xref: /freebsd/sys/net/if.c (revision b6de9e91bd2c47efaeec72a08642f8fd99cc7b20)
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/mac.h>
43 #include <sys/malloc.h>
44 #include <sys/sbuf.h>
45 #include <sys/bus.h>
46 #include <sys/mbuf.h>
47 #include <sys/systm.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_arp.h>
63 #include <net/if_clone.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/if_var.h>
67 #include <net/radix.h>
68 #include <net/route.h>
69 
70 #if defined(INET) || defined(INET6)
71 /*XXX*/
72 #include <netinet/in.h>
73 #include <netinet/in_var.h>
74 #ifdef INET6
75 #include <netinet6/in6_var.h>
76 #include <netinet6/in6_ifattach.h>
77 #endif
78 #endif
79 #ifdef INET
80 #include <netinet/if_ether.h>
81 #endif
82 #ifdef DEV_CARP
83 #include <netinet/ip_carp.h>
84 #endif
85 
86 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
87 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
88 
89 /* Log link state change events */
90 static int log_link_state_change = 1;
91 
92 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
93 	&log_link_state_change, 0,
94 	"log interface link state change events");
95 
96 void	(*bstp_linkstate_p)(struct ifnet *ifp, int state);
97 void	(*ng_ether_link_state_p)(struct ifnet *ifp, int state);
98 
99 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
100 
101 static void	if_attachdomain(void *);
102 static void	if_attachdomain1(struct ifnet *);
103 static int	ifconf(u_long, caddr_t);
104 static void	if_grow(void);
105 static void	if_init(void *);
106 static void	if_check(void *);
107 static void	if_qflush(struct ifaltq *);
108 static void	if_route(struct ifnet *, int flag, int fam);
109 static int	if_setflag(struct ifnet *, int, int, int *, int);
110 static void	if_slowtimo(void *);
111 static void	if_unroute(struct ifnet *, int flag, int fam);
112 static void	link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
113 static int	if_rtdel(struct radix_node *, void *);
114 static int	ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
115 static void	if_start_deferred(void *context, int pending);
116 static void	do_link_state_change(void *, int);
117 #ifdef INET6
118 /*
119  * XXX: declare here to avoid to include many inet6 related files..
120  * should be more generalized?
121  */
122 extern void	nd6_setmtu(struct ifnet *);
123 #endif
124 
125 int	if_index = 0;
126 struct	ifindex_entry *ifindex_table = NULL;
127 int	ifqmaxlen = IFQ_MAXLEN;
128 struct	ifnethead ifnet;	/* depend on static init XXX */
129 struct	mtx ifnet_lock;
130 static	if_com_alloc_t *if_com_alloc[256];
131 static	if_com_free_t *if_com_free[256];
132 
133 static int	if_indexlim = 8;
134 static struct	knlist ifklist;
135 
136 static void	filt_netdetach(struct knote *kn);
137 static int	filt_netdev(struct knote *kn, long hint);
138 
139 static struct filterops netdev_filtops =
140     { 1, NULL, filt_netdetach, filt_netdev };
141 
142 /*
143  * System initialization
144  */
145 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL)
146 SYSINIT(interface_check, SI_SUB_PROTO_IF, SI_ORDER_FIRST, if_check, NULL)
147 
148 MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
149 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
150 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
151 
152 static d_open_t		netopen;
153 static d_close_t	netclose;
154 static d_ioctl_t	netioctl;
155 static d_kqfilter_t	netkqfilter;
156 
157 static struct cdevsw net_cdevsw = {
158 	.d_version =	D_VERSION,
159 	.d_flags =	D_NEEDGIANT,
160 	.d_open =	netopen,
161 	.d_close =	netclose,
162 	.d_ioctl =	netioctl,
163 	.d_name =	"net",
164 	.d_kqfilter =	netkqfilter,
165 };
166 
167 static int
168 netopen(struct cdev *dev, int flag, int mode, struct thread *td)
169 {
170 	return (0);
171 }
172 
173 static int
174 netclose(struct cdev *dev, int flags, int fmt, struct thread *td)
175 {
176 	return (0);
177 }
178 
179 static int
180 netioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
181 {
182 	struct ifnet *ifp;
183 	int error, idx;
184 
185 	/* only support interface specific ioctls */
186 	if (IOCGROUP(cmd) != 'i')
187 		return (EOPNOTSUPP);
188 	idx = minor(dev);
189 	if (idx == 0) {
190 		/*
191 		 * special network device, not interface.
192 		 */
193 		if (cmd == SIOCGIFCONF)
194 			return (ifconf(cmd, data));	/* XXX remove cmd */
195 		return (EOPNOTSUPP);
196 	}
197 
198 	ifp = ifnet_byindex(idx);
199 	if (ifp == NULL)
200 		return (ENXIO);
201 
202 	error = ifhwioctl(cmd, ifp, data, td);
203 	if (error == ENOIOCTL)
204 		error = EOPNOTSUPP;
205 	return (error);
206 }
207 
208 static int
209 netkqfilter(struct cdev *dev, struct knote *kn)
210 {
211 	struct knlist *klist;
212 	struct ifnet *ifp;
213 	int idx;
214 
215 	switch (kn->kn_filter) {
216 	case EVFILT_NETDEV:
217 		kn->kn_fop = &netdev_filtops;
218 		break;
219 	default:
220 		return (EINVAL);
221 	}
222 
223 	idx = minor(dev);
224 	if (idx == 0) {
225 		klist = &ifklist;
226 	} else {
227 		ifp = ifnet_byindex(idx);
228 		if (ifp == NULL)
229 			return (1);
230 		klist = &ifp->if_klist;
231 	}
232 
233 	kn->kn_hook = (caddr_t)klist;
234 
235 	knlist_add(klist, kn, 0);
236 
237 	return (0);
238 }
239 
240 static void
241 filt_netdetach(struct knote *kn)
242 {
243 	struct knlist *klist = (struct knlist *)kn->kn_hook;
244 
245 	knlist_remove(klist, kn, 0);
246 }
247 
248 static int
249 filt_netdev(struct knote *kn, long hint)
250 {
251 	struct knlist *klist = (struct knlist *)kn->kn_hook;
252 
253 	/*
254 	 * Currently NOTE_EXIT is abused to indicate device detach.
255 	 */
256 	if (hint == NOTE_EXIT) {
257 		kn->kn_data = NOTE_LINKINV;
258 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
259 		knlist_remove_inevent(klist, kn);
260 		return (1);
261 	}
262 	if (hint != 0)
263 		kn->kn_data = hint;			/* current status */
264 	if (kn->kn_sfflags & hint)
265 		kn->kn_fflags |= hint;
266 	return (kn->kn_fflags != 0);
267 }
268 
269 /*
270  * Network interface utility routines.
271  *
272  * Routines with ifa_ifwith* names take sockaddr *'s as
273  * parameters.
274  */
275 /* ARGSUSED*/
276 static void
277 if_init(void *dummy __unused)
278 {
279 
280 	IFNET_LOCK_INIT();
281 	TAILQ_INIT(&ifnet);
282 	knlist_init(&ifklist, NULL, NULL, NULL, NULL);
283 	if_grow();				/* create initial table */
284 	ifdev_byindex(0) = make_dev(&net_cdevsw, 0,
285 	    UID_ROOT, GID_WHEEL, 0600, "network");
286 	if_clone_init();
287 }
288 
289 static void
290 if_grow(void)
291 {
292 	u_int n;
293 	struct ifindex_entry *e;
294 
295 	if_indexlim <<= 1;
296 	n = if_indexlim * sizeof(*e);
297 	e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
298 	if (ifindex_table != NULL) {
299 		memcpy((caddr_t)e, (caddr_t)ifindex_table, n/2);
300 		free((caddr_t)ifindex_table, M_IFNET);
301 	}
302 	ifindex_table = e;
303 }
304 
305 /* ARGSUSED*/
306 static void
307 if_check(void *dummy __unused)
308 {
309 	struct ifnet *ifp;
310 	int s;
311 
312 	s = splimp();
313 	IFNET_RLOCK();	/* could sleep on rare error; mostly okay XXX */
314 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
315 		if (ifp->if_snd.ifq_maxlen == 0) {
316 			if_printf(ifp, "XXX: driver didn't set ifq_maxlen\n");
317 			ifp->if_snd.ifq_maxlen = ifqmaxlen;
318 		}
319 		if (!mtx_initialized(&ifp->if_snd.ifq_mtx)) {
320 			if_printf(ifp,
321 			    "XXX: driver didn't initialize queue mtx\n");
322 			mtx_init(&ifp->if_snd.ifq_mtx, "unknown",
323 			    MTX_NETWORK_LOCK, MTX_DEF);
324 		}
325 	}
326 	IFNET_RUNLOCK();
327 	splx(s);
328 	if_slowtimo(0);
329 }
330 
331 /*
332  * Allocate a struct ifnet and in index for an interface.
333  */
334 struct ifnet*
335 if_alloc(u_char type)
336 {
337 	struct ifnet *ifp;
338 
339 	ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
340 
341 	/*
342 	 * Try to find an empty slot below if_index.  If we fail, take
343 	 * the next slot.
344 	 *
345 	 * XXX: should be locked!
