xref: /netbsd-src/sys/net/if.c (revision 7330f729ccf0bd976a06f95fad452fe774fc7fd1)
1 /*	$NetBSD: if.c,v 1.463 2019/10/06 15:11:17 uwe 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 1.463 2019/10/06 15:11:17 uwe Exp $");
94 
95 #if defined(_KERNEL_OPT)
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 #include "opt_atalk.h"
99 #include "opt_wlan.h"
100 #include "opt_net_mpsafe.h"
101 #include "opt_mrouting.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 #include <sys/module_hook.h>
123 #include <sys/compat_stub.h>
124 
125 #include <net/if.h>
126 #include <net/if_dl.h>
127 #include <net/if_ether.h>
128 #include <net/if_media.h>
129 #include <net80211/ieee80211.h>
130 #include <net80211/ieee80211_ioctl.h>
131 #include <net/if_types.h>
132 #include <net/route.h>
133 #include <net/netisr.h>
134 #include <sys/module.h>
135 #ifdef NETATALK
136 #include <netatalk/at_extern.h>
137 #include <netatalk/at.h>
138 #endif
139 #include <net/pfil.h>
140 #include <netinet/in.h>
141 #include <netinet/in_var.h>
142 #include <netinet/ip_encap.h>
143 #include <net/bpf.h>
144 
145 #ifdef INET6
146 #include <netinet6/in6_var.h>
147 #include <netinet6/nd6.h>
148 #endif
149 
150 #include "ether.h"
151 #include "fddi.h"
152 #include "token.h"
153 
154 #include "bridge.h"
155 #if NBRIDGE > 0
156 #include <net/if_bridgevar.h>
157 #endif
158 
159 #include "carp.h"
160 #if NCARP > 0
161 #include <netinet/ip_carp.h>
162 #endif
163 
164 #include <compat/sys/sockio.h>
165 
166 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
167 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
168 
169 /*
170  * Global list of interfaces.
171  */
172 /* DEPRECATED. Remove it once kvm(3) users disappeared */
173 struct ifnet_head		ifnet_list;
174 
175 struct pslist_head		ifnet_pslist;
176 static ifnet_t **		ifindex2ifnet = NULL;
177 static u_int			if_index = 1;
178 static size_t			if_indexlim = 0;
179 static uint64_t			index_gen;
180 /* Mutex to protect the above objects. */
181 kmutex_t			ifnet_mtx __cacheline_aligned;
182 static struct psref_class	*ifnet_psref_class __read_mostly;
183 static pserialize_t		ifnet_psz;
184 
185 static kmutex_t			if_clone_mtx;
186 
187 struct ifnet *lo0ifp;
188 int	ifqmaxlen = IFQ_MAXLEN;
189 
190 struct psref_class		*ifa_psref_class __read_mostly;
191 
192 static int	if_delroute_matcher(struct rtentry *, void *);
193 
194 static bool if_is_unit(const char *);
195 static struct if_clone *if_clone_lookup(const char *, int *);
196 
197 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
198 static int if_cloners_count;
199 
200 /* Packet filtering hook for interfaces. */
201 pfil_head_t *			if_pfil __read_mostly;
202 
203 static kauth_listener_t if_listener;
204 
205 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
206 static void if_detach_queues(struct ifnet *, struct ifqueue *);
207 static void sysctl_sndq_setup(struct sysctllog **, const char *,
208     struct ifaltq *);
209 static void if_slowtimo(void *);
210 static void if_attachdomain1(struct ifnet *);
211 static int ifconf(u_long, void *);
212 static int if_transmit(struct ifnet *, struct mbuf *);
213 static int if_clone_create(const char *);
214 static int if_clone_destroy(const char *);
215 static void if_link_state_change_si(void *);
216 static void if_up_locked(struct ifnet *);
217 static void _if_down(struct ifnet *);
218 static void if_down_deactivated(struct ifnet *);
219 
220 struct if_percpuq {
221 	struct ifnet	*ipq_ifp;
222 	void		*ipq_si;
223 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
224 };
225 
226 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
227 
228 static void if_percpuq_drops(void *, void *, struct cpu_info *);
229 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
230 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
231     struct if_percpuq *);
232 
233 struct if_deferred_start {
234 	struct ifnet	*ids_ifp;
235 	void		(*ids_if_start)(struct ifnet *);
236 	void		*ids_si;
237 };
238 
239 static void if_deferred_start_softint(void *);
240 static void if_deferred_start_common(struct ifnet *);
241 static void if_deferred_start_destroy(struct ifnet *);
242 
243 #if defined(INET) || defined(INET6)
244 static void sysctl_net_pktq_setup(struct sysctllog **, int);
245 #endif
246 
247 /*
248  * Hook for if_vlan - needed by if_agr
249  */
250 struct if_vlan_vlan_input_hook_t if_vlan_vlan_input_hook;
251 
252 static void if_sysctl_setup(struct sysctllog **);
253 
254 static int
255 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
256     void *arg0, void *arg1, void *arg2, void *arg3)
257 {
258 	int result;
259 	enum kauth_network_req req;
260 
261 	result = KAUTH_RESULT_DEFER;
262 	req = (enum kauth_network_req)arg1;
263 
264 	if (action != KAUTH_NETWORK_INTERFACE)
265 		return result;
266 
267 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
268 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
269 		result = KAUTH_RESULT_ALLOW;
270 
271 	return result;
272 }
273 
274 /*
275  * Network interface utility routines.
276  *
277  * Routines with ifa_ifwith* names take sockaddr *'s as
278  * parameters.
279  */
280 void
281 ifinit(void)
282 {
283 
284 #if (defined(INET) || defined(INET6))
285 	encapinit();
286 #endif
287 
288 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
289 	    if_listener_cb, NULL);
290 
291 	/* interfaces are available, inform socket code */
292 	ifioctl = doifioctl;
293 }
294 
295 /*
296  * XXX Initialization before configure().
297  * XXX hack to get pfil_add_hook working in autoconf.
298  */
299 void
300 ifinit1(void)
301 {
302 
303 #ifdef NET_MPSAFE
304 	printf("NET_MPSAFE enabled\n");
305 #endif
306 
307 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
308 
309 	TAILQ_INIT(&ifnet_list);
310 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
311 	ifnet_psz = pserialize_create();
312 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
313 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
314 	PSLIST_INIT(&ifnet_pslist);
315 
316 	if_indexlim = 8;
317 
318 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
319 	KASSERT(if_pfil != NULL);
320 
321 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
322 	etherinit();
323 #endif
324 }
325 
326 /* XXX must be after domaininit() */
327 void
328 ifinit_post(void)
329 {
330 
331 	if_sysctl_setup(NULL);
332 }
333 
334 ifnet_t *
335 if_alloc(u_char type)
336 {
337 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
338 }
339 
340 void
341 if_free(ifnet_t *ifp)
342 {
343 	kmem_free(ifp, sizeof(ifnet_t));
344 }
345 
346 void
347 if_initname(struct ifnet *ifp, const char *name, int unit)
348 {
349 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
350 	    "%s%d", name, unit);
351 }
352 
353 /*
354  * Null routines used while an interface is going away.  These routines
355  * just return an error.
