xref: /freebsd/sys/net/if.c (revision 36daf0495aa68d669ac6abf004940ec1b1e83e42)
1 /*-
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.5 (Berkeley) 1/9/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/malloc.h>
41 #include <sys/sbuf.h>
42 #include <sys/bus.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/refcount.h>
53 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
60 #include <sys/jail.h>
61 #include <sys/priv.h>
62 
63 #include <machine/stdarg.h>
64 #include <vm/uma.h>
65 
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_clone.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/vnet.h>
75 
76 #if defined(INET) || defined(INET6)
77 /*XXX*/
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip_carp.h>
81 #ifdef INET6
82 #include <netinet6/in6_var.h>
83 #include <netinet6/in6_ifattach.h>
84 #endif
85 #endif
86 #ifdef INET
87 #include <netinet/if_ether.h>
88 #endif
89 
90 #include <security/mac/mac_framework.h>
91 
92 #ifdef COMPAT_FREEBSD32
93 #include <sys/mount.h>
94 #include <compat/freebsd32/freebsd32.h>
95 #endif
96 
97 struct ifindex_entry {
98 	struct  ifnet *ife_ifnet;
99 };
100 
101 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
102 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
103 
104 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
105 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
106     &ifqmaxlen, 0, "max send queue size");
107 
108 /* Log link state change events */
109 static int log_link_state_change = 1;
110 
111 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
112 	&log_link_state_change, 0,
113 	"log interface link state change events");
114 
115 /* Interface description */
116 static unsigned int ifdescr_maxlen = 1024;
117 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
118 	&ifdescr_maxlen, 0,
119 	"administrative maximum length for interface description");
120 
121 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	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
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 #if defined(INET) || defined(INET6)
133 struct ifnet *(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
134 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
135     struct sockaddr *sa, struct rtentry *rt);
136 #endif
137 #ifdef INET
138 int (*carp_iamatch_p)(struct ifnet *, struct in_ifaddr *, struct in_addr *,
139     u_int8_t **);
140 #endif
141 #ifdef INET6
142 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
143 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
144     const struct in6_addr *taddr);
145 #endif
146 
147 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
148 
149 /*
150  * XXX: Style; these should be sorted alphabetically, and unprototyped
151  * static functions should be prototyped. Currently they are sorted by
152  * declaration order.
153  */
154 static void	if_attachdomain(void *);
155 static void	if_attachdomain1(struct ifnet *);
156 static int	ifconf(u_long, caddr_t);
157 static void	if_freemulti(struct ifmultiaddr *);
158 static void	if_init(void *);
159 static void	if_grow(void);
160 static void	if_route(struct ifnet *, int flag, int fam);
161 static int	if_setflag(struct ifnet *, int, int, int *, int);
162 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
163 static void	if_unroute(struct ifnet *, int flag, int fam);
164 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
165 static int	if_rtdel(struct radix_node *, void *);
166 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
167 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
168 static void	do_link_state_change(void *, int);
169 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
170 static int	if_getgroupmembers(struct ifgroupreq *);
171 static void	if_delgroups(struct ifnet *);
172 static void	if_attach_internal(struct ifnet *, int);
173 static void	if_detach_internal(struct ifnet *, int);
174 
175 #ifdef INET6
176 /*
177  * XXX: declare here to avoid to include many inet6 related files..
178  * should be more generalized?
179  */
180 extern void	nd6_setmtu(struct ifnet *);
181 #endif
182 
183 VNET_DEFINE(int, if_index);
184 int	ifqmaxlen = IFQ_MAXLEN;
185 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
186 VNET_DEFINE(struct ifgrouphead, ifg_head);
187 
188 static VNET_DEFINE(int, if_indexlim) = 8;
189 
190 /* Table of ifnet by index. */
191 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
192 
193 #define	V_if_indexlim		VNET(if_indexlim)
194 #define	V_ifindex_table		VNET(ifindex_table)
195 
196 /*
197  * The global network interface list (V_ifnet) and related state (such as
198  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
199  * an rwlock.  Either may be acquired shared to stablize the list, but both
200  * must be acquired writable to modify the list.  This model allows us to
201  * both stablize the interface list during interrupt thread processing, but
202  * also to stablize it over long-running ioctls, without introducing priority
203  * inversions and deadlocks.
204  */
205 struct rwlock ifnet_rwlock;
206 struct sx ifnet_sxlock;
207 
208 /*
209  * The allocation of network interfaces is a rather non-atomic affair; we
210  * need to select an index before we are ready to expose the interface for
211  * use, so will use this pointer value to indicate reservation.
212  */
213 #define	IFNET_HOLD	(void *)(uintptr_t)(-1)
214 
215 static	if_com_alloc_t *if_com_alloc[256];
216 static	if_com_free_t *if_com_free[256];
217 
218 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
219 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
220 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
221 
222 struct ifnet *
223 ifnet_byindex_locked(u_short idx)
224 {
225 
226 	if (idx > V_if_index)
227 		return (NULL);
228 	if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
229 		return (NULL);
230 	return (V_ifindex_table[idx].ife_ifnet);
231 }
232 
233 struct ifnet *
234 ifnet_byindex(u_short idx)
235 {
236 	struct ifnet *ifp;
237 
238 	IFNET_RLOCK_NOSLEEP();
239 	ifp = ifnet_byindex_locked(idx);
240 	IFNET_RUNLOCK_NOSLEEP();
241 	return (ifp);
242 }
243 
244 struct ifnet *
245 ifnet_byindex_ref(u_short idx)
246 {
247 	struct ifnet *ifp;
248 
249 	IFNET_RLOCK_NOSLEEP();
250 	ifp = ifnet_byindex_locked(idx);
251 	if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
252 		IFNET_RUNLOCK_NOSLEEP();
253 		return (NULL);
254 	}
255 	if_ref(ifp);
256 	IFNET_RUNLOCK_NOSLEEP();
257 	return (ifp);
258 }
259 
260 /*
261  * Allocate an ifindex array entry; return 0 on success or an error on
262  * failure.
263  */
264 static int
265 ifindex_alloc_locked(u_short *idxp)
266 {
267 	u_short idx;
268 
269 	IFNET_WLOCK_ASSERT();
270 
271 retry:
272 	/*
273 	 * Try to find an empty slot below V_if_index.  If we fail, take the
274 	 * next slot.
275 	 */
276 	for (idx = 1; idx <= V_if_index; idx++) {
277 		if (V_ifindex_table[idx].ife_ifnet == NULL)
278 			break;
279 	}
280 
281 	/* Catch if_index overflow. */
282 	if (idx < 1)
283 		return (ENOSPC);
284 	if (idx >= V_if_indexlim) {
285 		if_grow();
286 		goto retry;
287 	}
288 	if (idx > V_if_index)
289 		V_if_index = idx;
290 	*idxp = idx;
291 	return (0);
292 }
293 
294 static void
295 ifindex_free_locked(u_short idx)
296 {
297 
298 	IFNET_WLOCK_ASSERT();
299 
300 	V_ifindex_table[idx].ife_ifnet = NULL;
301 	while (V_if_index > 0 &&
302 	    V_ifindex_table[V_if_index].ife_ifnet == NULL)
303 		V_if_index--;
304 }
305 
306 static void
307 ifindex_free(u_short idx)
308 {
309 
310 	IFNET_WLOCK();
311 	ifindex_free_locked(idx);
312 	IFNET_WUNLOCK();
313 }
314 
315 static void
316 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
317 {
318 
319 	IFNET_WLOCK_ASSERT();
320 
321 	V_ifindex_table[idx].ife_ifnet = ifp;
322 }
323 
324 static void
325 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
326 {
327 
328 	IFNET_WLOCK();
329 	ifnet_setbyindex_locked(idx, ifp);
330 	IFNET_WUNLOCK();
331 }
332 
333 struct ifaddr *
334 ifaddr_byindex(u_short idx)
335 {
336 	struct ifaddr *ifa;
337 
338 	IFNET_RLOCK_NOSLEEP();
339 	ifa = ifnet_byindex_locked(idx)->if_addr;
340 	if (ifa != NULL)
341 		ifa_ref(ifa);
342 	IFNET_RUNLOCK_NOSLEEP();
343 	return (ifa);
344 }
345 
346 /*
347  * Network interface utility routines.
348  *
349  * Routines with ifa_ifwith* names take sockaddr *'s as
350  * parameters.
351  */
352 
353 static void
354 vnet_if_init(const void *unused __unused)
355 {
356 
357 	TAILQ_INIT(&V_ifnet);
358 	TAILQ_INIT(&V_ifg_head);
359 	IFNET_WLOCK();
360 	if_grow();				/* create initial table */
361 	IFNET_WUNLOCK();
362 	vnet_if_clone_init();
363 }
364 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
365     NULL);
366 
367 /* ARGSUSED*/
368 static void
369 if_init(void *dummy __unused)
370 {
371 
372 	IFNET_LOCK_INIT();
373 	if_clone_init();
374 }
375 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
376 
377 
378 #ifdef VIMAGE
379 static void
380 vnet_if_uninit(const void *unused __unused)
381 {
382 
383 	VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
384 	    "not empty", __func__, __LINE__, &V_ifnet));
385 	VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
386 	    "not empty", __func__, __LINE__, &V_ifg_head));
387 
388 	free((caddr_t)V_ifindex_table, M_IFNET);
389 }
390 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
391     vnet_if_uninit, NULL);
392 #endif
393 
394 static void
395 if_grow(void)
396 {
397 	int oldlim;
398 	u_int n;
399 	struct ifindex_entry *e;
400 
401 	IFNET_WLOCK_ASSERT();
402 	oldlim = V_if_indexlim;
403 	IFNET_WUNLOCK();
404 	n = (oldlim << 1) * sizeof(*e);
405 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
406 	IFNET_WLOCK();
407 	if (V_if_indexlim != oldlim) {
408 		free(e, M_IFNET);
409 		return;
410 	}
411 	if (V_ifindex_table != NULL) {
412 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
413 		free((caddr_t)V_ifindex_table, M_IFNET);
414 	}
415 	V_if_indexlim <<= 1;
416 	V_ifindex_table = e;
417 }
418 
419 /*
420  * Allocate a struct ifnet and an index for an interface.  A layer 2
421  * common structure will also be allocated if an allocation routine is
422  * registered for the passed type.
423  */
424 struct ifnet *
425 if_alloc(u_char type)
426 {
427 	struct ifnet *ifp;
428 	u_short idx;
429 
430 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
431 	IFNET_WLOCK();
432 	if (ifindex_alloc_locked(&idx) != 0) {
433 		IFNET_WUNLOCK();
434 		free(ifp, M_IFNET);
435 		return (NULL);
436 	}
437 	ifnet_setbyindex_locked(idx, IFNET_HOLD);
438 	IFNET_WUNLOCK();
439 	ifp->if_index = idx;
440 	ifp->if_type = type;
441 	ifp->if_alloctype = type;
442 	if (if_com_alloc[type] != NULL) {
443 		ifp->if_l2com = if_com_alloc[type](type, ifp);
444 		if (ifp->if_l2com == NULL) {
445 			free(ifp, M_IFNET);
446 			ifindex_free(idx);
447 			return (NULL);
448 		}
449 	}
450 
451 	IF_ADDR_LOCK_INIT(ifp);
452 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
453 	ifp->if_afdata_initialized = 0;
454 	IF_AFDATA_LOCK_INIT(ifp);
455 	TAILQ_INIT(&ifp->if_addrhead);
456 	TAILQ_INIT(&ifp->if_prefixhead);
457 	TAILQ_INIT(&ifp->if_multiaddrs);
458 	TAILQ_INIT(&ifp->if_groups);
459 #ifdef MAC
460 	mac_ifnet_init(ifp);
461 #endif
462 	ifq_init(&ifp->if_snd, ifp);
463 
464 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
465 	ifnet_setbyindex(ifp->if_index, ifp);
466 	return (ifp);
467 }
468 
469 /*
470  * Do the actual work of freeing a struct ifnet, and layer 2 common
471  * structure.  This call is made when the last reference to an
472  * interface is released.
473  */
474 static void
475 if_free_internal(struct ifnet *ifp)
476 {
477 
478 	KASSERT((ifp->if_flags & IFF_DYING),
479 	    ("if_free_internal: interface not dying"));
480 
481 	if (if_com_free[ifp->if_alloctype] != NULL)
482 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
483 		    ifp->if_alloctype);
484 
485 #ifdef MAC
486 	mac_ifnet_destroy(ifp);
487 #endif /* MAC */
488 	if (ifp->if_description != NULL)
489 		free(ifp->if_description, M_IFDESCR);
490 	IF_AFDATA_DESTROY(ifp);
491 	IF_ADDR_LOCK_DESTROY(ifp);
492 	ifq_delete(&ifp->if_snd);
493 	free(ifp, M_IFNET);
494 }
495 
496 /*
497  * This version should only be called by intefaces that switch their type
498  * after calling if_alloc().  if_free_type() will go away again now that we
499  * have if_alloctype to cache the original allocation type.  For now, assert
500  * that they match, since we require that in practice.
