xref: /freebsd/sys/net/if.c (revision 1c6d60de932c8553af44629218cb9697bc0f2ef1)
1 /*-
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.5 (Berkeley) 1/9/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/malloc.h>
41 #include <sys/sbuf.h>
42 #include <sys/bus.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/refcount.h>
53 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
60 #include <sys/jail.h>
61 #include <sys/priv.h>
62 
63 #include <machine/stdarg.h>
64 #include <vm/uma.h>
65 
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_clone.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/vnet.h>
75 
76 #if defined(INET) || defined(INET6)
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
79 #include <netinet/ip.h>
80 #include <netinet/ip_carp.h>
81 #ifdef INET
82 #include <netinet/if_ether.h>
83 #endif /* INET */
84 #ifdef INET6
85 #include <netinet6/in6_var.h>
86 #include <netinet6/in6_ifattach.h>
87 #endif /* INET6 */
88 #endif /* INET || INET6 */
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     const 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 	CURVNET_SET_QUIET(ifp->if_vnet);
509 	IFNET_WLOCK();
510 	KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
511 	    ("%s: freeing unallocated ifnet", ifp->if_xname));
512 
513 	ifindex_free_locked(ifp->if_index);
514 	IFNET_WUNLOCK();
515 
516 	if (refcount_release(&ifp->if_refcount))
517 		if_free_internal(ifp);
518 	CURVNET_RESTORE();
519 }
520 
521 /*
522  * Interfaces to keep an ifnet type-stable despite the possibility of the
523  * driver calling if_free().  If there are additional references, we defer
524  * freeing the underlying data structure.
525  */
526 void
527 if_ref(struct ifnet *ifp)
528 {
529 
530 	/* We don't assert the ifnet list lock here, but arguably should. */
531 	refcount_acquire(&ifp->if_refcount);
532 }
533 
534 void
535 if_rele(struct ifnet *ifp)
536 {
537 
538 	if (!refcount_release(&ifp->if_refcount))
539 		return;
540 	if_free_internal(ifp);
541 }
542 
543 void
544 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
545 {
546 
547 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
548 
549 	if (ifq->ifq_maxlen == 0)
550 		ifq->ifq_maxlen = ifqmaxlen;
551 
552 	ifq->altq_type = 0;
553 	ifq->altq_disc = NULL;
554 	ifq->altq_flags &= ALTQF_CANTCHANGE;
555 	ifq->altq_tbr  = NULL;
556 	ifq->altq_ifp  = ifp;
557 }
558 
559 void
560 ifq_delete(struct ifaltq *ifq)
561 {
562 	mtx_destroy(&ifq->ifq_mtx);
563 }
564 
565 /*
566  * Perform generic interface initalization tasks and attach the interface
567  * to the list of "active" interfaces.  If vmove flag is set on entry
568  * to if_attach_internal(), perform only a limited subset of initialization
569  * tasks, given that we are moving from one vnet to another an ifnet which
570  * has already been fully initialized.
571  *
572  * XXX:
573  *  - The decision to return void and thus require this function to
574  *    succeed is questionable.
575  *  - We should probably do more sanity checking.  For instance we don't
576  *    do anything to insure if_xname is unique or non-empty.
577  */
578 void
579 if_attach(struct ifnet *ifp)
580 {
581 
582 	if_attach_internal(ifp, 0);
583 }
584 
585 static void
586 if_attach_internal(struct ifnet *ifp, int vmove)
587 {
588 	unsigned socksize, ifasize;
589 	int namelen, masklen;
590 	struct sockaddr_dl *sdl;
591 	struct ifaddr *ifa;
592 
593 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
594 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
595 		    ifp->if_xname);
596 
597 #ifdef VIMAGE
598 	ifp->if_vnet = curvnet;
599 	if (ifp->if_home_vnet == NULL)
600 		ifp->if_home_vnet = curvnet;
601 #endif
602 
603 	if_addgroup(ifp, IFG_ALL);
604 
605 	getmicrotime(&ifp->if_lastchange);
606 	ifp->if_data.ifi_epoch = time_uptime;
607 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
608 
609 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
610 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
611 	    ("transmit and qflush must both either be set or both be NULL"));
612 	if (ifp->if_transmit == NULL) {
613 		ifp->if_transmit = if_transmit;
614 		ifp->if_qflush = if_qflush;
615 	}
616 
617 	if (!vmove) {
618 #ifdef MAC
619 		mac_ifnet_create(ifp);
620 #endif
621 
622 		/*
623 		 * Create a Link Level name for this device.
624 		 */
625 		namelen = strlen(ifp->if_xname);
626 		/*
627 		 * Always save enough space for any possiable name so we
628 		 * can do a rename in place later.
629 		 */
630 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
631 		socksize = masklen + ifp->if_addrlen;
632 		if (socksize < sizeof(*sdl))
633 			socksize = sizeof(*sdl);
634 		socksize = roundup2(socksize, sizeof(long));
635 		ifasize = sizeof(*ifa) + 2 * socksize;
636 		ifa = ifa_alloc(ifasize, M_WAITOK);
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 #if defined(INET) || defined(INET6)
658 		/* Initialize to max value. */
659 		if (ifp->if_hw_tsomax == 0)
660 			ifp->if_hw_tsomax = IP_MAXPACKET;
661 		KASSERT(ifp->if_hw_tsomax <= IP_MAXPACKET &&
662 		    ifp->if_hw_tsomax >= IP_MAXPACKET / 8,
663 		    ("%s: tsomax outside of range", __func__));
664 #endif
665 	}
666 #ifdef VIMAGE
667 	else {
668 		/*
669 		 * Update the interface index in the link layer address
670 		 * of the interface.
671 		 */
672 		for (ifa = ifp->if_addr; ifa != NULL;
673 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
674 			if (ifa->ifa_addr->sa_family == AF_LINK) {
675 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
676 				sdl->sdl_index = ifp->if_index;
677 			}
678 		}
679 	}
680 #endif
681 
682 	IFNET_WLOCK();
683 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
684 #ifdef VIMAGE
685 	curvnet->vnet_ifcnt++;
686 #endif
687 	IFNET_WUNLOCK();
688 
689 	if (domain_init_status >= 2)
690 		if_attachdomain1(ifp);
691 
692 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
693 	if (IS_DEFAULT_VNET(curvnet))
694 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
695 
696 	/* Announce the interface. */
697 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
698 }
699 
700 static void
701 if_attachdomain(void *dummy)
702 {
703 	struct ifnet *ifp;
704 
705 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
706 		if_attachdomain1(ifp);
707 }
708 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
709     if_attachdomain, NULL);
710 
711 static void
712 if_attachdomain1(struct ifnet *ifp)
713 {
714 	struct domain *dp;
715 
716 	/*
717 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
718 	 * cannot lock ifp->if_afdata initialization, entirely.
719 	 */
720 	if (IF_AFDATA_TRYLOCK(ifp) == 0)
721 		return;
722 	if (ifp->if_afdata_initialized >= domain_init_status) {
723 		IF_AFDATA_UNLOCK(ifp);
724 		log(LOG_WARNING, "%s called more than once on %s\n",
725 		    __func__, ifp->if_xname);
726 		return;
727 	}
728 	ifp->if_afdata_initialized = domain_init_status;
729 	IF_AFDATA_UNLOCK(ifp);
730 
731 	/* address family dependent data region */
732 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
733 	for (dp = domains; dp; dp = dp->dom_next) {
734 		if (dp->dom_ifattach)
735 			ifp->if_afdata[dp->dom_family] =
736 			    (*dp->dom_ifattach)(ifp);
737 	}
738 }
739 
740 /*
741  * Remove any unicast or broadcast network addresses from an interface.
742  */
743 void
744 if_purgeaddrs(struct ifnet *ifp)
745 {
746 	struct ifaddr *ifa, *next;
747 
748 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
749 		if (ifa->ifa_addr->sa_family == AF_LINK)
750 			continue;
751 #ifdef INET
752 		/* XXX: Ugly!! ad hoc just for INET */
753 		if (ifa->ifa_addr->sa_family == AF_INET) {
754 			struct ifaliasreq ifr;
755 
756 			bzero(&ifr, sizeof(ifr));
757 			ifr.ifra_addr = *ifa->ifa_addr;
758 			if (ifa->ifa_dstaddr)
759 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
760 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
761 			    NULL) == 0)
762 				continue;
763 		}
764 #endif /* INET */
765 #ifdef INET6
766 		if (ifa->ifa_addr->sa_family == AF_INET6) {
767 			in6_purgeaddr(ifa);
768 			/* ifp_addrhead is already updated */
769 			continue;
770 		}
771 #endif /* INET6 */
772 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
773 		ifa_free(ifa);
774 	}
775 }
776 
777 /*
778  * Remove any multicast network addresses from an interface when an ifnet
779  * is going away.
780  */
781 static void
782 if_purgemaddrs(struct ifnet *ifp)
783 {
784 	struct ifmultiaddr *ifma;
785 	struct ifmultiaddr *next;
786 
787 	IF_ADDR_WLOCK(ifp);
788 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
789 		if_delmulti_locked(ifp, ifma, 1);
790 	IF_ADDR_WUNLOCK(ifp);
791 }
792 
793 /*
794  * Detach an interface, removing it from the list of "active" interfaces.
795  * If vmove flag is set on entry to if_detach_internal(), perform only a
796  * limited subset of cleanup tasks, given that we are moving an ifnet from
797  * one vnet to another, where it must be fully operational.
798  *
799  * XXXRW: There are some significant questions about event ordering, and
800  * how to prevent things from starting to use the interface during detach.
801  */
802 void
803 if_detach(struct ifnet *ifp)
804 {
805 
806 	CURVNET_SET_QUIET(ifp->if_vnet);
807 	if_detach_internal(ifp, 0);
808 	CURVNET_RESTORE();
809 }
810 
811 static void
812 if_detach_internal(struct ifnet *ifp, int vmove)
813 {
814 	struct ifaddr *ifa;
815 	struct radix_node_head	*rnh;
816 	int i, j;
817 	struct domain *dp;
818  	struct ifnet *iter;
819  	int found = 0;
820 
821 	IFNET_WLOCK();
822 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
823 		if (iter == ifp) {
824 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
825 			found = 1;
826 			break;
827 		}
828 #ifdef VIMAGE
829 	if (found)
830 		curvnet->vnet_ifcnt--;
831 #endif
832 	IFNET_WUNLOCK();
833 	if (!found) {
834 		if (vmove)
835 			panic("%s: ifp=%p not on the ifnet tailq %p",
836 			    __func__, ifp, &V_ifnet);
837 		else
838 			return; /* XXX this should panic as well? */
839 	}
840 
841 	/*
842 	 * Remove/wait for pending events.
843 	 */
844 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
845 
846 	/*
847 	 * Remove routes and flush queues.
848 	 */
849 	if_down(ifp);
850 #ifdef ALTQ
851 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
852 		altq_disable(&ifp->if_snd);
853 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
854 		altq_detach(&ifp->if_snd);
855 #endif
856 
857 	if_purgeaddrs(ifp);
858 
859 #ifdef INET
860 	in_ifdetach(ifp);
861 #endif
862 
863 #ifdef INET6
864 	/*
865 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
866 	 * before removing routing entries below, since IPv6 interface direct
867 	 * routes are expected to be removed by the IPv6-specific kernel API.
868 	 * Otherwise, the kernel will detect some inconsistency and bark it.