346 	 */
347 	for (ifp->if_index = 1; ifp->if_index <= if_index; ifp->if_index++) {
348 		if (ifnet_byindex(ifp->if_index) == NULL)
349 			break;
350 	}
351 	/* Catch if_index overflow. */
352 	if (ifp->if_index < 1) {
353 		free(ifp, M_IFNET);
354 		return (NULL);
355 	}
356 	if (ifp->if_index > if_index)
357 		if_index = ifp->if_index;
358 	if (if_index >= if_indexlim)
359 		if_grow();
360 	ifnet_byindex(ifp->if_index) = ifp;
361 
362 	ifp->if_type = type;
363 
364 	if (if_com_alloc[type] != NULL) {
365 		ifp->if_l2com = if_com_alloc[type](type, ifp);
366 		if (ifp->if_l2com == NULL) {
367 			free(ifp, M_IFNET);
368 			return (NULL);
369 		}
370 	}
371 	IF_ADDR_LOCK_INIT(ifp);
372 
373 	return (ifp);
374 }
375 
376 void
377 if_free(struct ifnet *ifp)
378 {
379 
380 	/* Do not add code to this function!  Add it to if_free_type(). */
381 	if_free_type(ifp, ifp->if_type);
382 }
383 
384 void
385 if_free_type(struct ifnet *ifp, u_char type)
386 {
387 
388 	if (ifp != ifnet_byindex(ifp->if_index)) {
389 		if_printf(ifp, "%s: value was not if_alloced, skipping\n",
390 		    __func__);
391 		return;
392 	}
393 
394 	IF_ADDR_LOCK_DESTROY(ifp);
395 
396 	ifnet_byindex(ifp->if_index) = NULL;
397 
398 	/* XXX: should be locked with if_findindex() */
399 	while (if_index > 0 && ifnet_byindex(if_index) == NULL)
400 		if_index--;
401 
402 	if (if_com_free[type] != NULL)
403 		if_com_free[type](ifp->if_l2com, type);
404 
405 	free(ifp, M_IFNET);
406 };
407 
408 /*
409  * Attach an interface to the
410  * list of "active" interfaces.
411  */
412 void
413 if_attach(struct ifnet *ifp)
414 {
415 	unsigned socksize, ifasize;
416 	int namelen, masklen;
417 	struct sockaddr_dl *sdl;
418 	struct ifaddr *ifa;
419 
420 	if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
421 		panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
422 		    ifp->if_xname);
423 
424 	TASK_INIT(&ifp->if_starttask, 0, if_start_deferred, ifp);
425 	TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
426 	IF_AFDATA_LOCK_INIT(ifp);
427 	ifp->if_afdata_initialized = 0;
428 	IFNET_WLOCK();
429 	TAILQ_INSERT_TAIL(&ifnet, ifp, if_link);
430 	IFNET_WUNLOCK();
431 	/*
432 	 * XXX -
433 	 * The old code would work if the interface passed a pre-existing
434 	 * chain of ifaddrs to this code.  We don't trust our callers to
435 	 * properly initialize the tailq, however, so we no longer allow
436 	 * this unlikely case.
437 	 */
438 	TAILQ_INIT(&ifp->if_addrhead);
439 	TAILQ_INIT(&ifp->if_prefixhead);
440 	TAILQ_INIT(&ifp->if_multiaddrs);
441 	knlist_init(&ifp->if_klist, NULL, NULL, NULL, NULL);
442 	getmicrotime(&ifp->if_lastchange);
443 	ifp->if_data.ifi_epoch = time_uptime;
444 	ifp->if_data.ifi_datalen = sizeof(struct if_data);
445 
446 #ifdef MAC
447 	mac_init_ifnet(ifp);
448 	mac_create_ifnet(ifp);
449 #endif
450 
451 	ifdev_byindex(ifp->if_index) = make_dev(&net_cdevsw,
452 	    unit2minor(ifp->if_index),
453 	    UID_ROOT, GID_WHEEL, 0600, "%s/%s",
454 	    net_cdevsw.d_name, ifp->if_xname);
455 	make_dev_alias(ifdev_byindex(ifp->if_index), "%s%d",
456 	    net_cdevsw.d_name, ifp->if_index);
457 
458 	mtx_init(&ifp->if_snd.ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
459 
460 	/*
461 	 * create a Link Level name for this device
462 	 */
463 	namelen = strlen(ifp->if_xname);
464 	/*
465 	 * Always save enough space for any possiable name so we can do
466 	 * a rename in place later.
467 	 */
468 	masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
469 	socksize = masklen + ifp->if_addrlen;
470 	if (socksize < sizeof(*sdl))
471 		socksize = sizeof(*sdl);
472 	socksize = roundup2(socksize, sizeof(long));
473 	ifasize = sizeof(*ifa) + 2 * socksize;
474 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
475 	IFA_LOCK_INIT(ifa);
476 	sdl = (struct sockaddr_dl *)(ifa + 1);
477 	sdl->sdl_len = socksize;
478 	sdl->sdl_family = AF_LINK;
479 	bcopy(ifp->if_xname, sdl->sdl_data, namelen);
480 	sdl->sdl_nlen = namelen;
481 	sdl->sdl_index = ifp->if_index;
482 	sdl->sdl_type = ifp->if_type;
483 	ifaddr_byindex(ifp->if_index) = ifa;
484 	ifa->ifa_ifp = ifp;
485 	ifa->ifa_rtrequest = link_rtrequest;
486 	ifa->ifa_addr = (struct sockaddr *)sdl;
487 	sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
488 	ifa->ifa_netmask = (struct sockaddr *)sdl;
489 	sdl->sdl_len = masklen;
490 	while (namelen != 0)
491 		sdl->sdl_data[--namelen] = 0xff;
492 	ifa->ifa_refcnt = 1;
493 	TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
494 	ifp->if_broadcastaddr = NULL; /* reliably crash if used uninitialized */
495 	ifp->if_snd.altq_type = 0;
496 	ifp->if_snd.altq_disc = NULL;
497 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
498 	ifp->if_snd.altq_tbr  = NULL;
499 	ifp->if_snd.altq_ifp  = ifp;
500 
501 	if (domain_init_status >= 2)
502 		if_attachdomain1(ifp);
503 
504 	EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
505 
506 	/* Announce the interface. */
507 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
508 }
509 
510 static void
511 if_attachdomain(void *dummy)
512 {
513 	struct ifnet *ifp;
514 	int s;
515 
516 	s = splnet();
517 	TAILQ_FOREACH(ifp, &ifnet, if_link)
518 		if_attachdomain1(ifp);
519 	splx(s);
520 }
521 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
522     if_attachdomain, NULL);
523 
524 static void
525 if_attachdomain1(struct ifnet *ifp)
526 {
527 	struct domain *dp;
528 	int s;
529 
530 	s = splnet();
531 
532 	/*
533 	 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
534 	 * cannot lock ifp->if_afdata initialization, entirely.
535 	 */
536 	if (IF_AFDATA_TRYLOCK(ifp) == 0) {
537 		splx(s);
538 		return;
539 	}
540 	if (ifp->if_afdata_initialized >= domain_init_status) {
541 		IF_AFDATA_UNLOCK(ifp);
542 		splx(s);
543 		printf("if_attachdomain called more than once on %s\n",
544 		    ifp->if_xname);
545 		return;
546 	}
547 	ifp->if_afdata_initialized = domain_init_status;
548 	IF_AFDATA_UNLOCK(ifp);
549 
550 	/* address family dependent data region */
551 	bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
552 	for (dp = domains; dp; dp = dp->dom_next) {
553 		if (dp->dom_ifattach)
554 			ifp->if_afdata[dp->dom_family] =
555 			    (*dp->dom_ifattach)(ifp);
556 	}
557 
558 	splx(s);
559 }
560 
561 /*
562  * Remove any network addresses from an interface.
563  */
564 
565 void
566 if_purgeaddrs(struct ifnet *ifp)
567 {
568 	struct ifaddr *ifa, *next;
569 
570 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
571 
572 		if (ifa->ifa_addr->sa_family == AF_LINK)
573 			continue;
574 #ifdef INET
575 		/* XXX: Ugly!! ad hoc just for INET */
576 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
577 			struct ifaliasreq ifr;
578 
579 			bzero(&ifr, sizeof(ifr));
580 			ifr.ifra_addr = *ifa->ifa_addr;
581 			if (ifa->ifa_dstaddr)
582 				ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
583 			if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
584 			    NULL) == 0)
585 				continue;
586 		}
587 #endif /* INET */
588 #ifdef INET6
589 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
590 			in6_purgeaddr(ifa);
591 			/* ifp_addrhead is already updated */
592 			continue;
593 		}
594 #endif /* INET6 */
595 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
596 		IFAFREE(ifa);
597 	}
598 }
599 
600 /*
601  * Detach an interface, removing it from the
602  * list of "active" interfaces and freeing the struct ifnet.
603  *
604  * XXXRW: There are some significant questions about event ordering, and
605  * how to prevent things from starting to use the interface during detach.
606  */
607 void
608 if_detach(struct ifnet *ifp)
609 {
610 	struct ifaddr *ifa;
611 	struct radix_node_head	*rnh;
612 	int s;
613 	int i;
614 	struct domain *dp;
615  	struct ifnet *iter;
616  	int found;
617 
618 	/*
619 	 * Remove/wait for pending events.
620 	 */
621 	taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
622 
623 #ifdef DEV_CARP
624 	/* Maybe hook to the generalized departure handler above?!? */
625 	if (ifp->if_carp)
626 		carp_ifdetach(ifp);
627 #endif
628 
629 	/*
630 	 * Remove routes and flush queues.
