xref: /freebsd/sys/net/if.c (revision 10b9d77bf1ccf2f3affafa6261692cb92cf7e992)
1 /*-
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.5 (Berkeley) 1/9/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/malloc.h>
41 #include <sys/sbuf.h>
42 #include <sys/bus.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/refcount.h>
53 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
60 #include <sys/jail.h>
61 #include <machine/stdarg.h>
62 #include <vm/uma.h>
63 
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_clone.h>
67 #include <net/if_dl.h>
68 #include <net/if_types.h>
69 #include <net/if_var.h>
70 #include <net/radix.h>
71 #include <net/route.h>
72 #include <net/vnet.h>
73 
74 #if defined(INET) || defined(INET6)
75 /*XXX*/
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
78 #include <netinet/ip_carp.h>
79 #ifdef INET6
80 #include <netinet6/in6_var.h>
81 #include <netinet6/in6_ifattach.h>
82 #endif
83 #endif
84 #ifdef INET
85 #include <netinet/if_ether.h>
86 #endif
87 
88 #include <security/mac/mac_framework.h>
89 
90 #ifdef COMPAT_FREEBSD32
91 #include <sys/mount.h>
92 #include <compat/freebsd32/freebsd32.h>
93 #endif
94 
95 struct ifindex_entry {
96 	struct  ifnet *ife_ifnet;
97 };
98 
99 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
100 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
101 
102 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
103 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
104     &ifqmaxlen, 0, "max send queue size");
105 
106 /* Log link state change events */
107 static int log_link_state_change = 1;
108 
109 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
110 	&log_link_state_change, 0,
111 	"log interface link state change events");
112 
113 /* Interface description */
114 static unsigned int ifdescr_maxlen = 1024;
115 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
116 	&ifdescr_maxlen, 0,
117 	"administrative maximum length for interface description");
118 
119 MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
120 
121 /* global sx for non-critical path ifdescr */
122 static struct sx ifdescr_sx;
123 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
124 
125 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
126 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
127 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
128 /* These are external hooks for CARP. */
129 void	(*carp_linkstate_p)(struct ifnet *ifp);
130 #if defined(INET) || defined(INET6)
131 struct ifnet *(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
132 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
133     struct sockaddr *sa, struct rtentry *rt);
134 #endif
135 #ifdef INET
136 int (*carp_iamatch_p)(struct ifnet *, struct in_ifaddr *, struct in_addr *,
137     u_int8_t **);
138 #endif
139 #ifdef INET6
140 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
141 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
142     const struct in6_addr *taddr);
143 #endif
144 
145 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
146 
147 /*
148  * XXX: Style; these should be sorted alphabetically, and unprototyped
149  * static functions should be prototyped. Currently they are sorted by
150  * declaration order.
151  */
152 static void	if_attachdomain(void *);
153 static void	if_attachdomain1(struct ifnet *);
154 static int	ifconf(u_long, caddr_t);
155 static void	if_freemulti(struct ifmultiaddr *);
156 static void	if_init(void *);
157 static void	if_grow(void);
158 static void	if_route(struct ifnet *, int flag, int fam);
159 static int	if_setflag(struct ifnet *, int, int, int *, int);
160 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
161 static void	if_unroute(struct ifnet *, int flag, int fam);
162 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
163 static int	if_rtdel(struct radix_node *, void *);
164 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
165 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
166 static void	do_link_state_change(void *, int);
167 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
168 static int	if_getgroupmembers(struct ifgroupreq *);
169 static void	if_delgroups(struct ifnet *);
170 static void	if_attach_internal(struct ifnet *, int);
171 static void	if_detach_internal(struct ifnet *, int);
172 
173 #ifdef INET6
174 /*
175  * XXX: declare here to avoid to include many inet6 related files..
176  * should be more generalized?
177  */
178 extern void	nd6_setmtu(struct ifnet *);
179 #endif
180 
181 VNET_DEFINE(int, if_index);
182 int	ifqmaxlen = IFQ_MAXLEN;
183 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
184 VNET_DEFINE(struct ifgrouphead, ifg_head);
185 
186 static VNET_DEFINE(int, if_indexlim) = 8;
187 
188 /* Table of ifnet by index. */
189 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
190 
191 #define	V_if_indexlim		VNET(if_indexlim)
192 #define	V_ifindex_table		VNET(ifindex_table)
193 
194 /*
195  * The global network interface list (V_ifnet) and related state (such as
196  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
197  * an rwlock.  Either may be acquired shared to stablize the list, but both
198  * must be acquired writable to modify the list.  This model allows us to
199  * both stablize the interface list during interrupt thread processing, but
200  * also to stablize it over long-running ioctls, without introducing priority
201  * inversions and deadlocks.
202  */
203 struct rwlock ifnet_rwlock;
204 struct sx ifnet_sxlock;
205 
206 /*
207  * The allocation of network interfaces is a rather non-atomic affair; we
208  * need to select an index before we are ready to expose the interface for
209  * use, so will use this pointer value to indicate reservation.
210  */
211 #define	IFNET_HOLD	(void *)(uintptr_t)(-1)
212 
213 static	if_com_alloc_t *if_com_alloc[256];
214 static	if_com_free_t *if_com_free[256];
215 
216 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
217 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
218 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
219 
220 struct ifnet *
221 ifnet_byindex_locked(u_short idx)
222 {
223 
224 	if (idx > V_if_index)
225 		return (NULL);
226 	if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
227 		return (NULL);
228 	return (V_ifindex_table[idx].ife_ifnet);
229 }
230 
231 struct ifnet *
232 ifnet_byindex(u_short idx)
233 {
234 	struct ifnet *ifp;
235 
236 	IFNET_RLOCK_NOSLEEP();
237 	ifp = ifnet_byindex_locked(idx);
238 	IFNET_RUNLOCK_NOSLEEP();
239 	return (ifp);
240 }
241 
242 struct ifnet *
243 ifnet_byindex_ref(u_short idx)
244 {
245 	struct ifnet *ifp;
246 
247 	IFNET_RLOCK_NOSLEEP();
248 	ifp = ifnet_byindex_locked(idx);
249 	if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
250 		IFNET_RUNLOCK_NOSLEEP();
251 		return (NULL);
252 	}
253 	if_ref(ifp);
254 	IFNET_RUNLOCK_NOSLEEP();
255 	return (ifp);
256 }
257 
258 /*
259  * Allocate an ifindex array entry; return 0 on success or an error on
260  * failure.
261  */
262 static int
263 ifindex_alloc_locked(u_short *idxp)
264 {
265 	u_short idx;
266 
267 	IFNET_WLOCK_ASSERT();
268 
269 retry:
270 	/*
271 	 * Try to find an empty slot below V_if_index.  If we fail, take the
272 	 * next slot.
273 	 */
274 	for (idx = 1; idx <= V_if_index; idx++) {
275 		if (V_ifindex_table[idx].ife_ifnet == NULL)
276 			break;
277 	}
278 
279 	/* Catch if_index overflow. */
280 	if (idx < 1)
281 		return (ENOSPC);
282 	if (idx >= V_if_indexlim) {
283 		if_grow();
284 		goto retry;
285 	}
286 	if (idx > V_if_index)
287 		V_if_index = idx;
288 	*idxp = idx;
289 	return (0);
290 }
291 
292 static void
293 ifindex_free_locked(u_short idx)
294 {
295 
296 	IFNET_WLOCK_ASSERT();
297 
298 	V_ifindex_table[idx].ife_ifnet = NULL;
299 	while (V_if_index > 0 &&
300 	    V_ifindex_table[V_if_index].ife_ifnet == NULL)
301 		V_if_index--;
302 }
303 
304 static void
305 ifindex_free(u_short idx)
306 {
307 
308 	IFNET_WLOCK();
309 	ifindex_free_locked(idx);
310 	IFNET_WUNLOCK();
311 }
312 
313 static void
314 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
315 {
316 
317 	IFNET_WLOCK_ASSERT();
318 
319 	V_ifindex_table[idx].ife_ifnet = ifp;
320 }
321 
322 static void
323 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
324 {
325 
326 	IFNET_WLOCK();
327 	ifnet_setbyindex_locked(idx, ifp);
328 	IFNET_WUNLOCK();
329 }
330 
331 struct ifaddr *
332 ifaddr_byindex(u_short idx)
333 {
334 	struct ifaddr *ifa;
335 
336 	IFNET_RLOCK_NOSLEEP();
337 	ifa = ifnet_byindex_locked(idx)->if_addr;
338 	if (ifa != NULL)
339 		ifa_ref(ifa);
340 	IFNET_RUNLOCK_NOSLEEP();
341 	return (ifa);
342 }
343 
344 /*
345  * Network interface utility routines.
346  *
347  * Routines with ifa_ifwith* names take sockaddr *'s as
348  * parameters.
349  */
350 
351 static void
352 vnet_if_init(const void *unused __unused)
353 {
354 
355 	TAILQ_INIT(&V_ifnet);
356 	TAILQ_INIT(&V_ifg_head);
357 	IFNET_WLOCK();
358 	if_grow();				/* create initial table */
359 	IFNET_WUNLOCK();
360 	vnet_if_clone_init();
361 }
362 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
363     NULL);
364 
365 /* ARGSUSED*/
366 static void
367 if_init(void *dummy __unused)
368 {
369 
370 	IFNET_LOCK_INIT();
371 	if_clone_init();
372 }
373 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
374 
375 
376 #ifdef VIMAGE
377 static void
378 vnet_if_uninit(const void *unused __unused)
379 {
380 
381 	VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
382 	    "not empty", __func__, __LINE__, &V_ifnet));
383 	VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
384 	    "not empty", __func__, __LINE__, &V_ifg_head));
385 
386 	free((caddr_t)V_ifindex_table, M_IFNET);
387 }
388 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
389     vnet_if_uninit, NULL);
390 #endif
391 
392 static void
393 if_grow(void)
394 {
395 	int oldlim;
396 	u_int n;
397 	struct ifindex_entry *e;
398 
399 	IFNET_WLOCK_ASSERT();
400 	oldlim = V_if_indexlim;
401 	IFNET_WUNLOCK();
402 	n = (oldlim << 1) * sizeof(*e);
403 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
404 	IFNET_WLOCK();
405 	if (V_if_indexlim != oldlim) {
406 		free(e, M_IFNET);
407 		return;
408 	}
409 	if (V_ifindex_table != NULL) {
410 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
411 		free((caddr_t)V_ifindex_table, M_IFNET);
412 	}
413 	V_if_indexlim <<= 1;
414 	V_ifindex_table = e;
415 }
416 
417 /*
418  * Allocate a struct ifnet and an index for an interface.  A layer 2
419  * common structure will also be allocated if an allocation routine is
420  * registered for the passed type.
421  */
422 struct ifnet *
423 if_alloc(u_char type)
424 {
425 	struct ifnet *ifp;
426 	u_short idx;
427 
428 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
429 	IFNET_WLOCK();
430 	if (ifindex_alloc_locked(&idx) != 0) {
431 		IFNET_WUNLOCK();
432 		free(ifp, M_IFNET);
433 		return (NULL);
434 	}
435 	ifnet_setbyindex_locked(idx, IFNET_HOLD);
436 	IFNET_WUNLOCK();
437 	ifp->if_index = idx;
438 	ifp->if_type = type;
439 	ifp->if_alloctype = type;
440 	if (if_com_alloc[type] != NULL) {
441 		ifp->if_l2com = if_com_alloc[type](type, ifp);
442 		if (ifp->if_l2com == NULL) {
443 			free(ifp, M_IFNET);
444 			ifindex_free(idx);
445 			return (NULL);
446 		}
447 	}
448 
449 	IF_ADDR_LOCK_INIT(ifp);
450 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
451 	ifp->if_afdata_initialized = 0;
452 	IF_AFDATA_LOCK_INIT(ifp);
453 	TAILQ_INIT(&ifp->if_addrhead);
454 	TAILQ_INIT(&ifp->if_prefixhead);
455 	TAILQ_INIT(&ifp->if_multiaddrs);
456 	TAILQ_INIT(&ifp->if_groups);
457 #ifdef MAC
458 	mac_ifnet_init(ifp);
459 #endif
460 	ifq_init(&ifp->if_snd, ifp);
461 
462 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
463 	ifnet_setbyindex(ifp->if_index, ifp);
464 	return (ifp);
465 }
466 
467 /*
468  * Do the actual work of freeing a struct ifnet, associated index, and layer
469  * 2 common structure.  This call is made when the last reference to an
470  * interface is released.