356  */
357 
358 int
359 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
360     const struct sockaddr *so, const struct rtentry *rt)
361 {
362 
363 	return ENXIO;
364 }
365 
366 void
367 if_nullinput(struct ifnet *ifp, struct mbuf *m)
368 {
369 
370 	/* Nothing. */
371 }
372 
373 void
374 if_nullstart(struct ifnet *ifp)
375 {
376 
377 	/* Nothing. */
378 }
379 
380 int
381 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
382 {
383 
384 	m_freem(m);
385 	return ENXIO;
386 }
387 
388 int
389 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
390 {
391 
392 	return ENXIO;
393 }
394 
395 int
396 if_nullinit(struct ifnet *ifp)
397 {
398 
399 	return ENXIO;
400 }
401 
402 void
403 if_nullstop(struct ifnet *ifp, int disable)
404 {
405 
406 	/* Nothing. */
407 }
408 
409 void
410 if_nullslowtimo(struct ifnet *ifp)
411 {
412 
413 	/* Nothing. */
414 }
415 
416 void
417 if_nulldrain(struct ifnet *ifp)
418 {
419 
420 	/* Nothing. */
421 }
422 
423 void
424 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
425 {
426 	struct ifaddr *ifa;
427 	struct sockaddr_dl *sdl;
428 
429 	ifp->if_addrlen = addrlen;
430 	if_alloc_sadl(ifp);
431 	ifa = ifp->if_dl;
432 	sdl = satosdl(ifa->ifa_addr);
433 
434 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
435 	if (factory) {
436 		KASSERT(ifp->if_hwdl == NULL);
437 		ifp->if_hwdl = ifp->if_dl;
438 		ifaref(ifp->if_hwdl);
439 	}
440 	/* TBD routing socket */
441 }
442 
443 struct ifaddr *
444 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
445 {
446 	unsigned socksize, ifasize;
447 	int addrlen, namelen;
448 	struct sockaddr_dl *mask, *sdl;
449 	struct ifaddr *ifa;
450 
451 	namelen = strlen(ifp->if_xname);
452 	addrlen = ifp->if_addrlen;
453 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
454 	ifasize = sizeof(*ifa) + 2 * socksize;
455 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
456 
457 	sdl = (struct sockaddr_dl *)(ifa + 1);
458 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
459 
460 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
461 	    ifp->if_xname, namelen, NULL, addrlen);
462 	mask->sdl_family = AF_LINK;
463 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
464 	memset(&mask->sdl_data[0], 0xff, namelen);
465 	ifa->ifa_rtrequest = link_rtrequest;
466 	ifa->ifa_addr = (struct sockaddr *)sdl;
467 	ifa->ifa_netmask = (struct sockaddr *)mask;
468 	ifa_psref_init(ifa);
469 
470 	*sdlp = sdl;
471 
472 	return ifa;
473 }
474 
475 static void
476 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
477 {
478 	const struct sockaddr_dl *sdl;
479 
480 	ifp->if_dl = ifa;
481 	ifaref(ifa);
482 	sdl = satosdl(ifa->ifa_addr);
483 	ifp->if_sadl = sdl;
484 }
485 
486 /*
487  * Allocate the link level name for the specified interface.  This
488  * is an attachment helper.  It must be called after ifp->if_addrlen
489  * is initialized, which may not be the case when if_attach() is
490  * called.
491  */
492 void
493 if_alloc_sadl(struct ifnet *ifp)
494 {
495 	struct ifaddr *ifa;
496 	const struct sockaddr_dl *sdl;
497 
498 	/*
499 	 * If the interface already has a link name, release it
500 	 * now.  This is useful for interfaces that can change
501 	 * link types, and thus switch link names often.
502 	 */
503 	if (ifp->if_sadl != NULL)
504 		if_free_sadl(ifp, 0);
505 
506 	ifa = if_dl_create(ifp, &sdl);
507 
508 	ifa_insert(ifp, ifa);
509 	if_sadl_setrefs(ifp, ifa);
510 }
511 
512 static void
513 if_deactivate_sadl(struct ifnet *ifp)
514 {
515 	struct ifaddr *ifa;
516 
517 	KASSERT(ifp->if_dl != NULL);
518 
519 	ifa = ifp->if_dl;
520 
521 	ifp->if_sadl = NULL;
522 
523 	ifp->if_dl = NULL;
524 	ifafree(ifa);
525 }
526 
527 static void
528 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa)
529 {
530 	struct ifaddr *old;
531 
532 	KASSERT(ifp->if_dl != NULL);
533 
534 	old = ifp->if_dl;
535 
536 	ifaref(ifa);
537 	/* XXX Update if_dl and if_sadl atomically */
538 	ifp->if_dl = ifa;
539 	ifp->if_sadl = satosdl(ifa->ifa_addr);
540 
541 	ifafree(old);
542 }
543 
544 void
545 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
546     const struct sockaddr_dl *sdl)
547 {
548 	int s, ss;
549 	struct ifaddr *ifa;
550 	int bound = curlwp_bind();
551 
552 	KASSERT(ifa_held(ifa0));
553 
554 	s = splsoftnet();
555 
556 	if_replace_sadl(ifp, ifa0);
557 
558 	ss = pserialize_read_enter();
559 	IFADDR_READER_FOREACH(ifa, ifp) {
560 		struct psref psref;
561 		ifa_acquire(ifa, &psref);
562 		pserialize_read_exit(ss);
563 
564 		rtinit(ifa, RTM_LLINFO_UPD, 0);
565 
566 		ss = pserialize_read_enter();
567 		ifa_release(ifa, &psref);
568 	}
569 	pserialize_read_exit(ss);
570 
571 	splx(s);
572 	curlwp_bindx(bound);
573 }
574 
575 /*
576  * Free the link level name for the specified interface.  This is
577  * a detach helper.  This is called from if_detach().
578  */
579 void
580 if_free_sadl(struct ifnet *ifp, int factory)
581 {
582 	struct ifaddr *ifa;
583 	int s;
584 
585 	if (factory && ifp->if_hwdl != NULL) {
586 		ifa = ifp->if_hwdl;
587 		ifp->if_hwdl = NULL;
588 		ifafree(ifa);
589 	}
590 
591 	ifa = ifp->if_dl;
592 	if (ifa == NULL) {
593 		KASSERT(ifp->if_sadl == NULL);
594 		return;
595 	}
596 
597 	KASSERT(ifp->if_sadl != NULL);
598 
599 	s = splsoftnet();
600 	KASSERT(ifa->ifa_addr->sa_family == AF_LINK);
601 	ifa_remove(ifp, ifa);
602 	if_deactivate_sadl(ifp);
603 	splx(s);
604 }
605 
606 static void
607 if_getindex(ifnet_t *ifp)
608 {
609 	bool hitlimit = false;
610 
611 	ifp->if_index_gen = index_gen++;
612 
613 	ifp->if_index = if_index;
614 	if (ifindex2ifnet == NULL) {
615 		if_index++;
616 		goto skip;
617 	}
618 	while (if_byindex(ifp->if_index)) {
619 		/*
620 		 * If we hit USHRT_MAX, we skip back to 0 since
621 		 * there are a number of places where the value
622 		 * of if_index or if_index itself is compared
623 		 * to or stored in an unsigned short.  By
624 		 * jumping back, we won't botch those assignments
625 		 * or comparisons.
626 		 */
627 		if (++if_index == 0) {
628 			if_index = 1;
629 		} else if (if_index == USHRT_MAX) {
630 			/*
631 			 * However, if we have to jump back to
632 			 * zero *twice* without finding an empty
633 			 * slot in ifindex2ifnet[], then there
634 			 * there are too many (>65535) interfaces.
635 			 */
636 			if (hitlimit) {
637 				panic("too many interfaces");
638 			}
639 			hitlimit = true;
640 			if_index = 1;
641 		}
642 		ifp->if_index = if_index;
643 	}
644 skip:
645 	/*
646 	 * ifindex2ifnet is indexed by if_index. Since if_index will
647 	 * grow dynamically, it should grow too.
648 	 */
649 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
650 		size_t m, n, oldlim;
651 		void *q;
652 
653 		oldlim = if_indexlim;
654 		while (ifp->if_index >= if_indexlim)
655 			if_indexlim <<= 1;
656 
657 		/* grow ifindex2ifnet */
658 		m = oldlim * sizeof(struct ifnet *);
659 		n = if_indexlim * sizeof(struct ifnet *);
660 		q = malloc(n, M_IFADDR, M_WAITOK | M_ZERO);
661 		if (ifindex2ifnet != NULL) {
662 			memcpy(q, ifindex2ifnet, m);
663 			free(ifindex2ifnet, M_IFADDR);
664 		}
665 		ifindex2ifnet = (struct ifnet **)q;
666 	}
667 	ifindex2ifnet[ifp->if_index] = ifp;
668 }
669 
670 /*
671  * Initialize an interface and assign an index for it.
672  *
673  * It must be called prior to a device specific attach routine
674  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
675  * and be followed by if_register:
676  *
677  *     if_initialize(ifp);
678  *     ether_ifattach(ifp, enaddr);
679  *     if_register(ifp);
680  */
681 int
682 if_initialize(ifnet_t *ifp)
683 {
684 	int rv = 0;
685 
686 	KASSERT(if_indexlim > 0);
687 	TAILQ_INIT(&ifp->if_addrlist);
688 
689 	/*
690 	 * Link level name is allocated later by a separate call to
691 	 * if_alloc_sadl().