631 	 */
632 	s = splnet();
633 	if_down(ifp);
634 #ifdef ALTQ
635 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
636 		altq_disable(&ifp->if_snd);
637 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
638 		altq_detach(&ifp->if_snd);
639 #endif
640 
641 	if_purgeaddrs(ifp);
642 
643 #ifdef INET
644 	in_ifdetach(ifp);
645 #endif
646 
647 #ifdef INET6
648 	/*
649 	 * Remove all IPv6 kernel structs related to ifp.  This should be done
650 	 * before removing routing entries below, since IPv6 interface direct
651 	 * routes are expected to be removed by the IPv6-specific kernel API.
652 	 * Otherwise, the kernel will detect some inconsistency and bark it.
653 	 */
654 	in6_ifdetach(ifp);
655 #endif
656 	/*
657 	 * Remove address from ifindex_table[] and maybe decrement if_index.
658 	 * Clean up all addresses.
659 	 */
660 	ifaddr_byindex(ifp->if_index) = NULL;
661 	destroy_dev(ifdev_byindex(ifp->if_index));
662 	ifdev_byindex(ifp->if_index) = NULL;
663 
664 	/* We can now free link ifaddr. */
665 	if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
666 		ifa = TAILQ_FIRST(&ifp->if_addrhead);
667 		TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
668 		IFAFREE(ifa);
669 	}
670 
671 	/*
672 	 * Delete all remaining routes using this interface
673 	 * Unfortuneatly the only way to do this is to slog through
674 	 * the entire routing table looking for routes which point
675 	 * to this interface...oh well...
676 	 */
677 	for (i = 1; i <= AF_MAX; i++) {
678 		if ((rnh = rt_tables[i]) == NULL)
679 			continue;
680 		RADIX_NODE_HEAD_LOCK(rnh);
681 		(void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
682 		RADIX_NODE_HEAD_UNLOCK(rnh);
683 	}
684 
685 	/* Announce that the interface is gone. */
686 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
687 	EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
688 
689 	IF_AFDATA_LOCK(ifp);
690 	for (dp = domains; dp; dp = dp->dom_next) {
691 		if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
692 			(*dp->dom_ifdetach)(ifp,
693 			    ifp->if_afdata[dp->dom_family]);
694 	}
695 	IF_AFDATA_UNLOCK(ifp);
696 
697 #ifdef MAC
698 	mac_destroy_ifnet(ifp);
699 #endif /* MAC */
700 	KNOTE_UNLOCKED(&ifp->if_klist, NOTE_EXIT);
701 	knlist_clear(&ifp->if_klist, 0);
702 	knlist_destroy(&ifp->if_klist);
703 	IFNET_WLOCK();
704  	found = 0;
705  	TAILQ_FOREACH(iter, &ifnet, if_link)
706  		if (iter == ifp) {
707  			found = 1;
708  			break;
709  		}
710  	if (found)
711  		TAILQ_REMOVE(&ifnet, ifp, if_link);
712 	IFNET_WUNLOCK();
713 	mtx_destroy(&ifp->if_snd.ifq_mtx);
714 	IF_AFDATA_DESTROY(ifp);
715 	splx(s);
716 }
717 
718 /*
719  * Delete Routes for a Network Interface
720  *
721  * Called for each routing entry via the rnh->rnh_walktree() call above
722  * to delete all route entries referencing a detaching network interface.
723  *
724  * Arguments:
725  *	rn	pointer to node in the routing table
726  *	arg	argument passed to rnh->rnh_walktree() - detaching interface
727  *
728  * Returns:
729  *	0	successful
730  *	errno	failed - reason indicated
731  *
732  */
733 static int
734 if_rtdel(struct radix_node *rn, void *arg)
735 {
736 	struct rtentry	*rt = (struct rtentry *)rn;
737 	struct ifnet	*ifp = arg;
738 	int		err;
739 
740 	if (rt->rt_ifp == ifp) {
741 
742 		/*
743 		 * Protect (sorta) against walktree recursion problems
744 		 * with cloned routes
745 		 */
746 		if ((rt->rt_flags & RTF_UP) == 0)
747 			return (0);
748 
749 		err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
750 				rt_mask(rt), rt->rt_flags,
751 				(struct rtentry **) NULL);
752 		if (err) {
753 			log(LOG_WARNING, "if_rtdel: error %d\n", err);
754 		}
755 	}
756 
757 	return (0);
758 }
759 
760 #define	sa_equal(a1, a2)	(bcmp((a1), (a2), ((a1))->sa_len) == 0)
761 
762 /*
763  * Locate an interface based on a complete address.
764  */
765 /*ARGSUSED*/
766 struct ifaddr *
767 ifa_ifwithaddr(struct sockaddr *addr)
768 {
769 	struct ifnet *ifp;
770 	struct ifaddr *ifa;
771 
772 	IFNET_RLOCK();
773 	TAILQ_FOREACH(ifp, &ifnet, if_link)
774 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
775 			if (ifa->ifa_addr->sa_family != addr->sa_family)
776 				continue;
777 			if (sa_equal(addr, ifa->ifa_addr))
778 				goto done;
779 			/* IP6 doesn't have broadcast */
780 			if ((ifp->if_flags & IFF_BROADCAST) &&
781 			    ifa->ifa_broadaddr &&
782 			    ifa->ifa_broadaddr->sa_len != 0 &&
783 			    sa_equal(ifa->ifa_broadaddr, addr))
784 				goto done;
785 		}
786 	ifa = NULL;
787 done:
788 	IFNET_RUNLOCK();
789 	return (ifa);
790 }
791 
792 /*
793  * Locate the point to point interface with a given destination address.
794  */
795 /*ARGSUSED*/
796 struct ifaddr *
797 ifa_ifwithdstaddr(struct sockaddr *addr)
798 {
799 	struct ifnet *ifp;
800 	struct ifaddr *ifa;
801 
802 	IFNET_RLOCK();
803 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
804 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
805 			continue;
806 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
807 			if (ifa->ifa_addr->sa_family != addr->sa_family)
808 				continue;
809 			if (ifa->ifa_dstaddr &&
810 			    sa_equal(addr, ifa->ifa_dstaddr))
811 				goto done;
812 		}
813 	}
814 	ifa = NULL;
815 done:
816 	IFNET_RUNLOCK();
817 	return (ifa);
818 }
819 
820 /*
821  * Find an interface on a specific network.  If many, choice
822  * is most specific found.
823  */
824 struct ifaddr *
825 ifa_ifwithnet(struct sockaddr *addr)
826 {
827 	struct ifnet *ifp;
828 	struct ifaddr *ifa;
829 	struct ifaddr *ifa_maybe = (struct ifaddr *) 0;
830 	u_int af = addr->sa_family;
831 	char *addr_data = addr->sa_data, *cplim;
832 
833 	/*
834 	 * AF_LINK addresses can be looked up directly by their index number,
835 	 * so do that if we can.
836 	 */
837 	if (af == AF_LINK) {
838 	    struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
839 	    if (sdl->sdl_index && sdl->sdl_index <= if_index)
840 		return (ifaddr_byindex(sdl->sdl_index));
841 	}
842 
843 	/*
844 	 * Scan though each interface, looking for ones that have
845 	 * addresses in this address family.
846 	 */
847 	IFNET_RLOCK();
848 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
849 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
850 			char *cp, *cp2, *cp3;
851 
852 			if (ifa->ifa_addr->sa_family != af)
853 next:				continue;
854 			if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
855 				/*
856 				 * This is a bit broken as it doesn't
857 				 * take into account that the remote end may
858 				 * be a single node in the network we are
859 				 * looking for.
860 				 * The trouble is that we don't know the
861 				 * netmask for the remote end.
862 				 */
863 				if (ifa->ifa_dstaddr != 0 &&
864 				    sa_equal(addr, ifa->ifa_dstaddr))
865 					goto done;
866 			} else {
867 				/*
868 				 * if we have a special address handler,
869 				 * then use it instead of the generic one.
870 				 */
871 				if (ifa->ifa_claim_addr) {
872 					if ((*ifa->ifa_claim_addr)(ifa, addr))
873 						goto done;
874 					continue;
875 				}
876 
877 				/*
878 				 * Scan all the bits in the ifa's address.
879 				 * If a bit dissagrees with what we are
880 				 * looking for, mask it with the netmask
881 				 * to see if it really matters.
882 				 * (A byte at a time)
883 				 */
884 				if (ifa->ifa_netmask == 0)
885 					continue;
886 				cp = addr_data;
887 				cp2 = ifa->ifa_addr->sa_data;
888 				cp3 = ifa->ifa_netmask->sa_data;
889 				cplim = ifa->ifa_netmask->sa_len
890 					+ (char *)ifa->ifa_netmask;
891 				while (cp3 < cplim)
892 					if ((*cp++ ^ *cp2++) & *cp3++)
893 						goto next; /* next address! */
894 				/*
895 				 * If the netmask of what we just found
896 				 * is more specific than what we had before
897 				 * (if we had one) then remember the new one
898 				 * before continuing to search
899 				 * for an even better one.
900 				 */
901 				if (ifa_maybe == 0 ||
902 				    rn_refines((caddr_t)ifa->ifa_netmask,
903 				    (caddr_t)ifa_maybe->ifa_netmask))
904 					ifa_maybe = ifa;
905 			}
906 		}
907 	}
908 	ifa = ifa_maybe;
909 done:
910 	IFNET_RUNLOCK();
911 	return (ifa);
912 }
913 
914 /*
915  * Find an interface address specific to an interface best matching
916  * a given address.