501  */
502 void
503 if_free_type(struct ifnet *ifp, u_char type)
504 {
505 
506 	KASSERT(ifp->if_alloctype == type,
507 	    ("if_free_type: type (%d) != alloctype (%d)", type,
508 	    ifp->if_alloctype));
509 
510 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
511 
512 	IFNET_WLOCK();
513 	KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
514 	    ("%s: freeing unallocated ifnet", ifp->if_xname));
515 
516 	ifindex_free_locked(ifp->if_index);
517 	IFNET_WUNLOCK();
518 
519 	if (!refcount_release(&ifp->if_refcount))
520 		return;
521 	if_free_internal(ifp);
522 }
523 
524 /*
525  * This is the normal version of if_free(), used by device drivers to free a
526  * detached network interface.  The contents of if_free_type() will move into
527  * here when if_free_type() goes away.
528  */
529 void
530 if_free(struct ifnet *ifp)
531 {
532 
533 	if_free_type(ifp, ifp->if_alloctype);
534 }
535 
536 /*
537  * Interfaces to keep an ifnet type-stable despite the possibility of the
538  * driver calling if_free().  If there are additional references, we defer
539  * freeing the underlying data structure.
540  */
541 void
542 if_ref(struct ifnet *ifp)
543 {
544 
545 	/* We don't assert the ifnet list lock here, but arguably should. */
546 	refcount_acquire(&ifp->if_refcount);
547 }
548 
549 void
550 if_rele(struct ifnet *ifp)
551 {
552 
553 	if (!refcount_release(&ifp->if_refcount))
554 		return;
555 	if_free_internal(ifp);
556 }
557 
558 void
559 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
560 {
561 
562 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
563 
564 	if (ifq->ifq_maxlen == 0)
565 		ifq->ifq_maxlen = ifqmaxlen;
566 
567 	ifq->altq_type = 0;
568 	ifq->altq_disc = NULL;
569 	ifq->altq_flags &= ALTQF_CANTCHANGE;
570 	ifq->altq_tbr  = NULL;
571 	ifq->altq_ifp  = ifp;
572 }
573 
574 void
575 ifq_delete(struct ifaltq *ifq)
576 {
577 	mtx_destroy(&ifq->ifq_mtx);
578 }
579 
580 /*
581  * Perform generic interface initalization tasks and attach the interface
582  * to the list of "active" interfaces.  If vmove flag is set on entry
583  * to if_attach_internal(), perform only a limited subset of initialization
584  * tasks, given that we are moving from one vnet to another an ifnet which
585  * has already been fully initialized.
586  *
587  * XXX:
588  *  - The decision to return void and thus require this function to
589  *    succeed is questionable.
590  *  - We should probably do more sanity checking.  For instance we don't
591  *    do anything to insure if_xname is unique or non-empty.
592  */
593 void
594 if_attach(struct ifnet *ifp)
595 {
596 
597 	if_attach_internal(ifp, 0);
598 }
599 
600 static void
601 if_attach_internal(struct ifnet *ifp, int vmove)
602 {
603 	unsigned socksize, ifasize;
604 	int namelen, masklen;
605 	struct sockaddr_dl *sdl;
606 	struct ifaddr *ifa;
607 
608 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
609 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
610 		    ifp->if_xname);
611 
612 #ifdef VIMAGE
613 	ifp->if_vnet = curvnet;
614 	if (ifp->if_home_vnet == NULL)
615 		ifp->if_home_vnet = curvnet;
616 #endif
617 
618 	if_addgroup(ifp, IFG_ALL);
619 
620 	getmicrotime(&ifp->if_lastchange);
621 	ifp->if_data.ifi_epoch = time_uptime;
622 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
623 
624 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
625 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
626 	    ("transmit and qflush must both either be set or both be NULL"));
627 	if (ifp->if_transmit == NULL) {
628 		ifp->if_transmit = if_transmit;
629 		ifp->if_qflush = if_qflush;
630 	}
631 
632 	if (!vmove) {
633 #ifdef MAC
634 		mac_ifnet_create(ifp);
635 #endif
636 
637 		/*
638 		 * Create a Link Level name for this device.
639 		 */
640 		namelen = strlen(ifp->if_xname);
641 		/*
642 		 * Always save enough space for any possiable name so we
643 		 * can do a rename in place later.
644 		 */
645 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
646 		socksize = masklen + ifp->if_addrlen;
647 		if (socksize < sizeof(*sdl))
648 			socksize = sizeof(*sdl);
649 		socksize = roundup2(socksize, sizeof(long));
650 		ifasize = sizeof(*ifa) + 2 * socksize;
651 		ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
652 		ifa_init(ifa);
653 		sdl = (struct sockaddr_dl *)(ifa + 1);
654 		sdl->sdl_len = socksize;
655 		sdl->sdl_family = AF_LINK;
656 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
657 		sdl->sdl_nlen = namelen;
658 		sdl->sdl_index = ifp->if_index;
659 		sdl->sdl_type = ifp->if_type;
660 		ifp->if_addr = ifa;
661 		ifa->ifa_ifp = ifp;
662 		ifa->ifa_rtrequest = link_rtrequest;
663 		ifa->ifa_addr = (struct sockaddr *)sdl;
664 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
665 		ifa->ifa_netmask = (struct sockaddr *)sdl;
666 		sdl->sdl_len = masklen;
667 		while (namelen != 0)
668 			sdl->sdl_data[--namelen] = 0xff;
669 		TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
670 		/* Reliably crash if used uninitialized. */
671 		ifp->if_broadcastaddr = NULL;
672 	}
673 #ifdef VIMAGE
674 	else {
675 		/*
676 		 * Update the interface index in the link layer address
677 		 * of the interface.
678 		 */
679 		for (ifa = ifp->if_addr; ifa != NULL;
680 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
681 			if (ifa->ifa_addr->sa_family == AF_LINK) {
682 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
683 				sdl->sdl_index = ifp->if_index;
684 			}
685 		}
686 	}
687 #endif
688 
689 	IFNET_WLOCK();
690 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
691 #ifdef VIMAGE
692 	curvnet->vnet_ifcnt++;
693 #endif
694 	IFNET_WUNLOCK();
695 
696 	if (domain_init_status >= 2)
697 		if_attachdomain1(ifp);
698 
699 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
700 	if (IS_DEFAULT_VNET(curvnet))
701 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
702 
703 	/* Announce the interface. */
704 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
705 }
706 
707 static void
708 if_attachdomain(void *dummy)
709 {
710 	struct ifnet *ifp;
711 	int s;
712 
713 	s = splnet();
714 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
715 		if_attachdomain1(ifp);
716 	splx(s);
717 }
718 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
719     if_attachdomain, NULL);
720 
721 static void
722 if_attachdomain1(struct ifnet *ifp)
723 {
724 	struct domain *dp;
725 	int s;
726 
727 	s = splnet();
728 
729 	/*
730 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
731 	 * cannot lock ifp->if_afdata initialization, entirely.
732 	 */
733 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
734 		splx(s);
735 		return;
736 	}
737 	if (ifp->if_afdata_initialized >= domain_init_status) {
738 		IF_AFDATA_UNLOCK(ifp);
739 		splx(s);
740 		printf("if_attachdomain called more than once on %s\n",
741 		    ifp->if_xname);
742 		return;
743 	}
744 	ifp->if_afdata_initialized = domain_init_status;
745 	IF_AFDATA_UNLOCK(ifp);
746 
747 	/* address family dependent data region */
748 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
749 	for (dp = domains; dp; dp = dp->dom_next) {
750 		if (dp->dom_ifattach)
751 			ifp->if_afdata[dp->dom_family] =
752 			    (*dp->dom_ifattach)(ifp);
753 	}
754 
755 	splx(s);
756 }
757 
758 /*
759  * Remove any unicast or broadcast network addresses from an interface.
760  */
761 void
762 if_purgeaddrs(struct ifnet *ifp)
763 {
764 	struct ifaddr *ifa, *next;
765 
766 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
767 		if (ifa->ifa_addr->sa_family == AF_LINK)
768 			continue;
769 #ifdef INET
770 		/* XXX: Ugly!! ad hoc just for INET */
771 		if (ifa->ifa_addr->sa_family == AF_INET) {
772 			struct ifaliasreq ifr;
773 
774 			bzero(&ifr, sizeof(ifr));
775 			ifr.ifra_addr = *ifa->ifa_addr;
776 			if (ifa->ifa_dstaddr)
777 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
778 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
779 			    NULL) == 0)
780 				continue;
781 		}
782 #endif /* INET */
783 #ifdef INET6
784 		if (ifa->ifa_addr->sa_family == AF_INET6) {
785 			in6_purgeaddr(ifa);
786 			/* ifp_addrhead is already updated */
787 			continue;
788 		}
789 #endif /* INET6 */
790 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
791 		ifa_free(ifa);
792 	}
793 }
794 
795 /*
796  * Remove any multicast network addresses from an interface when an ifnet
797  * is going away.
798  */
799 static void
800 if_purgemaddrs(struct ifnet *ifp)
801 {
802 	struct ifmultiaddr *ifma;
803 	struct ifmultiaddr *next;
804 
805 	IF_ADDR_LOCK(ifp);
806 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
807 		if_delmulti_locked(ifp, ifma, 1);
808 	IF_ADDR_UNLOCK(ifp);
809 }
810 
811 /*
812  * Detach an interface, removing it from the list of "active" interfaces.
813  * If vmove flag is set on entry to if_detach_internal(), perform only a
814  * limited subset of cleanup tasks, given that we are moving an ifnet from
815  * one vnet to another, where it must be fully operational.
816  *
817  * XXXRW: There are some significant questions about event ordering, and
818  * how to prevent things from starting to use the interface during detach.
819  */
820 void
821 if_detach(struct ifnet *ifp)
822 {
823 
824 	if_detach_internal(ifp, 0);
825 }
826 
827 static void
828 if_detach_internal(struct ifnet *ifp, int vmove)
829 {
830 	struct ifaddr *ifa;
831 	struct radix_node_head	*rnh;
832 	int i, j;
833 	struct domain *dp;
834  	struct ifnet *iter;
835  	int found = 0;
836 
837 	IFNET_WLOCK();
838 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
839 		if (iter == ifp) {
840 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
841 			found = 1;
842 			break;
843 		}
844 #ifdef VIMAGE
845 	if (found)
846 		curvnet->vnet_ifcnt--;
847 #endif
848 	IFNET_WUNLOCK();
849 	if (!found) {
850 		if (vmove)
851 			panic("%s: ifp=%p not on the ifnet tailq %p",
852 			    __func__, ifp, &V_ifnet);
853 		else
854 			return; /* XXX this should panic as well? */
855 	}
856 
857 	/*
858 	 * Remove/wait for pending events.
859 	 */
860 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
861 
862 	/*
863 	 * Remove routes and flush queues.
864 	 */
865 	if_down(ifp);
866 #ifdef ALTQ
867 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
868 		altq_disable(&ifp->if_snd);
869 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
870 		altq_detach(&ifp->if_snd);
871 #endif
872 
873 	if_purgeaddrs(ifp);
874 
875 #ifdef INET
876 	in_ifdetach(ifp);
877 #endif
878 
879 #ifdef INET6
880 	/*
881 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
882 	 * before removing routing entries below, since IPv6 interface direct
883 	 * routes are expected to be removed by the IPv6-specific kernel API.
884 	 * Otherwise, the kernel will detect some inconsistency and bark it.
885 	 */
886 	in6_ifdetach(ifp);
887 #endif
888 	if_purgemaddrs(ifp);
889 
890 	if (!vmove) {
891 		/*
892 		 * Prevent further calls into the device driver via ifnet.
893 		 */
894 		if_dead(ifp);
895 
896 		/*
897 		 * Remove link ifaddr pointer and maybe decrement if_index.
898 		 * Clean up all addresses.
899 		 */
900 		ifp->if_addr = NULL;
901 
902 		/* We can now free link ifaddr. */
903 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
904 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
905 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
906 			ifa_free(ifa);
907 		}
908 	}
909 
910 	/*
911 	 * Delete all remaining routes using this interface
912 	 * Unfortuneatly the only way to do this is to slog through
913 	 * the entire routing table looking for routes which point
914 	 * to this interface...oh well...
915 	 */
916 	for (i = 1; i <= AF_MAX; i++) {
917 		for (j = 0; j < rt_numfibs; j++) {
918 			rnh = rt_tables_get_rnh(j, i);
919 			if (rnh == NULL)
920 				continue;
921 			RADIX_NODE_HEAD_LOCK(rnh);
922 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
923 			RADIX_NODE_HEAD_UNLOCK(rnh);
924 		}
925 	}
926 
927 	/* Announce that the interface is gone. */
928 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
929 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
930 	if (IS_DEFAULT_VNET(curvnet))
931 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
932 	if_delgroups(ifp);
933 
934 	/*
935 	 * We cannot hold the lock over dom_ifdetach calls as they might
936 	 * sleep, for example trying to drain a callout, thus open up the
937 	 * theoretical race with re-attaching.