692 	 */
693 
694 	if (ifp->if_snd.ifq_maxlen == 0)
695 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
696 
697 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
698 
699 	ifp->if_link_state = LINK_STATE_UNKNOWN;
700 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
701 
702 	ifp->if_capenable = 0;
703 	ifp->if_csum_flags_tx = 0;
704 	ifp->if_csum_flags_rx = 0;
705 
706 #ifdef ALTQ
707 	ifp->if_snd.altq_type = 0;
708 	ifp->if_snd.altq_disc = NULL;
709 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
710 	ifp->if_snd.altq_tbr  = NULL;
711 	ifp->if_snd.altq_ifp  = ifp;
712 #endif
713 
714 	IFQ_LOCK_INIT(&ifp->if_snd);
715 
716 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
717 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
718 
719 	IF_AFDATA_LOCK_INIT(ifp);
720 
721 	if (if_is_link_state_changeable(ifp)) {
722 		u_int flags = SOFTINT_NET;
723 		flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
724 		ifp->if_link_si = softint_establish(flags,
725 		    if_link_state_change_si, ifp);
726 		if (ifp->if_link_si == NULL) {
727 			rv = ENOMEM;
728 			goto fail;
729 		}
730 	}
731 
732 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
733 	PSLIST_INIT(&ifp->if_addr_pslist);
734 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
735 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
736 	LIST_INIT(&ifp->if_multiaddrs);
737 
738 	IFNET_GLOBAL_LOCK();
739 	if_getindex(ifp);
740 	IFNET_GLOBAL_UNLOCK();
741 
742 	return 0;
743 
744 fail:
745 	IF_AFDATA_LOCK_DESTROY(ifp);
746 
747 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
748 	(void)pfil_head_destroy(ifp->if_pfil);
749 
750 	IFQ_LOCK_DESTROY(&ifp->if_snd);
751 
752 	return rv;
753 }
754 
755 /*
756  * Register an interface to the list of "active" interfaces.
757  */
758 void
759 if_register(ifnet_t *ifp)
760 {
761 	/*
762 	 * If the driver has not supplied its own if_ioctl, then
763 	 * supply the default.
764 	 */
765 	if (ifp->if_ioctl == NULL)
766 		ifp->if_ioctl = ifioctl_common;
767 
768 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
769 
770 	if (!STAILQ_EMPTY(&domains))
771 		if_attachdomain1(ifp);
772 
773 	/* Announce the interface. */
774 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
775 
776 	if (ifp->if_slowtimo != NULL) {
777 		ifp->if_slowtimo_ch =
778 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
779 		callout_init(ifp->if_slowtimo_ch, 0);
780 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
781 		if_slowtimo(ifp);
782 	}
783 
784 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
785 		ifp->if_transmit = if_transmit;
786 
787 	IFNET_GLOBAL_LOCK();
788 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
789 	IFNET_WRITER_INSERT_TAIL(ifp);
790 	IFNET_GLOBAL_UNLOCK();
791 }
792 
793 /*
794  * The if_percpuq framework
795  *
796  * It allows network device drivers to execute the network stack
797  * in softint (so called softint-based if_input). It utilizes
798  * softint and percpu ifqueue. It doesn't distribute any packets
799  * between CPUs, unlike pktqueue(9).
800  *
801  * Currently we support two options for device drivers to apply the framework:
802  * - Use it implicitly with less changes
803  *   - If you use if_attach in driver's _attach function and if_input in
804  *     driver's Rx interrupt handler, a packet is queued and a softint handles
805  *     the packet implicitly
806  * - Use it explicitly in each driver (recommended)
807  *   - You can use if_percpuq_* directly in your driver
808  *   - In this case, you need to allocate struct if_percpuq in driver's softc
809  *   - See wm(4) as a reference implementation
810  */
811 
812 static void
813 if_percpuq_softint(void *arg)
814 {
815 	struct if_percpuq *ipq = arg;
816 	struct ifnet *ifp = ipq->ipq_ifp;
817 	struct mbuf *m;
818 
819 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
820 		ifp->if_ipackets++;
821 		bpf_mtap(ifp, m, BPF_D_IN);
822 
823 		ifp->_if_input(ifp, m);
824 	}
825 }
826 
827 static void
828 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
829 {
830 	struct ifqueue *const ifq = p;
831 
832 	memset(ifq, 0, sizeof(*ifq));
833 	ifq->ifq_maxlen = IFQ_MAXLEN;
834 }
835 
836 struct if_percpuq *
837 if_percpuq_create(struct ifnet *ifp)
838 {
839 	struct if_percpuq *ipq;
840 	u_int flags = SOFTINT_NET;
841 
842 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
843 
844 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
845 	ipq->ipq_ifp = ifp;
846 	ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq);
847 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
848 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
849 
850 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
851 
852 	return ipq;
853 }
854 
855 static struct mbuf *
856 if_percpuq_dequeue(struct if_percpuq *ipq)
857 {
858 	struct mbuf *m;
859 	struct ifqueue *ifq;
860 	int s;
861 
862 	s = splnet();
863 	ifq = percpu_getref(ipq->ipq_ifqs);
864 	IF_DEQUEUE(ifq, m);
865 	percpu_putref(ipq->ipq_ifqs);
866 	splx(s);
867 
868 	return m;
869 }
870 
871 static void
872 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
873 {
874 	struct ifqueue *const ifq = p;
875 
876 	IF_PURGE(ifq);
877 }
878 
879 void
880 if_percpuq_destroy(struct if_percpuq *ipq)
881 {
882 
883 	/* if_detach may already destroy it */
884 	if (ipq == NULL)
885 		return;
886 
887 	softint_disestablish(ipq->ipq_si);
888 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
889 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
890 	kmem_free(ipq, sizeof(*ipq));
891 }
892 
893 void
894 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
895 {
896 	struct ifqueue *ifq;
897 	int s;
898 
899 	KASSERT(ipq != NULL);
900 
901 	s = splnet();
902 	ifq = percpu_getref(ipq->ipq_ifqs);
903 	if (IF_QFULL(ifq)) {
904 		IF_DROP(ifq);
905 		percpu_putref(ipq->ipq_ifqs);
906 		m_freem(m);
907 		goto out;
908 	}
909 	IF_ENQUEUE(ifq, m);
910 	percpu_putref(ipq->ipq_ifqs);
911 
912 	softint_schedule(ipq->ipq_si);
913 out:
914 	splx(s);
915 }
916 
917 static void
918 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
919 {
920 	struct ifqueue *const ifq = p;
921 	int *sum = arg;
922 
923 	*sum += ifq->ifq_drops;
924 }
925 
926 static int
927 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
928 {
929 	struct sysctlnode node;
930 	struct if_percpuq *ipq;
931 	int sum = 0;
932 	int error;
933 
934 	node = *rnode;
935 	ipq = node.sysctl_data;
936 
937 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
938 
939 	node.sysctl_data = &sum;
940 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
941 	if (error != 0 || newp == NULL)
942 		return error;
943 
944 	return 0;
945 }
946 
947 static void
948 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
949     struct if_percpuq *ipq)
950 {
951 	const struct sysctlnode *cnode, *rnode;
952 
953 	if (sysctl_createv(clog, 0, NULL, &rnode,
954 		       CTLFLAG_PERMANENT,
955 		       CTLTYPE_NODE, "interfaces",
956 		       SYSCTL_DESCR("Per-interface controls"),
957 		       NULL, 0, NULL, 0,
958 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
959 		goto bad;
960 
961 	if (sysctl_createv(clog, 0, &rnode, &rnode,
962 		       CTLFLAG_PERMANENT,
963 		       CTLTYPE_NODE, ifname,
964 		       SYSCTL_DESCR("Interface controls"),
965 		       NULL, 0, NULL, 0,
966 		       CTL_CREATE, CTL_EOL) != 0)
967 		goto bad;
968 
969 	if (sysctl_createv(clog, 0, &rnode, &rnode,
970 		       CTLFLAG_PERMANENT,
971 		       CTLTYPE_NODE, "rcvq",
972 		       SYSCTL_DESCR("Interface input queue controls"),
973 		       NULL, 0, NULL, 0,
974 		       CTL_CREATE, CTL_EOL) != 0)
975 		goto bad;
976 
977 #ifdef NOTYET
978 	/* XXX Should show each per-CPU queue length? */
979 	if (sysctl_createv(clog, 0, &rnode, &rnode,
980 		       CTLFLAG_PERMANENT,
981 		       CTLTYPE_INT, "len",
982 		       SYSCTL_DESCR("Current input queue length"),
983 		       sysctl_percpuq_len, 0, NULL, 0,
984 		       CTL_CREATE, CTL_EOL) != 0)
985 		goto bad;
986 
987 	if (sysctl_createv(clog, 0, &rnode, &cnode,
988 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
989 		       CTLTYPE_INT, "maxlen",
990 		       SYSCTL_DESCR("Maximum allowed input queue length"),
991 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
992 		       CTL_CREATE, CTL_EOL) != 0)
993 		goto bad;
994 #endif
995 
996 	if (sysctl_createv(clog, 0, &rnode, &cnode,
997 		       CTLFLAG_PERMANENT,
998 		       CTLTYPE_INT, "drops",
999 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
1000 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
1001 		       CTL_CREATE, CTL_EOL) != 0)
1002 		goto bad;
1003 
1004 	return;
1005 bad:
1006 	printf("%s: could not attach sysctl nodes\n", ifname);
1007 	return;
1008 }
1009 
1010 /*
1011  * The deferred if_start framework
1012  *
1013  * The common APIs to defer if_start to softint when if_start is requested
1014  * from a device driver running in hardware interrupt context.