471  */
472 static void
473 if_free_internal(struct ifnet *ifp)
474 {
475 
476 	KASSERT((ifp->if_flags & IFF_DYING),
477 	    ("if_free_internal: interface not dying"));
478 
479 	IFNET_WLOCK();
480 	KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
481 	    ("%s: freeing unallocated ifnet", ifp->if_xname));
482 
483 	ifindex_free_locked(ifp->if_index);
484 	IFNET_WUNLOCK();
485 
486 	if (if_com_free[ifp->if_alloctype] != NULL)
487 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
488 		    ifp->if_alloctype);
489 
490 #ifdef MAC
491 	mac_ifnet_destroy(ifp);
492 #endif /* MAC */
493 	if (ifp->if_description != NULL)
494 		free(ifp->if_description, M_IFDESCR);
495 	IF_AFDATA_DESTROY(ifp);
496 	IF_ADDR_LOCK_DESTROY(ifp);
497 	ifq_delete(&ifp->if_snd);
498 	free(ifp, M_IFNET);
499 }
500 
501 /*
502  * This version should only be called by intefaces that switch their type
503  * after calling if_alloc().  if_free_type() will go away again now that we
504  * have if_alloctype to cache the original allocation type.  For now, assert
505  * that they match, since we require that in practice.
506  */
507 void
508 if_free_type(struct ifnet *ifp, u_char type)
509 {
510 
511 	KASSERT(ifp->if_alloctype == type,
512 	    ("if_free_type: type (%d) != alloctype (%d)", type,
513 	    ifp->if_alloctype));
514 
515 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
516 	if (!refcount_release(&ifp->if_refcount))
517 		return;
518 	if_free_internal(ifp);
519 }
520 
521 /*
522  * This is the normal version of if_free(), used by device drivers to free a
523  * detached network interface.  The contents of if_free_type() will move into
524  * here when if_free_type() goes away.
525  */
526 void
527 if_free(struct ifnet *ifp)
528 {
529 
530 	if_free_type(ifp, ifp->if_alloctype);
531 }
532 
533 /*
534  * Interfaces to keep an ifnet type-stable despite the possibility of the
535  * driver calling if_free().  If there are additional references, we defer
536  * freeing the underlying data structure.
537  */
538 void
539 if_ref(struct ifnet *ifp)
540 {
541 
542 	/* We don't assert the ifnet list lock here, but arguably should. */
543 	refcount_acquire(&ifp->if_refcount);
544 }
545 
546 void
547 if_rele(struct ifnet *ifp)
548 {
549 
550 	if (!refcount_release(&ifp->if_refcount))
551 		return;
552 	if_free_internal(ifp);
553 }
554 
555 void
556 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
557 {
558 
559 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
560 
561 	if (ifq->ifq_maxlen == 0)
562 		ifq->ifq_maxlen = ifqmaxlen;
563 
564 	ifq->altq_type = 0;
565 	ifq->altq_disc = NULL;
566 	ifq->altq_flags &= ALTQF_CANTCHANGE;
567 	ifq->altq_tbr  = NULL;
568 	ifq->altq_ifp  = ifp;
569 }
570 
571 void
572 ifq_delete(struct ifaltq *ifq)
573 {
574 	mtx_destroy(&ifq->ifq_mtx);
575 }
576 
577 /*
578  * Perform generic interface initalization tasks and attach the interface
579  * to the list of "active" interfaces.  If vmove flag is set on entry
580  * to if_attach_internal(), perform only a limited subset of initialization
581  * tasks, given that we are moving from one vnet to another an ifnet which
582  * has already been fully initialized.
583  *
584  * XXX:
585  *  - The decision to return void and thus require this function to
586  *    succeed is questionable.
587  *  - We should probably do more sanity checking.  For instance we don't
588  *    do anything to insure if_xname is unique or non-empty.
589  */
590 void
591 if_attach(struct ifnet *ifp)
592 {
593 
594 	if_attach_internal(ifp, 0);
595 }
596 
597 static void
598 if_attach_internal(struct ifnet *ifp, int vmove)
599 {
600 	unsigned socksize, ifasize;
601 	int namelen, masklen;
602 	struct sockaddr_dl *sdl;
603 	struct ifaddr *ifa;
604 
605 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
606 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
607 		    ifp->if_xname);
608 
609 #ifdef VIMAGE
610 	ifp->if_vnet = curvnet;
611 	if (ifp->if_home_vnet == NULL)
612 		ifp->if_home_vnet = curvnet;
613 #endif
614 
615 	if_addgroup(ifp, IFG_ALL);
616 
617 	getmicrotime(&ifp->if_lastchange);
618 	ifp->if_data.ifi_epoch = time_uptime;
619 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
620 
621 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
622 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
623 	    ("transmit and qflush must both either be set or both be NULL"));
624 	if (ifp->if_transmit == NULL) {
625 		ifp->if_transmit = if_transmit;
626 		ifp->if_qflush = if_qflush;
627 	}
628 
629 	if (!vmove) {
630 #ifdef MAC
631 		mac_ifnet_create(ifp);
632 #endif
633 
634 		/*
635 		 * Create a Link Level name for this device.
636 		 */
637 		namelen = strlen(ifp->if_xname);
638 		/*
639 		 * Always save enough space for any possiable name so we
640 		 * can do a rename in place later.
641 		 */
642 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
643 		socksize = masklen + ifp->if_addrlen;
644 		if (socksize < sizeof(*sdl))
645 			socksize = sizeof(*sdl);
646 		socksize = roundup2(socksize, sizeof(long));
647 		ifasize = sizeof(*ifa) + 2 * socksize;
648 		ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
649 		ifa_init(ifa);
650 		sdl = (struct sockaddr_dl *)(ifa + 1);
651 		sdl->sdl_len = socksize;
652 		sdl->sdl_family = AF_LINK;
653 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
654 		sdl->sdl_nlen = namelen;
655 		sdl->sdl_index = ifp->if_index;
656 		sdl->sdl_type = ifp->if_type;
657 		ifp->if_addr = ifa;
658 		ifa->ifa_ifp = ifp;
659 		ifa->ifa_rtrequest = link_rtrequest;
660 		ifa->ifa_addr = (struct sockaddr *)sdl;
661 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
662 		ifa->ifa_netmask = (struct sockaddr *)sdl;
663 		sdl->sdl_len = masklen;
664 		while (namelen != 0)
665 			sdl->sdl_data[--namelen] = 0xff;
666 		TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
667 		/* Reliably crash if used uninitialized. */
668 		ifp->if_broadcastaddr = NULL;
669 	}
670 #ifdef VIMAGE
671 	else {
672 		/*
673 		 * Update the interface index in the link layer address
674 		 * of the interface.
675 		 */
676 		for (ifa = ifp->if_addr; ifa != NULL;
677 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
678 			if (ifa->ifa_addr->sa_family == AF_LINK) {
679 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
680 				sdl->sdl_index = ifp->if_index;
681 			}
682 		}
683 	}
684 #endif
685 
686 	IFNET_WLOCK();
687 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
688 #ifdef VIMAGE
689 	curvnet->vnet_ifcnt++;
690 #endif
691 	IFNET_WUNLOCK();
692 
693 	if (domain_init_status >= 2)
694 		if_attachdomain1(ifp);
695 
696 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
697 	if (IS_DEFAULT_VNET(curvnet))
698 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
699 
700 	/* Announce the interface. */
701 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
702 }
703 
704 static void
705 if_attachdomain(void *dummy)
706 {
707 	struct ifnet *ifp;
708 	int s;
709 
710 	s = splnet();
711 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
712 		if_attachdomain1(ifp);
713 	splx(s);
714 }
715 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
716     if_attachdomain, NULL);
717 
718 static void
719 if_attachdomain1(struct ifnet *ifp)
720 {
721 	struct domain *dp;
722 	int s;
723 
724 	s = splnet();
725 
726 	/*
727 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
728 	 * cannot lock ifp->if_afdata initialization, entirely.
729 	 */
730 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
731 		splx(s);
732 		return;
733 	}
734 	if (ifp->if_afdata_initialized >= domain_init_status) {
735 		IF_AFDATA_UNLOCK(ifp);
736 		splx(s);
737 		printf("if_attachdomain called more than once on %s\n",
738 		    ifp->if_xname);
739 		return;
740 	}
741 	ifp->if_afdata_initialized = domain_init_status;
742 	IF_AFDATA_UNLOCK(ifp);
743 
744 	/* address family dependent data region */
745 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
746 	for (dp = domains; dp; dp = dp->dom_next) {
747 		if (dp->dom_ifattach)
748 			ifp->if_afdata[dp->dom_family] =
749 			    (*dp->dom_ifattach)(ifp);
750 	}
751 
752 	splx(s);
753 }
754 
755 /*
756  * Remove any unicast or broadcast network addresses from an interface.
757  */
758 void
759 if_purgeaddrs(struct ifnet *ifp)
760 {
761 	struct ifaddr *ifa, *next;
762 
763 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
764 		if (ifa->ifa_addr->sa_family == AF_LINK)
765 			continue;
766 #ifdef INET
767 		/* XXX: Ugly!! ad hoc just for INET */
768 		if (ifa->ifa_addr->sa_family == AF_INET) {
769 			struct ifaliasreq ifr;
770 
771 			bzero(&ifr, sizeof(ifr));
772 			ifr.ifra_addr = *ifa->ifa_addr;
773 			if (ifa->ifa_dstaddr)
774 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
775 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
776 			    NULL) == 0)
777 				continue;
778 		}
779 #endif /* INET */
780 #ifdef INET6
781 		if (ifa->ifa_addr->sa_family == AF_INET6) {
782 			in6_purgeaddr(ifa);
783 			/* ifp_addrhead is already updated */
784 			continue;
785 		}
786 #endif /* INET6 */
787 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
788 		ifa_free(ifa);
789 	}
790 }
791 
792 /*
793  * Remove any multicast network addresses from an interface when an ifnet
794  * is going away.
795  */
796 static void
797 if_purgemaddrs(struct ifnet *ifp)
798 {
799 	struct ifmultiaddr *ifma;
800 	struct ifmultiaddr *next;
801 
802 	IF_ADDR_LOCK(ifp);
803 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
804 		if_delmulti_locked(ifp, ifma, 1);
805 	IF_ADDR_UNLOCK(ifp);
806 }
807 
808 /*
809  * Detach an interface, removing it from the list of "active" interfaces.
810  * If vmove flag is set on entry to if_detach_internal(), perform only a
811  * limited subset of cleanup tasks, given that we are moving an ifnet from
812  * one vnet to another, where it must be fully operational.
813  *
814  * XXXRW: There are some significant questions about event ordering, and
815  * how to prevent things from starting to use the interface during detach.
816  */
817 void
818 if_detach(struct ifnet *ifp)
819 {
820 
821 	if_detach_internal(ifp, 0);
822 }
823 
824 static void
825 if_detach_internal(struct ifnet *ifp, int vmove)
826 {
827 	struct ifaddr *ifa;
828 	struct radix_node_head	*rnh;
829 	int i, j;
830 	struct domain *dp;
831  	struct ifnet *iter;
832  	int found = 0;
833 
834 	IFNET_WLOCK();
835 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
836 		if (iter == ifp) {
837 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
838 			found = 1;
839 			break;
840 		}
841 #ifdef VIMAGE
842 	if (found)
843 		curvnet->vnet_ifcnt--;
844 #endif
845 	IFNET_WUNLOCK();
846 	if (!found) {
847 		if (vmove)
848 			panic("%s: ifp=%p not on the ifnet tailq %p",
849 			    __func__, ifp, &V_ifnet);
850 		else
851 			return; /* XXX this should panic as well? */
852 	}
853 
854 	/*
855 	 * Remove/wait for pending events.
856 	 */
857 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
858 
859 	/*
860 	 * Remove routes and flush queues.
861 	 */
862 	if_down(ifp);
863 #ifdef ALTQ
864 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
865 		altq_disable(&ifp->if_snd);
866 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
867 		altq_detach(&ifp->if_snd);
868 #endif
869 
870 	if_purgeaddrs(ifp);
871 
872 #ifdef INET
873 	in_ifdetach(ifp);
874 #endif
875 
876 #ifdef INET6
877 	/*
878 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
879 	 * before removing routing entries below, since IPv6 interface direct
880 	 * routes are expected to be removed by the IPv6-specific kernel API.
881 	 * Otherwise, the kernel will detect some inconsistency and bark it.
882 	 */
883 	in6_ifdetach(ifp);
884 #endif
885 	if_purgemaddrs(ifp);
886 
887 	if (!vmove) {
888 		/*
889 		 * Prevent further calls into the device driver via ifnet.
890 		 */
891 		if_dead(ifp);
892 
893 		/*
894 		 * Remove link ifaddr pointer and maybe decrement if_index.
895 		 * Clean up all addresses.
896 		 */
897 		ifp->if_addr = NULL;
898 
899 		/* We can now free link ifaddr. */
900 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
901 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
902 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
903 			ifa_free(ifa);
904 		}
905 	}
906 
907 	/*
908 	 * Delete all remaining routes using this interface
909 	 * Unfortuneatly the only way to do this is to slog through
910 	 * the entire routing table looking for routes which point
911 	 * to this interface...oh well...
