xref: /freebsd/sys/net/if.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
1 /*-
2  * Copyright (c) 1980, 1986, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if.c	8.5 (Berkeley) 1/9/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/malloc.h>
41 #include <sys/sbuf.h>
42 #include <sys/bus.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/refcount.h>
53 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
60 #include <sys/jail.h>
61 #include <sys/priv.h>
62 
63 #include <machine/stdarg.h>
64 #include <vm/uma.h>
65 
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_clone.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/vnet.h>
75 
76 #if defined(INET) || defined(INET6)
77 /*XXX*/
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip_carp.h>
81 #ifdef INET6
82 #include <netinet6/in6_var.h>
83 #include <netinet6/in6_ifattach.h>
84 #endif
85 #endif
86 #ifdef INET
87 #include <netinet/if_ether.h>
88 #endif
89 
90 #include <security/mac/mac_framework.h>
91 
92 #ifdef COMPAT_FREEBSD32
93 #include <sys/mount.h>
94 #include <compat/freebsd32/freebsd32.h>
95 #endif
96 
97 struct ifindex_entry {
98 	struct  ifnet *ife_ifnet;
99 };
100 
101 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
102 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
103 
104 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
105 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
106     &ifqmaxlen, 0, "max send queue size");
107 
108 /* Log link state change events */
109 static int log_link_state_change = 1;
110 
111 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
112 	&log_link_state_change, 0,
113 	"log interface link state change events");
114 
115 /* Interface description */
116 static unsigned int ifdescr_maxlen = 1024;
117 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
118 	&ifdescr_maxlen, 0,
119 	"administrative maximum length for interface description");
120 
121 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
122 
123 /* global sx for non-critical path ifdescr */
124 static struct sx ifdescr_sx;
125 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
126 
127 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
128 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
129 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
130 /* These are external hooks for CARP. */
131 void	(*carp_linkstate_p)(struct ifnet *ifp);
132 void	(*carp_demote_adj_p)(int, char *);
133 #if defined(INET) || defined(INET6)
134 int	(*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
135 int	(*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
136     struct sockaddr *sa);
137 int	(*carp_ioctl_p)(struct ifreq *, u_long, struct thread *);
138 int	(*carp_attach_p)(struct ifaddr *, int);
139 void	(*carp_detach_p)(struct ifaddr *);
140 #endif
141 #ifdef INET
142 int	(*carp_iamatch_p)(struct ifaddr *, uint8_t **);
143 #endif
144 #ifdef INET6
145 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
146 caddr_t	(*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
147     const struct in6_addr *taddr);
148 #endif
149 
150 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
151 
152 /*
153  * XXX: Style; these should be sorted alphabetically, and unprototyped
154  * static functions should be prototyped. Currently they are sorted by
155  * declaration order.
156  */
157 static void	if_attachdomain(void *);
158 static void	if_attachdomain1(struct ifnet *);
159 static int	ifconf(u_long, caddr_t);
160 static void	if_freemulti(struct ifmultiaddr *);
161 static void	if_init(void *);
162 static void	if_grow(void);
163 static void	if_route(struct ifnet *, int flag, int fam);
164 static int	if_setflag(struct ifnet *, int, int, int *, int);
165 static int	if_transmit(struct ifnet *ifp, struct mbuf *m);
166 static void	if_unroute(struct ifnet *, int flag, int fam);
167 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
168 static int	if_rtdel(struct radix_node *, void *);
169 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
170 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
171 static void	do_link_state_change(void *, int);
172 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
173 static int	if_getgroupmembers(struct ifgroupreq *);
174 static void	if_delgroups(struct ifnet *);
175 static void	if_attach_internal(struct ifnet *, int);
176 static void	if_detach_internal(struct ifnet *, int);
177 
178 #ifdef INET6
179 /*
180  * XXX: declare here to avoid to include many inet6 related files..
181  * should be more generalized?
182  */
183 extern void	nd6_setmtu(struct ifnet *);
184 #endif
185 
186 VNET_DEFINE(int, if_index);
187 int	ifqmaxlen = IFQ_MAXLEN;
188 VNET_DEFINE(struct ifnethead, ifnet);	/* depend on static init XXX */
189 VNET_DEFINE(struct ifgrouphead, ifg_head);
190 
191 static VNET_DEFINE(int, if_indexlim) = 8;
192 
193 /* Table of ifnet by index. */
194 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
195 
196 #define	V_if_indexlim		VNET(if_indexlim)
197 #define	V_ifindex_table		VNET(ifindex_table)
198 
199 /*
200  * The global network interface list (V_ifnet) and related state (such as
201  * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
202  * an rwlock.  Either may be acquired shared to stablize the list, but both
203  * must be acquired writable to modify the list.  This model allows us to
204  * both stablize the interface list during interrupt thread processing, but
205  * also to stablize it over long-running ioctls, without introducing priority
206  * inversions and deadlocks.
207  */
208 struct rwlock ifnet_rwlock;
209 struct sx ifnet_sxlock;
210 
211 /*
212  * The allocation of network interfaces is a rather non-atomic affair; we
213  * need to select an index before we are ready to expose the interface for
214  * use, so will use this pointer value to indicate reservation.
215  */
216 #define	IFNET_HOLD	(void *)(uintptr_t)(-1)
217 
218 static	if_com_alloc_t *if_com_alloc[256];
219 static	if_com_free_t *if_com_free[256];
220 
221 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
222 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
223 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
224 
225 struct ifnet *
226 ifnet_byindex_locked(u_short idx)
227 {
228 
229 	if (idx > V_if_index)
230 		return (NULL);
231 	if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
232 		return (NULL);
233 	return (V_ifindex_table[idx].ife_ifnet);
234 }
235 
236 struct ifnet *
237 ifnet_byindex(u_short idx)
238 {
239 	struct ifnet *ifp;
240 
241 	IFNET_RLOCK_NOSLEEP();
242 	ifp = ifnet_byindex_locked(idx);
243 	IFNET_RUNLOCK_NOSLEEP();
244 	return (ifp);
245 }
246 
247 struct ifnet *
248 ifnet_byindex_ref(u_short idx)
249 {
250 	struct ifnet *ifp;
251 
252 	IFNET_RLOCK_NOSLEEP();
253 	ifp = ifnet_byindex_locked(idx);
254 	if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
255 		IFNET_RUNLOCK_NOSLEEP();
256 		return (NULL);
257 	}
258 	if_ref(ifp);
259 	IFNET_RUNLOCK_NOSLEEP();
260 	return (ifp);
261 }
262 
263 /*
264  * Allocate an ifindex array entry; return 0 on success or an error on
265  * failure.
266  */
267 static int
268 ifindex_alloc_locked(u_short *idxp)
269 {
270 	u_short idx;
271 
272 	IFNET_WLOCK_ASSERT();
273 
274 retry:
275 	/*
276 	 * Try to find an empty slot below V_if_index.  If we fail, take the
277 	 * next slot.
278 	 */
279 	for (idx = 1; idx <= V_if_index; idx++) {
280 		if (V_ifindex_table[idx].ife_ifnet == NULL)
281 			break;
282 	}
283 
284 	/* Catch if_index overflow. */
285 	if (idx < 1)
286 		return (ENOSPC);
287 	if (idx >= V_if_indexlim) {
288 		if_grow();
289 		goto retry;
290 	}
291 	if (idx > V_if_index)
292 		V_if_index = idx;
293 	*idxp = idx;
294 	return (0);
295 }
296 
297 static void
298 ifindex_free_locked(u_short idx)
299 {
300 
301 	IFNET_WLOCK_ASSERT();
302 
303 	V_ifindex_table[idx].ife_ifnet = NULL;
304 	while (V_if_index > 0 &&
305 	    V_ifindex_table[V_if_index].ife_ifnet == NULL)
306 		V_if_index--;
307 }
308 
309 static void
310 ifindex_free(u_short idx)
311 {
312 
313 	IFNET_WLOCK();
314 	ifindex_free_locked(idx);
315 	IFNET_WUNLOCK();
316 }
317 
318 static void
319 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
320 {
321 
322 	IFNET_WLOCK_ASSERT();
323 
324 	V_ifindex_table[idx].ife_ifnet = ifp;
325 }
326 
327 static void
328 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
329 {
330 
331 	IFNET_WLOCK();
332 	ifnet_setbyindex_locked(idx, ifp);
333 	IFNET_WUNLOCK();
334 }
335 
336 struct ifaddr *
337 ifaddr_byindex(u_short idx)
338 {
339 	struct ifaddr *ifa;
340 
341 	IFNET_RLOCK_NOSLEEP();
342 	ifa = ifnet_byindex_locked(idx)->if_addr;
343 	if (ifa != NULL)
344 		ifa_ref(ifa);
345 	IFNET_RUNLOCK_NOSLEEP();
346 	return (ifa);
347 }
348 
349 /*
350  * Network interface utility routines.
351  *
352  * Routines with ifa_ifwith* names take sockaddr *'s as
353  * parameters.
354  */
355 
356 static void
357 vnet_if_init(const void *unused __unused)
358 {
359 
360 	TAILQ_INIT(&V_ifnet);
361 	TAILQ_INIT(&V_ifg_head);
362 	IFNET_WLOCK();
363 	if_grow();				/* create initial table */
364 	IFNET_WUNLOCK();
365 	vnet_if_clone_init();
366 }
367 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
368     NULL);
369 
370 /* ARGSUSED*/
371 static void
372 if_init(void *dummy __unused)
373 {
374 
375 	IFNET_LOCK_INIT();
376 	if_clone_init();
377 }
378 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
379 
380 
381 #ifdef VIMAGE
382 static void
383 vnet_if_uninit(const void *unused __unused)
384 {
385 
386 	VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
387 	    "not empty", __func__, __LINE__, &V_ifnet));
388 	VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
389 	    "not empty", __func__, __LINE__, &V_ifg_head));
390 
391 	free((caddr_t)V_ifindex_table, M_IFNET);
392 }
393 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
394     vnet_if_uninit, NULL);
395 #endif
396 
397 static void
398 if_grow(void)
399 {
400 	int oldlim;
401 	u_int n;
402 	struct ifindex_entry *e;
403 
404 	IFNET_WLOCK_ASSERT();
405 	oldlim = V_if_indexlim;
406 	IFNET_WUNLOCK();
407 	n = (oldlim << 1) * sizeof(*e);
408 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
409 	IFNET_WLOCK();
410 	if (V_if_indexlim != oldlim) {
411 		free(e, M_IFNET);
412 		return;
413 	}
414 	if (V_ifindex_table != NULL) {
415 		memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
416 		free((caddr_t)V_ifindex_table, M_IFNET);
417 	}
418 	V_if_indexlim <<= 1;
419 	V_ifindex_table = e;
420 }
421 
422 /*
423  * Allocate a struct ifnet and an index for an interface.  A layer 2
424  * common structure will also be allocated if an allocation routine is
425  * registered for the passed type.
426  */
427 struct ifnet *
428 if_alloc(u_char type)
429 {
430 	struct ifnet *ifp;
431 	u_short idx;
432 
433 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
434 	IFNET_WLOCK();
435 	if (ifindex_alloc_locked(&idx) != 0) {
436 		IFNET_WUNLOCK();
437 		free(ifp, M_IFNET);
438 		return (NULL);
439 	}
440 	ifnet_setbyindex_locked(idx, IFNET_HOLD);
441 	IFNET_WUNLOCK();
442 	ifp->if_index = idx;
443 	ifp->if_type = type;
444 	ifp->if_alloctype = type;
445 	if (if_com_alloc[type] != NULL) {
446 		ifp->if_l2com = if_com_alloc[type](type, ifp);
447 		if (ifp->if_l2com == NULL) {
448 			free(ifp, M_IFNET);
449 			ifindex_free(idx);
450 			return (NULL);
451 		}
452 	}
453 
454 	IF_ADDR_LOCK_INIT(ifp);
455 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
456 	ifp->if_afdata_initialized = 0;
457 	IF_AFDATA_LOCK_INIT(ifp);
458 	TAILQ_INIT(&ifp->if_addrhead);
459 	TAILQ_INIT(&ifp->if_prefixhead);
460 	TAILQ_INIT(&ifp->if_multiaddrs);
461 	TAILQ_INIT(&ifp->if_groups);
462 #ifdef MAC
463 	mac_ifnet_init(ifp);
464 #endif
465 	ifq_init(&ifp->if_snd, ifp);
466 
467 	refcount_init(&ifp->if_refcount, 1);	/* Index reference. */
468 	ifnet_setbyindex(ifp->if_index, ifp);
469 	return (ifp);
470 }
471 
472 /*
473  * Do the actual work of freeing a struct ifnet, and layer 2 common
474  * structure.  This call is made when the last reference to an
475  * interface is released.
476  */
477 static void
478 if_free_internal(struct ifnet *ifp)
479 {
480 
481 	KASSERT((ifp->if_flags & IFF_DYING),
482 	    ("if_free_internal: interface not dying"));
483 
484 	if (if_com_free[ifp->if_alloctype] != NULL)
485 		if_com_free[ifp->if_alloctype](ifp->if_l2com,
486 		    ifp->if_alloctype);
487 
488 #ifdef MAC
489 	mac_ifnet_destroy(ifp);
490 #endif /* MAC */
491 	if (ifp->if_description != NULL)
492 		free(ifp->if_description, M_IFDESCR);
493 	IF_AFDATA_DESTROY(ifp);
494 	IF_ADDR_LOCK_DESTROY(ifp);
495 	ifq_delete(&ifp->if_snd);
496 	free(ifp, M_IFNET);
497 }
498 
499 /*
500  * Deregister an interface and free the associated storage.