1015  */
1016 /*
1017  * Call ifp->if_start (or equivalent) in a dedicated softint for
1018  * deferred if_start.
1019  */
1020 static void
1021 if_deferred_start_softint(void *arg)
1022 {
1023 	struct if_deferred_start *ids = arg;
1024 	struct ifnet *ifp = ids->ids_ifp;
1025 
1026 	ids->ids_if_start(ifp);
1027 }
1028 
1029 /*
1030  * The default callback function for deferred if_start.
1031  */
1032 static void
1033 if_deferred_start_common(struct ifnet *ifp)
1034 {
1035 	int s;
1036 
1037 	s = splnet();
1038 	if_start_lock(ifp);
1039 	splx(s);
1040 }
1041 
1042 static inline bool
1043 if_snd_is_used(struct ifnet *ifp)
1044 {
1045 
1046 	return ALTQ_IS_ENABLED(&ifp->if_snd) ||
1047 		ifp->if_transmit == if_transmit ||
1048 		ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit;
1049 }
1050 
1051 /*
1052  * Schedule deferred if_start.
1053  */
1054 void
1055 if_schedule_deferred_start(struct ifnet *ifp)
1056 {
1057 
1058 	KASSERT(ifp->if_deferred_start != NULL);
1059 
1060 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1061 		return;
1062 
1063 	softint_schedule(ifp->if_deferred_start->ids_si);
1064 }
1065 
1066 /*
1067  * Create an instance of deferred if_start. A driver should call the function
1068  * only if the driver needs deferred if_start. Drivers can setup their own
1069  * deferred if_start function via 2nd argument.
1070  */
1071 void
1072 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1073 {
1074 	struct if_deferred_start *ids;
1075 	u_int flags = SOFTINT_NET;
1076 
1077 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
1078 
1079 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1080 	ids->ids_ifp = ifp;
1081 	ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids);
1082 	if (func != NULL)
1083 		ids->ids_if_start = func;
1084 	else
1085 		ids->ids_if_start = if_deferred_start_common;
1086 
1087 	ifp->if_deferred_start = ids;
1088 }
1089 
1090 static void
1091 if_deferred_start_destroy(struct ifnet *ifp)
1092 {
1093 
1094 	if (ifp->if_deferred_start == NULL)
1095 		return;
1096 
1097 	softint_disestablish(ifp->if_deferred_start->ids_si);
1098 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1099 	ifp->if_deferred_start = NULL;
1100 }
1101 
1102 /*
1103  * The common interface input routine that is called by device drivers,
1104  * which should be used only when the driver's rx handler already runs
1105  * in softint.
1106  */
1107 void
1108 if_input(struct ifnet *ifp, struct mbuf *m)
1109 {
1110 
1111 	KASSERT(ifp->if_percpuq == NULL);
1112 	KASSERT(!cpu_intr_p());
1113 
1114 	ifp->if_ipackets++;
1115 	bpf_mtap(ifp, m, BPF_D_IN);
1116 
1117 	ifp->_if_input(ifp, m);
1118 }
1119 
1120 /*
1121  * DEPRECATED. Use if_initialize and if_register instead.
1122  * See the above comment of if_initialize.
1123  *
1124  * Note that it implicitly enables if_percpuq to make drivers easy to
1125  * migrate softint-based if_input without much changes. If you don't
1126  * want to enable it, use if_initialize instead.
1127  */
1128 int
1129 if_attach(ifnet_t *ifp)
1130 {
1131 	int rv;
1132 
1133 	rv = if_initialize(ifp);
1134 	if (rv != 0)
1135 		return rv;
1136 
1137 	ifp->if_percpuq = if_percpuq_create(ifp);
1138 	if_register(ifp);
1139 
1140 	return 0;
1141 }
1142 
1143 void
1144 if_attachdomain(void)
1145 {
1146 	struct ifnet *ifp;
1147 	int s;
1148 	int bound = curlwp_bind();
1149 
1150 	s = pserialize_read_enter();
1151 	IFNET_READER_FOREACH(ifp) {
1152 		struct psref psref;
1153 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1154 		pserialize_read_exit(s);
1155 		if_attachdomain1(ifp);
1156 		s = pserialize_read_enter();
1157 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1158 	}
1159 	pserialize_read_exit(s);
1160 	curlwp_bindx(bound);
1161 }
1162 
1163 static void
1164 if_attachdomain1(struct ifnet *ifp)
1165 {
1166 	struct domain *dp;
1167 	int s;
1168 
1169 	s = splsoftnet();
1170 
1171 	/* address family dependent data region */
1172 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1173 	DOMAIN_FOREACH(dp) {
1174 		if (dp->dom_ifattach != NULL)
1175 			ifp->if_afdata[dp->dom_family] =
1176 			    (*dp->dom_ifattach)(ifp);
1177 	}
1178 
1179 	splx(s);
1180 }
1181 
1182 /*
1183  * Deactivate an interface.  This points all of the procedure
1184  * handles at error stubs.  May be called from interrupt context.
1185  */
1186 void
1187 if_deactivate(struct ifnet *ifp)
1188 {
1189 	int s;
1190 
1191 	s = splsoftnet();
1192 
1193 	ifp->if_output	 = if_nulloutput;
1194 	ifp->_if_input	 = if_nullinput;
1195 	ifp->if_start	 = if_nullstart;
1196 	ifp->if_transmit = if_nulltransmit;
1197 	ifp->if_ioctl	 = if_nullioctl;
1198 	ifp->if_init	 = if_nullinit;
1199 	ifp->if_stop	 = if_nullstop;
1200 	ifp->if_slowtimo = if_nullslowtimo;
1201 	ifp->if_drain	 = if_nulldrain;
1202 
1203 	/* No more packets may be enqueued. */
1204 	ifp->if_snd.ifq_maxlen = 0;
1205 
1206 	splx(s);
1207 }
1208 
1209 bool
1210 if_is_deactivated(const struct ifnet *ifp)
1211 {
1212 
1213 	return ifp->if_output == if_nulloutput;
1214 }
1215 
1216 void
1217 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1218 {
1219 	struct ifaddr *ifa, *nifa;
1220 	int s;
1221 
1222 	s = pserialize_read_enter();
1223 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1224 		nifa = IFADDR_READER_NEXT(ifa);
1225 		if (ifa->ifa_addr->sa_family != family)
1226 			continue;
1227 		pserialize_read_exit(s);
1228 
1229 		(*purgeaddr)(ifa);
1230 
1231 		s = pserialize_read_enter();
1232 	}
1233 	pserialize_read_exit(s);
1234 }
1235 
1236 #ifdef IFAREF_DEBUG
1237 static struct ifaddr **ifa_list;
1238 static int ifa_list_size;
1239 
1240 /* Depends on only one if_attach runs at once */
1241 static void
1242 if_build_ifa_list(struct ifnet *ifp)
1243 {
1244 	struct ifaddr *ifa;
1245 	int i;
1246 
1247 	KASSERT(ifa_list == NULL);
1248 	KASSERT(ifa_list_size == 0);
1249 
1250 	IFADDR_READER_FOREACH(ifa, ifp)
1251 		ifa_list_size++;
1252 
1253 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1254 	i = 0;
1255 	IFADDR_READER_FOREACH(ifa, ifp) {
1256 		ifa_list[i++] = ifa;
1257 		ifaref(ifa);
1258 	}
1259 }
1260 
1261 static void
1262 if_check_and_free_ifa_list(struct ifnet *ifp)
1263 {
1264 	int i;
1265 	struct ifaddr *ifa;
1266 
1267 	if (ifa_list == NULL)
1268 		return;
1269 
1270 	for (i = 0; i < ifa_list_size; i++) {
1271 		char buf[64];
1272 
1273 		ifa = ifa_list[i];
1274 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1275 		if (ifa->ifa_refcnt > 1) {
1276 			log(LOG_WARNING,
1277 			    "ifa(%s) still referenced (refcnt=%d)\n",
1278 			    buf, ifa->ifa_refcnt - 1);
1279 		} else
1280 			log(LOG_DEBUG,
1281 			    "ifa(%s) not referenced (refcnt=%d)\n",
1282 			    buf, ifa->ifa_refcnt - 1);
1283 		ifafree(ifa);
1284 	}
1285 
1286 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1287 	ifa_list = NULL;
1288 	ifa_list_size = 0;
1289 }
1290 #endif
1291 
1292 /*
1293  * Detach an interface from the list of "active" interfaces,
1294  * freeing any resources as we go along.