917  */
918 struct ifaddr *
919 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
920 {
921 	struct ifaddr *ifa;
922 	char *cp, *cp2, *cp3;
923 	char *cplim;
924 	struct ifaddr *ifa_maybe = 0;
925 	u_int af = addr->sa_family;
926 
927 	if (af >= AF_MAX)
928 		return (0);
929 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
930 		if (ifa->ifa_addr->sa_family != af)
931 			continue;
932 		if (ifa_maybe == 0)
933 			ifa_maybe = ifa;
934 		if (ifa->ifa_netmask == 0) {
935 			if (sa_equal(addr, ifa->ifa_addr) ||
936 			    (ifa->ifa_dstaddr &&
937 			    sa_equal(addr, ifa->ifa_dstaddr)))
938 				goto done;
939 			continue;
940 		}
941 		if (ifp->if_flags & IFF_POINTOPOINT) {
942 			if (sa_equal(addr, ifa->ifa_dstaddr))
943 				goto done;
944 		} else {
945 			cp = addr->sa_data;
946 			cp2 = ifa->ifa_addr->sa_data;
947 			cp3 = ifa->ifa_netmask->sa_data;
948 			cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
949 			for (; cp3 < cplim; cp3++)
950 				if ((*cp++ ^ *cp2++) & *cp3)
951 					break;
952 			if (cp3 == cplim)
953 				goto done;
954 		}
955 	}
956 	ifa = ifa_maybe;
957 done:
958 	return (ifa);
959 }
960 
961 #include <net/route.h>
962 
963 /*
964  * Default action when installing a route with a Link Level gateway.
965  * Lookup an appropriate real ifa to point to.
966  * This should be moved to /sys/net/link.c eventually.
967  */
968 static void
969 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
970 {
971 	struct ifaddr *ifa, *oifa;
972 	struct sockaddr *dst;
973 	struct ifnet *ifp;
974 
975 	RT_LOCK_ASSERT(rt);
976 
977 	if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
978 	    ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
979 		return;
980 	ifa = ifaof_ifpforaddr(dst, ifp);
981 	if (ifa) {
982 		IFAREF(ifa);		/* XXX */
983 		oifa = rt->rt_ifa;
984 		rt->rt_ifa = ifa;
985 		IFAFREE(oifa);
986 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
987 			ifa->ifa_rtrequest(cmd, rt, info);
988 	}
989 }
990 
991 /*
992  * Mark an interface down and notify protocols of
993  * the transition.
994  * NOTE: must be called at splnet or eqivalent.
995  */
996 static void
997 if_unroute(struct ifnet *ifp, int flag, int fam)
998 {
999 	struct ifaddr *ifa;
1000 
1001 	KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
1002 
1003 	ifp->if_flags &= ~flag;
1004 	getmicrotime(&ifp->if_lastchange);
1005 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1006 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1007 			pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1008 	if_qflush(&ifp->if_snd);
1009 #ifdef DEV_CARP
1010 	if (ifp->if_carp)
1011 		carp_carpdev_state(ifp->if_carp);
1012 #endif
1013 	rt_ifmsg(ifp);
1014 }
1015 
1016 /*
1017  * Mark an interface up and notify protocols of
1018  * the transition.
1019  * NOTE: must be called at splnet or eqivalent.
1020  */
1021 static void
1022 if_route(struct ifnet *ifp, int flag, int fam)
1023 {
1024 	struct ifaddr *ifa;
1025 
1026 	KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
1027 
1028 	ifp->if_flags |= flag;
1029 	getmicrotime(&ifp->if_lastchange);
1030 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1031 		if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1032 			pfctlinput(PRC_IFUP, ifa->ifa_addr);
1033 #ifdef DEV_CARP
1034 	if (ifp->if_carp)
1035 		carp_carpdev_state(ifp->if_carp);
1036 #endif
1037 	rt_ifmsg(ifp);
1038 #ifdef INET6
1039 	in6_if_up(ifp);
1040 #endif
1041 }
1042 
1043 void	(*vlan_link_state_p)(struct ifnet *, int);	/* XXX: private from if_vlan */
1044 
1045 /*
1046  * Handle a change in the interface link state. To avoid LORs
1047  * between driver lock and upper layer locks, as well as possible
1048  * recursions, we post event to taskqueue, and all job
1049  * is done in static do_link_state_change().
1050  */
1051 void
1052 if_link_state_change(struct ifnet *ifp, int link_state)
1053 {
1054 	/* Return if state hasn't changed. */
1055 	if (ifp->if_link_state == link_state)
1056 		return;
1057 
1058 	ifp->if_link_state = link_state;
1059 
1060 	taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
1061 }
1062 
1063 static void
1064 do_link_state_change(void *arg, int pending)
1065 {
1066 	struct ifnet *ifp = (struct ifnet *)arg;
1067 	int link_state = ifp->if_link_state;
1068 	int link;
1069 
1070 	/* Notify that the link state has changed. */
1071 	rt_ifmsg(ifp);
1072 	if (link_state == LINK_STATE_UP)
1073 		link = NOTE_LINKUP;
1074 	else if (link_state == LINK_STATE_DOWN)
1075 		link = NOTE_LINKDOWN;
1076 	else
1077 		link = NOTE_LINKINV;
1078 	KNOTE_UNLOCKED(&ifp->if_klist, link);
1079 	if (ifp->if_nvlans != 0)
1080 		(*vlan_link_state_p)(ifp, link);
1081 
1082 	if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
1083 	    IFP2AC(ifp)->ac_netgraph != NULL)
1084 		(*ng_ether_link_state_p)(ifp, link_state);
1085 #ifdef DEV_CARP
1086 	if (ifp->if_carp)
1087 		carp_carpdev_state(ifp->if_carp);
1088 #endif
1089 	if (ifp->if_bridge) {
1090 		KASSERT(bstp_linkstate_p != NULL,("if_bridge bstp not loaded!"));
1091 		(*bstp_linkstate_p)(ifp, link_state);
1092 	}
1093 
1094 	devctl_notify("IFNET", ifp->if_xname,
1095 	    (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1096 	if (pending > 1)
1097 		if_printf(ifp, "%d link states coalesced\n", pending);
1098 	if (log_link_state_change)
1099 		log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
1100 		    (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
1101 }
1102 
1103 /*
1104  * Mark an interface down and notify protocols of
1105  * the transition.
1106  * NOTE: must be called at splnet or eqivalent.
1107  */
1108 void
1109 if_down(struct ifnet *ifp)
1110 {
1111 
1112 	if_unroute(ifp, IFF_UP, AF_UNSPEC);
1113 }
1114 
1115 /*
1116  * Mark an interface up and notify protocols of
1117  * the transition.
1118  * NOTE: must be called at splnet or eqivalent.
1119  */
1120 void
1121 if_up(struct ifnet *ifp)
1122 {
1123 
1124 	if_route(ifp, IFF_UP, AF_UNSPEC);
1125 }
1126 
1127 /*
1128  * Flush an interface queue.
1129  */
1130 static void
1131 if_qflush(struct ifaltq *ifq)
1132 {
1133 	struct mbuf *m, *n;
1134 
1135 	IFQ_LOCK(ifq);
1136 #ifdef ALTQ
1137 	if (ALTQ_IS_ENABLED(ifq))
1138 		ALTQ_PURGE(ifq);
1139 #endif
1140 	n = ifq->ifq_head;
1141 	while ((m = n) != 0) {
1142 		n = m->m_act;
1143 		m_freem(m);
1144 	}
1145 	ifq->ifq_head = 0;
1146 	ifq->ifq_tail = 0;
1147 	ifq->ifq_len = 0;
1148 	IFQ_UNLOCK(ifq);
1149 }
1150 
1151 /*
1152  * Handle interface watchdog timer routines.  Called
1153  * from softclock, we decrement timers (if set) and
1154  * call the appropriate interface routine on expiration.
1155  *
1156  * XXXRW: Note that because timeouts run with Giant, if_watchdog() is called
1157  * holding Giant.  If we switch to an MPSAFE callout, we likely need to grab
1158  * Giant before entering if_watchdog() on an IFF_NEEDSGIANT interface.
1159  */
1160 static void
1161 if_slowtimo(void *arg)
1162 {
1163 	struct ifnet *ifp;
1164 	int s = splimp();
1165 
1166 	IFNET_RLOCK();
1167 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1168 		if (ifp->if_timer == 0 || --ifp->if_timer)
1169 			continue;
1170 		if (ifp->if_watchdog)
1171 			(*ifp->if_watchdog)(ifp);
1172 	}
1173 	IFNET_RUNLOCK();
1174 	splx(s);
1175 	timeout(if_slowtimo, (void *)0, hz / IFNET_SLOWHZ);
1176 }
1177 
1178 /*
1179  * Map interface name to
1180  * interface structure pointer.
1181  */
1182 struct ifnet *
1183 ifunit(const char *name)
1184 {
1185 	struct ifnet *ifp;
1186 
1187 	IFNET_RLOCK();
1188 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1189 		if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
1190 			break;
1191 	}
1192 	IFNET_RUNLOCK();
1193 	return (ifp);
1194 }
1195 
1196 /*
1197  * Hardware specific interface ioctls.