501  */
502 void
503 if_free(struct ifnet *ifp)
504 {
505 
506 	ifp->if_flags |= IFF_DYING;			/* XXX: Locking */
507 
508 	IFNET_WLOCK();
509 	KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
510 	    ("%s: freeing unallocated ifnet", ifp->if_xname));
511 
512 	ifindex_free_locked(ifp->if_index);
513 	IFNET_WUNLOCK();
514 
515 	if (!refcount_release(&ifp->if_refcount))
516 		return;
517 	if_free_internal(ifp);
518 }
519 
520 /*
521  * Interfaces to keep an ifnet type-stable despite the possibility of the
522  * driver calling if_free().  If there are additional references, we defer
523  * freeing the underlying data structure.
524  */
525 void
526 if_ref(struct ifnet *ifp)
527 {
528 
529 	/* We don't assert the ifnet list lock here, but arguably should. */
530 	refcount_acquire(&ifp->if_refcount);
531 }
532 
533 void
534 if_rele(struct ifnet *ifp)
535 {
536 
537 	if (!refcount_release(&ifp->if_refcount))
538 		return;
539 	if_free_internal(ifp);
540 }
541 
542 void
543 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
544 {
545 
546 	mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
547 
548 	if (ifq->ifq_maxlen == 0)
549 		ifq->ifq_maxlen = ifqmaxlen;
550 
551 	ifq->altq_type = 0;
552 	ifq->altq_disc = NULL;
553 	ifq->altq_flags &= ALTQF_CANTCHANGE;
554 	ifq->altq_tbr  = NULL;
555 	ifq->altq_ifp  = ifp;
556 }
557 
558 void
559 ifq_delete(struct ifaltq *ifq)
560 {
561 	mtx_destroy(&ifq->ifq_mtx);
562 }
563 
564 /*
565  * Perform generic interface initalization tasks and attach the interface
566  * to the list of "active" interfaces.  If vmove flag is set on entry
567  * to if_attach_internal(), perform only a limited subset of initialization
568  * tasks, given that we are moving from one vnet to another an ifnet which
569  * has already been fully initialized.
570  *
571  * XXX:
572  *  - The decision to return void and thus require this function to
573  *    succeed is questionable.
574  *  - We should probably do more sanity checking.  For instance we don't
575  *    do anything to insure if_xname is unique or non-empty.
576  */
577 void
578 if_attach(struct ifnet *ifp)
579 {
580 
581 	if_attach_internal(ifp, 0);
582 }
583 
584 static void
585 if_attach_internal(struct ifnet *ifp, int vmove)
586 {
587 	unsigned socksize, ifasize;
588 	int namelen, masklen;
589 	struct sockaddr_dl *sdl;
590 	struct ifaddr *ifa;
591 
592 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
593 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
594 		    ifp->if_xname);
595 
596 #ifdef VIMAGE
597 	ifp->if_vnet = curvnet;
598 	if (ifp->if_home_vnet == NULL)
599 		ifp->if_home_vnet = curvnet;
600 #endif
601 
602 	if_addgroup(ifp, IFG_ALL);
603 
604 	getmicrotime(&ifp->if_lastchange);
605 	ifp->if_data.ifi_epoch = time_uptime;
606 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
607 
608 	KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
609 	    (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
610 	    ("transmit and qflush must both either be set or both be NULL"));
611 	if (ifp->if_transmit == NULL) {
612 		ifp->if_transmit = if_transmit;
613 		ifp->if_qflush = if_qflush;
614 	}
615 
616 	if (!vmove) {
617 #ifdef MAC
618 		mac_ifnet_create(ifp);
619 #endif
620 
621 		/*
622 		 * Create a Link Level name for this device.
623 		 */
624 		namelen = strlen(ifp->if_xname);
625 		/*
626 		 * Always save enough space for any possiable name so we
627 		 * can do a rename in place later.
628 		 */
629 		masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
630 		socksize = masklen + ifp->if_addrlen;
631 		if (socksize < sizeof(*sdl))
632 			socksize = sizeof(*sdl);
633 		socksize = roundup2(socksize, sizeof(long));
634 		ifasize = sizeof(*ifa) + 2 * socksize;
635 		ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
636 		ifa_init(ifa);
637 		sdl = (struct sockaddr_dl *)(ifa + 1);
638 		sdl->sdl_len = socksize;
639 		sdl->sdl_family = AF_LINK;
640 		bcopy(ifp->if_xname, sdl->sdl_data, namelen);
641 		sdl->sdl_nlen = namelen;
642 		sdl->sdl_index = ifp->if_index;
643 		sdl->sdl_type = ifp->if_type;
644 		ifp->if_addr = ifa;
645 		ifa->ifa_ifp = ifp;
646 		ifa->ifa_rtrequest = link_rtrequest;
647 		ifa->ifa_addr = (struct sockaddr *)sdl;
648 		sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
649 		ifa->ifa_netmask = (struct sockaddr *)sdl;
650 		sdl->sdl_len = masklen;
651 		while (namelen != 0)
652 			sdl->sdl_data[--namelen] = 0xff;
653 		TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
654 		/* Reliably crash if used uninitialized. */
655 		ifp->if_broadcastaddr = NULL;
656 	}
657 #ifdef VIMAGE
658 	else {
659 		/*
660 		 * Update the interface index in the link layer address
661 		 * of the interface.
662 		 */
663 		for (ifa = ifp->if_addr; ifa != NULL;
664 		    ifa = TAILQ_NEXT(ifa, ifa_link)) {
665 			if (ifa->ifa_addr->sa_family == AF_LINK) {
666 				sdl = (struct sockaddr_dl *)ifa->ifa_addr;
667 				sdl->sdl_index = ifp->if_index;
668 			}
669 		}
670 	}
671 #endif
672 
673 	IFNET_WLOCK();
674 	TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
675 #ifdef VIMAGE
676 	curvnet->vnet_ifcnt++;
677 #endif
678 	IFNET_WUNLOCK();
679 
680 	if (domain_init_status >= 2)
681 		if_attachdomain1(ifp);
682 
683 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
684 	if (IS_DEFAULT_VNET(curvnet))
685 		devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
686 
687 	/* Announce the interface. */
688 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
689 }
690 
691 static void
692 if_attachdomain(void *dummy)
693 {
694 	struct ifnet *ifp;
695 	int s;
696 
697 	s = splnet();
698 	TAILQ_FOREACH(ifp, &V_ifnet, if_link)
699 		if_attachdomain1(ifp);
700 	splx(s);
701 }
702 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
703     if_attachdomain, NULL);
704 
705 static void
706 if_attachdomain1(struct ifnet *ifp)
707 {
708 	struct domain *dp;
709 	int s;
710 
711 	s = splnet();
712 
713 	/*
714 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
715 	 * cannot lock ifp->if_afdata initialization, entirely.
716 	 */
717 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
718 		splx(s);
719 		return;
720 	}
721 	if (ifp->if_afdata_initialized >= domain_init_status) {
722 		IF_AFDATA_UNLOCK(ifp);
723 		splx(s);
724 		printf("if_attachdomain called more than once on %s\n",
725 		    ifp->if_xname);
726 		return;
727 	}
728 	ifp->if_afdata_initialized = domain_init_status;
729 	IF_AFDATA_UNLOCK(ifp);
730 
731 	/* address family dependent data region */
732 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
733 	for (dp = domains; dp; dp = dp->dom_next) {
734 		if (dp->dom_ifattach)
735 			ifp->if_afdata[dp->dom_family] =
736 			    (*dp->dom_ifattach)(ifp);
737 	}
738 
739 	splx(s);
740 }
741 
742 /*
743  * Remove any unicast or broadcast network addresses from an interface.
744  */
745 void
746 if_purgeaddrs(struct ifnet *ifp)
747 {
748 	struct ifaddr *ifa, *next;
749 
750 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
751 		if (ifa->ifa_addr->sa_family == AF_LINK)
752 			continue;
753 #ifdef INET
754 		/* XXX: Ugly!! ad hoc just for INET */
755 		if (ifa->ifa_addr->sa_family == AF_INET) {
756 			struct ifaliasreq ifr;
757 
758 			bzero(&ifr, sizeof(ifr));
759 			ifr.ifra_addr = *ifa->ifa_addr;
760 			if (ifa->ifa_dstaddr)
761 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
762 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
763 			    NULL) == 0)
764 				continue;
765 		}
766 #endif /* INET */
767 #ifdef INET6
768 		if (ifa->ifa_addr->sa_family == AF_INET6) {
769 			in6_purgeaddr(ifa);
770 			/* ifp_addrhead is already updated */
771 			continue;
772 		}
773 #endif /* INET6 */
774 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
775 		ifa_free(ifa);
776 	}
777 }
778 
779 /*
780  * Remove any multicast network addresses from an interface when an ifnet
781  * is going away.
782  */
783 static void
784 if_purgemaddrs(struct ifnet *ifp)
785 {
786 	struct ifmultiaddr *ifma;
787 	struct ifmultiaddr *next;
788 
789 	IF_ADDR_WLOCK(ifp);
790 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
791 		if_delmulti_locked(ifp, ifma, 1);
792 	IF_ADDR_WUNLOCK(ifp);
793 }
794 
795 /*
796  * Detach an interface, removing it from the list of "active" interfaces.
797  * If vmove flag is set on entry to if_detach_internal(), perform only a
798  * limited subset of cleanup tasks, given that we are moving an ifnet from
799  * one vnet to another, where it must be fully operational.
800  *
801  * XXXRW: There are some significant questions about event ordering, and
802  * how to prevent things from starting to use the interface during detach.
803  */
804 void
805 if_detach(struct ifnet *ifp)
806 {
807 
808 	if_detach_internal(ifp, 0);
809 }
810 
811 static void
812 if_detach_internal(struct ifnet *ifp, int vmove)
813 {
814 	struct ifaddr *ifa;
815 	struct radix_node_head	*rnh;
816 	int i, j;
817 	struct domain *dp;
818  	struct ifnet *iter;
819  	int found = 0;
820 
821 	IFNET_WLOCK();
822 	TAILQ_FOREACH(iter, &V_ifnet, if_link)
823 		if (iter == ifp) {
824 			TAILQ_REMOVE(&V_ifnet, ifp, if_link);
825 			found = 1;
826 			break;
827 		}
828 #ifdef VIMAGE
829 	if (found)
830 		curvnet->vnet_ifcnt--;
831 #endif
832 	IFNET_WUNLOCK();
833 	if (!found) {
834 		if (vmove)
835 			panic("%s: ifp=%p not on the ifnet tailq %p",
836 			    __func__, ifp, &V_ifnet);
837 		else
838 			return; /* XXX this should panic as well? */
839 	}
840 
841 	/*
842 	 * Remove/wait for pending events.
843 	 */
844 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
845 
846 	/*
847 	 * Remove routes and flush queues.
848 	 */
849 	if_down(ifp);
850 #ifdef ALTQ
851 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
852 		altq_disable(&ifp->if_snd);
853 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
854 		altq_detach(&ifp->if_snd);
855 #endif
856 
857 	if_purgeaddrs(ifp);
858 
859 #ifdef INET
860 	in_ifdetach(ifp);
861 #endif
862 
863 #ifdef INET6
864 	/*
865 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
866 	 * before removing routing entries below, since IPv6 interface direct
867 	 * routes are expected to be removed by the IPv6-specific kernel API.
868 	 * Otherwise, the kernel will detect some inconsistency and bark it.
869 	 */
870 	in6_ifdetach(ifp);
871 #endif
872 	if_purgemaddrs(ifp);
873 
874 	if (!vmove) {
875 		/*
876 		 * Prevent further calls into the device driver via ifnet.
877 		 */
878 		if_dead(ifp);
879 
880 		/*
881 		 * Remove link ifaddr pointer and maybe decrement if_index.
882 		 * Clean up all addresses.
883 		 */
884 		ifp->if_addr = NULL;
885 
886 		/* We can now free link ifaddr. */
887 		if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
888 			ifa = TAILQ_FIRST(&ifp->if_addrhead);
889 			TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
890 			ifa_free(ifa);
891 		}
892 	}
893 
894 	/*
895 	 * Delete all remaining routes using this interface
896 	 * Unfortuneatly the only way to do this is to slog through
897 	 * the entire routing table looking for routes which point
898 	 * to this interface...oh well...
899 	 */
900 	for (i = 1; i <= AF_MAX; i++) {
901 		for (j = 0; j < rt_numfibs; j++) {
902 			rnh = rt_tables_get_rnh(j, i);
903 			if (rnh == NULL)
904 				continue;
905 			RADIX_NODE_HEAD_LOCK(rnh);
906 			(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
907 			RADIX_NODE_HEAD_UNLOCK(rnh);
908 		}
909 	}
910 
911 	/* Announce that the interface is gone. */
912 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
913 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
914 	if (IS_DEFAULT_VNET(curvnet))
915 		devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
916 	if_delgroups(ifp);
917 
918 	/*
919 	 * We cannot hold the lock over dom_ifdetach calls as they might
920 	 * sleep, for example trying to drain a callout, thus open up the
921 	 * theoretical race with re-attaching.