869 	 */
870 	in6_ifdetach(ifp);
871 #endif
872 	if_purgemaddrs(ifp);
873 
874 	if (!vmove) {
875 		/*
876 		 * Prevent further calls into the device driver via ifnet.
877 		 */
878 		if_dead(ifp);
879 
880 		/*
881 		 * Remove link ifaddr pointer and maybe decrement if_index.
882 		 * Clean up all addresses.
883 		 */
884 		ifp->if_addr = NULL;
885 
886 		/* We can now free link ifaddr. */
887 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
888 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
889 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
890 			ifa_free(ifa);
891 		}
892 	}
893 
894 	/*
895 	 * Delete all remaining routes using this interface
896 	 * Unfortuneatly the only way to do this is to slog through
897 	 * the entire routing table looking for routes which point
898 	 * to this interface...oh well...
899 	 */
900 	for (i = 1; i <= AF_MAX; i++) {
901 		for (j = 0; j < rt_numfibs; j++) {
902 			rnh = rt_tables_get_rnh(j, i);
903 			if (rnh == NULL)
904 				continue;
905 			RADIX_NODE_HEAD_LOCK(rnh);
906 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
907 			RADIX_NODE_HEAD_UNLOCK(rnh);
908 		}
909 	}
910 
911 	/* Announce that the interface is gone. */
912 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
913 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
914 	if (IS_DEFAULT_VNET(curvnet))
915 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
916 	if_delgroups(ifp);
917 
918 	/*
919 	 * We cannot hold the lock over dom_ifdetach calls as they might
920 	 * sleep, for example trying to drain a callout, thus open up the
921 	 * theoretical race with re-attaching.
922 	 */
923 	IF_AFDATA_LOCK(ifp);
924 	i = ifp->if_afdata_initialized;
925 	ifp->if_afdata_initialized = 0;
926 	IF_AFDATA_UNLOCK(ifp);
927 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
928 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
929 			(*dp->dom_ifdetach)(ifp,
930 			    ifp->if_afdata[dp->dom_family]);
931 	}
932 }
933 
934 #ifdef VIMAGE
935 /*
936  * if_vmove() performs a limited version of if_detach() in current
937  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
938  * An attempt is made to shrink if_index in current vnet, find an
939  * unused if_index in target vnet and calls if_grow() if necessary,
940  * and finally find an unused if_xname for the target vnet.
941  */
942 void
943 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
944 {
945 	u_short idx;
946 
947 	/*
948 	 * Detach from current vnet, but preserve LLADDR info, do not
949 	 * mark as dead etc. so that the ifnet can be reattached later.
950 	 */
951 	if_detach_internal(ifp, 1);
952 
953 	/*
954 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
955 	 * the if_index for that vnet if possible.
956 	 *
957 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
958 	 * or we'd lock on one vnet and unlock on another.
959 	 */
960 	IFNET_WLOCK();
961 	ifindex_free_locked(ifp->if_index);
962 	IFNET_WUNLOCK();
963 
964 	/*
965 	 * Perform interface-specific reassignment tasks, if provided by
966 	 * the driver.
967 	 */
968 	if (ifp->if_reassign != NULL)
969 		ifp->if_reassign(ifp, new_vnet, NULL);
970 
971 	/*
972 	 * Switch to the context of the target vnet.
973 	 */
974 	CURVNET_SET_QUIET(new_vnet);
975 
976 	IFNET_WLOCK();
977 	if (ifindex_alloc_locked(&idx) != 0) {
978 		IFNET_WUNLOCK();
979 		panic("if_index overflow");
980 	}
981 	ifp->if_index = idx;
982 	ifnet_setbyindex_locked(ifp->if_index, ifp);
983 	IFNET_WUNLOCK();
984 
985 	if_attach_internal(ifp, 1);
986 
987 	CURVNET_RESTORE();
988 }
989 
990 /*
991  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
992  */
993 static int
994 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
995 {
996 	struct prison *pr;
997 	struct ifnet *difp;
998 
999 	/* Try to find the prison within our visibility. */
1000 	sx_slock(&allprison_lock);
1001 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1002 	sx_sunlock(&allprison_lock);
1003 	if (pr == NULL)
1004 		return (ENXIO);
1005 	prison_hold_locked(pr);
1006 	mtx_unlock(&pr->pr_mtx);
1007 
1008 	/* Do not try to move the iface from and to the same prison. */
1009 	if (pr->pr_vnet == ifp->if_vnet) {
1010 		prison_free(pr);
1011 		return (EEXIST);
1012 	}
1013 
1014 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1015 	/* XXX Lock interfaces to avoid races. */
1016 	CURVNET_SET_QUIET(pr->pr_vnet);
1017 	difp = ifunit(ifname);
1018 	CURVNET_RESTORE();
1019 	if (difp != NULL) {
1020 		prison_free(pr);
1021 		return (EEXIST);
1022 	}
1023 
1024 	/* Move the interface into the child jail/vnet. */
1025 	if_vmove(ifp, pr->pr_vnet);
1026 
1027 	/* Report the new if_xname back to the userland. */
1028 	sprintf(ifname, "%s", ifp->if_xname);
1029 
1030 	prison_free(pr);
1031 	return (0);
1032 }
1033 
1034 static int
1035 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1036 {
1037 	struct prison *pr;
1038 	struct vnet *vnet_dst;
1039 	struct ifnet *ifp;
1040 
1041 	/* Try to find the prison within our visibility. */
1042 	sx_slock(&allprison_lock);
1043 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1044 	sx_sunlock(&allprison_lock);
1045 	if (pr == NULL)
1046 		return (ENXIO);
1047 	prison_hold_locked(pr);
1048 	mtx_unlock(&pr->pr_mtx);
1049 
1050 	/* Make sure the named iface exists in the source prison/vnet. */
1051 	CURVNET_SET(pr->pr_vnet);
1052 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1053 	if (ifp == NULL) {
1054 		CURVNET_RESTORE();
1055 		prison_free(pr);
1056 		return (ENXIO);
1057 	}
1058 
1059 	/* Do not try to move the iface from and to the same prison. */
1060 	vnet_dst = TD_TO_VNET(td);
1061 	if (vnet_dst == ifp->if_vnet) {
1062 		CURVNET_RESTORE();
1063 		prison_free(pr);
1064 		return (EEXIST);
1065 	}
1066 
1067 	/* Get interface back from child jail/vnet. */
1068 	if_vmove(ifp, vnet_dst);
1069 	CURVNET_RESTORE();
1070 
1071 	/* Report the new if_xname back to the userland. */
1072 	sprintf(ifname, "%s", ifp->if_xname);
1073 
1074 	prison_free(pr);
1075 	return (0);
1076 }
1077 #endif /* VIMAGE */
1078 
1079 /*
1080  * Add a group to an interface
1081  */
1082 int
1083 if_addgroup(struct ifnet *ifp, const char *groupname)
1084 {
1085 	struct ifg_list		*ifgl;
1086 	struct ifg_group	*ifg = NULL;
1087 	struct ifg_member	*ifgm;
1088 	int 			 new = 0;
1089 
1090 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1091 	    groupname[strlen(groupname) - 1] <= '9')
1092 		return (EINVAL);
1093 
1094 	IFNET_WLOCK();
1095 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1096 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1097 			IFNET_WUNLOCK();
1098 			return (EEXIST);
1099 		}
1100 
1101 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1102 	    M_NOWAIT)) == NULL) {
1103 	    	IFNET_WUNLOCK();
1104 		return (ENOMEM);
1105 	}
1106 
1107 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1108 	    M_TEMP, M_NOWAIT)) == NULL) {
1109 		free(ifgl, M_TEMP);
1110 		IFNET_WUNLOCK();
1111 		return (ENOMEM);
1112 	}
1113 
1114 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1115 		if (!strcmp(ifg->ifg_group, groupname))
1116 			break;
1117 
1118 	if (ifg == NULL) {
1119 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1120 		    M_TEMP, M_NOWAIT)) == NULL) {
1121 			free(ifgl, M_TEMP);
1122 			free(ifgm, M_TEMP);
1123 			IFNET_WUNLOCK();
1124 			return (ENOMEM);
1125 		}
1126 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1127 		ifg->ifg_refcnt = 0;
1128 		TAILQ_INIT(&ifg->ifg_members);
1129 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1130 		new = 1;
1131 	}
1132 
1133 	ifg->ifg_refcnt++;
1134 	ifgl->ifgl_group = ifg;
1135 	ifgm->ifgm_ifp = ifp;
1136 
1137 	IF_ADDR_WLOCK(ifp);
1138 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1139 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1140 	IF_ADDR_WUNLOCK(ifp);
1141 
1142 	IFNET_WUNLOCK();
1143 
1144 	if (new)
1145 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1146 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Remove a group from an interface
1153  */
1154 int
1155 if_delgroup(struct ifnet *ifp, const char *groupname)
1156 {
1157 	struct ifg_list		*ifgl;
1158 	struct ifg_member	*ifgm;
1159 
1160 	IFNET_WLOCK();
1161 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1162 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1163 			break;
1164 	if (ifgl == NULL) {
1165 		IFNET_WUNLOCK();
1166 		return (ENOENT);
1167 	}
1168 
1169 	IF_ADDR_WLOCK(ifp);
1170 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1171 	IF_ADDR_WUNLOCK(ifp);
1172 
1173 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1174 		if (ifgm->ifgm_ifp == ifp)
1175 			break;
1176 
1177 	if (ifgm != NULL) {
1178 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1179 		free(ifgm, M_TEMP);
1180 	}
1181 
1182 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1183 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1184 		IFNET_WUNLOCK();
1185 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1186 		free(ifgl->ifgl_group, M_TEMP);
1187 	} else
1188 		IFNET_WUNLOCK();
1189 
1190 	free(ifgl, M_TEMP);
1191 
1192 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1193 
1194 	return (0);
1195 }
1196 
1197 /*
1198  * Remove an interface from all groups
1199  */
1200 static void
1201 if_delgroups(struct ifnet *ifp)
1202 {
1203 	struct ifg_list		*ifgl;
1204 	struct ifg_member	*ifgm;
1205 	char groupname[IFNAMSIZ];
1206 
1207 	IFNET_WLOCK();
1208 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1209 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1210 
1211 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1212 
1213 		IF_ADDR_WLOCK(ifp);
1214 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1215 		IF_ADDR_WUNLOCK(ifp);
1216 
1217 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1218 			if (ifgm->ifgm_ifp == ifp)
1219 				break;
1220 
1221 		if (ifgm != NULL) {
1222 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1223 			    ifgm_next);
1224 			free(ifgm, M_TEMP);
1225 		}
1226 
1227 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1228 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1229 			IFNET_WUNLOCK();
1230 			EVENTHANDLER_INVOKE(group_detach_event,
1231 			    ifgl->ifgl_group);
1232 			free(ifgl->ifgl_group, M_TEMP);
1233 		} else
1234 			IFNET_WUNLOCK();
1235 
1236 		free(ifgl, M_TEMP);
1237 
1238 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1239 
1240 		IFNET_WLOCK();
1241 	}
1242 	IFNET_WUNLOCK();
1243 }
1244 
1245 /*
1246  * Stores all groups from an interface in memory pointed
1247  * to by data
1248  */
1249 static int
1250 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1251 {
1252 	int			 len, error;
1253 	struct ifg_list		*ifgl;
1254 	struct ifg_req		 ifgrq, *ifgp;
1255 	struct ifgroupreq	*ifgr = data;
1256 
1257 	if (ifgr->ifgr_len == 0) {
1258 		IF_ADDR_RLOCK(ifp);
1259 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1260 			ifgr->ifgr_len += sizeof(struct ifg_req);
1261 		IF_ADDR_RUNLOCK(ifp);
1262 		return (0);
1263 	}
1264 
1265 	len = ifgr->ifgr_len;
1266 	ifgp = ifgr->ifgr_groups;
1267 	/* XXX: wire */
1268 	IF_ADDR_RLOCK(ifp);
1269 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1270 		if (len < sizeof(ifgrq)) {
1271 			IF_ADDR_RUNLOCK(ifp);
1272 			return (EINVAL);
1273 		}
1274 		bzero(&ifgrq, sizeof ifgrq);
1275 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1276 		    sizeof(ifgrq.ifgrq_group));
1277 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1278 		    	IF_ADDR_RUNLOCK(ifp);
1279 			return (error);
1280 		}
1281 		len -= sizeof(ifgrq);
1282 		ifgp++;
1283 	}
1284 	IF_ADDR_RUNLOCK(ifp);
1285 
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Stores all members of a group in memory pointed to by data
1291  */
1292 static int
1293 if_getgroupmembers(struct ifgroupreq *data)
1294 {
1295 	struct ifgroupreq	*ifgr = data;
1296 	struct ifg_group	*ifg;
1297 	struct ifg_member	*ifgm;
1298 	struct ifg_req		 ifgrq, *ifgp;
1299 	int			 len, error;
1300 
1301 	IFNET_RLOCK();
1302 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1303 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1304 			break;
1305 	if (ifg == NULL) {
1306 		IFNET_RUNLOCK();
1307 		return (ENOENT);
1308 	}
1309 
1310 	if (ifgr->ifgr_len == 0) {
1311 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1312 			ifgr->ifgr_len += sizeof(ifgrq);
1313 		IFNET_RUNLOCK();
1314 		return (0);
1315 	}
1316 
1317 	len = ifgr->ifgr_len;
1318 	ifgp = ifgr->ifgr_groups;
1319 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1320 		if (len < sizeof(ifgrq)) {
1321 			IFNET_RUNLOCK();
1322 			return (EINVAL);
1323 		}
1324 		bzero(&ifgrq, sizeof ifgrq);
1325 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1326 		    sizeof(ifgrq.ifgrq_member));
1327 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1328 			IFNET_RUNLOCK();
1329 			return (error);
1330 		}
1331 		len -= sizeof(ifgrq);
1332 		ifgp++;
1333 	}
1334 	IFNET_RUNLOCK();
1335 
1336 	return (0);
1337 }
1338 
1339 /*
1340  * Delete Routes for a Network Interface
1341  *
1342  * Called for each routing entry via the rnh->rnh_walktree() call above
1343  * to delete all route entries referencing a detaching network interface.
