xref: /netbsd-src/sys/net/if.c (revision a24efa7dea9f1f56c3bdb15a927d3516792ace1c)
1 /*	$NetBSD: if.c,v 1.337 2016/05/31 04:05:01 ozaki-r Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by William Studenmund and Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the project nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 /*
62  * Copyright (c) 1980, 1986, 1993
63  *	The Regents of the University of California.  All rights reserved.
64  *
65  * Redistribution and use in source and binary forms, with or without
66  * modification, are permitted provided that the following conditions
67  * are met:
68  * 1. Redistributions of source code must retain the above copyright
69  *    notice, this list of conditions and the following disclaimer.
70  * 2. Redistributions in binary form must reproduce the above copyright
71  *    notice, this list of conditions and the following disclaimer in the
72  *    documentation and/or other materials provided with the distribution.
73  * 3. Neither the name of the University nor the names of its contributors
74  *    may be used to endorse or promote products derived from this software
75  *    without specific prior written permission.
76  *
77  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
78  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
79  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
80  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
81  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
82  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
83  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
84  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
85  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
86  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
87  * SUCH DAMAGE.
88  *
89  *	@(#)if.c	8.5 (Berkeley) 1/9/95
90  */
91 
92 #include <sys/cdefs.h>
93 __KERNEL_RCSID(0, "$NetBSD: if.c,v 1.337 2016/05/31 04:05:01 ozaki-r Exp $");
94 
95 #if defined(_KERNEL_OPT)
96 #include "opt_inet.h"
97 
98 #include "opt_atalk.h"
99 #include "opt_natm.h"
100 #include "opt_wlan.h"
101 #include "opt_net_mpsafe.h"
102 #endif
103 
104 #include <sys/param.h>
105 #include <sys/mbuf.h>
106 #include <sys/systm.h>
107 #include <sys/callout.h>
108 #include <sys/proc.h>
109 #include <sys/socket.h>
110 #include <sys/socketvar.h>
111 #include <sys/domain.h>
112 #include <sys/protosw.h>
113 #include <sys/kernel.h>
114 #include <sys/ioctl.h>
115 #include <sys/sysctl.h>
116 #include <sys/syslog.h>
117 #include <sys/kauth.h>
118 #include <sys/kmem.h>
119 #include <sys/xcall.h>
120 #include <sys/cpu.h>
121 #include <sys/intr.h>
122 
123 #include <net/if.h>
124 #include <net/if_dl.h>
125 #include <net/if_ether.h>
126 #include <net/if_media.h>
127 #include <net80211/ieee80211.h>
128 #include <net80211/ieee80211_ioctl.h>
129 #include <net/if_types.h>
130 #include <net/route.h>
131 #include <net/netisr.h>
132 #include <sys/module.h>
133 #ifdef NETATALK
134 #include <netatalk/at_extern.h>
135 #include <netatalk/at.h>
136 #endif
137 #include <net/pfil.h>
138 #include <netinet/in.h>
139 #include <netinet/in_var.h>
140 
141 #ifdef INET6
142 #include <netinet6/in6_var.h>
143 #include <netinet6/nd6.h>
144 #endif
145 
146 #include "ether.h"
147 #include "fddi.h"
148 #include "token.h"
149 
150 #include "carp.h"
151 #if NCARP > 0
152 #include <netinet/ip_carp.h>
153 #endif
154 
155 #include <compat/sys/sockio.h>
156 #include <compat/sys/socket.h>
157 
158 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
159 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
160 
161 /*
162  * Global list of interfaces.
163  */
164 /* DEPRECATED. Remove it once kvm(3) users disappeared */
165 struct ifnet_head		ifnet_list;
166 
167 struct pslist_head		ifnet_pslist;
168 static ifnet_t **		ifindex2ifnet = NULL;
169 static u_int			if_index = 1;
170 static size_t			if_indexlim = 0;
171 static uint64_t			index_gen;
172 /* Mutex to protect the above objects. */
173 kmutex_t			ifnet_mtx __cacheline_aligned;
174 struct psref_class		*ifnet_psref_class __read_mostly;
175 static pserialize_t		ifnet_psz;
176 
177 static kmutex_t			if_clone_mtx;
178 
179 struct ifnet *lo0ifp;
180 int	ifqmaxlen = IFQ_MAXLEN;
181 
182 static int	if_rt_walktree(struct rtentry *, void *);
183 
184 static struct if_clone *if_clone_lookup(const char *, int *);
185 
186 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
187 static int if_cloners_count;
188 
189 /* Packet filtering hook for interfaces. */
190 pfil_head_t *	if_pfil;
191 
192 static kauth_listener_t if_listener;
193 
194 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
195 static void if_detach_queues(struct ifnet *, struct ifqueue *);
196 static void sysctl_sndq_setup(struct sysctllog **, const char *,
197     struct ifaltq *);
198 static void if_slowtimo(void *);
199 static void if_free_sadl(struct ifnet *);
200 static void if_attachdomain1(struct ifnet *);
201 static int ifconf(u_long, void *);
202 static int if_clone_create(const char *);
203 static int if_clone_destroy(const char *);
204 static void if_link_state_change_si(void *);
205 
206 struct if_percpuq {
207 	struct ifnet	*ipq_ifp;
208 	void		*ipq_si;
209 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
210 };
211 
212 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
213 
214 static void if_percpuq_drops(void *, void *, struct cpu_info *);
215 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
216 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
217     struct if_percpuq *);
218 
219 #if defined(INET) || defined(INET6)
220 static void sysctl_net_pktq_setup(struct sysctllog **, int);
221 #endif
222 
223 static int
224 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
225     void *arg0, void *arg1, void *arg2, void *arg3)
226 {
227 	int result;
228 	enum kauth_network_req req;
229 
230 	result = KAUTH_RESULT_DEFER;
231 	req = (enum kauth_network_req)arg1;
232 
233 	if (action != KAUTH_NETWORK_INTERFACE)
234 		return result;
235 
236 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
237 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
238 		result = KAUTH_RESULT_ALLOW;
239 
240 	return result;
241 }
242 
243 /*
244  * Network interface utility routines.
245  *
246  * Routines with ifa_ifwith* names take sockaddr *'s as
247  * parameters.
248  */
249 void
250 ifinit(void)
251 {
252 #if defined(INET)
253 	sysctl_net_pktq_setup(NULL, PF_INET);
254 #endif
255 #ifdef INET6
256 	if (in6_present)
257 		sysctl_net_pktq_setup(NULL, PF_INET6);
258 #endif
259 
260 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
261 	    if_listener_cb, NULL);
262 
263 	/* interfaces are available, inform socket code */
264 	ifioctl = doifioctl;
265 }
266 
267 /*
268  * XXX Initialization before configure().
269  * XXX hack to get pfil_add_hook working in autoconf.
270  */
271 void
272 ifinit1(void)
273 {
274 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
275 	TAILQ_INIT(&ifnet_list);
276 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
277 	ifnet_psz = pserialize_create();
278 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
279 	PSLIST_INIT(&ifnet_pslist);
280 	if_indexlim = 8;
281 
282 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
283 	KASSERT(if_pfil != NULL);
284 
285 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
286 	etherinit();
287 #endif
288 }
289 
290 ifnet_t *
291 if_alloc(u_char type)
292 {
293 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
294 }
295 
296 void
297 if_free(ifnet_t *ifp)
298 {
299 	kmem_free(ifp, sizeof(ifnet_t));
300 }
301 
302 void
303 if_initname(struct ifnet *ifp, const char *name, int unit)
304 {
305 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
306 	    "%s%d", name, unit);
307 }
308 
309 /*
310  * Null routines used while an interface is going away.  These routines
311  * just return an error.
312  */
313 
314 int
315 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
316     const struct sockaddr *so, const struct rtentry *rt)
317 {
318 
319 	return ENXIO;
320 }
321 
322 void
323 if_nullinput(struct ifnet *ifp, struct mbuf *m)
324 {
325 
326 	/* Nothing. */
327 }
328 
329 void
330 if_nullstart(struct ifnet *ifp)
331 {
332 
333 	/* Nothing. */
334 }
335 
336 int
337 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
338 {
339 
340 	return ENXIO;
341 }
342 
343 int
344 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
345 {
346 
347 	return ENXIO;
348 }
349 
350 int
351 if_nullinit(struct ifnet *ifp)
352 {
353 
354 	return ENXIO;
355 }
356 
357 void
358 if_nullstop(struct ifnet *ifp, int disable)
359 {
360 
361 	/* Nothing. */
362 }
363 
364 void
365 if_nullslowtimo(struct ifnet *ifp)
366 {
367 
368 	/* Nothing. */
369 }
370 
371 void
372 if_nulldrain(struct ifnet *ifp)
373 {
374 
375 	/* Nothing. */
376 }
377 
378 void
379 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
380 {
381 	struct ifaddr *ifa;
382 	struct sockaddr_dl *sdl;
383 
384 	ifp->if_addrlen = addrlen;
385 	if_alloc_sadl(ifp);
386 	ifa = ifp->if_dl;
387 	sdl = satosdl(ifa->ifa_addr);
388 
389 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
390 	if (factory) {
391 		ifp->if_hwdl = ifp->if_dl;
392 		ifaref(ifp->if_hwdl);
393 	}
394 	/* TBD routing socket */
395 }
396 
397 struct ifaddr *
398 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
399 {
400 	unsigned socksize, ifasize;
401 	int addrlen, namelen;
402 	struct sockaddr_dl *mask, *sdl;
403 	struct ifaddr *ifa;
404 
405 	namelen = strlen(ifp->if_xname);
406 	addrlen = ifp->if_addrlen;
407 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
408 	ifasize = sizeof(*ifa) + 2 * socksize;
409 	ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
410 
411 	sdl = (struct sockaddr_dl *)(ifa + 1);
412 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
413 
414 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
415 	    ifp->if_xname, namelen, NULL, addrlen);
416 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
417 	memset(&mask->sdl_data[0], 0xff, namelen);
418 	ifa->ifa_rtrequest = link_rtrequest;
419 	ifa->ifa_addr = (struct sockaddr *)sdl;
420 	ifa->ifa_netmask = (struct sockaddr *)mask;
421 
422 	*sdlp = sdl;
423 
424 	return ifa;
425 }
426 
427 static void
428 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
429 {
430 	const struct sockaddr_dl *sdl;
431 
432 	ifp->if_dl = ifa;
433 	ifaref(ifa);
434 	sdl = satosdl(ifa->ifa_addr);
435 	ifp->if_sadl = sdl;
436 }
437 
438 /*
439  * Allocate the link level name for the specified interface.  This
440  * is an attachment helper.  It must be called after ifp->if_addrlen
441  * is initialized, which may not be the case when if_attach() is
442  * called.
443  */
444 void
445 if_alloc_sadl(struct ifnet *ifp)
446 {
447 	struct ifaddr *ifa;
448 	const struct sockaddr_dl *sdl;
449 
450 	/*
451 	 * If the interface already has a link name, release it
452 	 * now.  This is useful for interfaces that can change
453 	 * link types, and thus switch link names often.
454 	 */
455 	if (ifp->if_sadl != NULL)
456 		if_free_sadl(ifp);
457 
458 	ifa = if_dl_create(ifp, &sdl);
459 
460 	ifa_insert(ifp, ifa);
461 	if_sadl_setrefs(ifp, ifa);
462 }
463 
464 static void
465 if_deactivate_sadl(struct ifnet *ifp)
466 {
467 	struct ifaddr *ifa;
468 
469 	KASSERT(ifp->if_dl != NULL);
470 
471 	ifa = ifp->if_dl;
472 
473 	ifp->if_sadl = NULL;
474 
475 	ifp->if_dl = NULL;
476 	ifafree(ifa);
477 }
478 
479 void
480 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa,
481     const struct sockaddr_dl *sdl)
482 {
483 	int s;
484 
485 	s = splnet();
486 
487 	if_deactivate_sadl(ifp);
488 
489 	if_sadl_setrefs(ifp, ifa);
490 	IFADDR_FOREACH(ifa, ifp)
491 		rtinit(ifa, RTM_LLINFO_UPD, 0);
492 	splx(s);
493 }
494 
495 /*
496  * Free the link level name for the specified interface.  This is
497  * a detach helper.  This is called from if_detach().