922 	 */
923 	IF_AFDATA_LOCK(ifp);
924 	i = ifp->if_afdata_initialized;
925 	ifp->if_afdata_initialized = 0;
926 	IF_AFDATA_UNLOCK(ifp);
927 	for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
928 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
929 			(*dp->dom_ifdetach)(ifp,
930 			    ifp->if_afdata[dp->dom_family]);
931 	}
932 }
933 
934 #ifdef VIMAGE
935 /*
936  * if_vmove() performs a limited version of if_detach() in current
937  * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
938  * An attempt is made to shrink if_index in current vnet, find an
939  * unused if_index in target vnet and calls if_grow() if necessary,
940  * and finally find an unused if_xname for the target vnet.
941  */
942 void
943 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
944 {
945 	u_short idx;
946 
947 	/*
948 	 * Detach from current vnet, but preserve LLADDR info, do not
949 	 * mark as dead etc. so that the ifnet can be reattached later.
950 	 */
951 	if_detach_internal(ifp, 1);
952 
953 	/*
954 	 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
955 	 * the if_index for that vnet if possible.
956 	 *
957 	 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
958 	 * or we'd lock on one vnet and unlock on another.
959 	 */
960 	IFNET_WLOCK();
961 	ifindex_free_locked(ifp->if_index);
962 	IFNET_WUNLOCK();
963 
964 	/*
965 	 * Perform interface-specific reassignment tasks, if provided by
966 	 * the driver.
967 	 */
968 	if (ifp->if_reassign != NULL)
969 		ifp->if_reassign(ifp, new_vnet, NULL);
970 
971 	/*
972 	 * Switch to the context of the target vnet.
973 	 */
974 	CURVNET_SET_QUIET(new_vnet);
975 
976 	IFNET_WLOCK();
977 	if (ifindex_alloc_locked(&idx) != 0) {
978 		IFNET_WUNLOCK();
979 		panic("if_index overflow");
980 	}
981 	ifp->if_index = idx;
982 	ifnet_setbyindex_locked(ifp->if_index, ifp);
983 	IFNET_WUNLOCK();
984 
985 	if_attach_internal(ifp, 1);
986 
987 	CURVNET_RESTORE();
988 }
989 
990 /*
991  * Move an ifnet to or from another child prison/vnet, specified by the jail id.
992  */
993 static int
994 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
995 {
996 	struct prison *pr;
997 	struct ifnet *difp;
998 
999 	/* Try to find the prison within our visibility. */
1000 	sx_slock(&allprison_lock);
1001 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1002 	sx_sunlock(&allprison_lock);
1003 	if (pr == NULL)
1004 		return (ENXIO);
1005 	prison_hold_locked(pr);
1006 	mtx_unlock(&pr->pr_mtx);
1007 
1008 	/* Do not try to move the iface from and to the same prison. */
1009 	if (pr->pr_vnet == ifp->if_vnet) {
1010 		prison_free(pr);
1011 		return (EEXIST);
1012 	}
1013 
1014 	/* Make sure the named iface does not exists in the dst. prison/vnet. */
1015 	/* XXX Lock interfaces to avoid races. */
1016 	CURVNET_SET_QUIET(pr->pr_vnet);
1017 	difp = ifunit(ifname);
1018 	CURVNET_RESTORE();
1019 	if (difp != NULL) {
1020 		prison_free(pr);
1021 		return (EEXIST);
1022 	}
1023 
1024 	/* Move the interface into the child jail/vnet. */
1025 	if_vmove(ifp, pr->pr_vnet);
1026 
1027 	/* Report the new if_xname back to the userland. */
1028 	sprintf(ifname, "%s", ifp->if_xname);
1029 
1030 	prison_free(pr);
1031 	return (0);
1032 }
1033 
1034 static int
1035 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1036 {
1037 	struct prison *pr;
1038 	struct vnet *vnet_dst;
1039 	struct ifnet *ifp;
1040 
1041 	/* Try to find the prison within our visibility. */
1042 	sx_slock(&allprison_lock);
1043 	pr = prison_find_child(td->td_ucred->cr_prison, jid);
1044 	sx_sunlock(&allprison_lock);
1045 	if (pr == NULL)
1046 		return (ENXIO);
1047 	prison_hold_locked(pr);
1048 	mtx_unlock(&pr->pr_mtx);
1049 
1050 	/* Make sure the named iface exists in the source prison/vnet. */
1051 	CURVNET_SET(pr->pr_vnet);
1052 	ifp = ifunit(ifname);		/* XXX Lock to avoid races. */
1053 	if (ifp == NULL) {
1054 		CURVNET_RESTORE();
1055 		prison_free(pr);
1056 		return (ENXIO);
1057 	}
1058 
1059 	/* Do not try to move the iface from and to the same prison. */
1060 	vnet_dst = TD_TO_VNET(td);
1061 	if (vnet_dst == ifp->if_vnet) {
1062 		CURVNET_RESTORE();
1063 		prison_free(pr);
1064 		return (EEXIST);
1065 	}
1066 
1067 	/* Get interface back from child jail/vnet. */
1068 	if_vmove(ifp, vnet_dst);
1069 	CURVNET_RESTORE();
1070 
1071 	/* Report the new if_xname back to the userland. */
1072 	sprintf(ifname, "%s", ifp->if_xname);
1073 
1074 	prison_free(pr);
1075 	return (0);
1076 }
1077 #endif /* VIMAGE */
1078 
1079 /*
1080  * Add a group to an interface
1081  */
1082 int
1083 if_addgroup(struct ifnet *ifp, const char *groupname)
1084 {
1085 	struct ifg_list		*ifgl;
1086 	struct ifg_group	*ifg = NULL;
1087 	struct ifg_member	*ifgm;
1088 
1089 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1090 	    groupname[strlen(groupname) - 1] <= '9')
1091 		return (EINVAL);
1092 
1093 	IFNET_WLOCK();
1094 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1095 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1096 			IFNET_WUNLOCK();
1097 			return (EEXIST);
1098 		}
1099 
1100 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1101 	    M_NOWAIT)) == NULL) {
1102 	    	IFNET_WUNLOCK();
1103 		return (ENOMEM);
1104 	}
1105 
1106 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1107 	    M_TEMP, M_NOWAIT)) == NULL) {
1108 		free(ifgl, M_TEMP);
1109 		IFNET_WUNLOCK();
1110 		return (ENOMEM);
1111 	}
1112 
1113 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1114 		if (!strcmp(ifg->ifg_group, groupname))
1115 			break;
1116 
1117 	if (ifg == NULL) {
1118 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1119 		    M_TEMP, M_NOWAIT)) == NULL) {
1120 			free(ifgl, M_TEMP);
1121 			free(ifgm, M_TEMP);
1122 			IFNET_WUNLOCK();
1123 			return (ENOMEM);
1124 		}
1125 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1126 		ifg->ifg_refcnt = 0;
1127 		TAILQ_INIT(&ifg->ifg_members);
1128 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
1129 		TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1130 	}
1131 
1132 	ifg->ifg_refcnt++;
1133 	ifgl->ifgl_group = ifg;
1134 	ifgm->ifgm_ifp = ifp;
1135 
1136 	IF_ADDR_WLOCK(ifp);
1137 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1138 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1139 	IF_ADDR_WUNLOCK(ifp);
1140 
1141 	IFNET_WUNLOCK();
1142 
1143 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1144 
1145 	return (0);
1146 }
1147 
1148 /*
1149  * Remove a group from an interface
1150  */
1151 int
1152 if_delgroup(struct ifnet *ifp, const char *groupname)
1153 {
1154 	struct ifg_list		*ifgl;
1155 	struct ifg_member	*ifgm;
1156 
1157 	IFNET_WLOCK();
1158 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1159 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1160 			break;
1161 	if (ifgl == NULL) {
1162 		IFNET_WUNLOCK();
1163 		return (ENOENT);
1164 	}
1165 
1166 	IF_ADDR_WLOCK(ifp);
1167 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1168 	IF_ADDR_WUNLOCK(ifp);
1169 
1170 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1171 		if (ifgm->ifgm_ifp == ifp)
1172 			break;
1173 
1174 	if (ifgm != NULL) {
1175 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1176 		free(ifgm, M_TEMP);
1177 	}
1178 
1179 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1180 		TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1181 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1182 		free(ifgl->ifgl_group, M_TEMP);
1183 	}
1184 	IFNET_WUNLOCK();
1185 
1186 	free(ifgl, M_TEMP);
1187 
1188 	EVENTHANDLER_INVOKE(group_change_event, groupname);
1189 
1190 	return (0);
1191 }
1192 
1193 /*
1194  * Remove an interface from all groups
1195  */
1196 static void
1197 if_delgroups(struct ifnet *ifp)
1198 {
1199 	struct ifg_list		*ifgl;
1200 	struct ifg_member	*ifgm;
1201 	char groupname[IFNAMSIZ];
1202 
1203 	IFNET_WLOCK();
1204 	while (!TAILQ_EMPTY(&ifp->if_groups)) {
1205 		ifgl = TAILQ_FIRST(&ifp->if_groups);
1206 
1207 		strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1208 
1209 		IF_ADDR_WLOCK(ifp);
1210 		TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1211 		IF_ADDR_WUNLOCK(ifp);
1212 
1213 		TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1214 			if (ifgm->ifgm_ifp == ifp)
1215 				break;
1216 
1217 		if (ifgm != NULL) {
1218 			TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1219 			    ifgm_next);
1220 			free(ifgm, M_TEMP);
1221 		}
1222 
1223 		if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1224 			TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1225 			EVENTHANDLER_INVOKE(group_detach_event,
1226 			    ifgl->ifgl_group);
1227 			free(ifgl->ifgl_group, M_TEMP);
1228 		}
1229 		IFNET_WUNLOCK();
1230 
1231 		free(ifgl, M_TEMP);
1232 
1233 		EVENTHANDLER_INVOKE(group_change_event, groupname);
1234 
1235 		IFNET_WLOCK();
1236 	}
1237 	IFNET_WUNLOCK();
1238 }
1239 
1240 /*
1241  * Stores all groups from an interface in memory pointed
1242  * to by data
1243  */
1244 static int
1245 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1246 {
1247 	int			 len, error;
1248 	struct ifg_list		*ifgl;
1249 	struct ifg_req		 ifgrq, *ifgp;
1250 	struct ifgroupreq	*ifgr = data;
1251 
1252 	if (ifgr->ifgr_len == 0) {
1253 		IF_ADDR_RLOCK(ifp);
1254 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1255 			ifgr->ifgr_len += sizeof(struct ifg_req);
1256 		IF_ADDR_RUNLOCK(ifp);
1257 		return (0);
1258 	}
1259 
1260 	len = ifgr->ifgr_len;
1261 	ifgp = ifgr->ifgr_groups;
1262 	/* XXX: wire */
1263 	IF_ADDR_RLOCK(ifp);
1264 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1265 		if (len < sizeof(ifgrq)) {
1266 			IF_ADDR_RUNLOCK(ifp);
1267 			return (EINVAL);
1268 		}
1269 		bzero(&ifgrq, sizeof ifgrq);
1270 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1271 		    sizeof(ifgrq.ifgrq_group));
1272 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1273 		    	IF_ADDR_RUNLOCK(ifp);
1274 			return (error);
1275 		}
1276 		len -= sizeof(ifgrq);
1277 		ifgp++;
1278 	}
1279 	IF_ADDR_RUNLOCK(ifp);
1280 
1281 	return (0);
1282 }
1283 
1284 /*
1285  * Stores all members of a group in memory pointed to by data
1286  */
1287 static int
1288 if_getgroupmembers(struct ifgroupreq *data)
1289 {
1290 	struct ifgroupreq	*ifgr = data;
1291 	struct ifg_group	*ifg;
1292 	struct ifg_member	*ifgm;
1293 	struct ifg_req		 ifgrq, *ifgp;
1294 	int			 len, error;
1295 
1296 	IFNET_RLOCK();
1297 	TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1298 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1299 			break;
1300 	if (ifg == NULL) {
1301 		IFNET_RUNLOCK();
1302 		return (ENOENT);
1303 	}
1304 
1305 	if (ifgr->ifgr_len == 0) {
1306 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1307 			ifgr->ifgr_len += sizeof(ifgrq);
1308 		IFNET_RUNLOCK();
1309 		return (0);
1310 	}
1311 
1312 	len = ifgr->ifgr_len;
1313 	ifgp = ifgr->ifgr_groups;
1314 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1315 		if (len < sizeof(ifgrq)) {
1316 			IFNET_RUNLOCK();
1317 			return (EINVAL);
1318 		}
1319 		bzero(&ifgrq, sizeof ifgrq);
1320 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1321 		    sizeof(ifgrq.ifgrq_member));
1322 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1323 			IFNET_RUNLOCK();
1324 			return (error);
1325 		}
1326 		len -= sizeof(ifgrq);
1327 		ifgp++;
1328 	}
1329 	IFNET_RUNLOCK();
1330 
1331 	return (0);
1332 }
1333 
1334 /*
1335  * Delete Routes for a Network Interface
1336  *
1337  * Called for each routing entry via the rnh->rnh_walktree() call above
1338  * to delete all route entries referencing a detaching network interface.
