xref: /freebsd/sys/net/if.c (revision 4fd2d3b6927878771635a3628ae1623daf810d39)
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 #include "opt_mac.h"
37 #include "opt_carp.h"
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/conf.h>
42 #include <sys/malloc.h>
43 #include <sys/sbuf.h>
44 #include <sys/bus.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/protosw.h>
52 #include <sys/kernel.h>
53 #include <sys/sockio.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
56 #include <sys/taskqueue.h>
57 #include <sys/domain.h>
58 #include <sys/jail.h>
59 #include <machine/stdarg.h>
60 
61 #include <net/if.h>
62 #include <net/if_clone.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/if_var.h>
66 #include <net/radix.h>
67 #include <net/route.h>
68 
69 #if defined(INET) || defined(INET6)
70 /*XXX*/
71 #include <netinet/in.h>
72 #include <netinet/in_var.h>
73 #ifdef INET6
74 #include <netinet6/in6_var.h>
75 #include <netinet6/in6_ifattach.h>
76 #endif
77 #endif
78 #ifdef INET
79 #include <netinet/if_ether.h>
80 #endif
81 #ifdef DEV_CARP
82 #include <netinet/ip_carp.h>
83 #endif
84 
85 #include <security/mac/mac_framework.h>
86 
87 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
88 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
89 
90 /* Log link state change events */
91 static int log_link_state_change = 1;
92 
93 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
94 	&log_link_state_change, 0,
95 	"log interface link state change events");
96 
97 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
98 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
99 void	(*lagg_linkstate_p)(struct ifnet *ifp, int state);
100 
101 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
102 
103 /*
104  * XXX: Style; these should be sorted alphabetically, and unprototyped
105  * static functions should be prototyped. Currently they are sorted by
106  * declaration order.
107  */
108 static void	if_attachdomain(void *);
109 static void	if_attachdomain1(struct ifnet *);
110 static void	if_purgemaddrs(struct ifnet *);
111 static int	ifconf(u_long, caddr_t);
112 static void	if_freemulti(struct ifmultiaddr *);
113 static void	if_grow(void);
114 static void	if_init(void *);
115 static void	if_check(void *);
116 static void	if_qflush(struct ifaltq *);
117 static void	if_route(struct ifnet *, int flag, int fam);
118 static int	if_setflag(struct ifnet *, int, int, int *, int);
119 static void	if_slowtimo(void *);
120 static void	if_unroute(struct ifnet *, int flag, int fam);
121 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
122 static int	if_rtdel(struct radix_node *, void *);
123 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
124 static int	if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
125 static void	if_start_deferred(void *context, int pending);
126 static void	do_link_state_change(void *, int);
127 static int	if_getgroup(struct ifgroupreq *, struct ifnet *);
128 static int	if_getgroupmembers(struct ifgroupreq *);
129 #ifdef INET6
130 /*
131  * XXX: declare here to avoid to include many inet6 related files..
132  * should be more generalized?
133  */
134 extern void	nd6_setmtu(struct ifnet *);
135 #endif
136 
137 int	if_index = 0;
138 struct	ifindex_entry *ifindex_table = NULL;
139 int	ifqmaxlen = IFQ_MAXLEN;
140 struct	ifnethead ifnet;	/* depend on static init XXX */
141 struct	ifgrouphead ifg_head;
142 struct	mtx ifnet_lock;
143 static	if_com_alloc_t *if_com_alloc[256];
144 static	if_com_free_t *if_com_free[256];
145 
146 static int	if_indexlim = 8;
147 static struct	knlist ifklist;
148 
149 static void	filt_netdetach(struct knote *kn);
150 static int	filt_netdev(struct knote *kn, long hint);
151 
152 static struct filterops netdev_filtops =
153     { 1, NULL, filt_netdetach, filt_netdev };
154 
155 /*
156  * System initialization
157  */
158 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
159 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_check, NULL);
160 
161 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
162 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
163 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
164 
165 static d_open_t		netopen;
166 static d_close_t	netclose;
167 static d_ioctl_t	netioctl;
168 static d_kqfilter_t	netkqfilter;
169 
170 static struct cdevsw net_cdevsw = {
171 	.d_version =	D_VERSION,
172 	.d_flags =	D_NEEDGIANT,
173 	.d_open =	netopen,
174 	.d_close =	netclose,
175 	.d_ioctl =	netioctl,
176 	.d_name =	"net",
177 	.d_kqfilter =	netkqfilter,
178 };
179 
180 static int
181 netopen(struct cdev *dev, int flag, int mode, struct thread *td)
182 {
183 	return (0);
184 }
185 
186 static int
187 netclose(struct cdev *dev, int flags, int fmt, struct thread *td)
188 {
189 	return (0);
190 }
191 
192 static int
193 netioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
194 {
195 	struct ifnet *ifp;
196 	int error, idx;
197 
198 	/* only support interface specific ioctls */
199 	if (IOCGROUP(cmd) != 'i')
200 		return (EOPNOTSUPP);
201 	idx = minor(dev);
202 	if (idx == 0) {
203 		/*
204 		 * special network device, not interface.
205 		 */
206 		if (cmd == SIOCGIFCONF)
207 			return (ifconf(cmd, data));	/* XXX remove cmd */
208 #ifdef __amd64__
209 		if (cmd == SIOCGIFCONF32)
210 			return (ifconf(cmd, data));	/* XXX remove cmd */
211 #endif
212 		return (EOPNOTSUPP);
213 	}
214 
215 	ifp = ifnet_byindex(idx);
216 	if (ifp == NULL)
217 		return (ENXIO);
218 
219 	error = ifhwioctl(cmd, ifp, data, td);
220 	if (error == ENOIOCTL)
221 		error = EOPNOTSUPP;
222 	return (error);
223 }
224 
225 static int
226 netkqfilter(struct cdev *dev, struct knote *kn)
227 {
228 	struct knlist *klist;
229 	struct ifnet *ifp;
230 	int idx;
231 
232 	switch (kn->kn_filter) {
233 	case EVFILT_NETDEV:
234 		kn->kn_fop = &netdev_filtops;
235 		break;
236 	default:
237 		return (EINVAL);
238 	}
239 
240 	idx = minor(dev);
241 	if (idx == 0) {
242 		klist = &ifklist;
243 	} else {
244 		ifp = ifnet_byindex(idx);
245 		if (ifp == NULL)
246 			return (1);
247 		klist = &ifp->if_klist;
248 	}
249 
250 	kn->kn_hook = (caddr_t)klist;
251 
252 	knlist_add(klist, kn, 0);
253 
254 	return (0);
255 }
256 
257 static void
258 filt_netdetach(struct knote *kn)
259 {
260 	struct knlist *klist = (struct knlist *)kn->kn_hook;
261 
262 	knlist_remove(klist, kn, 0);
263 }
264 
265 static int
266 filt_netdev(struct knote *kn, long hint)
267 {
268 	struct knlist *klist = (struct knlist *)kn->kn_hook;
269 
270 	/*
271 	 * Currently NOTE_EXIT is abused to indicate device detach.
272 	 */
273 	if (hint == NOTE_EXIT) {
274 		kn->kn_data = NOTE_LINKINV;
275 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
276 		knlist_remove_inevent(klist, kn);
277 		return (1);
278 	}
279 	if (hint != 0)
280 		kn->kn_data = hint;			/* current status */
281 	if (kn->kn_sfflags & hint)
282 		kn->kn_fflags |= hint;
283 	return (kn->kn_fflags != 0);
284 }
285 
286 /*
287  * Network interface utility routines.
288  *
289  * Routines with ifa_ifwith* names take sockaddr *'s as
290  * parameters.
291  */
292 
293 /* ARGSUSED*/
294 static void
295 if_init(void *dummy __unused)
296 {
297 
298 	IFNET_LOCK_INIT();
299 	TAILQ_INIT(&ifnet);
300 	TAILQ_INIT(&ifg_head);
301 	knlist_init(&ifklist, NULL, NULL, NULL, NULL);
302 	if_grow();				/* create initial table */
303 	ifdev_byindex(0) = make_dev(&net_cdevsw, 0,
304 	    UID_ROOT, GID_WHEEL, 0600, "network");
305 	if_clone_init();
306 }
307 
308 static void
309 if_grow(void)
310 {
311 	u_int n;
312 	struct ifindex_entry *e;
313 
314 	if_indexlim <<= 1;
315 	n = if_indexlim * sizeof(*e);
316 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
317 	if (ifindex_table != NULL) {
318 		memcpy((caddr_t)e, (caddr_t)ifindex_table, n/2);
319 		free((caddr_t)ifindex_table, M_IFNET);
320 	}
321 	ifindex_table = e;
322 }
323 
324 /* ARGSUSED*/
325 static void
326 if_check(void *dummy __unused)
327 {
328 	struct ifnet *ifp;
329 	int s;
330 
331 	s = splimp();
332 	IFNET_RLOCK();	/* could sleep on rare error; mostly okay XXX */
333 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
334 		if (ifp->if_snd.ifq_maxlen == 0) {
335 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
336 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
337 		}
338 		if (!mtx_initialized(&ifp->if_snd.ifq_mtx)) {
339 			if_printf(ifp,
340 			    "XXX: driver didn't initialize queue mtx\n");
341 			mtx_init(&ifp->if_snd.ifq_mtx, "unknown",
342 			    MTX_NETWORK_LOCK, MTX_DEF);
343 		}
344 	}
345 	IFNET_RUNLOCK();
346 	splx(s);
347 	if_slowtimo(0);
348 }
349 
350 /*
351  * Allocate a struct ifnet and an index for an interface.  A layer 2
352  * common structure will also be allocated if an allocation routine is
353  * registered for the passed type.
354  */
355 struct ifnet*
356 if_alloc(u_char type)
357 {
358 	struct ifnet *ifp;
359 
360 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
361 
362 	/*
363 	 * Try to find an empty slot below if_index.  If we fail, take
364 	 * the next slot.
365 	 *
366 	 * XXX: should be locked!
367 	 */
368 	for (ifp->if_index = 1; ifp->if_index <= if_index; ifp->if_index++) {
369 		if (ifnet_byindex(ifp->if_index) == NULL)
370 			break;
371 	}
372 	/* Catch if_index overflow. */
373 	if (ifp->if_index < 1) {
374 		free(ifp, M_IFNET);
375 		return (NULL);
376 	}
377 	if (ifp->if_index > if_index)
378 		if_index = ifp->if_index;
379 	if (if_index >= if_indexlim)
380 		if_grow();
381 	ifnet_byindex(ifp->if_index) = ifp;
382 
383 	ifp->if_type = type;
384 
385 	if (if_com_alloc[type] != NULL) {
386 		ifp->if_l2com = if_com_alloc[type](type, ifp);
387 		if (ifp->if_l2com == NULL) {
388 			free(ifp, M_IFNET);
389 			return (NULL);
390 		}
391 	}
392 	IF_ADDR_LOCK_INIT(ifp);
393 
394 	return (ifp);
395 }
396 
397 /*
398  * Free the struct ifnet, the associated index, and the layer 2 common
399  * structure if needed.  All the work is done in if_free_type().