1198  */
1199 static int
1200 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
1201 {
1202 	struct ifreq *ifr;
1203 	struct ifstat *ifs;
1204 	int error = 0;
1205 	int new_flags, temp_flags;
1206 	size_t namelen, onamelen;
1207 	char new_name[IFNAMSIZ];
1208 	struct ifaddr *ifa;
1209 	struct sockaddr_dl *sdl;
1210 
1211 	ifr = (struct ifreq *)data;
1212 	switch (cmd) {
1213 	case SIOCGIFINDEX:
1214 		ifr->ifr_index = ifp->if_index;
1215 		break;
1216 
1217 	case SIOCGIFFLAGS:
1218 		temp_flags = ifp->if_flags | ifp->if_drv_flags;
1219 		ifr->ifr_flags = temp_flags & 0xffff;
1220 		ifr->ifr_flagshigh = temp_flags >> 16;
1221 		break;
1222 
1223 	case SIOCGIFCAP:
1224 		ifr->ifr_reqcap = ifp->if_capabilities;
1225 		ifr->ifr_curcap = ifp->if_capenable;
1226 		break;
1227 
1228 #ifdef MAC
1229 	case SIOCGIFMAC:
1230 		error = mac_ioctl_ifnet_get(td->td_ucred, ifr, ifp);
1231 		break;
1232 #endif
1233 
1234 	case SIOCGIFMETRIC:
1235 		ifr->ifr_metric = ifp->if_metric;
1236 		break;
1237 
1238 	case SIOCGIFMTU:
1239 		ifr->ifr_mtu = ifp->if_mtu;
1240 		break;
1241 
1242 	case SIOCGIFPHYS:
1243 		ifr->ifr_phys = ifp->if_physical;
1244 		break;
1245 
1246 	case SIOCSIFFLAGS:
1247 		error = suser(td);
1248 		if (error)
1249 			return (error);
1250 		/*
1251 		 * Currently, no driver owned flags pass the IFF_CANTCHANGE
1252 		 * check, so we don't need special handling here yet.
1253 		 */
1254 		new_flags = (ifr->ifr_flags & 0xffff) |
1255 		    (ifr->ifr_flagshigh << 16);
1256 		if (ifp->if_flags & IFF_SMART) {
1257 			/* Smart drivers twiddle their own routes */
1258 		} else if (ifp->if_flags & IFF_UP &&
1259 		    (new_flags & IFF_UP) == 0) {
1260 			int s = splimp();
1261 			if_down(ifp);
1262 			splx(s);
1263 		} else if (new_flags & IFF_UP &&
1264 		    (ifp->if_flags & IFF_UP) == 0) {
1265 			int s = splimp();
1266 			if_up(ifp);
1267 			splx(s);
1268 		}
1269 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
1270 			(new_flags &~ IFF_CANTCHANGE);
1271 		if (new_flags & IFF_PPROMISC) {
1272 			/* Permanently promiscuous mode requested */
1273 			ifp->if_flags |= IFF_PROMISC;
1274 		} else if (ifp->if_pcount == 0) {
1275 			ifp->if_flags &= ~IFF_PROMISC;
1276 		}
1277 		if (ifp->if_ioctl) {
1278 			IFF_LOCKGIANT(ifp);
1279 			(void) (*ifp->if_ioctl)(ifp, cmd, data);
1280 			IFF_UNLOCKGIANT(ifp);
1281 		}
1282 		getmicrotime(&ifp->if_lastchange);
1283 		break;
1284 
1285 	case SIOCSIFCAP:
1286 		error = suser(td);
1287 		if (error)
1288 			return (error);
1289 		if (ifp->if_ioctl == NULL)
1290 			return (EOPNOTSUPP);
1291 		if (ifr->ifr_reqcap & ~ifp->if_capabilities)
1292 			return (EINVAL);
1293 		IFF_LOCKGIANT(ifp);
1294 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1295 		IFF_UNLOCKGIANT(ifp);
1296 		if (error == 0)
1297 			getmicrotime(&ifp->if_lastchange);
1298 		break;
1299 
1300 #ifdef MAC
1301 	case SIOCSIFMAC:
1302 		error = mac_ioctl_ifnet_set(td->td_ucred, ifr, ifp);
1303 		break;
1304 #endif
1305 
1306 	case SIOCSIFNAME:
1307 		error = suser(td);
1308 		if (error != 0)
1309 			return (error);
1310 		error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
1311 		if (error != 0)
1312 			return (error);
1313 		if (new_name[0] == '\0')
1314 			return (EINVAL);
1315 		if (ifunit(new_name) != NULL)
1316 			return (EEXIST);
1317 
1318 		/* Announce the departure of the interface. */
1319 		rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1320 		EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
1321 
1322 		log(LOG_INFO, "%s: changing name to '%s'\n",
1323 		    ifp->if_xname, new_name);
1324 
1325 		strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
1326 		ifa = ifaddr_byindex(ifp->if_index);
1327 		IFA_LOCK(ifa);
1328 		sdl = (struct sockaddr_dl *)ifa->ifa_addr;
1329 		namelen = strlen(new_name);
1330 		onamelen = sdl->sdl_nlen;
1331 		/*
1332 		 * Move the address if needed.  This is safe because we
1333 		 * allocate space for a name of length IFNAMSIZ when we
1334 		 * create this in if_attach().
1335 		 */
1336 		if (namelen != onamelen) {
1337 			bcopy(sdl->sdl_data + onamelen,
1338 			    sdl->sdl_data + namelen, sdl->sdl_alen);
1339 		}
1340 		bcopy(new_name, sdl->sdl_data, namelen);
1341 		sdl->sdl_nlen = namelen;
1342 		sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
1343 		bzero(sdl->sdl_data, onamelen);
1344 		while (namelen != 0)
1345 			sdl->sdl_data[--namelen] = 0xff;
1346 		IFA_UNLOCK(ifa);
1347 
1348 		EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
1349 		/* Announce the return of the interface. */
1350 		rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
1351 		break;
1352 
1353 	case SIOCSIFMETRIC:
1354 		error = suser(td);
1355 		if (error)
1356 			return (error);
1357 		ifp->if_metric = ifr->ifr_metric;
1358 		getmicrotime(&ifp->if_lastchange);
1359 		break;
1360 
1361 	case SIOCSIFPHYS:
1362 		error = suser(td);
1363 		if (error)
1364 			return (error);
1365 		if (ifp->if_ioctl == NULL)
1366 			return (EOPNOTSUPP);
1367 		IFF_LOCKGIANT(ifp);
1368 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1369 		IFF_UNLOCKGIANT(ifp);
1370 		if (error == 0)
1371 			getmicrotime(&ifp->if_lastchange);
1372 		break;
1373 
1374 	case SIOCSIFMTU:
1375 	{
1376 		u_long oldmtu = ifp->if_mtu;
1377 
1378 		error = suser(td);
1379 		if (error)
1380 			return (error);
1381 		if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
1382 			return (EINVAL);
1383 		if (ifp->if_ioctl == NULL)
1384 			return (EOPNOTSUPP);
1385 		IFF_LOCKGIANT(ifp);
1386 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1387 		IFF_UNLOCKGIANT(ifp);
1388 		if (error == 0) {
1389 			getmicrotime(&ifp->if_lastchange);
1390 			rt_ifmsg(ifp);
1391 		}
1392 		/*
1393 		 * If the link MTU changed, do network layer specific procedure.
1394 		 */
1395 		if (ifp->if_mtu != oldmtu) {
1396 #ifdef INET6
1397 			nd6_setmtu(ifp);
1398 #endif
1399 		}
1400 		break;
1401 	}
1402 
1403 	case SIOCADDMULTI:
1404 	case SIOCDELMULTI:
1405 		error = suser(td);
1406 		if (error)
1407 			return (error);
1408 
1409 		/* Don't allow group membership on non-multicast interfaces. */
1410 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
1411 			return (EOPNOTSUPP);
1412 
1413 		/* Don't let users screw up protocols' entries. */
1414 		if (ifr->ifr_addr.sa_family != AF_LINK)
1415 			return (EINVAL);
1416 
1417 		if (cmd == SIOCADDMULTI) {
1418 			struct ifmultiaddr *ifma;
1419 			error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
1420 		} else {
1421 			error = if_delmulti(ifp, &ifr->ifr_addr);
1422 		}
1423 		if (error == 0)
1424 			getmicrotime(&ifp->if_lastchange);
1425 		break;
1426 
1427 	case SIOCSIFPHYADDR:
1428 	case SIOCDIFPHYADDR:
1429 #ifdef INET6
1430 	case SIOCSIFPHYADDR_IN6:
1431 #endif
1432 	case SIOCSLIFPHYADDR:
1433 	case SIOCSIFMEDIA:
1434 	case SIOCSIFGENERIC:
1435 		error = suser(td);
1436 		if (error)
1437 			return (error);
1438 		if (ifp->if_ioctl == NULL)
1439 			return (EOPNOTSUPP);
1440 		IFF_LOCKGIANT(ifp);
1441 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1442 		IFF_UNLOCKGIANT(ifp);
1443 		if (error == 0)
1444 			getmicrotime(&ifp->if_lastchange);
1445 		break;
1446 
1447 	case SIOCGIFSTATUS:
1448 		ifs = (struct ifstat *)data;
1449 		ifs->ascii[0] = '\0';
1450 
1451 	case SIOCGIFPSRCADDR:
1452 	case SIOCGIFPDSTADDR:
1453 	case SIOCGLIFPHYADDR:
1454 	case SIOCGIFMEDIA:
1455 	case SIOCGIFGENERIC:
1456 		if (ifp->if_ioctl == NULL)
1457 			return (EOPNOTSUPP);
1458 		IFF_LOCKGIANT(ifp);
1459 		error = (*ifp->if_ioctl)(ifp, cmd, data);
1460 		IFF_UNLOCKGIANT(ifp);
1461 		break;
1462 
1463 	case SIOCSIFLLADDR:
1464 		error = suser(td);
1465 		if (error)
1466 			return (error);
1467 		error = if_setlladdr(ifp,
1468 		    ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
1469 		break;
1470 
1471 	default:
1472 		error = ENOIOCTL;
1473 		break;
1474 	}
1475 	return (error);
1476 }
1477 
1478 /*
1479  * Interface ioctls.