912 	 */
913 	for (i = 1; i <= AF_MAX; i++) {
914 		for (j = 0; j < rt_numfibs; j++) {
915 			rnh = rt_tables_get_rnh(j, i);
916 			if (rnh == NULL)
917 				continue;
918 			RADIX_NODE_HEAD_LOCK(rnh);
919 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
920 			RADIX_NODE_HEAD_UNLOCK(rnh);
921 		}
922 	}
923 
924 	/* Announce that the interface is gone. */
925 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
926 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
927 	if (IS_DEFAULT_VNET(curvnet))
928 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
929 	if_delgroups(ifp);
930 
931 	/*
932 	 * We cannot hold the lock over dom_ifdetach calls as they might
933 	 * sleep, for example trying to drain a callout, thus open up the
934 	 * theoretical race with re-attaching.
935 	 */
936 	IF_AFDATA_LOCK(ifp);
937 	i = ifp->if_afdata_initialized;
938 	ifp->if_afdata_initialized = 0;
939 	IF_AFDATA_UNLOCK(ifp);
940 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
941 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
942 			(*dp->dom_ifdetach)(ifp,
943 			    ifp->if_afdata[dp->dom_family]);
944 	}
945 }
946 
947 #ifdef VIMAGE
948 /*
949  * if_vmove() performs a limited version of if_detach() in current
950  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
951  * An attempt is made to shrink if_index in current vnet, find an
952  * unused if_index in target vnet and calls if_grow() if necessary,
953  * and finally find an unused if_xname for the target vnet.
954  */
955 void
956 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
957 {
958 	u_short idx;
959 
960 	/*
961 	 * Detach from current vnet, but preserve LLADDR info, do not
962 	 * mark as dead etc. so that the ifnet can be reattached later.
963 	 */
964 	if_detach_internal(ifp, 1);
965 
966 	/*
967 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
968 	 * the if_index for that vnet if possible.
969 	 *
970 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
971 	 * or we'd lock on one vnet and unlock on another.
972 	 */
973 	IFNET_WLOCK();
974 	ifindex_free_locked(ifp->if_index);
975 	IFNET_WUNLOCK();
976 
977 	/*
978 	 * Perform interface-specific reassignment tasks, if provided by
979 	 * the driver.
980 	 */
981 	if (ifp->if_reassign != NULL)
982 		ifp->if_reassign(ifp, new_vnet, NULL);
983 
984 	/*
985 	 * Switch to the context of the target vnet.
986 	 */
987 	CURVNET_SET_QUIET(new_vnet);
988 
989 	IFNET_WLOCK();
990 	if (ifindex_alloc_locked(&idx) != 0) {
991 		IFNET_WUNLOCK();
992 		panic("if_index overflow");
993 	}
994 	ifp->if_index = idx;
995 	ifnet_setbyindex_locked(ifp->if_index, ifp);
996 	IFNET_WUNLOCK();
997 
998 	if_attach_internal(ifp, 1);
999 
1000 	CURVNET_RESTORE();
1001 }
1002 
1003 /*
1004  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1005  */
1006 static int
1007 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1008 {
1009 	struct prison *pr;
1010 	struct ifnet *difp;
1011 
1012 	/* Try to find the prison within our visibility. */
1013 	sx_slock(&allprison_lock);
1014 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1015 	sx_sunlock(&allprison_lock);
1016 	if (pr == NULL)
1017 		return (ENXIO);
1018 	prison_hold_locked(pr);
1019 	mtx_unlock(&pr->pr_mtx);
1020 
1021 	/* Do not try to move the iface from and to the same prison. */
1022 	if (pr->pr_vnet == ifp->if_vnet) {
1023 		prison_free(pr);
1024 		return (EEXIST);
1025 	}
1026 
1027 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1028 	/* XXX Lock interfaces to avoid races. */
1029 	CURVNET_SET_QUIET(pr->pr_vnet);
1030 	difp = ifunit(ifname);
1031 	CURVNET_RESTORE();
1032 	if (difp != NULL) {
1033 		prison_free(pr);
1034 		return (EEXIST);
1035 	}
1036 
1037 	/* Move the interface into the child jail/vnet. */
1038 	if_vmove(ifp, pr->pr_vnet);
1039 
1040 	/* Report the new if_xname back to the userland. */
1041 	sprintf(ifname, "%s", ifp->if_xname);
1042 
1043 	prison_free(pr);
1044 	return (0);
1045 }
1046 
1047 static int
1048 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1049 {
1050 	struct prison *pr;
1051 	struct vnet *vnet_dst;
1052 	struct ifnet *ifp;
1053 
1054 	/* Try to find the prison within our visibility. */
1055 	sx_slock(&allprison_lock);
1056 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1057 	sx_sunlock(&allprison_lock);
1058 	if (pr == NULL)
1059 		return (ENXIO);
1060 	prison_hold_locked(pr);
1061 	mtx_unlock(&pr->pr_mtx);
1062 
1063 	/* Make sure the named iface exists in the source prison/vnet. */
1064 	CURVNET_SET(pr->pr_vnet);
1065 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1066 	if (ifp == NULL) {
1067 		CURVNET_RESTORE();
1068 		prison_free(pr);
1069 		return (ENXIO);
1070 	}
1071 
1072 	/* Do not try to move the iface from and to the same prison. */
1073 	vnet_dst = TD_TO_VNET(td);
1074 	if (vnet_dst == ifp->if_vnet) {
1075 		CURVNET_RESTORE();
1076 		prison_free(pr);
1077 		return (EEXIST);
1078 	}
1079 
1080 	/* Get interface back from child jail/vnet. */
1081 	if_vmove(ifp, vnet_dst);
1082 	CURVNET_RESTORE();
1083 
1084 	/* Report the new if_xname back to the userland. */
1085 	sprintf(ifname, "%s", ifp->if_xname);
1086 
1087 	prison_free(pr);
1088 	return (0);
1089 }
1090 #endif /* VIMAGE */
1091 
1092 /*
1093  * Add a group to an interface
1094  */
1095 int
1096 if_addgroup(struct ifnet *ifp, const char *groupname)
1097 {
1098 	struct ifg_list		*ifgl;
1099 	struct ifg_group	*ifg = NULL;
1100 	struct ifg_member	*ifgm;
1101 
1102 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1103 	    groupname[strlen(groupname) - 1] <= '9')
1104 		return (EINVAL);
1105 
1106 	IFNET_WLOCK();
1107 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1108 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1109 			IFNET_WUNLOCK();
1110 			return (EEXIST);
1111 		}
1112 
1113 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1114 	    M_NOWAIT)) == NULL) {
1115 	    	IFNET_WUNLOCK();
1116 		return (ENOMEM);
1117 	}
1118 
1119 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1120 	    M_TEMP, M_NOWAIT)) == NULL) {
1121 		free(ifgl, M_TEMP);
1122 		IFNET_WUNLOCK();
1123 		return (ENOMEM);
1124 	}
1125 
1126 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1127 		if (!strcmp(ifg->ifg_group, groupname))
1128 			break;
1129 
1130 	if (ifg == NULL) {
1131 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1132 		    M_TEMP, M_NOWAIT)) == NULL) {
1133 			free(ifgl, M_TEMP);
1134 			free(ifgm, M_TEMP);
1135 			IFNET_WUNLOCK();
1136 			return (ENOMEM);
1137 		}
1138 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1139 		ifg->ifg_refcnt = 0;
1140 		TAILQ_INIT(&ifg->ifg_members);
1141 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1142 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1143 	}
1144 
1145 	ifg->ifg_refcnt++;
1146 	ifgl->ifgl_group = ifg;
1147 	ifgm->ifgm_ifp = ifp;
1148 
1149 	IF_ADDR_LOCK(ifp);
1150 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1151 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1152 	IF_ADDR_UNLOCK(ifp);
1153 
1154 	IFNET_WUNLOCK();
1155 
1156 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1157 
1158 	return (0);
1159 }
1160 
1161 /*
1162  * Remove a group from an interface
1163  */
1164 int
1165 if_delgroup(struct ifnet *ifp, const char *groupname)
1166 {
1167 	struct ifg_list		*ifgl;
1168 	struct ifg_member	*ifgm;
1169 
1170 	IFNET_WLOCK();
1171 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1172 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1173 			break;
1174 	if (ifgl == NULL) {
1175 		IFNET_WUNLOCK();
1176 		return (ENOENT);
1177 	}
1178 
1179 	IF_ADDR_LOCK(ifp);
1180 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1181 	IF_ADDR_UNLOCK(ifp);
1182 
1183 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1184 		if (ifgm->ifgm_ifp == ifp)
1185 			break;
1186 
1187 	if (ifgm != NULL) {
1188 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1189 		free(ifgm, M_TEMP);
1190 	}
1191 
1192 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1193 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1194 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1195 		free(ifgl->ifgl_group, M_TEMP);
1196 	}
1197 	IFNET_WUNLOCK();
1198 
1199 	free(ifgl, M_TEMP);
1200 
1201 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1202 
1203 	return (0);
1204 }
1205 
1206 /*
1207  * Remove an interface from all groups
1208  */
1209 static void
1210 if_delgroups(struct ifnet *ifp)
1211 {
1212 	struct ifg_list		*ifgl;
1213 	struct ifg_member	*ifgm;
1214 	char groupname[IFNAMSIZ];
1215 
1216 	IFNET_WLOCK();
1217 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1218 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1219 
1220 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1221 
1222 		IF_ADDR_LOCK(ifp);
1223 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1224 		IF_ADDR_UNLOCK(ifp);
1225 
1226 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1227 			if (ifgm->ifgm_ifp == ifp)
1228 				break;
1229 
1230 		if (ifgm != NULL) {
1231 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1232 			    ifgm_next);
1233 			free(ifgm, M_TEMP);
1234 		}
1235 
1236 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1237 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1238 			EVENTHANDLER_INVOKE(group_detach_event,
1239 			    ifgl->ifgl_group);
1240 			free(ifgl->ifgl_group, M_TEMP);
1241 		}
1242 		IFNET_WUNLOCK();
1243 
1244 		free(ifgl, M_TEMP);
1245 
1246 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1247 
1248 		IFNET_WLOCK();
1249 	}
1250 	IFNET_WUNLOCK();
1251 }
1252 
1253 /*
1254  * Stores all groups from an interface in memory pointed
1255  * to by data
1256  */
1257 static int
1258 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1259 {
1260 	int			 len, error;
1261 	struct ifg_list		*ifgl;
1262 	struct ifg_req		 ifgrq, *ifgp;
1263 	struct ifgroupreq	*ifgr = data;
1264 
1265 	if (ifgr->ifgr_len == 0) {
1266 		IF_ADDR_LOCK(ifp);
1267 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1268 			ifgr->ifgr_len += sizeof(struct ifg_req);
1269 		IF_ADDR_UNLOCK(ifp);
1270 		return (0);
1271 	}
1272 
1273 	len = ifgr->ifgr_len;
1274 	ifgp = ifgr->ifgr_groups;
1275 	/* XXX: wire */
1276 	IF_ADDR_LOCK(ifp);
1277 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1278 		if (len < sizeof(ifgrq)) {
1279 			IF_ADDR_UNLOCK(ifp);
1280 			return (EINVAL);
1281 		}
1282 		bzero(&ifgrq, sizeof ifgrq);
1283 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1284 		    sizeof(ifgrq.ifgrq_group));
1285 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1286 		    	IF_ADDR_UNLOCK(ifp);
1287 			return (error);
1288 		}
1289 		len -= sizeof(ifgrq);
1290 		ifgp++;
1291 	}
1292 	IF_ADDR_UNLOCK(ifp);
1293 
1294 	return (0);
1295 }
1296 
1297 /*
1298  * Stores all members of a group in memory pointed to by data
1299  */
1300 static int
1301 if_getgroupmembers(struct ifgroupreq *data)
1302 {
1303 	struct ifgroupreq	*ifgr = data;
1304 	struct ifg_group	*ifg;
1305 	struct ifg_member	*ifgm;
1306 	struct ifg_req		 ifgrq, *ifgp;
1307 	int			 len, error;
1308 
1309 	IFNET_RLOCK();
1310 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1311 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1312 			break;
1313 	if (ifg == NULL) {
1314 		IFNET_RUNLOCK();
1315 		return (ENOENT);
1316 	}
1317 
1318 	if (ifgr->ifgr_len == 0) {
1319 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1320 			ifgr->ifgr_len += sizeof(ifgrq);
1321 		IFNET_RUNLOCK();
1322 		return (0);
1323 	}
1324 
1325 	len = ifgr->ifgr_len;
1326 	ifgp = ifgr->ifgr_groups;
1327 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1328 		if (len < sizeof(ifgrq)) {
1329 			IFNET_RUNLOCK();
1330 			return (EINVAL);
1331 		}
1332 		bzero(&ifgrq, sizeof ifgrq);
1333 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1334 		    sizeof(ifgrq.ifgrq_member));
1335 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1336 			IFNET_RUNLOCK();
1337 			return (error);
1338 		}
1339 		len -= sizeof(ifgrq);
1340 		ifgp++;
1341 	}
1342 	IFNET_RUNLOCK();
1343 
1344 	return (0);
1345 }
1346 
1347 /*
1348  * Delete Routes for a Network Interface
1349  *
1350  * Called for each routing entry via the rnh->rnh_walktree() call above
1351  * to delete all route entries referencing a detaching network interface.