400  *
401  * Do not add code to this function!  Add it to if_free_type().
402  */
403 void
404 if_free(struct ifnet *ifp)
405 {
406 
407 	if_free_type(ifp, ifp->if_type);
408 }
409 
410 /*
411  * Do the actual work of freeing a struct ifnet, associated index, and
412  * layer 2 common structure.  This version should only be called by
413  * intefaces that switch their type after calling if_alloc().
414  */
415 void
416 if_free_type(struct ifnet *ifp, u_char type)
417 {
418 
419 	if (ifp != ifnet_byindex(ifp->if_index)) {
420 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
421 		    __func__);
422 		return;
423 	}
424 
425 	IF_ADDR_LOCK_DESTROY(ifp);
426 
427 	ifnet_byindex(ifp->if_index) = NULL;
428 
429 	/* XXX: should be locked with if_findindex() */
430 	while (if_index > 0 && ifnet_byindex(if_index) == NULL)
431 		if_index--;
432 
433 	if (if_com_free[type] != NULL)
434 		if_com_free[type](ifp->if_l2com, type);
435 
436 	free(ifp, M_IFNET);
437 };
438 
439 /*
440  * Perform generic interface initalization tasks and attach the interface
441  * to the list of "active" interfaces.
442  *
443  * XXX:
444  *  - The decision to return void and thus require this function to
445  *    succeed is questionable.
446  *  - We do more initialization here then is probably a good idea.
447  *    Some of this should probably move to if_alloc().
448  *  - We should probably do more sanity checking.  For instance we don't
449  *    do anything to insure if_xname is unique or non-empty.
450  */
451 void
452 if_attach(struct ifnet *ifp)
453 {
454 	unsigned socksize, ifasize;
455 	int namelen, masklen;
456 	struct sockaddr_dl *sdl;
457 	struct ifaddr *ifa;
458 
459 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
460 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
461 		    ifp->if_xname);
462 
463 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
464 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
465 	IF_AFDATA_LOCK_INIT(ifp);
466 	ifp->if_afdata_initialized = 0;
467 
468 	TAILQ_INIT(&ifp->if_addrhead);
469 	TAILQ_INIT(&ifp->if_prefixhead);
470 	TAILQ_INIT(&ifp->if_multiaddrs);
471 	TAILQ_INIT(&ifp->if_groups);
472 
473 	if_addgroup(ifp, IFG_ALL);
474 
475 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
476 	getmicrotime(&ifp->if_lastchange);
477 	ifp->if_data.ifi_epoch = time_uptime;
478 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
479 
480 #ifdef MAC
481 	mac_ifnet_init(ifp);
482 	mac_ifnet_create(ifp);
483 #endif
484 
485 	ifdev_byindex(ifp->if_index) = make_dev(&net_cdevsw,
486 	    unit2minor(ifp->if_index),
487 	    UID_ROOT, GID_WHEEL, 0600, "%s/%s",
488 	    net_cdevsw.d_name, ifp->if_xname);
489 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
490 	    net_cdevsw.d_name, ifp->if_index);
491 
492 	mtx_init(&ifp->if_snd.ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
493 
494 	/*
495 	 * create a Link Level name for this device
496 	 */
497 	namelen = strlen(ifp->if_xname);
498 	/*
499 	 * Always save enough space for any possiable name so we can do
500 	 * a rename in place later.
501 	 */
502 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
503 	socksize = masklen + ifp->if_addrlen;
504 	if (socksize < sizeof(*sdl))
505 		socksize = sizeof(*sdl);
506 	socksize = roundup2(socksize, sizeof(long));
507 	ifasize = sizeof(*ifa) + 2 * socksize;
508 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
509 	IFA_LOCK_INIT(ifa);
510 	sdl = (struct sockaddr_dl *)(ifa + 1);
511 	sdl->sdl_len = socksize;
512 	sdl->sdl_family = AF_LINK;
513 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
514 	sdl->sdl_nlen = namelen;
515 	sdl->sdl_index = ifp->if_index;
516 	sdl->sdl_type = ifp->if_type;
517 	ifp->if_addr = ifa;
518 	ifa->ifa_ifp = ifp;
519 	ifa->ifa_rtrequest = link_rtrequest;
520 	ifa->ifa_addr = (struct sockaddr *)sdl;
521 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
522 	ifa->ifa_netmask = (struct sockaddr *)sdl;
523 	sdl->sdl_len = masklen;
524 	while (namelen != 0)
525 		sdl->sdl_data[--namelen] = 0xff;
526 	ifa->ifa_refcnt = 1;
527 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
528 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
529 	ifp->if_snd.altq_type = 0;
530 	ifp->if_snd.altq_disc = NULL;
531 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
532 	ifp->if_snd.altq_tbr  = NULL;
533 	ifp->if_snd.altq_ifp  = ifp;
534 
535 	IFNET_WLOCK();
536 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
537 	IFNET_WUNLOCK();
538 
539 	if (domain_init_status >= 2)
540 		if_attachdomain1(ifp);
541 
542 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
543 	devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
544 
545 	/* Announce the interface. */
546 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
547 
548 	if (ifp->if_watchdog != NULL)
549 		if_printf(ifp,
550 		    "WARNING: using obsoleted if_watchdog interface\n");
551 	if (ifp->if_flags & IFF_NEEDSGIANT)
552 		if_printf(ifp,
553 		    "WARNING: using obsoleted IFF_NEEDSGIANT flag\n");
554 }
555 
556 static void
557 if_attachdomain(void *dummy)
558 {
559 	struct ifnet *ifp;
560 	int s;
561 
562 	s = splnet();
563 	TAILQ_FOREACH(ifp, &ifnet, if_link)
564 		if_attachdomain1(ifp);
565 	splx(s);
566 }
567 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
568     if_attachdomain, NULL);
569 
570 static void
571 if_attachdomain1(struct ifnet *ifp)
572 {
573 	struct domain *dp;
574 	int s;
575 
576 	s = splnet();
577 
578 	/*
579 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
580 	 * cannot lock ifp->if_afdata initialization, entirely.
581 	 */
582 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
583 		splx(s);
584 		return;
585 	}
586 	if (ifp->if_afdata_initialized >= domain_init_status) {
587 		IF_AFDATA_UNLOCK(ifp);
588 		splx(s);
589 		printf("if_attachdomain called more than once on %s\n",
590 		    ifp->if_xname);
591 		return;
592 	}
593 	ifp->if_afdata_initialized = domain_init_status;
594 	IF_AFDATA_UNLOCK(ifp);
595 
596 	/* address family dependent data region */
597 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
598 	for (dp = domains; dp; dp = dp->dom_next) {
599 		if (dp->dom_ifattach)
600 			ifp->if_afdata[dp->dom_family] =
601 			    (*dp->dom_ifattach)(ifp);
602 	}
603 
604 	splx(s);
605 }
606 
607 /*
608  * Remove any unicast or broadcast network addresses from an interface.
609  */
610 void
611 if_purgeaddrs(struct ifnet *ifp)
612 {
613 	struct ifaddr *ifa, *next;
614 
615 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
616 		if (ifa->ifa_addr->sa_family == AF_LINK)
617 			continue;
618 #ifdef INET
619 		/* XXX: Ugly!! ad hoc just for INET */
620 		if (ifa->ifa_addr->sa_family == AF_INET) {
621 			struct ifaliasreq ifr;
622 
623 			bzero(&ifr, sizeof(ifr));
624 			ifr.ifra_addr = *ifa->ifa_addr;
625 			if (ifa->ifa_dstaddr)
626 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
627 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
628 			    NULL) == 0)
629 				continue;
630 		}
631 #endif /* INET */
632 #ifdef INET6
633 		if (ifa->ifa_addr->sa_family == AF_INET6) {
634 			in6_purgeaddr(ifa);
635 			/* ifp_addrhead is already updated */
636 			continue;
637 		}
638 #endif /* INET6 */
639 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
640 		IFAFREE(ifa);
641 	}
642 }
643 
644 /*
645  * Remove any multicast network addresses from an interface.
646  */
647 static void
648 if_purgemaddrs(struct ifnet *ifp)
649 {
650 	struct ifmultiaddr *ifma;
651 	struct ifmultiaddr *next;
652 
653 	IF_ADDR_LOCK(ifp);
654 	TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
655 		if_delmulti_locked(ifp, ifma, 1);
656 	IF_ADDR_UNLOCK(ifp);
657 }
658 
659 /*
660  * Detach an interface, removing it from the
661  * list of "active" interfaces.
662  *
663  * XXXRW: There are some significant questions about event ordering, and
664  * how to prevent things from starting to use the interface during detach.
665  */
666 void
667 if_detach(struct ifnet *ifp)
668 {
669 	struct ifaddr *ifa;
670 	struct radix_node_head	*rnh;
671 	int s;
672 	int i;
673 	struct domain *dp;
674  	struct ifnet *iter;
675  	int found = 0;
676 
677 	IFNET_WLOCK();
678 	TAILQ_FOREACH(iter, &ifnet, if_link)
679 		if (iter == ifp) {
680 			TAILQ_REMOVE(&ifnet, ifp, if_link);
681 			found = 1;
682 			break;
683 		}
684 	IFNET_WUNLOCK();
685 	if (!found)
686 		return;
687 
688 	/*
689 	 * Remove/wait for pending events.
690 	 */
691 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
692 
693 	/*
694 	 * Remove routes and flush queues.
695 	 */
696 	s = splnet();
697 	if_down(ifp);
698 #ifdef ALTQ
699 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
700 		altq_disable(&ifp->if_snd);
701 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
702 		altq_detach(&ifp->if_snd);
703 #endif
704 
705 	if_purgeaddrs(ifp);
706 
707 #ifdef INET
708 	in_ifdetach(ifp);
709 #endif
710 
711 #ifdef INET6
712 	/*
713 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
714 	 * before removing routing entries below, since IPv6 interface direct
715 	 * routes are expected to be removed by the IPv6-specific kernel API.
716 	 * Otherwise, the kernel will detect some inconsistency and bark it.
717 	 */
718 	in6_ifdetach(ifp);
719 #endif
720 	if_purgemaddrs(ifp);
721 
722 	/*
723 	 * Remove link ifaddr pointer and maybe decrement if_index.
724 	 * Clean up all addresses.
725 	 */
726 	ifp->if_addr = NULL;
727 	destroy_dev(ifdev_byindex(ifp->if_index));
728 	ifdev_byindex(ifp->if_index) = NULL;
729 
730 	/* We can now free link ifaddr. */
731 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
732 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
733 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
734 		IFAFREE(ifa);
735 	}
736 
737 	/*
738 	 * Delete all remaining routes using this interface
739 	 * Unfortuneatly the only way to do this is to slog through
740 	 * the entire routing table looking for routes which point
741 	 * to this interface...oh well...