1344  *
1345  * Arguments:
1346  *	rn	pointer to node in the routing table
1347  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1348  *
1349  * Returns:
1350  *	0	successful
1351  *	errno	failed - reason indicated
1352  *
1353  */
1354 static int
1355 if_rtdel(struct radix_node *rn, void *arg)
1356 {
1357 	struct rtentry	*rt = (struct rtentry *)rn;
1358 	struct ifnet	*ifp = arg;
1359 	int		err;
1360 
1361 	if (rt->rt_ifp == ifp) {
1362 
1363 		/*
1364 		 * Protect (sorta) against walktree recursion problems
1365 		 * with cloned routes
1366 		 */
1367 		if ((rt->rt_flags & RTF_UP) == 0)
1368 			return (0);
1369 
1370 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1371 				rt_mask(rt),
1372 				rt->rt_flags|RTF_RNH_LOCKED|RTF_PINNED,
1373 				(struct rtentry **) NULL, rt->rt_fibnum);
1374 		if (err) {
1375 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1376 		}
1377 	}
1378 
1379 	return (0);
1380 }
1381 
1382 /*
1383  * Wrapper functions for struct ifnet address list locking macros.  These are
1384  * used by kernel modules to avoid encoding programming interface or binary
1385  * interface assumptions that may be violated when kernel-internal locking
1386  * approaches change.
1387  */
1388 void
1389 if_addr_rlock(struct ifnet *ifp)
1390 {
1391 
1392 	IF_ADDR_RLOCK(ifp);
1393 }
1394 
1395 void
1396 if_addr_runlock(struct ifnet *ifp)
1397 {
1398 
1399 	IF_ADDR_RUNLOCK(ifp);
1400 }
1401 
1402 void
1403 if_maddr_rlock(struct ifnet *ifp)
1404 {
1405 
1406 	IF_ADDR_RLOCK(ifp);
1407 }
1408 
1409 void
1410 if_maddr_runlock(struct ifnet *ifp)
1411 {
1412 
1413 	IF_ADDR_RUNLOCK(ifp);
1414 }
1415 
1416 /*
1417  * Initialization, destruction and refcounting functions for ifaddrs.
1418  */
1419 struct ifaddr *
1420 ifa_alloc(size_t size, int flags)
1421 {
1422 	struct ifaddr *ifa;
1423 
1424 	KASSERT(size >= sizeof(struct ifaddr),
1425 	    ("%s: invalid size %zu", __func__, size));
1426 
1427 	ifa = malloc(size, M_IFADDR, M_ZERO | flags);
1428 	if (ifa == NULL)
1429 		return (NULL);
1430 
1431 	if ((ifa->ifa_opackets = counter_u64_alloc(flags)) == NULL)
1432 		goto fail;
1433 	if ((ifa->ifa_ipackets = counter_u64_alloc(flags)) == NULL)
1434 		goto fail;
1435 	if ((ifa->ifa_obytes = counter_u64_alloc(flags)) == NULL)
1436 		goto fail;
1437 	if ((ifa->ifa_ibytes = counter_u64_alloc(flags)) == NULL)
1438 		goto fail;
1439 
1440 	refcount_init(&ifa->ifa_refcnt, 1);
1441 
1442 	return (ifa);
1443 
1444 fail:
1445 	/* free(NULL) is okay */
1446 	counter_u64_free(ifa->ifa_opackets);
1447 	counter_u64_free(ifa->ifa_ipackets);
1448 	counter_u64_free(ifa->ifa_obytes);
1449 	counter_u64_free(ifa->ifa_ibytes);
1450 	free(ifa, M_IFADDR);
1451 
1452 	return (NULL);
1453 }
1454 
1455 void
1456 ifa_ref(struct ifaddr *ifa)
1457 {
1458 
1459 	refcount_acquire(&ifa->ifa_refcnt);
1460 }
1461 
1462 void
1463 ifa_free(struct ifaddr *ifa)
1464 {
1465 
1466 	if (refcount_release(&ifa->ifa_refcnt)) {
1467 		counter_u64_free(ifa->ifa_opackets);
1468 		counter_u64_free(ifa->ifa_ipackets);
1469 		counter_u64_free(ifa->ifa_obytes);
1470 		counter_u64_free(ifa->ifa_ibytes);
1471 		free(ifa, M_IFADDR);
1472 	}
1473 }
1474 
1475 int
1476 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1477 {
1478 	int error = 0;
1479 	struct rtentry *rt = NULL;
1480 	struct rt_addrinfo info;
1481 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1482 
1483 	bzero(&info, sizeof(info));
1484 	info.rti_ifp = V_loif;
1485 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1486 	info.rti_info[RTAX_DST] = ia;
1487 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1488 	error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1489 
1490 	if (error == 0 && rt != NULL) {
1491 		RT_LOCK(rt);
1492 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1493 			ifa->ifa_ifp->if_type;
1494 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1495 			ifa->ifa_ifp->if_index;
1496 		RT_REMREF(rt);
1497 		RT_UNLOCK(rt);
1498 	} else if (error != 0)
1499 		log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1500 
1501 	return (error);
1502 }
1503 
1504 int
1505 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1506 {
1507 	int error = 0;
1508 	struct rt_addrinfo info;
1509 	struct sockaddr_dl null_sdl;
1510 
1511 	bzero(&null_sdl, sizeof(null_sdl));
1512 	null_sdl.sdl_len = sizeof(null_sdl);
1513 	null_sdl.sdl_family = AF_LINK;
1514 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1515 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1516 	bzero(&info, sizeof(info));
1517 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1518 	info.rti_info[RTAX_DST] = ia;
1519 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1520 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1521 
1522 	if (error != 0)
1523 		log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1524 
1525 	return (error);
1526 }
1527 
1528 int
1529 ifa_switch_loopback_route(struct ifaddr *ifa, struct sockaddr *sa)
1530 {
1531 	struct rtentry *rt;
1532 
1533 	rt = rtalloc1_fib(sa, 0, 0, 0);
1534 	if (rt == NULL) {
1535 		log(LOG_DEBUG, "%s: fail", __func__);
1536 		return (EHOSTUNREACH);
1537 	}
1538 	((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1539 	    ifa->ifa_ifp->if_type;
1540 	((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1541 	    ifa->ifa_ifp->if_index;
1542 	RTFREE_LOCKED(rt);
1543 
1544 	return (0);
1545 }
1546 
1547 /*
1548  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1549  * structs used to represent other address families, it is necessary
1550  * to perform a different comparison.
1551  */
1552 
1553 #define	sa_equal(a1, a2)	\
1554 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1555 
1556 #define	sa_dl_equal(a1, a2)	\
1557 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1558 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1559 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1560 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1561 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1562 
1563 /*
1564  * Locate an interface based on a complete address.
1565  */
1566 /*ARGSUSED*/
1567 static struct ifaddr *
1568 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1569 {
1570 	struct ifnet *ifp;
1571 	struct ifaddr *ifa;
1572 
1573 	IFNET_RLOCK_NOSLEEP();
1574 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1575 		IF_ADDR_RLOCK(ifp);
1576 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1577 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1578 				continue;
1579 			if (sa_equal(addr, ifa->ifa_addr)) {
1580 				if (getref)
1581 					ifa_ref(ifa);
1582 				IF_ADDR_RUNLOCK(ifp);
1583 				goto done;
1584 			}
1585 			/* IP6 doesn't have broadcast */
1586 			if ((ifp->if_flags & IFF_BROADCAST) &&
1587 			    ifa->ifa_broadaddr &&
1588 			    ifa->ifa_broadaddr->sa_len != 0 &&
1589 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1590 				if (getref)
1591 					ifa_ref(ifa);
1592 				IF_ADDR_RUNLOCK(ifp);
1593 				goto done;
1594 			}
1595 		}
1596 		IF_ADDR_RUNLOCK(ifp);
1597 	}
1598 	ifa = NULL;
1599 done:
1600 	IFNET_RUNLOCK_NOSLEEP();
1601 	return (ifa);
1602 }
1603 
1604 struct ifaddr *
1605 ifa_ifwithaddr(struct sockaddr *addr)
1606 {
1607 
1608 	return (ifa_ifwithaddr_internal(addr, 1));
1609 }
1610 
1611 int
1612 ifa_ifwithaddr_check(struct sockaddr *addr)
1613 {
1614 
1615 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1616 }
1617 
1618 /*
1619  * Locate an interface based on the broadcast address.