1352  *
1353  * Arguments:
1354  *	rn	pointer to node in the routing table
1355  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1356  *
1357  * Returns:
1358  *	0	successful
1359  *	errno	failed - reason indicated
1360  *
1361  */
1362 static int
1363 if_rtdel(struct radix_node *rn, void *arg)
1364 {
1365 	struct rtentry	*rt = (struct rtentry *)rn;
1366 	struct ifnet	*ifp = arg;
1367 	int		err;
1368 
1369 	if (rt->rt_ifp == ifp) {
1370 
1371 		/*
1372 		 * Protect (sorta) against walktree recursion problems
1373 		 * with cloned routes
1374 		 */
1375 		if ((rt->rt_flags & RTF_UP) == 0)
1376 			return (0);
1377 
1378 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1379 				rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1380 				(struct rtentry **) NULL, rt->rt_fibnum);
1381 		if (err) {
1382 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1383 		}
1384 	}
1385 
1386 	return (0);
1387 }
1388 
1389 /*
1390  * Wrapper functions for struct ifnet address list locking macros.  These are
1391  * used by kernel modules to avoid encoding programming interface or binary
1392  * interface assumptions that may be violated when kernel-internal locking
1393  * approaches change.
1394  */
1395 void
1396 if_addr_rlock(struct ifnet *ifp)
1397 {
1398 
1399 	IF_ADDR_LOCK(ifp);
1400 }
1401 
1402 void
1403 if_addr_runlock(struct ifnet *ifp)
1404 {
1405 
1406 	IF_ADDR_UNLOCK(ifp);
1407 }
1408 
1409 void
1410 if_maddr_rlock(struct ifnet *ifp)
1411 {
1412 
1413 	IF_ADDR_LOCK(ifp);
1414 }
1415 
1416 void
1417 if_maddr_runlock(struct ifnet *ifp)
1418 {
1419 
1420 	IF_ADDR_UNLOCK(ifp);
1421 }
1422 
1423 /*
1424  * Reference count functions for ifaddrs.
1425  */
1426 void
1427 ifa_init(struct ifaddr *ifa)
1428 {
1429 
1430 	mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1431 	refcount_init(&ifa->ifa_refcnt, 1);
1432 }
1433 
1434 void
1435 ifa_ref(struct ifaddr *ifa)
1436 {
1437 
1438 	refcount_acquire(&ifa->ifa_refcnt);
1439 }
1440 
1441 void
1442 ifa_free(struct ifaddr *ifa)
1443 {
1444 
1445 	if (refcount_release(&ifa->ifa_refcnt)) {
1446 		mtx_destroy(&ifa->ifa_mtx);
1447 		free(ifa, M_IFADDR);
1448 	}
1449 }
1450 
1451 int
1452 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1453 {
1454 	int error = 0;
1455 	struct rtentry *rt = NULL;
1456 	struct rt_addrinfo info;
1457 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1458 
1459 	bzero(&info, sizeof(info));
1460 	info.rti_ifp = V_loif;
1461 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1462 	info.rti_info[RTAX_DST] = ia;
1463 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1464 	error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1465 
1466 	if (error == 0 && rt != NULL) {
1467 		RT_LOCK(rt);
1468 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1469 			ifa->ifa_ifp->if_type;
1470 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1471 			ifa->ifa_ifp->if_index;
1472 		RT_REMREF(rt);
1473 		RT_UNLOCK(rt);
1474 	} else if (error != 0)
1475 		log(LOG_INFO, "ifa_add_loopback_route: insertion failed\n");
1476 
1477 	return (error);
1478 }
1479 
1480 int
1481 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1482 {
1483 	int error = 0;
1484 	struct rt_addrinfo info;
1485 	struct sockaddr_dl null_sdl;
1486 
1487 	bzero(&null_sdl, sizeof(null_sdl));
1488 	null_sdl.sdl_len = sizeof(null_sdl);
1489 	null_sdl.sdl_family = AF_LINK;
1490 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1491 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1492 	bzero(&info, sizeof(info));
1493 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1494 	info.rti_info[RTAX_DST] = ia;
1495 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1496 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1497 
1498 	if (error != 0)
1499 		log(LOG_INFO, "ifa_del_loopback_route: deletion failed\n");
1500 
1501 	return (error);
1502 }
1503 
1504 /*
1505  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1506  * structs used to represent other address families, it is necessary
1507  * to perform a different comparison.
1508  */
1509 
1510 #define	sa_equal(a1, a2)	\
1511 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1512 
1513 #define	sa_dl_equal(a1, a2)	\
1514 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1515 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1516 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1517 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1518 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1519 
1520 /*
1521  * Locate an interface based on a complete address.
1522  */
1523 /*ARGSUSED*/
1524 static struct ifaddr *
1525 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1526 {
1527 	struct ifnet *ifp;
1528 	struct ifaddr *ifa;
1529 
1530 	IFNET_RLOCK_NOSLEEP();
1531 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1532 		IF_ADDR_LOCK(ifp);
1533 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1534 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1535 				continue;
1536 			if (sa_equal(addr, ifa->ifa_addr)) {
1537 				if (getref)
1538 					ifa_ref(ifa);
1539 				IF_ADDR_UNLOCK(ifp);
1540 				goto done;
1541 			}
1542 			/* IP6 doesn't have broadcast */
1543 			if ((ifp->if_flags & IFF_BROADCAST) &&
1544 			    ifa->ifa_broadaddr &&
1545 			    ifa->ifa_broadaddr->sa_len != 0 &&
1546 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1547 				if (getref)
1548 					ifa_ref(ifa);
1549 				IF_ADDR_UNLOCK(ifp);
1550 				goto done;
1551 			}
1552 		}
1553 		IF_ADDR_UNLOCK(ifp);
1554 	}
1555 	ifa = NULL;
1556 done:
1557 	IFNET_RUNLOCK_NOSLEEP();
1558 	return (ifa);
1559 }
1560 
1561 struct ifaddr *
1562 ifa_ifwithaddr(struct sockaddr *addr)
1563 {
1564 
1565 	return (ifa_ifwithaddr_internal(addr, 1));
1566 }
1567 
1568 int
1569 ifa_ifwithaddr_check(struct sockaddr *addr)
1570 {
1571 
1572 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1573 }
1574 
1575 /*
1576  * Locate an interface based on the broadcast address.
1577  */
1578 /* ARGSUSED */
1579 struct ifaddr *
1580 ifa_ifwithbroadaddr(struct sockaddr *addr)
1581 {
1582 	struct ifnet *ifp;
1583 	struct ifaddr *ifa;
1584 
1585 	IFNET_RLOCK_NOSLEEP();
1586 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1587 		IF_ADDR_LOCK(ifp);
1588 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1589 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1590 				continue;
1591 			if ((ifp->if_flags & IFF_BROADCAST) &&
1592 			    ifa->ifa_broadaddr &&
1593 			    ifa->ifa_broadaddr->sa_len != 0 &&
1594 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1595 				ifa_ref(ifa);
1596 				IF_ADDR_UNLOCK(ifp);
1597 				goto done;
1598 			}
1599 		}
1600 		IF_ADDR_UNLOCK(ifp);
1601 	}
1602 	ifa = NULL;
1603 done:
1604 	IFNET_RUNLOCK_NOSLEEP();
1605 	return (ifa);
1606 }
1607 
1608 /*
1609  * Locate the point to point interface with a given destination address.
1610  */
1611 /*ARGSUSED*/
1612 struct ifaddr *
1613 ifa_ifwithdstaddr(struct sockaddr *addr)
1614 {
1615 	struct ifnet *ifp;
1616 	struct ifaddr *ifa;
1617 
1618 	IFNET_RLOCK_NOSLEEP();
1619 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1620 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1621 			continue;
1622 		IF_ADDR_LOCK(ifp);
1623 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1624 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1625 				continue;
1626 			if (ifa->ifa_dstaddr != NULL &&
1627 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1628 				ifa_ref(ifa);
1629 				IF_ADDR_UNLOCK(ifp);
1630 				goto done;
1631 			}
1632 		}
1633 		IF_ADDR_UNLOCK(ifp);
1634 	}
1635 	ifa = NULL;
1636 done:
1637 	IFNET_RUNLOCK_NOSLEEP();
1638 	return (ifa);
1639 }
1640 
1641 /*
1642  * Find an interface on a specific network.  If many, choice
1643  * is most specific found.
1644  */
1645 struct ifaddr *
1646 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1647 {
1648 	struct ifnet *ifp;
1649 	struct ifaddr *ifa;
1650 	struct ifaddr *ifa_maybe = NULL;
1651 	u_int af = addr->sa_family;
1652 	char *addr_data = addr->sa_data, *cplim;
1653 
1654 	/*
1655 	 * AF_LINK addresses can be looked up directly by their index number,
1656 	 * so do that if we can.
1657 	 */
1658 	if (af == AF_LINK) {
1659 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1660 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1661 		return (ifaddr_byindex(sdl->sdl_index));
1662 	}
1663 
1664 	/*
1665 	 * Scan though each interface, looking for ones that have addresses
1666 	 * in this address family.  Maintain a reference on ifa_maybe once
1667 	 * we find one, as we release the IF_ADDR_LOCK() that kept it stable
1668 	 * when we move onto the next interface.
1669 	 */
1670 	IFNET_RLOCK_NOSLEEP();
1671 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1672 		IF_ADDR_LOCK(ifp);
1673 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1674 			char *cp, *cp2, *cp3;
1675 
1676 			if (ifa->ifa_addr->sa_family != af)
1677 next:				continue;
1678 			if (af == AF_INET &&
1679 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1680 				/*
1681 				 * This is a bit broken as it doesn't
1682 				 * take into account that the remote end may
1683 				 * be a single node in the network we are
1684 				 * looking for.
1685 				 * The trouble is that we don't know the
1686 				 * netmask for the remote end.
1687 				 */
1688 				if (ifa->ifa_dstaddr != NULL &&
1689 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1690 					ifa_ref(ifa);
1691 					IF_ADDR_UNLOCK(ifp);
1692 					goto done;
1693 				}
1694 			} else {
1695 				/*
1696 				 * if we have a special address handler,
1697 				 * then use it instead of the generic one.
1698 				 */
1699 				if (ifa->ifa_claim_addr) {
1700 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1701 						ifa_ref(ifa);
1702 						IF_ADDR_UNLOCK(ifp);
1703 						goto done;
1704 					}
1705 					continue;
1706 				}
1707 
1708 				/*
1709 				 * Scan all the bits in the ifa's address.
1710 				 * If a bit dissagrees with what we are
1711 				 * looking for, mask it with the netmask
1712 				 * to see if it really matters.
1713 				 * (A byte at a time)
1714 				 */
1715 				if (ifa->ifa_netmask == 0)
1716 					continue;
1717 				cp = addr_data;
1718 				cp2 = ifa->ifa_addr->sa_data;
1719 				cp3 = ifa->ifa_netmask->sa_data;
1720 				cplim = ifa->ifa_netmask->sa_len
1721 					+ (char *)ifa->ifa_netmask;
1722 				while (cp3 < cplim)
1723 					if ((*cp++ ^ *cp2++) & *cp3++)
1724 						goto next; /* next address! */
1725 				/*
1726 				 * If the netmask of what we just found
1727 				 * is more specific than what we had before
1728 				 * (if we had one) then remember the new one
1729 				 * before continuing to search
1730 				 * for an even better one.
1731 				 */
1732 				if (ifa_maybe == NULL ||
1733 				    rn_refines((caddr_t)ifa->ifa_netmask,
1734 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1735 					if (ifa_maybe != NULL)
1736 						ifa_free(ifa_maybe);
1737 					ifa_maybe = ifa;
1738 					ifa_ref(ifa_maybe);
1739 				}
1740 			}
1741 		}
1742 		IF_ADDR_UNLOCK(ifp);
1743 	}
1744 	ifa = ifa_maybe;
1745 	ifa_maybe = NULL;
1746 done:
1747 	IFNET_RUNLOCK_NOSLEEP();
1748 	if (ifa_maybe != NULL)
1749 		ifa_free(ifa_maybe);
1750 	return (ifa);
1751 }
1752 
1753 /*
1754  * Find an interface address specific to an interface best matching
1755  * a given address.