1339  *
1340  * Arguments:
1341  *	rn	pointer to node in the routing table
1342  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
1343  *
1344  * Returns:
1345  *	0	successful
1346  *	errno	failed - reason indicated
1347  *
1348  */
1349 static int
1350 if_rtdel(struct radix_node *rn, void *arg)
1351 {
1352 	struct rtentry	*rt = (struct rtentry *)rn;
1353 	struct ifnet	*ifp = arg;
1354 	int		err;
1355 
1356 	if (rt->rt_ifp == ifp) {
1357 
1358 		/*
1359 		 * Protect (sorta) against walktree recursion problems
1360 		 * with cloned routes
1361 		 */
1362 		if ((rt->rt_flags & RTF_UP) == 0)
1363 			return (0);
1364 
1365 		err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1366 				rt_mask(rt), rt->rt_flags|RTF_RNH_LOCKED,
1367 				(struct rtentry **) NULL, rt->rt_fibnum);
1368 		if (err) {
1369 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1370 		}
1371 	}
1372 
1373 	return (0);
1374 }
1375 
1376 /*
1377  * Wrapper functions for struct ifnet address list locking macros.  These are
1378  * used by kernel modules to avoid encoding programming interface or binary
1379  * interface assumptions that may be violated when kernel-internal locking
1380  * approaches change.
1381  */
1382 void
1383 if_addr_rlock(struct ifnet *ifp)
1384 {
1385 
1386 	IF_ADDR_RLOCK(ifp);
1387 }
1388 
1389 void
1390 if_addr_runlock(struct ifnet *ifp)
1391 {
1392 
1393 	IF_ADDR_RUNLOCK(ifp);
1394 }
1395 
1396 void
1397 if_maddr_rlock(struct ifnet *ifp)
1398 {
1399 
1400 	IF_ADDR_RLOCK(ifp);
1401 }
1402 
1403 void
1404 if_maddr_runlock(struct ifnet *ifp)
1405 {
1406 
1407 	IF_ADDR_RUNLOCK(ifp);
1408 }
1409 
1410 /*
1411  * Reference count functions for ifaddrs.
1412  */
1413 void
1414 ifa_init(struct ifaddr *ifa)
1415 {
1416 
1417 	mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1418 	refcount_init(&ifa->ifa_refcnt, 1);
1419 }
1420 
1421 void
1422 ifa_ref(struct ifaddr *ifa)
1423 {
1424 
1425 	refcount_acquire(&ifa->ifa_refcnt);
1426 }
1427 
1428 void
1429 ifa_free(struct ifaddr *ifa)
1430 {
1431 
1432 	if (refcount_release(&ifa->ifa_refcnt)) {
1433 		mtx_destroy(&ifa->ifa_mtx);
1434 		free(ifa, M_IFADDR);
1435 	}
1436 }
1437 
1438 int
1439 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1440 {
1441 	int error = 0;
1442 	struct rtentry *rt = NULL;
1443 	struct rt_addrinfo info;
1444 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1445 
1446 	bzero(&info, sizeof(info));
1447 	info.rti_ifp = V_loif;
1448 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1449 	info.rti_info[RTAX_DST] = ia;
1450 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1451 	error = rtrequest1_fib(RTM_ADD, &info, &rt, 0);
1452 
1453 	if (error == 0 && rt != NULL) {
1454 		RT_LOCK(rt);
1455 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_type  =
1456 			ifa->ifa_ifp->if_type;
1457 		((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1458 			ifa->ifa_ifp->if_index;
1459 		RT_REMREF(rt);
1460 		RT_UNLOCK(rt);
1461 	} else if (error != 0)
1462 		log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1463 
1464 	return (error);
1465 }
1466 
1467 int
1468 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1469 {
1470 	int error = 0;
1471 	struct rt_addrinfo info;
1472 	struct sockaddr_dl null_sdl;
1473 
1474 	bzero(&null_sdl, sizeof(null_sdl));
1475 	null_sdl.sdl_len = sizeof(null_sdl);
1476 	null_sdl.sdl_family = AF_LINK;
1477 	null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1478 	null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1479 	bzero(&info, sizeof(info));
1480 	info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC;
1481 	info.rti_info[RTAX_DST] = ia;
1482 	info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1483 	error = rtrequest1_fib(RTM_DELETE, &info, NULL, 0);
1484 
1485 	if (error != 0)
1486 		log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1487 
1488 	return (error);
1489 }
1490 
1491 /*
1492  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1493  * structs used to represent other address families, it is necessary
1494  * to perform a different comparison.
1495  */
1496 
1497 #define	sa_equal(a1, a2)	\
1498 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1499 
1500 #define	sa_dl_equal(a1, a2)	\
1501 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1502 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1503 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1504 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1505 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1506 
1507 /*
1508  * Locate an interface based on a complete address.
1509  */
1510 /*ARGSUSED*/
1511 static struct ifaddr *
1512 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1513 {
1514 	struct ifnet *ifp;
1515 	struct ifaddr *ifa;
1516 
1517 	IFNET_RLOCK_NOSLEEP();
1518 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1519 		IF_ADDR_RLOCK(ifp);
1520 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1521 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1522 				continue;
1523 			if (sa_equal(addr, ifa->ifa_addr)) {
1524 				if (getref)
1525 					ifa_ref(ifa);
1526 				IF_ADDR_RUNLOCK(ifp);
1527 				goto done;
1528 			}
1529 			/* IP6 doesn't have broadcast */
1530 			if ((ifp->if_flags & IFF_BROADCAST) &&
1531 			    ifa->ifa_broadaddr &&
1532 			    ifa->ifa_broadaddr->sa_len != 0 &&
1533 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1534 				if (getref)
1535 					ifa_ref(ifa);
1536 				IF_ADDR_RUNLOCK(ifp);
1537 				goto done;
1538 			}
1539 		}
1540 		IF_ADDR_RUNLOCK(ifp);
1541 	}
1542 	ifa = NULL;
1543 done:
1544 	IFNET_RUNLOCK_NOSLEEP();
1545 	return (ifa);
1546 }
1547 
1548 struct ifaddr *
1549 ifa_ifwithaddr(struct sockaddr *addr)
1550 {
1551 
1552 	return (ifa_ifwithaddr_internal(addr, 1));
1553 }
1554 
1555 int
1556 ifa_ifwithaddr_check(struct sockaddr *addr)
1557 {
1558 
1559 	return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1560 }
1561 
1562 /*
1563  * Locate an interface based on the broadcast address.
1564  */
1565 /* ARGSUSED */
1566 struct ifaddr *
1567 ifa_ifwithbroadaddr(struct sockaddr *addr)
1568 {
1569 	struct ifnet *ifp;
1570 	struct ifaddr *ifa;
1571 
1572 	IFNET_RLOCK_NOSLEEP();
1573 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1574 		IF_ADDR_RLOCK(ifp);
1575 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1576 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1577 				continue;
1578 			if ((ifp->if_flags & IFF_BROADCAST) &&
1579 			    ifa->ifa_broadaddr &&
1580 			    ifa->ifa_broadaddr->sa_len != 0 &&
1581 			    sa_equal(ifa->ifa_broadaddr, addr)) {
1582 				ifa_ref(ifa);
1583 				IF_ADDR_RUNLOCK(ifp);
1584 				goto done;
1585 			}
1586 		}
1587 		IF_ADDR_RUNLOCK(ifp);
1588 	}
1589 	ifa = NULL;
1590 done:
1591 	IFNET_RUNLOCK_NOSLEEP();
1592 	return (ifa);
1593 }
1594 
1595 /*
1596  * Locate the point to point interface with a given destination address.
1597  */
1598 /*ARGSUSED*/
1599 struct ifaddr *
1600 ifa_ifwithdstaddr(struct sockaddr *addr)
1601 {
1602 	struct ifnet *ifp;
1603 	struct ifaddr *ifa;
1604 
1605 	IFNET_RLOCK_NOSLEEP();
1606 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1607 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1608 			continue;
1609 		IF_ADDR_RLOCK(ifp);
1610 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1611 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1612 				continue;
1613 			if (ifa->ifa_dstaddr != NULL &&
1614 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1615 				ifa_ref(ifa);
1616 				IF_ADDR_RUNLOCK(ifp);
1617 				goto done;
1618 			}
1619 		}
1620 		IF_ADDR_RUNLOCK(ifp);
1621 	}
1622 	ifa = NULL;
1623 done:
1624 	IFNET_RUNLOCK_NOSLEEP();
1625 	return (ifa);
1626 }
1627 
1628 /*
1629  * Find an interface on a specific network.  If many, choice
1630  * is most specific found.
1631  */
1632 struct ifaddr *
1633 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1634 {
1635 	struct ifnet *ifp;
1636 	struct ifaddr *ifa;
1637 	struct ifaddr *ifa_maybe = NULL;
1638 	u_int af = addr->sa_family;
1639 	char *addr_data = addr->sa_data, *cplim;
1640 
1641 	/*
1642 	 * AF_LINK addresses can be looked up directly by their index number,
1643 	 * so do that if we can.
1644 	 */
1645 	if (af == AF_LINK) {
1646 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1647 	    if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1648 		return (ifaddr_byindex(sdl->sdl_index));
1649 	}
1650 
1651 	/*
1652 	 * Scan though each interface, looking for ones that have addresses
1653 	 * in this address family.  Maintain a reference on ifa_maybe once
1654 	 * we find one, as we release the IF_ADDR_RLOCK() that kept it stable
1655 	 * when we move onto the next interface.
1656 	 */
1657 	IFNET_RLOCK_NOSLEEP();
1658 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1659 		IF_ADDR_RLOCK(ifp);
1660 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1661 			char *cp, *cp2, *cp3;
1662 
1663 			if (ifa->ifa_addr->sa_family != af)
1664 next:				continue;
1665 			if (af == AF_INET &&
1666 			    ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1667 				/*
1668 				 * This is a bit broken as it doesn't
1669 				 * take into account that the remote end may
1670 				 * be a single node in the network we are
1671 				 * looking for.
1672 				 * The trouble is that we don't know the
1673 				 * netmask for the remote end.
1674 				 */
1675 				if (ifa->ifa_dstaddr != NULL &&
1676 				    sa_equal(addr, ifa->ifa_dstaddr)) {
1677 					ifa_ref(ifa);
1678 					IF_ADDR_RUNLOCK(ifp);
1679 					goto done;
1680 				}
1681 			} else {
1682 				/*
1683 				 * if we have a special address handler,
1684 				 * then use it instead of the generic one.
1685 				 */
1686 				if (ifa->ifa_claim_addr) {
1687 					if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1688 						ifa_ref(ifa);
1689 						IF_ADDR_RUNLOCK(ifp);
1690 						goto done;
1691 					}
1692 					continue;
1693 				}
1694 
1695 				/*
1696 				 * Scan all the bits in the ifa's address.
1697 				 * If a bit dissagrees with what we are
1698 				 * looking for, mask it with the netmask
1699 				 * to see if it really matters.
1700 				 * (A byte at a time)
1701 				 */
1702 				if (ifa->ifa_netmask == 0)
1703 					continue;
1704 				cp = addr_data;
1705 				cp2 = ifa->ifa_addr->sa_data;
1706 				cp3 = ifa->ifa_netmask->sa_data;
1707 				cplim = ifa->ifa_netmask->sa_len
1708 					+ (char *)ifa->ifa_netmask;
1709 				while (cp3 < cplim)
1710 					if ((*cp++ ^ *cp2++) & *cp3++)
1711 						goto next; /* next address! */
1712 				/*
1713 				 * If the netmask of what we just found
1714 				 * is more specific than what we had before
1715 				 * (if we had one) then remember the new one
1716 				 * before continuing to search
1717 				 * for an even better one.
1718 				 */
1719 				if (ifa_maybe == NULL ||
1720 				    rn_refines((caddr_t)ifa->ifa_netmask,
1721 				    (caddr_t)ifa_maybe->ifa_netmask)) {
1722 					if (ifa_maybe != NULL)
1723 						ifa_free(ifa_maybe);
1724 					ifa_maybe = ifa;
1725 					ifa_ref(ifa_maybe);
1726 				}
1727 			}
1728 		}
1729 		IF_ADDR_RUNLOCK(ifp);
1730 	}
1731 	ifa = ifa_maybe;
1732 	ifa_maybe = NULL;
1733 done:
1734 	IFNET_RUNLOCK_NOSLEEP();
1735 	if (ifa_maybe != NULL)
1736 		ifa_free(ifa_maybe);
1737 	return (ifa);
1738 }
1739 
1740 /*
1741  * Find an interface address specific to an interface best matching
1742  * a given address.
1743  */
1744 struct ifaddr *
1745 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1746 {
1747 	struct ifaddr *ifa;
1748 	char *cp, *cp2, *cp3;
1749 	char *cplim;
1750 	struct ifaddr *ifa_maybe = NULL;
1751 	u_int af = addr->sa_family;
1752 
1753 	if (af >= AF_MAX)
1754 		return (NULL);
1755 	IF_ADDR_RLOCK(ifp);
1756 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1757 		if (ifa->ifa_addr->sa_family != af)
1758 			continue;
1759 		if (ifa_maybe == NULL)
1760 			ifa_maybe = ifa;
1761 		if (ifa->ifa_netmask == 0) {
1762 			if (sa_equal(addr, ifa->ifa_addr) ||
1763 			    (ifa->ifa_dstaddr &&
1764 			    sa_equal(addr, ifa->ifa_dstaddr)))
1765 				goto done;
1766 			continue;
1767 		}
1768 		if (ifp->if_flags & IFF_POINTOPOINT) {
1769 			if (sa_equal(addr, ifa->ifa_dstaddr))
1770 				goto done;
1771 		} else {
1772 			cp = addr->sa_data;
1773 			cp2 = ifa->ifa_addr->sa_data;
1774 			cp3 = ifa->ifa_netmask->sa_data;
1775 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1776 			for (; cp3 < cplim; cp3++)
1777 				if ((*cp++ ^ *cp2++) & *cp3)
1778 					break;
1779 			if (cp3 == cplim)
1780 				goto done;
1781 		}
1782 	}
1783 	ifa = ifa_maybe;
1784 done:
1785 	if (ifa != NULL)
1786 		ifa_ref(ifa);
1787 	IF_ADDR_RUNLOCK(ifp);
1788 	return (ifa);
1789 }
1790 
1791 #include <net/if_llatbl.h>
1792 
1793 /*
1794  * Default action when installing a route with a Link Level gateway.