498  */
499 static void
500 if_free_sadl(struct ifnet *ifp)
501 {
502 	struct ifaddr *ifa;
503 	int s;
504 
505 	ifa = ifp->if_dl;
506 	if (ifa == NULL) {
507 		KASSERT(ifp->if_sadl == NULL);
508 		return;
509 	}
510 
511 	KASSERT(ifp->if_sadl != NULL);
512 
513 	s = splnet();
514 	rtinit(ifa, RTM_DELETE, 0);
515 	ifa_remove(ifp, ifa);
516 	if_deactivate_sadl(ifp);
517 	if (ifp->if_hwdl == ifa) {
518 		ifafree(ifa);
519 		ifp->if_hwdl = NULL;
520 	}
521 	splx(s);
522 }
523 
524 static void
525 if_getindex(ifnet_t *ifp)
526 {
527 	bool hitlimit = false;
528 
529 	ifp->if_index_gen = index_gen++;
530 
531 	ifp->if_index = if_index;
532 	if (ifindex2ifnet == NULL) {
533 		if_index++;
534 		goto skip;
535 	}
536 	while (if_byindex(ifp->if_index)) {
537 		/*
538 		 * If we hit USHRT_MAX, we skip back to 0 since
539 		 * there are a number of places where the value
540 		 * of if_index or if_index itself is compared
541 		 * to or stored in an unsigned short.  By
542 		 * jumping back, we won't botch those assignments
543 		 * or comparisons.
544 		 */
545 		if (++if_index == 0) {
546 			if_index = 1;
547 		} else if (if_index == USHRT_MAX) {
548 			/*
549 			 * However, if we have to jump back to
550 			 * zero *twice* without finding an empty
551 			 * slot in ifindex2ifnet[], then there
552 			 * there are too many (>65535) interfaces.
553 			 */
554 			if (hitlimit) {
555 				panic("too many interfaces");
556 			}
557 			hitlimit = true;
558 			if_index = 1;
559 		}
560 		ifp->if_index = if_index;
561 	}
562 skip:
563 	/*
564 	 * ifindex2ifnet is indexed by if_index. Since if_index will
565 	 * grow dynamically, it should grow too.
566 	 */
567 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
568 		size_t m, n, oldlim;
569 		void *q;
570 
571 		oldlim = if_indexlim;
572 		while (ifp->if_index >= if_indexlim)
573 			if_indexlim <<= 1;
574 
575 		/* grow ifindex2ifnet */
576 		m = oldlim * sizeof(struct ifnet *);
577 		n = if_indexlim * sizeof(struct ifnet *);
578 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
579 		if (ifindex2ifnet != NULL) {
580 			memcpy(q, ifindex2ifnet, m);
581 			free(ifindex2ifnet, M_IFADDR);
582 		}
583 		ifindex2ifnet = (struct ifnet **)q;
584 	}
585 	ifindex2ifnet[ifp->if_index] = ifp;
586 }
587 
588 /*
589  * Initialize an interface and assign an index for it.
590  *
591  * It must be called prior to a device specific attach routine
592  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
593  * and be followed by if_register:
594  *
595  *     if_initialize(ifp);
596  *     ether_ifattach(ifp, enaddr);
597  *     if_register(ifp);
598  */
599 void
600 if_initialize(ifnet_t *ifp)
601 {
602 	KASSERT(if_indexlim > 0);
603 	TAILQ_INIT(&ifp->if_addrlist);
604 
605 	/*
606 	 * Link level name is allocated later by a separate call to
607 	 * if_alloc_sadl().
608 	 */
609 
610 	if (ifp->if_snd.ifq_maxlen == 0)
611 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
612 
613 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
614 
615 	ifp->if_link_state = LINK_STATE_UNKNOWN;
616 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
617 
618 	ifp->if_capenable = 0;
619 	ifp->if_csum_flags_tx = 0;
620 	ifp->if_csum_flags_rx = 0;
621 
622 #ifdef ALTQ
623 	ifp->if_snd.altq_type = 0;
624 	ifp->if_snd.altq_disc = NULL;
625 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
626 	ifp->if_snd.altq_tbr  = NULL;
627 	ifp->if_snd.altq_ifp  = ifp;
628 #endif
629 
630 #ifdef NET_MPSAFE
631 	ifp->if_snd.ifq_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
632 #else
633 	ifp->if_snd.ifq_lock = NULL;
634 #endif
635 
636 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
637 	(void)pfil_run_hooks(if_pfil,
638 	    (struct mbuf **)PFIL_IFNET_ATTACH, ifp, PFIL_IFNET);
639 
640 	IF_AFDATA_LOCK_INIT(ifp);
641 
642 	ifp->if_link_si = softint_establish(SOFTINT_NET, if_link_state_change_si, ifp);
643 	if (ifp->if_link_si == NULL)
644 		panic("%s: softint_establish() failed", __func__);
645 
646 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
647 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
648 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
649 
650 	IFNET_LOCK();
651 	if_getindex(ifp);
652 	IFNET_UNLOCK();
653 }
654 
655 /*
656  * Register an interface to the list of "active" interfaces.
657  */
658 void
659 if_register(ifnet_t *ifp)
660 {
661 	/*
662 	 * If the driver has not supplied its own if_ioctl, then
663 	 * supply the default.
664 	 */
665 	if (ifp->if_ioctl == NULL)
666 		ifp->if_ioctl = ifioctl_common;
667 
668 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
669 
670 	if (!STAILQ_EMPTY(&domains))
671 		if_attachdomain1(ifp);
672 
673 	/* Announce the interface. */
674 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
675 
676 	if (ifp->if_slowtimo != NULL) {
677 		ifp->if_slowtimo_ch =
678 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
679 		callout_init(ifp->if_slowtimo_ch, 0);
680 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
681 		if_slowtimo(ifp);
682 	}
683 
684 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
685 		ifp->if_transmit = if_transmit;
686 
687 	IFNET_LOCK();
688 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
689 	IFNET_WRITER_INSERT_TAIL(ifp);
690 	IFNET_UNLOCK();
691 }
692 
693 /*
694  * The if_percpuq framework
695  *
696  * It allows network device drivers to execute the network stack
697  * in softint (so called softint-based if_input). It utilizes
698  * softint and percpu ifqueue. It doesn't distribute any packets
699  * between CPUs, unlike pktqueue(9).
700  *
701  * Currently we support two options for device drivers to apply the framework:
702  * - Use it implicitly with less changes
703  *   - If you use if_attach in driver's _attach function and if_input in
704  *     driver's Rx interrupt handler, a packet is queued and a softint handles
705  *     the packet implicitly
706  * - Use it explicitly in each driver (recommended)
707  *   - You can use if_percpuq_* directly in your driver
708  *   - In this case, you need to allocate struct if_percpuq in driver's softc
709  *   - See wm(4) as a reference implementation
710  */
711 
712 static void
713 if_percpuq_softint(void *arg)
714 {
715 	struct if_percpuq *ipq = arg;
716 	struct ifnet *ifp = ipq->ipq_ifp;
717 	struct mbuf *m;
718 
719 	while ((m = if_percpuq_dequeue(ipq)) != NULL)
720 		ifp->_if_input(ifp, m);
721 }
722 
723 static void
724 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
725 {
726 	struct ifqueue *const ifq = p;
727 
728 	memset(ifq, 0, sizeof(*ifq));
729 	ifq->ifq_maxlen = IFQ_MAXLEN;
730 }
731 
732 struct if_percpuq *
733 if_percpuq_create(struct ifnet *ifp)
734 {
735 	struct if_percpuq *ipq;
736 
737 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
738 	if (ipq == NULL)
739 		panic("kmem_zalloc failed");
740 
741 	ipq->ipq_ifp = ifp;
742 	ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
743 	    if_percpuq_softint, ipq);
744 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
745 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
746 
747 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
748 
749 	return ipq;
750 }
751 
752 static struct mbuf *
753 if_percpuq_dequeue(struct if_percpuq *ipq)
754 {
755 	struct mbuf *m;
756 	struct ifqueue *ifq;
757 	int s;
758 
759 	s = splnet();
760 	ifq = percpu_getref(ipq->ipq_ifqs);
761 	IF_DEQUEUE(ifq, m);
762 	percpu_putref(ipq->ipq_ifqs);
763 	splx(s);
764 
765 	return m;
766 }
767 
768 static void
769 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
770 {
771 	struct ifqueue *const ifq = p;
772 
773 	IF_PURGE(ifq);
774 }
775 
776 void
777 if_percpuq_destroy(struct if_percpuq *ipq)
778 {
779 
780 	/* if_detach may already destroy it */
781 	if (ipq == NULL)
782 		return;
783 
784 	softint_disestablish(ipq->ipq_si);
785 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
786 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
787 }
788 
789 void
790 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
791 {
792 	struct ifqueue *ifq;
793 	int s;
794 
795 	KASSERT(ipq != NULL);
796 
797 	s = splnet();
798 	ifq = percpu_getref(ipq->ipq_ifqs);
799 	if (IF_QFULL(ifq)) {
800 		IF_DROP(ifq);
801 		percpu_putref(ipq->ipq_ifqs);
802 		m_freem(m);
803 		goto out;
804 	}
805 	IF_ENQUEUE(ifq, m);
806 	percpu_putref(ipq->ipq_ifqs);
807 
808 	softint_schedule(ipq->ipq_si);
809 out:
810 	splx(s);
811 }
812 
813 static void
814 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
815 {
816 	struct ifqueue *const ifq = p;
817 	int *sum = arg;
818 
819 	*sum += ifq->ifq_drops;
820 }
821 
822 static int
823 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
824 {
825 	struct sysctlnode node;
826 	struct if_percpuq *ipq;
827 	int sum = 0;
828 	int error;
829 
830 	node = *rnode;
831 	ipq = node.sysctl_data;
832 
833 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
834 
835 	node.sysctl_data = &sum;
836 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
837 	if (error != 0 || newp == NULL)
838 		return error;
839 
840 	return 0;
841 }
842 
843 static void
844 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
845     struct if_percpuq *ipq)
846 {
847 	const struct sysctlnode *cnode, *rnode;
848 
849 	if (sysctl_createv(clog, 0, NULL, &rnode,
850 		       CTLFLAG_PERMANENT,
851 		       CTLTYPE_NODE, "interfaces",
852 		       SYSCTL_DESCR("Per-interface controls"),
853 		       NULL, 0, NULL, 0,
854 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
855 		goto bad;
856 
857 	if (sysctl_createv(clog, 0, &rnode, &rnode,
858 		       CTLFLAG_PERMANENT,
859 		       CTLTYPE_NODE, ifname,
860 		       SYSCTL_DESCR("Interface controls"),
861 		       NULL, 0, NULL, 0,
862 		       CTL_CREATE, CTL_EOL) != 0)
863 		goto bad;
864 
865 	if (sysctl_createv(clog, 0, &rnode, &rnode,
866 		       CTLFLAG_PERMANENT,
867 		       CTLTYPE_NODE, "rcvq",
868 		       SYSCTL_DESCR("Interface input queue controls"),
869 		       NULL, 0, NULL, 0,
870 		       CTL_CREATE, CTL_EOL) != 0)
871 		goto bad;
872 
873 #ifdef NOTYET
874 	/* XXX Should show each per-CPU queue length? */
875 	if (sysctl_createv(clog, 0, &rnode, &rnode,
876 		       CTLFLAG_PERMANENT,
877 		       CTLTYPE_INT, "len",
878 		       SYSCTL_DESCR("Current input queue length"),
879 		       sysctl_percpuq_len, 0, NULL, 0,
880 		       CTL_CREATE, CTL_EOL) != 0)
881 		goto bad;
882 
883 	if (sysctl_createv(clog, 0, &rnode, &cnode,
884 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
885 		       CTLTYPE_INT, "maxlen",
886 		       SYSCTL_DESCR("Maximum allowed input queue length"),
887 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
888 		       CTL_CREATE, CTL_EOL) != 0)
889 		goto bad;
890 #endif
891 
892 	if (sysctl_createv(clog, 0, &rnode, &cnode,
893 		       CTLFLAG_PERMANENT,
894 		       CTLTYPE_INT, "drops",
895 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
896 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
897 		       CTL_CREATE, CTL_EOL) != 0)
898 		goto bad;
899 
900 	return;
901 bad:
902 	printf("%s: could not attach sysctl nodes\n", ifname);
903 	return;
904 }
905 
906 
907 /*
908  * The common interface input routine that is called by device drivers,
909  * which should be used only when the driver's rx handler already runs
910  * in softint.