1295  *
1296  * NOTE: This routine must be called with a valid thread context,
1297  * as it may block.
1298  */
1299 void
1300 if_detach(struct ifnet *ifp)
1301 {
1302 	struct socket so;
1303 	struct ifaddr *ifa;
1304 #ifdef IFAREF_DEBUG
1305 	struct ifaddr *last_ifa = NULL;
1306 #endif
1307 	struct domain *dp;
1308 	const struct protosw *pr;
1309 	int s, i, family, purged;
1310 
1311 #ifdef IFAREF_DEBUG
1312 	if_build_ifa_list(ifp);
1313 #endif
1314 	/*
1315 	 * XXX It's kind of lame that we have to have the
1316 	 * XXX socket structure...
1317 	 */
1318 	memset(&so, 0, sizeof(so));
1319 
1320 	s = splnet();
1321 
1322 	sysctl_teardown(&ifp->if_sysctl_log);
1323 	IFNET_LOCK(ifp);
1324 	if_deactivate(ifp);
1325 	IFNET_UNLOCK(ifp);
1326 
1327 	/*
1328 	 * Unlink from the list and wait for all readers to leave
1329 	 * from pserialize read sections.  Note that we can't do
1330 	 * psref_target_destroy here.  See below.
1331 	 */
1332 	IFNET_GLOBAL_LOCK();
1333 	ifindex2ifnet[ifp->if_index] = NULL;
1334 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1335 	IFNET_WRITER_REMOVE(ifp);
1336 	pserialize_perform(ifnet_psz);
1337 	IFNET_GLOBAL_UNLOCK();
1338 
1339 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1340 		ifp->if_slowtimo = NULL;
1341 		callout_halt(ifp->if_slowtimo_ch, NULL);
1342 		callout_destroy(ifp->if_slowtimo_ch);
1343 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1344 	}
1345 	if_deferred_start_destroy(ifp);
1346 
1347 	/*
1348 	 * Do an if_down() to give protocols a chance to do something.
1349 	 */
1350 	if_down_deactivated(ifp);
1351 
1352 #ifdef ALTQ
1353 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1354 		altq_disable(&ifp->if_snd);
1355 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1356 		altq_detach(&ifp->if_snd);
1357 #endif
1358 
1359 #if NCARP > 0
1360 	/* Remove the interface from any carp group it is a part of.  */
1361 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1362 		carp_ifdetach(ifp);
1363 #endif
1364 
1365 	/*
1366 	 * Rip all the addresses off the interface.  This should make
1367 	 * all of the routes go away.
1368 	 *
1369 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1370 	 * from the list, including our "cursor", ifa.  For safety,
1371 	 * and to honor the TAILQ abstraction, I just restart the
1372 	 * loop after each removal.  Note that the loop will exit
1373 	 * when all of the remaining ifaddrs belong to the AF_LINK
1374 	 * family.  I am counting on the historical fact that at
1375 	 * least one pr_usrreq in each address domain removes at
1376 	 * least one ifaddr.
1377 	 */
1378 again:
1379 	/*
1380 	 * At this point, no other one tries to remove ifa in the list,
1381 	 * so we don't need to take a lock or psref.  Avoid using
1382 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1383 	 * violations of pserialize.
1384 	 */
1385 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1386 		family = ifa->ifa_addr->sa_family;
1387 #ifdef IFAREF_DEBUG
1388 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1389 		    ifa, family, ifa->ifa_refcnt);
1390 		if (last_ifa != NULL && ifa == last_ifa)
1391 			panic("if_detach: loop detected");
1392 		last_ifa = ifa;
1393 #endif
1394 		if (family == AF_LINK)
1395 			continue;
1396 		dp = pffinddomain(family);
1397 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1398 		/*
1399 		 * XXX These PURGEIF calls are redundant with the
1400 		 * purge-all-families calls below, but are left in for
1401 		 * now both to make a smaller change, and to avoid
1402 		 * unplanned interactions with clearing of
1403 		 * ifp->if_addrlist.
1404 		 */
1405 		purged = 0;
1406 		for (pr = dp->dom_protosw;
1407 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1408 			so.so_proto = pr;
1409 			if (pr->pr_usrreqs) {
1410 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1411 				purged = 1;
1412 			}
1413 		}
1414 		if (purged == 0) {
1415 			/*
1416 			 * XXX What's really the best thing to do
1417 			 * XXX here?  --thorpej@NetBSD.org
1418 			 */
1419 			printf("if_detach: WARNING: AF %d not purged\n",
1420 			    family);
1421 			ifa_remove(ifp, ifa);
1422 		}
1423 		goto again;
1424 	}
1425 
1426 	if_free_sadl(ifp, 1);
1427 
1428 restart:
1429 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1430 		family = ifa->ifa_addr->sa_family;
1431 		KASSERT(family == AF_LINK);
1432 		ifa_remove(ifp, ifa);
1433 		goto restart;
1434 	}
1435 
1436 	/* Delete stray routes from the routing table. */
1437 	for (i = 0; i <= AF_MAX; i++)
1438 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1439 
1440 	DOMAIN_FOREACH(dp) {
1441 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1442 		{
1443 			void *p = ifp->if_afdata[dp->dom_family];
1444 			if (p) {
1445 				ifp->if_afdata[dp->dom_family] = NULL;
1446 				(*dp->dom_ifdetach)(ifp, p);
1447 			}
1448 		}
1449 
1450 		/*
1451 		 * One would expect multicast memberships (INET and
1452 		 * INET6) on UDP sockets to be purged by the PURGEIF
1453 		 * calls above, but if all addresses were removed from
1454 		 * the interface prior to destruction, the calls will
1455 		 * not be made (e.g. ppp, for which pppd(8) generally
1456 		 * removes addresses before destroying the interface).
1457 		 * Because there is no invariant that multicast
1458 		 * memberships only exist for interfaces with IPv4
1459 		 * addresses, we must call PURGEIF regardless of
1460 		 * addresses.  (Protocols which might store ifnet
1461 		 * pointers are marked with PR_PURGEIF.)
1462 		 */
1463 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1464 			so.so_proto = pr;
1465 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1466 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1467 		}
1468 	}
1469 
1470 	/*
1471 	 * Must be done after the above pr_purgeif because if_psref may be
1472 	 * still used in pr_purgeif.
1473 	 */
1474 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1475 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1476 
1477 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1478 	(void)pfil_head_destroy(ifp->if_pfil);
1479 
1480 	/* Announce that the interface is gone. */
1481 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1482 
1483 	IF_AFDATA_LOCK_DESTROY(ifp);
1484 
1485 	if (if_is_link_state_changeable(ifp)) {
1486 		softint_disestablish(ifp->if_link_si);
1487 		ifp->if_link_si = NULL;
1488 	}
1489 
1490 	/*
1491 	 * remove packets that came from ifp, from software interrupt queues.
1492 	 */
1493 	DOMAIN_FOREACH(dp) {
1494 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1495 			struct ifqueue *iq = dp->dom_ifqueues[i];
1496 			if (iq == NULL)
1497 				break;
1498 			dp->dom_ifqueues[i] = NULL;
1499 			if_detach_queues(ifp, iq);
1500 		}
1501 	}
1502 
1503 	/*
1504 	 * IP queues have to be processed separately: net-queue barrier
1505 	 * ensures that the packets are dequeued while a cross-call will
1506 	 * ensure that the interrupts have completed. FIXME: not quite..