742 	 */
743 	for (i = 1; i <= AF_MAX; i++) {
744 		if ((rnh = rt_tables[i]) == NULL)
745 			continue;
746 		RADIX_NODE_HEAD_LOCK(rnh);
747 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
748 		RADIX_NODE_HEAD_UNLOCK(rnh);
749 	}
750 
751 	/* Announce that the interface is gone. */
752 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
753 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
754 	devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
755 
756 	IF_AFDATA_LOCK(ifp);
757 	for (dp = domains; dp; dp = dp->dom_next) {
758 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
759 			(*dp->dom_ifdetach)(ifp,
760 			    ifp->if_afdata[dp->dom_family]);
761 	}
762 	IF_AFDATA_UNLOCK(ifp);
763 
764 #ifdef MAC
765 	mac_ifnet_destroy(ifp);
766 #endif /* MAC */
767 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
768 	knlist_clear(&ifp->if_klist, 0);
769 	knlist_destroy(&ifp->if_klist);
770 	mtx_destroy(&ifp->if_snd.ifq_mtx);
771 	IF_AFDATA_DESTROY(ifp);
772 	splx(s);
773 }
774 
775 /*
776  * Add a group to an interface
777  */
778 int
779 if_addgroup(struct ifnet *ifp, const char *groupname)
780 {
781 	struct ifg_list		*ifgl;
782 	struct ifg_group	*ifg = NULL;
783 	struct ifg_member	*ifgm;
784 
785 	if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
786 	    groupname[strlen(groupname) - 1] <= '9')
787 		return (EINVAL);
788 
789 	IFNET_WLOCK();
790 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
791 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
792 			IFNET_WUNLOCK();
793 			return (EEXIST);
794 		}
795 
796 	if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
797 	    M_NOWAIT)) == NULL) {
798 	    	IFNET_WUNLOCK();
799 		return (ENOMEM);
800 	}
801 
802 	if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
803 	    M_TEMP, M_NOWAIT)) == NULL) {
804 		free(ifgl, M_TEMP);
805 		IFNET_WUNLOCK();
806 		return (ENOMEM);
807 	}
808 
809 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
810 		if (!strcmp(ifg->ifg_group, groupname))
811 			break;
812 
813 	if (ifg == NULL) {
814 		if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
815 		    M_TEMP, M_NOWAIT)) == NULL) {
816 			free(ifgl, M_TEMP);
817 			free(ifgm, M_TEMP);
818 			IFNET_WUNLOCK();
819 			return (ENOMEM);
820 		}
821 		strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
822 		ifg->ifg_refcnt = 0;
823 		TAILQ_INIT(&ifg->ifg_members);
824 		EVENTHANDLER_INVOKE(group_attach_event, ifg);
825 		TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
826 	}
827 
828 	ifg->ifg_refcnt++;
829 	ifgl->ifgl_group = ifg;
830 	ifgm->ifgm_ifp = ifp;
831 
832 	IF_ADDR_LOCK(ifp);
833 	TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
834 	TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
835 	IF_ADDR_UNLOCK(ifp);
836 
837 	IFNET_WUNLOCK();
838 
839 	EVENTHANDLER_INVOKE(group_change_event, groupname);
840 
841 	return (0);
842 }
843 
844 /*
845  * Remove a group from an interface
846  */
847 int
848 if_delgroup(struct ifnet *ifp, const char *groupname)
849 {
850 	struct ifg_list		*ifgl;
851 	struct ifg_member	*ifgm;
852 
853 	IFNET_WLOCK();
854 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
855 		if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
856 			break;
857 	if (ifgl == NULL) {
858 		IFNET_WUNLOCK();
859 		return (ENOENT);
860 	}
861 
862 	IF_ADDR_LOCK(ifp);
863 	TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
864 	IF_ADDR_UNLOCK(ifp);
865 
866 	TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
867 		if (ifgm->ifgm_ifp == ifp)
868 			break;
869 
870 	if (ifgm != NULL) {
871 		TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
872 		free(ifgm, M_TEMP);
873 	}
874 
875 	if (--ifgl->ifgl_group->ifg_refcnt == 0) {
876 		TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
877 		EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
878 		free(ifgl->ifgl_group, M_TEMP);
879 	}
880 	IFNET_WUNLOCK();
881 
882 	free(ifgl, M_TEMP);
883 
884 	EVENTHANDLER_INVOKE(group_change_event, groupname);
885 
886 	return (0);
887 }
888 
889 /*
890  * Stores all groups from an interface in memory pointed
891  * to by data
892  */
893 static int
894 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
895 {
896 	int			 len, error;
897 	struct ifg_list		*ifgl;
898 	struct ifg_req		 ifgrq, *ifgp;
899 	struct ifgroupreq	*ifgr = data;
900 
901 	if (ifgr->ifgr_len == 0) {
902 		IF_ADDR_LOCK(ifp);
903 		TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
904 			ifgr->ifgr_len += sizeof(struct ifg_req);
905 		IF_ADDR_UNLOCK(ifp);
906 		return (0);
907 	}
908 
909 	len = ifgr->ifgr_len;
910 	ifgp = ifgr->ifgr_groups;
911 	/* XXX: wire */
912 	IF_ADDR_LOCK(ifp);
913 	TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
914 		if (len < sizeof(ifgrq)) {
915 			IF_ADDR_UNLOCK(ifp);
916 			return (EINVAL);
917 		}
918 		bzero(&ifgrq, sizeof ifgrq);
919 		strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
920 		    sizeof(ifgrq.ifgrq_group));
921 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
922 		    	IF_ADDR_UNLOCK(ifp);
923 			return (error);
924 		}
925 		len -= sizeof(ifgrq);
926 		ifgp++;
927 	}
928 	IF_ADDR_UNLOCK(ifp);
929 
930 	return (0);
931 }
932 
933 /*
934  * Stores all members of a group in memory pointed to by data
935  */
936 static int
937 if_getgroupmembers(struct ifgroupreq *data)
938 {
939 	struct ifgroupreq	*ifgr = data;
940 	struct ifg_group	*ifg;
941 	struct ifg_member	*ifgm;
942 	struct ifg_req		 ifgrq, *ifgp;
943 	int			 len, error;
944 
945 	IFNET_RLOCK();
946 	TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
947 		if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
948 			break;
949 	if (ifg == NULL) {
950 		IFNET_RUNLOCK();
951 		return (ENOENT);
952 	}
953 
954 	if (ifgr->ifgr_len == 0) {
955 		TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
956 			ifgr->ifgr_len += sizeof(ifgrq);
957 		IFNET_RUNLOCK();
958 		return (0);
959 	}
960 
961 	len = ifgr->ifgr_len;
962 	ifgp = ifgr->ifgr_groups;
963 	TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
964 		if (len < sizeof(ifgrq)) {
965 			IFNET_RUNLOCK();
966 			return (EINVAL);
967 		}
968 		bzero(&ifgrq, sizeof ifgrq);
969 		strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
970 		    sizeof(ifgrq.ifgrq_member));
971 		if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
972 			IFNET_RUNLOCK();
973 			return (error);
974 		}
975 		len -= sizeof(ifgrq);
976 		ifgp++;
977 	}
978 	IFNET_RUNLOCK();
979 
980 	return (0);
981 }
982 
983 /*
984  * Delete Routes for a Network Interface
985  *
986  * Called for each routing entry via the rnh->rnh_walktree() call above
987  * to delete all route entries referencing a detaching network interface.
988  *
989  * Arguments:
990  *	rn	pointer to node in the routing table
991  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
992  *
993  * Returns:
994  *	0	successful
995  *	errno	failed - reason indicated
996  *
997  */
998 static int
999 if_rtdel(struct radix_node *rn, void *arg)
1000 {
1001 	struct rtentry	*rt = (struct rtentry *)rn;
1002 	struct ifnet	*ifp = arg;
1003 	int		err;
1004 
1005 	if (rt->rt_ifp == ifp) {
1006 
1007 		/*
1008 		 * Protect (sorta) against walktree recursion problems
1009 		 * with cloned routes
1010 		 */
1011 		if ((rt->rt_flags & RTF_UP) == 0)
1012 			return (0);
1013 
1014 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1015 				rt_mask(rt), rt->rt_flags,
1016 				(struct rtentry **) NULL);
1017 		if (err) {
1018 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
1019 		}
1020 	}
1021 
1022 	return (0);
1023 }
1024 
1025 /*
1026  * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1027  * structs used to represent other address families, it is necessary
1028  * to perform a different comparison.
1029  */
1030 
1031 #define	sa_equal(a1, a2)	\
1032 	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
1033 
1034 #define	sa_dl_equal(a1, a2)	\
1035 	((((struct sockaddr_dl *)(a1))->sdl_len ==			\
1036 	 ((struct sockaddr_dl *)(a2))->sdl_len) &&			\
1037 	 (bcmp(LLADDR((struct sockaddr_dl *)(a1)),			\
1038 	       LLADDR((struct sockaddr_dl *)(a2)),			\
1039 	       ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1040 
1041 /*
1042  * Locate an interface based on a complete address.
1043  */
1044 /*ARGSUSED*/
1045 struct ifaddr *
1046 ifa_ifwithaddr(struct sockaddr *addr)
1047 {
1048 	struct ifnet *ifp;
1049 	struct ifaddr *ifa;
1050 
1051 	IFNET_RLOCK();
1052 	TAILQ_FOREACH(ifp, &ifnet, if_link)
1053 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1054 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1055 				continue;
1056 			if (sa_equal(addr, ifa->ifa_addr))
1057 				goto done;
1058 			/* IP6 doesn't have broadcast */
1059 			if ((ifp->if_flags & IFF_BROADCAST) &&
1060 			    ifa->ifa_broadaddr &&
1061 			    ifa->ifa_broadaddr->sa_len != 0 &&
1062 			    sa_equal(ifa->ifa_broadaddr, addr))
1063 				goto done;
1064 		}
1065 	ifa = NULL;
1066 done:
1067 	IFNET_RUNLOCK();
1068 	return (ifa);
1069 }
1070 
1071 /*
1072  * Locate an interface based on the broadcast address.
1073  */
1074 /* ARGSUSED */
1075 struct ifaddr *
1076 ifa_ifwithbroadaddr(struct sockaddr *addr)
1077 {
1078 	struct ifnet *ifp;
1079 	struct ifaddr *ifa;
1080 
1081 	IFNET_RLOCK();
1082 	TAILQ_FOREACH(ifp, &ifnet, if_link)
1083 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1084 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1085 				continue;
1086 			if ((ifp->if_flags & IFF_BROADCAST) &&
1087 			    ifa->ifa_broadaddr &&
1088 			    ifa->ifa_broadaddr->sa_len != 0 &&
1089 			    sa_equal(ifa->ifa_broadaddr, addr))
1090 				goto done;
1091 		}
1092 	ifa = NULL;
1093 done:
1094 	IFNET_RUNLOCK();
1095 	return (ifa);
1096 }
1097 
1098 /*
1099  * Locate the point to point interface with a given destination address.
