #ifndef COMPAT_43
2544 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2545 								 data,
2546 								 ifp, td));
2547 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2548 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2549 #else
2550 	{
2551 		u_long ocmd = cmd;
2552 
2553 		switch (cmd) {
2554 
2555 		case SIOCSIFDSTADDR:
2556 		case SIOCSIFADDR:
2557 		case SIOCSIFBRDADDR:
2558 		case SIOCSIFNETMASK:
2559 #if BYTE_ORDER != BIG_ENDIAN
2560 			if (ifr->ifr_addr.sa_family == 0 &&
2561 			    ifr->ifr_addr.sa_len < 16) {
2562 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2563 				ifr->ifr_addr.sa_len = 16;
2564 			}
2565 #else
2566 			if (ifr->ifr_addr.sa_len == 0)
2567 				ifr->ifr_addr.sa_len = 16;
2568 #endif
2569 			break;
2570 
2571 		case OSIOCGIFADDR:
2572 			cmd = SIOCGIFADDR;
2573 			break;
2574 
2575 		case OSIOCGIFDSTADDR:
2576 			cmd = SIOCGIFDSTADDR;
2577 			break;
2578 
2579 		case OSIOCGIFBRDADDR:
2580 			cmd = SIOCGIFBRDADDR;
2581 			break;
2582 
2583 		case OSIOCGIFNETMASK:
2584 			cmd = SIOCGIFNETMASK;
2585 		}
2586 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2587 								   cmd,
2588 								   data,
2589 								   ifp, td));
2590 		if (error == EOPNOTSUPP && ifp != NULL &&
2591 		    ifp->if_ioctl != NULL)
2592 			error = (*ifp->if_ioctl)(ifp, cmd, data);
2593 		switch (ocmd) {
2594 
2595 		case OSIOCGIFADDR:
2596 		case OSIOCGIFDSTADDR:
2597 		case OSIOCGIFBRDADDR:
2598 		case OSIOCGIFNETMASK:
2599 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2600 
2601 		}
2602 	}
2603 #endif /* COMPAT_43 */
2604 
2605 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2606 #ifdef INET6
2607 		if (ifp->if_flags & IFF_UP)
2608 			in6_if_up(ifp);
2609 #endif
2610 	}
2611 	if_rele(ifp);
2612 	CURVNET_RESTORE();
2613 	return (error);
2614 }
2615 
2616 /*
2617  * The code common to handling reference counted flags,
2618  * e.g., in ifpromisc() and if_allmulti().
2619  * The "pflag" argument can specify a permanent mode flag to check,
2620  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2621  *
2622  * Only to be used on stack-owned flags, not driver-owned flags.
2623  */
2624 static int
2625 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2626 {
2627 	struct ifreq ifr;
2628 	int error;
2629 	int oldflags, oldcount;
2630 
2631 	/* Sanity checks to catch programming errors */
2632 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2633 	    ("%s: setting driver-owned flag %d", __func__, flag));
2634 
2635 	if (onswitch)
2636 		KASSERT(*refcount >= 0,
2637 		    ("%s: increment negative refcount %d for flag %d",
2638 		    __func__, *refcount, flag));
2639 	else
2640 		KASSERT(*refcount > 0,
2641 		    ("%s: decrement non-positive refcount %d for flag %d",
2642 		    __func__, *refcount, flag));
2643 
2644 	/* In case this mode is permanent, just touch refcount */
2645 	if (ifp->if_flags & pflag) {
2646 		*refcount += onswitch ? 1 : -1;
2647 		return (0);
2648 	}
2649 
2650 	/* Save ifnet parameters for if_ioctl() may fail */
2651 	oldcount = *refcount;
2652 	oldflags = ifp->if_flags;
2653 
2654 	/*
2655 	 * See if we aren't the only and touching refcount is enough.
2656 	 * Actually toggle interface flag if we are the first or last.