911  */
912 void
913 if_input(struct ifnet *ifp, struct mbuf *m)
914 {
915 
916 	KASSERT(ifp->if_percpuq == NULL);
917 	KASSERT(!cpu_intr_p());
918 
919 	ifp->_if_input(ifp, m);
920 }
921 
922 /*
923  * DEPRECATED. Use if_initialize and if_register instead.
924  * See the above comment of if_initialize.
925  *
926  * Note that it implicitly enables if_percpuq to make drivers easy to
927  * migrate softint-based if_input without much changes. If you don't
928  * want to enable it, use if_initialize instead.
929  */
930 void
931 if_attach(ifnet_t *ifp)
932 {
933 
934 	if_initialize(ifp);
935 	ifp->if_percpuq = if_percpuq_create(ifp);
936 	if_register(ifp);
937 }
938 
939 void
940 if_attachdomain(void)
941 {
942 	struct ifnet *ifp;
943 	int s;
944 	int bound = curlwp->l_pflag & LP_BOUND;
945 
946 	curlwp->l_pflag |= LP_BOUND;
947 	s = pserialize_read_enter();
948 	IFNET_READER_FOREACH(ifp) {
949 		struct psref psref;
950 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
951 		pserialize_read_exit(s);
952 		if_attachdomain1(ifp);
953 		s = pserialize_read_enter();
954 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
955 	}
956 	pserialize_read_exit(s);
957 	curlwp->l_pflag ^= bound ^ LP_BOUND;
958 }
959 
960 static void
961 if_attachdomain1(struct ifnet *ifp)
962 {
963 	struct domain *dp;
964 	int s;
965 
966 	s = splnet();
967 
968 	/* address family dependent data region */
969 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
970 	DOMAIN_FOREACH(dp) {
971 		if (dp->dom_ifattach != NULL)
972 			ifp->if_afdata[dp->dom_family] =
973 			    (*dp->dom_ifattach)(ifp);
974 	}
975 
976 	splx(s);
977 }
978 
979 /*
980  * Deactivate an interface.  This points all of the procedure
981  * handles at error stubs.  May be called from interrupt context.
982  */
983 void
984 if_deactivate(struct ifnet *ifp)
985 {
986 	int s;
987 
988 	s = splnet();
989 
990 	ifp->if_output	 = if_nulloutput;
991 	ifp->_if_input	 = if_nullinput;
992 	ifp->if_start	 = if_nullstart;
993 	ifp->if_transmit = if_nulltransmit;
994 	ifp->if_ioctl	 = if_nullioctl;
995 	ifp->if_init	 = if_nullinit;
996 	ifp->if_stop	 = if_nullstop;
997 	ifp->if_slowtimo = if_nullslowtimo;
998 	ifp->if_drain	 = if_nulldrain;
999 
1000 	/* No more packets may be enqueued. */
1001 	ifp->if_snd.ifq_maxlen = 0;
1002 
1003 	splx(s);
1004 }
1005 
1006 void
1007 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1008 {
1009 	struct ifaddr *ifa, *nifa;
1010 
1011 	IFADDR_FOREACH_SAFE(ifa, ifp, nifa) {
1012 		if (ifa->ifa_addr->sa_family != family)
1013 			continue;
1014 		(*purgeaddr)(ifa);
1015 	}
1016 }
1017 
1018 /*
1019  * Detach an interface from the list of "active" interfaces,
1020  * freeing any resources as we go along.
1021  *
1022  * NOTE: This routine must be called with a valid thread context,
1023  * as it may block.
1024  */
1025 void
1026 if_detach(struct ifnet *ifp)
1027 {
1028 	struct socket so;
1029 	struct ifaddr *ifa;
1030 #ifdef IFAREF_DEBUG
1031 	struct ifaddr *last_ifa = NULL;
1032 #endif
1033 	struct domain *dp;
1034 	const struct protosw *pr;
1035 	int s, i, family, purged;
1036 	uint64_t xc;
1037 
1038 	/*
1039 	 * XXX It's kind of lame that we have to have the
1040 	 * XXX socket structure...
1041 	 */
1042 	memset(&so, 0, sizeof(so));
1043 
1044 	s = splnet();
1045 
1046 	sysctl_teardown(&ifp->if_sysctl_log);
1047 	mutex_enter(ifp->if_ioctl_lock);
1048 	ifp->if_ioctl = if_nullioctl;
1049 	mutex_exit(ifp->if_ioctl_lock);
1050 
1051 	IFNET_LOCK();
1052 	ifindex2ifnet[ifp->if_index] = NULL;
1053 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1054 	IFNET_WRITER_REMOVE(ifp);
1055 	pserialize_perform(ifnet_psz);
1056 	IFNET_UNLOCK();
1057 
1058 	/* Wait for all readers to drain before freeing.  */
1059 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1060 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1061 
1062 	mutex_obj_free(ifp->if_ioctl_lock);
1063 	ifp->if_ioctl_lock = NULL;
1064 
1065 	if (ifp->if_slowtimo != NULL) {
1066 		ifp->if_slowtimo = NULL;
1067 		callout_halt(ifp->if_slowtimo_ch, NULL);
1068 		callout_destroy(ifp->if_slowtimo_ch);
1069 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1070 	}
1071 
1072 	/*
1073 	 * Do an if_down() to give protocols a chance to do something.
1074 	 */
1075 	if_down(ifp);
1076 
1077 #ifdef ALTQ
1078 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1079 		altq_disable(&ifp->if_snd);
1080 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1081 		altq_detach(&ifp->if_snd);
1082 #endif
1083 
1084 	if (ifp->if_snd.ifq_lock)
1085 		mutex_obj_free(ifp->if_snd.ifq_lock);
1086 
1087 #if NCARP > 0
1088 	/* Remove the interface from any carp group it is a part of.  */
1089 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1090 		carp_ifdetach(ifp);
1091 #endif
1092 
1093 	/*
1094 	 * Rip all the addresses off the interface.  This should make
1095 	 * all of the routes go away.
1096 	 *
1097 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1098 	 * from the list, including our "cursor", ifa.  For safety,
1099 	 * and to honor the TAILQ abstraction, I just restart the
1100 	 * loop after each removal.  Note that the loop will exit
1101 	 * when all of the remaining ifaddrs belong to the AF_LINK
1102 	 * family.  I am counting on the historical fact that at
1103 	 * least one pr_usrreq in each address domain removes at
1104 	 * least one ifaddr.
1105 	 */
1106 again:
1107 	IFADDR_FOREACH(ifa, ifp) {
1108 		family = ifa->ifa_addr->sa_family;
1109 #ifdef IFAREF_DEBUG
1110 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1111 		    ifa, family, ifa->ifa_refcnt);
1112 		if (last_ifa != NULL && ifa == last_ifa)
1113 			panic("if_detach: loop detected");
1114 		last_ifa = ifa;
1115 #endif
1116 		if (family == AF_LINK)
1117 			continue;
1118 		dp = pffinddomain(family);
1119 #ifdef DIAGNOSTIC
1120 		if (dp == NULL)
1121 			panic("if_detach: no domain for AF %d",
1122 			    family);
1123 #endif
1124 		/*
1125 		 * XXX These PURGEIF calls are redundant with the
1126 		 * purge-all-families calls below, but are left in for
1127 		 * now both to make a smaller change, and to avoid
1128 		 * unplanned interactions with clearing of
1129 		 * ifp->if_addrlist.
1130 		 */
1131 		purged = 0;
1132 		for (pr = dp->dom_protosw;
1133 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1134 			so.so_proto = pr;
1135 			if (pr->pr_usrreqs) {
1136 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1137 				purged = 1;
1138 			}
1139 		}
1140 		if (purged == 0) {
1141 			/*
1142 			 * XXX What's really the best thing to do
1143 			 * XXX here?  --thorpej@NetBSD.org
1144 			 */
1145 			printf("if_detach: WARNING: AF %d not purged\n",
1146 			    family);
1147 			ifa_remove(ifp, ifa);
1148 		}
1149 		goto again;
1150 	}
1151 
1152 	if_free_sadl(ifp);
1153 
1154 	/* Walk the routing table looking for stragglers. */
1155 	for (i = 0; i <= AF_MAX; i++) {
1156 		while (rt_walktree(i, if_rt_walktree, ifp) == ERESTART)
1157 			continue;
1158 	}
1159 
1160 	DOMAIN_FOREACH(dp) {
1161 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1162 		{
1163 			void *p = ifp->if_afdata[dp->dom_family];
1164 			if (p) {
1165 				ifp->if_afdata[dp->dom_family] = NULL;
1166 				(*dp->dom_ifdetach)(ifp, p);
1167 			}
1168 		}
1169 
1170 		/*
1171 		 * One would expect multicast memberships (INET and
1172 		 * INET6) on UDP sockets to be purged by the PURGEIF
1173 		 * calls above, but if all addresses were removed from
1174 		 * the interface prior to destruction, the calls will
1175 		 * not be made (e.g. ppp, for which pppd(8) generally
1176 		 * removes addresses before destroying the interface).
1177 		 * Because there is no invariant that multicast
1178 		 * memberships only exist for interfaces with IPv4
1179 		 * addresses, we must call PURGEIF regardless of
1180 		 * addresses.  (Protocols which might store ifnet
1181 		 * pointers are marked with PR_PURGEIF.)
1182 		 */
1183 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1184 			so.so_proto = pr;
1185 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1186 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1187 		}
1188 	}
1189 
1190 	(void)pfil_run_hooks(if_pfil,
1191 	    (struct mbuf **)PFIL_IFNET_DETACH, ifp, PFIL_IFNET);
1192 	(void)pfil_head_destroy(ifp->if_pfil);
1193 
1194 	/* Announce that the interface is gone. */
1195 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1196 
1197 	IF_AFDATA_LOCK_DESTROY(ifp);
1198 
1199 	softint_disestablish(ifp->if_link_si);
1200 	ifp->if_link_si = NULL;
1201 
1202 	/*
1203 	 * remove packets that came from ifp, from software interrupt queues.
1204 	 */
1205 	DOMAIN_FOREACH(dp) {
1206 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1207 			struct ifqueue *iq = dp->dom_ifqueues[i];
1208 			if (iq == NULL)
1209 				break;
1210 			dp->dom_ifqueues[i] = NULL;
1211 			if_detach_queues(ifp, iq);
1212 		}
1213 	}
1214 
1215 	/*
1216 	 * IP queues have to be processed separately: net-queue barrier
1217 	 * ensures that the packets are dequeued while a cross-call will
1218 	 * ensure that the interrupts have completed. FIXME: not quite..