1507 	 */
1508 #ifdef INET
1509 	pktq_barrier(ip_pktq);
1510 #endif
1511 #ifdef INET6
1512 	if (in6_present)
1513 		pktq_barrier(ip6_pktq);
1514 #endif
1515 	xc_barrier(0);
1516 
1517 	if (ifp->if_percpuq != NULL) {
1518 		if_percpuq_destroy(ifp->if_percpuq);
1519 		ifp->if_percpuq = NULL;
1520 	}
1521 
1522 	mutex_obj_free(ifp->if_ioctl_lock);
1523 	ifp->if_ioctl_lock = NULL;
1524 	mutex_obj_free(ifp->if_snd.ifq_lock);
1525 
1526 	splx(s);
1527 
1528 #ifdef IFAREF_DEBUG
1529 	if_check_and_free_ifa_list(ifp);
1530 #endif
1531 }
1532 
1533 static void
1534 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1535 {
1536 	struct mbuf *m, *prev, *next;
1537 
1538 	prev = NULL;
1539 	for (m = q->ifq_head; m != NULL; m = next) {
1540 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1541 
1542 		next = m->m_nextpkt;
1543 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1544 			prev = m;
1545 			continue;
1546 		}
1547 
1548 		if (prev != NULL)
1549 			prev->m_nextpkt = m->m_nextpkt;
1550 		else
1551 			q->ifq_head = m->m_nextpkt;
1552 		if (q->ifq_tail == m)
1553 			q->ifq_tail = prev;
1554 		q->ifq_len--;
1555 
1556 		m->m_nextpkt = NULL;
1557 		m_freem(m);
1558 		IF_DROP(q);
1559 	}
1560 }
1561 
1562 /*
1563  * Callback for a radix tree walk to delete all references to an
1564  * ifnet.
1565  */
1566 static int
1567 if_delroute_matcher(struct rtentry *rt, void *v)
1568 {
1569 	struct ifnet *ifp = (struct ifnet *)v;
1570 
1571 	if (rt->rt_ifp == ifp)
1572 		return 1;
1573 	else
1574 		return 0;
1575 }
1576 
1577 /*
1578  * Create a clone network interface.
1579  */
1580 static int
1581 if_clone_create(const char *name)
1582 {
1583 	struct if_clone *ifc;
1584 	int unit;
1585 	struct ifnet *ifp;
1586 	struct psref psref;
1587 
1588 	KASSERT(mutex_owned(&if_clone_mtx));
1589 
1590 	ifc = if_clone_lookup(name, &unit);
1591 	if (ifc == NULL)
1592 		return EINVAL;
1593 
1594 	ifp = if_get(name, &psref);
1595 	if (ifp != NULL) {
1596 		if_put(ifp, &psref);
1597 		return EEXIST;
1598 	}
1599 
1600 	return (*ifc->ifc_create)(ifc, unit);
1601 }
1602 
1603 /*
1604  * Destroy a clone network interface.
1605  */
1606 static int
1607 if_clone_destroy(const char *name)
1608 {
1609 	struct if_clone *ifc;
1610 	struct ifnet *ifp;
1611 	struct psref psref;
1612 	int error;
1613 	int (*if_ioctl)(struct ifnet *, u_long, void *);
1614 
1615 	KASSERT(mutex_owned(&if_clone_mtx));
1616 
1617 	ifc = if_clone_lookup(name, NULL);
1618 	if (ifc == NULL)
1619 		return EINVAL;
1620 
1621 	if (ifc->ifc_destroy == NULL)
1622 		return EOPNOTSUPP;
1623 
1624 	ifp = if_get(name, &psref);
1625 	if (ifp == NULL)
1626 		return ENXIO;
1627 
1628 	/* We have to disable ioctls here */
1629 	IFNET_LOCK(ifp);
1630 	if_ioctl = ifp->if_ioctl;
1631 	ifp->if_ioctl = if_nullioctl;
1632 	IFNET_UNLOCK(ifp);
1633 
1634 	/*
1635 	 * We cannot call ifc_destroy with holding ifp.
1636 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1637 	 */
1638 	if_put(ifp, &psref);
1639 
1640 	error = (*ifc->ifc_destroy)(ifp);
1641 
1642 	if (error != 0) {
1643 		/* We have to restore if_ioctl on error */
1644 		IFNET_LOCK(ifp);
1645 		ifp->if_ioctl = if_ioctl;
1646 		IFNET_UNLOCK(ifp);
1647 	}
1648 
1649 	return error;
1650 }
1651 
1652 static bool
1653 if_is_unit(const char *name)
1654 {
1655 
1656 	while (*name != '\0') {
1657 		if (*name < '0' || *name > '9')
1658 			return false;
1659 		name++;
1660 	}
1661 
1662 	return true;
1663 }
1664 
1665 /*
1666  * Look up a network interface cloner.
1667  */
1668 static struct if_clone *
1669 if_clone_lookup(const char *name, int *unitp)
1670 {
1671 	struct if_clone *ifc;
1672 	const char *cp;
1673 	char *dp, ifname[IFNAMSIZ + 3];
1674 	int unit;
1675 
1676 	KASSERT(mutex_owned(&if_clone_mtx));
1677 
1678 	strcpy(ifname, "if_");
1679 	/* separate interface name from unit */
1680 	/* TODO: search unit number from backward */
1681 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1682 	    *cp && !if_is_unit(cp);)
1683 		*dp++ = *cp++;
1684 
1685 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1686 		return NULL;	/* No name or unit number */
1687 	*dp++ = '\0';
1688 
1689 again:
1690 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1691 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1692 			break;
1693 	}
1694 
1695 	if (ifc == NULL) {
1696 		int error;
1697 		if (*ifname == '\0')
1698 			return NULL;
1699 		mutex_exit(&if_clone_mtx);
1700 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1701 		mutex_enter(&if_clone_mtx);
1702 		if (error)
1703 			return NULL;
1704 		*ifname = '\0';
1705 		goto again;
1706 	}
1707 
1708 	unit = 0;
1709 	while (cp - name < IFNAMSIZ && *cp) {
1710 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1711 			/* Bogus unit number. */
1712 			return NULL;
1713 		}
1714 		unit = (unit * 10) + (*cp++ - '0');
1715 	}
1716 
1717 	if (unitp != NULL)
1718 		*unitp = unit;
1719 	return ifc;
1720 }
1721 
1722 /*
1723  * Register a network interface cloner.
1724  */
1725 void
1726 if_clone_attach(struct if_clone *ifc)
1727 {
1728 
1729 	mutex_enter(&if_clone_mtx);
1730 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1731 	if_cloners_count++;
1732 	mutex_exit(&if_clone_mtx);
1733 }
1734 
1735 /*
1736  * Unregister a network interface cloner.
1737  */
1738 void
1739 if_clone_detach(struct if_clone *ifc)
1740 {
1741 
1742 	mutex_enter(&if_clone_mtx);
1743 	LIST_REMOVE(ifc, ifc_list);
1744 	if_cloners_count--;
1745 	mutex_exit(&if_clone_mtx);
1746 }
1747 
1748 /*
1749  * Provide list of interface cloners to userspace.
1750  */
1751 int
1752 if_clone_list(int buf_count, char *buffer, int *total)
1753 {
1754 	char outbuf[IFNAMSIZ], *dst;
1755 	struct if_clone *ifc;
1756 	int count, error = 0;
1757 
1758 	mutex_enter(&if_clone_mtx);
1759 	*total = if_cloners_count;
1760 	if ((dst = buffer) == NULL) {
1761 		/* Just asking how many there are. */
1762 		goto out;
1763 	}
1764 
1765 	if (buf_count < 0) {
1766 		error = EINVAL;
1767 		goto out;
1768 	}
1769 
1770 	count = (if_cloners_count < buf_count) ?
1771 	    if_cloners_count : buf_count;
1772 
1773 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1774 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1775 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1776 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1777 			error = ENAMETOOLONG;
1778 			goto out;
1779 		}
1780 		error = copyout(outbuf, dst, sizeof(outbuf));
1781 		if (error != 0)
1782 			break;
1783 	}
1784 
1785 out:
1786 	mutex_exit(&if_clone_mtx);
1787 	return error;
1788 }
1789 
1790 void
1791 ifa_psref_init(struct ifaddr *ifa)
1792 {
1793 
1794 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1795 }
1796 
1797 void
1798 ifaref(struct ifaddr *ifa)
1799 {
1800 
1801 	atomic_inc_uint(&ifa->ifa_refcnt);
1802 }
1803 
1804 void
1805 ifafree(struct ifaddr *ifa)
1806 {
1807 	KASSERT(ifa != NULL);
1808 	KASSERTMSG(ifa->ifa_refcnt > 0, "ifa_refcnt=%d", ifa->ifa_refcnt);
1809 
1810 	if (atomic_dec_uint_nv(&ifa->ifa_refcnt) == 0) {
1811 		free(ifa, M_IFADDR);
1812 	}
1813 }
1814 
1815 bool
1816 ifa_is_destroying(struct ifaddr *ifa)
1817 {
1818 
1819 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1820 }
1821 
1822 void
1823 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1824 {
1825 
1826 	ifa->ifa_ifp = ifp;
1827 
1828 	/*
1829 	 * Check MP-safety for IFEF_MPSAFE drivers.