1480  */
1481 int
1482 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
1483 {
1484 	struct ifnet *ifp;
1485 	struct ifreq *ifr;
1486 	int error;
1487 	int oif_flags;
1488 
1489 	switch (cmd) {
1490 	case SIOCGIFCONF:
1491 	case OSIOCGIFCONF:
1492 		return (ifconf(cmd, data));
1493 	}
1494 	ifr = (struct ifreq *)data;
1495 
1496 	switch (cmd) {
1497 	case SIOCIFCREATE:
1498 	case SIOCIFDESTROY:
1499 		if ((error = suser(td)) != 0)
1500 			return (error);
1501 		return ((cmd == SIOCIFCREATE) ?
1502 			if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name)) :
1503 			if_clone_destroy(ifr->ifr_name));
1504 
1505 	case SIOCIFGCLONERS:
1506 		return (if_clone_list((struct if_clonereq *)data));
1507 	}
1508 
1509 	ifp = ifunit(ifr->ifr_name);
1510 	if (ifp == 0)
1511 		return (ENXIO);
1512 
1513 	error = ifhwioctl(cmd, ifp, data, td);
1514 	if (error != ENOIOCTL)
1515 		return (error);
1516 
1517 	oif_flags = ifp->if_flags;
1518 	if (so->so_proto == 0)
1519 		return (EOPNOTSUPP);
1520 #ifndef COMPAT_43
1521 	error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
1522 								 data,
1523 								 ifp, td));
1524 #else
1525 	{
1526 		int ocmd = cmd;
1527 
1528 		switch (cmd) {
1529 
1530 		case SIOCSIFDSTADDR:
1531 		case SIOCSIFADDR:
1532 		case SIOCSIFBRDADDR:
1533 		case SIOCSIFNETMASK:
1534 #if BYTE_ORDER != BIG_ENDIAN
1535 			if (ifr->ifr_addr.sa_family == 0 &&
1536 			    ifr->ifr_addr.sa_len < 16) {
1537 				ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
1538 				ifr->ifr_addr.sa_len = 16;
1539 			}
1540 #else
1541 			if (ifr->ifr_addr.sa_len == 0)
1542 				ifr->ifr_addr.sa_len = 16;
1543 #endif
1544 			break;
1545 
1546 		case OSIOCGIFADDR:
1547 			cmd = SIOCGIFADDR;
1548 			break;
1549 
1550 		case OSIOCGIFDSTADDR:
1551 			cmd = SIOCGIFDSTADDR;
1552 			break;
1553 
1554 		case OSIOCGIFBRDADDR:
1555 			cmd = SIOCGIFBRDADDR;
1556 			break;
1557 
1558 		case OSIOCGIFNETMASK:
1559 			cmd = SIOCGIFNETMASK;
1560 		}
1561 		error =  ((*so->so_proto->pr_usrreqs->pru_control)(so,
1562 								   cmd,
1563 								   data,
1564 								   ifp, td));
1565 		switch (ocmd) {
1566 
1567 		case OSIOCGIFADDR:
1568 		case OSIOCGIFDSTADDR:
1569 		case OSIOCGIFBRDADDR:
1570 		case OSIOCGIFNETMASK:
1571 			*(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
1572 
1573 		}
1574 	}
1575 #endif /* COMPAT_43 */
1576 
1577 	if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
1578 #ifdef INET6
1579 		DELAY(100);/* XXX: temporary workaround for fxp issue*/
1580 		if (ifp->if_flags & IFF_UP) {
1581 			int s = splimp();
1582 			in6_if_up(ifp);
1583 			splx(s);
1584 		}
1585 #endif
1586 	}
1587 	return (error);
1588 }
1589 
1590 /*
1591  * The code common to handling reference counted flags,
1592  * e.g., in ifpromisc() and if_allmulti().
1593  * The "pflag" argument can specify a permanent mode flag,
1594  * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
1595  *
1596  * Only to be used on stack-owned flags, not driver-owned flags.
1597  */
1598 static int
1599 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
1600 {
1601 	struct ifreq ifr;
1602 	int error;
1603 	int oldflags, oldcount;
1604 
1605 	KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
1606 	    ("if_setflag: setting driver-ownded flag %d", flag));
1607 
1608 	/* Sanity checks to catch programming errors */
1609 	if (onswitch) {
1610 		if (*refcount < 0) {
1611 			if_printf(ifp,
1612 			    "refusing to increment negative refcount %d "
1613 			    "for interface flag %d\n", *refcount, flag);
1614 			return (EINVAL);
1615 		}
1616 	} else {
1617 		if (*refcount <= 0) {
1618 			if_printf(ifp,
1619 			    "refusing to decrement non-positive refcount %d"
1620 			    "for interface flag %d\n", *refcount, flag);
1621 			return (EINVAL);
1622 		}
1623 	}
1624 
1625 	/* In case this mode is permanent, just touch refcount */
1626 	if (ifp->if_flags & pflag) {
1627 		*refcount += onswitch ? 1 : -1;
1628 		return (0);
1629 	}
1630 
1631 	/* Save ifnet parameters for if_ioctl() may fail */
1632 	oldcount = *refcount;
1633 	oldflags = ifp->if_flags;
1634 
1635 	/*
1636 	 * See if we aren't the only and touching refcount is enough.
1637 	 * Actually toggle interface flag if we are the first or last.
1638 	 */
1639 	if (onswitch) {
1640 		if ((*refcount)++)
1641 			return (0);
1642 		ifp->if_flags |= flag;
1643 	} else {
1644 		if (--(*refcount))
1645 			return (0);
1646 		ifp->if_flags &= ~flag;
1647 	}
1648 
1649 	/* Call down the driver since we've changed interface flags */
1650 	if (ifp->if_ioctl == NULL) {
1651 		error = EOPNOTSUPP;
1652 		goto recover;
1653 	}
1654 	ifr.ifr_flags = ifp->if_flags & 0xffff;
1655 	ifr.ifr_flagshigh = ifp->if_flags >> 16;
1656 	IFF_LOCKGIANT(ifp);
1657 	error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1658 	IFF_UNLOCKGIANT(ifp);
1659 	if (error)
1660 		goto recover;
1661 	/* Notify userland that interface flags have changed */
1662 	rt_ifmsg(ifp);
1663 	return (0);
1664 
1665 recover:
1666 	/* Recover after driver error */
1667 	*refcount = oldcount;
1668 	ifp->if_flags = oldflags;
1669 	return (error);
1670 }
1671 
1672 /*
1673  * Set/clear promiscuous mode on interface ifp based on the truth value
1674  * of pswitch.  The calls are reference counted so that only the first
1675  * "on" request actually has an effect, as does the final "off" request.
1676  * Results are undefined if the "off" and "on" requests are not matched.
1677  */
1678 int
1679 ifpromisc(struct ifnet *ifp, int pswitch)
1680 {
1681 	int error;
1682 	int oldflags = ifp->if_flags;
1683 
1684 	error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
1685 			   &ifp->if_pcount, pswitch);
1686 	/* If promiscuous mode status has changed, log a message */
1687 	if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC))
1688 		log(LOG_INFO, "%s: promiscuous mode %s\n",
1689 		    ifp->if_xname,
1690 		    (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
1691 	return (error);
1692 }
1693 
1694 /*
1695  * Return interface configuration
1696  * of system.  List may be used
1697  * in later ioctl's (above) to get
1698  * other information.
1699  */
1700 /*ARGSUSED*/
1701 static int
1702 ifconf(u_long cmd, caddr_t data)
1703 {
1704 	struct ifconf *ifc = (struct ifconf *)data;
1705 	struct ifnet *ifp;
1706 	struct ifaddr *ifa;
1707 	struct ifreq ifr;
1708 	struct sbuf *sb;
1709 	int error, full = 0, valid_len, max_len;
1710 
1711 	/* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
1712 	max_len = MAXPHYS - 1;
1713 
1714 	/* Prevent hostile input from being able to crash the system */
1715 	if (ifc->ifc_len <= 0)
1716 		return (EINVAL);
1717 
1718 again:
1719 	if (ifc->ifc_len <= max_len) {
1720 		max_len = ifc->ifc_len;
1721 		full = 1;
1722 	}
1723 	sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
1724 	max_len = 0;
1725 	valid_len = 0;
1726 
1727 	IFNET_RLOCK();		/* could sleep XXX */
1728 	TAILQ_FOREACH(ifp, &ifnet, if_link) {
1729 		int addrs;
1730 
1731 		/*
1732 		 * Zero the ifr_name buffer to make sure we don't
1733 		 * disclose the contents of the stack.