1756  */
1757 struct ifaddr *
1758 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1759 {
1760 	struct ifaddr *ifa;
1761 	char *cp, *cp2, *cp3;
1762 	char *cplim;
1763 	struct ifaddr *ifa_maybe = NULL;
1764 	u_int af = addr->sa_family;
1765 
1766 	if (af >= AF_MAX)
1767 		return (NULL);
1768 	IF_ADDR_LOCK(ifp);
1769 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1770 		if (ifa->ifa_addr->sa_family != af)
1771 			continue;
1772 		if (ifa_maybe == NULL)
1773 			ifa_maybe = ifa;
1774 		if (ifa->ifa_netmask == 0) {
1775 			if (sa_equal(addr, ifa->ifa_addr) ||
1776 			    (ifa->ifa_dstaddr &&
1777 			    sa_equal(addr, ifa->ifa_dstaddr)))
1778 				goto done;
1779 			continue;
1780 		}
1781 		if (ifp->if_flags & IFF_POINTOPOINT) {
1782 			if (sa_equal(addr, ifa->ifa_dstaddr))
1783 				goto done;
1784 		} else {
1785 			cp = addr->sa_data;
1786 			cp2 = ifa->ifa_addr->sa_data;
1787 			cp3 = ifa->ifa_netmask->sa_data;
1788 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1789 			for (; cp3 < cplim; cp3++)
1790 				if ((*cp++ ^ *cp2++) & *cp3)
1791 					break;
1792 			if (cp3 == cplim)
1793 				goto done;
1794 		}
1795 	}
1796 	ifa = ifa_maybe;
1797 done:
1798 	if (ifa != NULL)
1799 		ifa_ref(ifa);
1800 	IF_ADDR_UNLOCK(ifp);
1801 	return (ifa);
1802 }
1803 
1804 #include <net/if_llatbl.h>
1805 
1806 /*
1807  * Default action when installing a route with a Link Level gateway.
1808  * Lookup an appropriate real ifa to point to.
1809  * This should be moved to /sys/net/link.c eventually.
1810  */
1811 static void
1812 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1813 {
1814 	struct ifaddr *ifa, *oifa;
1815 	struct sockaddr *dst;
1816 	struct ifnet *ifp;
1817 
1818 	RT_LOCK_ASSERT(rt);
1819 
1820 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1821 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1822 		return;
1823 	ifa = ifaof_ifpforaddr(dst, ifp);
1824 	if (ifa) {
1825 		oifa = rt->rt_ifa;
1826 		rt->rt_ifa = ifa;
1827 		ifa_free(oifa);
1828 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1829 			ifa->ifa_rtrequest(cmd, rt, info);
1830 	}
1831 }
1832 
1833 /*
1834  * Mark an interface down and notify protocols of
1835  * the transition.
1836  * NOTE: must be called at splnet or eqivalent.
1837  */
1838 static void
1839 if_unroute(struct ifnet *ifp, int flag, int fam)
1840 {
1841 	struct ifaddr *ifa;
1842 
1843 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1844 
1845 	ifp->if_flags &= ~flag;
1846 	getmicrotime(&ifp->if_lastchange);
1847 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1848 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1849 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1850 	ifp->if_qflush(ifp);
1851 
1852 	if (ifp->if_carp)
1853 		(*carp_linkstate_p)(ifp);
1854 	rt_ifmsg(ifp);
1855 }
1856 
1857 /*
1858  * Mark an interface up and notify protocols of
1859  * the transition.
1860  * NOTE: must be called at splnet or eqivalent.
1861  */
1862 static void
1863 if_route(struct ifnet *ifp, int flag, int fam)
1864 {
1865 	struct ifaddr *ifa;
1866 
1867 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1868 
1869 	ifp->if_flags |= flag;
1870 	getmicrotime(&ifp->if_lastchange);
1871 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1872 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1873 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1874 	if (ifp->if_carp)
1875 		(*carp_linkstate_p)(ifp);
1876 	rt_ifmsg(ifp);
1877 #ifdef INET6
1878 	in6_if_up(ifp);
1879 #endif
1880 }
1881 
1882 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1883 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1884 
1885 /*
1886  * Handle a change in the interface link state. To avoid LORs
1887  * between driver lock and upper layer locks, as well as possible
1888  * recursions, we post event to taskqueue, and all job
1889  * is done in static do_link_state_change().
1890  */
1891 void
1892 if_link_state_change(struct ifnet *ifp, int link_state)
1893 {
1894 	/* Return if state hasn't changed. */
1895 	if (ifp->if_link_state == link_state)
1896 		return;
1897 
1898 	ifp->if_link_state = link_state;
1899 
1900 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1901 }
1902 
1903 static void
1904 do_link_state_change(void *arg, int pending)
1905 {
1906 	struct ifnet *ifp = (struct ifnet *)arg;
1907 	int link_state = ifp->if_link_state;
1908 	CURVNET_SET(ifp->if_vnet);
1909 
1910 	/* Notify that the link state has changed. */
1911 	rt_ifmsg(ifp);
1912 	if (ifp->if_vlantrunk != NULL)
1913 		(*vlan_link_state_p)(ifp);
1914 
1915 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1916 	    IFP2AC(ifp)->ac_netgraph != NULL)
1917 		(*ng_ether_link_state_p)(ifp, link_state);
1918 	if (ifp->if_carp)
1919 		(*carp_linkstate_p)(ifp);
1920 	if (ifp->if_bridge) {
1921 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1922 		(*bstp_linkstate_p)(ifp, link_state);
1923 	}
1924 	if (ifp->if_lagg) {
1925 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1926 		(*lagg_linkstate_p)(ifp, link_state);
1927 	}
1928 
1929 	if (IS_DEFAULT_VNET(curvnet))
1930 		devctl_notify("IFNET", ifp->if_xname,
1931 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1932 		    NULL);
1933 	if (pending > 1)
1934 		if_printf(ifp, "%d link states coalesced\n", pending);
1935 	if (log_link_state_change)
1936 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1937 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1938 	CURVNET_RESTORE();
1939 }
1940 
1941 /*
1942  * Mark an interface down and notify protocols of
1943  * the transition.
1944  * NOTE: must be called at splnet or eqivalent.
1945  */
1946 void
1947 if_down(struct ifnet *ifp)
1948 {
1949 
1950 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1951 }
1952 
1953 /*
1954  * Mark an interface up and notify protocols of
1955  * the transition.
1956  * NOTE: must be called at splnet or eqivalent.
1957  */
1958 void
1959 if_up(struct ifnet *ifp)
1960 {
1961 
1962 	if_route(ifp, IFF_UP, AF_UNSPEC);
1963 }
1964 
1965 /*
1966  * Flush an interface queue.
1967  */
1968 void
1969 if_qflush(struct ifnet *ifp)
1970 {
1971 	struct mbuf *m, *n;
1972 	struct ifaltq *ifq;
1973 
1974 	ifq = &ifp->if_snd;
1975 	IFQ_LOCK(ifq);
1976 #ifdef ALTQ
1977 	if (ALTQ_IS_ENABLED(ifq))
1978 		ALTQ_PURGE(ifq);
1979 #endif
1980 	n = ifq->ifq_head;
1981 	while ((m = n) != 0) {
1982 		n = m->m_act;
1983 		m_freem(m);
1984 	}
1985 	ifq->ifq_head = 0;
1986 	ifq->ifq_tail = 0;
1987 	ifq->ifq_len = 0;
1988 	IFQ_UNLOCK(ifq);
1989 }
1990 
1991 /*
1992  * Map interface name to interface structure pointer, with or without
1993  * returning a reference.
1994  */
1995 struct ifnet *
1996 ifunit_ref(const char *name)
1997 {
1998 	struct ifnet *ifp;
1999 
2000 	IFNET_RLOCK_NOSLEEP();
2001 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2002 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2003 		    !(ifp->if_flags & IFF_DYING))
2004 			break;
2005 	}
2006 	if (ifp != NULL)
2007 		if_ref(ifp);
2008 	IFNET_RUNLOCK_NOSLEEP();
2009 	return (ifp);
2010 }
2011 
2012 struct ifnet *
2013 ifunit(const char *name)
2014 {
2015 	struct ifnet *ifp;
2016 
2017 	IFNET_RLOCK_NOSLEEP();
2018 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2019 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2020 			break;
2021 	}
2022 	IFNET_RUNLOCK_NOSLEEP();
2023 	return (ifp);
2024 }
2025 
2026 /*
2027  * Hardware specific interface ioctls.
2028  */
2029 static int
2030 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2031 {
2032 	struct ifreq *ifr;
2033 	struct ifstat *ifs;
2034 	int error = 0;
2035 	int new_flags, temp_flags;
2036 	size_t namelen, onamelen;
2037 	size_t descrlen;
2038 	char *descrbuf, *odescrbuf;
2039 	char new_name[IFNAMSIZ];
2040 	struct ifaddr *ifa;
2041 	struct sockaddr_dl *sdl;
2042 
2043 	ifr = (struct ifreq *)data;
2044 	switch (cmd) {
2045 	case SIOCGIFINDEX:
2046 		ifr->ifr_index = ifp->if_index;
2047 		break;
2048 
2049 	case SIOCGIFFLAGS:
2050 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2051 		ifr->ifr_flags = temp_flags & 0xffff;
2052 		ifr->ifr_flagshigh = temp_flags >> 16;
2053 		break;
2054 
2055 	case SIOCGIFCAP:
2056 		ifr->ifr_reqcap = ifp->if_capabilities;
2057 		ifr->ifr_curcap = ifp->if_capenable;
2058 		break;
2059 
2060 #ifdef MAC
2061 	case SIOCGIFMAC:
2062 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2063 		break;
2064 #endif
2065 
2066 	case SIOCGIFMETRIC:
2067 		ifr->ifr_metric = ifp->if_metric;
2068 		break;
2069 
2070 	case SIOCGIFMTU:
2071 		ifr->ifr_mtu = ifp->if_mtu;
2072 		break;
2073 
2074 	case SIOCGIFPHYS:
2075 		ifr->ifr_phys = ifp->if_physical;
2076 		break;
2077 
2078 	case SIOCGIFDESCR:
2079 		error = 0;
2080 		sx_slock(&ifdescr_sx);
2081 		if (ifp->if_description == NULL)
2082 			error = ENOMSG;
2083 		else {
2084 			/* space for terminating nul */
2085 			descrlen = strlen(ifp->if_description) + 1;
2086 			if (ifr->ifr_buffer.length < descrlen)
2087 				ifr->ifr_buffer.buffer = NULL;
2088 			else
2089 				error = copyout(ifp->if_description,
2090 				    ifr->ifr_buffer.buffer, descrlen);
2091 			ifr->ifr_buffer.length = descrlen;
2092 		}
2093 		sx_sunlock(&ifdescr_sx);
2094 		break;
2095 
2096 	case SIOCSIFDESCR:
2097 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2098 		if (error)
2099 			return (error);
2100 
2101 		/*
2102 		 * Copy only (length-1) bytes to make sure that
2103 		 * if_description is always nul terminated.  The
2104 		 * length parameter is supposed to count the
2105 		 * terminating nul in.
2106 		 */
2107 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2108 			return (ENAMETOOLONG);
2109 		else if (ifr->ifr_buffer.length == 0)
2110 			descrbuf = NULL;
2111 		else {
2112 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2113 			    M_WAITOK | M_ZERO);
2114 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2115 			    ifr->ifr_buffer.length - 1);
2116 			if (error) {
2117 				free(descrbuf, M_IFDESCR);
2118 				break;
2119 			}
2120 		}
2121 
2122 		sx_xlock(&ifdescr_sx);
2123 		odescrbuf = ifp->if_description;
2124 		ifp->if_description = descrbuf;
2125 		sx_xunlock(&ifdescr_sx);
2126 
2127 		getmicrotime(&ifp->if_lastchange);
2128 		free(odescrbuf, M_IFDESCR);
2129 		break;
2130 
2131 	case SIOCSIFFLAGS:
2132 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2133 		if (error)
2134 			return (error);
2135 		/*
2136 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2137 		 * check, so we don't need special handling here yet.
2138 		 */
2139 		new_flags = (ifr->ifr_flags & 0xffff) |
2140 		    (ifr->ifr_flagshigh << 16);
2141 		if (ifp->if_flags & IFF_SMART) {
2142 			/* Smart drivers twiddle their own routes */
2143 		} else if (ifp->if_flags & IFF_UP &&
2144 		    (new_flags & IFF_UP) == 0) {
2145 			int s = splimp();
2146 			if_down(ifp);
2147 			splx(s);
2148 		} else if (new_flags & IFF_UP &&
2149 		    (ifp->if_flags & IFF_UP) == 0) {
2150 			int s = splimp();
2151 			if_up(ifp);
2152 			splx(s);
2153 		}
2154 		/* See if permanently promiscuous mode bit is about to flip */
2155 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2156 			if (new_flags & IFF_PPROMISC)
2157 				ifp->if_flags |= IFF_PROMISC;
2158 			else if (ifp->if_pcount == 0)
2159 				ifp->if_flags &= ~IFF_PROMISC;
2160 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2161 			    ifp->if_xname,
2162 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2163 		}
2164 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2165 			(new_flags &~ IFF_CANTCHANGE);
2166 		if (ifp->if_ioctl) {
2167 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2168 		}
2169 		getmicrotime(&ifp->if_lastchange);
2170 		break;
2171 
2172 	case SIOCSIFCAP:
2173 		error = priv_check(td, PRIV_NET_SETIFCAP);
2174 		if (error)
2175 			return (error);
2176 		if (ifp->if_ioctl == NULL)
2177 			return (EOPNOTSUPP);
2178 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2179 			return (EINVAL);
2180 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2181 		if (error == 0)
2182 			getmicrotime(&ifp->if_lastchange);
2183 		break;
2184 
2185 #ifdef MAC
2186 	case SIOCSIFMAC:
2187 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2188 		break;
2189 #endif
2190 
2191 	case SIOCSIFNAME:
2192 		error = priv_check(td, PRIV_NET_SETIFNAME);
2193 		if (error)
2194 			return (error);
2195 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2196 		if (error != 0)
2197 			return (error);
2198 		if (new_name[0] == '\0')
2199 			return (EINVAL);
2200 		if (ifunit(new_name) != NULL)
2201 			return (EEXIST);
2202 
2203 		/*
2204 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2205 		 * and there are numerous other races with the
2206 		 * ifunit() checks not being atomic with namespace
2207 		 * changes (renames, vmoves, if_attach, etc).