1620  */
1621 /* ARGSUSED */
1622 struct ifaddr *
1623 ifa_ifwithbroadaddr(struct sockaddr *addr)
1624 {
1625 	struct ifnet *ifp;
1626 	struct ifaddr *ifa;
1627 
1628 	IFNET_RLOCK_NOSLEEP();
1629 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1630 		IF_ADDR_RLOCK(ifp);
1631 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1632 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1633 				continue;
1634 			if ((ifp->if_flags & IFF_BROADCAST) &&
1635 			    ifa->ifa_broadaddr &&
1636 			    ifa->ifa_broadaddr->sa_len != 0 &&
1637 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1638 				ifa_ref(ifa);
1639 				IF_ADDR_RUNLOCK(ifp);
1640 				goto done;
1641 			}
1642 		}
1643 		IF_ADDR_RUNLOCK(ifp);
1644 	}
1645 	ifa = NULL;
1646 done:
1647 	IFNET_RUNLOCK_NOSLEEP();
1648 	return (ifa);
1649 }
1650 
1651 /*
1652  * Locate the point to point interface with a given destination address.
1653  */
1654 /*ARGSUSED*/
1655 struct ifaddr *
1656 ifa_ifwithdstaddr(struct sockaddr *addr)
1657 {
1658 	struct ifnet *ifp;
1659 	struct ifaddr *ifa;
1660 
1661 	IFNET_RLOCK_NOSLEEP();
1662 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1663 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1664 			continue;
1665 		IF_ADDR_RLOCK(ifp);
1666 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1667 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1668 				continue;
1669 			if (ifa->ifa_dstaddr != NULL &&
1670 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1671 				ifa_ref(ifa);
1672 				IF_ADDR_RUNLOCK(ifp);
1673 				goto done;
1674 			}
1675 		}
1676 		IF_ADDR_RUNLOCK(ifp);
1677 	}
1678 	ifa = NULL;
1679 done:
1680 	IFNET_RUNLOCK_NOSLEEP();
1681 	return (ifa);
1682 }
1683 
1684 /*
1685  * Find an interface on a specific network.  If many, choice
1686  * is most specific found.
1687  */
1688 struct ifaddr *
1689 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1690 {
1691 	struct ifnet *ifp;
1692 	struct ifaddr *ifa;
1693 	struct ifaddr *ifa_maybe = NULL;
1694 	u_int af = addr->sa_family;
1695 	char *addr_data = addr->sa_data, *cplim;
1696 
1697 	/*
1698 	 * AF_LINK addresses can be looked up directly by their index number,
1699 	 * so do that if we can.
1700 	 */
1701 	if (af == AF_LINK) {
1702 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1703 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1704 		return (ifaddr_byindex(sdl->sdl_index));
1705 	}
1706 
1707 	/*
1708 	 * Scan though each interface, looking for ones that have addresses
1709 	 * in this address family.  Maintain a reference on ifa_maybe once
1710 	 * we find one, as we release the IF_ADDR_RLOCK() that kept it stable
1711 	 * when we move onto the next interface.
1712 	 */
1713 	IFNET_RLOCK_NOSLEEP();
1714 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1715 		IF_ADDR_RLOCK(ifp);
1716 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1717 			char *cp, *cp2, *cp3;
1718 
1719 			if (ifa->ifa_addr->sa_family != af)
1720 next:				continue;
1721 			if (af == AF_INET &&
1722 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1723 				/*
1724 				 * This is a bit broken as it doesn't
1725 				 * take into account that the remote end may
1726 				 * be a single node in the network we are
1727 				 * looking for.
1728 				 * The trouble is that we don't know the
1729 				 * netmask for the remote end.
1730 				 */
1731 				if (ifa->ifa_dstaddr != NULL &&
1732 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1733 					ifa_ref(ifa);
1734 					IF_ADDR_RUNLOCK(ifp);
1735 					goto done;
1736 				}
1737 			} else {
1738 				/*
1739 				 * if we have a special address handler,
1740 				 * then use it instead of the generic one.
1741 				 */
1742 				if (ifa->ifa_claim_addr) {
1743 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1744 						ifa_ref(ifa);
1745 						IF_ADDR_RUNLOCK(ifp);
1746 						goto done;
1747 					}
1748 					continue;
1749 				}
1750 
1751 				/*
1752 				 * Scan all the bits in the ifa's address.
1753 				 * If a bit dissagrees with what we are
1754 				 * looking for, mask it with the netmask
1755 				 * to see if it really matters.
1756 				 * (A byte at a time)
1757 				 */
1758 				if (ifa->ifa_netmask == 0)
1759 					continue;
1760 				cp = addr_data;
1761 				cp2 = ifa->ifa_addr->sa_data;
1762 				cp3 = ifa->ifa_netmask->sa_data;
1763 				cplim = ifa->ifa_netmask->sa_len
1764 					+ (char *)ifa->ifa_netmask;
1765 				while (cp3 < cplim)
1766 					if ((*cp++ ^ *cp2++) & *cp3++)
1767 						goto next; /* next address! */
1768 				/*
1769 				 * If the netmask of what we just found
1770 				 * is more specific than what we had before
1771 				 * (if we had one), or if the virtual status
1772 				 * of new prefix is better than of the old one,
1773 				 * then remember the new one before continuing
1774 				 * to search for an even better one.
1775 				 */
1776 				if (ifa_maybe == NULL ||
1777 				    ifa_preferred(ifa_maybe, ifa) ||
1778 				    rn_refines((caddr_t)ifa->ifa_netmask,
1779 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1780 					if (ifa_maybe != NULL)
1781 						ifa_free(ifa_maybe);
1782 					ifa_maybe = ifa;
1783 					ifa_ref(ifa_maybe);
1784 				}
1785 			}
1786 		}
1787 		IF_ADDR_RUNLOCK(ifp);
1788 	}
1789 	ifa = ifa_maybe;
1790 	ifa_maybe = NULL;
1791 done:
1792 	IFNET_RUNLOCK_NOSLEEP();
1793 	if (ifa_maybe != NULL)
1794 		ifa_free(ifa_maybe);
1795 	return (ifa);
1796 }
1797 
1798 /*
1799  * Find an interface address specific to an interface best matching
1800  * a given address.
1801  */
1802 struct ifaddr *
1803 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1804 {
1805 	struct ifaddr *ifa;
1806 	char *cp, *cp2, *cp3;
1807 	char *cplim;
1808 	struct ifaddr *ifa_maybe = NULL;
1809 	u_int af = addr->sa_family;
1810 
1811 	if (af >= AF_MAX)
1812 		return (NULL);
1813 	IF_ADDR_RLOCK(ifp);
1814 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1815 		if (ifa->ifa_addr->sa_family != af)
1816 			continue;
1817 		if (ifa_maybe == NULL)
1818 			ifa_maybe = ifa;
1819 		if (ifa->ifa_netmask == 0) {
1820 			if (sa_equal(addr, ifa->ifa_addr) ||
1821 			    (ifa->ifa_dstaddr &&
1822 			    sa_equal(addr, ifa->ifa_dstaddr)))
1823 				goto done;
1824 			continue;
1825 		}
1826 		if (ifp->if_flags & IFF_POINTOPOINT) {
1827 			if (sa_equal(addr, ifa->ifa_dstaddr))
1828 				goto done;
1829 		} else {
1830 			cp = addr->sa_data;
1831 			cp2 = ifa->ifa_addr->sa_data;
1832 			cp3 = ifa->ifa_netmask->sa_data;
1833 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1834 			for (; cp3 < cplim; cp3++)
1835 				if ((*cp++ ^ *cp2++) & *cp3)
1836 					break;
1837 			if (cp3 == cplim)
1838 				goto done;
1839 		}
1840 	}
1841 	ifa = ifa_maybe;
1842 done:
1843 	if (ifa != NULL)
1844 		ifa_ref(ifa);
1845 	IF_ADDR_RUNLOCK(ifp);
1846 	return (ifa);
1847 }
1848 
1849 /*
1850  * See whether new ifa is better than current one:
1851  * 1) A non-virtual one is preferred over virtual.
1852  * 2) A virtual in master state preferred over any other state.
1853  *
1854  * Used in several address selecting functions.
1855  */
1856 int
1857 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
1858 {
1859 
1860 	return (cur->ifa_carp && (!next->ifa_carp ||
1861 	    ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
1862 }
1863 
1864 #include <net/if_llatbl.h>
1865 
1866 /*
1867  * Default action when installing a route with a Link Level gateway.
1868  * Lookup an appropriate real ifa to point to.
1869  * This should be moved to /sys/net/link.c eventually.
1870  */
1871 static void
1872 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1873 {
1874 	struct ifaddr *ifa, *oifa;
1875 	struct sockaddr *dst;
1876 	struct ifnet *ifp;
1877 
1878 	RT_LOCK_ASSERT(rt);
1879 
1880 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1881 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1882 		return;
1883 	ifa = ifaof_ifpforaddr(dst, ifp);
1884 	if (ifa) {
1885 		oifa = rt->rt_ifa;
1886 		rt->rt_ifa = ifa;
1887 		ifa_free(oifa);
1888 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1889 			ifa->ifa_rtrequest(cmd, rt, info);
1890 	}
1891 }
1892 
1893 /*
1894  * Mark an interface down and notify protocols of
1895  * the transition.
1896  */
1897 static void
1898 if_unroute(struct ifnet *ifp, int flag, int fam)
1899 {
1900 	struct ifaddr *ifa;
1901 
1902 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1903 
1904 	ifp->if_flags &= ~flag;
1905 	getmicrotime(&ifp->if_lastchange);
1906 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1907 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1908 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1909 	ifp->if_qflush(ifp);
1910 
1911 	if (ifp->if_carp)
1912 		(*carp_linkstate_p)(ifp);
1913 	rt_ifmsg(ifp);
1914 }
1915 
1916 /*
1917  * Mark an interface up and notify protocols of
1918  * the transition.
1919  */
1920 static void
1921 if_route(struct ifnet *ifp, int flag, int fam)
1922 {
1923 	struct ifaddr *ifa;
1924 
1925 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1926 
1927 	ifp->if_flags |= flag;
1928 	getmicrotime(&ifp->if_lastchange);
1929 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1930 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1931 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1932 	if (ifp->if_carp)
1933 		(*carp_linkstate_p)(ifp);
1934 	rt_ifmsg(ifp);
1935 #ifdef INET6
1936 	in6_if_up(ifp);
1937 #endif
1938 }
1939 
1940 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1941 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1942 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1943 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1944 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
1945 int	(*vlan_setcookie_p)(struct ifnet *, void *);
1946 void	*(*vlan_cookie_p)(struct ifnet *);
1947 
1948 /*
1949  * Handle a change in the interface link state. To avoid LORs
1950  * between driver lock and upper layer locks, as well as possible
1951  * recursions, we post event to taskqueue, and all job
1952  * is done in static do_link_state_change().
1953  */
1954 void
1955 if_link_state_change(struct ifnet *ifp, int link_state)
1956 {
1957 	/* Return if state hasn't changed. */
1958 	if (ifp->if_link_state == link_state)
1959 		return;
1960 
1961 	ifp->if_link_state = link_state;
1962 
1963 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1964 }
1965 
1966 static void
1967 do_link_state_change(void *arg, int pending)
1968 {
1969 	struct ifnet *ifp = (struct ifnet *)arg;
1970 	int link_state = ifp->if_link_state;
1971 	CURVNET_SET(ifp->if_vnet);
1972 
1973 	/* Notify that the link state has changed. */
1974 	rt_ifmsg(ifp);
1975 	if (ifp->if_vlantrunk != NULL)
1976 		(*vlan_link_state_p)(ifp);
1977 
1978 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1979 	    IFP2AC(ifp)->ac_netgraph != NULL)
1980 		(*ng_ether_link_state_p)(ifp, link_state);
1981 	if (ifp->if_carp)
1982 		(*carp_linkstate_p)(ifp);
1983 	if (ifp->if_bridge)
1984 		(*bridge_linkstate_p)(ifp);
1985 	if (ifp->if_lagg)
1986 		(*lagg_linkstate_p)(ifp, link_state);
1987 
1988 	if (IS_DEFAULT_VNET(curvnet))
1989 		devctl_notify("IFNET", ifp->if_xname,
1990 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1991 		    NULL);
1992 	if (pending > 1)
1993 		if_printf(ifp, "%d link states coalesced\n", pending);
1994 	if (log_link_state_change)
1995 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1996 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1997 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
1998 	CURVNET_RESTORE();
1999 }
2000 
2001 /*
2002  * Mark an interface down and notify protocols of
2003  * the transition.