2657 	 */
2658 	if (onswitch) {
2659 		if ((*refcount)++)
2660 			return (0);
2661 		ifp->if_flags |= flag;
2662 	} else {
2663 		if (--(*refcount))
2664 			return (0);
2665 		ifp->if_flags &= ~flag;
2666 	}
2667 
2668 	/* Call down the driver since we've changed interface flags */
2669 	if (ifp->if_ioctl == NULL) {
2670 		error = EOPNOTSUPP;
2671 		goto recover;
2672 	}
2673 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2674 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2675 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2676 	if (error)
2677 		goto recover;
2678 	/* Notify userland that interface flags have changed */
2679 	rt_ifmsg(ifp);
2680 	return (0);
2681 
2682 recover:
2683 	/* Recover after driver error */
2684 	*refcount = oldcount;
2685 	ifp->if_flags = oldflags;
2686 	return (error);
2687 }
2688 
2689 /*
2690  * Set/clear promiscuous mode on interface ifp based on the truth value
2691  * of pswitch.  The calls are reference counted so that only the first
2692  * "on" request actually has an effect, as does the final "off" request.
2693  * Results are undefined if the "off" and "on" requests are not matched.
2694  */
2695 int
2696 ifpromisc(struct ifnet *ifp, int pswitch)
2697 {
2698 	int error;
2699 	int oldflags = ifp->if_flags;
2700 
2701 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2702 			   &ifp->if_pcount, pswitch);
2703 	/* If promiscuous mode status has changed, log a message */
2704 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2705 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2706 		    ifp->if_xname,
2707 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2708 	return (error);
2709 }
2710 
2711 /*
2712  * Return interface configuration
2713  * of system.  List may be used
2714  * in later ioctl's (above) to get
2715  * other information.
2716  */
2717 /*ARGSUSED*/
2718 static int
2719 ifconf(u_long cmd, caddr_t data)
2720 {
2721 	struct ifconf *ifc = (struct ifconf *)data;
2722 	struct ifnet *ifp;
2723 	struct ifaddr *ifa;
2724 	struct ifreq ifr;
2725 	struct sbuf *sb;
2726 	int error, full = 0, valid_len, max_len;
2727 
2728 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2729 	max_len = MAXPHYS - 1;
2730 
2731 	/* Prevent hostile input from being able to crash the system */
2732 	if (ifc->ifc_len <= 0)
2733 		return (EINVAL);
2734 
2735 again:
2736 	if (ifc->ifc_len <= max_len) {
2737 		max_len = ifc->ifc_len;
2738 		full = 1;
2739 	}
2740 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2741 	max_len = 0;
2742 	valid_len = 0;
2743 
2744 	IFNET_RLOCK();
2745 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2746 		int addrs;
2747 
2748 		/*
2749 		 * Zero the ifr_name buffer to make sure we don't
2750 		 * disclose the contents of the stack.
2751 		 */
2752 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2753 
2754 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2755 		    >= sizeof(ifr.ifr_name)) {
2756 			sbuf_delete(sb);
2757 			IFNET_RUNLOCK();
2758 			return (ENAMETOOLONG);
2759 		}
2760 
2761 		addrs = 0;
2762 		IF_ADDR_RLOCK(ifp);
2763 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2764 			struct sockaddr *sa = ifa->ifa_addr;
2765 
2766 			if (prison_if(curthread->td_ucred, sa) != 0)
2767 				continue;
2768 			addrs++;
2769 #ifdef COMPAT_43
2770 			if (cmd == OSIOCGIFCONF) {
2771 				struct osockaddr *osa =
2772 					 (struct osockaddr *)&ifr.ifr_addr;
2773 				ifr.ifr_addr = *sa;
2774 				osa->sa_family = sa->sa_family;
2775 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2776 				max_len += sizeof(ifr);
2777 			} else
2778 #endif
2779 			if (sa->sa_len <= sizeof(*sa)) {
2780 				ifr.ifr_addr = *sa;
2781 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2782 				max_len += sizeof(ifr);
2783 			} else {
2784 				sbuf_bcat(sb, &ifr,
2785 				    offsetof(struct ifreq, ifr_addr));
2786 				max_len += offsetof(struct ifreq, ifr_addr);
2787 				sbuf_bcat(sb, sa, sa->sa_len);
2788 				max_len += sa->sa_len;
2789 			}
2790 
2791 			if (sbuf_error(sb) == 0)
2792 				valid_len = sbuf_len(sb);
2793 		}
2794 		IF_ADDR_RUNLOCK(ifp);
2795 		if (addrs == 0) {
2796 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2797 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2798 			max_len += sizeof(ifr);
2799 
2800 			if (sbuf_error(sb) == 0)
2801 				valid_len = sbuf_len(sb);
2802 		}
2803 	}
2804 	IFNET_RUNLOCK();
2805 
2806 	/*
2807 	 * If we didn't allocate enough space (uncommon), try again.  If
2808 	 * we have already allocated as much space as we are allowed,
2809 	 * return what we've got.