938 	 */
939 	IF_AFDATA_LOCK(ifp);
940 	i = ifp->if_afdata_initialized;
941 	ifp->if_afdata_initialized = 0;
942 	IF_AFDATA_UNLOCK(ifp);
943 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
944 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
945 			(*dp->dom_ifdetach)(ifp,
946 			    ifp->if_afdata[dp->dom_family]);
947 	}
948 }
949 
950 #ifdef VIMAGE
951 /*
952  * if_vmove() performs a limited version of if_detach() in current
953  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
954  * An attempt is made to shrink if_index in current vnet, find an
955  * unused if_index in target vnet and calls if_grow() if necessary,
956  * and finally find an unused if_xname for the target vnet.
957  */
958 void
959 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
960 {
961 	u_short idx;
962 
963 	/*
964 	 * Detach from current vnet, but preserve LLADDR info, do not
965 	 * mark as dead etc. so that the ifnet can be reattached later.
966 	 */
967 	if_detach_internal(ifp, 1);
968 
969 	/*
970 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
971 	 * the if_index for that vnet if possible.
972 	 *
973 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
974 	 * or we'd lock on one vnet and unlock on another.
975 	 */
976 	IFNET_WLOCK();
977 	ifindex_free_locked(ifp->if_index);
978 	IFNET_WUNLOCK();
979 
980 	/*
981 	 * Perform interface-specific reassignment tasks, if provided by
982 	 * the driver.
983 	 */
984 	if (ifp->if_reassign != NULL)
985 		ifp->if_reassign(ifp, new_vnet, NULL);
986 
987 	/*
988 	 * Switch to the context of the target vnet.
989 	 */
990 	CURVNET_SET_QUIET(new_vnet);
991 
992 	IFNET_WLOCK();
993 	if (ifindex_alloc_locked(&idx) != 0) {
994 		IFNET_WUNLOCK();
995 		panic("if_index overflow");
996 	}
997 	ifp->if_index = idx;
998 	ifnet_setbyindex_locked(ifp->if_index, ifp);
999 	IFNET_WUNLOCK();
1000 
1001 	if_attach_internal(ifp, 1);
1002 
1003 	CURVNET_RESTORE();
1004 }
1005 
1006 /*
1007  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1008  */
1009 static int
1010 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1011 {
1012 	struct prison *pr;
1013 	struct ifnet *difp;
1014 
1015 	/* Try to find the prison within our visibility. */
1016 	sx_slock(&allprison_lock);
1017 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1018 	sx_sunlock(&allprison_lock);
1019 	if (pr == NULL)
1020 		return (ENXIO);
1021 	prison_hold_locked(pr);
1022 	mtx_unlock(&pr->pr_mtx);
1023 
1024 	/* Do not try to move the iface from and to the same prison. */
1025 	if (pr->pr_vnet == ifp->if_vnet) {
1026 		prison_free(pr);
1027 		return (EEXIST);
1028 	}
1029 
1030 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1031 	/* XXX Lock interfaces to avoid races. */
1032 	CURVNET_SET_QUIET(pr->pr_vnet);
1033 	difp = ifunit(ifname);
1034 	CURVNET_RESTORE();
1035 	if (difp != NULL) {
1036 		prison_free(pr);
1037 		return (EEXIST);
1038 	}
1039 
1040 	/* Move the interface into the child jail/vnet. */
1041 	if_vmove(ifp, pr->pr_vnet);
1042 
1043 	/* Report the new if_xname back to the userland. */
1044 	sprintf(ifname, "%s", ifp->if_xname);
1045 
1046 	prison_free(pr);
1047 	return (0);
1048 }
1049 
1050 static int
1051 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1052 {
1053 	struct prison *pr;
1054 	struct vnet *vnet_dst;
1055 	struct ifnet *ifp;
1056 
1057 	/* Try to find the prison within our visibility. */
1058 	sx_slock(&allprison_lock);
1059 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1060 	sx_sunlock(&allprison_lock);
1061 	if (pr == NULL)
1062 		return (ENXIO);
1063 	prison_hold_locked(pr);
1064 	mtx_unlock(&pr->pr_mtx);
1065 
1066 	/* Make sure the named iface exists in the source prison/vnet. */
1067 	CURVNET_SET(pr->pr_vnet);
1068 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1069 	if (ifp == NULL) {
1070 		CURVNET_RESTORE();
1071 		prison_free(pr);
1072 		return (ENXIO);
1073 	}
1074 
1075 	/* Do not try to move the iface from and to the same prison. */
1076 	vnet_dst = TD_TO_VNET(td);
1077 	if (vnet_dst == ifp->if_vnet) {
1078 		CURVNET_RESTORE();
1079 		prison_free(pr);
1080 		return (EEXIST);
1081 	}
1082 
1083 	/* Get interface back from child jail/vnet. */
1084 	if_vmove(ifp, vnet_dst);
1085 	CURVNET_RESTORE();
1086 
1087 	/* Report the new if_xname back to the userland. */
1088 	sprintf(ifname, "%s", ifp->if_xname);
1089 
1090 	prison_free(pr);
1091 	return (0);
1092 }
1093 #endif /* VIMAGE */
1094 
1095 /*
1096  * Add a group to an interface
1097  */
1098 int
1099 if_addgroup(struct ifnet *ifp, const char *groupname)
1100 {
1101 	struct ifg_list		*ifgl;
1102 	struct ifg_group	*ifg = NULL;
1103 	struct ifg_member	*ifgm;
1104 
1105 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1106 	    groupname[strlen(groupname) - 1] <= '9')
1107 		return (EINVAL);
1108 
1109 	IFNET_WLOCK();
1110 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1111 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1112 			IFNET_WUNLOCK();
1113 			return (EEXIST);
1114 		}
1115 
1116 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1117 	    M_NOWAIT)) == NULL) {
1118 	    	IFNET_WUNLOCK();
1119 		return (ENOMEM);
1120 	}
1121 
1122 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1123 	    M_TEMP, M_NOWAIT)) == NULL) {
1124 		free(ifgl, M_TEMP);
1125 		IFNET_WUNLOCK();
1126 		return (ENOMEM);
1127 	}
1128 
1129 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1130 		if (!strcmp(ifg->ifg_group, groupname))
1131 			break;
1132 
1133 	if (ifg == NULL) {
1134 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1135 		    M_TEMP, M_NOWAIT)) == NULL) {
1136 			free(ifgl, M_TEMP);
1137 			free(ifgm, M_TEMP);
1138 			IFNET_WUNLOCK();
1139 			return (ENOMEM);
1140 		}
1141 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1142 		ifg->ifg_refcnt = 0;
1143 		TAILQ_INIT(&ifg->ifg_members);
1144 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1145 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1146 	}
1147 
1148 	ifg->ifg_refcnt++;
1149 	ifgl->ifgl_group = ifg;
1150 	ifgm->ifgm_ifp = ifp;
1151 
1152 	IF_ADDR_LOCK(ifp);
1153 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1154 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1155 	IF_ADDR_UNLOCK(ifp);
1156 
1157 	IFNET_WUNLOCK();
1158 
1159 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1160 
1161 	return (0);
1162 }
1163 
1164 /*
1165  * Remove a group from an interface
1166  */
1167 int
1168 if_delgroup(struct ifnet *ifp, const char *groupname)
1169 {
1170 	struct ifg_list		*ifgl;
1171 	struct ifg_member	*ifgm;
1172 
1173 	IFNET_WLOCK();
1174 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1175 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1176 			break;
1177 	if (ifgl == NULL) {
1178 		IFNET_WUNLOCK();
1179 		return (ENOENT);
1180 	}
1181 
1182 	IF_ADDR_LOCK(ifp);
1183 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1184 	IF_ADDR_UNLOCK(ifp);
1185 
1186 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1187 		if (ifgm->ifgm_ifp == ifp)
1188 			break;
1189 
1190 	if (ifgm != NULL) {
1191 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1192 		free(ifgm, M_TEMP);
1193 	}
1194 
1195 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1196 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1197 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1198 		free(ifgl->ifgl_group, M_TEMP);
1199 	}
1200 	IFNET_WUNLOCK();
1201 
1202 	free(ifgl, M_TEMP);
1203 
1204 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1205 
1206 	return (0);
1207 }
1208 
1209 /*
1210  * Remove an interface from all groups
1211  */
1212 static void
1213 if_delgroups(struct ifnet *ifp)
1214 {
1215 	struct ifg_list		*ifgl;
1216 	struct ifg_member	*ifgm;
1217 	char groupname[IFNAMSIZ];
1218 
1219 	IFNET_WLOCK();
1220 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1221 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1222 
1223 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1224 
1225 		IF_ADDR_LOCK(ifp);
1226 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1227 		IF_ADDR_UNLOCK(ifp);
1228 
1229 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1230 			if (ifgm->ifgm_ifp == ifp)
1231 				break;
1232 
1233 		if (ifgm != NULL) {
1234 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1235 			    ifgm_next);
1236 			free(ifgm, M_TEMP);
1237 		}
1238 
1239 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1240 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1241 			EVENTHANDLER_INVOKE(group_detach_event,
1242 			    ifgl->ifgl_group);
1243 			free(ifgl->ifgl_group, M_TEMP);
1244 		}
1245 		IFNET_WUNLOCK();
1246 
1247 		free(ifgl, M_TEMP);
1248 
1249 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1250 
1251 		IFNET_WLOCK();
1252 	}
1253 	IFNET_WUNLOCK();
1254 }
1255 
1256 /*
1257  * Stores all groups from an interface in memory pointed
1258  * to by data
1259  */
1260 static int
1261 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1262 {
1263 	int			 len, error;
1264 	struct ifg_list		*ifgl;
1265 	struct ifg_req		 ifgrq, *ifgp;
1266 	struct ifgroupreq	*ifgr = data;
1267 
1268 	if (ifgr->ifgr_len == 0) {
1269 		IF_ADDR_LOCK(ifp);
1270 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1271 			ifgr->ifgr_len += sizeof(struct ifg_req);
1272 		IF_ADDR_UNLOCK(ifp);
1273 		return (0);
1274 	}
1275 
1276 	len = ifgr->ifgr_len;
1277 	ifgp = ifgr->ifgr_groups;
1278 	/* XXX: wire */
1279 	IF_ADDR_LOCK(ifp);
1280 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1281 		if (len < sizeof(ifgrq)) {
1282 			IF_ADDR_UNLOCK(ifp);
1283 			return (EINVAL);
1284 		}
1285 		bzero(&ifgrq, sizeof ifgrq);
1286 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1287 		    sizeof(ifgrq.ifgrq_group));
1288 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1289 		    	IF_ADDR_UNLOCK(ifp);
1290 			return (error);
1291 		}
1292 		len -= sizeof(ifgrq);
1293 		ifgp++;
1294 	}
1295 	IF_ADDR_UNLOCK(ifp);
1296 
1297 	return (0);
1298 }
1299 
1300 /*
1301  * Stores all members of a group in memory pointed to by data
1302  */
1303 static int
1304 if_getgroupmembers(struct ifgroupreq *data)
1305 {
1306 	struct ifgroupreq	*ifgr = data;
1307 	struct ifg_group	*ifg;
1308 	struct ifg_member	*ifgm;
1309 	struct ifg_req		 ifgrq, *ifgp;
1310 	int			 len, error;
1311 
1312 	IFNET_RLOCK();
1313 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1314 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1315 			break;
1316 	if (ifg == NULL) {
1317 		IFNET_RUNLOCK();
1318 		return (ENOENT);
1319 	}
1320 
1321 	if (ifgr->ifgr_len == 0) {
1322 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1323 			ifgr->ifgr_len += sizeof(ifgrq);
1324 		IFNET_RUNLOCK();
1325 		return (0);
1326 	}
1327 
1328 	len = ifgr->ifgr_len;
1329 	ifgp = ifgr->ifgr_groups;
1330 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1331 		if (len < sizeof(ifgrq)) {
1332 			IFNET_RUNLOCK();
1333 			return (EINVAL);
1334 		}
1335 		bzero(&ifgrq, sizeof ifgrq);
1336 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1337 		    sizeof(ifgrq.ifgrq_member));
1338 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1339 			IFNET_RUNLOCK();
1340 			return (error);
1341 		}
1342 		len -= sizeof(ifgrq);
1343 		ifgp++;
1344 	}
1345 	IFNET_RUNLOCK();
1346 
1347 	return (0);
1348 }
1349 
1350 /*
1351  * Delete Routes for a Network Interface
1352  *
1353  * Called for each routing entry via the rnh->rnh_walktree() call above
1354  * to delete all route entries referencing a detaching network interface.