1734 		 */
1735 		memset(ifr.ifr_name, 0, sizeof(ifr.ifr_name));
1736 
1737 		if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
1738 		    >= sizeof(ifr.ifr_name)) {
1739 			sbuf_delete(sb);
1740 			IFNET_RUNLOCK();
1741 			return (ENAMETOOLONG);
1742 		}
1743 
1744 		addrs = 0;
1745 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1746 			struct sockaddr *sa = ifa->ifa_addr;
1747 
1748 			if (jailed(curthread->td_ucred) &&
1749 			    prison_if(curthread->td_ucred, sa))
1750 				continue;
1751 			addrs++;
1752 #ifdef COMPAT_43
1753 			if (cmd == OSIOCGIFCONF) {
1754 				struct osockaddr *osa =
1755 					 (struct osockaddr *)&ifr.ifr_addr;
1756 				ifr.ifr_addr = *sa;
1757 				osa->sa_family = sa->sa_family;
1758 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1759 				max_len += sizeof(ifr);
1760 			} else
1761 #endif
1762 			if (sa->sa_len <= sizeof(*sa)) {
1763 				ifr.ifr_addr = *sa;
1764 				sbuf_bcat(sb, &ifr, sizeof(ifr));
1765 				max_len += sizeof(ifr);
1766 			} else {
1767 				sbuf_bcat(sb, &ifr,
1768 				    offsetof(struct ifreq, ifr_addr));
1769 				max_len += offsetof(struct ifreq, ifr_addr);
1770 				sbuf_bcat(sb, sa, sa->sa_len);
1771 				max_len += sa->sa_len;
1772 			}
1773 
1774 			if (!sbuf_overflowed(sb))
1775 				valid_len = sbuf_len(sb);
1776 		}
1777 		if (addrs == 0) {
1778 			bzero((caddr_t)&ifr.ifr_addr, sizeof(ifr.ifr_addr));
1779 			sbuf_bcat(sb, &ifr, sizeof(ifr));
1780 			max_len += sizeof(ifr);
1781 
1782 			if (!sbuf_overflowed(sb))
1783 				valid_len = sbuf_len(sb);
1784 		}
1785 	}
1786 	IFNET_RUNLOCK();
1787 
1788 	/*
1789 	 * If we didn't allocate enough space (uncommon), try again.  If
1790 	 * we have already allocated as much space as we are allowed,
1791 	 * return what we've got.
1792 	 */
1793 	if (valid_len != max_len && !full) {
1794 		sbuf_delete(sb);
1795 		goto again;
1796 	}
1797 
1798 	ifc->ifc_len = valid_len;
1799 	sbuf_finish(sb);
1800 	error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
1801 	sbuf_delete(sb);
1802 	return (error);
1803 }
1804 
1805 /*
1806  * Just like ifpromisc(), but for all-multicast-reception mode.
1807  */
1808 int
1809 if_allmulti(struct ifnet *ifp, int onswitch)
1810 {
1811 
1812 	return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
1813 }
1814 
1815 static struct ifmultiaddr *
1816 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
1817 {
1818 	struct ifmultiaddr *ifma;
1819 
1820 	IF_ADDR_LOCK_ASSERT(ifp);
1821 
1822 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1823 		if (sa_equal(ifma->ifma_addr, sa))
1824 			break;
1825 	}
1826 
1827 	return ifma;
1828 }
1829 
1830 /*
1831  * Allocate a new ifmultiaddr and initialize based on passed arguments.  We
1832  * make copies of passed sockaddrs.  The ifmultiaddr will not be added to
1833  * the ifnet multicast address list here, so the caller must do that and
1834  * other setup work (such as notifying the device driver).  The reference
1835  * count is initialized to 1.
1836  */
1837 static struct ifmultiaddr *
1838 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
1839     int mflags)
1840 {
1841 	struct ifmultiaddr *ifma;
1842 	struct sockaddr *dupsa;
1843 
1844 	MALLOC(ifma, struct ifmultiaddr *, sizeof *ifma, M_IFMADDR, mflags |
1845 	    M_ZERO);
1846 	if (ifma == NULL)
1847 		return (NULL);
1848 
1849 	MALLOC(dupsa, struct sockaddr *, sa->sa_len, M_IFMADDR, mflags);
1850 	if (dupsa == NULL) {
1851 		FREE(ifma, M_IFMADDR);
1852 		return (NULL);
1853 	}
1854 	bcopy(sa, dupsa, sa->sa_len);
1855 	ifma->ifma_addr = dupsa;
1856 
1857 	ifma->ifma_ifp = ifp;
1858 	ifma->ifma_refcount = 1;
1859 	ifma->ifma_protospec = NULL;
1860 
1861 	if (llsa == NULL) {
1862 		ifma->ifma_lladdr = NULL;
1863 		return (ifma);
1864 	}
1865 
1866 	MALLOC(dupsa, struct sockaddr *, llsa->sa_len, M_IFMADDR, mflags);
1867 	if (dupsa == NULL) {
1868 		FREE(ifma->ifma_addr, M_IFMADDR);
1869 		FREE(ifma, M_IFMADDR);
1870 		return (NULL);
1871 	}
1872 	bcopy(llsa, dupsa, llsa->sa_len);
1873 	ifma->ifma_lladdr = dupsa;
1874 
1875 	return (ifma);
1876 }
1877 
1878 /*
1879  * if_freemulti: free ifmultiaddr structure and possibly attached related
1880  * addresses.  The caller is responsible for implementing reference
1881  * counting, notifying the driver, handling routing messages, and releasing
1882  * any dependent link layer state.
1883  */
1884 static void
1885 if_freemulti(struct ifmultiaddr *ifma)
1886 {
1887 
1888 	KASSERT(ifma->ifma_refcount == 1, ("if_freemulti: refcount %d",
1889 	    ifma->ifma_refcount));
1890 	KASSERT(ifma->ifma_protospec == NULL,
1891 	    ("if_freemulti: protospec not NULL"));
1892 
1893 	if (ifma->ifma_lladdr != NULL)
1894 		FREE(ifma->ifma_lladdr, M_IFMADDR);
1895 	FREE(ifma->ifma_addr, M_IFMADDR);
1896 	FREE(ifma, M_IFMADDR);
1897 }
1898 
1899 /*
1900  * Register an additional multicast address with a network interface.
1901  *
1902  * - If the address is already present, bump the reference count on the
1903  *   address and return.
1904  * - If the address is not link-layer, look up a link layer address.
1905  * - Allocate address structures for one or both addresses, and attach to the
1906  *   multicast address list on the interface.  If automatically adding a link
1907  *   layer address, the protocol address will own a reference to the link
1908  *   layer address, to be freed when it is freed.
1909  * - Notify the network device driver of an addition to the multicast address
1910  *   list.
1911  *
1912  * 'sa' points to caller-owned memory with the desired multicast address.
1913  *
1914  * 'retifma' will be used to return a pointer to the resulting multicast
1915  * address reference, if desired.
1916  */
1917 int
1918 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
1919     struct ifmultiaddr **retifma)
1920 {
1921 	struct ifmultiaddr *ifma, *ll_ifma;
1922 	struct sockaddr *llsa;
1923 	int error;
1924 
1925 	/*
1926 	 * If the address is already present, return a new reference to it;
1927 	 * otherwise, allocate storage and set up a new address.
1928 	 */
1929 	IF_ADDR_LOCK(ifp);
1930 	ifma = if_findmulti(ifp, sa);
1931 	if (ifma != NULL) {
1932 		ifma->ifma_refcount++;
1933 		if (retifma != NULL)
1934 			*retifma = ifma;
1935 		IF_ADDR_UNLOCK(ifp);
1936 		return (0);
1937 	}
1938 
1939 	/*
1940 	 * The address isn't already present; resolve the protocol address
1941 	 * into a link layer address, and then look that up, bump its
1942 	 * refcount or allocate an ifma for that also.  If 'llsa' was
1943 	 * returned, we will need to free it later.
1944 	 */
1945 	llsa = NULL;
1946 	ll_ifma = NULL;
1947 	if (ifp->if_resolvemulti != NULL) {
1948 		error = ifp->if_resolvemulti(ifp, &llsa, sa);
1949 		if (error)
1950 			goto unlock_out;
1951 	}
1952 
1953 	/*
1954 	 * Allocate the new address.  Don't hook it up yet, as we may also
1955 	 * need to allocate a link layer multicast address.
1956 	 */
1957 	ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
1958 	if (ifma == NULL) {
1959 		error = ENOMEM;
1960 		goto free_llsa_out;
1961 	}
1962 
1963 	/*
1964 	 * If a link layer address is found, we'll need to see if it's
1965 	 * already present in the address list, or allocate is as well.
1966 	 * When this block finishes, the link layer address will be on the
1967 	 * list.
1968 	 */
1969 	if (llsa != NULL) {
1970 		ll_ifma = if_findmulti(ifp, llsa);
1971 		if (ll_ifma == NULL) {
1972 			ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
1973 			if (ll_ifma == NULL) {
1974 				if_freemulti(ifma);
1975 				error = ENOMEM;
1976 				goto free_llsa_out;
1977 			}
1978 			TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
1979 			    ifma_link);
1980 		} else
1981 			ll_ifma->ifma_refcount++;
1982 	}
1983 
1984 	/*
1985 	 * We now have a new multicast address, ifma, and possibly a new or
1986 	 * referenced link layer address.  Add the primary address to the
1987 	 * ifnet address list.
1988 	 */
1989 	TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
1990 
1991 	if (retifma != NULL)
1992 		*retifma = ifma;
1993 
1994 	/*
1995 	 * Must generate the message while holding the lock so that 'ifma'
1996 	 * pointer is still valid.
1997 	 *
1998 	 * XXXRW: How come we don't announce ll_ifma?
1999 	 */
2000 	rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2001 	IF_ADDR_UNLOCK(ifp);
2002 
2003 	/*
2004 	 * We are certain we have added something, so call down to the
2005 	 * interface to let them know about it.