2208 		 */
2209 		ifp->if_flags |= IFF_RENAMING;
2210 
2211 		/* Announce the departure of the interface. */
2212 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2213 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2214 
2215 		log(LOG_INFO, "%s: changing name to '%s'\n",
2216 		    ifp->if_xname, new_name);
2217 
2218 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2219 		ifa = ifp->if_addr;
2220 		IFA_LOCK(ifa);
2221 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2222 		namelen = strlen(new_name);
2223 		onamelen = sdl->sdl_nlen;
2224 		/*
2225 		 * Move the address if needed.  This is safe because we
2226 		 * allocate space for a name of length IFNAMSIZ when we
2227 		 * create this in if_attach().
2228 		 */
2229 		if (namelen != onamelen) {
2230 			bcopy(sdl->sdl_data + onamelen,
2231 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2232 		}
2233 		bcopy(new_name, sdl->sdl_data, namelen);
2234 		sdl->sdl_nlen = namelen;
2235 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2236 		bzero(sdl->sdl_data, onamelen);
2237 		while (namelen != 0)
2238 			sdl->sdl_data[--namelen] = 0xff;
2239 		IFA_UNLOCK(ifa);
2240 
2241 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2242 		/* Announce the return of the interface. */
2243 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2244 
2245 		ifp->if_flags &= ~IFF_RENAMING;
2246 		break;
2247 
2248 #ifdef VIMAGE
2249 	case SIOCSIFVNET:
2250 		error = priv_check(td, PRIV_NET_SETIFVNET);
2251 		if (error)
2252 			return (error);
2253 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2254 		break;
2255 #endif
2256 
2257 	case SIOCSIFMETRIC:
2258 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2259 		if (error)
2260 			return (error);
2261 		ifp->if_metric = ifr->ifr_metric;
2262 		getmicrotime(&ifp->if_lastchange);
2263 		break;
2264 
2265 	case SIOCSIFPHYS:
2266 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2267 		if (error)
2268 			return (error);
2269 		if (ifp->if_ioctl == NULL)
2270 			return (EOPNOTSUPP);
2271 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2272 		if (error == 0)
2273 			getmicrotime(&ifp->if_lastchange);
2274 		break;
2275 
2276 	case SIOCSIFMTU:
2277 	{
2278 		u_long oldmtu = ifp->if_mtu;
2279 
2280 		error = priv_check(td, PRIV_NET_SETIFMTU);
2281 		if (error)
2282 			return (error);
2283 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2284 			return (EINVAL);
2285 		if (ifp->if_ioctl == NULL)
2286 			return (EOPNOTSUPP);
2287 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2288 		if (error == 0) {
2289 			getmicrotime(&ifp->if_lastchange);
2290 			rt_ifmsg(ifp);
2291 		}
2292 		/*
2293 		 * If the link MTU changed, do network layer specific procedure.
2294 		 */
2295 		if (ifp->if_mtu != oldmtu) {
2296 #ifdef INET6
2297 			nd6_setmtu(ifp);
2298 #endif
2299 		}
2300 		break;
2301 	}
2302 
2303 	case SIOCADDMULTI:
2304 	case SIOCDELMULTI:
2305 		if (cmd == SIOCADDMULTI)
2306 			error = priv_check(td, PRIV_NET_ADDMULTI);
2307 		else
2308 			error = priv_check(td, PRIV_NET_DELMULTI);
2309 		if (error)
2310 			return (error);
2311 
2312 		/* Don't allow group membership on non-multicast interfaces. */
2313 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2314 			return (EOPNOTSUPP);
2315 
2316 		/* Don't let users screw up protocols' entries. */
2317 		if (ifr->ifr_addr.sa_family != AF_LINK)
2318 			return (EINVAL);
2319 
2320 		if (cmd == SIOCADDMULTI) {
2321 			struct ifmultiaddr *ifma;
2322 
2323 			/*
2324 			 * Userland is only permitted to join groups once
2325 			 * via the if_addmulti() KPI, because it cannot hold
2326 			 * struct ifmultiaddr * between calls. It may also
2327 			 * lose a race while we check if the membership
2328 			 * already exists.
2329 			 */
2330 			IF_ADDR_LOCK(ifp);
2331 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2332 			IF_ADDR_UNLOCK(ifp);
2333 			if (ifma != NULL)
2334 				error = EADDRINUSE;
2335 			else
2336 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2337 		} else {
2338 			error = if_delmulti(ifp, &ifr->ifr_addr);
2339 		}
2340 		if (error == 0)
2341 			getmicrotime(&ifp->if_lastchange);
2342 		break;
2343 
2344 	case SIOCSIFPHYADDR:
2345 	case SIOCDIFPHYADDR:
2346 #ifdef INET6
2347 	case SIOCSIFPHYADDR_IN6:
2348 #endif
2349 	case SIOCSLIFPHYADDR:
2350 	case SIOCSIFMEDIA:
2351 	case SIOCSIFGENERIC:
2352 		error = priv_check(td, PRIV_NET_HWIOCTL);
2353 		if (error)
2354 			return (error);
2355 		if (ifp->if_ioctl == NULL)
2356 			return (EOPNOTSUPP);
2357 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2358 		if (error == 0)
2359 			getmicrotime(&ifp->if_lastchange);
2360 		break;
2361 
2362 	case SIOCGIFSTATUS:
2363 		ifs = (struct ifstat *)data;
2364 		ifs->ascii[0] = '\0';
2365 
2366 	case SIOCGIFPSRCADDR:
2367 	case SIOCGIFPDSTADDR:
2368 	case SIOCGLIFPHYADDR:
2369 	case SIOCGIFMEDIA:
2370 	case SIOCGIFGENERIC:
2371 		if (ifp->if_ioctl == NULL)
2372 			return (EOPNOTSUPP);
2373 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2374 		break;
2375 
2376 	case SIOCSIFLLADDR:
2377 		error = priv_check(td, PRIV_NET_SETLLADDR);
2378 		if (error)
2379 			return (error);
2380 		error = if_setlladdr(ifp,
2381 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2382 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2383 		break;
2384 
2385 	case SIOCAIFGROUP:
2386 	{
2387 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2388 
2389 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2390 		if (error)
2391 			return (error);
2392 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2393 			return (error);
2394 		break;
2395 	}
2396 
2397 	case SIOCGIFGROUP:
2398 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2399 			return (error);
2400 		break;
2401 
2402 	case SIOCDIFGROUP:
2403 	{
2404 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2405 
2406 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2407 		if (error)
2408 			return (error);
2409 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2410 			return (error);
2411 		break;
2412 	}
2413 
2414 	default:
2415 		error = ENOIOCTL;
2416 		break;
2417 	}
2418 	return (error);
2419 }
2420 
2421 #ifdef COMPAT_FREEBSD32
2422 struct ifconf32 {
2423 	int32_t	ifc_len;
2424 	union {
2425 		uint32_t	ifcu_buf;
2426 		uint32_t	ifcu_req;
2427 	} ifc_ifcu;
2428 };
2429 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2430 #endif
2431 
2432 /*
2433  * Interface ioctls.
2434  */
2435 int
2436 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2437 {
2438 	struct ifnet *ifp;
2439 	struct ifreq *ifr;
2440 	int error;
2441 	int oif_flags;
2442 
2443 	CURVNET_SET(so->so_vnet);
2444 	switch (cmd) {
2445 	case SIOCGIFCONF:
2446 	case OSIOCGIFCONF:
2447 		error = ifconf(cmd, data);
2448 		CURVNET_RESTORE();
2449 		return (error);
2450 
2451 #ifdef COMPAT_FREEBSD32
2452 	case SIOCGIFCONF32:
2453 		{
2454 			struct ifconf32 *ifc32;
2455 			struct ifconf ifc;
2456 
2457 			ifc32 = (struct ifconf32 *)data;
2458 			ifc.ifc_len = ifc32->ifc_len;
2459 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2460 
2461 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2462 			CURVNET_RESTORE();
2463 			return (error);
2464 		}
2465 #endif
2466 	}
2467 	ifr = (struct ifreq *)data;
2468 
2469 	switch (cmd) {
2470 #ifdef VIMAGE
2471 	case SIOCSIFRVNET:
2472 		error = priv_check(td, PRIV_NET_SETIFVNET);
2473 		if (error == 0)
2474 			error = if_vmove_reclaim(td, ifr->ifr_name,
2475 			    ifr->ifr_jid);
2476 		CURVNET_RESTORE();
2477 		return (error);
2478 #endif
2479 	case SIOCIFCREATE:
2480 	case SIOCIFCREATE2:
2481 		error = priv_check(td, PRIV_NET_IFCREATE);
2482 		if (error == 0)
2483 			error = if_clone_create(ifr->ifr_name,
2484 			    sizeof(ifr->ifr_name),
2485 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2486 		CURVNET_RESTORE();
2487 		return (error);
2488 	case SIOCIFDESTROY:
2489 		error = priv_check(td, PRIV_NET_IFDESTROY);
2490 		if (error == 0)
2491 			error = if_clone_destroy(ifr->ifr_name);
2492 		CURVNET_RESTORE();
2493 		return (error);
2494 
2495 	case SIOCIFGCLONERS:
2496 		error = if_clone_list((struct if_clonereq *)data);
2497 		CURVNET_RESTORE();
2498 		return (error);
2499 	case SIOCGIFGMEMB:
2500 		error = if_getgroupmembers((struct ifgroupreq *)data);
2501 		CURVNET_RESTORE();
2502 		return (error);
2503 	}
2504 
2505 	ifp = ifunit_ref(ifr->ifr_name);
2506 	if (ifp == NULL) {
2507 		CURVNET_RESTORE();
2508 		return (ENXIO);
2509 	}
2510 
2511 	error = ifhwioctl(cmd, ifp, data, td);
2512 	if (error != ENOIOCTL) {
2513 		if_rele(ifp);
2514 		CURVNET_RESTORE();
2515 		return (error);
2516 	}
2517 
2518 	oif_flags = ifp->if_flags;
2519 	if (so->so_proto == NULL) {
2520 		if_rele(ifp);
2521 		CURVNET_RESTORE();
2522 		return (EOPNOTSUPP);
2523 	}
2524 #ifndef COMPAT_43
2525 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2526 								 data,
2527 								 ifp, td));
2528 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2529 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2530 #else
2531 	{
2532 		u_long ocmd = cmd;
2533 
2534 		switch (cmd) {
2535 
2536 		case SIOCSIFDSTADDR:
2537 		case SIOCSIFADDR:
2538 		case SIOCSIFBRDADDR:
2539 		case SIOCSIFNETMASK:
2540 #if BYTE_ORDER != BIG_ENDIAN
2541 			if (ifr->ifr_addr.sa_family == 0 &&
2542 			    ifr->ifr_addr.sa_len < 16) {
2543 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2544 				ifr->ifr_addr.sa_len = 16;
2545 			}
2546 #else
2547 			if (ifr->ifr_addr.sa_len == 0)
2548 				ifr->ifr_addr.sa_len = 16;
2549 #endif
2550 			break;
2551 
2552 		case OSIOCGIFADDR:
2553 			cmd = SIOCGIFADDR;
2554 			break;
2555 
2556 		case OSIOCGIFDSTADDR:
2557 			cmd = SIOCGIFDSTADDR;
2558 			break;
2559 
2560 		case OSIOCGIFBRDADDR:
2561 			cmd = SIOCGIFBRDADDR;
2562 			break;
2563 
2564 		case OSIOCGIFNETMASK:
2565 			cmd = SIOCGIFNETMASK;
2566 		}
2567 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2568 								   cmd,
2569 								   data,
2570 								   ifp, td));
2571 		if (error == EOPNOTSUPP && ifp != NULL &&
2572 		    ifp->if_ioctl != NULL)
2573 			error = (*ifp->if_ioctl)(ifp, cmd, data);
2574 		switch (ocmd) {
2575 
2576 		case OSIOCGIFADDR:
2577 		case OSIOCGIFDSTADDR:
2578 		case OSIOCGIFBRDADDR:
2579 		case OSIOCGIFNETMASK:
2580 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2581 
2582 		}
2583 	}
2584 #endif /* COMPAT_43 */
2585 
2586 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2587 #ifdef INET6
2588 		if (ifp->if_flags & IFF_UP) {
2589 			int s = splimp();
2590 			in6_if_up(ifp);
2591 			splx(s);
2592 		}
2593 #endif
2594 	}
2595 	if_rele(ifp);
2596 	CURVNET_RESTORE();
2597 	return (error);
2598 }
2599 
2600 /*
2601  * The code common to handling reference counted flags,
2602  * e.g., in ifpromisc() and if_allmulti().