1355  *
1356  * Arguments:
1357  *	rn	pointer to node in the routing table
1358  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1359  *
1360  * Returns:
1361  *	0	successful
1362  *	errno	failed - reason indicated
1363  *
1364  */
1365 static int
1366 if_rtdel(struct radix_node *rn, void *arg)
1367 {
1368 	struct rtentry	*rt = (struct rtentry *)rn;
1369 	struct ifnet	*ifp = arg;
1370 	int		err;
1371 
1372 	if (rt->rt_ifp == ifp) {
1373 
1374 		/*
1375 		 * Protect (sorta) against walktree recursion problems
1376 		 * with cloned routes
1377 		 */
1378 		if ((rt->rt_flags & RTF_UP) == 0)
1379 			return (0);
1380 
1381 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1382 				rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1383 				(struct rtentry **) NULL, rt->rt_fibnum);
1384 		if (err) {
1385 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1386 		}
1387 	}
1388 
1389 	return (0);
1390 }
1391 
1392 /*
1393  * Wrapper functions for struct ifnet address list locking macros.  These are
1394  * used by kernel modules to avoid encoding programming interface or binary
1395  * interface assumptions that may be violated when kernel-internal locking
1396  * approaches change.
1397  */
1398 void
1399 if_addr_rlock(struct ifnet *ifp)
1400 {
1401 
1402 	IF_ADDR_LOCK(ifp);
1403 }
1404 
1405 void
1406 if_addr_runlock(struct ifnet *ifp)
1407 {
1408 
1409 	IF_ADDR_UNLOCK(ifp);
1410 }
1411 
1412 void
1413 if_maddr_rlock(struct ifnet *ifp)
1414 {
1415 
1416 	IF_ADDR_LOCK(ifp);
1417 }
1418 
1419 void
1420 if_maddr_runlock(struct ifnet *ifp)
1421 {
1422 
1423 	IF_ADDR_UNLOCK(ifp);
1424 }
1425 
1426 /*
1427  * Reference count functions for ifaddrs.
1428  */
1429 void
1430 ifa_init(struct ifaddr *ifa)
1431 {
1432 
1433 	mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1434 	refcount_init(&ifa->ifa_refcnt, 1);
1435 }
1436 
1437 void
1438 ifa_ref(struct ifaddr *ifa)
1439 {
1440 
1441 	refcount_acquire(&ifa->ifa_refcnt);
1442 }
1443 
1444 void
1445 ifa_free(struct ifaddr *ifa)
1446 {
1447 
1448 	if (refcount_release(&ifa->ifa_refcnt)) {
1449 		mtx_destroy(&ifa->ifa_mtx);
1450 		free(ifa, M_IFADDR);
1451 	}
1452 }
1453 
1454 int
1455 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1456 {
1457 	int error = 0;
1458 	struct rtentry *rt = NULL;
1459 	struct rt_addrinfo info;
1460 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1461 
1462 	bzero(&info, sizeof(info));
1463 	info.rti_ifp = V_loif;
1464 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1465 	info.rti_info[RTAX_DST] = ia;
1466 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1467 	error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1468 
1469 	if (error == 0 && rt != NULL) {
1470 		RT_LOCK(rt);
1471 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1472 			ifa->ifa_ifp->if_type;
1473 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1474 			ifa->ifa_ifp->if_index;
1475 		RT_REMREF(rt);
1476 		RT_UNLOCK(rt);
1477 	} else if (error != 0)
1478 		log(LOG_INFO, "ifa_add_loopback_route: insertion failed\n");
1479 
1480 	return (error);
1481 }
1482 
1483 int
1484 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1485 {
1486 	int error = 0;
1487 	struct rt_addrinfo info;
1488 	struct sockaddr_dl null_sdl;
1489 
1490 	bzero(&null_sdl, sizeof(null_sdl));
1491 	null_sdl.sdl_len = sizeof(null_sdl);
1492 	null_sdl.sdl_family = AF_LINK;
1493 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1494 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1495 	bzero(&info, sizeof(info));
1496 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1497 	info.rti_info[RTAX_DST] = ia;
1498 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1499 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1500 
1501 	if (error != 0)
1502 		log(LOG_INFO, "ifa_del_loopback_route: deletion failed\n");
1503 
1504 	return (error);
1505 }
1506 
1507 /*
1508  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1509  * structs used to represent other address families, it is necessary
1510  * to perform a different comparison.
1511  */
1512 
1513 #define	sa_equal(a1, a2)	\
1514 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1515 
1516 #define	sa_dl_equal(a1, a2)	\
1517 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1518 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1519 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1520 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1521 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1522 
1523 /*
1524  * Locate an interface based on a complete address.
1525  */
1526 /*ARGSUSED*/
1527 static struct ifaddr *
1528 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1529 {
1530 	struct ifnet *ifp;
1531 	struct ifaddr *ifa;
1532 
1533 	IFNET_RLOCK_NOSLEEP();
1534 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1535 		IF_ADDR_LOCK(ifp);
1536 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1537 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1538 				continue;
1539 			if (sa_equal(addr, ifa->ifa_addr)) {
1540 				if (getref)
1541 					ifa_ref(ifa);
1542 				IF_ADDR_UNLOCK(ifp);
1543 				goto done;
1544 			}
1545 			/* IP6 doesn't have broadcast */
1546 			if ((ifp->if_flags & IFF_BROADCAST) &&
1547 			    ifa->ifa_broadaddr &&
1548 			    ifa->ifa_broadaddr->sa_len != 0 &&
1549 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1550 				if (getref)
1551 					ifa_ref(ifa);
1552 				IF_ADDR_UNLOCK(ifp);
1553 				goto done;
1554 			}
1555 		}
1556 		IF_ADDR_UNLOCK(ifp);
1557 	}
1558 	ifa = NULL;
1559 done:
1560 	IFNET_RUNLOCK_NOSLEEP();
1561 	return (ifa);
1562 }
1563 
1564 struct ifaddr *
1565 ifa_ifwithaddr(struct sockaddr *addr)
1566 {
1567 
1568 	return (ifa_ifwithaddr_internal(addr, 1));
1569 }
1570 
1571 int
1572 ifa_ifwithaddr_check(struct sockaddr *addr)
1573 {
1574 
1575 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1576 }
1577 
1578 /*
1579  * Locate an interface based on the broadcast address.
1580  */
1581 /* ARGSUSED */
1582 struct ifaddr *
1583 ifa_ifwithbroadaddr(struct sockaddr *addr)
1584 {
1585 	struct ifnet *ifp;
1586 	struct ifaddr *ifa;
1587 
1588 	IFNET_RLOCK_NOSLEEP();
1589 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1590 		IF_ADDR_LOCK(ifp);
1591 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1592 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1593 				continue;
1594 			if ((ifp->if_flags & IFF_BROADCAST) &&
1595 			    ifa->ifa_broadaddr &&
1596 			    ifa->ifa_broadaddr->sa_len != 0 &&
1597 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1598 				ifa_ref(ifa);
1599 				IF_ADDR_UNLOCK(ifp);
1600 				goto done;
1601 			}
1602 		}
1603 		IF_ADDR_UNLOCK(ifp);
1604 	}
1605 	ifa = NULL;
1606 done:
1607 	IFNET_RUNLOCK_NOSLEEP();
1608 	return (ifa);
1609 }
1610 
1611 /*
1612  * Locate the point to point interface with a given destination address.
1613  */
1614 /*ARGSUSED*/
1615 struct ifaddr *
1616 ifa_ifwithdstaddr(struct sockaddr *addr)
1617 {
1618 	struct ifnet *ifp;
1619 	struct ifaddr *ifa;
1620 
1621 	IFNET_RLOCK_NOSLEEP();
1622 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1623 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1624 			continue;
1625 		IF_ADDR_LOCK(ifp);
1626 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1627 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1628 				continue;
1629 			if (ifa->ifa_dstaddr != NULL &&
1630 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1631 				ifa_ref(ifa);
1632 				IF_ADDR_UNLOCK(ifp);
1633 				goto done;
1634 			}
1635 		}
1636 		IF_ADDR_UNLOCK(ifp);
1637 	}
1638 	ifa = NULL;
1639 done:
1640 	IFNET_RUNLOCK_NOSLEEP();
1641 	return (ifa);
1642 }
1643 
1644 /*
1645  * Find an interface on a specific network.  If many, choice
1646  * is most specific found.
1647  */
1648 struct ifaddr *
1649 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1650 {
1651 	struct ifnet *ifp;
1652 	struct ifaddr *ifa;
1653 	struct ifaddr *ifa_maybe = NULL;
1654 	u_int af = addr->sa_family;
1655 	char *addr_data = addr->sa_data, *cplim;
1656 
1657 	/*
1658 	 * AF_LINK addresses can be looked up directly by their index number,
1659 	 * so do that if we can.
1660 	 */
1661 	if (af == AF_LINK) {
1662 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1663 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1664 		return (ifaddr_byindex(sdl->sdl_index));
1665 	}
1666 
1667 	/*
1668 	 * Scan though each interface, looking for ones that have addresses
1669 	 * in this address family.  Maintain a reference on ifa_maybe once
1670 	 * we find one, as we release the IF_ADDR_LOCK() that kept it stable
1671 	 * when we move onto the next interface.
1672 	 */
1673 	IFNET_RLOCK_NOSLEEP();
1674 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1675 		IF_ADDR_LOCK(ifp);
1676 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1677 			char *cp, *cp2, *cp3;
1678 
1679 			if (ifa->ifa_addr->sa_family != af)
1680 next:				continue;
1681 			if (af == AF_INET &&
1682 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1683 				/*
1684 				 * This is a bit broken as it doesn't
1685 				 * take into account that the remote end may
1686 				 * be a single node in the network we are
1687 				 * looking for.
1688 				 * The trouble is that we don't know the
1689 				 * netmask for the remote end.
1690 				 */
1691 				if (ifa->ifa_dstaddr != NULL &&
1692 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1693 					ifa_ref(ifa);
1694 					IF_ADDR_UNLOCK(ifp);
1695 					goto done;
1696 				}
1697 			} else {
1698 				/*
1699 				 * if we have a special address handler,
1700 				 * then use it instead of the generic one.
1701 				 */
1702 				if (ifa->ifa_claim_addr) {
1703 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1704 						ifa_ref(ifa);
1705 						IF_ADDR_UNLOCK(ifp);
1706 						goto done;
1707 					}
1708 					continue;
1709 				}
1710 
1711 				/*
1712 				 * Scan all the bits in the ifa's address.
1713 				 * If a bit dissagrees with what we are
1714 				 * looking for, mask it with the netmask
1715 				 * to see if it really matters.
1716 				 * (A byte at a time)
1717 				 */
1718 				if (ifa->ifa_netmask == 0)
1719 					continue;
1720 				cp = addr_data;
1721 				cp2 = ifa->ifa_addr->sa_data;
1722 				cp3 = ifa->ifa_netmask->sa_data;
1723 				cplim = ifa->ifa_netmask->sa_len
1724 					+ (char *)ifa->ifa_netmask;
1725 				while (cp3 < cplim)
1726 					if ((*cp++ ^ *cp2++) & *cp3++)
1727 						goto next; /* next address! */
1728 				/*
1729 				 * If the netmask of what we just found
1730 				 * is more specific than what we had before
1731 				 * (if we had one) then remember the new one
1732 				 * before continuing to search
1733 				 * for an even better one.
1734 				 */
1735 				if (ifa_maybe == NULL ||
1736 				    rn_refines((caddr_t)ifa->ifa_netmask,
1737 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1738 					if (ifa_maybe != NULL)
1739 						ifa_free(ifa_maybe);
1740 					ifa_maybe = ifa;
1741 					ifa_ref(ifa_maybe);
1742 				}
1743 			}
1744 		}
1745 		IF_ADDR_UNLOCK(ifp);
1746 	}
1747 	ifa = ifa_maybe;
1748 	ifa_maybe = NULL;
1749 done:
1750 	IFNET_RUNLOCK_NOSLEEP();
1751 	if (ifa_maybe != NULL)
1752 		ifa_free(ifa_maybe);
1753 	return (ifa);
1754 }
1755 
1756 /*
1757  * Find an interface address specific to an interface best matching
1758  * a given address.
1759  */
1760 struct ifaddr *
1761 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1762 {
1763 	struct ifaddr *ifa;
1764 	char *cp, *cp2, *cp3;
1765 	char *cplim;
1766 	struct ifaddr *ifa_maybe = NULL;
1767 	u_int af = addr->sa_family;
1768 
1769 	if (af >= AF_MAX)
1770 		return (NULL);
1771 	IF_ADDR_LOCK(ifp);
1772 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1773 		if (ifa->ifa_addr->sa_family != af)
1774 			continue;
1775 		if (ifa_maybe == NULL)
1776 			ifa_maybe = ifa;
1777 		if (ifa->ifa_netmask == 0) {
1778 			if (sa_equal(addr, ifa->ifa_addr) ||
1779 			    (ifa->ifa_dstaddr &&
1780 			    sa_equal(addr, ifa->ifa_dstaddr)))
1781 				goto done;
1782 			continue;
1783 		}
1784 		if (ifp->if_flags & IFF_POINTOPOINT) {
1785 			if (sa_equal(addr, ifa->ifa_dstaddr))
1786 				goto done;
1787 		} else {
1788 			cp = addr->sa_data;
1789 			cp2 = ifa->ifa_addr->sa_data;
1790 			cp3 = ifa->ifa_netmask->sa_data;
1791 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1792 			for (; cp3 < cplim; cp3++)
1793 				if ((*cp++ ^ *cp2++) & *cp3)
1794 					break;
1795 			if (cp3 == cplim)
1796 				goto done;
1797 		}
1798 	}
1799 	ifa = ifa_maybe;
1800 done:
1801 	if (ifa != NULL)
1802 		ifa_ref(ifa);
1803 	IF_ADDR_UNLOCK(ifp);
1804 	return (ifa);
1805 }
1806 
1807 #include <net/if_llatbl.h>
1808 
1809 /*
1810  * Default action when installing a route with a Link Level gateway.