1100  */
1101 /*ARGSUSED*/
1102 struct ifaddr *
1103 ifa_ifwithdstaddr(struct sockaddr *addr)
1104 {
1105 	struct ifnet *ifp;
1106 	struct ifaddr *ifa;
1107 
1108 	IFNET_RLOCK();
1109 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1110 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1111 			continue;
1112 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1113 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1114 				continue;
1115 			if (ifa->ifa_dstaddr &&
1116 			    sa_equal(addr, ifa->ifa_dstaddr))
1117 				goto done;
1118 		}
1119 	}
1120 	ifa = NULL;
1121 done:
1122 	IFNET_RUNLOCK();
1123 	return (ifa);
1124 }
1125 
1126 /*
1127  * Find an interface on a specific network.  If many, choice
1128  * is most specific found.
1129  */
1130 struct ifaddr *
1131 ifa_ifwithnet(struct sockaddr *addr)
1132 {
1133 	struct ifnet *ifp;
1134 	struct ifaddr *ifa;
1135 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
1136 	u_int af = addr->sa_family;
1137 	char *addr_data = addr->sa_data, *cplim;
1138 
1139 	/*
1140 	 * AF_LINK addresses can be looked up directly by their index number,
1141 	 * so do that if we can.
1142 	 */
1143 	if (af == AF_LINK) {
1144 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1145 	    if (sdl->sdl_index && sdl->sdl_index <= if_index)
1146 		return (ifaddr_byindex(sdl->sdl_index));
1147 	}
1148 
1149 	/*
1150 	 * Scan though each interface, looking for ones that have
1151 	 * addresses in this address family.
1152 	 */
1153 	IFNET_RLOCK();
1154 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1155 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1156 			char *cp, *cp2, *cp3;
1157 
1158 			if (ifa->ifa_addr->sa_family != af)
1159 next:				continue;
1160 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1161 				/*
1162 				 * This is a bit broken as it doesn't
1163 				 * take into account that the remote end may
1164 				 * be a single node in the network we are
1165 				 * looking for.
1166 				 * The trouble is that we don't know the
1167 				 * netmask for the remote end.
1168 				 */
1169 				if (ifa->ifa_dstaddr != 0 &&
1170 				    sa_equal(addr, ifa->ifa_dstaddr))
1171 					goto done;
1172 			} else {
1173 				/*
1174 				 * if we have a special address handler,
1175 				 * then use it instead of the generic one.
1176 				 */
1177 				if (ifa->ifa_claim_addr) {
1178 					if ((*ifa->ifa_claim_addr)(ifa, addr))
1179 						goto done;
1180 					continue;
1181 				}
1182 
1183 				/*
1184 				 * Scan all the bits in the ifa's address.
1185 				 * If a bit dissagrees with what we are
1186 				 * looking for, mask it with the netmask
1187 				 * to see if it really matters.
1188 				 * (A byte at a time)
1189 				 */
1190 				if (ifa->ifa_netmask == 0)
1191 					continue;
1192 				cp = addr_data;
1193 				cp2 = ifa->ifa_addr->sa_data;
1194 				cp3 = ifa->ifa_netmask->sa_data;
1195 				cplim = ifa->ifa_netmask->sa_len
1196 					+ (char *)ifa->ifa_netmask;
1197 				while (cp3 < cplim)
1198 					if ((*cp++ ^ *cp2++) & *cp3++)
1199 						goto next; /* next address! */
1200 				/*
1201 				 * If the netmask of what we just found
1202 				 * is more specific than what we had before
1203 				 * (if we had one) then remember the new one
1204 				 * before continuing to search
1205 				 * for an even better one.
1206 				 */
1207 				if (ifa_maybe == 0 ||
1208 				    rn_refines((caddr_t)ifa->ifa_netmask,
1209 				    (caddr_t)ifa_maybe->ifa_netmask))
1210 					ifa_maybe = ifa;
1211 			}
1212 		}
1213 	}
1214 	ifa = ifa_maybe;
1215 done:
1216 	IFNET_RUNLOCK();
1217 	return (ifa);
1218 }
1219 
1220 /*
1221  * Find an interface address specific to an interface best matching
1222  * a given address.
1223  */
1224 struct ifaddr *
1225 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1226 {
1227 	struct ifaddr *ifa;
1228 	char *cp, *cp2, *cp3;
1229 	char *cplim;
1230 	struct ifaddr *ifa_maybe = 0;
1231 	u_int af = addr->sa_family;
1232 
1233 	if (af >= AF_MAX)
1234 		return (0);
1235 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1236 		if (ifa->ifa_addr->sa_family != af)
1237 			continue;
1238 		if (ifa_maybe == 0)
1239 			ifa_maybe = ifa;
1240 		if (ifa->ifa_netmask == 0) {
1241 			if (sa_equal(addr, ifa->ifa_addr) ||
1242 			    (ifa->ifa_dstaddr &&
1243 			    sa_equal(addr, ifa->ifa_dstaddr)))
1244 				goto done;
1245 			continue;
1246 		}
1247 		if (ifp->if_flags & IFF_POINTOPOINT) {
1248 			if (sa_equal(addr, ifa->ifa_dstaddr))
1249 				goto done;
1250 		} else {
1251 			cp = addr->sa_data;
1252 			cp2 = ifa->ifa_addr->sa_data;
1253 			cp3 = ifa->ifa_netmask->sa_data;
1254 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1255 			for (; cp3 < cplim; cp3++)
1256 				if ((*cp++ ^ *cp2++) & *cp3)
1257 					break;
1258 			if (cp3 == cplim)
1259 				goto done;
1260 		}
1261 	}
1262 	ifa = ifa_maybe;
1263 done:
1264 	return (ifa);
1265 }
1266 
1267 #include <net/route.h>
1268 
1269 /*
1270  * Default action when installing a route with a Link Level gateway.
1271  * Lookup an appropriate real ifa to point to.
1272  * This should be moved to /sys/net/link.c eventually.
1273  */
1274 static void
1275 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
1276 {
1277 	struct ifaddr *ifa, *oifa;
1278 	struct sockaddr *dst;
1279 	struct ifnet *ifp;
1280 
1281 	RT_LOCK_ASSERT(rt);
1282 
1283 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
1284 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
1285 		return;
1286 	ifa = ifaof_ifpforaddr(dst, ifp);
1287 	if (ifa) {
1288 		IFAREF(ifa);		/* XXX */
1289 		oifa = rt->rt_ifa;
1290 		rt->rt_ifa = ifa;
1291 		IFAFREE(oifa);
1292 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1293 			ifa->ifa_rtrequest(cmd, rt, info);
1294 	}
1295 }
1296 
1297 /*
1298  * Mark an interface down and notify protocols of
1299  * the transition.
1300  * NOTE: must be called at splnet or eqivalent.
1301  */
1302 static void
1303 if_unroute(struct ifnet *ifp, int flag, int fam)
1304 {
1305 	struct ifaddr *ifa;
1306 
1307 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1308 
1309 	ifp->if_flags &= ~flag;
1310 	getmicrotime(&ifp->if_lastchange);
1311 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1312 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1313 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1314 	if_qflush(&ifp->if_snd);
1315 #ifdef DEV_CARP
1316 	if (ifp->if_carp)
1317 		carp_carpdev_state(ifp->if_carp);
1318 #endif
1319 	rt_ifmsg(ifp);
1320 }
1321 
1322 /*
1323  * Mark an interface up and notify protocols of
1324  * the transition.
1325  * NOTE: must be called at splnet or eqivalent.
1326  */
1327 static void
1328 if_route(struct ifnet *ifp, int flag, int fam)
1329 {
1330 	struct ifaddr *ifa;
1331 
1332 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1333 
1334 	ifp->if_flags |= flag;
1335 	getmicrotime(&ifp->if_lastchange);
1336 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1337 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1338 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1339 #ifdef DEV_CARP
1340 	if (ifp->if_carp)
1341 		carp_carpdev_state(ifp->if_carp);
1342 #endif
1343 	rt_ifmsg(ifp);
1344 #ifdef INET6
1345 	in6_if_up(ifp);
1346 #endif
1347 }
1348 
1349 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1350 void	(*vlan_trunk_cap_p)(struct ifnet *);		/* XXX: private from if_vlan */
1351 
1352 /*
1353  * Handle a change in the interface link state. To avoid LORs
1354  * between driver lock and upper layer locks, as well as possible
1355  * recursions, we post event to taskqueue, and all job
1356  * is done in static do_link_state_change().
1357  */
1358 void
1359 if_link_state_change(struct ifnet *ifp, int link_state)
1360 {
1361 	/* Return if state hasn't changed. */
1362 	if (ifp->if_link_state == link_state)
1363 		return;
1364 
1365 	ifp->if_link_state = link_state;
1366 
1367 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1368 }
1369 
1370 static void
1371 do_link_state_change(void *arg, int pending)
1372 {
1373 	struct ifnet *ifp = (struct ifnet *)arg;
1374 	int link_state = ifp->if_link_state;
1375 	int link;
1376 
1377 	/* Notify that the link state has changed. */
1378 	rt_ifmsg(ifp);
1379 	if (link_state == LINK_STATE_UP)
1380 		link = NOTE_LINKUP;
1381 	else if (link_state == LINK_STATE_DOWN)
1382 		link = NOTE_LINKDOWN;
1383 	else
1384 		link = NOTE_LINKINV;
1385 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1386 	if (ifp->if_vlantrunk != NULL)
1387 		(*vlan_link_state_p)(ifp, link);
1388 
1389 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1390 	    IFP2AC(ifp)->ac_netgraph != NULL)
1391 		(*ng_ether_link_state_p)(ifp, link_state);
1392 #ifdef DEV_CARP
1393 	if (ifp->if_carp)
1394 		carp_carpdev_state(ifp->if_carp);
1395 #endif
1396 	if (ifp->if_bridge) {
1397 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1398 		(*bstp_linkstate_p)(ifp, link_state);
1399 	}
1400 	if (ifp->if_lagg) {
1401 		KASSERT(lagg_linkstate_p != NULL,("if_lagg not loaded!"));
1402 		(*lagg_linkstate_p)(ifp, link_state);
1403 	}
1404 
1405 	devctl_notify("IFNET", ifp->if_xname,
1406 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1407 	if (pending > 1)
1408 		if_printf(ifp, "%d link states coalesced\n", pending);
1409 	if (log_link_state_change)
1410 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1411 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1412 }
1413 
1414 /*
1415  * Mark an interface down and notify protocols of
1416  * the transition.
1417  * NOTE: must be called at splnet or eqivalent.
1418  */
1419 void
1420 if_down(struct ifnet *ifp)
1421 {
1422 
1423 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1424 }
1425 
1426 /*
1427  * Mark an interface up and notify protocols of
1428  * the transition.
1429  * NOTE: must be called at splnet or eqivalent.
1430  */
1431 void
1432 if_up(struct ifnet *ifp)
1433 {
1434 
1435 	if_route(ifp, IFF_UP, AF_UNSPEC);
1436 }
1437 
1438 /*
1439  * Flush an interface queue.