2603  * The "pflag" argument can specify a permanent mode flag to check,
2604  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2605  *
2606  * Only to be used on stack-owned flags, not driver-owned flags.
2607  */
2608 static int
2609 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2610 {
2611 	struct ifreq ifr;
2612 	int error;
2613 	int oldflags, oldcount;
2614 
2615 	/* Sanity checks to catch programming errors */
2616 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2617 	    ("%s: setting driver-owned flag %d", __func__, flag));
2618 
2619 	if (onswitch)
2620 		KASSERT(*refcount >= 0,
2621 		    ("%s: increment negative refcount %d for flag %d",
2622 		    __func__, *refcount, flag));
2623 	else
2624 		KASSERT(*refcount > 0,
2625 		    ("%s: decrement non-positive refcount %d for flag %d",
2626 		    __func__, *refcount, flag));
2627 
2628 	/* In case this mode is permanent, just touch refcount */
2629 	if (ifp->if_flags & pflag) {
2630 		*refcount += onswitch ? 1 : -1;
2631 		return (0);
2632 	}
2633 
2634 	/* Save ifnet parameters for if_ioctl() may fail */
2635 	oldcount = *refcount;
2636 	oldflags = ifp->if_flags;
2637 
2638 	/*
2639 	 * See if we aren't the only and touching refcount is enough.
2640 	 * Actually toggle interface flag if we are the first or last.
2641 	 */
2642 	if (onswitch) {
2643 		if ((*refcount)++)
2644 			return (0);
2645 		ifp->if_flags |= flag;
2646 	} else {
2647 		if (--(*refcount))
2648 			return (0);
2649 		ifp->if_flags &= ~flag;
2650 	}
2651 
2652 	/* Call down the driver since we've changed interface flags */
2653 	if (ifp->if_ioctl == NULL) {
2654 		error = EOPNOTSUPP;
2655 		goto recover;
2656 	}
2657 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2658 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2659 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2660 	if (error)
2661 		goto recover;
2662 	/* Notify userland that interface flags have changed */
2663 	rt_ifmsg(ifp);
2664 	return (0);
2665 
2666 recover:
2667 	/* Recover after driver error */
2668 	*refcount = oldcount;
2669 	ifp->if_flags = oldflags;
2670 	return (error);
2671 }
2672 
2673 /*
2674  * Set/clear promiscuous mode on interface ifp based on the truth value
2675  * of pswitch.  The calls are reference counted so that only the first
2676  * "on" request actually has an effect, as does the final "off" request.
2677  * Results are undefined if the "off" and "on" requests are not matched.
2678  */
2679 int
2680 ifpromisc(struct ifnet *ifp, int pswitch)
2681 {
2682 	int error;
2683 	int oldflags = ifp->if_flags;
2684 
2685 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2686 			   &ifp->if_pcount, pswitch);
2687 	/* If promiscuous mode status has changed, log a message */
2688 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2689 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2690 		    ifp->if_xname,
2691 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2692 	return (error);
2693 }
2694 
2695 /*
2696  * Return interface configuration
2697  * of system.  List may be used
2698  * in later ioctl's (above) to get
2699  * other information.
2700  */
2701 /*ARGSUSED*/
2702 static int
2703 ifconf(u_long cmd, caddr_t data)
2704 {
2705 	struct ifconf *ifc = (struct ifconf *)data;
2706 	struct ifnet *ifp;
2707 	struct ifaddr *ifa;
2708 	struct ifreq ifr;
2709 	struct sbuf *sb;
2710 	int error, full = 0, valid_len, max_len;
2711 
2712 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2713 	max_len = MAXPHYS - 1;
2714 
2715 	/* Prevent hostile input from being able to crash the system */
2716 	if (ifc->ifc_len <= 0)
2717 		return (EINVAL);
2718 
2719 again:
2720 	if (ifc->ifc_len <= max_len) {
2721 		max_len = ifc->ifc_len;
2722 		full = 1;
2723 	}
2724 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2725 	max_len = 0;
2726 	valid_len = 0;
2727 
2728 	IFNET_RLOCK();
2729 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2730 		int addrs;
2731 
2732 		/*
2733 		 * Zero the ifr_name buffer to make sure we don't
2734 		 * disclose the contents of the stack.
2735 		 */
2736 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2737 
2738 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2739 		    >= sizeof(ifr.ifr_name)) {
2740 			sbuf_delete(sb);
2741 			IFNET_RUNLOCK();
2742 			return (ENAMETOOLONG);
2743 		}
2744 
2745 		addrs = 0;
2746 		IF_ADDR_LOCK(ifp);
2747 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2748 			struct sockaddr *sa = ifa->ifa_addr;
2749 
2750 			if (prison_if(curthread->td_ucred, sa) != 0)
2751 				continue;
2752 			addrs++;
2753 #ifdef COMPAT_43
2754 			if (cmd == OSIOCGIFCONF) {
2755 				struct osockaddr *osa =
2756 					 (struct osockaddr *)&ifr.ifr_addr;
2757 				ifr.ifr_addr = *sa;
2758 				osa->sa_family = sa->sa_family;
2759 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2760 				max_len += sizeof(ifr);
2761 			} else
2762 #endif
2763 			if (sa->sa_len <= sizeof(*sa)) {
2764 				ifr.ifr_addr = *sa;
2765 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2766 				max_len += sizeof(ifr);
2767 			} else {
2768 				sbuf_bcat(sb, &ifr,
2769 				    offsetof(struct ifreq, ifr_addr));
2770 				max_len += offsetof(struct ifreq, ifr_addr);
2771 				sbuf_bcat(sb, sa, sa->sa_len);
2772 				max_len += sa->sa_len;
2773 			}
2774 
2775 			if (sbuf_error(sb) == 0)
2776 				valid_len = sbuf_len(sb);
2777 		}
2778 		IF_ADDR_UNLOCK(ifp);
2779 		if (addrs == 0) {
2780 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2781 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2782 			max_len += sizeof(ifr);
2783 
2784 			if (sbuf_error(sb) == 0)
2785 				valid_len = sbuf_len(sb);
2786 		}
2787 	}
2788 	IFNET_RUNLOCK();
2789 
2790 	/*
2791 	 * If we didn't allocate enough space (uncommon), try again.  If
2792 	 * we have already allocated as much space as we are allowed,
2793 	 * return what we've got.
2794 	 */
2795 	if (valid_len != max_len && !full) {
2796 		sbuf_delete(sb);
2797 		goto again;
2798 	}
2799 
2800 	ifc->ifc_len = valid_len;
2801 	sbuf_finish(sb);
2802 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2803 	sbuf_delete(sb);
2804 	return (error);
2805 }
2806 
2807 /*
2808  * Just like ifpromisc(), but for all-multicast-reception mode.
2809  */
2810 int
2811 if_allmulti(struct ifnet *ifp, int onswitch)
2812 {
2813 
2814 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2815 }
2816 
2817 struct ifmultiaddr *
2818 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2819 {
2820 	struct ifmultiaddr *ifma;
2821 
2822 	IF_ADDR_LOCK_ASSERT(ifp);
2823 
2824 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2825 		if (sa->sa_family == AF_LINK) {
2826 			if (sa_dl_equal(ifma->ifma_addr, sa))
2827 				break;
2828 		} else {
2829 			if (sa_equal(ifma->ifma_addr, sa))
2830 				break;
2831 		}
2832 	}
2833 
2834 	return ifma;
2835 }
2836 
2837 /*
2838  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2839  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2840  * the ifnet multicast address list here, so the caller must do that and
2841  * other setup work (such as notifying the device driver).  The reference
2842  * count is initialized to 1.
2843  */
2844 static struct ifmultiaddr *
2845 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2846     int mflags)
2847 {
2848 	struct ifmultiaddr *ifma;
2849 	struct sockaddr *dupsa;
2850 
2851 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2852 	    M_ZERO);
2853 	if (ifma == NULL)
2854 		return (NULL);
2855 
2856 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2857 	if (dupsa == NULL) {
2858 		free(ifma, M_IFMADDR);
2859 		return (NULL);
2860 	}
2861 	bcopy(sa, dupsa, sa->sa_len);
2862 	ifma->ifma_addr = dupsa;
2863 
2864 	ifma->ifma_ifp = ifp;
2865 	ifma->ifma_refcount = 1;
2866 	ifma->ifma_protospec = NULL;
2867 
2868 	if (llsa == NULL) {
2869 		ifma->ifma_lladdr = NULL;
2870 		return (ifma);
2871 	}
2872 
2873 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2874 	if (dupsa == NULL) {
2875 		free(ifma->ifma_addr, M_IFMADDR);
2876 		free(ifma, M_IFMADDR);
2877 		return (NULL);
2878 	}
2879 	bcopy(llsa, dupsa, llsa->sa_len);
2880 	ifma->ifma_lladdr = dupsa;
2881 
2882 	return (ifma);
2883 }
2884 
2885 /*
2886  * if_freemulti: free ifmultiaddr structure and possibly attached related
2887  * addresses.  The caller is responsible for implementing reference
2888  * counting, notifying the driver, handling routing messages, and releasing
2889  * any dependent link layer state.
2890  */
2891 static void
2892 if_freemulti(struct ifmultiaddr *ifma)
2893 {
2894 
2895 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2896 	    ifma->ifma_refcount));
2897 	KASSERT(ifma->ifma_protospec == NULL,
2898 	    ("if_freemulti: protospec not NULL"));
2899 
2900 	if (ifma->ifma_lladdr != NULL)
2901 		free(ifma->ifma_lladdr, M_IFMADDR);
2902 	free(ifma->ifma_addr, M_IFMADDR);
2903 	free(ifma, M_IFMADDR);
2904 }
2905 
2906 /*
2907  * Register an additional multicast address with a network interface.
2908  *
2909  * - If the address is already present, bump the reference count on the
2910  *   address and return.
2911  * - If the address is not link-layer, look up a link layer address.
2912  * - Allocate address structures for one or both addresses, and attach to the
2913  *   multicast address list on the interface.  If automatically adding a link
2914  *   layer address, the protocol address will own a reference to the link
2915  *   layer address, to be freed when it is freed.
2916  * - Notify the network device driver of an addition to the multicast address
2917  *   list.
2918  *
2919  * 'sa' points to caller-owned memory with the desired multicast address.
2920  *
2921  * 'retifma' will be used to return a pointer to the resulting multicast
2922  * address reference, if desired.
2923  */
2924 int
2925 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2926     struct ifmultiaddr **retifma)
2927 {
2928 	struct ifmultiaddr *ifma, *ll_ifma;
2929 	struct sockaddr *llsa;
2930 	int error;
2931 
2932 	/*
2933 	 * If the address is already present, return a new reference to it;
2934 	 * otherwise, allocate storage and set up a new address.
2935 	 */
2936 	IF_ADDR_LOCK(ifp);
2937 	ifma = if_findmulti(ifp, sa);
2938 	if (ifma != NULL) {
2939 		ifma->ifma_refcount++;
2940 		if (retifma != NULL)
2941 			*retifma = ifma;
2942 		IF_ADDR_UNLOCK(ifp);
2943 		return (0);
2944 	}
2945 
2946 	/*
2947 	 * The address isn't already present; resolve the protocol address
2948 	 * into a link layer address, and then look that up, bump its
2949 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2950 	 * returned, we will need to free it later.
2951 	 */
2952 	llsa = NULL;
2953 	ll_ifma = NULL;
2954 	if (ifp->if_resolvemulti != NULL) {
2955 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2956 		if (error)
2957 			goto unlock_out;
2958 	}
2959 
2960 	/*
2961 	 * Allocate the new address.  Don't hook it up yet, as we may also
2962 	 * need to allocate a link layer multicast address.
2963 	 */
2964 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2965 	if (ifma == NULL) {
2966 		error = ENOMEM;
2967 		goto free_llsa_out;
2968 	}
2969 
2970 	/*
2971 	 * If a link layer address is found, we'll need to see if it's
2972 	 * already present in the address list, or allocate is as well.
2973 	 * When this block finishes, the link layer address will be on the
2974 	 * list.