1219 	 */
1220 #ifdef INET
1221 	pktq_barrier(ip_pktq);
1222 #endif
1223 #ifdef INET6
1224 	if (in6_present)
1225 		pktq_barrier(ip6_pktq);
1226 #endif
1227 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1228 	xc_wait(xc);
1229 
1230 	if (ifp->if_percpuq != NULL) {
1231 		if_percpuq_destroy(ifp->if_percpuq);
1232 		ifp->if_percpuq = NULL;
1233 	}
1234 
1235 	splx(s);
1236 }
1237 
1238 static void
1239 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1240 {
1241 	struct mbuf *m, *prev, *next;
1242 
1243 	prev = NULL;
1244 	for (m = q->ifq_head; m != NULL; m = next) {
1245 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1246 
1247 		next = m->m_nextpkt;
1248 		if (m->m_pkthdr.rcvif != ifp) {
1249 			prev = m;
1250 			continue;
1251 		}
1252 
1253 		if (prev != NULL)
1254 			prev->m_nextpkt = m->m_nextpkt;
1255 		else
1256 			q->ifq_head = m->m_nextpkt;
1257 		if (q->ifq_tail == m)
1258 			q->ifq_tail = prev;
1259 		q->ifq_len--;
1260 
1261 		m->m_nextpkt = NULL;
1262 		m_freem(m);
1263 		IF_DROP(q);
1264 	}
1265 }
1266 
1267 /*
1268  * Callback for a radix tree walk to delete all references to an
1269  * ifnet.
1270  */
1271 static int
1272 if_rt_walktree(struct rtentry *rt, void *v)
1273 {
1274 	struct ifnet *ifp = (struct ifnet *)v;
1275 	int error;
1276 	struct rtentry *retrt;
1277 
1278 	if (rt->rt_ifp != ifp)
1279 		return 0;
1280 
1281 	/* Delete the entry. */
1282 	error = rtrequest(RTM_DELETE, rt_getkey(rt), rt->rt_gateway,
1283 	    rt_mask(rt), rt->rt_flags, &retrt);
1284 	if (error == 0) {
1285 		KASSERT(retrt == rt);
1286 		KASSERT((retrt->rt_flags & RTF_UP) == 0);
1287 		retrt->rt_ifp = NULL;
1288 		rtfree(retrt);
1289 	} else {
1290 		printf("%s: warning: unable to delete rtentry @ %p, "
1291 		    "error = %d\n", ifp->if_xname, rt, error);
1292 	}
1293 	return ERESTART;
1294 }
1295 
1296 /*
1297  * Create a clone network interface.
1298  */
1299 static int
1300 if_clone_create(const char *name)
1301 {
1302 	struct if_clone *ifc;
1303 	int unit;
1304 	struct ifnet *ifp;
1305 	struct psref psref;
1306 
1307 	ifc = if_clone_lookup(name, &unit);
1308 	if (ifc == NULL)
1309 		return EINVAL;
1310 
1311 	ifp = if_get(name, &psref);
1312 	if (ifp != NULL) {
1313 		if_put(ifp, &psref);
1314 		return EEXIST;
1315 	}
1316 
1317 	return (*ifc->ifc_create)(ifc, unit);
1318 }
1319 
1320 /*
1321  * Destroy a clone network interface.
1322  */
1323 static int
1324 if_clone_destroy(const char *name)
1325 {
1326 	struct if_clone *ifc;
1327 	struct ifnet *ifp;
1328 	struct psref psref;
1329 
1330 	ifc = if_clone_lookup(name, NULL);
1331 	if (ifc == NULL)
1332 		return EINVAL;
1333 
1334 	if (ifc->ifc_destroy == NULL)
1335 		return EOPNOTSUPP;
1336 
1337 	ifp = if_get(name, &psref);
1338 	if (ifp == NULL)
1339 		return ENXIO;
1340 
1341 	/* We have to disable ioctls here */
1342 	mutex_enter(ifp->if_ioctl_lock);
1343 	ifp->if_ioctl = if_nullioctl;
1344 	mutex_exit(ifp->if_ioctl_lock);
1345 
1346 	/*
1347 	 * We cannot call ifc_destroy with holding ifp.
1348 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1349 	 */
1350 	if_put(ifp, &psref);
1351 
1352 	return (*ifc->ifc_destroy)(ifp);
1353 }
1354 
1355 /*
1356  * Look up a network interface cloner.
1357  */
1358 static struct if_clone *
1359 if_clone_lookup(const char *name, int *unitp)
1360 {
1361 	struct if_clone *ifc;
1362 	const char *cp;
1363 	char *dp, ifname[IFNAMSIZ + 3];
1364 	int unit;
1365 
1366 	strcpy(ifname, "if_");
1367 	/* separate interface name from unit */
1368 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1369 	    *cp && (*cp < '0' || *cp > '9');)
1370 		*dp++ = *cp++;
1371 
1372 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1373 		return NULL;	/* No name or unit number */
1374 	*dp++ = '\0';
1375 
1376 again:
1377 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1378 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1379 			break;
1380 	}
1381 
1382 	if (ifc == NULL) {
1383 		if (*ifname == '\0' ||
1384 		    module_autoload(ifname, MODULE_CLASS_DRIVER))
1385 			return NULL;
1386 		*ifname = '\0';
1387 		goto again;
1388 	}
1389 
1390 	unit = 0;
1391 	while (cp - name < IFNAMSIZ && *cp) {
1392 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1393 			/* Bogus unit number. */
1394 			return NULL;
1395 		}
1396 		unit = (unit * 10) + (*cp++ - '0');
1397 	}
1398 
1399 	if (unitp != NULL)
1400 		*unitp = unit;
1401 	return ifc;
1402 }
1403 
1404 /*
1405  * Register a network interface cloner.
1406  */
1407 void
1408 if_clone_attach(struct if_clone *ifc)
1409 {
1410 
1411 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1412 	if_cloners_count++;
1413 }
1414 
1415 /*
1416  * Unregister a network interface cloner.
1417  */
1418 void
1419 if_clone_detach(struct if_clone *ifc)
1420 {
1421 
1422 	LIST_REMOVE(ifc, ifc_list);
1423 	if_cloners_count--;
1424 }
1425 
1426 /*
1427  * Provide list of interface cloners to userspace.
1428  */
1429 int
1430 if_clone_list(int buf_count, char *buffer, int *total)
1431 {
1432 	char outbuf[IFNAMSIZ], *dst;
1433 	struct if_clone *ifc;
1434 	int count, error = 0;
1435 
1436 	*total = if_cloners_count;
1437 	if ((dst = buffer) == NULL) {
1438 		/* Just asking how many there are. */
1439 		return 0;
1440 	}
1441 
1442 	if (buf_count < 0)
1443 		return EINVAL;
1444 
1445 	count = (if_cloners_count < buf_count) ?
1446 	    if_cloners_count : buf_count;
1447 
1448 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1449 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1450 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1451 		if (outbuf[sizeof(outbuf) - 1] != '\0')
1452 			return ENAMETOOLONG;
1453 		error = copyout(outbuf, dst, sizeof(outbuf));
1454 		if (error != 0)
1455 			break;
1456 	}
1457 
1458 	return error;
1459 }
1460 
1461 void
1462 ifaref(struct ifaddr *ifa)
1463 {
1464 	ifa->ifa_refcnt++;
1465 }
1466 
1467 void
1468 ifafree(struct ifaddr *ifa)
1469 {
1470 	KASSERT(ifa != NULL);
1471 	KASSERT(ifa->ifa_refcnt > 0);
1472 
1473 	if (--ifa->ifa_refcnt == 0) {
1474 		free(ifa, M_IFADDR);
1475 	}
1476 }
1477 
1478 void
1479 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1480 {
1481 	ifa->ifa_ifp = ifp;
1482 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1483 	ifaref(ifa);
1484 }
1485 
1486 void
1487 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1488 {
1489 	KASSERT(ifa->ifa_ifp == ifp);
1490 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1491 	ifafree(ifa);
1492 }
1493 
1494 static inline int
1495 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1496 {
1497 	return sockaddr_cmp(sa1, sa2) == 0;
1498 }
1499 
1500 /*
1501  * Locate an interface based on a complete address.
1502  */
1503 /*ARGSUSED*/
1504 struct ifaddr *
1505 ifa_ifwithaddr(const struct sockaddr *addr)
1506 {
1507 	struct ifnet *ifp;
1508 	struct ifaddr *ifa;
1509 	int s;
1510 
1511 	s = pserialize_read_enter();
1512 	IFNET_READER_FOREACH(ifp) {
1513 		if (ifp->if_output == if_nulloutput)
1514 			continue;
1515 		IFADDR_FOREACH(ifa, ifp) {
1516 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1517 				continue;
1518 			if (equal(addr, ifa->ifa_addr))
1519 				return ifa;
1520 			if ((ifp->if_flags & IFF_BROADCAST) &&
1521 			    ifa->ifa_broadaddr &&
1522 			    /* IP6 doesn't have broadcast */
1523 			    ifa->ifa_broadaddr->sa_len != 0 &&
1524 			    equal(ifa->ifa_broadaddr, addr))
1525 				return ifa;
1526 		}
1527 	}
1528 	pserialize_read_exit(s);
1529 	return NULL;
1530 }
1531 
1532 /*
1533  * Locate the point to point interface with a given destination address.
1534  */
1535 /*ARGSUSED*/
1536 struct ifaddr *
1537 ifa_ifwithdstaddr(const struct sockaddr *addr)
1538 {
1539 	struct ifnet *ifp;
1540 	struct ifaddr *ifa;
1541 	int s;
1542 
1543 	s = pserialize_read_enter();
1544 	IFNET_READER_FOREACH(ifp) {
1545 		if (ifp->if_output == if_nulloutput)
1546 			continue;
1547 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1548 			continue;
1549 		IFADDR_FOREACH(ifa, ifp) {
1550 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1551 			    ifa->ifa_dstaddr == NULL)
1552 				continue;
1553 			if (equal(addr, ifa->ifa_dstaddr))
1554 				return ifa;
1555 		}
1556 	}
1557 	pserialize_read_exit(s);
1558 	return NULL;
1559 }
1560 
1561 /*
1562  * Find an interface on a specific network.  If many, choice
1563  * is most specific found.
1564  */
1565 struct ifaddr *
1566 ifa_ifwithnet(const struct sockaddr *addr)
1567 {
1568 	struct ifnet *ifp;
1569 	struct ifaddr *ifa;
1570 	const struct sockaddr_dl *sdl;
1571 	struct ifaddr *ifa_maybe = 0;
1572 	u_int af = addr->sa_family;
1573 	const char *addr_data = addr->sa_data, *cplim;
1574 	int s;
1575 
1576 	if (af == AF_LINK) {
1577 		sdl = satocsdl(addr);
1578 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1579 		    ifindex2ifnet[sdl->sdl_index] &&
1580 		    ifindex2ifnet[sdl->sdl_index]->if_output != if_nulloutput) {
1581 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1582 		}
1583 	}
1584 #ifdef NETATALK
1585 	if (af == AF_APPLETALK) {
1586 		const struct sockaddr_at *sat, *sat2;
1587 		sat = (const struct sockaddr_at *)addr;
1588 		s = pserialize_read_enter();
1589 		IFNET_READER_FOREACH(ifp) {
1590 			if (ifp->if_output == if_nulloutput)
1591 				continue;
1592 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1593 			if (ifa == NULL)
1594 				continue;
1595 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1596 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1597 				return ifa; /* exact match */
1598 			if (ifa_maybe == NULL) {
1599 				/* else keep the if with the right range */
1600 				ifa_maybe = ifa;
1601 			}
1602 		}
1603 		pserialize_read_exit(s);
1604 		return ifa_maybe;
1605 	}
1606 #endif
1607 	s = pserialize_read_enter();
1608 	IFNET_READER_FOREACH(ifp) {
1609 		if (ifp->if_output == if_nulloutput)
1610 			continue;
1611 		IFADDR_FOREACH(ifa, ifp) {
1612 			const char *cp, *cp2, *cp3;
1613 
1614 			if (ifa->ifa_addr->sa_family != af ||
1615 			    ifa->ifa_netmask == NULL)
1616  next:				continue;
1617 			cp = addr_data;
1618 			cp2 = ifa->ifa_addr->sa_data;
1619 			cp3 = ifa->ifa_netmask->sa_data;
1620 			cplim = (const char *)ifa->ifa_netmask +
1621 			    ifa->ifa_netmask->sa_len;
1622 			while (cp3 < cplim) {
1623 				if ((*cp++ ^ *cp2++) & *cp3++) {
1624 					/* want to continue for() loop */
1625 					goto next;
1626 				}
1627 			}
1628 			if (ifa_maybe == NULL ||
1629 			    rt_refines(ifa->ifa_netmask,
1630 			               ifa_maybe->ifa_netmask))
1631 				ifa_maybe = ifa;
1632 		}
1633 	}
1634 	pserialize_read_exit(s);
1635 	return ifa_maybe;
1636 }
1637 
1638 /*
1639  * Find the interface of the addresss.