2810 	 */
2811 	if (valid_len != max_len && !full) {
2812 		sbuf_delete(sb);
2813 		goto again;
2814 	}
2815 
2816 	ifc->ifc_len = valid_len;
2817 	sbuf_finish(sb);
2818 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2819 	sbuf_delete(sb);
2820 	return (error);
2821 }
2822 
2823 /*
2824  * Just like ifpromisc(), but for all-multicast-reception mode.
2825  */
2826 int
2827 if_allmulti(struct ifnet *ifp, int onswitch)
2828 {
2829 
2830 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2831 }
2832 
2833 struct ifmultiaddr *
2834 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2835 {
2836 	struct ifmultiaddr *ifma;
2837 
2838 	IF_ADDR_LOCK_ASSERT(ifp);
2839 
2840 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2841 		if (sa->sa_family == AF_LINK) {
2842 			if (sa_dl_equal(ifma->ifma_addr, sa))
2843 				break;
2844 		} else {
2845 			if (sa_equal(ifma->ifma_addr, sa))
2846 				break;
2847 		}
2848 	}
2849 
2850 	return ifma;
2851 }
2852 
2853 /*
2854  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2855  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2856  * the ifnet multicast address list here, so the caller must do that and
2857  * other setup work (such as notifying the device driver).  The reference
2858  * count is initialized to 1.
2859  */
2860 static struct ifmultiaddr *
2861 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2862     int mflags)
2863 {
2864 	struct ifmultiaddr *ifma;
2865 	struct sockaddr *dupsa;
2866 
2867 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2868 	    M_ZERO);
2869 	if (ifma == NULL)
2870 		return (NULL);
2871 
2872 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2873 	if (dupsa == NULL) {
2874 		free(ifma, M_IFMADDR);
2875 		return (NULL);
2876 	}
2877 	bcopy(sa, dupsa, sa->sa_len);
2878 	ifma->ifma_addr = dupsa;
2879 
2880 	ifma->ifma_ifp = ifp;
2881 	ifma->ifma_refcount = 1;
2882 	ifma->ifma_protospec = NULL;
2883 
2884 	if (llsa == NULL) {
2885 		ifma->ifma_lladdr = NULL;
2886 		return (ifma);
2887 	}
2888 
2889 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2890 	if (dupsa == NULL) {
2891 		free(ifma->ifma_addr, M_IFMADDR);
2892 		free(ifma, M_IFMADDR);
2893 		return (NULL);
2894 	}
2895 	bcopy(llsa, dupsa, llsa->sa_len);
2896 	ifma->ifma_lladdr = dupsa;
2897 
2898 	return (ifma);
2899 }
2900 
2901 /*
2902  * if_freemulti: free ifmultiaddr structure and possibly attached related
2903  * addresses.  The caller is responsible for implementing reference
2904  * counting, notifying the driver, handling routing messages, and releasing
2905  * any dependent link layer state.
2906  */
2907 static void
2908 if_freemulti(struct ifmultiaddr *ifma)
2909 {
2910 
2911 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2912 	    ifma->ifma_refcount));
2913 	KASSERT(ifma->ifma_protospec == NULL,
2914 	    ("if_freemulti: protospec not NULL"));
2915 
2916 	if (ifma->ifma_lladdr != NULL)
2917 		free(ifma->ifma_lladdr, M_IFMADDR);
2918 	free(ifma->ifma_addr, M_IFMADDR);
2919 	free(ifma, M_IFMADDR);
2920 }
2921 
2922 /*
2923  * Register an additional multicast address with a network interface.