1440  */
1441 static void
1442 if_qflush(struct ifaltq *ifq)
1443 {
1444 	struct mbuf *m, *n;
1445 
1446 	IFQ_LOCK(ifq);
1447 #ifdef ALTQ
1448 	if (ALTQ_IS_ENABLED(ifq))
1449 		ALTQ_PURGE(ifq);
1450 #endif
1451 	n = ifq->ifq_head;
1452 	while ((m = n) != 0) {
1453 		n = m->m_act;
1454 		m_freem(m);
1455 	}
1456 	ifq->ifq_head = 0;
1457 	ifq->ifq_tail = 0;
1458 	ifq->ifq_len = 0;
1459 	IFQ_UNLOCK(ifq);
1460 }
1461 
1462 /*
1463  * Handle interface watchdog timer routines.  Called
1464  * from softclock, we decrement timers (if set) and
1465  * call the appropriate interface routine on expiration.
1466  *
1467  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1468  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1469  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1470  */
1471 static void
1472 if_slowtimo(void *arg)
1473 {
1474 	struct ifnet *ifp;
1475 	int s = splimp();
1476 
1477 	IFNET_RLOCK();
1478 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1479 		if (ifp->if_timer == 0 || --ifp->if_timer)
1480 			continue;
1481 		if (ifp->if_watchdog)
1482 			(*ifp->if_watchdog)(ifp);
1483 	}
1484 	IFNET_RUNLOCK();
1485 	splx(s);
1486 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1487 }
1488 
1489 /*
1490  * Map interface name to
1491  * interface structure pointer.
1492  */
1493 struct ifnet *
1494 ifunit(const char *name)
1495 {
1496 	struct ifnet *ifp;
1497 
1498 	IFNET_RLOCK();
1499 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1500 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1501 			break;
1502 	}
1503 	IFNET_RUNLOCK();
1504 	return (ifp);
1505 }
1506 
1507 /*
1508  * Hardware specific interface ioctls.
1509  */
1510 static int
1511 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1512 {
1513 	struct ifreq *ifr;
1514 	struct ifstat *ifs;
1515 	int error = 0;
1516 	int new_flags, temp_flags;
1517 	size_t namelen, onamelen;
1518 	char new_name[IFNAMSIZ];
1519 	struct ifaddr *ifa;
1520 	struct sockaddr_dl *sdl;
1521 
1522 	ifr = (struct ifreq *)data;
1523 	switch (cmd) {
1524 	case SIOCGIFINDEX:
1525 		ifr->ifr_index = ifp->if_index;
1526 		break;
1527 
1528 	case SIOCGIFFLAGS:
1529 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1530 		ifr->ifr_flags = temp_flags & 0xffff;
1531 		ifr->ifr_flagshigh = temp_flags >> 16;
1532 		break;
1533 
1534 	case SIOCGIFCAP:
1535 		ifr->ifr_reqcap = ifp->if_capabilities;
1536 		ifr->ifr_curcap = ifp->if_capenable;
1537 		break;
1538 
1539 #ifdef MAC
1540 	case SIOCGIFMAC:
1541 		error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
1542 		break;
1543 #endif
1544 
1545 	case SIOCGIFMETRIC:
1546 		ifr->ifr_metric = ifp->if_metric;
1547 		break;
1548 
1549 	case SIOCGIFMTU:
1550 		ifr->ifr_mtu = ifp->if_mtu;
1551 		break;
1552 
1553 	case SIOCGIFPHYS:
1554 		ifr->ifr_phys = ifp->if_physical;
1555 		break;
1556 
1557 	case SIOCSIFFLAGS:
1558 		error = priv_check(td, PRIV_NET_SETIFFLAGS);
1559 		if (error)
1560 			return (error);
1561 		/*
1562 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1563 		 * check, so we don't need special handling here yet.
1564 		 */
1565 		new_flags = (ifr->ifr_flags & 0xffff) |
1566 		    (ifr->ifr_flagshigh << 16);
1567 		if (ifp->if_flags & IFF_SMART) {
1568 			/* Smart drivers twiddle their own routes */
1569 		} else if (ifp->if_flags & IFF_UP &&
1570 		    (new_flags & IFF_UP) == 0) {
1571 			int s = splimp();
1572 			if_down(ifp);
1573 			splx(s);
1574 		} else if (new_flags & IFF_UP &&
1575 		    (ifp->if_flags & IFF_UP) == 0) {
1576 			int s = splimp();
1577 			if_up(ifp);
1578 			splx(s);
1579 		}
1580 		/* See if permanently promiscuous mode bit is about to flip */
1581 		if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
1582 			if (new_flags & IFF_PPROMISC)
1583 				ifp->if_flags |= IFF_PROMISC;
1584 			else if (ifp->if_pcount == 0)
1585 				ifp->if_flags &= ~IFF_PROMISC;
1586 			log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
1587 			    ifp->if_xname,
1588 			    (new_flags & IFF_PPROMISC) ? "enabled" : "disabled");
1589 		}
1590 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1591 			(new_flags &~ IFF_CANTCHANGE);
1592 		if (ifp->if_ioctl) {
1593 			IFF_LOCKGIANT(ifp);
1594 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1595 			IFF_UNLOCKGIANT(ifp);
1596 		}
1597 		getmicrotime(&ifp->if_lastchange);
1598 		break;
1599 
1600 	case SIOCSIFCAP:
1601 		error = priv_check(td, PRIV_NET_SETIFCAP);
1602 		if (error)
1603 			return (error);
1604 		if (ifp->if_ioctl == NULL)
1605 			return (EOPNOTSUPP);
1606 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1607 			return (EINVAL);
1608 		IFF_LOCKGIANT(ifp);
1609 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1610 		IFF_UNLOCKGIANT(ifp);
1611 		if (error == 0)
1612 			getmicrotime(&ifp->if_lastchange);
1613 		break;
1614 
1615 #ifdef MAC
1616 	case SIOCSIFMAC:
1617 		error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
1618 		break;
1619 #endif
1620 
1621 	case SIOCSIFNAME:
1622 		error = priv_check(td, PRIV_NET_SETIFNAME);
1623 		if (error)
1624 			return (error);
1625 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1626 		if (error != 0)
1627 			return (error);
1628 		if (new_name[0] == '\0')
1629 			return (EINVAL);
1630 		if (ifunit(new_name) != NULL)
1631 			return (EEXIST);
1632 
1633 		/* Announce the departure of the interface. */
1634 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1635 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1636 
1637 		log(LOG_INFO, "%s: changing name to '%s'\n",
1638 		    ifp->if_xname, new_name);
1639 
1640 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1641 		ifa = ifp->if_addr;
1642 		IFA_LOCK(ifa);
1643 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1644 		namelen = strlen(new_name);
1645 		onamelen = sdl->sdl_nlen;
1646 		/*
1647 		 * Move the address if needed.  This is safe because we
1648 		 * allocate space for a name of length IFNAMSIZ when we
1649 		 * create this in if_attach().
1650 		 */
1651 		if (namelen != onamelen) {
1652 			bcopy(sdl->sdl_data + onamelen,
1653 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1654 		}
1655 		bcopy(new_name, sdl->sdl_data, namelen);
1656 		sdl->sdl_nlen = namelen;
1657 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1658 		bzero(sdl->sdl_data, onamelen);
1659 		while (namelen != 0)
1660 			sdl->sdl_data[--namelen] = 0xff;
1661 		IFA_UNLOCK(ifa);
1662 
1663 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1664 		/* Announce the return of the interface. */
1665 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1666 		break;
1667 
1668 	case SIOCSIFMETRIC:
1669 		error = priv_check(td, PRIV_NET_SETIFMETRIC);
1670 		if (error)
1671 			return (error);
1672 		ifp->if_metric = ifr->ifr_metric;
1673 		getmicrotime(&ifp->if_lastchange);
1674 		break;
1675 
1676 	case SIOCSIFPHYS:
1677 		error = priv_check(td, PRIV_NET_SETIFPHYS);
1678 		if (error)
1679 			return (error);
1680 		if (ifp->if_ioctl == NULL)
1681 			return (EOPNOTSUPP);
1682 		IFF_LOCKGIANT(ifp);
1683 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1684 		IFF_UNLOCKGIANT(ifp);
1685 		if (error == 0)
1686 			getmicrotime(&ifp->if_lastchange);
1687 		break;
1688 
1689 	case SIOCSIFMTU:
1690 	{
1691 		u_long oldmtu = ifp->if_mtu;
1692 
1693 		error = priv_check(td, PRIV_NET_SETIFMTU);
1694 		if (error)
1695 			return (error);
1696 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1697 			return (EINVAL);
1698 		if (ifp->if_ioctl == NULL)
1699 			return (EOPNOTSUPP);
1700 		IFF_LOCKGIANT(ifp);
1701 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1702 		IFF_UNLOCKGIANT(ifp);
1703 		if (error == 0) {
1704 			getmicrotime(&ifp->if_lastchange);
1705 			rt_ifmsg(ifp);
1706 		}
1707 		/*
1708 		 * If the link MTU changed, do network layer specific procedure.
1709 		 */
1710 		if (ifp->if_mtu != oldmtu) {
1711 #ifdef INET6
1712 			nd6_setmtu(ifp);
1713 #endif
1714 		}
1715 		break;
1716 	}
1717 
1718 	case SIOCADDMULTI:
1719 	case SIOCDELMULTI:
1720 		if (cmd == SIOCADDMULTI)
1721 			error = priv_check(td, PRIV_NET_ADDMULTI);
1722 		else
1723 			error = priv_check(td, PRIV_NET_DELMULTI);
1724 		if (error)
1725 			return (error);
1726 
1727 		/* Don't allow group membership on non-multicast interfaces. */
1728 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1729 			return (EOPNOTSUPP);
1730 
1731 		/* Don't let users screw up protocols' entries. */
1732 		if (ifr->ifr_addr.sa_family != AF_LINK)
1733 			return (EINVAL);
1734 
1735 		if (cmd == SIOCADDMULTI) {
1736 			struct ifmultiaddr *ifma;
1737 
1738 			/*
1739 			 * Userland is only permitted to join groups once
1740 			 * via the if_addmulti() KPI, because it cannot hold
1741 			 * struct ifmultiaddr * between calls. It may also
1742 			 * lose a race while we check if the membership
1743 			 * already exists.