2006 	 */
2007 	if (ifp->if_ioctl != NULL) {
2008 		IFF_LOCKGIANT(ifp);
2009 		(void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
2010 		IFF_UNLOCKGIANT(ifp);
2011 	}
2012 
2013 	if (llsa != NULL)
2014 		FREE(llsa, M_IFMADDR);
2015 
2016 	return (0);
2017 
2018 free_llsa_out:
2019 	if (llsa != NULL)
2020 		FREE(llsa, M_IFMADDR);
2021 
2022 unlock_out:
2023 	IF_ADDR_UNLOCK(ifp);
2024 	return (error);
2025 }
2026 
2027 /*
2028  * Remove a reference to a multicast address on this interface.  Yell
2029  * if the request does not match an existing membership.
2030  */
2031 int
2032 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2033 {
2034 	struct ifmultiaddr *ifma, *ll_ifma;
2035 
2036 	IF_ADDR_LOCK(ifp);
2037 	ifma = if_findmulti(ifp, sa);
2038 	if (ifma == NULL) {
2039 		IF_ADDR_UNLOCK(ifp);
2040 		return ENOENT;
2041 	}
2042 
2043 	if (ifma->ifma_refcount > 1) {
2044 		ifma->ifma_refcount--;
2045 		IF_ADDR_UNLOCK(ifp);
2046 		return 0;
2047 	}
2048 
2049 	sa = ifma->ifma_lladdr;
2050 	if (sa != NULL)
2051 		ll_ifma = if_findmulti(ifp, sa);
2052 	else
2053 		ll_ifma = NULL;
2054 
2055 	/*
2056 	 * XXXRW: How come we don't announce ll_ifma?
2057 	 */
2058 	rt_newmaddrmsg(RTM_DELMADDR, ifma);
2059 
2060 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2061 	if_freemulti(ifma);
2062 
2063 	if (ll_ifma != NULL) {
2064 		if (ll_ifma->ifma_refcount == 1) {
2065 			TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifma_link);
2066 			if_freemulti(ll_ifma);
2067 		} else
2068 			ll_ifma->ifma_refcount--;
2069 	}
2070 	IF_ADDR_UNLOCK(ifp);
2071 
2072 	/*
2073 	 * Make sure the interface driver is notified
2074 	 * in the case of a link layer mcast group being left.
2075 	 */
2076 	if (ifp->if_ioctl) {
2077 		IFF_LOCKGIANT(ifp);
2078 		(void) (*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
2079 		IFF_UNLOCKGIANT(ifp);
2080 	}
2081 
2082 	return 0;
2083 }
2084 
2085 /*
2086  * Set the link layer address on an interface.
2087  *
2088  * At this time we only support certain types of interfaces,
2089  * and we don't allow the length of the address to change.
2090  */
2091 int
2092 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2093 {
2094 	struct sockaddr_dl *sdl;
2095 	struct ifaddr *ifa;
2096 	struct ifreq ifr;
2097 
2098 	ifa = ifaddr_byindex(ifp->if_index);
2099 	if (ifa == NULL)
2100 		return (EINVAL);
2101 	sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2102 	if (sdl == NULL)
2103 		return (EINVAL);
2104 	if (len != sdl->sdl_alen)	/* don't allow length to change */
2105 		return (EINVAL);
2106 	switch (ifp->if_type) {
2107 	case IFT_ETHER:			/* these types use struct arpcom */
2108 	case IFT_FDDI:
2109 	case IFT_XETHER:
2110 	case IFT_ISO88025:
2111 	case IFT_L2VLAN:
2112 	case IFT_BRIDGE:
2113 		bcopy(lladdr, IFP2ENADDR(ifp), len);
2114 		/*
2115 		 * XXX We also need to store the lladdr in LLADDR(sdl),
2116 		 * which is done below. This is a pain because we must
2117 		 * remember to keep the info in sync.
2118 		 */
2119 		/* FALLTHROUGH */
2120 	case IFT_ARCNET:
2121 		bcopy(lladdr, LLADDR(sdl), len);
2122 		break;
2123 	default:
2124 		return (ENODEV);
2125 	}
2126 	/*
2127 	 * If the interface is already up, we need
2128 	 * to re-init it in order to reprogram its
2129 	 * address filter.
2130 	 */
2131 	if ((ifp->if_flags & IFF_UP) != 0) {
2132 		if (ifp->if_ioctl) {
2133 			IFF_LOCKGIANT(ifp);
2134 			ifp->if_flags &= ~IFF_UP;
2135 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2136 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2137 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2138 			ifp->if_flags |= IFF_UP;
2139 			ifr.ifr_flags = ifp->if_flags & 0xffff;
2140 			ifr.ifr_flagshigh = ifp->if_flags >> 16;
2141 			(*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2142 			IFF_UNLOCKGIANT(ifp);
2143 		}
2144 #ifdef INET
2145 		/*
2146 		 * Also send gratuitous ARPs to notify other nodes about
2147 		 * the address change.
2148 		 */
2149 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2150 			if (ifa->ifa_addr != NULL &&
2151 			    ifa->ifa_addr->sa_family == AF_INET)
2152 				arp_ifinit(ifp, ifa);
2153 		}
2154 #endif
2155 	}
2156 	return (0);
2157 }
2158 
2159 /*
2160  * The name argument must be a pointer to storage which will last as
2161  * long as the interface does.  For physical devices, the result of
2162  * device_get_name(dev) is a good choice and for pseudo-devices a
2163  * static string works well.
2164  */
2165 void
2166 if_initname(struct ifnet *ifp, const char *name, int unit)
2167 {
2168 	ifp->if_dname = name;
2169 	ifp->if_dunit = unit;
2170 	if (unit != IF_DUNIT_NONE)
2171 		snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2172 	else
2173 		strlcpy(ifp->if_xname, name, IFNAMSIZ);
2174 }
2175 
2176 int
2177 if_printf(struct ifnet *ifp, const char * fmt, ...)
2178 {
2179 	va_list ap;
2180 	int retval;
2181 
2182 	retval = printf("%s: ", ifp->if_xname);
2183 	va_start(ap, fmt);
2184 	retval += vprintf(fmt, ap);
2185 	va_end(ap);
2186 	return (retval);
2187 }
2188 
2189 /*
2190  * When an interface is marked IFF_NEEDSGIANT, its if_start() routine cannot
2191  * be called without Giant.  However, we often can't acquire the Giant lock
2192  * at those points; instead, we run it via a task queue that holds Giant via
2193  * if_start_deferred.
2194  *
2195  * XXXRW: We need to make sure that the ifnet isn't fully detached until any
2196  * outstanding if_start_deferred() tasks that will run after the free.  This
2197  * probably means waiting in if_detach().
2198  */
2199 void
2200 if_start(struct ifnet *ifp)
2201 {
2202 
2203 	NET_ASSERT_GIANT();
2204 
2205 	if ((ifp->if_flags & IFF_NEEDSGIANT) != 0 && debug_mpsafenet != 0) {
2206 		if (mtx_owned(&Giant))
2207 			(*(ifp)->if_start)(ifp);
2208 		else
2209 			taskqueue_enqueue(taskqueue_swi_giant,
2210 			    &ifp->if_starttask);
2211 	} else
2212 		(*(ifp)->if_start)(ifp);
2213 }
2214 
2215 static void
2216 if_start_deferred(void *context, int pending)
2217 {
2218 	struct ifnet *ifp;
2219 
2220 	/*
2221 	 * This code must be entered with Giant, and should never run if
2222 	 * we're not running with debug.mpsafenet.
2223 	 */
2224 	KASSERT(debug_mpsafenet != 0, ("if_start_deferred: debug.mpsafenet"));
2225 	GIANT_REQUIRED;
2226 
2227 	ifp = context;
2228 	(ifp->if_start)(ifp);
2229 }
2230 
2231 int
2232 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
2233 {
2234 	int active = 0;
2235 
2236 	IF_LOCK(ifq);
2237 	if (_IF_QFULL(ifq)) {
2238 		_IF_DROP(ifq);
2239 		IF_UNLOCK(ifq);
2240 		m_freem(m);
2241 		return (0);
2242 	}
2243 	if (ifp != NULL) {
2244 		ifp->if_obytes += m->m_pkthdr.len + adjust;
2245 		if (m->m_flags & (M_BCAST|M_MCAST))
2246 			ifp->if_omcasts++;
2247 		active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
2248 	}
2249 	_IF_ENQUEUE(ifq, m);
2250 	IF_UNLOCK(ifq);
2251 	if (ifp != NULL && !active)
2252 		if_start(ifp);
2253 	return (1);
2254 }
2255 
2256 void
2257 if_register_com_alloc(u_char type,
2258     if_com_alloc_t *a, if_com_free_t *f)
2259 {
2260 
2261 	KASSERT(if_com_alloc[type] == NULL,
2262 	    ("if_register_com_alloc: %d already registered", type));
2263 	KASSERT(if_com_free[type] == NULL,
2264 	    ("if_register_com_alloc: %d free already registered", type));
2265 
2266 	if_com_alloc[type] = a;
2267 	if_com_free[type] = f;
2268 }
2269 
2270 void
2271 if_deregister_com_alloc(u_char type)
2272 {
2273 
2274 	KASSERT(if_com_alloc[type] == NULL,
2275 	    ("if_deregister_com_alloc: %d not registered", type));
2276 	KASSERT(if_com_free[type] == NULL,
2277 	    ("if_deregister_com_alloc: %d free not registered", type));
2278 	if_com_alloc[type] = NULL;
2279 	if_com_free[type] = NULL;
2280 }
2281