1640  */
1641 struct ifaddr *
1642 ifa_ifwithladdr(const struct sockaddr *addr)
1643 {
1644 	struct ifaddr *ia;
1645 
1646 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
1647 	    (ia = ifa_ifwithnet(addr)))
1648 		return ia;
1649 	return NULL;
1650 }
1651 
1652 /*
1653  * Find an interface using a specific address family
1654  */
1655 struct ifaddr *
1656 ifa_ifwithaf(int af)
1657 {
1658 	struct ifnet *ifp;
1659 	struct ifaddr *ifa = NULL;
1660 	int s;
1661 
1662 	s = pserialize_read_enter();
1663 	IFNET_READER_FOREACH(ifp) {
1664 		if (ifp->if_output == if_nulloutput)
1665 			continue;
1666 		IFADDR_FOREACH(ifa, ifp) {
1667 			if (ifa->ifa_addr->sa_family == af)
1668 				goto out;
1669 		}
1670 	}
1671 out:
1672 	pserialize_read_exit(s);
1673 	return ifa;
1674 }
1675 
1676 /*
1677  * Find an interface address specific to an interface best matching
1678  * a given address.
1679  */
1680 struct ifaddr *
1681 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
1682 {
1683 	struct ifaddr *ifa;
1684 	const char *cp, *cp2, *cp3;
1685 	const char *cplim;
1686 	struct ifaddr *ifa_maybe = 0;
1687 	u_int af = addr->sa_family;
1688 
1689 	if (ifp->if_output == if_nulloutput)
1690 		return NULL;
1691 
1692 	if (af >= AF_MAX)
1693 		return NULL;
1694 
1695 	IFADDR_FOREACH(ifa, ifp) {
1696 		if (ifa->ifa_addr->sa_family != af)
1697 			continue;
1698 		ifa_maybe = ifa;
1699 		if (ifa->ifa_netmask == NULL) {
1700 			if (equal(addr, ifa->ifa_addr) ||
1701 			    (ifa->ifa_dstaddr &&
1702 			     equal(addr, ifa->ifa_dstaddr)))
1703 				return ifa;
1704 			continue;
1705 		}
1706 		cp = addr->sa_data;
1707 		cp2 = ifa->ifa_addr->sa_data;
1708 		cp3 = ifa->ifa_netmask->sa_data;
1709 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1710 		for (; cp3 < cplim; cp3++) {
1711 			if ((*cp++ ^ *cp2++) & *cp3)
1712 				break;
1713 		}
1714 		if (cp3 == cplim)
1715 			return ifa;
1716 	}
1717 	return ifa_maybe;
1718 }
1719 
1720 /*
1721  * Default action when installing a route with a Link Level gateway.
1722  * Lookup an appropriate real ifa to point to.
1723  * This should be moved to /sys/net/link.c eventually.
1724  */
1725 void
1726 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
1727 {
1728 	struct ifaddr *ifa;
1729 	const struct sockaddr *dst;
1730 	struct ifnet *ifp;
1731 
1732 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1733 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
1734 		return;
1735 	if ((ifa = ifaof_ifpforaddr(dst, ifp)) != NULL) {
1736 		rt_replace_ifa(rt, ifa);
1737 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1738 			ifa->ifa_rtrequest(cmd, rt, info);
1739 	}
1740 }
1741 
1742 /*
1743  * bitmask macros to manage a densely packed link_state change queue.
1744  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
1745  * LINK_STATE_UP(2) we need 2 bits for each state change.
1746  * As a state change to store is 0, treat all bits set as an unset item.
1747  */
1748 #define LQ_ITEM_BITS		2
1749 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
1750 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
1751 #define LINK_STATE_UNSET	LQ_ITEM_MASK
1752 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
1753 #define LQ_STORE(q, i, v)						      \
1754 	do {								      \
1755 		(q) &= ~LQ_MASK((i));					      \
1756 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
1757 	} while (0 /* CONSTCOND */)
1758 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
1759 #define LQ_POP(q, v)							      \
1760 	do {								      \
1761 		(v) = LQ_ITEM((q), 0);					      \
1762 		(q) >>= LQ_ITEM_BITS;					      \
1763 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
1764 	} while (0 /* CONSTCOND */)
1765 #define LQ_PUSH(q, v)							      \
1766 	do {								      \
1767 		(q) >>= LQ_ITEM_BITS;					      \
1768 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
1769 	} while (0 /* CONSTCOND */)
1770 #define LQ_FIND_UNSET(q, i)						      \
1771 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
1772 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
1773 			break;						      \
1774 	}
1775 /*
1776  * Handle a change in the interface link state and
1777  * queue notifications.
1778  */
1779 void
1780 if_link_state_change(struct ifnet *ifp, int link_state)
1781 {
1782 	int s, idx;
1783 
1784 	/* Ensure change is to a valid state */
1785 	switch (link_state) {
1786 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
1787 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
1788 	case LINK_STATE_UP:
1789 		break;
1790 	default:
1791 #ifdef DEBUG
1792 		printf("%s: invalid link state %d\n",
1793 		    ifp->if_xname, link_state);
1794 #endif
1795 		return;
1796 	}
1797 
1798 	s = splnet();
1799 
1800 	/* Find the last unset event in the queue. */
1801 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
1802 
1803 	/*
1804 	 * Ensure link_state doesn't match the last event in the queue.
1805 	 * ifp->if_link_state is not checked and set here because
1806 	 * that would present an inconsistent picture to the system.
1807 	 */
1808 	if (idx != 0 &&
1809 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
1810 		goto out;
1811 
1812 	/* Handle queue overflow. */
1813 	if (idx == LQ_MAX(ifp->if_link_queue)) {
1814 		uint8_t lost;
1815 
1816 		/*
1817 		 * The DOWN state must be protected from being pushed off
1818 		 * the queue to ensure that userland will always be
1819 		 * in a sane state.
1820 		 * Because DOWN is protected, there is no need to protect
1821 		 * UNKNOWN.
1822 		 * It should be invalid to change from any other state to
1823 		 * UNKNOWN anyway ...
1824 		 */
1825 		lost = LQ_ITEM(ifp->if_link_queue, 0);
1826 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
1827 		if (lost == LINK_STATE_DOWN) {
1828 			lost = LQ_ITEM(ifp->if_link_queue, 0);
1829 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
1830 		}
1831 		printf("%s: lost link state change %s\n",
1832 		    ifp->if_xname,
1833 		    lost == LINK_STATE_UP ? "UP" :
1834 		    lost == LINK_STATE_DOWN ? "DOWN" :
1835 		    "UNKNOWN");
1836 	} else
1837 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
1838 
1839 	softint_schedule(ifp->if_link_si);
1840 
1841 out:
1842 	splx(s);
1843 }
1844 
1845 /*
1846  * Handle interface link state change notifications.
1847  * Must be called at splnet().
1848  */
1849 static void
1850 if_link_state_change0(struct ifnet *ifp, int link_state)
1851 {
1852 	struct domain *dp;
1853 
1854 	/* Ensure the change is still valid. */
1855 	if (ifp->if_link_state == link_state)
1856 		return;
1857 
1858 #ifdef DEBUG
1859 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
1860 		link_state == LINK_STATE_UP ? "UP" :
1861 		link_state == LINK_STATE_DOWN ? "DOWN" :
1862 		"UNKNOWN",
1863 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
1864 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
1865 		"UNKNOWN");
1866 #endif
1867 
1868 	/*
1869 	 * When going from UNKNOWN to UP, we need to mark existing
1870 	 * addresses as tentative and restart DAD as we may have
1871 	 * erroneously not found a duplicate.
1872 	 *
1873 	 * This needs to happen before rt_ifmsg to avoid a race where
1874 	 * listeners would have an address and expect it to work right
1875 	 * away.
1876 	 */
1877 	if (link_state == LINK_STATE_UP &&
1878 	    ifp->if_link_state == LINK_STATE_UNKNOWN)
1879 	{
1880 		DOMAIN_FOREACH(dp) {
1881 			if (dp->dom_if_link_state_change != NULL)
1882 				dp->dom_if_link_state_change(ifp,
1883 				    LINK_STATE_DOWN);
1884 		}
1885 	}
1886 
1887 	ifp->if_link_state = link_state;
1888 
1889 	/* Notify that the link state has changed. */
1890 	rt_ifmsg(ifp);
1891 
1892 #if NCARP > 0
1893 	if (ifp->if_carp)
1894 		carp_carpdev_state(ifp);
1895 #endif
1896 
1897 	DOMAIN_FOREACH(dp) {
1898 		if (dp->dom_if_link_state_change != NULL)
1899 			dp->dom_if_link_state_change(ifp, link_state);
1900 	}
1901 }
1902 
1903 /*
1904  * Process the interface link state change queue.
1905  */
1906 static void
1907 if_link_state_change_si(void *arg)
1908 {
1909 	struct ifnet *ifp = arg;
1910 	int s;
1911 	uint8_t state;
1912 
1913 	s = splnet();
1914 
1915 	/* Pop a link state change from the queue and process it. */
1916 	LQ_POP(ifp->if_link_queue, state);
1917 	if_link_state_change0(ifp, state);
1918 
1919 	/* If there is a link state change to come, schedule it. */
1920 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET)
1921 		softint_schedule(ifp->if_link_si);
1922 
1923 	splx(s);
1924 }
1925 
1926 /*
1927  * Default action when installing a local route on a point-to-point
1928  * interface.
1929  */
1930 void
1931 p2p_rtrequest(int req, struct rtentry *rt,
1932     __unused const struct rt_addrinfo *info)
1933 {
1934 	struct ifnet *ifp = rt->rt_ifp;
1935 	struct ifaddr *ifa, *lo0ifa;
1936 
1937 	switch (req) {
1938 	case RTM_ADD:
1939 		if ((rt->rt_flags & RTF_LOCAL) == 0)
1940 			break;
1941 
1942 		IFADDR_FOREACH(ifa, ifp) {
1943 			if (equal(rt_getkey(rt), ifa->ifa_addr))
1944 				break;
1945 		}
1946 		if (ifa == NULL)
1947 			break;
1948 
1949 		/*
1950 		 * Ensure lo0 has an address of the same family.
1951 		 */
1952 		IFADDR_FOREACH(lo0ifa, lo0ifp) {
1953 			if (lo0ifa->ifa_addr->sa_family ==
1954 			    ifa->ifa_addr->sa_family)
1955 				break;
1956 		}
1957 		if (lo0ifa == NULL)
1958 			break;
1959 
1960 		rt->rt_ifp = lo0ifp;
1961 
1962 		/*
1963 		 * Make sure to set rt->rt_ifa to the interface
1964 		 * address we are using, otherwise we will have trouble
1965 		 * with source address selection.
1966 		 */
1967 		if (ifa != rt->rt_ifa)
1968 			rt_replace_ifa(rt, ifa);
1969 		break;
1970 	case RTM_DELETE:
1971 	default:
1972 		break;
1973 	}
1974 }
1975 
1976 /*
1977  * Mark an interface down and notify protocols of
1978  * the transition.
1979  * NOTE: must be called at splsoftnet or equivalent.
1980  */
1981 void
1982 if_down(struct ifnet *ifp)
1983 {
1984 	struct ifaddr *ifa;
1985 	struct domain *dp;
1986 
1987 	ifp->if_flags &= ~IFF_UP;
1988 	nanotime(&ifp->if_lastchange);
1989 	IFADDR_FOREACH(ifa, ifp)
1990 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1991 	IFQ_PURGE(&ifp->if_snd);
1992 #if NCARP > 0
1993 	if (ifp->if_carp)
1994 		carp_carpdev_state(ifp);
1995 #endif
1996 	rt_ifmsg(ifp);
1997 	DOMAIN_FOREACH(dp) {
1998 		if (dp->dom_if_down)
1999 			dp->dom_if_down(ifp);
2000 	}
2001 }
2002 
2003 /*
2004  * Mark an interface up and notify protocols of
2005  * the transition.