1744 			 */
1745 			IF_ADDR_LOCK(ifp);
1746 			ifma = if_findmulti(ifp, &ifr->ifr_addr);
1747 			IF_ADDR_UNLOCK(ifp);
1748 			if (ifma != NULL)
1749 				error = EADDRINUSE;
1750 			else
1751 				error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1752 		} else {
1753 			error = if_delmulti(ifp, &ifr->ifr_addr);
1754 		}
1755 		if (error == 0)
1756 			getmicrotime(&ifp->if_lastchange);
1757 		break;
1758 
1759 	case SIOCSIFPHYADDR:
1760 	case SIOCDIFPHYADDR:
1761 #ifdef INET6
1762 	case SIOCSIFPHYADDR_IN6:
1763 #endif
1764 	case SIOCSLIFPHYADDR:
1765 	case SIOCSIFMEDIA:
1766 	case SIOCSIFGENERIC:
1767 		error = priv_check(td, PRIV_NET_HWIOCTL);
1768 		if (error)
1769 			return (error);
1770 		if (ifp->if_ioctl == NULL)
1771 			return (EOPNOTSUPP);
1772 		IFF_LOCKGIANT(ifp);
1773 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1774 		IFF_UNLOCKGIANT(ifp);
1775 		if (error == 0)
1776 			getmicrotime(&ifp->if_lastchange);
1777 		break;
1778 
1779 	case SIOCGIFSTATUS:
1780 		ifs = (struct ifstat *)data;
1781 		ifs->ascii[0] = '\0';
1782 
1783 	case SIOCGIFPSRCADDR:
1784 	case SIOCGIFPDSTADDR:
1785 	case SIOCGLIFPHYADDR:
1786 	case SIOCGIFMEDIA:
1787 	case SIOCGIFGENERIC:
1788 		if (ifp->if_ioctl == NULL)
1789 			return (EOPNOTSUPP);
1790 		IFF_LOCKGIANT(ifp);
1791 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1792 		IFF_UNLOCKGIANT(ifp);
1793 		break;
1794 
1795 	case SIOCSIFLLADDR:
1796 		error = priv_check(td, PRIV_NET_SETLLADDR);
1797 		if (error)
1798 			return (error);
1799 		error = if_setlladdr(ifp,
1800 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1801 		break;
1802 
1803 	case SIOCAIFGROUP:
1804 	{
1805 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1806 
1807 		error = priv_check(td, PRIV_NET_ADDIFGROUP);
1808 		if (error)
1809 			return (error);
1810 		if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
1811 			return (error);
1812 		break;
1813 	}
1814 
1815 	case SIOCGIFGROUP:
1816 		if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
1817 			return (error);
1818 		break;
1819 
1820 	case SIOCDIFGROUP:
1821 	{
1822 		struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
1823 
1824 		error = priv_check(td, PRIV_NET_DELIFGROUP);
1825 		if (error)
1826 			return (error);
1827 		if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
1828 			return (error);
1829 		break;
1830 	}
1831 
1832 	default:
1833 		error = ENOIOCTL;
1834 		break;
1835 	}
1836 	return (error);
1837 }
1838 
1839 /*
1840  * Interface ioctls.
1841  */
1842 int
1843 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1844 {
1845 	struct ifnet *ifp;
1846 	struct ifreq *ifr;
1847 	int error;
1848 	int oif_flags;
1849 
1850 	switch (cmd) {
1851 	case SIOCGIFCONF:
1852 	case OSIOCGIFCONF:
1853 #ifdef __amd64__
1854 	case SIOCGIFCONF32:
1855 #endif
1856 		return (ifconf(cmd, data));
1857 	}
1858 	ifr = (struct ifreq *)data;
1859 
1860 	switch (cmd) {
1861 	case SIOCIFCREATE:
1862 	case SIOCIFCREATE2:
1863 		error = priv_check(td, PRIV_NET_IFCREATE);
1864 		if (error)
1865 			return (error);
1866 		return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1867 			cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1868 	case SIOCIFDESTROY:
1869 		error = priv_check(td, PRIV_NET_IFDESTROY);
1870 		if (error)
1871 			return (error);
1872 		return if_clone_destroy(ifr->ifr_name);
1873 
1874 	case SIOCIFGCLONERS:
1875 		return (if_clone_list((struct if_clonereq *)data));
1876 	case SIOCGIFGMEMB:
1877 		return (if_getgroupmembers((struct ifgroupreq *)data));
1878 	}
1879 
1880 	ifp = ifunit(ifr->ifr_name);
1881 	if (ifp == 0)
1882 		return (ENXIO);
1883 
1884 	error = ifhwioctl(cmd, ifp, data, td);
1885 	if (error != ENOIOCTL)
1886 		return (error);
1887 
1888 	oif_flags = ifp->if_flags;
1889 	if (so->so_proto == 0)
1890 		return (EOPNOTSUPP);
1891 #ifndef COMPAT_43
1892 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1893 								 data,
1894 								 ifp, td));
1895 #else
1896 	{
1897 		int ocmd = cmd;
1898 
1899 		switch (cmd) {
1900 
1901 		case SIOCSIFDSTADDR:
1902 		case SIOCSIFADDR:
1903 		case SIOCSIFBRDADDR:
1904 		case SIOCSIFNETMASK:
1905 #if BYTE_ORDER != BIG_ENDIAN
1906 			if (ifr->ifr_addr.sa_family == 0 &&
1907 			    ifr->ifr_addr.sa_len < 16) {
1908 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1909 				ifr->ifr_addr.sa_len = 16;
1910 			}
1911 #else
1912 			if (ifr->ifr_addr.sa_len == 0)
1913 				ifr->ifr_addr.sa_len = 16;
1914 #endif
1915 			break;
1916 
1917 		case OSIOCGIFADDR:
1918 			cmd = SIOCGIFADDR;
1919 			break;
1920 
1921 		case OSIOCGIFDSTADDR:
1922 			cmd = SIOCGIFDSTADDR;
1923 			break;
1924 
1925 		case OSIOCGIFBRDADDR:
1926 			cmd = SIOCGIFBRDADDR;
1927 			break;
1928 
1929 		case OSIOCGIFNETMASK:
1930 			cmd = SIOCGIFNETMASK;
1931 		}
1932 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
1933 								   cmd,
1934 								   data,
1935 								   ifp, td));
1936 		switch (ocmd) {
1937 
1938 		case OSIOCGIFADDR:
1939 		case OSIOCGIFDSTADDR:
1940 		case OSIOCGIFBRDADDR:
1941 		case OSIOCGIFNETMASK:
1942 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1943 
1944 		}
1945 	}
1946 #endif /* COMPAT_43 */
1947 
1948 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1949 #ifdef INET6
1950 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
1951 		if (ifp->if_flags & IFF_UP) {
1952 			int s = splimp();
1953 			in6_if_up(ifp);
1954 			splx(s);
1955 		}
1956 #endif
1957 	}
1958 	return (error);
1959 }
1960 
1961 /*
1962  * The code common to handling reference counted flags,
1963  * e.g., in ifpromisc() and if_allmulti().
1964  * The "pflag" argument can specify a permanent mode flag to check,
1965  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
1966  *
1967  * Only to be used on stack-owned flags, not driver-owned flags.
1968  */
1969 static int
1970 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
1971 {
1972 	struct ifreq ifr;
1973 	int error;
1974 	int oldflags, oldcount;
1975 
1976 	/* Sanity checks to catch programming errors */
1977 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
1978 	    ("%s: setting driver-owned flag %d", __func__, flag));
1979 
1980 	if (onswitch)
1981 		KASSERT(*refcount >= 0,
1982 		    ("%s: increment negative refcount %d for flag %d",
1983 		    __func__, *refcount, flag));
1984 	else
1985 		KASSERT(*refcount > 0,
1986 		    ("%s: decrement non-positive refcount %d for flag %d",
1987 		    __func__, *refcount, flag));
1988 
1989 	/* In case this mode is permanent, just touch refcount */
1990 	if (ifp->if_flags & pflag) {
1991 		*refcount += onswitch ? 1 : -1;
1992 		return (0);
1993 	}
1994 
1995 	/* Save ifnet parameters for if_ioctl() may fail */
1996 	oldcount = *refcount;
1997 	oldflags = ifp->if_flags;
1998 
1999 	/*
2000 	 * See if we aren't the only and touching refcount is enough.
2001 	 * Actually toggle interface flag if we are the first or last.
2002 	 */
2003 	if (onswitch) {
2004 		if ((*refcount)++)
2005 			return (0);
2006 		ifp->if_flags |= flag;
2007 	} else {
2008 		if (--(*refcount))
2009 			return (0);
2010 		ifp->if_flags &= ~flag;
2011 	}
2012 
2013 	/* Call down the driver since we've changed interface flags */
2014 	if (ifp->if_ioctl == NULL) {
2015 		error = EOPNOTSUPP;
2016 		goto recover;
2017 	}
2018 	ifr.ifr_flags = ifp->if_flags & 0xffff;
2019 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
2020 	IFF_LOCKGIANT(ifp);
2021 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2022 	IFF_UNLOCKGIANT(ifp);
2023 	if (error)
2024 		goto recover;
2025 	/* Notify userland that interface flags have changed */
2026 	rt_ifmsg(ifp);
2027 	return (0);
2028 
2029 recover:
2030 	/* Recover after driver error */
2031 	*refcount = oldcount;
2032 	ifp->if_flags = oldflags;
2033 	return (error);
2034 }
2035 
2036 /*
2037  * Set/clear promiscuous mode on interface ifp based on the truth value
2038  * of pswitch.  The calls are reference counted so that only the first
2039  * "on" request actually has an effect, as does the final "off" request.
2040  * Results are undefined if the "off" and "on" requests are not matched.
2041  */
2042 int
2043 ifpromisc(struct ifnet *ifp, int pswitch)
2044 {
2045 	int error;
2046 	int oldflags = ifp->if_flags;
2047 
2048 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2049 			   &ifp->if_pcount, pswitch);
2050 	/* If promiscuous mode status has changed, log a message */
2051 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
2052 		log(LOG_INFO, "%s: promiscuous mode %s\n",
2053 		    ifp->if_xname,
2054 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2055 	return (error);
2056 }
2057 
2058 /*
2059  * Return interface configuration
2060  * of system.  List may be used
2061  * in later ioctl's (above) to get
2062  * other information.
2063  */
2064 /*ARGSUSED*/
2065 static int
2066 ifconf(u_long cmd, caddr_t data)
2067 {
2068 	struct ifconf *ifc = (struct ifconf *)data;
2069 #ifdef __amd64__
2070 	struct ifconf32 *ifc32 = (struct ifconf32 *)data;
2071 	struct ifconf ifc_swab;
2072 #endif
2073 	struct ifnet *ifp;
2074 	struct ifaddr *ifa;
2075 	struct ifreq ifr;
2076 	struct sbuf *sb;
2077 	int error, full = 0, valid_len, max_len;
2078 
2079 #ifdef __amd64__
2080 	if (cmd == SIOCGIFCONF32) {
2081 		ifc_swab.ifc_len = ifc32->ifc_len;
2082 		ifc_swab.ifc_buf = (caddr_t)(uintptr_t)ifc32->ifc_buf;
2083 		ifc = &ifc_swab;
2084 	}
2085 #endif
2086 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
2087 	max_len = MAXPHYS - 1;
2088 
2089 	/* Prevent hostile input from being able to crash the system */
2090 	if (ifc->ifc_len <= 0)
2091 		return (EINVAL);
2092 
2093 again:
2094 	if (ifc->ifc_len <= max_len) {
2095 		max_len = ifc->ifc_len;
2096 		full = 1;
2097 	}
2098 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
2099 	max_len = 0;
2100 	valid_len = 0;
2101 
2102 	IFNET_RLOCK();		/* could sleep XXX */
2103 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
2104 		int addrs;
2105 
2106 		/*
2107 		 * Zero the ifr_name buffer to make sure we don't
2108 		 * disclose the contents of the stack.