1830 	 * Check !IFF_RUNNING for initialization routines that normally don't
1831 	 * take IFNET_LOCK but it's safe because there is no competitor.
1832 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1833 	 * if_stop.
1834 	 */
1835 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1836 	    IFNET_LOCKED(ifp));
1837 
1838 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1839 	IFADDR_ENTRY_INIT(ifa);
1840 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1841 
1842 	ifaref(ifa);
1843 }
1844 
1845 void
1846 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1847 {
1848 
1849 	KASSERT(ifa->ifa_ifp == ifp);
1850 	/*
1851 	 * Check MP-safety for IFEF_MPSAFE drivers.
1852 	 * if_is_deactivated indicates ifa_remove is called form if_detach
1853 	 * where is safe even if IFNET_LOCK isn't held.
1854 	 */
1855 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp));
1856 
1857 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1858 	IFADDR_WRITER_REMOVE(ifa);
1859 #ifdef NET_MPSAFE
1860 	IFNET_GLOBAL_LOCK();
1861 	pserialize_perform(ifnet_psz);
1862 	IFNET_GLOBAL_UNLOCK();
1863 #endif
1864 
1865 #ifdef NET_MPSAFE
1866 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1867 #endif
1868 	IFADDR_ENTRY_DESTROY(ifa);
1869 	ifafree(ifa);
1870 }
1871 
1872 void
1873 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1874 {
1875 
1876 	PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
1877 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1878 }
1879 
1880 void
1881 ifa_release(struct ifaddr *ifa, struct psref *psref)
1882 {
1883 
1884 	if (ifa == NULL)
1885 		return;
1886 
1887 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1888 }
1889 
1890 bool
1891 ifa_held(struct ifaddr *ifa)
1892 {
1893 
1894 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1895 }
1896 
1897 static inline int
1898 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1899 {
1900 	return sockaddr_cmp(sa1, sa2) == 0;
1901 }
1902 
1903 /*
1904  * Locate an interface based on a complete address.
1905  */
1906 /*ARGSUSED*/
1907 struct ifaddr *
1908 ifa_ifwithaddr(const struct sockaddr *addr)
1909 {
1910 	struct ifnet *ifp;
1911 	struct ifaddr *ifa;
1912 
1913 	IFNET_READER_FOREACH(ifp) {
1914 		if (if_is_deactivated(ifp))
1915 			continue;
1916 		IFADDR_READER_FOREACH(ifa, ifp) {
1917 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1918 				continue;
1919 			if (equal(addr, ifa->ifa_addr))
1920 				return ifa;
1921 			if ((ifp->if_flags & IFF_BROADCAST) &&
1922 			    ifa->ifa_broadaddr &&
1923 			    /* IP6 doesn't have broadcast */
1924 			    ifa->ifa_broadaddr->sa_len != 0 &&
1925 			    equal(ifa->ifa_broadaddr, addr))
1926 				return ifa;
1927 		}
1928 	}
1929 	return NULL;
1930 }
1931 
1932 struct ifaddr *
1933 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1934 {
1935 	struct ifaddr *ifa;
1936 	int s = pserialize_read_enter();
1937 
1938 	ifa = ifa_ifwithaddr(addr);
1939 	if (ifa != NULL)
1940 		ifa_acquire(ifa, psref);
1941 	pserialize_read_exit(s);
1942 
1943 	return ifa;
1944 }
1945 
1946 /*
1947  * Locate the point to point interface with a given destination address.
1948  */
1949 /*ARGSUSED*/
1950 struct ifaddr *
1951 ifa_ifwithdstaddr(const struct sockaddr *addr)
1952 {
1953 	struct ifnet *ifp;
1954 	struct ifaddr *ifa;
1955 
1956 	IFNET_READER_FOREACH(ifp) {
1957 		if (if_is_deactivated(ifp))
1958 			continue;
1959 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1960 			continue;
1961 		IFADDR_READER_FOREACH(ifa, ifp) {
1962 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1963 			    ifa->ifa_dstaddr == NULL)
1964 				continue;
1965 			if (equal(addr, ifa->ifa_dstaddr))
1966 				return ifa;
1967 		}
1968 	}
1969 
1970 	return NULL;
1971 }
1972 
1973 struct ifaddr *
1974 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1975 {
1976 	struct ifaddr *ifa;
1977 	int s;
1978 
1979 	s = pserialize_read_enter();
1980 	ifa = ifa_ifwithdstaddr(addr);
1981 	if (ifa != NULL)
1982 		ifa_acquire(ifa, psref);
1983 	pserialize_read_exit(s);
1984 
1985 	return ifa;
1986 }
1987 
1988 /*
1989  * Find an interface on a specific network.  If many, choice
1990  * is most specific found.
1991  */
1992 struct ifaddr *
1993 ifa_ifwithnet(const struct sockaddr *addr)
1994 {
1995 	struct ifnet *ifp;
1996 	struct ifaddr *ifa, *ifa_maybe = NULL;
1997 	const struct sockaddr_dl *sdl;
1998 	u_int af = addr->sa_family;
1999 	const char *addr_data = addr->sa_data, *cplim;
2000 
2001 	if (af == AF_LINK) {
2002 		sdl = satocsdl(addr);
2003 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
2004 		    ifindex2ifnet[sdl->sdl_index] &&
2005 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
2006 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
2007 		}
2008 	}
2009 #ifdef NETATALK
2010 	if (af == AF_APPLETALK) {
2011 		const struct sockaddr_at *sat, *sat2;
2012 		sat = (const struct sockaddr_at *)addr;
2013 		IFNET_READER_FOREACH(ifp) {
2014 			if (if_is_deactivated(ifp))
2015 				continue;
2016 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
2017 			if (ifa == NULL)
2018 				continue;
2019 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
2020 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
2021 				return ifa; /* exact match */
2022 			if (ifa_maybe == NULL) {
2023 				/* else keep the if with the right range */
2024 				ifa_maybe = ifa;
2025 			}
2026 		}
2027 		return ifa_maybe;
2028 	}
2029 #endif
2030 	IFNET_READER_FOREACH(ifp) {
2031 		if (if_is_deactivated(ifp))
2032 			continue;
2033 		IFADDR_READER_FOREACH(ifa, ifp) {
2034 			const char *cp, *cp2, *cp3;
2035 
2036 			if (ifa->ifa_addr->sa_family != af ||
2037 			    ifa->ifa_netmask == NULL)
2038  next:				continue;
2039 			cp = addr_data;
2040 			cp2 = ifa->ifa_addr->sa_data;
2041 			cp3 = ifa->ifa_netmask->sa_data;
2042 			cplim = (const char *)ifa->ifa_netmask +
2043 			    ifa->ifa_netmask->sa_len;
2044 			while (cp3 < cplim) {
2045 				if ((*cp++ ^ *cp2++) & *cp3++) {
2046 					/* want to continue for() loop */
2047 					goto next;
2048 				}
2049 			}
2050 			if (ifa_maybe == NULL ||
2051 			    rt_refines(ifa->ifa_netmask,
2052 			               ifa_maybe->ifa_netmask))
2053 				ifa_maybe = ifa;
2054 		}
2055 	}
2056 	return ifa_maybe;
2057 }
2058 
2059 struct ifaddr *
2060 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2061 {
2062 	struct ifaddr *ifa;
2063 	int s;
2064 
2065 	s = pserialize_read_enter();
2066 	ifa = ifa_ifwithnet(addr);
2067 	if (ifa != NULL)
2068 		ifa_acquire(ifa, psref);
2069 	pserialize_read_exit(s);
2070 
2071 	return ifa;
2072 }
2073 
2074 /*
2075  * Find the interface of the addresss.