2006  * NOTE: must be called at splsoftnet or equivalent.
2007  */
2008 void
2009 if_up(struct ifnet *ifp)
2010 {
2011 #ifdef notyet
2012 	struct ifaddr *ifa;
2013 #endif
2014 	struct domain *dp;
2015 
2016 	ifp->if_flags |= IFF_UP;
2017 	nanotime(&ifp->if_lastchange);
2018 #ifdef notyet
2019 	/* this has no effect on IP, and will kill all ISO connections XXX */
2020 	IFADDR_FOREACH(ifa, ifp)
2021 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2022 #endif
2023 #if NCARP > 0
2024 	if (ifp->if_carp)
2025 		carp_carpdev_state(ifp);
2026 #endif
2027 	rt_ifmsg(ifp);
2028 	DOMAIN_FOREACH(dp) {
2029 		if (dp->dom_if_up)
2030 			dp->dom_if_up(ifp);
2031 	}
2032 }
2033 
2034 /*
2035  * Handle interface slowtimo timer routine.  Called
2036  * from softclock, we decrement timer (if set) and
2037  * call the appropriate interface routine on expiration.
2038  */
2039 static void
2040 if_slowtimo(void *arg)
2041 {
2042 	void (*slowtimo)(struct ifnet *);
2043 	struct ifnet *ifp = arg;
2044 	int s;
2045 
2046 	slowtimo = ifp->if_slowtimo;
2047 	if (__predict_false(slowtimo == NULL))
2048 		return;
2049 
2050 	s = splnet();
2051 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2052 		(*slowtimo)(ifp);
2053 
2054 	splx(s);
2055 
2056 	if (__predict_true(ifp->if_slowtimo != NULL))
2057 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2058 }
2059 
2060 /*
2061  * Set/clear promiscuous mode on interface ifp based on the truth value
2062  * of pswitch.  The calls are reference counted so that only the first
2063  * "on" request actually has an effect, as does the final "off" request.
2064  * Results are undefined if the "off" and "on" requests are not matched.
2065  */
2066 int
2067 ifpromisc(struct ifnet *ifp, int pswitch)
2068 {
2069 	int pcount, ret;
2070 	short nflags;
2071 
2072 	pcount = ifp->if_pcount;
2073 	if (pswitch) {
2074 		/*
2075 		 * Allow the device to be "placed" into promiscuous
2076 		 * mode even if it is not configured up.  It will
2077 		 * consult IFF_PROMISC when it is brought up.
2078 		 */
2079 		if (ifp->if_pcount++ != 0)
2080 			return 0;
2081 		nflags = ifp->if_flags | IFF_PROMISC;
2082 	} else {
2083 		if (--ifp->if_pcount > 0)
2084 			return 0;
2085 		nflags = ifp->if_flags & ~IFF_PROMISC;
2086 	}
2087 	ret = if_flags_set(ifp, nflags);
2088 	/* Restore interface state if not successful. */
2089 	if (ret != 0) {
2090 		ifp->if_pcount = pcount;
2091 	}
2092 	return ret;
2093 }
2094 
2095 /*
2096  * Map interface name to
2097  * interface structure pointer.
2098  */
2099 struct ifnet *
2100 ifunit(const char *name)
2101 {
2102 	struct ifnet *ifp;
2103 	const char *cp = name;
2104 	u_int unit = 0;
2105 	u_int i;
2106 	int s;
2107 
2108 	/*
2109 	 * If the entire name is a number, treat it as an ifindex.
2110 	 */
2111 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2112 		unit = unit * 10 + (*cp - '0');
2113 	}
2114 
2115 	/*
2116 	 * If the number took all of the name, then it's a valid ifindex.
2117 	 */
2118 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2119 		if (unit >= if_indexlim)
2120 			return NULL;
2121 		ifp = ifindex2ifnet[unit];
2122 		if (ifp == NULL || ifp->if_output == if_nulloutput)
2123 			return NULL;
2124 		return ifp;
2125 	}
2126 
2127 	ifp = NULL;
2128 	s = pserialize_read_enter();
2129 	IFNET_READER_FOREACH(ifp) {
2130 		if (ifp->if_output == if_nulloutput)
2131 			continue;
2132 	 	if (strcmp(ifp->if_xname, name) == 0)
2133 			goto out;
2134 	}
2135 out:
2136 	pserialize_read_exit(s);
2137 	return ifp;
2138 }
2139 
2140 /*
2141  * Get a reference of an ifnet object by an interface name.
2142  * The returned reference is protected by psref(9). The caller
2143  * must release a returned reference by if_put after use.
2144  */
2145 struct ifnet *
2146 if_get(const char *name, struct psref *psref)
2147 {
2148 	struct ifnet *ifp;
2149 	const char *cp = name;
2150 	u_int unit = 0;
2151 	u_int i;
2152 	int s;
2153 
2154 	/*
2155 	 * If the entire name is a number, treat it as an ifindex.
2156 	 */
2157 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2158 		unit = unit * 10 + (*cp - '0');
2159 	}
2160 
2161 	/*
2162 	 * If the number took all of the name, then it's a valid ifindex.
2163 	 */
2164 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2165 		if (unit >= if_indexlim)
2166 			return NULL;
2167 		ifp = ifindex2ifnet[unit];
2168 		if (ifp == NULL || ifp->if_output == if_nulloutput)
2169 			return NULL;
2170 		return ifp;
2171 	}
2172 
2173 	ifp = NULL;
2174 	s = pserialize_read_enter();
2175 	IFNET_READER_FOREACH(ifp) {
2176 		if (ifp->if_output == if_nulloutput)
2177 			continue;
2178 		if (strcmp(ifp->if_xname, name) == 0) {
2179 			psref_acquire(psref, &ifp->if_psref,
2180 			    ifnet_psref_class);
2181 			goto out;
2182 		}
2183 	}
2184 out:
2185 	pserialize_read_exit(s);
2186 	return ifp;
2187 }
2188 
2189 /*
2190  * Release a reference of an ifnet object given by if_get or
2191  * if_get_byindex.
2192  */
2193 void
2194 if_put(const struct ifnet *ifp, struct psref *psref)
2195 {
2196 
2197 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2198 }
2199 
2200 ifnet_t *
2201 if_byindex(u_int idx)
2202 {
2203 	return (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL;
2204 }
2205 
2206 /*
2207  * Get a reference of an ifnet object by an interface index.
2208  * The returned reference is protected by psref(9). The caller
2209  * must release a returned reference by if_put after use.
2210  */
2211 ifnet_t *
2212 if_get_byindex(u_int idx, struct psref *psref)
2213 {
2214 	ifnet_t *ifp;
2215 	int s;
2216 
2217 	s = pserialize_read_enter();
2218 	ifp = (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2219 	if (__predict_true(ifp != NULL))
2220 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2221 	pserialize_read_exit(s);
2222 
2223 	return ifp;
2224 }
2225 
2226 /* common */
2227 int
2228 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2229 {
2230 	int s;
2231 	struct ifreq *ifr;
2232 	struct ifcapreq *ifcr;
2233 	struct ifdatareq *ifdr;
2234 
2235 	switch (cmd) {
2236 	case SIOCSIFCAP:
2237 		ifcr = data;
2238 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2239 			return EINVAL;
2240 
2241 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2242 			return 0;
2243 
2244 		ifp->if_capenable = ifcr->ifcr_capenable;
2245 
2246 		/* Pre-compute the checksum flags mask. */
2247 		ifp->if_csum_flags_tx = 0;
2248 		ifp->if_csum_flags_rx = 0;
2249 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2250 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2251 		}
2252 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2253 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2254 		}
2255 
2256 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2257 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2258 		}
2259 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2260 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2261 		}
2262 
2263 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2264 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2265 		}
2266 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2267 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2268 		}
2269 
2270 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2271 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2272 		}
2273 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2274 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2275 		}
2276 
2277 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2278 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2279 		}
2280 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2281 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2282 		}
2283 		if (ifp->if_flags & IFF_UP)
2284 			return ENETRESET;
2285 		return 0;
2286 	case SIOCSIFFLAGS:
2287 		ifr = data;
2288 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2289 			s = splnet();
2290 			if_down(ifp);
2291 			splx(s);
2292 		}
2293 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2294 			s = splnet();
2295 			if_up(ifp);
2296 			splx(s);
2297 		}
2298 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2299 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2300 		break;
2301 	case SIOCGIFFLAGS:
2302 		ifr = data;
2303 		ifr->ifr_flags = ifp->if_flags;
2304 		break;
2305 
2306 	case SIOCGIFMETRIC:
2307 		ifr = data;
2308 		ifr->ifr_metric = ifp->if_metric;
2309 		break;
2310 
2311 	case SIOCGIFMTU:
2312 		ifr = data;
2313 		ifr->ifr_mtu = ifp->if_mtu;
2314 		break;
2315 
2316 	case SIOCGIFDLT:
2317 		ifr = data;
2318 		ifr->ifr_dlt = ifp->if_dlt;
2319 		break;
2320 
2321 	case SIOCGIFCAP:
2322 		ifcr = data;
2323 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2324 		ifcr->ifcr_capenable = ifp->if_capenable;
2325 		break;
2326 
2327 	case SIOCSIFMETRIC:
2328 		ifr = data;
2329 		ifp->if_metric = ifr->ifr_metric;
2330 		break;
2331 
2332 	case SIOCGIFDATA:
2333 		ifdr = data;
2334 		ifdr->ifdr_data = ifp->if_data;
2335 		break;
2336 
2337 	case SIOCGIFINDEX:
2338 		ifr = data;
2339 		ifr->ifr_index = ifp->if_index;
2340 		break;
2341 
2342 	case SIOCZIFDATA:
2343 		ifdr = data;
2344 		ifdr->ifdr_data = ifp->if_data;
2345 		/*
2346 		 * Assumes that the volatile counters that can be
2347 		 * zero'ed are at the end of if_data.
2348 		 */
2349 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
2350 		    offsetof(struct if_data, ifi_ipackets));
2351 		/*
2352 		 * The memset() clears to the bottm of if_data. In the area,
2353 		 * if_lastchange is included. Please be careful if new entry
2354 		 * will be added into if_data or rewite this.
2355 		 *
2356 		 * And also, update if_lastchnage.
2357 		 */
2358 		getnanotime(&ifp->if_lastchange);
2359 		break;
2360 	case SIOCSIFMTU:
2361 		ifr = data;
2362 		if (ifp->if_mtu == ifr->ifr_mtu)
2363 			break;
2364 		ifp->if_mtu = ifr->ifr_mtu;
2365 		/*
2366 		 * If the link MTU changed, do network layer specific procedure.