2004  */
2005 void
2006 if_down(struct ifnet *ifp)
2007 {
2008 
2009 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
2010 }
2011 
2012 /*
2013  * Mark an interface up and notify protocols of
2014  * the transition.
2015  */
2016 void
2017 if_up(struct ifnet *ifp)
2018 {
2019 
2020 	if_route(ifp, IFF_UP, AF_UNSPEC);
2021 }
2022 
2023 /*
2024  * Flush an interface queue.
2025  */
2026 void
2027 if_qflush(struct ifnet *ifp)
2028 {
2029 	struct mbuf *m, *n;
2030 	struct ifaltq *ifq;
2031 
2032 	ifq = &ifp->if_snd;
2033 	IFQ_LOCK(ifq);
2034 #ifdef ALTQ
2035 	if (ALTQ_IS_ENABLED(ifq))
2036 		ALTQ_PURGE(ifq);
2037 #endif
2038 	n = ifq->ifq_head;
2039 	while ((m = n) != 0) {
2040 		n = m->m_act;
2041 		m_freem(m);
2042 	}
2043 	ifq->ifq_head = 0;
2044 	ifq->ifq_tail = 0;
2045 	ifq->ifq_len = 0;
2046 	IFQ_UNLOCK(ifq);
2047 }
2048 
2049 /*
2050  * Map interface name to interface structure pointer, with or without
2051  * returning a reference.
2052  */
2053 struct ifnet *
2054 ifunit_ref(const char *name)
2055 {
2056 	struct ifnet *ifp;
2057 
2058 	IFNET_RLOCK_NOSLEEP();
2059 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2060 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2061 		    !(ifp->if_flags & IFF_DYING))
2062 			break;
2063 	}
2064 	if (ifp != NULL)
2065 		if_ref(ifp);
2066 	IFNET_RUNLOCK_NOSLEEP();
2067 	return (ifp);
2068 }
2069 
2070 struct ifnet *
2071 ifunit(const char *name)
2072 {
2073 	struct ifnet *ifp;
2074 
2075 	IFNET_RLOCK_NOSLEEP();
2076 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2077 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2078 			break;
2079 	}
2080 	IFNET_RUNLOCK_NOSLEEP();
2081 	return (ifp);
2082 }
2083 
2084 /*
2085  * Hardware specific interface ioctls.
2086  */
2087 static int
2088 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2089 {
2090 	struct ifreq *ifr;
2091 	struct ifstat *ifs;
2092 	int error = 0;
2093 	int new_flags, temp_flags;
2094 	size_t namelen, onamelen;
2095 	size_t descrlen;
2096 	char *descrbuf, *odescrbuf;
2097 	char new_name[IFNAMSIZ];
2098 	struct ifaddr *ifa;
2099 	struct sockaddr_dl *sdl;
2100 
2101 	ifr = (struct ifreq *)data;
2102 	switch (cmd) {
2103 	case SIOCGIFINDEX:
2104 		ifr->ifr_index = ifp->if_index;
2105 		break;
2106 
2107 	case SIOCGIFFLAGS:
2108 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2109 		ifr->ifr_flags = temp_flags & 0xffff;
2110 		ifr->ifr_flagshigh = temp_flags >> 16;
2111 		break;
2112 
2113 	case SIOCGIFCAP:
2114 		ifr->ifr_reqcap = ifp->if_capabilities;
2115 		ifr->ifr_curcap = ifp->if_capenable;
2116 		break;
2117 
2118 #ifdef MAC
2119 	case SIOCGIFMAC:
2120 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2121 		break;
2122 #endif
2123 
2124 	case SIOCGIFMETRIC:
2125 		ifr->ifr_metric = ifp->if_metric;
2126 		break;
2127 
2128 	case SIOCGIFMTU:
2129 		ifr->ifr_mtu = ifp->if_mtu;
2130 		break;
2131 
2132 	case SIOCGIFPHYS:
2133 		ifr->ifr_phys = ifp->if_physical;
2134 		break;
2135 
2136 	case SIOCGIFDESCR:
2137 		error = 0;
2138 		sx_slock(&ifdescr_sx);
2139 		if (ifp->if_description == NULL)
2140 			error = ENOMSG;
2141 		else {
2142 			/* space for terminating nul */
2143 			descrlen = strlen(ifp->if_description) + 1;
2144 			if (ifr->ifr_buffer.length < descrlen)
2145 				ifr->ifr_buffer.buffer = NULL;
2146 			else
2147 				error = copyout(ifp->if_description,
2148 				    ifr->ifr_buffer.buffer, descrlen);
2149 			ifr->ifr_buffer.length = descrlen;
2150 		}
2151 		sx_sunlock(&ifdescr_sx);
2152 		break;
2153 
2154 	case SIOCSIFDESCR:
2155 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2156 		if (error)
2157 			return (error);
2158 
2159 		/*
2160 		 * Copy only (length-1) bytes to make sure that
2161 		 * if_description is always nul terminated.  The
2162 		 * length parameter is supposed to count the
2163 		 * terminating nul in.
2164 		 */
2165 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2166 			return (ENAMETOOLONG);
2167 		else if (ifr->ifr_buffer.length == 0)
2168 			descrbuf = NULL;
2169 		else {
2170 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2171 			    M_WAITOK | M_ZERO);
2172 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2173 			    ifr->ifr_buffer.length - 1);
2174 			if (error) {
2175 				free(descrbuf, M_IFDESCR);
2176 				break;
2177 			}
2178 		}
2179 
2180 		sx_xlock(&ifdescr_sx);
2181 		odescrbuf = ifp->if_description;
2182 		ifp->if_description = descrbuf;
2183 		sx_xunlock(&ifdescr_sx);
2184 
2185 		getmicrotime(&ifp->if_lastchange);
2186 		free(odescrbuf, M_IFDESCR);
2187 		break;
2188 
2189 	case SIOCGIFFIB:
2190 		ifr->ifr_fib = ifp->if_fib;
2191 		break;
2192 
2193 	case SIOCSIFFIB:
2194 		error = priv_check(td, PRIV_NET_SETIFFIB);
2195 		if (error)
2196 			return (error);
2197 		if (ifr->ifr_fib >= rt_numfibs)
2198 			return (EINVAL);
2199 
2200 		ifp->if_fib = ifr->ifr_fib;
2201 		break;
2202 
2203 	case SIOCSIFFLAGS:
2204 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2205 		if (error)
2206 			return (error);
2207 		/*
2208 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2209 		 * check, so we don't need special handling here yet.
2210 		 */
2211 		new_flags = (ifr->ifr_flags & 0xffff) |
2212 		    (ifr->ifr_flagshigh << 16);
2213 		if (ifp->if_flags & IFF_UP &&
2214 		    (new_flags & IFF_UP) == 0) {
2215 			if_down(ifp);
2216 		} else if (new_flags & IFF_UP &&
2217 		    (ifp->if_flags & IFF_UP) == 0) {
2218 			if_up(ifp);
2219 		}
2220 		/* See if permanently promiscuous mode bit is about to flip */
2221 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2222 			if (new_flags & IFF_PPROMISC)
2223 				ifp->if_flags |= IFF_PROMISC;
2224 			else if (ifp->if_pcount == 0)
2225 				ifp->if_flags &= ~IFF_PROMISC;
2226 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2227 			    ifp->if_xname,
2228 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2229 		}
2230 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2231 			(new_flags &~ IFF_CANTCHANGE);
2232 		if (ifp->if_ioctl) {
2233 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2234 		}
2235 		getmicrotime(&ifp->if_lastchange);
2236 		break;
2237 
2238 	case SIOCSIFCAP:
2239 		error = priv_check(td, PRIV_NET_SETIFCAP);
2240 		if (error)
2241 			return (error);
2242 		if (ifp->if_ioctl == NULL)
2243 			return (EOPNOTSUPP);
2244 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2245 			return (EINVAL);
2246 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2247 		if (error == 0)
2248 			getmicrotime(&ifp->if_lastchange);
2249 		break;
2250 
2251 #ifdef MAC
2252 	case SIOCSIFMAC:
2253 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2254 		break;
2255 #endif
2256 
2257 	case SIOCSIFNAME:
2258 		error = priv_check(td, PRIV_NET_SETIFNAME);
2259 		if (error)
2260 			return (error);
2261 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2262 		if (error != 0)
2263 			return (error);
2264 		if (new_name[0] == '\0')
2265 			return (EINVAL);
2266 		if (ifunit(new_name) != NULL)
2267 			return (EEXIST);
2268 
2269 		/*
2270 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2271 		 * and there are numerous other races with the
2272 		 * ifunit() checks not being atomic with namespace
2273 		 * changes (renames, vmoves, if_attach, etc).
2274 		 */
2275 		ifp->if_flags |= IFF_RENAMING;
2276 
2277 		/* Announce the departure of the interface. */
2278 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2279 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2280 
2281 		log(LOG_INFO, "%s: changing name to '%s'\n",
2282 		    ifp->if_xname, new_name);
2283 
2284 		IF_ADDR_WLOCK(ifp);
2285 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2286 		ifa = ifp->if_addr;
2287 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2288 		namelen = strlen(new_name);
2289 		onamelen = sdl->sdl_nlen;
2290 		/*
2291 		 * Move the address if needed.  This is safe because we
2292 		 * allocate space for a name of length IFNAMSIZ when we
2293 		 * create this in if_attach().
2294 		 */
2295 		if (namelen != onamelen) {
2296 			bcopy(sdl->sdl_data + onamelen,
2297 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2298 		}
2299 		bcopy(new_name, sdl->sdl_data, namelen);
2300 		sdl->sdl_nlen = namelen;
2301 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2302 		bzero(sdl->sdl_data, onamelen);
2303 		while (namelen != 0)
2304 			sdl->sdl_data[--namelen] = 0xff;
2305 		IF_ADDR_WUNLOCK(ifp);
2306 
2307 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2308 		/* Announce the return of the interface. */
2309 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2310 
2311 		ifp->if_flags &= ~IFF_RENAMING;
2312 		break;
2313 
2314 #ifdef VIMAGE
2315 	case SIOCSIFVNET:
2316 		error = priv_check(td, PRIV_NET_SETIFVNET);
2317 		if (error)
2318 			return (error);
2319 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2320 		break;
2321 #endif
2322 
2323 	case SIOCSIFMETRIC:
2324 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2325 		if (error)
2326 			return (error);
2327 		ifp->if_metric = ifr->ifr_metric;
2328 		getmicrotime(&ifp->if_lastchange);
2329 		break;
2330 
2331 	case SIOCSIFPHYS:
2332 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2333 		if (error)
2334 			return (error);
2335 		if (ifp->if_ioctl == NULL)
2336 			return (EOPNOTSUPP);
2337 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2338 		if (error == 0)
2339 			getmicrotime(&ifp->if_lastchange);
2340 		break;
2341 
2342 	case SIOCSIFMTU:
2343 	{
2344 		u_long oldmtu = ifp->if_mtu;
2345 
2346 		error = priv_check(td, PRIV_NET_SETIFMTU);
2347 		if (error)
2348 			return (error);
2349 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2350 			return (EINVAL);
2351 		if (ifp->if_ioctl == NULL)
2352 			return (EOPNOTSUPP);
2353 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2354 		if (error == 0) {
2355 			getmicrotime(&ifp->if_lastchange);
2356 			rt_ifmsg(ifp);
2357 		}
2358 		/*
2359 		 * If the link MTU changed, do network layer specific procedure.