2924  *
2925  * - If the address is already present, bump the reference count on the
2926  *   address and return.
2927  * - If the address is not link-layer, look up a link layer address.
2928  * - Allocate address structures for one or both addresses, and attach to the
2929  *   multicast address list on the interface.  If automatically adding a link
2930  *   layer address, the protocol address will own a reference to the link
2931  *   layer address, to be freed when it is freed.
2932  * - Notify the network device driver of an addition to the multicast address
2933  *   list.
2934  *
2935  * 'sa' points to caller-owned memory with the desired multicast address.
2936  *
2937  * 'retifma' will be used to return a pointer to the resulting multicast
2938  * address reference, if desired.
2939  */
2940 int
2941 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2942     struct ifmultiaddr **retifma)
2943 {
2944 	struct ifmultiaddr *ifma, *ll_ifma;
2945 	struct sockaddr *llsa;
2946 	int error;
2947 
2948 	/*
2949 	 * If the address is already present, return a new reference to it;
2950 	 * otherwise, allocate storage and set up a new address.
2951 	 */
2952 	IF_ADDR_WLOCK(ifp);
2953 	ifma = if_findmulti(ifp, sa);
2954 	if (ifma != NULL) {
2955 		ifma->ifma_refcount++;
2956 		if (retifma != NULL)
2957 			*retifma = ifma;
2958 		IF_ADDR_WUNLOCK(ifp);
2959 		return (0);
2960 	}
2961 
2962 	/*
2963 	 * The address isn't already present; resolve the protocol address
2964 	 * into a link layer address, and then look that up, bump its
2965 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2966 	 * returned, we will need to free it later.
2967 	 */
2968 	llsa = NULL;
2969 	ll_ifma = NULL;
2970 	if (ifp->if_resolvemulti != NULL) {
2971 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2972 		if (error)
2973 			goto unlock_out;
2974 	}
2975 
2976 	/*
2977 	 * Allocate the new address.  Don't hook it up yet, as we may also
2978 	 * need to allocate a link layer multicast address.
2979 	 */
2980 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2981 	if (ifma == NULL) {
2982 		error = ENOMEM;
2983 		goto free_llsa_out;
2984 	}
2985 
2986 	/*
2987 	 * If a link layer address is found, we'll need to see if it's
2988 	 * already present in the address list, or allocate is as well.
2989 	 * When this block finishes, the link layer address will be on the
2990 	 * list.
2991 	 */
2992 	if (llsa != NULL) {
2993 		ll_ifma = if_findmulti(ifp, llsa);
2994 		if (ll_ifma == NULL) {
2995 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2996 			if (ll_ifma == NULL) {
2997 				--ifma->ifma_refcount;
2998 				if_freemulti(ifma);
2999 				error = ENOMEM;
3000 				goto free_llsa_out;
3001 			}
3002 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3003 			    ifma_link);
3004 		} else
3005 			ll_ifma->ifma_refcount++;
3006 		ifma->ifma_llifma = ll_ifma;
3007 	}
3008 
3009 	/*
3010 	 * We now have a new multicast address, ifma, and possibly a new or
3011 	 * referenced link layer address.  Add the primary address to the
3012 	 * ifnet address list.
3013 	 */
3014 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3015 
3016 	if (retifma != NULL)
3017 		*retifma = ifma;
3018 
3019 	/*
3020 	 * Must generate the message while holding the lock so that 'ifma'
3021 	 * pointer is still valid.
3022 	 */
3023 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3024 	IF_ADDR_WUNLOCK(ifp);
3025 
3026 	/*
3027 	 * We are certain we have added something, so call down to the
3028 	 * interface to let them know about it.