2367 		 */
2368 #ifdef INET6
2369 		if (in6_present)
2370 			nd6_setmtu(ifp);
2371 #endif
2372 		return ENETRESET;
2373 	default:
2374 		return ENOTTY;
2375 	}
2376 	return 0;
2377 }
2378 
2379 int
2380 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
2381 {
2382 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
2383 	struct ifaddr *ifa;
2384 	const struct sockaddr *any, *sa;
2385 	union {
2386 		struct sockaddr sa;
2387 		struct sockaddr_storage ss;
2388 	} u, v;
2389 
2390 	switch (cmd) {
2391 	case SIOCSIFADDRPREF:
2392 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
2393 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
2394 		    NULL) != 0)
2395 			return EPERM;
2396 	case SIOCGIFADDRPREF:
2397 		break;
2398 	default:
2399 		return EOPNOTSUPP;
2400 	}
2401 
2402 	/* sanity checks */
2403 	if (data == NULL || ifp == NULL) {
2404 		panic("invalid argument to %s", __func__);
2405 		/*NOTREACHED*/
2406 	}
2407 
2408 	/* address must be specified on ADD and DELETE */
2409 	sa = sstocsa(&ifap->ifap_addr);
2410 	if (sa->sa_family != sofamily(so))
2411 		return EINVAL;
2412 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
2413 		return EINVAL;
2414 
2415 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
2416 
2417 	IFADDR_FOREACH(ifa, ifp) {
2418 		if (ifa->ifa_addr->sa_family != sa->sa_family)
2419 			continue;
2420 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
2421 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
2422 			break;
2423 	}
2424 	if (ifa == NULL)
2425 		return EADDRNOTAVAIL;
2426 
2427 	switch (cmd) {
2428 	case SIOCSIFADDRPREF:
2429 		ifa->ifa_preference = ifap->ifap_preference;
2430 		return 0;
2431 	case SIOCGIFADDRPREF:
2432 		/* fill in the if_laddrreq structure */
2433 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
2434 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
2435 		ifap->ifap_preference = ifa->ifa_preference;
2436 		return 0;
2437 	default:
2438 		return EOPNOTSUPP;
2439 	}
2440 }
2441 
2442 /*
2443  * Interface ioctls.
2444  */
2445 static int
2446 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
2447 {
2448 	struct ifnet *ifp;
2449 	struct ifreq *ifr;
2450 	int error = 0;
2451 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
2452 	u_long ocmd = cmd;
2453 #endif
2454 	short oif_flags;
2455 #ifdef COMPAT_OIFREQ
2456 	struct ifreq ifrb;
2457 	struct oifreq *oifr = NULL;
2458 #endif
2459 	int r;
2460 	struct psref psref;
2461 	int bound = curlwp->l_pflag & LP_BOUND;
2462 
2463 	switch (cmd) {
2464 #ifdef COMPAT_OIFREQ
2465 	case OSIOCGIFCONF:
2466 	case OOSIOCGIFCONF:
2467 		return compat_ifconf(cmd, data);
2468 #endif
2469 #ifdef COMPAT_OIFDATA
2470 	case OSIOCGIFDATA:
2471 	case OSIOCZIFDATA:
2472 		return compat_ifdatareq(l, cmd, data);
2473 #endif
2474 	case SIOCGIFCONF:
2475 		return ifconf(cmd, data);
2476 	case SIOCINITIFADDR:
2477 		return EPERM;
2478 	}
2479 
2480 #ifdef COMPAT_OIFREQ
2481 	cmd = compat_cvtcmd(cmd);
2482 	if (cmd != ocmd) {
2483 		oifr = data;
2484 		data = ifr = &ifrb;
2485 		ifreqo2n(oifr, ifr);
2486 	} else
2487 #endif
2488 		ifr = data;
2489 
2490 	switch (cmd) {
2491 	case SIOCIFCREATE:
2492 	case SIOCIFDESTROY:
2493 		curlwp->l_pflag |= LP_BOUND;
2494 		if (l != NULL) {
2495 			ifp = if_get(ifr->ifr_name, &psref);
2496 			error = kauth_authorize_network(l->l_cred,
2497 			    KAUTH_NETWORK_INTERFACE,
2498 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2499 			    (void *)cmd, NULL);
2500 			if (ifp != NULL)
2501 				if_put(ifp, &psref);
2502 			if (error != 0) {
2503 				curlwp->l_pflag ^= bound ^ LP_BOUND;
2504 				return error;
2505 			}
2506 		}
2507 		mutex_enter(&if_clone_mtx);
2508 		r = (cmd == SIOCIFCREATE) ?
2509 			if_clone_create(ifr->ifr_name) :
2510 			if_clone_destroy(ifr->ifr_name);
2511 		mutex_exit(&if_clone_mtx);
2512 		curlwp->l_pflag ^= bound ^ LP_BOUND;
2513 		return r;
2514 
2515 	case SIOCIFGCLONERS:
2516 		{
2517 			struct if_clonereq *req = (struct if_clonereq *)data;
2518 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
2519 			    &req->ifcr_total);
2520 		}
2521 	}
2522 
2523 	curlwp->l_pflag |= LP_BOUND;
2524 	ifp = if_get(ifr->ifr_name, &psref);
2525 	if (ifp == NULL) {
2526 		curlwp->l_pflag ^= bound ^ LP_BOUND;
2527 		return ENXIO;
2528 	}
2529 
2530 	switch (cmd) {
2531 	case SIOCALIFADDR:
2532 	case SIOCDLIFADDR:
2533 	case SIOCSIFADDRPREF:
2534 	case SIOCSIFFLAGS:
2535 	case SIOCSIFCAP:
2536 	case SIOCSIFMETRIC:
2537 	case SIOCZIFDATA:
2538 	case SIOCSIFMTU:
2539 	case SIOCSIFPHYADDR:
2540 	case SIOCDIFPHYADDR:
2541 #ifdef INET6
2542 	case SIOCSIFPHYADDR_IN6:
2543 #endif
2544 	case SIOCSLIFPHYADDR:
2545 	case SIOCADDMULTI:
2546 	case SIOCDELMULTI:
2547 	case SIOCSIFMEDIA:
2548 	case SIOCSDRVSPEC:
2549 	case SIOCG80211:
2550 	case SIOCS80211:
2551 	case SIOCS80211NWID:
2552 	case SIOCS80211NWKEY:
2553 	case SIOCS80211POWER:
2554 	case SIOCS80211BSSID:
2555 	case SIOCS80211CHANNEL:
2556 	case SIOCSLINKSTR:
2557 		if (l != NULL) {
2558 			error = kauth_authorize_network(l->l_cred,
2559 			    KAUTH_NETWORK_INTERFACE,
2560 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2561 			    (void *)cmd, NULL);
2562 			if (error != 0)
2563 				goto out;
2564 		}
2565 	}
2566 
2567 	oif_flags = ifp->if_flags;
2568 
2569 	mutex_enter(ifp->if_ioctl_lock);
2570 
2571 	error = (*ifp->if_ioctl)(ifp, cmd, data);
2572 	if (error != ENOTTY)
2573 		;
2574 	else if (so->so_proto == NULL)
2575 		error = EOPNOTSUPP;
2576 	else {
2577 #ifdef COMPAT_OSOCK
2578 		error = compat_ifioctl(so, ocmd, cmd, data, l);
2579 #else
2580 		error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
2581 		    cmd, data, ifp);
2582 #endif
2583 	}
2584 
2585 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
2586 		if ((ifp->if_flags & IFF_UP) != 0) {
2587 			int s = splnet();
2588 			if_up(ifp);
2589 			splx(s);
2590 		}
2591 	}
2592 #ifdef COMPAT_OIFREQ
2593 	if (cmd != ocmd)
2594 		ifreqn2o(oifr, ifr);
2595 #endif
2596 
2597 	mutex_exit(ifp->if_ioctl_lock);
2598 out:
2599 	if_put(ifp, &psref);
2600 	curlwp->l_pflag ^= bound ^ LP_BOUND;
2601 	return error;
2602 }
2603 
2604 /*
2605  * Return interface configuration
2606  * of system.  List may be used
2607  * in later ioctl's (above) to get
2608  * other information.
2609  *
2610  * Each record is a struct ifreq.  Before the addition of
2611  * sockaddr_storage, the API rule was that sockaddr flavors that did
2612  * not fit would extend beyond the struct ifreq, with the next struct
2613  * ifreq starting sa_len beyond the struct sockaddr.  Because the
2614  * union in struct ifreq includes struct sockaddr_storage, every kind
2615  * of sockaddr must fit.  Thus, there are no longer any overlength
2616  * records.
2617  *
2618  * Records are added to the user buffer if they fit, and ifc_len is
2619  * adjusted to the length that was written.  Thus, the user is only
2620  * assured of getting the complete list if ifc_len on return is at
2621  * least sizeof(struct ifreq) less than it was on entry.
2622  *
2623  * If the user buffer pointer is NULL, this routine copies no data and
2624  * returns the amount of space that would be needed.
2625  *
2626  * Invariants:
2627  * ifrp points to the next part of the user's buffer to be used.  If
2628  * ifrp != NULL, space holds the number of bytes remaining that we may
2629  * write at ifrp.  Otherwise, space holds the number of bytes that
2630  * would have been written had there been adequate space.
2631  */
2632 /*ARGSUSED*/
2633 static int
2634 ifconf(u_long cmd, void *data)
2635 {
2636 	struct ifconf *ifc = (struct ifconf *)data;
2637 	struct ifnet *ifp;
2638 	struct ifaddr *ifa;
2639 	struct ifreq ifr, *ifrp = NULL;
2640 	int space = 0, error = 0;
2641 	const int sz = (int)sizeof(struct ifreq);
2642 	const bool docopy = ifc->ifc_req != NULL;
2643 	int s;
2644 	int bound = curlwp->l_pflag & LP_BOUND;
2645 	struct psref psref;
2646 
2647 	if (docopy) {
2648 		space = ifc->ifc_len;
2649 		ifrp = ifc->ifc_req;
2650 	}
2651 
2652 	curlwp->l_pflag |= LP_BOUND;
2653 	s = pserialize_read_enter();
2654 	IFNET_READER_FOREACH(ifp) {
2655 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
2656 		pserialize_read_exit(s);
2657 
2658 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
2659 		    sizeof(ifr.ifr_name));
2660 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
2661 			error = ENAMETOOLONG;
2662 			goto release_exit;
2663 		}
2664 		if (IFADDR_EMPTY(ifp)) {
2665 			/* Interface with no addresses - send zero sockaddr. */
2666 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
2667 			if (!docopy) {
2668 				space += sz;
2669 				continue;
2670 			}
2671 			if (space >= sz) {
2672 				error = copyout(&ifr, ifrp, sz);
2673 				if (error != 0)
2674 					goto release_exit;
2675 				ifrp++;
2676 				space -= sz;
2677 			}
2678 		}
2679 
2680 		IFADDR_FOREACH(ifa, ifp) {
2681 			struct sockaddr *sa = ifa->ifa_addr;
2682 			/* all sockaddrs must fit in sockaddr_storage */
2683 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
2684 
2685 			if (!docopy) {
2686 				space += sz;
2687 				continue;
2688 			}
2689 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
2690 			if (space >= sz) {
2691 				error = copyout(&ifr, ifrp, sz);
2692 				if (error != 0)
2693 					goto release_exit;
2694 				ifrp++; space -= sz;
2695 			}
2696 		}
2697 
2698 		s = pserialize_read_enter();
2699 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
2700 	}
2701 	pserialize_read_exit(s);
2702 	curlwp->l_pflag ^= bound ^ LP_BOUND;
2703 
2704 	if (docopy) {
2705 		KASSERT(0 <= space && space <= ifc->ifc_len);
2706 		ifc->ifc_len -= space;
2707 	} else {
2708 		KASSERT(space >= 0);
2709 		ifc->ifc_len = space;
2710 	}
2711 	return (0);
2712 
2713 release_exit:
2714 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
2715 	curlwp->l_pflag ^= bound ^ LP_BOUND;
2716 	return error;
2717 }
2718 
2719 int
2720 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
2721 {
2722 	uint8_t len;
2723 #ifdef COMPAT_OIFREQ
2724 	struct ifreq ifrb;
2725 	struct oifreq *oifr = NULL;
2726 	u_long ocmd = cmd;
2727 	cmd = compat_cvtcmd(cmd);
2728 	if (cmd != ocmd) {
2729 		oifr = (struct oifreq *)(void *)ifr;
2730 		ifr = &ifrb;
2731 		ifreqo2n(oifr, ifr);
2732 		len = sizeof(oifr->ifr_addr);
2733 	} else
2734 #endif
2735 		len = sizeof(ifr->ifr_ifru.ifru_space);
2736 
2737 	if (len < sa->sa_len)
2738 		return EFBIG;
2739 
2740 	memset(&ifr->ifr_addr, 0, len);
2741 	sockaddr_copy(&ifr->ifr_addr, len, sa);
2742 
2743 #ifdef COMPAT_OIFREQ
2744 	if (cmd != ocmd)
2745 		ifreqn2o(oifr, ifr);
2746 #endif
2747 	return 0;
2748 }
2749 
2750 /*
2751  * wrapper function for the drivers which doesn't have if_transmit().