2109 		 */
2110 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
2111 
2112 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2113 		    >= sizeof(ifr.ifr_name)) {
2114 			sbuf_delete(sb);
2115 			IFNET_RUNLOCK();
2116 			return (ENAMETOOLONG);
2117 		}
2118 
2119 		addrs = 0;
2120 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2121 			struct sockaddr *sa = ifa->ifa_addr;
2122 
2123 			if (jailed(curthread->td_ucred) &&
2124 			    prison_if(curthread->td_ucred, sa))
2125 				continue;
2126 			addrs++;
2127 #ifdef COMPAT_43
2128 			if (cmd == OSIOCGIFCONF) {
2129 				struct osockaddr *osa =
2130 					 (struct osockaddr *)&ifr.ifr_addr;
2131 				ifr.ifr_addr = *sa;
2132 				osa->sa_family = sa->sa_family;
2133 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2134 				max_len += sizeof(ifr);
2135 			} else
2136 #endif
2137 			if (sa->sa_len <= sizeof(*sa)) {
2138 				ifr.ifr_addr = *sa;
2139 				sbuf_bcat(sb, &ifr, sizeof(ifr));
2140 				max_len += sizeof(ifr);
2141 			} else {
2142 				sbuf_bcat(sb, &ifr,
2143 				    offsetof(struct ifreq, ifr_addr));
2144 				max_len += offsetof(struct ifreq, ifr_addr);
2145 				sbuf_bcat(sb, sa, sa->sa_len);
2146 				max_len += sa->sa_len;
2147 			}
2148 
2149 			if (!sbuf_overflowed(sb))
2150 				valid_len = sbuf_len(sb);
2151 		}
2152 		if (addrs == 0) {
2153 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
2154 			sbuf_bcat(sb, &ifr, sizeof(ifr));
2155 			max_len += sizeof(ifr);
2156 
2157 			if (!sbuf_overflowed(sb))
2158 				valid_len = sbuf_len(sb);
2159 		}
2160 	}
2161 	IFNET_RUNLOCK();
2162 
2163 	/*
2164 	 * If we didn't allocate enough space (uncommon), try again.  If
2165 	 * we have already allocated as much space as we are allowed,
2166 	 * return what we've got.
2167 	 */
2168 	if (valid_len != max_len && !full) {
2169 		sbuf_delete(sb);
2170 		goto again;
2171 	}
2172 
2173 	ifc->ifc_len = valid_len;
2174 #ifdef __amd64__
2175 	if (cmd == SIOCGIFCONF32)
2176 		ifc32->ifc_len = valid_len;
2177 #endif
2178 	sbuf_finish(sb);
2179 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
2180 	sbuf_delete(sb);
2181 	return (error);
2182 }
2183 
2184 /*
2185  * Just like ifpromisc(), but for all-multicast-reception mode.
2186  */
2187 int
2188 if_allmulti(struct ifnet *ifp, int onswitch)
2189 {
2190 
2191 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
2192 }
2193 
2194 struct ifmultiaddr *
2195 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
2196 {
2197 	struct ifmultiaddr *ifma;
2198 
2199 	IF_ADDR_LOCK_ASSERT(ifp);
2200 
2201 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2202 		if (sa->sa_family == AF_LINK) {
2203 			if (sa_dl_equal(ifma->ifma_addr, sa))
2204 				break;
2205 		} else {
2206 			if (sa_equal(ifma->ifma_addr, sa))
2207 				break;
2208 		}
2209 	}
2210 
2211 	return ifma;
2212 }
2213 
2214 /*
2215  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
2216  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
2217  * the ifnet multicast address list here, so the caller must do that and
2218  * other setup work (such as notifying the device driver).  The reference
2219  * count is initialized to 1.
2220  */
2221 static struct ifmultiaddr *
2222 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
2223     int mflags)
2224 {
2225 	struct ifmultiaddr *ifma;
2226 	struct sockaddr *dupsa;
2227 
2228 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, mflags |
2229 	    M_ZERO);
2230 	if (ifma == NULL)
2231 		return (NULL);
2232 
2233 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, mflags);
2234 	if (dupsa == NULL) {
2235 		FREE(ifma, M_IFMADDR);
2236 		return (NULL);
2237 	}
2238 	bcopy(sa, dupsa, sa->sa_len);
2239 	ifma->ifma_addr = dupsa;
2240 
2241 	ifma->ifma_ifp = ifp;
2242 	ifma->ifma_refcount = 1;
2243 	ifma->ifma_protospec = NULL;
2244 
2245 	if (llsa == NULL) {
2246 		ifma->ifma_lladdr = NULL;
2247 		return (ifma);
2248 	}
2249 
2250 	MALLOC(dupsa, struct sockaddr *, llsa->sa_len, M_IFMADDR, mflags);
2251 	if (dupsa == NULL) {
2252 		FREE(ifma->ifma_addr, M_IFMADDR);
2253 		FREE(ifma, M_IFMADDR);
2254 		return (NULL);
2255 	}
2256 	bcopy(llsa, dupsa, llsa->sa_len);
2257 	ifma->ifma_lladdr = dupsa;
2258 
2259 	return (ifma);
2260 }
2261 
2262 /*
2263  * if_freemulti: free ifmultiaddr structure and possibly attached related
2264  * addresses.  The caller is responsible for implementing reference
2265  * counting, notifying the driver, handling routing messages, and releasing
2266  * any dependent link layer state.
2267  */
2268 static void
2269 if_freemulti(struct ifmultiaddr *ifma)
2270 {
2271 
2272 	KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
2273 	    ifma->ifma_refcount));
2274 	KASSERT(ifma->ifma_protospec == NULL,
2275 	    ("if_freemulti: protospec not NULL"));
2276 
2277 	if (ifma->ifma_lladdr != NULL)
2278 		FREE(ifma->ifma_lladdr, M_IFMADDR);
2279 	FREE(ifma->ifma_addr, M_IFMADDR);
2280 	FREE(ifma, M_IFMADDR);
2281 }
2282 
2283 /*
2284  * Register an additional multicast address with a network interface.
2285  *
2286  * - If the address is already present, bump the reference count on the
2287  *   address and return.
2288  * - If the address is not link-layer, look up a link layer address.
2289  * - Allocate address structures for one or both addresses, and attach to the
2290  *   multicast address list on the interface.  If automatically adding a link
2291  *   layer address, the protocol address will own a reference to the link
2292  *   layer address, to be freed when it is freed.
2293  * - Notify the network device driver of an addition to the multicast address
2294  *   list.
2295  *
2296  * 'sa' points to caller-owned memory with the desired multicast address.
2297  *
2298  * 'retifma' will be used to return a pointer to the resulting multicast
2299  * address reference, if desired.
2300  */
2301 int
2302 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2303     struct ifmultiaddr **retifma)
2304 {
2305 	struct ifmultiaddr *ifma, *ll_ifma;
2306 	struct sockaddr *llsa;
2307 	int error;
2308 
2309 	/*
2310 	 * If the address is already present, return a new reference to it;
2311 	 * otherwise, allocate storage and set up a new address.
2312 	 */
2313 	IF_ADDR_LOCK(ifp);
2314 	ifma = if_findmulti(ifp, sa);
2315 	if (ifma != NULL) {
2316 		ifma->ifma_refcount++;
2317 		if (retifma != NULL)
2318 			*retifma = ifma;
2319 		IF_ADDR_UNLOCK(ifp);
2320 		return (0);
2321 	}
2322 
2323 	/*
2324 	 * The address isn't already present; resolve the protocol address
2325 	 * into a link layer address, and then look that up, bump its
2326 	 * refcount or allocate an ifma for that also.  If 'llsa' was
2327 	 * returned, we will need to free it later.
2328 	 */
2329 	llsa = NULL;
2330 	ll_ifma = NULL;
2331 	if (ifp->if_resolvemulti != NULL) {
2332 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
2333 		if (error)
2334 			goto unlock_out;
2335 	}
2336 
2337 	/*
2338 	 * Allocate the new address.  Don't hook it up yet, as we may also
2339 	 * need to allocate a link layer multicast address.
2340 	 */
2341 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
2342 	if (ifma == NULL) {
2343 		error = ENOMEM;
2344 		goto free_llsa_out;
2345 	}
2346 
2347 	/*
2348 	 * If a link layer address is found, we'll need to see if it's
2349 	 * already present in the address list, or allocate is as well.
2350 	 * When this block finishes, the link layer address will be on the
2351 	 * list.
2352 	 */
2353 	if (llsa != NULL) {
2354 		ll_ifma = if_findmulti(ifp, llsa);
2355 		if (ll_ifma == NULL) {
2356 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
2357 			if (ll_ifma == NULL) {
2358 				--ifma->ifma_refcount;
2359 				if_freemulti(ifma);
2360 				error = ENOMEM;
2361 				goto free_llsa_out;
2362 			}
2363 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
2364 			    ifma_link);
2365 		} else
2366 			ll_ifma->ifma_refcount++;
2367 		ifma->ifma_llifma = ll_ifma;
2368 	}
2369 
2370 	/*
2371 	 * We now have a new multicast address, ifma, and possibly a new or
2372 	 * referenced link layer address.  Add the primary address to the
2373 	 * ifnet address list.
2374 	 */
2375 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2376 
2377 	if (retifma != NULL)
2378 		*retifma = ifma;
2379 
2380 	/*
2381 	 * Must generate the message while holding the lock so that 'ifma'
2382 	 * pointer is still valid.
2383 	 */
2384 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2385 	IF_ADDR_UNLOCK(ifp);
2386 
2387 	/*
2388 	 * We are certain we have added something, so call down to the
2389 	 * interface to let them know about it.
2390 	 */
2391 	if (ifp->if_ioctl != NULL) {
2392 		IFF_LOCKGIANT(ifp);
2393 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2394 		IFF_UNLOCKGIANT(ifp);
2395 	}
2396 
2397 	if (llsa != NULL)
2398 		FREE(llsa, M_IFMADDR);
2399 
2400 	return (0);
2401 
2402 free_llsa_out:
2403 	if (llsa != NULL)
2404 		FREE(llsa, M_IFMADDR);
2405 
2406 unlock_out:
2407 	IF_ADDR_UNLOCK(ifp);
2408 	return (error);
2409 }
2410 
2411 /*
2412  * Delete a multicast group membership by network-layer group address.
2413  *
2414  * Returns ENOENT if the entry could not be found. If ifp no longer
2415  * exists, results are undefined. This entry point should only be used
2416  * from subsystems which do appropriate locking to hold ifp for the
2417  * duration of the call.
2418  * Network-layer protocol domains must use if_delmulti_ifma().
2419  */
2420 int
2421 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2422 {
2423 	struct ifmultiaddr *ifma;
2424 	int lastref;
2425 #ifdef INVARIANTS
2426 	struct ifnet *oifp;
2427 
2428 	IFNET_RLOCK();
2429 	TAILQ_FOREACH(oifp, &ifnet, if_link)
2430 		if (ifp == oifp)
2431 			break;
2432 	if (ifp != oifp)
2433 		ifp = NULL;
2434 	IFNET_RUNLOCK();
2435 
2436 	KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
2437 #endif
2438 	if (ifp == NULL)
2439 		return (ENOENT);
2440 
2441 	IF_ADDR_LOCK(ifp);
2442 	lastref = 0;
2443 	ifma = if_findmulti(ifp, sa);
2444 	if (ifma != NULL)
2445 		lastref = if_delmulti_locked(ifp, ifma, 0);
2446 	IF_ADDR_UNLOCK(ifp);
2447 
2448 	if (ifma == NULL)
2449 		return (ENOENT);
2450 
2451 	if (lastref && ifp->if_ioctl != NULL) {
2452 		IFF_LOCKGIANT(ifp);
2453 		(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2454 		IFF_UNLOCKGIANT(ifp);
2455 	}
2456 
2457 	return (0);
2458 }
2459 
2460 /*
2461  * Delete a multicast group membership by group membership pointer.