3029 	 */
3030 	if (ifp->if_ioctl != NULL) {
3031 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3032 	}
3033 
3034 	if (llsa != NULL)
3035 		free(llsa, M_IFMADDR);
3036 
3037 	return (0);
3038 
3039 free_llsa_out:
3040 	if (llsa != NULL)
3041 		free(llsa, M_IFMADDR);
3042 
3043 unlock_out:
3044 	IF_ADDR_WUNLOCK(ifp);
3045 	return (error);
3046 }
3047 
3048 /*
3049  * Delete a multicast group membership by network-layer group address.
3050  *
3051  * Returns ENOENT if the entry could not be found. If ifp no longer
3052  * exists, results are undefined. This entry point should only be used
3053  * from subsystems which do appropriate locking to hold ifp for the
3054  * duration of the call.
3055  * Network-layer protocol domains must use if_delmulti_ifma().
3056  */
3057 int
3058 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3059 {
3060 	struct ifmultiaddr *ifma;
3061 	int lastref;
3062 #ifdef INVARIANTS
3063 	struct ifnet *oifp;
3064 
3065 	IFNET_RLOCK_NOSLEEP();
3066 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3067 		if (ifp == oifp)
3068 			break;
3069 	if (ifp != oifp)
3070 		ifp = NULL;
3071 	IFNET_RUNLOCK_NOSLEEP();
3072 
3073 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3074 #endif
3075 	if (ifp == NULL)
3076 		return (ENOENT);
3077 
3078 	IF_ADDR_WLOCK(ifp);
3079 	lastref = 0;
3080 	ifma = if_findmulti(ifp, sa);
3081 	if (ifma != NULL)
3082 		lastref = if_delmulti_locked(ifp, ifma, 0);
3083 	IF_ADDR_WUNLOCK(ifp);
3084 
3085 	if (ifma == NULL)
3086 		return (ENOENT);
3087 
3088 	if (lastref && ifp->if_ioctl != NULL) {
3089 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3090 	}
3091 
3092 	return (0);
3093 }
3094 
3095 /*
3096  * Delete all multicast group membership for an interface.
3097  * Should be used to quickly flush all multicast filters.
3098  */
3099 void
3100 if_delallmulti(struct ifnet *ifp)
3101 {
3102 	struct ifmultiaddr *ifma;
3103 	struct ifmultiaddr *next;
3104 
3105 	IF_ADDR_WLOCK(ifp);
3106 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3107 		if_delmulti_locked(ifp, ifma, 0);
3108 	IF_ADDR_WUNLOCK(ifp);
3109 }
3110 
3111 /*
3112  * Delete a multicast group membership by group membership pointer.
3113  * Network-layer protocol domains must use this routine.
3114  *
3115  * It is safe to call this routine if the ifp disappeared.
3116  */
3117 void
3118 if_delmulti_ifma(struct ifmultiaddr *ifma)
3119 {
3120 	struct ifnet *ifp;
3121 	int lastref;
3122 
3123 	ifp = ifma->ifma_ifp;
3124 #ifdef DIAGNOSTIC
3125 	if (ifp == NULL) {
3126 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3127 	} else {
3128 		struct ifnet *oifp;
3129 
3130 		IFNET_RLOCK_NOSLEEP();
3131 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3132 			if (ifp == oifp)
3133 				break;
3134 		if (ifp != oifp) {
3135 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3136 			ifp = NULL;
3137 		}
3138 		IFNET_RUNLOCK_NOSLEEP();
3139 	}
3140 #endif
3141 	/*
3142 	 * If and only if the ifnet instance exists: Acquire the address lock.
3143 	 */
3144 	if (ifp != NULL)
3145 		IF_ADDR_WLOCK(ifp);
3146 
3147 	lastref = if_delmulti_locked(ifp, ifma, 0);
3148 
3149 	if (ifp != NULL) {
3150 		/*
3151 		 * If and only if the ifnet instance exists:
3152 		 *  Release the address lock.
3153 		 *  If the group was left: update the hardware hash filter.
3154 		 */
3155 		IF_ADDR_WUNLOCK(ifp);
3156 		if (lastref && ifp->if_ioctl != NULL) {
3157 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3158 		}
3159 	}
3160 }
3161 
3162 /*
3163  * Perform deletion of network-layer and/or link-layer multicast address.