1811  * Lookup an appropriate real ifa to point to.
1812  * This should be moved to /sys/net/link.c eventually.
1813  */
1814 static void
1815 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1816 {
1817 	struct ifaddr *ifa, *oifa;
1818 	struct sockaddr *dst;
1819 	struct ifnet *ifp;
1820 
1821 	RT_LOCK_ASSERT(rt);
1822 
1823 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1824 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1825 		return;
1826 	ifa = ifaof_ifpforaddr(dst, ifp);
1827 	if (ifa) {
1828 		oifa = rt->rt_ifa;
1829 		rt->rt_ifa = ifa;
1830 		ifa_free(oifa);
1831 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1832 			ifa->ifa_rtrequest(cmd, rt, info);
1833 	}
1834 }
1835 
1836 /*
1837  * Mark an interface down and notify protocols of
1838  * the transition.
1839  * NOTE: must be called at splnet or eqivalent.
1840  */
1841 static void
1842 if_unroute(struct ifnet *ifp, int flag, int fam)
1843 {
1844 	struct ifaddr *ifa;
1845 
1846 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1847 
1848 	ifp->if_flags &= ~flag;
1849 	getmicrotime(&ifp->if_lastchange);
1850 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1851 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1852 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1853 	ifp->if_qflush(ifp);
1854 
1855 	if (ifp->if_carp)
1856 		(*carp_linkstate_p)(ifp);
1857 	rt_ifmsg(ifp);
1858 }
1859 
1860 /*
1861  * Mark an interface up and notify protocols of
1862  * the transition.
1863  * NOTE: must be called at splnet or eqivalent.
1864  */
1865 static void
1866 if_route(struct ifnet *ifp, int flag, int fam)
1867 {
1868 	struct ifaddr *ifa;
1869 
1870 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1871 
1872 	ifp->if_flags |= flag;
1873 	getmicrotime(&ifp->if_lastchange);
1874 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1875 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1876 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1877 	if (ifp->if_carp)
1878 		(*carp_linkstate_p)(ifp);
1879 	rt_ifmsg(ifp);
1880 #ifdef INET6
1881 	in6_if_up(ifp);
1882 #endif
1883 }
1884 
1885 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1886 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1887 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1888 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1889 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
1890 int	(*vlan_setcookie_p)(struct ifnet *, void *);
1891 void	*(*vlan_cookie_p)(struct ifnet *);
1892 
1893 /*
1894  * Handle a change in the interface link state. To avoid LORs
1895  * between driver lock and upper layer locks, as well as possible
1896  * recursions, we post event to taskqueue, and all job
1897  * is done in static do_link_state_change().
1898  */
1899 void
1900 if_link_state_change(struct ifnet *ifp, int link_state)
1901 {
1902 	/* Return if state hasn't changed. */
1903 	if (ifp->if_link_state == link_state)
1904 		return;
1905 
1906 	ifp->if_link_state = link_state;
1907 
1908 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1909 }
1910 
1911 static void
1912 do_link_state_change(void *arg, int pending)
1913 {
1914 	struct ifnet *ifp = (struct ifnet *)arg;
1915 	int link_state = ifp->if_link_state;
1916 	CURVNET_SET(ifp->if_vnet);
1917 
1918 	/* Notify that the link state has changed. */
1919 	rt_ifmsg(ifp);
1920 	if (ifp->if_vlantrunk != NULL)
1921 		(*vlan_link_state_p)(ifp);
1922 
1923 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1924 	    IFP2AC(ifp)->ac_netgraph != NULL)
1925 		(*ng_ether_link_state_p)(ifp, link_state);
1926 	if (ifp->if_carp)
1927 		(*carp_linkstate_p)(ifp);
1928 	if (ifp->if_bridge) {
1929 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1930 		(*bstp_linkstate_p)(ifp, link_state);
1931 	}
1932 	if (ifp->if_lagg) {
1933 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1934 		(*lagg_linkstate_p)(ifp, link_state);
1935 	}
1936 
1937 	if (IS_DEFAULT_VNET(curvnet))
1938 		devctl_notify("IFNET", ifp->if_xname,
1939 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1940 		    NULL);
1941 	if (pending > 1)
1942 		if_printf(ifp, "%d link states coalesced\n", pending);
1943 	if (log_link_state_change)
1944 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1945 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1946 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
1947 	CURVNET_RESTORE();
1948 }
1949 
1950 /*
1951  * Mark an interface down and notify protocols of
1952  * the transition.
1953  * NOTE: must be called at splnet or eqivalent.
1954  */
1955 void
1956 if_down(struct ifnet *ifp)
1957 {
1958 
1959 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1960 }
1961 
1962 /*
1963  * Mark an interface up and notify protocols of
1964  * the transition.
1965  * NOTE: must be called at splnet or eqivalent.
1966  */
1967 void
1968 if_up(struct ifnet *ifp)
1969 {
1970 
1971 	if_route(ifp, IFF_UP, AF_UNSPEC);
1972 }
1973 
1974 /*
1975  * Flush an interface queue.
1976  */
1977 void
1978 if_qflush(struct ifnet *ifp)
1979 {
1980 	struct mbuf *m, *n;
1981 	struct ifaltq *ifq;
1982 
1983 	ifq = &ifp->if_snd;
1984 	IFQ_LOCK(ifq);
1985 #ifdef ALTQ
1986 	if (ALTQ_IS_ENABLED(ifq))
1987 		ALTQ_PURGE(ifq);
1988 #endif
1989 	n = ifq->ifq_head;
1990 	while ((m = n) != 0) {
1991 		n = m->m_act;
1992 		m_freem(m);
1993 	}
1994 	ifq->ifq_head = 0;
1995 	ifq->ifq_tail = 0;
1996 	ifq->ifq_len = 0;
1997 	IFQ_UNLOCK(ifq);
1998 }
1999 
2000 /*
2001  * Map interface name to interface structure pointer, with or without
2002  * returning a reference.
2003  */
2004 struct ifnet *
2005 ifunit_ref(const char *name)
2006 {
2007 	struct ifnet *ifp;
2008 
2009 	IFNET_RLOCK_NOSLEEP();
2010 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2011 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2012 		    !(ifp->if_flags & IFF_DYING))
2013 			break;
2014 	}
2015 	if (ifp != NULL)
2016 		if_ref(ifp);
2017 	IFNET_RUNLOCK_NOSLEEP();
2018 	return (ifp);
2019 }
2020 
2021 struct ifnet *
2022 ifunit(const char *name)
2023 {
2024 	struct ifnet *ifp;
2025 
2026 	IFNET_RLOCK_NOSLEEP();
2027 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2028 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2029 			break;
2030 	}
2031 	IFNET_RUNLOCK_NOSLEEP();
2032 	return (ifp);
2033 }
2034 
2035 /*
2036  * Hardware specific interface ioctls.
2037  */
2038 static int
2039 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2040 {
2041 	struct ifreq *ifr;
2042 	struct ifstat *ifs;
2043 	int error = 0;
2044 	int new_flags, temp_flags;
2045 	size_t namelen, onamelen;
2046 	size_t descrlen;
2047 	char *descrbuf, *odescrbuf;
2048 	char new_name[IFNAMSIZ];
2049 	struct ifaddr *ifa;
2050 	struct sockaddr_dl *sdl;
2051 
2052 	ifr = (struct ifreq *)data;
2053 	switch (cmd) {
2054 	case SIOCGIFINDEX:
2055 		ifr->ifr_index = ifp->if_index;
2056 		break;
2057 
2058 	case SIOCGIFFLAGS:
2059 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2060 		ifr->ifr_flags = temp_flags & 0xffff;
2061 		ifr->ifr_flagshigh = temp_flags >> 16;
2062 		break;
2063 
2064 	case SIOCGIFCAP:
2065 		ifr->ifr_reqcap = ifp->if_capabilities;
2066 		ifr->ifr_curcap = ifp->if_capenable;
2067 		break;
2068 
2069 #ifdef MAC
2070 	case SIOCGIFMAC:
2071 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2072 		break;
2073 #endif
2074 
2075 	case SIOCGIFMETRIC:
2076 		ifr->ifr_metric = ifp->if_metric;
2077 		break;
2078 
2079 	case SIOCGIFMTU:
2080 		ifr->ifr_mtu = ifp->if_mtu;
2081 		break;
2082 
2083 	case SIOCGIFPHYS:
2084 		ifr->ifr_phys = ifp->if_physical;
2085 		break;
2086 
2087 	case SIOCGIFDESCR:
2088 		error = 0;
2089 		sx_slock(&ifdescr_sx);
2090 		if (ifp->if_description == NULL)
2091 			error = ENOMSG;
2092 		else {
2093 			/* space for terminating nul */
2094 			descrlen = strlen(ifp->if_description) + 1;
2095 			if (ifr->ifr_buffer.length < descrlen)
2096 				ifr->ifr_buffer.buffer = NULL;
2097 			else
2098 				error = copyout(ifp->if_description,
2099 				    ifr->ifr_buffer.buffer, descrlen);
2100 			ifr->ifr_buffer.length = descrlen;
2101 		}
2102 		sx_sunlock(&ifdescr_sx);
2103 		break;
2104 
2105 	case SIOCSIFDESCR:
2106 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2107 		if (error)
2108 			return (error);
2109 
2110 		/*
2111 		 * Copy only (length-1) bytes to make sure that
2112 		 * if_description is always nul terminated.  The
2113 		 * length parameter is supposed to count the
2114 		 * terminating nul in.
2115 		 */
2116 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2117 			return (ENAMETOOLONG);
2118 		else if (ifr->ifr_buffer.length == 0)
2119 			descrbuf = NULL;
2120 		else {
2121 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2122 			    M_WAITOK | M_ZERO);
2123 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2124 			    ifr->ifr_buffer.length - 1);
2125 			if (error) {
2126 				free(descrbuf, M_IFDESCR);
2127 				break;
2128 			}
2129 		}
2130 
2131 		sx_xlock(&ifdescr_sx);
2132 		odescrbuf = ifp->if_description;
2133 		ifp->if_description = descrbuf;
2134 		sx_xunlock(&ifdescr_sx);
2135 
2136 		getmicrotime(&ifp->if_lastchange);
2137 		free(odescrbuf, M_IFDESCR);
2138 		break;
2139 
2140 	case SIOCGIFFIB:
2141 		ifr->ifr_fib = ifp->if_fib;
2142 		break;
2143 
2144 	case SIOCSIFFIB:
2145 		error = priv_check(td, PRIV_NET_SETIFFIB);
2146 		if (error)
2147 			return (error);
2148 		if (ifr->ifr_fib >= rt_numfibs)
2149 			return (EINVAL);
2150 
2151 		ifp->if_fib = ifr->ifr_fib;
2152 		break;
2153 
2154 	case SIOCSIFFLAGS:
2155 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2156 		if (error)
2157 			return (error);
2158 		/*
2159 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2160 		 * check, so we don't need special handling here yet.
2161 		 */
2162 		new_flags = (ifr->ifr_flags & 0xffff) |
2163 		    (ifr->ifr_flagshigh << 16);
2164 		if (ifp->if_flags & IFF_SMART) {
2165 			/* Smart drivers twiddle their own routes */
2166 		} else if (ifp->if_flags & IFF_UP &&
2167 		    (new_flags & IFF_UP) == 0) {
2168 			int s = splimp();
2169 			if_down(ifp);
2170 			splx(s);
2171 		} else if (new_flags & IFF_UP &&
2172 		    (ifp->if_flags & IFF_UP) == 0) {
2173 			int s = splimp();
2174 			if_up(ifp);
2175 			splx(s);
2176 		}
2177 		/* See if permanently promiscuous mode bit is about to flip */
2178 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2179 			if (new_flags & IFF_PPROMISC)
2180 				ifp->if_flags |= IFF_PROMISC;
2181 			else if (ifp->if_pcount == 0)
2182 				ifp->if_flags &= ~IFF_PROMISC;
2183 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2184 			    ifp->if_xname,
2185 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2186 		}
2187 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2188 			(new_flags &~ IFF_CANTCHANGE);
2189 		if (ifp->if_ioctl) {
2190 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2191 		}
2192 		getmicrotime(&ifp->if_lastchange);
2193 		break;
2194 
2195 	case SIOCSIFCAP:
2196 		error = priv_check(td, PRIV_NET_SETIFCAP);
2197 		if (error)
2198 			return (error);
2199 		if (ifp->if_ioctl == NULL)
2200 			return (EOPNOTSUPP);
2201 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2202 			return (EINVAL);
2203 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2204 		if (error == 0)
2205 			getmicrotime(&ifp->if_lastchange);
2206 		break;
2207 
2208 #ifdef MAC
2209 	case SIOCSIFMAC:
2210 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2211 		break;
2212 #endif
2213 
2214 	case SIOCSIFNAME:
2215 		error = priv_check(td, PRIV_NET_SETIFNAME);
2216 		if (error)
2217 			return (error);
2218 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2219 		if (error != 0)
2220 			return (error);
2221 		if (new_name[0] == '\0')
2222 			return (EINVAL);
2223 		if (ifunit(new_name) != NULL)
2224 			return (EEXIST);
2225 
2226 		/*
2227 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2228 		 * and there are numerous other races with the
2229 		 * ifunit() checks not being atomic with namespace
2230 		 * changes (renames, vmoves, if_attach, etc).