2462  * Network-layer protocol domains must use this routine.
2463  *
2464  * It is safe to call this routine if the ifp disappeared. Callers should
2465  * hold IFF_LOCKGIANT() to avoid a LOR in case the hardware needs to be
2466  * reconfigured.
2467  */
2468 void
2469 if_delmulti_ifma(struct ifmultiaddr *ifma)
2470 {
2471 	struct ifnet *ifp;
2472 	int lastref;
2473 
2474 	ifp = ifma->ifma_ifp;
2475 #ifdef DIAGNOSTIC
2476 	if (ifp == NULL) {
2477 		printf("%s: ifma_ifp seems to be detached\n", __func__);
2478 	} else {
2479 		struct ifnet *oifp;
2480 
2481 		IFNET_RLOCK();
2482 		TAILQ_FOREACH(oifp, &ifnet, if_link)
2483 			if (ifp == oifp)
2484 				break;
2485 		if (ifp != oifp) {
2486 			printf("%s: ifnet %p disappeared\n", __func__, ifp);
2487 			ifp = NULL;
2488 		}
2489 		IFNET_RUNLOCK();
2490 	}
2491 #endif
2492 	/*
2493 	 * If and only if the ifnet instance exists: Acquire the address lock.
2494 	 */
2495 	if (ifp != NULL)
2496 		IF_ADDR_LOCK(ifp);
2497 
2498 	lastref = if_delmulti_locked(ifp, ifma, 0);
2499 
2500 	if (ifp != NULL) {
2501 		/*
2502 		 * If and only if the ifnet instance exists:
2503 		 *  Release the address lock.
2504 		 *  If the group was left: update the hardware hash filter.
2505 		 */
2506 		IF_ADDR_UNLOCK(ifp);
2507 		if (lastref && ifp->if_ioctl != NULL) {
2508 			IFF_LOCKGIANT(ifp);
2509 			(void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2510 			IFF_UNLOCKGIANT(ifp);
2511 		}
2512 	}
2513 }
2514 
2515 /*
2516  * Perform deletion of network-layer and/or link-layer multicast address.
2517  *
2518  * Return 0 if the reference count was decremented.
2519  * Return 1 if the final reference was released, indicating that the
2520  * hardware hash filter should be reprogrammed.
2521  */
2522 static int
2523 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
2524 {
2525 	struct ifmultiaddr *ll_ifma;
2526 
2527 	if (ifp != NULL && ifma->ifma_ifp != NULL) {
2528 		KASSERT(ifma->ifma_ifp == ifp,
2529 		    ("%s: inconsistent ifp %p", __func__, ifp));
2530 		IF_ADDR_LOCK_ASSERT(ifp);
2531 	}
2532 
2533 	ifp = ifma->ifma_ifp;
2534 
2535 	/*
2536 	 * If the ifnet is detaching, null out references to ifnet,
2537 	 * so that upper protocol layers will notice, and not attempt
2538 	 * to obtain locks for an ifnet which no longer exists. The
2539 	 * routing socket announcement must happen before the ifnet
2540 	 * instance is detached from the system.
2541 	 */
2542 	if (detaching) {
2543 #ifdef DIAGNOSTIC
2544 		printf("%s: detaching ifnet instance %p\n", __func__, ifp);
2545 #endif
2546 		/*
2547 		 * ifp may already be nulled out if we are being reentered
2548 		 * to delete the ll_ifma.
2549 		 */
2550 		if (ifp != NULL) {
2551 			rt_newmaddrmsg(RTM_DELMADDR, ifma);
2552 			ifma->ifma_ifp = NULL;
2553 		}
2554 	}
2555 
2556 	if (--ifma->ifma_refcount > 0)
2557 		return 0;
2558 
2559 	/*
2560 	 * If this ifma is a network-layer ifma, a link-layer ifma may
2561 	 * have been associated with it. Release it first if so.
2562 	 */
2563 	ll_ifma = ifma->ifma_llifma;
2564 	if (ll_ifma != NULL) {
2565 		KASSERT(ifma->ifma_lladdr != NULL,
2566 		    ("%s: llifma w/o lladdr", __func__));
2567 		if (detaching)
2568 			ll_ifma->ifma_ifp = NULL;	/* XXX */
2569 		if (--ll_ifma->ifma_refcount == 0) {
2570 			if (ifp != NULL) {
2571 				TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
2572 				    ifma_link);
2573 			}
2574 			if_freemulti(ll_ifma);
2575 		}
2576 	}
2577 
2578 	if (ifp != NULL)
2579 		TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2580 
2581 	if_freemulti(ifma);
2582 
2583 	/*
2584 	 * The last reference to this instance of struct ifmultiaddr
2585 	 * was released; the hardware should be notified of this change.
2586 	 */
2587 	return 1;
2588 }
2589 
2590 /*
2591  * Set the link layer address on an interface.
2592  *
2593  * At this time we only support certain types of interfaces,
2594  * and we don't allow the length of the address to change.
2595  */
2596 int
2597 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2598 {
2599 	struct sockaddr_dl *sdl;
2600 	struct ifaddr *ifa;
2601 	struct ifreq ifr;
2602 
2603 	ifa = ifp->if_addr;
2604 	if (ifa == NULL)
2605 		return (EINVAL);
2606 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2607 	if (sdl == NULL)
2608 		return (EINVAL);
2609 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2610 		return (EINVAL);
2611 	switch (ifp->if_type) {
2612 	case IFT_ETHER:
2613 	case IFT_FDDI:
2614 	case IFT_XETHER:
2615 	case IFT_ISO88025:
2616 	case IFT_L2VLAN:
2617 	case IFT_BRIDGE:
2618 	case IFT_ARCNET:
2619 	case IFT_IEEE8023ADLAG:
2620 		bcopy(lladdr, LLADDR(sdl), len);
2621 		break;
2622 	default:
2623 		return (ENODEV);
2624 	}
2625 	/*
2626 	 * If the interface is already up, we need
2627 	 * to re-init it in order to reprogram its
2628 	 * address filter.
2629 	 */
2630 	if ((ifp->if_flags & IFF_UP) != 0) {
2631 		if (ifp->if_ioctl) {
2632 			IFF_LOCKGIANT(ifp);
2633 			ifp->if_flags &= ~IFF_UP;
2634 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2635 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2636 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2637 			ifp->if_flags |= IFF_UP;
2638 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2639 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2640 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2641 			IFF_UNLOCKGIANT(ifp);
2642 		}
2643 #ifdef INET
2644 		/*
2645 		 * Also send gratuitous ARPs to notify other nodes about
2646 		 * the address change.
2647 		 */
2648 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2649 			if (ifa->ifa_addr->sa_family == AF_INET)
2650 				arp_ifinit(ifp, ifa);
2651 		}
2652 #endif
2653 	}
2654 	return (0);
2655 }
2656 
2657 /*
2658  * The name argument must be a pointer to storage which will last as
2659  * long as the interface does.  For physical devices, the result of
2660  * device_get_name(dev) is a good choice and for pseudo-devices a
2661  * static string works well.
2662  */
2663 void
2664 if_initname(struct ifnet *ifp, const char *name, int unit)
2665 {
2666 	ifp->if_dname = name;
2667 	ifp->if_dunit = unit;
2668 	if (unit != IF_DUNIT_NONE)
2669 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2670 	else
2671 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2672 }
2673 
2674 int
2675 if_printf(struct ifnet *ifp, const char * fmt, ...)
2676 {
2677 	va_list ap;
2678 	int retval;
2679 
2680 	retval = printf("%s: ", ifp->if_xname);
2681 	va_start(ap, fmt);
2682 	retval += vprintf(fmt, ap);
2683 	va_end(ap);
2684 	return (retval);
2685 }
2686 
2687 /*
2688  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2689  * be called without Giant.  However, we often can't acquire the Giant lock
2690  * at those points; instead, we run it via a task queue that holds Giant via
2691  * if_start_deferred.
2692  *
2693  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2694  * outstanding if_start_deferred() tasks that will run after the free.  This
2695  * probably means waiting in if_detach().
2696  */
2697 void
2698 if_start(struct ifnet *ifp)
2699 {
2700 
2701 	if (ifp->if_flags & IFF_NEEDSGIANT) {
2702 		if (mtx_owned(&Giant))
2703 			(*(ifp)->if_start)(ifp);
2704 		else
2705 			taskqueue_enqueue(taskqueue_swi_giant,
2706 			    &ifp->if_starttask);
2707 	} else
2708 		(*(ifp)->if_start)(ifp);
2709 }
2710 
2711 static void
2712 if_start_deferred(void *context, int pending)
2713 {
2714 	struct ifnet *ifp;
2715 
2716 	GIANT_REQUIRED;
2717 
2718 	ifp = context;
2719 	(ifp->if_start)(ifp);
2720 }
2721 
2722 int
2723 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2724 {
2725 	int active = 0;
2726 
2727 	IF_LOCK(ifq);
2728 	if (_IF_QFULL(ifq)) {
2729 		_IF_DROP(ifq);
2730 		IF_UNLOCK(ifq);
2731 		m_freem(m);
2732 		return (0);
2733 	}
2734 	if (ifp != NULL) {
2735 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2736 		if (m->m_flags & (M_BCAST|M_MCAST))
2737 			ifp->if_omcasts++;
2738 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2739 	}
2740 	_IF_ENQUEUE(ifq, m);
2741 	IF_UNLOCK(ifq);
2742 	if (ifp != NULL && !active)
2743 		if_start(ifp);
2744 	return (1);
2745 }
2746 
2747 void
2748 if_register_com_alloc(u_char type,
2749     if_com_alloc_t *a, if_com_free_t *f)
2750 {
2751 
2752 	KASSERT(if_com_alloc[type] == NULL,
2753 	    ("if_register_com_alloc: %d already registered", type));
2754 	KASSERT(if_com_free[type] == NULL,
2755 	    ("if_register_com_alloc: %d free already registered", type));
2756 
2757 	if_com_alloc[type] = a;
2758 	if_com_free[type] = f;
2759 }
2760 
2761 void
2762 if_deregister_com_alloc(u_char type)
2763 {
2764 
2765 	KASSERT(if_com_alloc[type] != NULL,
2766 	    ("if_deregister_com_alloc: %d not registered", type));
2767 	KASSERT(if_com_free[type] != NULL,
2768 	    ("if_deregister_com_alloc: %d free not registered", type));
2769 	if_com_alloc[type] = NULL;
2770 	if_com_free[type] = NULL;
2771 }
2772