3164  *
3165  * Return 0 if the reference count was decremented.
3166  * Return 1 if the final reference was released, indicating that the
3167  * hardware hash filter should be reprogrammed.
3168  */
3169 static int
3170 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3171 {
3172 	struct ifmultiaddr *ll_ifma;
3173 
3174 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3175 		KASSERT(ifma->ifma_ifp == ifp,
3176 		    ("%s: inconsistent ifp %p", __func__, ifp));
3177 		IF_ADDR_WLOCK_ASSERT(ifp);
3178 	}
3179 
3180 	ifp = ifma->ifma_ifp;
3181 
3182 	/*
3183 	 * If the ifnet is detaching, null out references to ifnet,
3184 	 * so that upper protocol layers will notice, and not attempt
3185 	 * to obtain locks for an ifnet which no longer exists. The
3186 	 * routing socket announcement must happen before the ifnet
3187 	 * instance is detached from the system.
3188 	 */
3189 	if (detaching) {
3190 #ifdef DIAGNOSTIC
3191 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3192 #endif
3193 		/*
3194 		 * ifp may already be nulled out if we are being reentered
3195 		 * to delete the ll_ifma.
3196 		 */
3197 		if (ifp != NULL) {
3198 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3199 			ifma->ifma_ifp = NULL;
3200 		}
3201 	}
3202 
3203 	if (--ifma->ifma_refcount > 0)
3204 		return 0;
3205 
3206 	/*
3207 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3208 	 * have been associated with it. Release it first if so.
3209 	 */
3210 	ll_ifma = ifma->ifma_llifma;
3211 	if (ll_ifma != NULL) {
3212 		KASSERT(ifma->ifma_lladdr != NULL,
3213 		    ("%s: llifma w/o lladdr", __func__));
3214 		if (detaching)
3215 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3216 		if (--ll_ifma->ifma_refcount == 0) {
3217 			if (ifp != NULL) {
3218 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3219 				    ifma_link);
3220 			}
3221 			if_freemulti(ll_ifma);
3222 		}
3223 	}
3224 
3225 	if (ifp != NULL)
3226 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3227 
3228 	if_freemulti(ifma);
3229 
3230 	/*
3231 	 * The last reference to this instance of struct ifmultiaddr
3232 	 * was released; the hardware should be notified of this change.
3233 	 */
3234 	return 1;
3235 }
3236 
3237 /*
3238  * Set the link layer address on an interface.
3239  *
3240  * At this time we only support certain types of interfaces,
3241  * and we don't allow the length of the address to change.
3242  */
3243 int
3244 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3245 {
3246 	struct sockaddr_dl *sdl;
3247 	struct ifaddr *ifa;
3248 	struct ifreq ifr;
3249 
3250 	IF_ADDR_RLOCK(ifp);
3251 	ifa = ifp->if_addr;
3252 	if (ifa == NULL) {
3253 		IF_ADDR_RUNLOCK(ifp);
3254 		return (EINVAL);
3255 	}
3256 	ifa_ref(ifa);
3257 	IF_ADDR_RUNLOCK(ifp);
3258 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3259 	if (sdl == NULL) {
3260 		ifa_free(ifa);
3261 		return (EINVAL);
3262 	}
3263 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3264 		ifa_free(ifa);
3265 		return (EINVAL);
3266 	}
3267 	switch (ifp->if_type) {
3268 	case IFT_ETHER:
3269 	case IFT_FDDI:
3270 	case IFT_XETHER:
3271 	case IFT_ISO88025:
3272 	case IFT_L2VLAN:
3273 	case IFT_BRIDGE:
3274 	case IFT_ARCNET:
3275 	case IFT_IEEE8023ADLAG:
3276 	case IFT_IEEE80211:
3277 		bcopy(lladdr, LLADDR(sdl), len);
3278 		ifa_free(ifa);
3279 		break;
3280 	default:
3281 		ifa_free(ifa);
3282 		return (ENODEV);
3283 	}
3284 
3285 	/*
3286 	 * If the interface is already up, we need
3287 	 * to re-init it in order to reprogram its
3288 	 * address filter.