1795  * Lookup an appropriate real ifa to point to.
1796  * This should be moved to /sys/net/link.c eventually.
1797  */
1798 static void
1799 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1800 {
1801 	struct ifaddr *ifa, *oifa;
1802 	struct sockaddr *dst;
1803 	struct ifnet *ifp;
1804 
1805 	RT_LOCK_ASSERT(rt);
1806 
1807 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1808 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1809 		return;
1810 	ifa = ifaof_ifpforaddr(dst, ifp);
1811 	if (ifa) {
1812 		oifa = rt->rt_ifa;
1813 		rt->rt_ifa = ifa;
1814 		ifa_free(oifa);
1815 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1816 			ifa->ifa_rtrequest(cmd, rt, info);
1817 	}
1818 }
1819 
1820 /*
1821  * Mark an interface down and notify protocols of
1822  * the transition.
1823  * NOTE: must be called at splnet or eqivalent.
1824  */
1825 static void
1826 if_unroute(struct ifnet *ifp, int flag, int fam)
1827 {
1828 	struct ifaddr *ifa;
1829 
1830 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1831 
1832 	ifp->if_flags &= ~flag;
1833 	getmicrotime(&ifp->if_lastchange);
1834 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1835 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1836 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1837 	ifp->if_qflush(ifp);
1838 
1839 	if (ifp->if_carp)
1840 		(*carp_linkstate_p)(ifp);
1841 	rt_ifmsg(ifp);
1842 }
1843 
1844 /*
1845  * Mark an interface up and notify protocols of
1846  * the transition.
1847  * NOTE: must be called at splnet or eqivalent.
1848  */
1849 static void
1850 if_route(struct ifnet *ifp, int flag, int fam)
1851 {
1852 	struct ifaddr *ifa;
1853 
1854 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1855 
1856 	ifp->if_flags |= flag;
1857 	getmicrotime(&ifp->if_lastchange);
1858 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1859 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1860 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1861 	if (ifp->if_carp)
1862 		(*carp_linkstate_p)(ifp);
1863 	rt_ifmsg(ifp);
1864 #ifdef INET6
1865 	in6_if_up(ifp);
1866 #endif
1867 }
1868 
1869 void	(*vlan_link_state_p)(struct ifnet *);	/* XXX: private from if_vlan */
1870 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1871 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
1872 struct	ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
1873 int	(*vlan_tag_p)(struct ifnet *, uint16_t *);
1874 int	(*vlan_setcookie_p)(struct ifnet *, void *);
1875 void	*(*vlan_cookie_p)(struct ifnet *);
1876 
1877 /*
1878  * Handle a change in the interface link state. To avoid LORs
1879  * between driver lock and upper layer locks, as well as possible
1880  * recursions, we post event to taskqueue, and all job
1881  * is done in static do_link_state_change().
1882  */
1883 void
1884 if_link_state_change(struct ifnet *ifp, int link_state)
1885 {
1886 	/* Return if state hasn't changed. */
1887 	if (ifp->if_link_state == link_state)
1888 		return;
1889 
1890 	ifp->if_link_state = link_state;
1891 
1892 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1893 }
1894 
1895 static void
1896 do_link_state_change(void *arg, int pending)
1897 {
1898 	struct ifnet *ifp = (struct ifnet *)arg;
1899 	int link_state = ifp->if_link_state;
1900 	CURVNET_SET(ifp->if_vnet);
1901 
1902 	/* Notify that the link state has changed. */
1903 	rt_ifmsg(ifp);
1904 	if (ifp->if_vlantrunk != NULL)
1905 		(*vlan_link_state_p)(ifp);
1906 
1907 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1908 	    IFP2AC(ifp)->ac_netgraph != NULL)
1909 		(*ng_ether_link_state_p)(ifp, link_state);
1910 	if (ifp->if_carp)
1911 		(*carp_linkstate_p)(ifp);
1912 	if (ifp->if_bridge) {
1913 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1914 		(*bstp_linkstate_p)(ifp, link_state);
1915 	}
1916 	if (ifp->if_lagg) {
1917 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1918 		(*lagg_linkstate_p)(ifp, link_state);
1919 	}
1920 
1921 	if (IS_DEFAULT_VNET(curvnet))
1922 		devctl_notify("IFNET", ifp->if_xname,
1923 		    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
1924 		    NULL);
1925 	if (pending > 1)
1926 		if_printf(ifp, "%d link states coalesced\n", pending);
1927 	if (log_link_state_change)
1928 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1929 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1930 	EVENTHANDLER_INVOKE(ifnet_link_event, ifp, ifp->if_link_state);
1931 	CURVNET_RESTORE();
1932 }
1933 
1934 /*
1935  * Mark an interface down and notify protocols of
1936  * the transition.
1937  * NOTE: must be called at splnet or eqivalent.
1938  */
1939 void
1940 if_down(struct ifnet *ifp)
1941 {
1942 
1943 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1944 }
1945 
1946 /*
1947  * Mark an interface up and notify protocols of
1948  * the transition.
1949  * NOTE: must be called at splnet or eqivalent.
1950  */
1951 void
1952 if_up(struct ifnet *ifp)
1953 {
1954 
1955 	if_route(ifp, IFF_UP, AF_UNSPEC);
1956 }
1957 
1958 /*
1959  * Flush an interface queue.
1960  */
1961 void
1962 if_qflush(struct ifnet *ifp)
1963 {
1964 	struct mbuf *m, *n;
1965 	struct ifaltq *ifq;
1966 
1967 	ifq = &ifp->if_snd;
1968 	IFQ_LOCK(ifq);
1969 #ifdef ALTQ
1970 	if (ALTQ_IS_ENABLED(ifq))
1971 		ALTQ_PURGE(ifq);
1972 #endif
1973 	n = ifq->ifq_head;
1974 	while ((m = n) != 0) {
1975 		n = m->m_act;
1976 		m_freem(m);
1977 	}
1978 	ifq->ifq_head = 0;
1979 	ifq->ifq_tail = 0;
1980 	ifq->ifq_len = 0;
1981 	IFQ_UNLOCK(ifq);
1982 }
1983 
1984 /*
1985  * Map interface name to interface structure pointer, with or without
1986  * returning a reference.
1987  */
1988 struct ifnet *
1989 ifunit_ref(const char *name)
1990 {
1991 	struct ifnet *ifp;
1992 
1993 	IFNET_RLOCK_NOSLEEP();
1994 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1995 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
1996 		    !(ifp->if_flags & IFF_DYING))
1997 			break;
1998 	}
1999 	if (ifp != NULL)
2000 		if_ref(ifp);
2001 	IFNET_RUNLOCK_NOSLEEP();
2002 	return (ifp);
2003 }
2004 
2005 struct ifnet *
2006 ifunit(const char *name)
2007 {
2008 	struct ifnet *ifp;
2009 
2010 	IFNET_RLOCK_NOSLEEP();
2011 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2012 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2013 			break;
2014 	}
2015 	IFNET_RUNLOCK_NOSLEEP();
2016 	return (ifp);
2017 }
2018 
2019 /*
2020  * Hardware specific interface ioctls.
2021  */
2022 static int
2023 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2024 {
2025 	struct ifreq *ifr;
2026 	struct ifstat *ifs;
2027 	int error = 0;
2028 	int new_flags, temp_flags;
2029 	size_t namelen, onamelen;
2030 	size_t descrlen;
2031 	char *descrbuf, *odescrbuf;
2032 	char new_name[IFNAMSIZ];
2033 	struct ifaddr *ifa;
2034 	struct sockaddr_dl *sdl;
2035 
2036 	ifr = (struct ifreq *)data;
2037 	switch (cmd) {
2038 	case SIOCGIFINDEX:
2039 		ifr->ifr_index = ifp->if_index;
2040 		break;
2041 
2042 	case SIOCGIFFLAGS:
2043 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
2044 		ifr->ifr_flags = temp_flags & 0xffff;
2045 		ifr->ifr_flagshigh = temp_flags >> 16;
2046 		break;
2047 
2048 	case SIOCGIFCAP:
2049 		ifr->ifr_reqcap = ifp->if_capabilities;
2050 		ifr->ifr_curcap = ifp->if_capenable;
2051 		break;
2052 
2053 #ifdef MAC
2054 	case SIOCGIFMAC:
2055 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2056 		break;
2057 #endif
2058 
2059 	case SIOCGIFMETRIC:
2060 		ifr->ifr_metric = ifp->if_metric;
2061 		break;
2062 
2063 	case SIOCGIFMTU:
2064 		ifr->ifr_mtu = ifp->if_mtu;
2065 		break;
2066 
2067 	case SIOCGIFPHYS:
2068 		ifr->ifr_phys = ifp->if_physical;
2069 		break;
2070 
2071 	case SIOCGIFDESCR:
2072 		error = 0;
2073 		sx_slock(&ifdescr_sx);
2074 		if (ifp->if_description == NULL)
2075 			error = ENOMSG;
2076 		else {
2077 			/* space for terminating nul */
2078 			descrlen = strlen(ifp->if_description) + 1;
2079 			if (ifr->ifr_buffer.length < descrlen)
2080 				ifr->ifr_buffer.buffer = NULL;
2081 			else
2082 				error = copyout(ifp->if_description,
2083 				    ifr->ifr_buffer.buffer, descrlen);
2084 			ifr->ifr_buffer.length = descrlen;
2085 		}
2086 		sx_sunlock(&ifdescr_sx);
2087 		break;
2088 
2089 	case SIOCSIFDESCR:
2090 		error = priv_check(td, PRIV_NET_SETIFDESCR);
2091 		if (error)
2092 			return (error);
2093 
2094 		/*
2095 		 * Copy only (length-1) bytes to make sure that
2096 		 * if_description is always nul terminated.  The
2097 		 * length parameter is supposed to count the
2098 		 * terminating nul in.
2099 		 */
2100 		if (ifr->ifr_buffer.length > ifdescr_maxlen)
2101 			return (ENAMETOOLONG);
2102 		else if (ifr->ifr_buffer.length == 0)
2103 			descrbuf = NULL;
2104 		else {
2105 			descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2106 			    M_WAITOK | M_ZERO);
2107 			error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2108 			    ifr->ifr_buffer.length - 1);
2109 			if (error) {
2110 				free(descrbuf, M_IFDESCR);
2111 				break;
2112 			}
2113 		}
2114 
2115 		sx_xlock(&ifdescr_sx);
2116 		odescrbuf = ifp->if_description;
2117 		ifp->if_description = descrbuf;
2118 		sx_xunlock(&ifdescr_sx);
2119 
2120 		getmicrotime(&ifp->if_lastchange);
2121 		free(odescrbuf, M_IFDESCR);
2122 		break;
2123 
2124 	case SIOCGIFFIB:
2125 		ifr->ifr_fib = ifp->if_fib;
2126 		break;
2127 
2128 	case SIOCSIFFIB:
2129 		error = priv_check(td, PRIV_NET_SETIFFIB);
2130 		if (error)
2131 			return (error);
2132 		if (ifr->ifr_fib >= rt_numfibs)
2133 			return (EINVAL);
2134 
2135 		ifp->if_fib = ifr->ifr_fib;
2136 		break;
2137 
2138 	case SIOCSIFFLAGS:
2139 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
2140 		if (error)
2141 			return (error);
2142 		/*
2143 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2144 		 * check, so we don't need special handling here yet.
2145 		 */
2146 		new_flags = (ifr->ifr_flags & 0xffff) |
2147 		    (ifr->ifr_flagshigh << 16);
2148 		if (ifp->if_flags & IFF_SMART) {
2149 			/* Smart drivers twiddle their own routes */
2150 		} else if (ifp->if_flags & IFF_UP &&
2151 		    (new_flags & IFF_UP) == 0) {
2152 			int s = splimp();
2153 			if_down(ifp);
2154 			splx(s);
2155 		} else if (new_flags & IFF_UP &&
2156 		    (ifp->if_flags & IFF_UP) == 0) {
2157 			int s = splimp();
2158 			if_up(ifp);
2159 			splx(s);
2160 		}
2161 		/* See if permanently promiscuous mode bit is about to flip */
2162 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2163 			if (new_flags & IFF_PPROMISC)
2164 				ifp->if_flags |= IFF_PROMISC;
2165 			else if (ifp->if_pcount == 0)
2166 				ifp->if_flags &= ~IFF_PROMISC;
2167 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2168 			    ifp->if_xname,
2169 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
2170 		}
2171 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2172 			(new_flags &~ IFF_CANTCHANGE);
2173 		if (ifp->if_ioctl) {
2174 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
2175 		}
2176 		getmicrotime(&ifp->if_lastchange);
2177 		break;
2178 
2179 	case SIOCSIFCAP:
2180 		error = priv_check(td, PRIV_NET_SETIFCAP);
2181 		if (error)
2182 			return (error);
2183 		if (ifp->if_ioctl == NULL)
2184 			return (EOPNOTSUPP);
2185 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2186 			return (EINVAL);
2187 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2188 		if (error == 0)
2189 			getmicrotime(&ifp->if_lastchange);
2190 		break;
2191 
2192 #ifdef MAC
2193 	case SIOCSIFMAC:
2194 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2195 		break;
2196 #endif
2197 
2198 	case SIOCSIFNAME:
2199 		error = priv_check(td, PRIV_NET_SETIFNAME);
2200 		if (error)
2201 			return (error);
2202 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2203 		if (error != 0)
2204 			return (error);
2205 		if (new_name[0] == '\0')
2206 			return (EINVAL);
2207 		if (ifunit(new_name) != NULL)
2208 			return (EEXIST);
2209 
2210 		/*
2211 		 * XXX: Locking.  Nothing else seems to lock if_flags,
2212 		 * and there are numerous other races with the
2213 		 * ifunit() checks not being atomic with namespace
2214 		 * changes (renames, vmoves, if_attach, etc).