2231 		 */
2232 		ifp->if_flags |= IFF_RENAMING;
2233 
2234 		/* Announce the departure of the interface. */
2235 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2236 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2237 
2238 		log(LOG_INFO, "%s: changing name to '%s'\n",
2239 		    ifp->if_xname, new_name);
2240 
2241 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2242 		ifa = ifp->if_addr;
2243 		IFA_LOCK(ifa);
2244 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2245 		namelen = strlen(new_name);
2246 		onamelen = sdl->sdl_nlen;
2247 		/*
2248 		 * Move the address if needed.  This is safe because we
2249 		 * allocate space for a name of length IFNAMSIZ when we
2250 		 * create this in if_attach().
2251 		 */
2252 		if (namelen != onamelen) {
2253 			bcopy(sdl->sdl_data + onamelen,
2254 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2255 		}
2256 		bcopy(new_name, sdl->sdl_data, namelen);
2257 		sdl->sdl_nlen = namelen;
2258 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2259 		bzero(sdl->sdl_data, onamelen);
2260 		while (namelen != 0)
2261 			sdl->sdl_data[--namelen] = 0xff;
2262 		IFA_UNLOCK(ifa);
2263 
2264 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2265 		/* Announce the return of the interface. */
2266 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2267 
2268 		ifp->if_flags &= ~IFF_RENAMING;
2269 		break;
2270 
2271 #ifdef VIMAGE
2272 	case SIOCSIFVNET:
2273 		error = priv_check(td, PRIV_NET_SETIFVNET);
2274 		if (error)
2275 			return (error);
2276 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2277 		break;
2278 #endif
2279 
2280 	case SIOCSIFMETRIC:
2281 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2282 		if (error)
2283 			return (error);
2284 		ifp->if_metric = ifr->ifr_metric;
2285 		getmicrotime(&ifp->if_lastchange);
2286 		break;
2287 
2288 	case SIOCSIFPHYS:
2289 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2290 		if (error)
2291 			return (error);
2292 		if (ifp->if_ioctl == NULL)
2293 			return (EOPNOTSUPP);
2294 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2295 		if (error == 0)
2296 			getmicrotime(&ifp->if_lastchange);
2297 		break;
2298 
2299 	case SIOCSIFMTU:
2300 	{
2301 		u_long oldmtu = ifp->if_mtu;
2302 
2303 		error = priv_check(td, PRIV_NET_SETIFMTU);
2304 		if (error)
2305 			return (error);
2306 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2307 			return (EINVAL);
2308 		if (ifp->if_ioctl == NULL)
2309 			return (EOPNOTSUPP);
2310 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2311 		if (error == 0) {
2312 			getmicrotime(&ifp->if_lastchange);
2313 			rt_ifmsg(ifp);
2314 		}
2315 		/*
2316 		 * If the link MTU changed, do network layer specific procedure.
2317 		 */
2318 		if (ifp->if_mtu != oldmtu) {
2319 #ifdef INET6
2320 			nd6_setmtu(ifp);
2321 #endif
2322 		}
2323 		break;
2324 	}
2325 
2326 	case SIOCADDMULTI:
2327 	case SIOCDELMULTI:
2328 		if (cmd == SIOCADDMULTI)
2329 			error = priv_check(td, PRIV_NET_ADDMULTI);
2330 		else
2331 			error = priv_check(td, PRIV_NET_DELMULTI);
2332 		if (error)
2333 			return (error);
2334 
2335 		/* Don't allow group membership on non-multicast interfaces. */
2336 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2337 			return (EOPNOTSUPP);
2338 
2339 		/* Don't let users screw up protocols' entries. */
2340 		if (ifr->ifr_addr.sa_family != AF_LINK)
2341 			return (EINVAL);
2342 
2343 		if (cmd == SIOCADDMULTI) {
2344 			struct ifmultiaddr *ifma;
2345 
2346 			/*
2347 			 * Userland is only permitted to join groups once
2348 			 * via the if_addmulti() KPI, because it cannot hold
2349 			 * struct ifmultiaddr * between calls. It may also
2350 			 * lose a race while we check if the membership
2351 			 * already exists.
2352 			 */
2353 			IF_ADDR_LOCK(ifp);
2354 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2355 			IF_ADDR_UNLOCK(ifp);
2356 			if (ifma != NULL)
2357 				error = EADDRINUSE;
2358 			else
2359 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2360 		} else {
2361 			error = if_delmulti(ifp, &ifr->ifr_addr);
2362 		}
2363 		if (error == 0)
2364 			getmicrotime(&ifp->if_lastchange);
2365 		break;
2366 
2367 	case SIOCSIFPHYADDR:
2368 	case SIOCDIFPHYADDR:
2369 #ifdef INET6
2370 	case SIOCSIFPHYADDR_IN6:
2371 #endif
2372 	case SIOCSLIFPHYADDR:
2373 	case SIOCSIFMEDIA:
2374 	case SIOCSIFGENERIC:
2375 		error = priv_check(td, PRIV_NET_HWIOCTL);
2376 		if (error)
2377 			return (error);
2378 		if (ifp->if_ioctl == NULL)
2379 			return (EOPNOTSUPP);
2380 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2381 		if (error == 0)
2382 			getmicrotime(&ifp->if_lastchange);
2383 		break;
2384 
2385 	case SIOCGIFSTATUS:
2386 		ifs = (struct ifstat *)data;
2387 		ifs->ascii[0] = '\0';
2388 
2389 	case SIOCGIFPSRCADDR:
2390 	case SIOCGIFPDSTADDR:
2391 	case SIOCGLIFPHYADDR:
2392 	case SIOCGIFMEDIA:
2393 	case SIOCGIFGENERIC:
2394 		if (ifp->if_ioctl == NULL)
2395 			return (EOPNOTSUPP);
2396 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2397 		break;
2398 
2399 	case SIOCSIFLLADDR:
2400 		error = priv_check(td, PRIV_NET_SETLLADDR);
2401 		if (error)
2402 			return (error);
2403 		error = if_setlladdr(ifp,
2404 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2405 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2406 		break;
2407 
2408 	case SIOCAIFGROUP:
2409 	{
2410 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2411 
2412 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2413 		if (error)
2414 			return (error);
2415 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2416 			return (error);
2417 		break;
2418 	}
2419 
2420 	case SIOCGIFGROUP:
2421 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2422 			return (error);
2423 		break;
2424 
2425 	case SIOCDIFGROUP:
2426 	{
2427 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2428 
2429 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2430 		if (error)
2431 			return (error);
2432 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2433 			return (error);
2434 		break;
2435 	}
2436 
2437 	default:
2438 		error = ENOIOCTL;
2439 		break;
2440 	}
2441 	return (error);
2442 }
2443 
2444 #ifdef COMPAT_FREEBSD32
2445 struct ifconf32 {
2446 	int32_t	ifc_len;
2447 	union {
2448 		uint32_t	ifcu_buf;
2449 		uint32_t	ifcu_req;
2450 	} ifc_ifcu;
2451 };
2452 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2453 #endif
2454 
2455 /*
2456  * Interface ioctls.
2457  */
2458 int
2459 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2460 {
2461 	struct ifnet *ifp;
2462 	struct ifreq *ifr;
2463 	int error;
2464 	int oif_flags;
2465 
2466 	CURVNET_SET(so->so_vnet);
2467 	switch (cmd) {
2468 	case SIOCGIFCONF:
2469 	case OSIOCGIFCONF:
2470 		error = ifconf(cmd, data);
2471 		CURVNET_RESTORE();
2472 		return (error);
2473 
2474 #ifdef COMPAT_FREEBSD32
2475 	case SIOCGIFCONF32:
2476 		{
2477 			struct ifconf32 *ifc32;
2478 			struct ifconf ifc;
2479 
2480 			ifc32 = (struct ifconf32 *)data;
2481 			ifc.ifc_len = ifc32->ifc_len;
2482 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2483 
2484 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2485 			CURVNET_RESTORE();
2486 			if (error == 0)
2487 				ifc32->ifc_len = ifc.ifc_len;
2488 			return (error);
2489 		}
2490 #endif
2491 	}
2492 	ifr = (struct ifreq *)data;
2493 
2494 	switch (cmd) {
2495 #ifdef VIMAGE
2496 	case SIOCSIFRVNET:
2497 		error = priv_check(td, PRIV_NET_SETIFVNET);
2498 		if (error == 0)
2499 			error = if_vmove_reclaim(td, ifr->ifr_name,
2500 			    ifr->ifr_jid);
2501 		CURVNET_RESTORE();
2502 		return (error);
2503 #endif
2504 	case SIOCIFCREATE:
2505 	case SIOCIFCREATE2:
2506 		error = priv_check(td, PRIV_NET_IFCREATE);
2507 		if (error == 0)
2508 			error = if_clone_create(ifr->ifr_name,
2509 			    sizeof(ifr->ifr_name),
2510 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2511 		CURVNET_RESTORE();
2512 		return (error);
2513 	case SIOCIFDESTROY:
2514 		error = priv_check(td, PRIV_NET_IFDESTROY);
2515 		if (error == 0)
2516 			error = if_clone_destroy(ifr->ifr_name);
2517 		CURVNET_RESTORE();
2518 		return (error);
2519 
2520 	case SIOCIFGCLONERS:
2521 		error = if_clone_list((struct if_clonereq *)data);
2522 		CURVNET_RESTORE();
2523 		return (error);
2524 	case SIOCGIFGMEMB:
2525 		error = if_getgroupmembers((struct ifgroupreq *)data);
2526 		CURVNET_RESTORE();
2527 		return (error);
2528 	}
2529 
2530 	ifp = ifunit_ref(ifr->ifr_name);
2531 	if (ifp == NULL) {
2532 		CURVNET_RESTORE();
2533 		return (ENXIO);
2534 	}
2535 
2536 	error = ifhwioctl(cmd, ifp, data, td);
2537 	if (error != ENOIOCTL) {
2538 		if_rele(ifp);
2539 		CURVNET_RESTORE();
2540 		return (error);
2541 	}
2542 
2543 	oif_flags = ifp->if_flags;
2544 	if (so->so_proto == NULL) {
2545 		if_rele(ifp);
2546 		CURVNET_RESTORE();
2547 		return (EOPNOTSUPP);
2548 	}
2549 #ifndef COMPAT_43
2550 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2551 								 data,
2552 								 ifp, td));
2553 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2554 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2555 #else
2556 	{
2557 		u_long ocmd = cmd;
2558 
2559 		switch (cmd) {
2560 
2561 		case SIOCSIFDSTADDR:
2562 		case SIOCSIFADDR:
2563 		case SIOCSIFBRDADDR:
2564 		case SIOCSIFNETMASK:
2565 #if BYTE_ORDER != BIG_ENDIAN
2566 			if (ifr->ifr_addr.sa_family == 0 &&
2567 			    ifr->ifr_addr.sa_len < 16) {
2568 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2569 				ifr->ifr_addr.sa_len = 16;
2570 			}
2571 #else
2572 			if (ifr->ifr_addr.sa_len == 0)
2573 				ifr->ifr_addr.sa_len = 16;
2574 #endif
2575 			break;
2576 
2577 		case OSIOCGIFADDR:
2578 			cmd = SIOCGIFADDR;
2579 			break;
2580 
2581 		case OSIOCGIFDSTADDR:
2582 			cmd = SIOCGIFDSTADDR;
2583 			break;
2584 
2585 		case OSIOCGIFBRDADDR:
2586 			cmd = SIOCGIFBRDADDR;
2587 			break;
2588 
2589 		case OSIOCGIFNETMASK:
2590 			cmd = SIOCGIFNETMASK;
2591 		}
2592 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2593 								   cmd,
2594 								   data,
2595 								   ifp, td));
2596 		if (error == EOPNOTSUPP && ifp != NULL &&
2597 		    ifp->if_ioctl != NULL)
2598 			error = (*ifp->if_ioctl)(ifp, cmd, data);
2599 		switch (ocmd) {
2600 
2601 		case OSIOCGIFADDR:
2602 		case OSIOCGIFDSTADDR:
2603 		case OSIOCGIFBRDADDR:
2604 		case OSIOCGIFNETMASK:
2605 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2606 
2607 		}
2608 	}
2609 #endif /* COMPAT_43 */
2610 
2611 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2612 #ifdef INET6
2613 		if (ifp->if_flags & IFF_UP) {
2614 			int s = splimp();
2615 			in6_if_up(ifp);
2616 			splx(s);
2617 		}
2618 #endif
2619 	}
2620 	if_rele(ifp);
2621 	CURVNET_RESTORE();
2622 	return (error);
2623 }
2624 
2625 /*
2626  * The code common to handling reference counted flags,
2627  * e.g., in ifpromisc() and if_allmulti().