2752  */
2753 int
2754 if_transmit(struct ifnet *ifp, struct mbuf *m)
2755 {
2756 	int s, error;
2757 
2758 	s = splnet();
2759 
2760 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
2761 	if (error != 0) {
2762 		/* mbuf is already freed */
2763 		goto out;
2764 	}
2765 
2766 	ifp->if_obytes += m->m_pkthdr.len;;
2767 	if (m->m_flags & M_MCAST)
2768 		ifp->if_omcasts++;
2769 
2770 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
2771 		(*ifp->if_start)(ifp);
2772 out:
2773 	splx(s);
2774 
2775 	return error;
2776 }
2777 
2778 /*
2779  * Queue message on interface, and start output if interface
2780  * not yet active.
2781  */
2782 int
2783 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
2784 {
2785 
2786 	return (*ifp->if_transmit)(ifp, m);
2787 }
2788 
2789 /*
2790  * Queue message on interface, possibly using a second fast queue
2791  */
2792 int
2793 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
2794 {
2795 	int error = 0;
2796 
2797 	if (ifq != NULL
2798 #ifdef ALTQ
2799 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
2800 #endif
2801 	    ) {
2802 		if (IF_QFULL(ifq)) {
2803 			IF_DROP(&ifp->if_snd);
2804 			m_freem(m);
2805 			if (error == 0)
2806 				error = ENOBUFS;
2807 		} else
2808 			IF_ENQUEUE(ifq, m);
2809 	} else
2810 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
2811 	if (error != 0) {
2812 		++ifp->if_oerrors;
2813 		return error;
2814 	}
2815 	return 0;
2816 }
2817 
2818 int
2819 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
2820 {
2821 	int rc;
2822 
2823 	if (ifp->if_initaddr != NULL)
2824 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
2825 	else if (src ||
2826 		/* FIXME: may not hold if_ioctl_lock */
2827 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
2828 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
2829 
2830 	return rc;
2831 }
2832 
2833 int
2834 if_do_dad(struct ifnet *ifp)
2835 {
2836 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2837 		return 0;
2838 
2839 	switch (ifp->if_type) {
2840 	case IFT_FAITH:
2841 		/*
2842 		 * These interfaces do not have the IFF_LOOPBACK flag,
2843 		 * but loop packets back.  We do not have to do DAD on such
2844 		 * interfaces.  We should even omit it, because loop-backed
2845 		 * responses would confuse the DAD procedure.
2846 		 */
2847 		return 0;
2848 	default:
2849 		/*
2850 		 * Our DAD routine requires the interface up and running.
2851 		 * However, some interfaces can be up before the RUNNING
2852 		 * status.  Additionaly, users may try to assign addresses
2853 		 * before the interface becomes up (or running).
2854 		 * We simply skip DAD in such a case as a work around.
2855 		 * XXX: we should rather mark "tentative" on such addresses,
2856 		 * and do DAD after the interface becomes ready.
2857 		 */
2858 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
2859 		    (IFF_UP|IFF_RUNNING))
2860 			return 0;
2861 
2862 		return 1;
2863 	}
2864 }
2865 
2866 int
2867 if_flags_set(ifnet_t *ifp, const short flags)
2868 {
2869 	int rc;
2870 
2871 	if (ifp->if_setflags != NULL)
2872 		rc = (*ifp->if_setflags)(ifp, flags);
2873 	else {
2874 		short cantflags, chgdflags;
2875 		struct ifreq ifr;
2876 
2877 		chgdflags = ifp->if_flags ^ flags;
2878 		cantflags = chgdflags & IFF_CANTCHANGE;
2879 
2880 		if (cantflags != 0)
2881 			ifp->if_flags ^= cantflags;
2882 
2883                 /* Traditionally, we do not call if_ioctl after
2884                  * setting/clearing only IFF_PROMISC if the interface
2885                  * isn't IFF_UP.  Uphold that tradition.
2886 		 */
2887 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
2888 			return 0;
2889 
2890 		memset(&ifr, 0, sizeof(ifr));
2891 
2892 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
2893 		/* FIXME: may not hold if_ioctl_lock */
2894 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
2895 
2896 		if (rc != 0 && cantflags != 0)
2897 			ifp->if_flags ^= cantflags;
2898 	}
2899 
2900 	return rc;
2901 }
2902 
2903 int
2904 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
2905 {
2906 	int rc;
2907 	struct ifreq ifr;
2908 
2909 	if (ifp->if_mcastop != NULL)
2910 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
2911 	else {
2912 		ifreq_setaddr(cmd, &ifr, sa);
2913 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
2914 	}
2915 
2916 	return rc;
2917 }
2918 
2919 static void
2920 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
2921     struct ifaltq *ifq)
2922 {
2923 	const struct sysctlnode *cnode, *rnode;
2924 
2925 	if (sysctl_createv(clog, 0, NULL, &rnode,
2926 		       CTLFLAG_PERMANENT,
2927 		       CTLTYPE_NODE, "interfaces",
2928 		       SYSCTL_DESCR("Per-interface controls"),
2929 		       NULL, 0, NULL, 0,
2930 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
2931 		goto bad;
2932 
2933 	if (sysctl_createv(clog, 0, &rnode, &rnode,
2934 		       CTLFLAG_PERMANENT,
2935 		       CTLTYPE_NODE, ifname,
2936 		       SYSCTL_DESCR("Interface controls"),
2937 		       NULL, 0, NULL, 0,
2938 		       CTL_CREATE, CTL_EOL) != 0)
2939 		goto bad;
2940 
2941 	if (sysctl_createv(clog, 0, &rnode, &rnode,
2942 		       CTLFLAG_PERMANENT,
2943 		       CTLTYPE_NODE, "sndq",
2944 		       SYSCTL_DESCR("Interface output queue controls"),
2945 		       NULL, 0, NULL, 0,
2946 		       CTL_CREATE, CTL_EOL) != 0)
2947 		goto bad;
2948 
2949 	if (sysctl_createv(clog, 0, &rnode, &cnode,
2950 		       CTLFLAG_PERMANENT,
2951 		       CTLTYPE_INT, "len",
2952 		       SYSCTL_DESCR("Current output queue length"),
2953 		       NULL, 0, &ifq->ifq_len, 0,
2954 		       CTL_CREATE, CTL_EOL) != 0)
2955 		goto bad;
2956 
2957 	if (sysctl_createv(clog, 0, &rnode, &cnode,
2958 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2959 		       CTLTYPE_INT, "maxlen",
2960 		       SYSCTL_DESCR("Maximum allowed output queue length"),
2961 		       NULL, 0, &ifq->ifq_maxlen, 0,
2962 		       CTL_CREATE, CTL_EOL) != 0)
2963 		goto bad;
2964 
2965 	if (sysctl_createv(clog, 0, &rnode, &cnode,
2966 		       CTLFLAG_PERMANENT,
2967 		       CTLTYPE_INT, "drops",
2968 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
2969 		       NULL, 0, &ifq->ifq_drops, 0,
2970 		       CTL_CREATE, CTL_EOL) != 0)
2971 		goto bad;
2972 
2973 	return;
2974 bad:
2975 	printf("%s: could not attach sysctl nodes\n", ifname);
2976 	return;
2977 }
2978 
2979 #if defined(INET) || defined(INET6)
2980 
2981 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
2982 	static int							\
2983 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
2984 	{								\
2985 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
2986 	}
2987 
2988 #if defined(INET)
2989 static int
2990 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
2991 {
2992 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
2993 }
2994 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
2995 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
2996 #endif
2997 
2998 #if defined(INET6)
2999 static int
3000 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3001 {
3002 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3003 }
3004 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3005 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3006 #endif
3007 
3008 static void
3009 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3010 {
3011 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3012 	const char *pfname = NULL, *ipname = NULL;
3013 	int ipn = 0, qid = 0;
3014 
3015 	switch (pf) {
3016 #if defined(INET)
3017 	case PF_INET:
3018 		len_func = sysctl_net_ip_pktq_items;
3019 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3020 		drops_func = sysctl_net_ip_pktq_drops;
3021 		pfname = "inet", ipn = IPPROTO_IP;
3022 		ipname = "ip", qid = IPCTL_IFQ;
3023 		break;
3024 #endif
3025 #if defined(INET6)
3026 	case PF_INET6:
3027 		len_func = sysctl_net_ip6_pktq_items;
3028 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3029 		drops_func = sysctl_net_ip6_pktq_drops;
3030 		pfname = "inet6", ipn = IPPROTO_IPV6;
3031 		ipname = "ip6", qid = IPV6CTL_IFQ;
3032 		break;
3033 #endif
3034 	default:
3035 		KASSERT(false);
3036 	}
3037 
3038 	sysctl_createv(clog, 0, NULL, NULL,
3039 		       CTLFLAG_PERMANENT,
3040 		       CTLTYPE_NODE, pfname, NULL,
3041 		       NULL, 0, NULL, 0,
3042 		       CTL_NET, pf, CTL_EOL);
3043 	sysctl_createv(clog, 0, NULL, NULL,
3044 		       CTLFLAG_PERMANENT,
3045 		       CTLTYPE_NODE, ipname, NULL,
3046 		       NULL, 0, NULL, 0,
3047 		       CTL_NET, pf, ipn, CTL_EOL);
3048 	sysctl_createv(clog, 0, NULL, NULL,
3049 		       CTLFLAG_PERMANENT,
3050 		       CTLTYPE_NODE, "ifq",
3051 		       SYSCTL_DESCR("Protocol input queue controls"),
3052 		       NULL, 0, NULL, 0,
3053 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3054 
3055 	sysctl_createv(clog, 0, NULL, NULL,
3056 		       CTLFLAG_PERMANENT,
3057 		       CTLTYPE_INT, "len",
3058 		       SYSCTL_DESCR("Current input queue length"),
3059 		       len_func, 0, NULL, 0,
3060 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3061 	sysctl_createv(clog, 0, NULL, NULL,
3062 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3063 		       CTLTYPE_INT, "maxlen",
3064 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3065 		       maxlen_func, 0, NULL, 0,
3066 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3067 	sysctl_createv(clog, 0, NULL, NULL,
3068 		       CTLFLAG_PERMANENT,
3069 		       CTLTYPE_INT, "drops",
3070 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3071 		       drops_func, 0, NULL, 0,
3072 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3073 }
3074 #endif /* INET || INET6 */
3075 
3076 static int
3077 if_sdl_sysctl(SYSCTLFN_ARGS)
3078 {
3079 	struct ifnet *ifp;
3080 	const struct sockaddr_dl *sdl;
3081 
3082 	if (namelen != 1)
3083 		return EINVAL;
3084 
3085 	ifp = if_byindex(name[0]);
3086 	if (ifp == NULL)
3087 		return ENODEV;
3088 
3089 	sdl = ifp->if_sadl;
3090 	if (sdl == NULL) {
3091 		*oldlenp = 0;
3092 		return 0;
3093 	}
3094 
3095 	if (oldp == NULL) {
3096 		*oldlenp = sdl->sdl_alen;
3097 		return 0;
3098 	}
3099 
3100 	if (*oldlenp >= sdl->sdl_alen)
3101 		*oldlenp = sdl->sdl_alen;
3102 	return sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3103 }
3104 
3105 SYSCTL_SETUP(sysctl_net_sdl_setup, "sysctl net.sdl subtree setup")
3106 {
3107 	const struct sysctlnode *rnode = NULL;
3108 
3109 	sysctl_createv(clog, 0, NULL, &rnode,
3110 		       CTLFLAG_PERMANENT,
3111 		       CTLTYPE_NODE, "sdl",
3112 		       SYSCTL_DESCR("Get active link-layer address"),
3113 		       if_sdl_sysctl, 0, NULL, 0,
3114 		       CTL_NET, CTL_CREATE, CTL_EOL);
3115 }
3116