2215 		 */
2216 		ifp->if_flags |= IFF_RENAMING;
2217 
2218 		/* Announce the departure of the interface. */
2219 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2220 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2221 
2222 		log(LOG_INFO, "%s: changing name to '%s'\n",
2223 		    ifp->if_xname, new_name);
2224 
2225 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2226 		ifa = ifp->if_addr;
2227 		IFA_LOCK(ifa);
2228 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2229 		namelen = strlen(new_name);
2230 		onamelen = sdl->sdl_nlen;
2231 		/*
2232 		 * Move the address if needed.  This is safe because we
2233 		 * allocate space for a name of length IFNAMSIZ when we
2234 		 * create this in if_attach().
2235 		 */
2236 		if (namelen != onamelen) {
2237 			bcopy(sdl->sdl_data + onamelen,
2238 			    sdl->sdl_data + namelen, sdl->sdl_alen);
2239 		}
2240 		bcopy(new_name, sdl->sdl_data, namelen);
2241 		sdl->sdl_nlen = namelen;
2242 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2243 		bzero(sdl->sdl_data, onamelen);
2244 		while (namelen != 0)
2245 			sdl->sdl_data[--namelen] = 0xff;
2246 		IFA_UNLOCK(ifa);
2247 
2248 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2249 		/* Announce the return of the interface. */
2250 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2251 
2252 		ifp->if_flags &= ~IFF_RENAMING;
2253 		break;
2254 
2255 #ifdef VIMAGE
2256 	case SIOCSIFVNET:
2257 		error = priv_check(td, PRIV_NET_SETIFVNET);
2258 		if (error)
2259 			return (error);
2260 		error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2261 		break;
2262 #endif
2263 
2264 	case SIOCSIFMETRIC:
2265 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
2266 		if (error)
2267 			return (error);
2268 		ifp->if_metric = ifr->ifr_metric;
2269 		getmicrotime(&ifp->if_lastchange);
2270 		break;
2271 
2272 	case SIOCSIFPHYS:
2273 		error = priv_check(td, PRIV_NET_SETIFPHYS);
2274 		if (error)
2275 			return (error);
2276 		if (ifp->if_ioctl == NULL)
2277 			return (EOPNOTSUPP);
2278 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2279 		if (error == 0)
2280 			getmicrotime(&ifp->if_lastchange);
2281 		break;
2282 
2283 	case SIOCSIFMTU:
2284 	{
2285 		u_long oldmtu = ifp->if_mtu;
2286 
2287 		error = priv_check(td, PRIV_NET_SETIFMTU);
2288 		if (error)
2289 			return (error);
2290 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2291 			return (EINVAL);
2292 		if (ifp->if_ioctl == NULL)
2293 			return (EOPNOTSUPP);
2294 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2295 		if (error == 0) {
2296 			getmicrotime(&ifp->if_lastchange);
2297 			rt_ifmsg(ifp);
2298 		}
2299 		/*
2300 		 * If the link MTU changed, do network layer specific procedure.
2301 		 */
2302 		if (ifp->if_mtu != oldmtu) {
2303 #ifdef INET6
2304 			nd6_setmtu(ifp);
2305 #endif
2306 		}
2307 		break;
2308 	}
2309 
2310 	case SIOCADDMULTI:
2311 	case SIOCDELMULTI:
2312 		if (cmd == SIOCADDMULTI)
2313 			error = priv_check(td, PRIV_NET_ADDMULTI);
2314 		else
2315 			error = priv_check(td, PRIV_NET_DELMULTI);
2316 		if (error)
2317 			return (error);
2318 
2319 		/* Don't allow group membership on non-multicast interfaces. */
2320 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2321 			return (EOPNOTSUPP);
2322 
2323 		/* Don't let users screw up protocols' entries. */
2324 		if (ifr->ifr_addr.sa_family != AF_LINK)
2325 			return (EINVAL);
2326 
2327 		if (cmd == SIOCADDMULTI) {
2328 			struct ifmultiaddr *ifma;
2329 
2330 			/*
2331 			 * Userland is only permitted to join groups once
2332 			 * via the if_addmulti() KPI, because it cannot hold
2333 			 * struct ifmultiaddr * between calls. It may also
2334 			 * lose a race while we check if the membership
2335 			 * already exists.
2336 			 */
2337 			IF_ADDR_RLOCK(ifp);
2338 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
2339 			IF_ADDR_RUNLOCK(ifp);
2340 			if (ifma != NULL)
2341 				error = EADDRINUSE;
2342 			else
2343 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2344 		} else {
2345 			error = if_delmulti(ifp, &ifr->ifr_addr);
2346 		}
2347 		if (error == 0)
2348 			getmicrotime(&ifp->if_lastchange);
2349 		break;
2350 
2351 	case SIOCSIFPHYADDR:
2352 	case SIOCDIFPHYADDR:
2353 #ifdef INET6
2354 	case SIOCSIFPHYADDR_IN6:
2355 #endif
2356 	case SIOCSLIFPHYADDR:
2357 	case SIOCSIFMEDIA:
2358 	case SIOCSIFGENERIC:
2359 		error = priv_check(td, PRIV_NET_HWIOCTL);
2360 		if (error)
2361 			return (error);
2362 		if (ifp->if_ioctl == NULL)
2363 			return (EOPNOTSUPP);
2364 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2365 		if (error == 0)
2366 			getmicrotime(&ifp->if_lastchange);
2367 		break;
2368 
2369 	case SIOCGIFSTATUS:
2370 		ifs = (struct ifstat *)data;
2371 		ifs->ascii[0] = '\0';
2372 
2373 	case SIOCGIFPSRCADDR:
2374 	case SIOCGIFPDSTADDR:
2375 	case SIOCGLIFPHYADDR:
2376 	case SIOCGIFMEDIA:
2377 	case SIOCGIFGENERIC:
2378 		if (ifp->if_ioctl == NULL)
2379 			return (EOPNOTSUPP);
2380 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2381 		break;
2382 
2383 	case SIOCSIFLLADDR:
2384 		error = priv_check(td, PRIV_NET_SETLLADDR);
2385 		if (error)
2386 			return (error);
2387 		error = if_setlladdr(ifp,
2388 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2389 		EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2390 		break;
2391 
2392 	case SIOCAIFGROUP:
2393 	{
2394 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2395 
2396 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
2397 		if (error)
2398 			return (error);
2399 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2400 			return (error);
2401 		break;
2402 	}
2403 
2404 	case SIOCGIFGROUP:
2405 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2406 			return (error);
2407 		break;
2408 
2409 	case SIOCDIFGROUP:
2410 	{
2411 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2412 
2413 		error = priv_check(td, PRIV_NET_DELIFGROUP);
2414 		if (error)
2415 			return (error);
2416 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2417 			return (error);
2418 		break;
2419 	}
2420 
2421 	default:
2422 		error = ENOIOCTL;
2423 		break;
2424 	}
2425 	return (error);
2426 }
2427 
2428 #ifdef COMPAT_FREEBSD32
2429 struct ifconf32 {
2430 	int32_t	ifc_len;
2431 	union {
2432 		uint32_t	ifcu_buf;
2433 		uint32_t	ifcu_req;
2434 	} ifc_ifcu;
2435 };
2436 #define	SIOCGIFCONF32	_IOWR('i', 36, struct ifconf32)
2437 #endif
2438 
2439 /*
2440  * Interface ioctls.
2441  */
2442 int
2443 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2444 {
2445 	struct ifnet *ifp;
2446 	struct ifreq *ifr;
2447 	int error;
2448 	int oif_flags;
2449 
2450 	CURVNET_SET(so->so_vnet);
2451 	switch (cmd) {
2452 	case SIOCGIFCONF:
2453 	case OSIOCGIFCONF:
2454 		error = ifconf(cmd, data);
2455 		CURVNET_RESTORE();
2456 		return (error);
2457 
2458 #ifdef COMPAT_FREEBSD32
2459 	case SIOCGIFCONF32:
2460 		{
2461 			struct ifconf32 *ifc32;
2462 			struct ifconf ifc;
2463 
2464 			ifc32 = (struct ifconf32 *)data;
2465 			ifc.ifc_len = ifc32->ifc_len;
2466 			ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2467 
2468 			error = ifconf(SIOCGIFCONF, (void *)&ifc);
2469 			CURVNET_RESTORE();
2470 			if (error == 0)
2471 				ifc32->ifc_len = ifc.ifc_len;
2472 			return (error);
2473 		}
2474 #endif
2475 	}
2476 	ifr = (struct ifreq *)data;
2477 
2478 	switch (cmd) {
2479 #ifdef VIMAGE
2480 	case SIOCSIFRVNET:
2481 		error = priv_check(td, PRIV_NET_SETIFVNET);
2482 		if (error == 0)
2483 			error = if_vmove_reclaim(td, ifr->ifr_name,
2484 			    ifr->ifr_jid);
2485 		CURVNET_RESTORE();
2486 		return (error);
2487 #endif
2488 	case SIOCIFCREATE:
2489 	case SIOCIFCREATE2:
2490 		error = priv_check(td, PRIV_NET_IFCREATE);
2491 		if (error == 0)
2492 			error = if_clone_create(ifr->ifr_name,
2493 			    sizeof(ifr->ifr_name),
2494 			    cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2495 		CURVNET_RESTORE();
2496 		return (error);
2497 	case SIOCIFDESTROY:
2498 		error = priv_check(td, PRIV_NET_IFDESTROY);
2499 		if (error == 0)
2500 			error = if_clone_destroy(ifr->ifr_name);
2501 		CURVNET_RESTORE();
2502 		return (error);
2503 
2504 	case SIOCIFGCLONERS:
2505 		error = if_clone_list((struct if_clonereq *)data);
2506 		CURVNET_RESTORE();
2507 		return (error);
2508 	case SIOCGIFGMEMB:
2509 		error = if_getgroupmembers((struct ifgroupreq *)data);
2510 		CURVNET_RESTORE();
2511 		return (error);
2512 #if defined(INET) || defined(INET6)
2513 	case SIOCSVH:
2514 	case SIOCGVH:
2515 		if (carp_ioctl_p == NULL)
2516 			error = EPROTONOSUPPORT;
2517 		else
2518 			error = (*carp_ioctl_p)(ifr, cmd, td);
2519 		CURVNET_RESTORE();
2520 		return (error);
2521 #endif
2522 	}
2523 
2524 	ifp = ifunit_ref(ifr->ifr_name);
2525 	if (ifp == NULL) {
2526 		CURVNET_RESTORE();
2527 		return (ENXIO);
2528 	}
2529 
2530 	error = ifhwioctl(cmd, ifp, data, td);
2531 	if (error != ENOIOCTL) {
2532 		if_rele(ifp);
2533 		CURVNET_RESTORE();
2534 		return (error);
2535 	}
2536 
2537 	oif_flags = ifp->if_flags;
2538 	if (so->so_proto == NULL) {
2539 		if_rele(ifp);
2540 		CURVNET_RESTORE();
2541 		return (EOPNOTSUPP);
2542 	}
2543 #ifndef COMPAT_43
2544 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2545 								 data,
2546 								 ifp, td));
2547 	if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL)
2548 		error = (*ifp->if_ioctl)(ifp, cmd, data);
2549 #else
2550 	{
2551 		u_long ocmd = cmd;
2552 
2553 		switch (cmd) {
2554 
2555 		case SIOCSIFDSTADDR:
2556 		case SIOCSIFADDR:
2557 		case SIOCSIFBRDADDR:
2558 		case SIOCSIFNETMASK:
2559 #if BYTE_ORDER != BIG_ENDIAN
2560 			if (ifr->ifr_addr.sa_family == 0 &&
2561 			    ifr->ifr_addr.sa_len < 16) {
2562 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2563 				ifr->ifr_addr.sa_len = 16;
2564 			}
2565 #else
2566 			if (ifr->ifr_addr.sa_len == 0)
2567 				ifr->ifr_addr.sa_len = 16;
2568 #endif
2569 			break;
2570 
2571 		case OSIOCGIFADDR:
2572 			cmd = SIOCGIFADDR;
2573 			break;
2574 
2575 		case OSIOCGIFDSTADDR:
2576 			cmd = SIOCGIFDSTADDR;
2577 			break;
2578 
2579 		case OSIOCGIFBRDADDR:
2580 			cmd = SIOCGIFBRDADDR;
2581 			break;
2582 
2583 		case OSIOCGIFNETMASK:
2584 			cmd = SIOCGIFNETMASK;
2585 		}
2586 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
2587 								   cmd,
2588 								   data,
2589 								   ifp, td));
2590 		if (error == EOPNOTSUPP && ifp != NULL &&
2591 		    ifp->if_ioctl != NULL)
2592 			error = (*ifp->if_ioctl)(ifp, cmd, data);
2593 		switch (ocmd) {
2594 
2595 		case OSIOCGIFADDR:
2596 		case OSIOCGIFDSTADDR:
2597 		case OSIOCGIFBRDADDR:
2598 		case OSIOCGIFNETMASK:
2599 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2600 
2601 		}
2602 	}
2603 #endif /* COMPAT_43 */
2604 
2605 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2606 #ifdef INET6
2607 		if (ifp->if_flags & IFF_UP) {
2608 			int s = splimp();
2609 			in6_if_up(ifp);
2610 			splx(s);
2611 		}
2612 #endif
2613 	}
2614 	if_rele(ifp);
2615 	CURVNET_RESTORE();
2616 	return (error);
2617 }
2618 
2619 /*
2620  * The code common to handling reference counted flags,
2621  * e.g., in ifpromisc() and if_allmulti().