2628  * The "pflag" argument can specify a permanent mode flag to check,
2629  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2630  *
2631  * Only to be used on stack-owned flags, not driver-owned flags.
2632  */
2633 static int
2634 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2635 {
2636 	struct ifreq ifr;
2637 	int error;
2638 	int oldflags, oldcount;
2639 
2640 	/* Sanity checks to catch programming errors */
2641 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2642 	    ("%s: setting driver-owned flag %d", __func__, flag));
2643 
2644 	if (onswitch)
2645 		KASSERT(*refcount >= 0,
2646 		    ("%s: increment negative refcount %d for flag %d",
2647 		    __func__, *refcount, flag));
2648 	else
2649 		KASSERT(*refcount > 0,
2650 		    ("%s: decrement non-positive refcount %d for flag %d",
2651 		    __func__, *refcount, flag));
2652 
2653 	/* In case this mode is permanent, just touch refcount */
2654 	if (ifp->if_flags & pflag) {
2655 		*refcount += onswitch ? 1 : -1;
2656 		return (0);
2657 	}
2658 
2659 	/* Save ifnet parameters for if_ioctl() may fail */
2660 	oldcount = *refcount;
2661 	oldflags = ifp->if_flags;
2662 
2663 	/*
2664 	 * See if we aren't the only and touching refcount is enough.
2665 	 * Actually toggle interface flag if we are the first or last.
2666 	 */
2667 	if (onswitch) {
2668 		if ((*refcount)++)
2669 			return (0);
2670 		ifp->if_flags |= flag;
2671 	} else {
2672 		if (--(*refcount))
2673 			return (0);
2674 		ifp->if_flags &= ~flag;
2675 	}
2676 
2677 	/* Call down the driver since we've changed interface flags */
2678 	if (ifp->if_ioctl == NULL) {
2679 		error = EOPNOTSUPP;
2680 		goto recover;
2681 	}
2682 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2683 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2684 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2685 	if (error)
2686 		goto recover;
2687 	/* Notify userland that interface flags have changed */
2688 	rt_ifmsg(ifp);
2689 	return (0);
2690 
2691 recover:
2692 	/* Recover after driver error */
2693 	*refcount = oldcount;
2694 	ifp->if_flags = oldflags;
2695 	return (error);
2696 }
2697 
2698 /*
2699  * Set/clear promiscuous mode on interface ifp based on the truth value
2700  * of pswitch.  The calls are reference counted so that only the first
2701  * "on" request actually has an effect, as does the final "off" request.
2702  * Results are undefined if the "off" and "on" requests are not matched.
2703  */
2704 int
2705 ifpromisc(struct ifnet *ifp, int pswitch)
2706 {
2707 	int error;
2708 	int oldflags = ifp->if_flags;
2709 
2710 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2711 			   &ifp->if_pcount, pswitch);
2712 	/* If promiscuous mode status has changed, log a message */
2713 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2714 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2715 		    ifp->if_xname,
2716 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2717 	return (error);
2718 }
2719 
2720 /*
2721  * Return interface configuration
2722  * of system.  List may be used
2723  * in later ioctl's (above) to get
2724  * other information.
2725  */
2726 /*ARGSUSED*/
2727 static int
2728 ifconf(u_long cmd, caddr_t data)
2729 {
2730 	struct ifconf *ifc = (struct ifconf *)data;
2731 	struct ifnet *ifp;
2732 	struct ifaddr *ifa;
2733 	struct ifreq ifr;
2734 	struct sbuf *sb;
2735 	int error, full = 0, valid_len, max_len;
2736 
2737 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2738 	max_len = MAXPHYS - 1;
2739 
2740 	/* Prevent hostile input from being able to crash the system */
2741 	if (ifc->ifc_len <= 0)
2742 		return (EINVAL);
2743 
2744 again:
2745 	if (ifc->ifc_len <= max_len) {
2746 		max_len = ifc->ifc_len;
2747 		full = 1;
2748 	}
2749 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2750 	max_len = 0;
2751 	valid_len = 0;
2752 
2753 	IFNET_RLOCK();
2754 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2755 		int addrs;
2756 
2757 		/*
2758 		 * Zero the ifr_name buffer to make sure we don't
2759 		 * disclose the contents of the stack.
2760 		 */
2761 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2762 
2763 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2764 		    >= sizeof(ifr.ifr_name)) {
2765 			sbuf_delete(sb);
2766 			IFNET_RUNLOCK();
2767 			return (ENAMETOOLONG);
2768 		}
2769 
2770 		addrs = 0;
2771 		IF_ADDR_LOCK(ifp);
2772 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2773 			struct sockaddr *sa = ifa->ifa_addr;
2774 
2775 			if (prison_if(curthread->td_ucred, sa) != 0)
2776 				continue;
2777 			addrs++;
2778 #ifdef COMPAT_43
2779 			if (cmd == OSIOCGIFCONF) {
2780 				struct osockaddr *osa =
2781 					 (struct osockaddr *)&ifr.ifr_addr;
2782 				ifr.ifr_addr = *sa;
2783 				osa->sa_family = sa->sa_family;
2784 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2785 				max_len += sizeof(ifr);
2786 			} else
2787 #endif
2788 			if (sa->sa_len <= sizeof(*sa)) {
2789 				ifr.ifr_addr = *sa;
2790 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2791 				max_len += sizeof(ifr);
2792 			} else {
2793 				sbuf_bcat(sb, &ifr,
2794 				    offsetof(struct ifreq, ifr_addr));
2795 				max_len += offsetof(struct ifreq, ifr_addr);
2796 				sbuf_bcat(sb, sa, sa->sa_len);
2797 				max_len += sa->sa_len;
2798 			}
2799 
2800 			if (sbuf_error(sb) == 0)
2801 				valid_len = sbuf_len(sb);
2802 		}
2803 		IF_ADDR_UNLOCK(ifp);
2804 		if (addrs == 0) {
2805 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2806 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2807 			max_len += sizeof(ifr);
2808 
2809 			if (sbuf_error(sb) == 0)
2810 				valid_len = sbuf_len(sb);
2811 		}
2812 	}
2813 	IFNET_RUNLOCK();
2814 
2815 	/*
2816 	 * If we didn't allocate enough space (uncommon), try again.  If
2817 	 * we have already allocated as much space as we are allowed,
2818 	 * return what we've got.
2819 	 */
2820 	if (valid_len != max_len && !full) {
2821 		sbuf_delete(sb);
2822 		goto again;
2823 	}
2824 
2825 	ifc->ifc_len = valid_len;
2826 	sbuf_finish(sb);
2827 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2828 	sbuf_delete(sb);
2829 	return (error);
2830 }
2831 
2832 /*
2833  * Just like ifpromisc(), but for all-multicast-reception mode.
2834  */
2835 int
2836 if_allmulti(struct ifnet *ifp, int onswitch)
2837 {
2838 
2839 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2840 }
2841 
2842 struct ifmultiaddr *
2843 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2844 {
2845 	struct ifmultiaddr *ifma;
2846 
2847 	IF_ADDR_LOCK_ASSERT(ifp);
2848 
2849 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2850 		if (sa->sa_family == AF_LINK) {
2851 			if (sa_dl_equal(ifma->ifma_addr, sa))
2852 				break;
2853 		} else {
2854 			if (sa_equal(ifma->ifma_addr, sa))
2855 				break;
2856 		}
2857 	}
2858 
2859 	return ifma;
2860 }
2861 
2862 /*
2863  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2864  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2865  * the ifnet multicast address list here, so the caller must do that and
2866  * other setup work (such as notifying the device driver).  The reference
2867  * count is initialized to 1.
2868  */
2869 static struct ifmultiaddr *
2870 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2871     int mflags)
2872 {
2873 	struct ifmultiaddr *ifma;
2874 	struct sockaddr *dupsa;
2875 
2876 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2877 	    M_ZERO);
2878 	if (ifma == NULL)
2879 		return (NULL);
2880 
2881 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2882 	if (dupsa == NULL) {
2883 		free(ifma, M_IFMADDR);
2884 		return (NULL);
2885 	}
2886 	bcopy(sa, dupsa, sa->sa_len);
2887 	ifma->ifma_addr = dupsa;
2888 
2889 	ifma->ifma_ifp = ifp;
2890 	ifma->ifma_refcount = 1;
2891 	ifma->ifma_protospec = NULL;
2892 
2893 	if (llsa == NULL) {
2894 		ifma->ifma_lladdr = NULL;
2895 		return (ifma);
2896 	}
2897 
2898 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2899 	if (dupsa == NULL) {
2900 		free(ifma->ifma_addr, M_IFMADDR);
2901 		free(ifma, M_IFMADDR);
2902 		return (NULL);
2903 	}
2904 	bcopy(llsa, dupsa, llsa->sa_len);
2905 	ifma->ifma_lladdr = dupsa;
2906 
2907 	return (ifma);
2908 }
2909 
2910 /*
2911  * if_freemulti: free ifmultiaddr structure and possibly attached related
2912  * addresses.  The caller is responsible for implementing reference
2913  * counting, notifying the driver, handling routing messages, and releasing
2914  * any dependent link layer state.
2915  */
2916 static void
2917 if_freemulti(struct ifmultiaddr *ifma)
2918 {
2919 
2920 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2921 	    ifma->ifma_refcount));
2922 	KASSERT(ifma->ifma_protospec == NULL,
2923 	    ("if_freemulti: protospec not NULL"));
2924 
2925 	if (ifma->ifma_lladdr != NULL)
2926 		free(ifma->ifma_lladdr, M_IFMADDR);
2927 	free(ifma->ifma_addr, M_IFMADDR);
2928 	free(ifma, M_IFMADDR);
2929 }
2930 
2931 /*
2932  * Register an additional multicast address with a network interface.
2933  *
2934  * - If the address is already present, bump the reference count on the
2935  *   address and return.
2936  * - If the address is not link-layer, look up a link layer address.
2937  * - Allocate address structures for one or both addresses, and attach to the
2938  *   multicast address list on the interface.  If automatically adding a link
2939  *   layer address, the protocol address will own a reference to the link
2940  *   layer address, to be freed when it is freed.
2941  * - Notify the network device driver of an addition to the multicast address
2942  *   list.
2943  *
2944  * 'sa' points to caller-owned memory with the desired multicast address.
2945  *
2946  * 'retifma' will be used to return a pointer to the resulting multicast
2947  * address reference, if desired.
2948  */
2949 int
2950 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2951     struct ifmultiaddr **retifma)
2952 {
2953 	struct ifmultiaddr *ifma, *ll_ifma;
2954 	struct sockaddr *llsa;
2955 	int error;
2956 
2957 	/*
2958 	 * If the address is already present, return a new reference to it;
2959 	 * otherwise, allocate storage and set up a new address.
2960 	 */
2961 	IF_ADDR_LOCK(ifp);
2962 	ifma = if_findmulti(ifp, sa);
2963 	if (ifma != NULL) {
2964 		ifma->ifma_refcount++;
2965 		if (retifma != NULL)
2966 			*retifma = ifma;
2967 		IF_ADDR_UNLOCK(ifp);
2968 		return (0);
2969 	}
2970 
2971 	/*
2972 	 * The address isn't already present; resolve the protocol address
2973 	 * into a link layer address, and then look that up, bump its
2974 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2975 	 * returned, we will need to free it later.
2976 	 */
2977 	llsa = NULL;
2978 	ll_ifma = NULL;
2979 	if (ifp->if_resolvemulti != NULL) {
2980 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2981 		if (error)
2982 			goto unlock_out;
2983 	}
2984 
2985 	/*
2986 	 * Allocate the new address.  Don't hook it up yet, as we may also
2987 	 * need to allocate a link layer multicast address.
2988 	 */
2989 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2990 	if (ifma == NULL) {
2991 		error = ENOMEM;
2992 		goto free_llsa_out;
2993 	}
2994 
2995 	/*
2996 	 * If a link layer address is found, we'll need to see if it's
2997 	 * already present in the address list, or allocate is as well.
2998 	 * When this block finishes, the link layer address will be on the
2999 	 * list.