2360 		 */
2361 		if (ifp->if_mtu != oldmtu) {
2362 #ifdef INET6
2363 			nd6_setmtu(ifp);
2364 #endif
2365 		}
2366 		break;
2367 	}
2368 
2369 	case SIOCADDMULTI:
2370 	case SIOCDELMULTI:
2371 		if (cmd == SIOCADDMULTI)
2372 			error = priv_check(td, PRIV_NET_ADDMULTI);
2373 		else
2374 			error = priv_check(td, PRIV_NET_DELMULTI);
2375 		if (error)
2376 			return (error);
2377 
2378 		/* Don't allow group membership on non-multicast interfaces. */
2379 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2380 			return (EOPNOTSUPP);
2381 
2382 		/* Don't let users screw up protocols' entries. */
2383 		if (ifr->ifr_addr.sa_family != AF_LINK)
2384 			return (EINVAL);
2385 
2386 		if (cmd == SIOCADDMULTI) {
2387 			struct ifmultiaddr *ifma;
2388 
2389 			/*
2390 			 * Userland is only permitted to join groups once
2391 			 * via the if_addmulti() KPI, because it cannot hold
2392 			 * struct ifmultiaddr * between calls. It may also
2393 			 * lose a race while we check if the membership
2394 			 * already exists.
2395 			 */
2396 			IF_ADDR_RLOCK(ifp);
2397 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2398 			IF_ADDR_RUNLOCK(ifp);
2399 			if (ifma != NULL)
2400 				error = EADDRINUSE;
2401 			else
2402 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2403 		} else {
2404 			error = if_delmulti(ifp, &ifr->ifr_addr);
2405 		}
2406 		if (error == 0)
2407 			getmicrotime(&ifp->if_lastchange);
2408 		break;
2409 
2410 	case SIOCSIFPHYADDR:
2411 	case SIOCDIFPHYADDR:
2412 #ifdef INET6
2413 	case SIOCSIFPHYADDR_IN6:
2414 #endif
2415 	case SIOCSLIFPHYADDR:
2416 	case SIOCSIFMEDIA:
2417 	case SIOCSIFGENERIC:
2418 		error = priv_check(td, PRIV_NET_HWIOCTL);
2419 		if (error)
2420 			return (error);
2421 		if (ifp->if_ioctl == NULL)
2422 			return (EOPNOTSUPP);
2423 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2424 		if (error == 0)
2425 			getmicrotime(&ifp->if_lastchange);
2426 		break;
2427 
2428 	case SIOCGIFSTATUS:
2429 		ifs = (struct ifstat *)data;
2430 		ifs->ascii[0] = '\0';
2431 
2432 	case SIOCGIFPSRCADDR:
2433 	case SIOCGIFPDSTADDR:
2434 	case SIOCGLIFPHYADDR:
2435 	case SIOCGIFMEDIA:
2436 	case SIOCGIFGENERIC:
2437 		if (ifp->if_ioctl == NULL)
2438 			return (EOPNOTSUPP);
2439 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2440 		break;
2441 
2442 	case SIOCSIFLLADDR:
2443 		error = priv_check(td, PRIV_NET_SETLLADDR);
2444 		if (error)
2445 			return (error);
2446 		error = if_setlladdr(ifp,
2447 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2448 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2449 		break;
2450 
2451 	case SIOCAIFGROUP:
2452 	{
2453 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2454 
2455 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2456 		if (error)
2457 			return (error);
2458 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2459 			return (error);
2460 		break;
2461 	}
2462 
2463 	case SIOCGIFGROUP:
2464 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2465 			return (error);
2466 		break;
2467 
2468 	case SIOCDIFGROUP:
2469 	{
2470 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2471 
2472 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2473 		if (error)
2474 			return (error);
2475 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2476 			return (error);
2477 		break;
2478 	}
2479 
2480 	default:
2481 		error = ENOIOCTL;
2482 		break;
2483 	}
2484 	return (error);
2485 }
2486 
2487 #ifdef COMPAT_FREEBSD32
2488 struct ifconf32 {
2489 	int32_t	ifc_len;
2490 	union {
2491 		uint32_t	ifcu_buf;
2492 		uint32_t	ifcu_req;
2493 	} ifc_ifcu;
2494 };
2495 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2496 #endif
2497 
2498 /*
2499  * Interface ioctls.
2500  */
2501 int
2502 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2503 {
2504 	struct ifnet *ifp;
2505 	struct ifreq *ifr;
2506 	int error;
2507 	int oif_flags;
2508 
2509 	CURVNET_SET(so->so_vnet);
2510 	switch (cmd) {
2511 	case SIOCGIFCONF:
2512 		error = ifconf(cmd, data);
2513 		CURVNET_RESTORE();
2514 		return (error);
2515 
2516 #ifdef COMPAT_FREEBSD32
2517 	case SIOCGIFCONF32:
2518 		{
2519 			struct ifconf32 *ifc32;
2520 			struct ifconf ifc;
2521 
2522 			ifc32 = (struct ifconf32 *)data;
2523 			ifc.ifc_len = ifc32->ifc_len;
2524 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2525 
2526 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2527 			CURVNET_RESTORE();
2528 			if (error == 0)
2529 				ifc32->ifc_len = ifc.ifc_len;
2530 			return (error);
2531 		}
2532 #endif
2533 	}
2534 	ifr = (struct ifreq *)data;
2535 
2536 	switch (cmd) {
2537 #ifdef VIMAGE
2538 	case SIOCSIFRVNET:
2539 		error = priv_check(td, PRIV_NET_SETIFVNET);
2540 		if (error == 0)
2541 			error = if_vmove_reclaim(td, ifr->ifr_name,
2542 			    ifr->ifr_jid);
2543 		CURVNET_RESTORE();
2544 		return (error);
2545 #endif
2546 	case SIOCIFCREATE:
2547 	case SIOCIFCREATE2:
2548 		error = priv_check(td, PRIV_NET_IFCREATE);
2549 		if (error == 0)
2550 			error = if_clone_create(ifr->ifr_name,
2551 			    sizeof(ifr->ifr_name),
2552 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2553 		CURVNET_RESTORE();
2554 		return (error);
2555 	case SIOCIFDESTROY:
2556 		error = priv_check(td, PRIV_NET_IFDESTROY);
2557 		if (error == 0)
2558 			error = if_clone_destroy(ifr->ifr_name);
2559 		CURVNET_RESTORE();
2560 		return (error);
2561 
2562 	case SIOCIFGCLONERS:
2563 		error = if_clone_list((struct if_clonereq *)data);
2564 		CURVNET_RESTORE();
2565 		return (error);
2566 	case SIOCGIFGMEMB:
2567 		error = if_getgroupmembers((struct ifgroupreq *)data);
2568 		CURVNET_RESTORE();
2569 		return (error);
2570 #if defined(INET) || defined(INET6)
2571 	case SIOCSVH:
2572 	case SIOCGVH:
2573 		if (carp_ioctl_p == NULL)
2574 			error = EPROTONOSUPPORT;
2575 		else
2576 			error = (*carp_ioctl_p)(ifr, cmd, td);
2577 		CURVNET_RESTORE();
2578 		return (error);
2579 #endif
2580 	}
2581 
2582 	ifp = ifunit_ref(ifr->ifr_name);
2583 	if (ifp == NULL) {
2584 		CURVNET_RESTORE();
2585 		return (ENXIO);
2586 	}
2587 
2588 	error = ifhwioctl(cmd, ifp, data, td);
2589 	if (error != ENOIOCTL) {
2590 		if_rele(ifp);
2591 		CURVNET_RESTORE();
2592 		return (error);
2593 	}
2594 
2595 	oif_flags = ifp->if_flags;
2596 	if (so->so_proto == NULL) {
2597 		if_rele(ifp);
2598 		CURVNET_RESTORE();
2599 		return (EOPNOTSUPP);
2600 	}
2601 
2602 	/*
2603 	 * Pass the request on to the socket control method, and if the
2604 	 * latter returns EOPNOTSUPP, directly to the interface.
2605 	 *
2606 	 * Make an exception for the legacy SIOCSIF* requests.  Drivers
2607 	 * trust SIOCSIFADDR et al to come from an already privileged
2608 	 * layer, and do not perform any credentials checks or input
2609 	 * validation.
2610 	 */
2611 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd, data,
2612 	    ifp, td));
2613 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
2614 	    cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2615 	    cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2616 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2617 
2618 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2619 #ifdef INET6
2620 		if (ifp->if_flags & IFF_UP)
2621 			in6_if_up(ifp);
2622 #endif
2623 	}
2624 	if_rele(ifp);
2625 	CURVNET_RESTORE();
2626 	return (error);
2627 }
2628 
2629 /*
2630  * The code common to handling reference counted flags,
2631  * e.g., in ifpromisc() and if_allmulti().
2632  * The "pflag" argument can specify a permanent mode flag to check,
2633  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2634  *
2635  * Only to be used on stack-owned flags, not driver-owned flags.
2636  */
2637 static int
2638 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2639 {
2640 	struct ifreq ifr;
2641 	int error;
2642 	int oldflags, oldcount;
2643 
2644 	/* Sanity checks to catch programming errors */
2645 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2646 	    ("%s: setting driver-owned flag %d", __func__, flag));
2647 
2648 	if (onswitch)
2649 		KASSERT(*refcount >= 0,
2650 		    ("%s: increment negative refcount %d for flag %d",
2651 		    __func__, *refcount, flag));
2652 	else
2653 		KASSERT(*refcount > 0,
2654 		    ("%s: decrement non-positive refcount %d for flag %d",
2655 		    __func__, *refcount, flag));
2656 
2657 	/* In case this mode is permanent, just touch refcount */
2658 	if (ifp->if_flags & pflag) {
2659 		*refcount += onswitch ? 1 : -1;
2660 		return (0);
2661 	}
2662 
2663 	/* Save ifnet parameters for if_ioctl() may fail */
2664 	oldcount = *refcount;
2665 	oldflags = ifp->if_flags;
2666 
2667 	/*
2668 	 * See if we aren't the only and touching refcount is enough.
2669 	 * Actually toggle interface flag if we are the first or last.
2670 	 */
2671 	if (onswitch) {
2672 		if ((*refcount)++)
2673 			return (0);
2674 		ifp->if_flags |= flag;
2675 	} else {
2676 		if (--(*refcount))
2677 			return (0);
2678 		ifp->if_flags &= ~flag;
2679 	}
2680 
2681 	/* Call down the driver since we've changed interface flags */
2682 	if (ifp->if_ioctl == NULL) {
2683 		error = EOPNOTSUPP;
2684 		goto recover;
2685 	}
2686 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2687 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2688 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2689 	if (error)
2690 		goto recover;
2691 	/* Notify userland that interface flags have changed */
2692 	rt_ifmsg(ifp);
2693 	return (0);
2694 
2695 recover:
2696 	/* Recover after driver error */
2697 	*refcount = oldcount;
2698 	ifp->if_flags = oldflags;
2699 	return (error);
2700 }
2701 
2702 /*
2703  * Set/clear promiscuous mode on interface ifp based on the truth value
2704  * of pswitch.  The calls are reference counted so that only the first
2705  * "on" request actually has an effect, as does the final "off" request.