2622  * The "pflag" argument can specify a permanent mode flag to check,
2623  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2624  *
2625  * Only to be used on stack-owned flags, not driver-owned flags.
2626  */
2627 static int
2628 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2629 {
2630 	struct ifreq ifr;
2631 	int error;
2632 	int oldflags, oldcount;
2633 
2634 	/* Sanity checks to catch programming errors */
2635 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2636 	    ("%s: setting driver-owned flag %d", __func__, flag));
2637 
2638 	if (onswitch)
2639 		KASSERT(*refcount >= 0,
2640 		    ("%s: increment negative refcount %d for flag %d",
2641 		    __func__, *refcount, flag));
2642 	else
2643 		KASSERT(*refcount > 0,
2644 		    ("%s: decrement non-positive refcount %d for flag %d",
2645 		    __func__, *refcount, flag));
2646 
2647 	/* In case this mode is permanent, just touch refcount */
2648 	if (ifp->if_flags & pflag) {
2649 		*refcount += onswitch ? 1 : -1;
2650 		return (0);
2651 	}
2652 
2653 	/* Save ifnet parameters for if_ioctl() may fail */
2654 	oldcount = *refcount;
2655 	oldflags = ifp->if_flags;
2656 
2657 	/*
2658 	 * See if we aren't the only and touching refcount is enough.
2659 	 * Actually toggle interface flag if we are the first or last.
2660 	 */
2661 	if (onswitch) {
2662 		if ((*refcount)++)
2663 			return (0);
2664 		ifp->if_flags |= flag;
2665 	} else {
2666 		if (--(*refcount))
2667 			return (0);
2668 		ifp->if_flags &= ~flag;
2669 	}
2670 
2671 	/* Call down the driver since we've changed interface flags */
2672 	if (ifp->if_ioctl == NULL) {
2673 		error = EOPNOTSUPP;
2674 		goto recover;
2675 	}
2676 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2677 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2678 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2679 	if (error)
2680 		goto recover;
2681 	/* Notify userland that interface flags have changed */
2682 	rt_ifmsg(ifp);
2683 	return (0);
2684 
2685 recover:
2686 	/* Recover after driver error */
2687 	*refcount = oldcount;
2688 	ifp->if_flags = oldflags;
2689 	return (error);
2690 }
2691 
2692 /*
2693  * Set/clear promiscuous mode on interface ifp based on the truth value
2694  * of pswitch.  The calls are reference counted so that only the first
2695  * "on" request actually has an effect, as does the final "off" request.
2696  * Results are undefined if the "off" and "on" requests are not matched.
2697  */
2698 int
2699 ifpromisc(struct ifnet *ifp, int pswitch)
2700 {
2701 	int error;
2702 	int oldflags = ifp->if_flags;
2703 
2704 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2705 			   &ifp->if_pcount, pswitch);
2706 	/* If promiscuous mode status has changed, log a message */
2707 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2708 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2709 		    ifp->if_xname,
2710 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2711 	return (error);
2712 }
2713 
2714 /*
2715  * Return interface configuration
2716  * of system.  List may be used
2717  * in later ioctl's (above) to get
2718  * other information.
2719  */
2720 /*ARGSUSED*/
2721 static int
2722 ifconf(u_long cmd, caddr_t data)
2723 {
2724 	struct ifconf *ifc = (struct ifconf *)data;
2725 	struct ifnet *ifp;
2726 	struct ifaddr *ifa;
2727 	struct ifreq ifr;
2728 	struct sbuf *sb;
2729 	int error, full = 0, valid_len, max_len;
2730 
2731 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2732 	max_len = MAXPHYS - 1;
2733 
2734 	/* Prevent hostile input from being able to crash the system */
2735 	if (ifc->ifc_len <= 0)
2736 		return (EINVAL);
2737 
2738 again:
2739 	if (ifc->ifc_len <= max_len) {
2740 		max_len = ifc->ifc_len;
2741 		full = 1;
2742 	}
2743 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2744 	max_len = 0;
2745 	valid_len = 0;
2746 
2747 	IFNET_RLOCK();
2748 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2749 		int addrs;
2750 
2751 		/*
2752 		 * Zero the ifr_name buffer to make sure we don't
2753 		 * disclose the contents of the stack.
2754 		 */
2755 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2756 
2757 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2758 		    >= sizeof(ifr.ifr_name)) {
2759 			sbuf_delete(sb);
2760 			IFNET_RUNLOCK();
2761 			return (ENAMETOOLONG);
2762 		}
2763 
2764 		addrs = 0;
2765 		IF_ADDR_RLOCK(ifp);
2766 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2767 			struct sockaddr *sa = ifa->ifa_addr;
2768 
2769 			if (prison_if(curthread->td_ucred, sa) != 0)
2770 				continue;
2771 			addrs++;
2772 #ifdef COMPAT_43
2773 			if (cmd == OSIOCGIFCONF) {
2774 				struct osockaddr *osa =
2775 					 (struct osockaddr *)&ifr.ifr_addr;
2776 				ifr.ifr_addr = *sa;
2777 				osa->sa_family = sa->sa_family;
2778 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2779 				max_len += sizeof(ifr);
2780 			} else
2781 #endif
2782 			if (sa->sa_len <= sizeof(*sa)) {
2783 				ifr.ifr_addr = *sa;
2784 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2785 				max_len += sizeof(ifr);
2786 			} else {
2787 				sbuf_bcat(sb, &ifr,
2788 				    offsetof(struct ifreq, ifr_addr));
2789 				max_len += offsetof(struct ifreq, ifr_addr);
2790 				sbuf_bcat(sb, sa, sa->sa_len);
2791 				max_len += sa->sa_len;
2792 			}
2793 
2794 			if (sbuf_error(sb) == 0)
2795 				valid_len = sbuf_len(sb);
2796 		}
2797 		IF_ADDR_RUNLOCK(ifp);
2798 		if (addrs == 0) {
2799 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2800 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2801 			max_len += sizeof(ifr);
2802 
2803 			if (sbuf_error(sb) == 0)
2804 				valid_len = sbuf_len(sb);
2805 		}
2806 	}
2807 	IFNET_RUNLOCK();
2808 
2809 	/*
2810 	 * If we didn't allocate enough space (uncommon), try again.  If
2811 	 * we have already allocated as much space as we are allowed,
2812 	 * return what we've got.
2813 	 */
2814 	if (valid_len != max_len && !full) {
2815 		sbuf_delete(sb);
2816 		goto again;
2817 	}
2818 
2819 	ifc->ifc_len = valid_len;
2820 	sbuf_finish(sb);
2821 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2822 	sbuf_delete(sb);
2823 	return (error);
2824 }
2825 
2826 /*
2827  * Just like ifpromisc(), but for all-multicast-reception mode.
2828  */
2829 int
2830 if_allmulti(struct ifnet *ifp, int onswitch)
2831 {
2832 
2833 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2834 }
2835 
2836 struct ifmultiaddr *
2837 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2838 {
2839 	struct ifmultiaddr *ifma;
2840 
2841 	IF_ADDR_LOCK_ASSERT(ifp);
2842 
2843 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2844 		if (sa->sa_family == AF_LINK) {
2845 			if (sa_dl_equal(ifma->ifma_addr, sa))
2846 				break;
2847 		} else {
2848 			if (sa_equal(ifma->ifma_addr, sa))
2849 				break;
2850 		}
2851 	}
2852 
2853 	return ifma;
2854 }
2855 
2856 /*
2857  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2858  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2859  * the ifnet multicast address list here, so the caller must do that and
2860  * other setup work (such as notifying the device driver).  The reference
2861  * count is initialized to 1.
2862  */
2863 static struct ifmultiaddr *
2864 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2865     int mflags)
2866 {
2867 	struct ifmultiaddr *ifma;
2868 	struct sockaddr *dupsa;
2869 
2870 	ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
2871 	    M_ZERO);
2872 	if (ifma == NULL)
2873 		return (NULL);
2874 
2875 	dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
2876 	if (dupsa == NULL) {
2877 		free(ifma, M_IFMADDR);
2878 		return (NULL);
2879 	}
2880 	bcopy(sa, dupsa, sa->sa_len);
2881 	ifma->ifma_addr = dupsa;
2882 
2883 	ifma->ifma_ifp = ifp;
2884 	ifma->ifma_refcount = 1;
2885 	ifma->ifma_protospec = NULL;
2886 
2887 	if (llsa == NULL) {
2888 		ifma->ifma_lladdr = NULL;
2889 		return (ifma);
2890 	}
2891 
2892 	dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
2893 	if (dupsa == NULL) {
2894 		free(ifma->ifma_addr, M_IFMADDR);
2895 		free(ifma, M_IFMADDR);
2896 		return (NULL);
2897 	}
2898 	bcopy(llsa, dupsa, llsa->sa_len);
2899 	ifma->ifma_lladdr = dupsa;
2900 
2901 	return (ifma);
2902 }
2903 
2904 /*
2905  * if_freemulti: free ifmultiaddr structure and possibly attached related
2906  * addresses.  The caller is responsible for implementing reference
2907  * counting, notifying the driver, handling routing messages, and releasing
2908  * any dependent link layer state.
2909  */
2910 static void
2911 if_freemulti(struct ifmultiaddr *ifma)
2912 {
2913 
2914 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2915 	    ifma->ifma_refcount));
2916 	KASSERT(ifma->ifma_protospec == NULL,
2917 	    ("if_freemulti: protospec not NULL"));
2918 
2919 	if (ifma->ifma_lladdr != NULL)
2920 		free(ifma->ifma_lladdr, M_IFMADDR);
2921 	free(ifma->ifma_addr, M_IFMADDR);
2922 	free(ifma, M_IFMADDR);
2923 }
2924 
2925 /*
2926  * Register an additional multicast address with a network interface.
2927  *
2928  * - If the address is already present, bump the reference count on the
2929  *   address and return.
2930  * - If the address is not link-layer, look up a link layer address.
2931  * - Allocate address structures for one or both addresses, and attach to the
2932  *   multicast address list on the interface.  If automatically adding a link
2933  *   layer address, the protocol address will own a reference to the link
2934  *   layer address, to be freed when it is freed.
2935  * - Notify the network device driver of an addition to the multicast address
2936  *   list.
2937  *
2938  * 'sa' points to caller-owned memory with the desired multicast address.
2939  *
2940  * 'retifma' will be used to return a pointer to the resulting multicast
2941  * address reference, if desired.
2942  */
2943 int
2944 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2945     struct ifmultiaddr **retifma)
2946 {
2947 	struct ifmultiaddr *ifma, *ll_ifma;
2948 	struct sockaddr *llsa;
2949 	int error;
2950 
2951 	/*
2952 	 * If the address is already present, return a new reference to it;
2953 	 * otherwise, allocate storage and set up a new address.
2954 	 */
2955 	IF_ADDR_WLOCK(ifp);
2956 	ifma = if_findmulti(ifp, sa);
2957 	if (ifma != NULL) {
2958 		ifma->ifma_refcount++;
2959 		if (retifma != NULL)
2960 			*retifma = ifma;
2961 		IF_ADDR_WUNLOCK(ifp);
2962 		return (0);
2963 	}
2964 
2965 	/*
2966 	 * The address isn't already present; resolve the protocol address
2967 	 * into a link layer address, and then look that up, bump its
2968 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2969 	 * returned, we will need to free it later.
2970 	 */
2971 	llsa = NULL;
2972 	ll_ifma = NULL;
2973 	if (ifp->if_resolvemulti != NULL) {
2974 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2975 		if (error)
2976 			goto unlock_out;
2977 	}
2978 
2979 	/*
2980 	 * Allocate the new address.  Don't hook it up yet, as we may also
2981 	 * need to allocate a link layer multicast address.
2982 	 */
2983 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2984 	if (ifma == NULL) {
2985 		error = ENOMEM;
2986 		goto free_llsa_out;
2987 	}
2988 
2989 	/*
2990 	 * If a link layer address is found, we'll need to see if it's
2991 	 * already present in the address list, or allocate is as well.
2992 	 * When this block finishes, the link layer address will be on the
2993 	 * list.