3000 	 */
3001 	if (llsa != NULL) {
3002 		ll_ifma = if_findmulti(ifp, llsa);
3003 		if (ll_ifma == NULL) {
3004 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3005 			if (ll_ifma == NULL) {
3006 				--ifma->ifma_refcount;
3007 				if_freemulti(ifma);
3008 				error = ENOMEM;
3009 				goto free_llsa_out;
3010 			}
3011 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3012 			    ifma_link);
3013 		} else
3014 			ll_ifma->ifma_refcount++;
3015 		ifma->ifma_llifma = ll_ifma;
3016 	}
3017 
3018 	/*
3019 	 * We now have a new multicast address, ifma, and possibly a new or
3020 	 * referenced link layer address.  Add the primary address to the
3021 	 * ifnet address list.
3022 	 */
3023 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3024 
3025 	if (retifma != NULL)
3026 		*retifma = ifma;
3027 
3028 	/*
3029 	 * Must generate the message while holding the lock so that 'ifma'
3030 	 * pointer is still valid.
3031 	 */
3032 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3033 	IF_ADDR_UNLOCK(ifp);
3034 
3035 	/*
3036 	 * We are certain we have added something, so call down to the
3037 	 * interface to let them know about it.
3038 	 */
3039 	if (ifp->if_ioctl != NULL) {
3040 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3041 	}
3042 
3043 	if (llsa != NULL)
3044 		free(llsa, M_IFMADDR);
3045 
3046 	return (0);
3047 
3048 free_llsa_out:
3049 	if (llsa != NULL)
3050 		free(llsa, M_IFMADDR);
3051 
3052 unlock_out:
3053 	IF_ADDR_UNLOCK(ifp);
3054 	return (error);
3055 }
3056 
3057 /*
3058  * Delete a multicast group membership by network-layer group address.
3059  *
3060  * Returns ENOENT if the entry could not be found. If ifp no longer
3061  * exists, results are undefined. This entry point should only be used
3062  * from subsystems which do appropriate locking to hold ifp for the
3063  * duration of the call.
3064  * Network-layer protocol domains must use if_delmulti_ifma().
3065  */
3066 int
3067 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3068 {
3069 	struct ifmultiaddr *ifma;
3070 	int lastref;
3071 #ifdef INVARIANTS
3072 	struct ifnet *oifp;
3073 
3074 	IFNET_RLOCK_NOSLEEP();
3075 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3076 		if (ifp == oifp)
3077 			break;
3078 	if (ifp != oifp)
3079 		ifp = NULL;
3080 	IFNET_RUNLOCK_NOSLEEP();
3081 
3082 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3083 #endif
3084 	if (ifp == NULL)
3085 		return (ENOENT);
3086 
3087 	IF_ADDR_LOCK(ifp);
3088 	lastref = 0;
3089 	ifma = if_findmulti(ifp, sa);
3090 	if (ifma != NULL)
3091 		lastref = if_delmulti_locked(ifp, ifma, 0);
3092 	IF_ADDR_UNLOCK(ifp);
3093 
3094 	if (ifma == NULL)
3095 		return (ENOENT);
3096 
3097 	if (lastref && ifp->if_ioctl != NULL) {
3098 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3099 	}
3100 
3101 	return (0);
3102 }
3103 
3104 /*
3105  * Delete all multicast group membership for an interface.
3106  * Should be used to quickly flush all multicast filters.
3107  */
3108 void
3109 if_delallmulti(struct ifnet *ifp)
3110 {
3111 	struct ifmultiaddr *ifma;
3112 	struct ifmultiaddr *next;
3113 
3114 	IF_ADDR_LOCK(ifp);
3115 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3116 		if_delmulti_locked(ifp, ifma, 0);
3117 	IF_ADDR_UNLOCK(ifp);
3118 }
3119 
3120 /*
3121  * Delete a multicast group membership by group membership pointer.
3122  * Network-layer protocol domains must use this routine.
3123  *
3124  * It is safe to call this routine if the ifp disappeared.
3125  */
3126 void
3127 if_delmulti_ifma(struct ifmultiaddr *ifma)
3128 {
3129 	struct ifnet *ifp;
3130 	int lastref;
3131 
3132 	ifp = ifma->ifma_ifp;
3133 #ifdef DIAGNOSTIC
3134 	if (ifp == NULL) {
3135 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3136 	} else {
3137 		struct ifnet *oifp;
3138 
3139 		IFNET_RLOCK_NOSLEEP();
3140 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3141 			if (ifp == oifp)
3142 				break;
3143 		if (ifp != oifp) {
3144 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3145 			ifp = NULL;
3146 		}
3147 		IFNET_RUNLOCK_NOSLEEP();
3148 	}
3149 #endif
3150 	/*
3151 	 * If and only if the ifnet instance exists: Acquire the address lock.
3152 	 */
3153 	if (ifp != NULL)
3154 		IF_ADDR_LOCK(ifp);
3155 
3156 	lastref = if_delmulti_locked(ifp, ifma, 0);
3157 
3158 	if (ifp != NULL) {
3159 		/*
3160 		 * If and only if the ifnet instance exists:
3161 		 *  Release the address lock.
3162 		 *  If the group was left: update the hardware hash filter.
3163 		 */
3164 		IF_ADDR_UNLOCK(ifp);
3165 		if (lastref && ifp->if_ioctl != NULL) {
3166 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3167 		}
3168 	}
3169 }
3170 
3171 /*
3172  * Perform deletion of network-layer and/or link-layer multicast address.
3173  *
3174  * Return 0 if the reference count was decremented.
3175  * Return 1 if the final reference was released, indicating that the
3176  * hardware hash filter should be reprogrammed.
3177  */
3178 static int
3179 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3180 {
3181 	struct ifmultiaddr *ll_ifma;
3182 
3183 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3184 		KASSERT(ifma->ifma_ifp == ifp,
3185 		    ("%s: inconsistent ifp %p", __func__, ifp));
3186 		IF_ADDR_LOCK_ASSERT(ifp);
3187 	}
3188 
3189 	ifp = ifma->ifma_ifp;
3190 
3191 	/*
3192 	 * If the ifnet is detaching, null out references to ifnet,
3193 	 * so that upper protocol layers will notice, and not attempt
3194 	 * to obtain locks for an ifnet which no longer exists. The
3195 	 * routing socket announcement must happen before the ifnet
3196 	 * instance is detached from the system.
3197 	 */
3198 	if (detaching) {
3199 #ifdef DIAGNOSTIC
3200 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3201 #endif
3202 		/*
3203 		 * ifp may already be nulled out if we are being reentered
3204 		 * to delete the ll_ifma.
3205 		 */
3206 		if (ifp != NULL) {
3207 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3208 			ifma->ifma_ifp = NULL;
3209 		}
3210 	}
3211 
3212 	if (--ifma->ifma_refcount > 0)
3213 		return 0;
3214 
3215 	/*
3216 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3217 	 * have been associated with it. Release it first if so.
3218 	 */
3219 	ll_ifma = ifma->ifma_llifma;
3220 	if (ll_ifma != NULL) {
3221 		KASSERT(ifma->ifma_lladdr != NULL,
3222 		    ("%s: llifma w/o lladdr", __func__));
3223 		if (detaching)
3224 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3225 		if (--ll_ifma->ifma_refcount == 0) {
3226 			if (ifp != NULL) {
3227 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3228 				    ifma_link);
3229 			}
3230 			if_freemulti(ll_ifma);
3231 		}
3232 	}
3233 
3234 	if (ifp != NULL)
3235 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3236 
3237 	if_freemulti(ifma);
3238 
3239 	/*
3240 	 * The last reference to this instance of struct ifmultiaddr
3241 	 * was released; the hardware should be notified of this change.
3242 	 */
3243 	return 1;
3244 }
3245 
3246 /*
3247  * Set the link layer address on an interface.
3248  *
3249  * At this time we only support certain types of interfaces,
3250  * and we don't allow the length of the address to change.
3251  */
3252 int
3253 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3254 {
3255 	struct sockaddr_dl *sdl;
3256 	struct ifaddr *ifa;
3257 	struct ifreq ifr;
3258 
3259 	IF_ADDR_LOCK(ifp);
3260 	ifa = ifp->if_addr;
3261 	if (ifa == NULL) {
3262 		IF_ADDR_UNLOCK(ifp);
3263 		return (EINVAL);
3264 	}
3265 	ifa_ref(ifa);
3266 	IF_ADDR_UNLOCK(ifp);
3267 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3268 	if (sdl == NULL) {
3269 		ifa_free(ifa);
3270 		return (EINVAL);
3271 	}
3272 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3273 		ifa_free(ifa);
3274 		return (EINVAL);
3275 	}
3276 	switch (ifp->if_type) {
3277 	case IFT_ETHER:
3278 	case IFT_FDDI:
3279 	case IFT_XETHER:
3280 	case IFT_ISO88025:
3281 	case IFT_L2VLAN:
3282 	case IFT_BRIDGE:
3283 	case IFT_ARCNET:
3284 	case IFT_IEEE8023ADLAG:
3285 	case IFT_IEEE80211:
3286 		bcopy(lladdr, LLADDR(sdl), len);
3287 		ifa_free(ifa);
3288 		break;
3289 	default:
3290 		ifa_free(ifa);
3291 		return (ENODEV);
3292 	}
3293 
3294 	/*
3295 	 * If the interface is already up, we need
3296 	 * to re-init it in order to reprogram its
3297 	 * address filter.
3298 	 */
3299 	if ((ifp->if_flags & IFF_UP) != 0) {
3300 		if (ifp->if_ioctl) {
3301 			ifp->if_flags &= ~IFF_UP;
3302 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3303 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3304 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3305 			ifp->if_flags |= IFF_UP;
3306 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3307 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3308 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3309 		}
3310 #ifdef INET
3311 		/*
3312 		 * Also send gratuitous ARPs to notify other nodes about
3313 		 * the address change.
3314 		 */
3315 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3316 			if (ifa->ifa_addr->sa_family == AF_INET)
3317 				arp_ifinit(ifp, ifa);
3318 		}
3319 #endif
3320 	}
3321 	return (0);
3322 }
3323 
3324 /*
3325  * The name argument must be a pointer to storage which will last as
3326  * long as the interface does.  For physical devices, the result of
3327  * device_get_name(dev) is a good choice and for pseudo-devices a
3328  * static string works well.
3329  */
3330 void
3331 if_initname(struct ifnet *ifp, const char *name, int unit)
3332 {
3333 	ifp->if_dname = name;
3334 	ifp->if_dunit = unit;
3335 	if (unit != IF_DUNIT_NONE)
3336 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3337 	else
3338 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3339 }
3340 
3341 int
3342 if_printf(struct ifnet *ifp, const char * fmt, ...)
3343 {
3344 	va_list ap;
3345 	int retval;
3346 
3347 	retval = printf("%s: ", ifp->if_xname);
3348 	va_start(ap, fmt);
3349 	retval += vprintf(fmt, ap);
3350 	va_end(ap);
3351 	return (retval);
3352 }
3353 
3354 void
3355 if_start(struct ifnet *ifp)
3356 {
3357 
3358 	(*(ifp)->if_start)(ifp);
3359 }
3360 
3361 /*
3362  * Backwards compatibility interface for drivers
3363  * that have not implemented it
3364  */
3365 static int
3366 if_transmit(struct ifnet *ifp, struct mbuf *m)
3367 {
3368 	int error;
3369 
3370 	IFQ_HANDOFF(ifp, m, error);
3371 	return (error);
3372 }
3373 
3374 int
3375 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3376 {
3377 	int active = 0;
3378 
3379 	IF_LOCK(ifq);
3380 	if (_IF_QFULL(ifq)) {
3381 		_IF_DROP(ifq);
3382 		IF_UNLOCK(ifq);
3383 		m_freem(m);
3384 		return (0);
3385 	}
3386 	if (ifp != NULL) {
3387 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3388 		if (m->m_flags & (M_BCAST|M_MCAST))
3389 			ifp->if_omcasts++;
3390 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3391 	}
3392 	_IF_ENQUEUE(ifq, m);
3393 	IF_UNLOCK(ifq);
3394 	if (ifp != NULL && !active)
3395 		(*(ifp)->if_start)(ifp);
3396 	return (1);
3397 }
3398 
3399 void
3400 if_register_com_alloc(u_char type,
3401     if_com_alloc_t *a, if_com_free_t *f)
3402 {
3403 
3404 	KASSERT(if_com_alloc[type] == NULL,
3405 	    ("if_register_com_alloc: %d already registered", type));
3406 	KASSERT(if_com_free[type] == NULL,
3407 	    ("if_register_com_alloc: %d free already registered", type));
3408 
3409 	if_com_alloc[type] = a;
3410 	if_com_free[type] = f;
3411 }
3412 
3413 void
3414 if_deregister_com_alloc(u_char type)
3415 {
3416 
3417 	KASSERT(if_com_alloc[type] != NULL,
3418 	    ("if_deregister_com_alloc: %d not registered", type));
3419 	KASSERT(if_com_free[type] != NULL,
3420 	    ("if_deregister_com_alloc: %d free not registered", type));
3421 	if_com_alloc[type] = NULL;
3422 	if_com_free[type] = NULL;
3423 }
3424