2706  * Results are undefined if the "off" and "on" requests are not matched.
2707  */
2708 int
2709 ifpromisc(struct ifnet *ifp, int pswitch)
2710 {
2711 	int error;
2712 	int oldflags = ifp->if_flags;
2713 
2714 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2715 			   &ifp->if_pcount, pswitch);
2716 	/* If promiscuous mode status has changed, log a message */
2717 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2718 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2719 		    ifp->if_xname,
2720 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2721 	return (error);
2722 }
2723 
2724 /*
2725  * Return interface configuration
2726  * of system.  List may be used
2727  * in later ioctl's (above) to get
2728  * other information.
2729  */
2730 /*ARGSUSED*/
2731 static int
2732 ifconf(u_long cmd, caddr_t data)
2733 {
2734 	struct ifconf *ifc = (struct ifconf *)data;
2735 	struct ifnet *ifp;
2736 	struct ifaddr *ifa;
2737 	struct ifreq ifr;
2738 	struct sbuf *sb;
2739 	int error, full = 0, valid_len, max_len;
2740 
2741 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2742 	max_len = MAXPHYS - 1;
2743 
2744 	/* Prevent hostile input from being able to crash the system */
2745 	if (ifc->ifc_len <= 0)
2746 		return (EINVAL);
2747 
2748 again:
2749 	if (ifc->ifc_len <= max_len) {
2750 		max_len = ifc->ifc_len;
2751 		full = 1;
2752 	}
2753 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2754 	max_len = 0;
2755 	valid_len = 0;
2756 
2757 	IFNET_RLOCK();
2758 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2759 		int addrs;
2760 
2761 		/*
2762 		 * Zero the ifr_name buffer to make sure we don't
2763 		 * disclose the contents of the stack.
2764 		 */
2765 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2766 
2767 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2768 		    >= sizeof(ifr.ifr_name)) {
2769 			sbuf_delete(sb);
2770 			IFNET_RUNLOCK();
2771 			return (ENAMETOOLONG);
2772 		}
2773 
2774 		addrs = 0;
2775 		IF_ADDR_RLOCK(ifp);
2776 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2777 			struct sockaddr *sa = ifa->ifa_addr;
2778 
2779 			if (prison_if(curthread->td_ucred, sa) != 0)
2780 				continue;
2781 			addrs++;
2782 			if (sa->sa_len <= sizeof(*sa)) {
2783 				ifr.ifr_addr = *sa;
2784 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2785 				max_len += sizeof(ifr);
2786 			} else {
2787 				sbuf_bcat(sb, &ifr,
2788 				    offsetof(struct ifreq, ifr_addr));
2789 				max_len += offsetof(struct ifreq, ifr_addr);
2790 				sbuf_bcat(sb, sa, sa->sa_len);
2791 				max_len += sa->sa_len;
2792 			}
2793 
2794 			if (sbuf_error(sb) == 0)
2795 				valid_len = sbuf_len(sb);
2796 		}
2797 		IF_ADDR_RUNLOCK(ifp);
2798 		if (addrs == 0) {
2799 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2800 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2801 			max_len += sizeof(ifr);
2802 
2803 			if (sbuf_error(sb) == 0)
2804 				valid_len = sbuf_len(sb);
2805 		}
2806 	}
2807 	IFNET_RUNLOCK();
2808 
2809 	/*
2810 	 * If we didn't allocate enough space (uncommon), try again.  If
2811 	 * we have already allocated as much space as we are allowed,
2812 	 * return what we've got.
2813 	 */
2814 	if (valid_len != max_len && !full) {
2815 		sbuf_delete(sb);
2816 		goto again;
2817 	}
2818 
2819 	ifc->ifc_len = valid_len;
2820 	sbuf_finish(sb);
2821 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2822 	sbuf_delete(sb);
2823 	return (error);
2824 }
2825 
2826 /*
2827  * Just like ifpromisc(), but for all-multicast-reception mode.
2828  */
2829 int
2830 if_allmulti(struct ifnet *ifp, int onswitch)
2831 {
2832 
2833 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2834 }
2835 
2836 struct ifmultiaddr *
2837 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2838 {
2839 	struct ifmultiaddr *ifma;
2840 
2841 	IF_ADDR_LOCK_ASSERT(ifp);
2842 
2843 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2844 		if (sa->sa_family == AF_LINK) {
2845 			if (sa_dl_equal(ifma->ifma_addr, sa))
2846 				break;
2847 		} else {
2848 			if (sa_equal(ifma->ifma_addr, sa))
2849 				break;
2850 		}
2851 	}
2852 
2853 	return ifma;
2854 }
2855 
2856 /*
2857  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2858  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2859  * the ifnet multicast address list here, so the caller must do that and
2860  * other setup work (such as notifying the device driver).  The reference
2861  * count is initialized to 1.
2862  */
2863 static struct ifmultiaddr *
2864 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2865     int mflags)
2866 {
2867 	struct ifmultiaddr *ifma;
2868 	struct sockaddr *dupsa;
2869 
2870 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2871 	    M_ZERO);
2872 	if (ifma == NULL)
2873 		return (NULL);
2874 
2875 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2876 	if (dupsa == NULL) {
2877 		free(ifma, M_IFMADDR);
2878 		return (NULL);
2879 	}
2880 	bcopy(sa, dupsa, sa->sa_len);
2881 	ifma->ifma_addr = dupsa;
2882 
2883 	ifma->ifma_ifp = ifp;
2884 	ifma->ifma_refcount = 1;
2885 	ifma->ifma_protospec = NULL;
2886 
2887 	if (llsa == NULL) {
2888 		ifma->ifma_lladdr = NULL;
2889 		return (ifma);
2890 	}
2891 
2892 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2893 	if (dupsa == NULL) {
2894 		free(ifma->ifma_addr, M_IFMADDR);
2895 		free(ifma, M_IFMADDR);
2896 		return (NULL);
2897 	}
2898 	bcopy(llsa, dupsa, llsa->sa_len);
2899 	ifma->ifma_lladdr = dupsa;
2900 
2901 	return (ifma);
2902 }
2903 
2904 /*
2905  * if_freemulti: free ifmultiaddr structure and possibly attached related
2906  * addresses.  The caller is responsible for implementing reference
2907  * counting, notifying the driver, handling routing messages, and releasing
2908  * any dependent link layer state.
2909  */
2910 static void
2911 if_freemulti(struct ifmultiaddr *ifma)
2912 {
2913 
2914 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2915 	    ifma->ifma_refcount));
2916 	KASSERT(ifma->ifma_protospec == NULL,
2917 	    ("if_freemulti: protospec not NULL"));
2918 
2919 	if (ifma->ifma_lladdr != NULL)
2920 		free(ifma->ifma_lladdr, M_IFMADDR);
2921 	free(ifma->ifma_addr, M_IFMADDR);
2922 	free(ifma, M_IFMADDR);
2923 }
2924 
2925 /*
2926  * Register an additional multicast address with a network interface.
2927  *
2928  * - If the address is already present, bump the reference count on the
2929  *   address and return.
2930  * - If the address is not link-layer, look up a link layer address.
2931  * - Allocate address structures for one or both addresses, and attach to the
2932  *   multicast address list on the interface.  If automatically adding a link
2933  *   layer address, the protocol address will own a reference to the link
2934  *   layer address, to be freed when it is freed.
2935  * - Notify the network device driver of an addition to the multicast address
2936  *   list.
2937  *
2938  * 'sa' points to caller-owned memory with the desired multicast address.
2939  *
2940  * 'retifma' will be used to return a pointer to the resulting multicast
2941  * address reference, if desired.
2942  */
2943 int
2944 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2945     struct ifmultiaddr **retifma)
2946 {
2947 	struct ifmultiaddr *ifma, *ll_ifma;
2948 	struct sockaddr *llsa;
2949 	int error;
2950 
2951 	/*
2952 	 * If the address is already present, return a new reference to it;
2953 	 * otherwise, allocate storage and set up a new address.
2954 	 */
2955 	IF_ADDR_WLOCK(ifp);
2956 	ifma = if_findmulti(ifp, sa);
2957 	if (ifma != NULL) {
2958 		ifma->ifma_refcount++;
2959 		if (retifma != NULL)
2960 			*retifma = ifma;
2961 		IF_ADDR_WUNLOCK(ifp);
2962 		return (0);
2963 	}
2964 
2965 	/*
2966 	 * The address isn't already present; resolve the protocol address
2967 	 * into a link layer address, and then look that up, bump its
2968 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2969 	 * returned, we will need to free it later.
2970 	 */
2971 	llsa = NULL;
2972 	ll_ifma = NULL;
2973 	if (ifp->if_resolvemulti != NULL) {
2974 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2975 		if (error)
2976 			goto unlock_out;
2977 	}
2978 
2979 	/*
2980 	 * Allocate the new address.  Don't hook it up yet, as we may also
2981 	 * need to allocate a link layer multicast address.
2982 	 */
2983 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2984 	if (ifma == NULL) {
2985 		error = ENOMEM;
2986 		goto free_llsa_out;
2987 	}
2988 
2989 	/*
2990 	 * If a link layer address is found, we'll need to see if it's
2991 	 * already present in the address list, or allocate is as well.
2992 	 * When this block finishes, the link layer address will be on the
2993 	 * list.
2994 	 */
2995 	if (llsa != NULL) {
2996 		ll_ifma = if_findmulti(ifp, llsa);
2997 		if (ll_ifma == NULL) {
2998 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2999 			if (ll_ifma == NULL) {
3000 				--ifma->ifma_refcount;
3001 				if_freemulti(ifma);
3002 				error = ENOMEM;
3003 				goto free_llsa_out;
3004 			}
3005 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3006 			    ifma_link);
3007 		} else
3008 			ll_ifma->ifma_refcount++;
3009 		ifma->ifma_llifma = ll_ifma;
3010 	}
3011 
3012 	/*
3013 	 * We now have a new multicast address, ifma, and possibly a new or
3014 	 * referenced link layer address.  Add the primary address to the
3015 	 * ifnet address list.
3016 	 */
3017 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3018 
3019 	if (retifma != NULL)
3020 		*retifma = ifma;
3021 
3022 	/*
3023 	 * Must generate the message while holding the lock so that 'ifma'
3024 	 * pointer is still valid.
3025 	 */
3026 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3027 	IF_ADDR_WUNLOCK(ifp);
3028 
3029 	/*
3030 	 * We are certain we have added something, so call down to the
3031 	 * interface to let them know about it.
3032 	 */
3033 	if (ifp->if_ioctl != NULL) {
3034 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3035 	}
3036 
3037 	if (llsa != NULL)
3038 		free(llsa, M_IFMADDR);
3039 
3040 	return (0);
3041 
3042 free_llsa_out:
3043 	if (llsa != NULL)
3044 		free(llsa, M_IFMADDR);
3045 
3046 unlock_out:
3047 	IF_ADDR_WUNLOCK(ifp);
3048 	return (error);
3049 }
3050 
3051 /*
3052  * Delete a multicast group membership by network-layer group address.