2994 	 */
2995 	if (llsa != NULL) {
2996 		ll_ifma = if_findmulti(ifp, llsa);
2997 		if (ll_ifma == NULL) {
2998 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2999 			if (ll_ifma == NULL) {
3000 				--ifma->ifma_refcount;
3001 				if_freemulti(ifma);
3002 				error = ENOMEM;
3003 				goto free_llsa_out;
3004 			}
3005 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3006 			    ifma_link);
3007 		} else
3008 			ll_ifma->ifma_refcount++;
3009 		ifma->ifma_llifma = ll_ifma;
3010 	}
3011 
3012 	/*
3013 	 * We now have a new multicast address, ifma, and possibly a new or
3014 	 * referenced link layer address.  Add the primary address to the
3015 	 * ifnet address list.
3016 	 */
3017 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3018 
3019 	if (retifma != NULL)
3020 		*retifma = ifma;
3021 
3022 	/*
3023 	 * Must generate the message while holding the lock so that 'ifma'
3024 	 * pointer is still valid.
3025 	 */
3026 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3027 	IF_ADDR_WUNLOCK(ifp);
3028 
3029 	/*
3030 	 * We are certain we have added something, so call down to the
3031 	 * interface to let them know about it.
3032 	 */
3033 	if (ifp->if_ioctl != NULL) {
3034 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3035 	}
3036 
3037 	if (llsa != NULL)
3038 		free(llsa, M_IFMADDR);
3039 
3040 	return (0);
3041 
3042 free_llsa_out:
3043 	if (llsa != NULL)
3044 		free(llsa, M_IFMADDR);
3045 
3046 unlock_out:
3047 	IF_ADDR_WUNLOCK(ifp);
3048 	return (error);
3049 }
3050 
3051 /*
3052  * Delete a multicast group membership by network-layer group address.
3053  *
3054  * Returns ENOENT if the entry could not be found. If ifp no longer
3055  * exists, results are undefined. This entry point should only be used
3056  * from subsystems which do appropriate locking to hold ifp for the
3057  * duration of the call.
3058  * Network-layer protocol domains must use if_delmulti_ifma().
3059  */
3060 int
3061 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3062 {
3063 	struct ifmultiaddr *ifma;
3064 	int lastref;
3065 #ifdef INVARIANTS
3066 	struct ifnet *oifp;
3067 
3068 	IFNET_RLOCK_NOSLEEP();
3069 	TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3070 		if (ifp == oifp)
3071 			break;
3072 	if (ifp != oifp)
3073 		ifp = NULL;
3074 	IFNET_RUNLOCK_NOSLEEP();
3075 
3076 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3077 #endif
3078 	if (ifp == NULL)
3079 		return (ENOENT);
3080 
3081 	IF_ADDR_WLOCK(ifp);
3082 	lastref = 0;
3083 	ifma = if_findmulti(ifp, sa);
3084 	if (ifma != NULL)
3085 		lastref = if_delmulti_locked(ifp, ifma, 0);
3086 	IF_ADDR_WUNLOCK(ifp);
3087 
3088 	if (ifma == NULL)
3089 		return (ENOENT);
3090 
3091 	if (lastref && ifp->if_ioctl != NULL) {
3092 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3093 	}
3094 
3095 	return (0);
3096 }
3097 
3098 /*
3099  * Delete all multicast group membership for an interface.
3100  * Should be used to quickly flush all multicast filters.
3101  */
3102 void
3103 if_delallmulti(struct ifnet *ifp)
3104 {
3105 	struct ifmultiaddr *ifma;
3106 	struct ifmultiaddr *next;
3107 
3108 	IF_ADDR_WLOCK(ifp);
3109 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3110 		if_delmulti_locked(ifp, ifma, 0);
3111 	IF_ADDR_WUNLOCK(ifp);
3112 }
3113 
3114 /*
3115  * Delete a multicast group membership by group membership pointer.
3116  * Network-layer protocol domains must use this routine.
3117  *
3118  * It is safe to call this routine if the ifp disappeared.
3119  */
3120 void
3121 if_delmulti_ifma(struct ifmultiaddr *ifma)
3122 {
3123 	struct ifnet *ifp;
3124 	int lastref;
3125 
3126 	ifp = ifma->ifma_ifp;
3127 #ifdef DIAGNOSTIC
3128 	if (ifp == NULL) {
3129 		printf("%s: ifma_ifp seems to be detached\n", __func__);
3130 	} else {
3131 		struct ifnet *oifp;
3132 
3133 		IFNET_RLOCK_NOSLEEP();
3134 		TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3135 			if (ifp == oifp)
3136 				break;
3137 		if (ifp != oifp) {
3138 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
3139 			ifp = NULL;
3140 		}
3141 		IFNET_RUNLOCK_NOSLEEP();
3142 	}
3143 #endif
3144 	/*
3145 	 * If and only if the ifnet instance exists: Acquire the address lock.
3146 	 */
3147 	if (ifp != NULL)
3148 		IF_ADDR_WLOCK(ifp);
3149 
3150 	lastref = if_delmulti_locked(ifp, ifma, 0);
3151 
3152 	if (ifp != NULL) {
3153 		/*
3154 		 * If and only if the ifnet instance exists:
3155 		 *  Release the address lock.
3156 		 *  If the group was left: update the hardware hash filter.
3157 		 */
3158 		IF_ADDR_WUNLOCK(ifp);
3159 		if (lastref && ifp->if_ioctl != NULL) {
3160 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3161 		}
3162 	}
3163 }
3164 
3165 /*
3166  * Perform deletion of network-layer and/or link-layer multicast address.
3167  *
3168  * Return 0 if the reference count was decremented.
3169  * Return 1 if the final reference was released, indicating that the
3170  * hardware hash filter should be reprogrammed.
3171  */
3172 static int
3173 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3174 {
3175 	struct ifmultiaddr *ll_ifma;
3176 
3177 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
3178 		KASSERT(ifma->ifma_ifp == ifp,
3179 		    ("%s: inconsistent ifp %p", __func__, ifp));
3180 		IF_ADDR_WLOCK_ASSERT(ifp);
3181 	}
3182 
3183 	ifp = ifma->ifma_ifp;
3184 
3185 	/*
3186 	 * If the ifnet is detaching, null out references to ifnet,
3187 	 * so that upper protocol layers will notice, and not attempt
3188 	 * to obtain locks for an ifnet which no longer exists. The
3189 	 * routing socket announcement must happen before the ifnet
3190 	 * instance is detached from the system.
3191 	 */
3192 	if (detaching) {
3193 #ifdef DIAGNOSTIC
3194 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3195 #endif
3196 		/*
3197 		 * ifp may already be nulled out if we are being reentered
3198 		 * to delete the ll_ifma.
3199 		 */
3200 		if (ifp != NULL) {
3201 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
3202 			ifma->ifma_ifp = NULL;
3203 		}
3204 	}
3205 
3206 	if (--ifma->ifma_refcount > 0)
3207 		return 0;
3208 
3209 	/*
3210 	 * If this ifma is a network-layer ifma, a link-layer ifma may
3211 	 * have been associated with it. Release it first if so.
3212 	 */
3213 	ll_ifma = ifma->ifma_llifma;
3214 	if (ll_ifma != NULL) {
3215 		KASSERT(ifma->ifma_lladdr != NULL,
3216 		    ("%s: llifma w/o lladdr", __func__));
3217 		if (detaching)
3218 			ll_ifma->ifma_ifp = NULL;	/* XXX */
3219 		if (--ll_ifma->ifma_refcount == 0) {
3220 			if (ifp != NULL) {
3221 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3222 				    ifma_link);
3223 			}
3224 			if_freemulti(ll_ifma);
3225 		}
3226 	}
3227 
3228 	if (ifp != NULL)
3229 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3230 
3231 	if_freemulti(ifma);
3232 
3233 	/*
3234 	 * The last reference to this instance of struct ifmultiaddr
3235 	 * was released; the hardware should be notified of this change.
3236 	 */
3237 	return 1;
3238 }
3239 
3240 /*
3241  * Set the link layer address on an interface.
3242  *
3243  * At this time we only support certain types of interfaces,
3244  * and we don't allow the length of the address to change.
3245  */
3246 int
3247 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3248 {
3249 	struct sockaddr_dl *sdl;
3250 	struct ifaddr *ifa;
3251 	struct ifreq ifr;
3252 
3253 	IF_ADDR_RLOCK(ifp);
3254 	ifa = ifp->if_addr;
3255 	if (ifa == NULL) {
3256 		IF_ADDR_RUNLOCK(ifp);
3257 		return (EINVAL);
3258 	}
3259 	ifa_ref(ifa);
3260 	IF_ADDR_RUNLOCK(ifp);
3261 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3262 	if (sdl == NULL) {
3263 		ifa_free(ifa);
3264 		return (EINVAL);
3265 	}
3266 	if (len != sdl->sdl_alen) {	/* don't allow length to change */
3267 		ifa_free(ifa);
3268 		return (EINVAL);
3269 	}
3270 	switch (ifp->if_type) {
3271 	case IFT_ETHER:
3272 	case IFT_FDDI:
3273 	case IFT_XETHER:
3274 	case IFT_ISO88025:
3275 	case IFT_L2VLAN:
3276 	case IFT_BRIDGE:
3277 	case IFT_ARCNET:
3278 	case IFT_IEEE8023ADLAG:
3279 	case IFT_IEEE80211:
3280 		bcopy(lladdr, LLADDR(sdl), len);
3281 		ifa_free(ifa);
3282 		break;
3283 	default:
3284 		ifa_free(ifa);
3285 		return (ENODEV);
3286 	}
3287 
3288 	/*
3289 	 * If the interface is already up, we need
3290 	 * to re-init it in order to reprogram its
3291 	 * address filter.
3292 	 */
3293 	if ((ifp->if_flags & IFF_UP) != 0) {
3294 		if (ifp->if_ioctl) {
3295 			ifp->if_flags &= ~IFF_UP;
3296 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3297 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3298 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3299 			ifp->if_flags |= IFF_UP;
3300 			ifr.ifr_flags = ifp->if_flags & 0xffff;
3301 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
3302 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3303 		}
3304 #ifdef INET
3305 		/*
3306 		 * Also send gratuitous ARPs to notify other nodes about
3307 		 * the address change.
3308 		 */
3309 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3310 			if (ifa->ifa_addr->sa_family == AF_INET)
3311 				arp_ifinit(ifp, ifa);
3312 		}
3313 #endif
3314 	}
3315 	return (0);
3316 }
3317 
3318 /*
3319  * The name argument must be a pointer to storage which will last as
3320  * long as the interface does.  For physical devices, the result of
3321  * device_get_name(dev) is a good choice and for pseudo-devices a
3322  * static string works well.
3323  */
3324 void
3325 if_initname(struct ifnet *ifp, const char *name, int unit)
3326 {
3327 	ifp->if_dname = name;
3328 	ifp->if_dunit = unit;
3329 	if (unit != IF_DUNIT_NONE)
3330 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3331 	else
3332 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
3333 }
3334 
3335 int
3336 if_printf(struct ifnet *ifp, const char * fmt, ...)
3337 {
3338 	va_list ap;
3339 	int retval;
3340 
3341 	retval = printf("%s: ", ifp->if_xname);
3342 	va_start(ap, fmt);
3343 	retval += vprintf(fmt, ap);
3344 	va_end(ap);
3345 	return (retval);
3346 }
3347 
3348 void
3349 if_start(struct ifnet *ifp)
3350 {
3351 
3352 	(*(ifp)->if_start)(ifp);
3353 }
3354 
3355 /*
3356  * Backwards compatibility interface for drivers
3357  * that have not implemented it
3358  */
3359 static int
3360 if_transmit(struct ifnet *ifp, struct mbuf *m)
3361 {
3362 	int error;
3363 
3364 	IFQ_HANDOFF(ifp, m, error);
3365 	return (error);
3366 }
3367 
3368 int
3369 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3370 {
3371 	int active = 0;
3372 
3373 	IF_LOCK(ifq);
3374 	if (_IF_QFULL(ifq)) {
3375 		_IF_DROP(ifq);
3376 		IF_UNLOCK(ifq);
3377 		m_freem(m);
3378 		return (0);
3379 	}
3380 	if (ifp != NULL) {
3381 		ifp->if_obytes += m->m_pkthdr.len + adjust;
3382 		if (m->m_flags & (M_BCAST|M_MCAST))
3383 			ifp->if_omcasts++;
3384 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3385 	}
3386 	_IF_ENQUEUE(ifq, m);
3387 	IF_UNLOCK(ifq);
3388 	if (ifp != NULL && !active)
3389 		(*(ifp)->if_start)(ifp);
3390 	return (1);
3391 }
3392 
3393 void
3394 if_register_com_alloc(u_char type,
3395     if_com_alloc_t *a, if_com_free_t *f)
3396 {
3397 
3398 	KASSERT(if_com_alloc[type] == NULL,
3399 	    ("if_register_com_alloc: %d already registered", type));
3400 	KASSERT(if_com_free[type] == NULL,
3401 	    ("if_register_com_alloc: %d free already registered", type));
3402 
3403 	if_com_alloc[type] = a;
3404 	if_com_free[type] = f;
3405 }
3406 
3407 void
3408 if_deregister_com_alloc(u_char type)
3409 {
3410 
3411 	KASSERT(if_com_alloc[type] != NULL,
3412 	    ("if_deregister_com_alloc: %d not registered", type));
3413 	KASSERT(if_com_free[type] != NULL,
3414 	    ("if_deregister_com_alloc: %d free not registered", type));
3415 	if_com_alloc[type] = NULL;
3416 	if_com_free[type] = NULL;
3417 }
3418