2975 	 */
2976 	if (llsa != NULL) {
2977 		ll_ifma = if_findmulti(ifp, llsa);
2978 		if (ll_ifma == NULL) {
2979 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2980 			if (ll_ifma == NULL) {
2981 				--ifma->ifma_refcount;
2982 				if_freemulti(ifma);
2983 				error = ENOMEM;
2984 				goto free_llsa_out;
2985 			}
2986 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
2987 			    ifma_link);
2988 		} else
2989 			ll_ifma->ifma_refcount++;
2990 		ifma->ifma_llifma = ll_ifma;
2991 	}
2992 
2993 	/*
2994 	 * We now have a new multicast address, ifma, and possibly a new or
2995 	 * referenced link layer address.  Add the primary address to the
2996 	 * ifnet address list.
2997 	 */
2998 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2999 
3000 	if (retifma != NULL)
3001 		*retifma = ifma;
3002 
3003 	/*
3004 	 * Must generate the message while holding the lock so that 'ifma'
3005 	 * pointer is still valid.
3006 	 */
3007 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3008 	IF_ADDR_UNLOCK(ifp);
3009 
3010 	/*
3011 	 * We are certain we have added something, so call down to the
3012 	 * interface to let them know about it.
3013 	 */
3014 	if (ifp->if_ioctl != NULL) {
3015 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3016 	}
3017 
3018 	if (llsa != NULL)
3019 		free(llsa, M_IFMADDR);
3020 
3021 	return (0);
3022 
3023 free_llsa_out:
3024 	if (llsa != NULL)
3025 		free(llsa, M_IFMADDR);
3026 
3027 unlock_out:
3028 	IF_ADDR_UNLOCK(ifp);
3029 	return (error);
3030 }
3031 
3032 /*
3033  * Delete a multicast group membership by network-layer group address.
3034  *
3035  * Returns ENOENT if the entry could not be found. If ifp no longer
3036  * exists, results are undefined. This entry point should only be used
3037  * from subsystems which do appropriate locking to hold ifp for the
3038  * duration of the call.
3039  * Network-layer protocol domains must use if_delmulti_ifma().
3040  */
3041 int
3042 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3043 {
3044 	struct ifmultiaddr *ifma;
3045 	int lastref;
3046 #ifdef INVARIANTS
3047 	struct ifnet *oifp;
3048 
3049 	IFNET_RLOCK_NOSLEEP();
3050 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3051 		if (ifp == oifp)
3052 			break;
3053 	if (ifp != oifp)
3054 		ifp = NULL;
3055 	IFNET_RUNLOCK_NOSLEEP();
3056 
3057 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3058 #endif
3059 	if (ifp == NULL)
3060 		return (ENOENT);
3061 
3062 	IF_ADDR_LOCK(ifp);
3063 	lastref = 0;
3064 	ifma = if_findmulti(ifp, sa);
3065 	if (ifma != NULL)
3066 		lastref = if_delmulti_locked(ifp, ifma, 0);
3067 	IF_ADDR_UNLOCK(ifp);
3068 
3069 	if (ifma == NULL)
3070 		return (ENOENT);
3071 
3072 	if (lastref && ifp->if_ioctl != NULL) {
3073 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3074 	}
3075 
3076 	return (0);
3077 }
3078 
3079 /*
3080  * Delete all multicast group membership for an interface.
3081  * Should be used to quickly flush all multicast filters.
3082  */
3083 void
3084 if_delallmulti(struct ifnet *ifp)
3085 {
3086 	struct ifmultiaddr *ifma;
3087 	struct ifmultiaddr *next;
3088 
3089 	IF_ADDR_LOCK(ifp);
3090 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3091 		if_delmulti_locked(ifp, ifma, 0);
3092 	IF_ADDR_UNLOCK(ifp);
3093 }
3094 
3095 /*
3096  * Delete a multicast group membership by group membership pointer.
3097  * Network-layer protocol domains must use this routine.
3098  *
3099  * It is safe to call this routine if the ifp disappeared.
3100  */
3101 void
3102 if_delmulti_ifma(struct ifmultiaddr *ifma)
3103 {
3104 	struct ifnet *ifp;
3105 	int lastref;
3106 
3107 	ifp = ifma->ifma_ifp;
3108 #ifdef DIAGNOSTIC
3109 	if (ifp == NULL) {
3110 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3111 	} else {
3112 		struct ifnet *oifp;
3113 
3114 		IFNET_RLOCK_NOSLEEP();
3115 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3116 			if (ifp == oifp)
3117 				break;
3118 		if (ifp != oifp) {
3119 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3120 			ifp = NULL;
3121 		}
3122 		IFNET_RUNLOCK_NOSLEEP();
3123 	}
3124 #endif
3125 	/*
3126 	 * If and only if the ifnet instance exists: Acquire the address lock.
3127 	 */
3128 	if (ifp != NULL)
3129 		IF_ADDR_LOCK(ifp);
3130 
3131 	lastref = if_delmulti_locked(ifp, ifma, 0);
3132 
3133 	if (ifp != NULL) {
3134 		/*
3135 		 * If and only if the ifnet instance exists:
3136 		 *  Release the address lock.
3137 		 *  If the group was left: update the hardware hash filter.
3138 		 */
3139 		IF_ADDR_UNLOCK(ifp);
3140 		if (lastref && ifp->if_ioctl != NULL) {
3141 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3142 		}
3143 	}
3144 }
3145 
3146 /*
3147  * Perform deletion of network-layer and/or link-layer multicast address.
3148  *
3149  * Return 0 if the reference count was decremented.
3150  * Return 1 if the final reference was released, indicating that the
3151  * hardware hash filter should be reprogrammed.
3152  */
3153 static int
3154 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3155 {
3156 	struct ifmultiaddr *ll_ifma;
3157 
3158 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3159 		KASSERT(ifma->ifma_ifp == ifp,
3160 		    ("%s: inconsistent ifp %p", __func__, ifp));
3161 		IF_ADDR_LOCK_ASSERT(ifp);
3162 	}
3163 
3164 	ifp = ifma->ifma_ifp;
3165 
3166 	/*
3167 	 * If the ifnet is detaching, null out references to ifnet,
3168 	 * so that upper protocol layers will notice, and not attempt
3169 	 * to obtain locks for an ifnet which no longer exists. The
3170 	 * routing socket announcement must happen before the ifnet
3171 	 * instance is detached from the system.
3172 	 */
3173 	if (detaching) {
3174 #ifdef DIAGNOSTIC
3175 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3176 #endif
3177 		/*
3178 		 * ifp may already be nulled out if we are being reentered
3179 		 * to delete the ll_ifma.
3180 		 */
3181 		if (ifp != NULL) {
3182 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3183 			ifma->ifma_ifp = NULL;
3184 		}
3185 	}
3186 
3187 	if (--ifma->ifma_refcount > 0)
3188 		return 0;
3189 
3190 	/*
3191 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3192 	 * have been associated with it. Release it first if so.
3193 	 */
3194 	ll_ifma = ifma->ifma_llifma;
3195 	if (ll_ifma != NULL) {
3196 		KASSERT(ifma->ifma_lladdr != NULL,
3197 		    ("%s: llifma w/o lladdr", __func__));
3198 		if (detaching)
3199 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3200 		if (--ll_ifma->ifma_refcount == 0) {
3201 			if (ifp != NULL) {
3202 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3203 				    ifma_link);
3204 			}
3205 			if_freemulti(ll_ifma);
3206 		}
3207 	}
3208 
3209 	if (ifp != NULL)
3210 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3211 
3212 	if_freemulti(ifma);
3213 
3214 	/*
3215 	 * The last reference to this instance of struct ifmultiaddr
3216 	 * was released; the hardware should be notified of this change.
3217 	 */
3218 	return 1;
3219 }
3220 
3221 /*
3222  * Set the link layer address on an interface.
3223  *
3224  * At this time we only support certain types of interfaces,
3225  * and we don't allow the length of the address to change.
3226  */
3227 int
3228 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3229 {
3230 	struct sockaddr_dl *sdl;
3231 	struct ifaddr *ifa;
3232 	struct ifreq ifr;
3233 
3234 	IF_ADDR_LOCK(ifp);
3235 	ifa = ifp->if_addr;
3236 	if (ifa == NULL) {
3237 		IF_ADDR_UNLOCK(ifp);
3238 		return (EINVAL);
3239 	}
3240 	ifa_ref(ifa);
3241 	IF_ADDR_UNLOCK(ifp);
3242 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3243 	if (sdl == NULL) {
3244 		ifa_free(ifa);
3245 		return (EINVAL);
3246 	}
3247 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3248 		ifa_free(ifa);
3249 		return (EINVAL);
3250 	}
3251 	switch (ifp->if_type) {
3252 	case IFT_ETHER:
3253 	case IFT_FDDI:
3254 	case IFT_XETHER:
3255 	case IFT_ISO88025:
3256 	case IFT_L2VLAN:
3257 	case IFT_BRIDGE:
3258 	case IFT_ARCNET:
3259 	case IFT_IEEE8023ADLAG:
3260 	case IFT_IEEE80211:
3261 		bcopy(lladdr, LLADDR(sdl), len);
3262 		ifa_free(ifa);
3263 		break;
3264 	default:
3265 		ifa_free(ifa);
3266 		return (ENODEV);
3267 	}
3268 
3269 	/*
3270 	 * If the interface is already up, we need
3271 	 * to re-init it in order to reprogram its
3272 	 * address filter.
3273 	 */
3274 	if ((ifp->if_flags & IFF_UP) != 0) {
3275 		if (ifp->if_ioctl) {
3276 			ifp->if_flags &= ~IFF_UP;
3277 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3278 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3279 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3280 			ifp->if_flags |= IFF_UP;
3281 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3282 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3283 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3284 		}
3285 #ifdef INET
3286 		/*
3287 		 * Also send gratuitous ARPs to notify other nodes about
3288 		 * the address change.
3289 		 */
3290 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3291 			if (ifa->ifa_addr->sa_family == AF_INET)
3292 				arp_ifinit(ifp, ifa);
3293 		}
3294 #endif
3295 	}
3296 	return (0);
3297 }
3298 
3299 /*
3300  * The name argument must be a pointer to storage which will last as
3301  * long as the interface does.  For physical devices, the result of
3302  * device_get_name(dev) is a good choice and for pseudo-devices a
3303  * static string works well.
3304  */
3305 void
3306 if_initname(struct ifnet *ifp, const char *name, int unit)
3307 {
3308 	ifp->if_dname = name;
3309 	ifp->if_dunit = unit;
3310 	if (unit != IF_DUNIT_NONE)
3311 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3312 	else
3313 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3314 }
3315 
3316 int
3317 if_printf(struct ifnet *ifp, const char * fmt, ...)
3318 {
3319 	va_list ap;
3320 	int retval;
3321 
3322 	retval = printf("%s: ", ifp->if_xname);
3323 	va_start(ap, fmt);
3324 	retval += vprintf(fmt, ap);
3325 	va_end(ap);
3326 	return (retval);
3327 }
3328 
3329 void
3330 if_start(struct ifnet *ifp)
3331 {
3332 
3333 	(*(ifp)->if_start)(ifp);
3334 }
3335 
3336 /*
3337  * Backwards compatibility interface for drivers
3338  * that have not implemented it
3339  */
3340 static int
3341 if_transmit(struct ifnet *ifp, struct mbuf *m)
3342 {
3343 	int error;
3344 
3345 	IFQ_HANDOFF(ifp, m, error);
3346 	return (error);
3347 }
3348 
3349 int
3350 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3351 {
3352 	int active = 0;
3353 
3354 	IF_LOCK(ifq);
3355 	if (_IF_QFULL(ifq)) {
3356 		_IF_DROP(ifq);
3357 		IF_UNLOCK(ifq);
3358 		m_freem(m);
3359 		return (0);
3360 	}
3361 	if (ifp != NULL) {
3362 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3363 		if (m->m_flags & (M_BCAST|M_MCAST))
3364 			ifp->if_omcasts++;
3365 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3366 	}
3367 	_IF_ENQUEUE(ifq, m);
3368 	IF_UNLOCK(ifq);
3369 	if (ifp != NULL && !active)
3370 		(*(ifp)->if_start)(ifp);
3371 	return (1);
3372 }
3373 
3374 void
3375 if_register_com_alloc(u_char type,
3376     if_com_alloc_t *a, if_com_free_t *f)
3377 {
3378 
3379 	KASSERT(if_com_alloc[type] == NULL,
3380 	    ("if_register_com_alloc: %d already registered", type));
3381 	KASSERT(if_com_free[type] == NULL,
3382 	    ("if_register_com_alloc: %d free already registered", type));
3383 
3384 	if_com_alloc[type] = a;
3385 	if_com_free[type] = f;
3386 }
3387 
3388 void
3389 if_deregister_com_alloc(u_char type)
3390 {
3391 
3392 	KASSERT(if_com_alloc[type] != NULL,
3393 	    ("if_deregister_com_alloc: %d not registered", type));
3394 	KASSERT(if_com_free[type] != NULL,
3395 	    ("if_deregister_com_alloc: %d free not registered", type));
3396 	if_com_alloc[type] = NULL;
3397 	if_com_free[type] = NULL;
3398 }
3399