3289 	 */
3290 	if ((ifp->if_flags & IFF_UP) != 0) {
3291 		if (ifp->if_ioctl) {
3292 			ifp->if_flags &= ~IFF_UP;
3293 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3294 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3295 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3296 			ifp->if_flags |= IFF_UP;
3297 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3298 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3299 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3300 		}
3301 #ifdef INET
3302 		/*
3303 		 * Also send gratuitous ARPs to notify other nodes about
3304 		 * the address change.
3305 		 */
3306 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3307 			if (ifa->ifa_addr->sa_family == AF_INET)
3308 				arp_ifinit(ifp, ifa);
3309 		}
3310 #endif
3311 	}
3312 	return (0);
3313 }
3314 
3315 /*
3316  * The name argument must be a pointer to storage which will last as
3317  * long as the interface does.  For physical devices, the result of
3318  * device_get_name(dev) is a good choice and for pseudo-devices a
3319  * static string works well.
3320  */
3321 void
3322 if_initname(struct ifnet *ifp, const char *name, int unit)
3323 {
3324 	ifp->if_dname = name;
3325 	ifp->if_dunit = unit;
3326 	if (unit != IF_DUNIT_NONE)
3327 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3328 	else
3329 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3330 }
3331 
3332 int
3333 if_printf(struct ifnet *ifp, const char * fmt, ...)
3334 {
3335 	va_list ap;
3336 	int retval;
3337 
3338 	retval = printf("%s: ", ifp->if_xname);
3339 	va_start(ap, fmt);
3340 	retval += vprintf(fmt, ap);
3341 	va_end(ap);
3342 	return (retval);
3343 }
3344 
3345 void
3346 if_start(struct ifnet *ifp)
3347 {
3348 
3349 	(*(ifp)->if_start)(ifp);
3350 }
3351 
3352 /*
3353  * Backwards compatibility interface for drivers
3354  * that have not implemented it
3355  */
3356 static int
3357 if_transmit(struct ifnet *ifp, struct mbuf *m)
3358 {
3359 	int error;
3360 
3361 	IFQ_HANDOFF(ifp, m, error);
3362 	return (error);
3363 }
3364 
3365 int
3366 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3367 {
3368 	int active = 0;
3369 
3370 	IF_LOCK(ifq);
3371 	if (_IF_QFULL(ifq)) {
3372 		_IF_DROP(ifq);
3373 		IF_UNLOCK(ifq);
3374 		m_freem(m);
3375 		return (0);
3376 	}
3377 	if (ifp != NULL) {
3378 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3379 		if (m->m_flags & (M_BCAST|M_MCAST))
3380 			ifp->if_omcasts++;
3381 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3382 	}
3383 	_IF_ENQUEUE(ifq, m);
3384 	IF_UNLOCK(ifq);
3385 	if (ifp != NULL && !active)
3386 		(*(ifp)->if_start)(ifp);
3387 	return (1);
3388 }
3389 
3390 void
3391 if_register_com_alloc(u_char type,
3392     if_com_alloc_t *a, if_com_free_t *f)
3393 {
3394 
3395 	KASSERT(if_com_alloc[type] == NULL,
3396 	    ("if_register_com_alloc: %d already registered", type));
3397 	KASSERT(if_com_free[type] == NULL,
3398 	    ("if_register_com_alloc: %d free already registered", type));
3399 
3400 	if_com_alloc[type] = a;
3401 	if_com_free[type] = f;
3402 }
3403 
3404 void
3405 if_deregister_com_alloc(u_char type)
3406 {
3407 
3408 	KASSERT(if_com_alloc[type] != NULL,
3409 	    ("if_deregister_com_alloc: %d not registered", type));
3410 	KASSERT(if_com_free[type] != NULL,
3411 	    ("if_deregister_com_alloc: %d free not registered", type));
3412 	if_com_alloc[type] = NULL;
3413 	if_com_free[type] = NULL;
3414 }
3415