3053  *
3054  * Returns ENOENT if the entry could not be found. If ifp no longer
3055  * exists, results are undefined. This entry point should only be used
3056  * from subsystems which do appropriate locking to hold ifp for the
3057  * duration of the call.
3058  * Network-layer protocol domains must use if_delmulti_ifma().
3059  */
3060 int
3061 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3062 {
3063 	struct ifmultiaddr *ifma;
3064 	int lastref;
3065 #ifdef INVARIANTS
3066 	struct ifnet *oifp;
3067 
3068 	IFNET_RLOCK_NOSLEEP();
3069 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3070 		if (ifp == oifp)
3071 			break;
3072 	if (ifp != oifp)
3073 		ifp = NULL;
3074 	IFNET_RUNLOCK_NOSLEEP();
3075 
3076 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3077 #endif
3078 	if (ifp == NULL)
3079 		return (ENOENT);
3080 
3081 	IF_ADDR_WLOCK(ifp);
3082 	lastref = 0;
3083 	ifma = if_findmulti(ifp, sa);
3084 	if (ifma != NULL)
3085 		lastref = if_delmulti_locked(ifp, ifma, 0);
3086 	IF_ADDR_WUNLOCK(ifp);
3087 
3088 	if (ifma == NULL)
3089 		return (ENOENT);
3090 
3091 	if (lastref && ifp->if_ioctl != NULL) {
3092 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3093 	}
3094 
3095 	return (0);
3096 }
3097 
3098 /*
3099  * Delete all multicast group membership for an interface.
3100  * Should be used to quickly flush all multicast filters.
3101  */
3102 void
3103 if_delallmulti(struct ifnet *ifp)
3104 {
3105 	struct ifmultiaddr *ifma;
3106 	struct ifmultiaddr *next;
3107 
3108 	IF_ADDR_WLOCK(ifp);
3109 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3110 		if_delmulti_locked(ifp, ifma, 0);
3111 	IF_ADDR_WUNLOCK(ifp);
3112 }
3113 
3114 /*
3115  * Delete a multicast group membership by group membership pointer.
3116  * Network-layer protocol domains must use this routine.
3117  *
3118  * It is safe to call this routine if the ifp disappeared.
3119  */
3120 void
3121 if_delmulti_ifma(struct ifmultiaddr *ifma)
3122 {
3123 	struct ifnet *ifp;
3124 	int lastref;
3125 
3126 	ifp = ifma->ifma_ifp;
3127 #ifdef DIAGNOSTIC
3128 	if (ifp == NULL) {
3129 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3130 	} else {
3131 		struct ifnet *oifp;
3132 
3133 		IFNET_RLOCK_NOSLEEP();
3134 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3135 			if (ifp == oifp)
3136 				break;
3137 		if (ifp != oifp) {
3138 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3139 			ifp = NULL;
3140 		}
3141 		IFNET_RUNLOCK_NOSLEEP();
3142 	}
3143 #endif
3144 	/*
3145 	 * If and only if the ifnet instance exists: Acquire the address lock.
3146 	 */
3147 	if (ifp != NULL)
3148 		IF_ADDR_WLOCK(ifp);
3149 
3150 	lastref = if_delmulti_locked(ifp, ifma, 0);
3151 
3152 	if (ifp != NULL) {
3153 		/*
3154 		 * If and only if the ifnet instance exists:
3155 		 *  Release the address lock.
3156 		 *  If the group was left: update the hardware hash filter.
3157 		 */
3158 		IF_ADDR_WUNLOCK(ifp);
3159 		if (lastref && ifp->if_ioctl != NULL) {
3160 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3161 		}
3162 	}
3163 }
3164 
3165 /*
3166  * Perform deletion of network-layer and/or link-layer multicast address.
3167  *
3168  * Return 0 if the reference count was decremented.
3169  * Return 1 if the final reference was released, indicating that the
3170  * hardware hash filter should be reprogrammed.
3171  */
3172 static int
3173 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3174 {
3175 	struct ifmultiaddr *ll_ifma;
3176 
3177 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3178 		KASSERT(ifma->ifma_ifp == ifp,
3179 		    ("%s: inconsistent ifp %p", __func__, ifp));
3180 		IF_ADDR_WLOCK_ASSERT(ifp);
3181 	}
3182 
3183 	ifp = ifma->ifma_ifp;
3184 
3185 	/*
3186 	 * If the ifnet is detaching, null out references to ifnet,
3187 	 * so that upper protocol layers will notice, and not attempt
3188 	 * to obtain locks for an ifnet which no longer exists. The
3189 	 * routing socket announcement must happen before the ifnet
3190 	 * instance is detached from the system.
3191 	 */
3192 	if (detaching) {
3193 #ifdef DIAGNOSTIC
3194 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3195 #endif
3196 		/*
3197 		 * ifp may already be nulled out if we are being reentered
3198 		 * to delete the ll_ifma.
3199 		 */
3200 		if (ifp != NULL) {
3201 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3202 			ifma->ifma_ifp = NULL;
3203 		}
3204 	}
3205 
3206 	if (--ifma->ifma_refcount > 0)
3207 		return 0;
3208 
3209 	/*
3210 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3211 	 * have been associated with it. Release it first if so.
3212 	 */
3213 	ll_ifma = ifma->ifma_llifma;
3214 	if (ll_ifma != NULL) {
3215 		KASSERT(ifma->ifma_lladdr != NULL,
3216 		    ("%s: llifma w/o lladdr", __func__));
3217 		if (detaching)
3218 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3219 		if (--ll_ifma->ifma_refcount == 0) {
3220 			if (ifp != NULL) {
3221 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3222 				    ifma_link);
3223 			}
3224 			if_freemulti(ll_ifma);
3225 		}
3226 	}
3227 
3228 	if (ifp != NULL)
3229 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3230 
3231 	if_freemulti(ifma);
3232 
3233 	/*
3234 	 * The last reference to this instance of struct ifmultiaddr
3235 	 * was released; the hardware should be notified of this change.
3236 	 */
3237 	return 1;
3238 }
3239 
3240 /*
3241  * Set the link layer address on an interface.
3242  *
3243  * At this time we only support certain types of interfaces,
3244  * and we don't allow the length of the address to change.
3245  */
3246 int
3247 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3248 {
3249 	struct sockaddr_dl *sdl;
3250 	struct ifaddr *ifa;
3251 	struct ifreq ifr;
3252 
3253 	IF_ADDR_RLOCK(ifp);
3254 	ifa = ifp->if_addr;
3255 	if (ifa == NULL) {
3256 		IF_ADDR_RUNLOCK(ifp);
3257 		return (EINVAL);
3258 	}
3259 	ifa_ref(ifa);
3260 	IF_ADDR_RUNLOCK(ifp);
3261 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3262 	if (sdl == NULL) {
3263 		ifa_free(ifa);
3264 		return (EINVAL);
3265 	}
3266 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3267 		ifa_free(ifa);
3268 		return (EINVAL);
3269 	}
3270 	switch (ifp->if_type) {
3271 	case IFT_ETHER:
3272 	case IFT_FDDI:
3273 	case IFT_XETHER:
3274 	case IFT_ISO88025:
3275 	case IFT_L2VLAN:
3276 	case IFT_BRIDGE:
3277 	case IFT_ARCNET:
3278 	case IFT_IEEE8023ADLAG:
3279 	case IFT_IEEE80211:
3280 		bcopy(lladdr, LLADDR(sdl), len);
3281 		ifa_free(ifa);
3282 		break;
3283 	default:
3284 		ifa_free(ifa);
3285 		return (ENODEV);
3286 	}
3287 
3288 	/*
3289 	 * If the interface is already up, we need
3290 	 * to re-init it in order to reprogram its
3291 	 * address filter.
3292 	 */
3293 	if ((ifp->if_flags & IFF_UP) != 0) {
3294 		if (ifp->if_ioctl) {
3295 			ifp->if_flags &= ~IFF_UP;
3296 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3297 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3298 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3299 			ifp->if_flags |= IFF_UP;
3300 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3301 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3302 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3303 		}
3304 #ifdef INET
3305 		/*
3306 		 * Also send gratuitous ARPs to notify other nodes about
3307 		 * the address change.
3308 		 */
3309 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3310 			if (ifa->ifa_addr->sa_family == AF_INET)
3311 				arp_ifinit(ifp, ifa);
3312 		}
3313 #endif
3314 	}
3315 	return (0);
3316 }
3317 
3318 /*
3319  * The name argument must be a pointer to storage which will last as
3320  * long as the interface does.  For physical devices, the result of
3321  * device_get_name(dev) is a good choice and for pseudo-devices a
3322  * static string works well.
3323  */
3324 void
3325 if_initname(struct ifnet *ifp, const char *name, int unit)
3326 {
3327 	ifp->if_dname = name;
3328 	ifp->if_dunit = unit;
3329 	if (unit != IF_DUNIT_NONE)
3330 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3331 	else
3332 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3333 }
3334 
3335 int
3336 if_printf(struct ifnet *ifp, const char * fmt, ...)
3337 {
3338 	va_list ap;
3339 	int retval;
3340 
3341 	retval = printf("%s: ", ifp->if_xname);
3342 	va_start(ap, fmt);
3343 	retval += vprintf(fmt, ap);
3344 	va_end(ap);
3345 	return (retval);
3346 }
3347 
3348 void
3349 if_start(struct ifnet *ifp)
3350 {
3351 
3352 	(*(ifp)->if_start)(ifp);
3353 }
3354 
3355 /*
3356  * Backwards compatibility interface for drivers
3357  * that have not implemented it
3358  */
3359 static int
3360 if_transmit(struct ifnet *ifp, struct mbuf *m)
3361 {
3362 	int error;
3363 
3364 	IFQ_HANDOFF(ifp, m, error);
3365 	return (error);
3366 }
3367 
3368 int
3369 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3370 {
3371 	int active = 0;
3372 
3373 	IF_LOCK(ifq);
3374 	if (_IF_QFULL(ifq)) {
3375 		_IF_DROP(ifq);
3376 		IF_UNLOCK(ifq);
3377 		m_freem(m);
3378 		return (0);
3379 	}
3380 	if (ifp != NULL) {
3381 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3382 		if (m->m_flags & (M_BCAST|M_MCAST))
3383 			ifp->if_omcasts++;
3384 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3385 	}
3386 	_IF_ENQUEUE(ifq, m);
3387 	IF_UNLOCK(ifq);
3388 	if (ifp != NULL && !active)
3389 		(*(ifp)->if_start)(ifp);
3390 	return (1);
3391 }
3392 
3393 void
3394 if_register_com_alloc(u_char type,
3395     if_com_alloc_t *a, if_com_free_t *f)
3396 {
3397 
3398 	KASSERT(if_com_alloc[type] == NULL,
3399 	    ("if_register_com_alloc: %d already registered", type));
3400 	KASSERT(if_com_free[type] == NULL,
3401 	    ("if_register_com_alloc: %d free already registered", type));
3402 
3403 	if_com_alloc[type] = a;
3404 	if_com_free[type] = f;
3405 }
3406 
3407 void
3408 if_deregister_com_alloc(u_char type)
3409 {
3410 
3411 	KASSERT(if_com_alloc[type] != NULL,
3412 	    ("if_deregister_com_alloc: %d not registered", type));
3413 	KASSERT(if_com_free[type] != NULL,
3414 	    ("if_deregister_com_alloc: %d free not registered", type));
3415 	if_com_alloc[type] = NULL;
3416 	if_com_free[type] = NULL;
3417 }
3418