2076  */
2077 struct ifaddr *
2078 ifa_ifwithladdr(const struct sockaddr *addr)
2079 {
2080 	struct ifaddr *ia;
2081 
2082 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2083 	    (ia = ifa_ifwithnet(addr)))
2084 		return ia;
2085 	return NULL;
2086 }
2087 
2088 struct ifaddr *
2089 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2090 {
2091 	struct ifaddr *ifa;
2092 	int s;
2093 
2094 	s = pserialize_read_enter();
2095 	ifa = ifa_ifwithladdr(addr);
2096 	if (ifa != NULL)
2097 		ifa_acquire(ifa, psref);
2098 	pserialize_read_exit(s);
2099 
2100 	return ifa;
2101 }
2102 
2103 /*
2104  * Find an interface using a specific address family
2105  */
2106 struct ifaddr *
2107 ifa_ifwithaf(int af)
2108 {
2109 	struct ifnet *ifp;
2110 	struct ifaddr *ifa = NULL;
2111 	int s;
2112 
2113 	s = pserialize_read_enter();
2114 	IFNET_READER_FOREACH(ifp) {
2115 		if (if_is_deactivated(ifp))
2116 			continue;
2117 		IFADDR_READER_FOREACH(ifa, ifp) {
2118 			if (ifa->ifa_addr->sa_family == af)
2119 				goto out;
2120 		}
2121 	}
2122 out:
2123 	pserialize_read_exit(s);
2124 	return ifa;
2125 }
2126 
2127 /*
2128  * Find an interface address specific to an interface best matching
2129  * a given address.
2130  */
2131 struct ifaddr *
2132 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2133 {
2134 	struct ifaddr *ifa;
2135 	const char *cp, *cp2, *cp3;
2136 	const char *cplim;
2137 	struct ifaddr *ifa_maybe = 0;
2138 	u_int af = addr->sa_family;
2139 
2140 	if (if_is_deactivated(ifp))
2141 		return NULL;
2142 
2143 	if (af >= AF_MAX)
2144 		return NULL;
2145 
2146 	IFADDR_READER_FOREACH(ifa, ifp) {
2147 		if (ifa->ifa_addr->sa_family != af)
2148 			continue;
2149 		ifa_maybe = ifa;
2150 		if (ifa->ifa_netmask == NULL) {
2151 			if (equal(addr, ifa->ifa_addr) ||
2152 			    (ifa->ifa_dstaddr &&
2153 			     equal(addr, ifa->ifa_dstaddr)))
2154 				return ifa;
2155 			continue;
2156 		}
2157 		cp = addr->sa_data;
2158 		cp2 = ifa->ifa_addr->sa_data;
2159 		cp3 = ifa->ifa_netmask->sa_data;
2160 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2161 		for (; cp3 < cplim; cp3++) {
2162 			if ((*cp++ ^ *cp2++) & *cp3)
2163 				break;
2164 		}
2165 		if (cp3 == cplim)
2166 			return ifa;
2167 	}
2168 	return ifa_maybe;
2169 }
2170 
2171 struct ifaddr *
2172 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2173     struct psref *psref)
2174 {
2175 	struct ifaddr *ifa;
2176 	int s;
2177 
2178 	s = pserialize_read_enter();
2179 	ifa = ifaof_ifpforaddr(addr, ifp);
2180 	if (ifa != NULL)
2181 		ifa_acquire(ifa, psref);
2182 	pserialize_read_exit(s);
2183 
2184 	return ifa;
2185 }
2186 
2187 /*
2188  * Default action when installing a route with a Link Level gateway.
2189  * Lookup an appropriate real ifa to point to.
2190  * This should be moved to /sys/net/link.c eventually.
2191  */
2192 void
2193 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2194 {
2195 	struct ifaddr *ifa;
2196 	const struct sockaddr *dst;
2197 	struct ifnet *ifp;
2198 	struct psref psref;
2199 
2200 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2201 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL ||
2202 	    ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2203 		return;
2204 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2205 		rt_replace_ifa(rt, ifa);
2206 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2207 			ifa->ifa_rtrequest(cmd, rt, info);
2208 		ifa_release(ifa, &psref);
2209 	}
2210 }
2211 
2212 /*
2213  * bitmask macros to manage a densely packed link_state change queue.
2214  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2215  * LINK_STATE_UP(2) we need 2 bits for each state change.
2216  * As a state change to store is 0, treat all bits set as an unset item.
2217  */
2218 #define LQ_ITEM_BITS		2
2219 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2220 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2221 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2222 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2223 #define LQ_STORE(q, i, v)						      \
2224 	do {								      \
2225 		(q) &= ~LQ_MASK((i));					      \
2226 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2227 	} while (0 /* CONSTCOND */)
2228 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2229 #define LQ_POP(q, v)							      \
2230 	do {								      \
2231 		(v) = LQ_ITEM((q), 0);					      \
2232 		(q) >>= LQ_ITEM_BITS;					      \
2233 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2234 	} while (0 /* CONSTCOND */)
2235 #define LQ_PUSH(q, v)							      \
2236 	do {								      \
2237 		(q) >>= LQ_ITEM_BITS;					      \
2238 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2239 	} while (0 /* CONSTCOND */)
2240 #define LQ_FIND_UNSET(q, i)						      \
2241 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2242 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2243 			break;						      \
2244 	}
2245 
2246 /*
2247  * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex
2248  * for each ifnet.  It doesn't matter because:
2249  * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on
2250  *   ifq_lock don't happen
2251  * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock
2252  *   because if_snd, if_link_state_change and if_link_state_change_softint
2253  *   are all called with KERNEL_LOCK
2254  */
2255 #define IF_LINK_STATE_CHANGE_LOCK(ifp)		\
2256 	mutex_enter((ifp)->if_snd.ifq_lock)
2257 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp)	\
2258 	mutex_exit((ifp)->if_snd.ifq_lock)
2259 
2260 /*
2261  * Handle a change in the interface link state and
2262  * queue notifications.
2263  */
2264 void
2265 if_link_state_change(struct ifnet *ifp, int link_state)
2266 {
2267 	int idx;
2268 
2269 	KASSERTMSG(if_is_link_state_changeable(ifp),
2270 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2271 	    ifp->if_xname, ifp->if_extflags);
2272 
2273 	/* Ensure change is to a valid state */
2274 	switch (link_state) {
2275 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2276 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2277 	case LINK_STATE_UP:
2278 		break;
2279 	default:
2280 #ifdef DEBUG
2281 		printf("%s: invalid link state %d\n",
2282 		    ifp->if_xname, link_state);
2283 #endif
2284 		return;
2285 	}
2286 
2287 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2288 
2289 	/* Find the last unset event in the queue. */
2290 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2291 
2292 	/*
2293 	 * Ensure link_state doesn't match the last event in the queue.
2294 	 * ifp->if_link_state is not checked and set here because
2295 	 * that would present an inconsistent picture to the system.
2296 	 */
2297 	if (idx != 0 &&
2298 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2299 		goto out;
2300 
2301 	/* Handle queue overflow. */
2302 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2303 		uint8_t lost;
2304 
2305 		/*
2306 		 * The DOWN state must be protected from being pushed off
2307 		 * the queue to ensure that userland will always be
2308 		 * in a sane state.
2309 		 * Because DOWN is protected, there is no need to protect
2310 		 * UNKNOWN.
2311 		 * It should be invalid to change from any other state to
2312 		 * UNKNOWN anyway ...
2313 		 */
2314 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2315 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2316 		if (lost == LINK_STATE_DOWN) {
2317 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2318 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2319 		}
2320 		printf("%s: lost link state change %s\n",
2321 		    ifp->if_xname,
2322 		    lost == LINK_STATE_UP ? "UP" :
2323 		    lost == LINK_STATE_DOWN ? "DOWN" :
2324 		    "UNKNOWN");
2325 	} else
2326 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2327 
2328 	softint_schedule(ifp->if_link_si);
2329 
2330 out:
2331 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2332 }
2333 
2334 /*
2335  * Handle interface link state change notifications.
2336  */
2337 void
2338 if_link_state_change_softint(struct ifnet *ifp, int link_state)
2339 {
2340 	struct domain *dp;
2341 	int s = splnet();
2342 	bool notify;
2343 
2344 	KASSERT(!cpu_intr_p());
2345 
2346 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2347 
2348 	/* Ensure the change is still valid. */
2349 	if (ifp->if_link_state == link_state) {
2350 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2351 		splx(s);
2352 		return;
2353 	}
2354 
2355 #ifdef DEBUG
2356 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2357 		link_state == LINK_STATE_UP ? "UP" :
2358 		link_state == LINK_STATE_DOWN ? "DOWN" :
2359 		"UNKNOWN",
2360 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2361 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2362 		"UNKNOWN");
2363 #endif
2364 
2365 	/*
2366 	 * When going from UNKNOWN to UP, we need to mark existing
2367 	 * addresses as tentative and restart DAD as we may have
2368 	 * erroneously not found a duplicate.
2369 	 *
2370 	 * This needs to happen before rt_ifmsg to avoid a race where
2371 	 * listeners would have an address and expect it to work right
2372 	 * away.
2373 	 */
2374 	notify = (link_state == LINK_STATE_UP &&
2375 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2376 	ifp->if_link_state = link_state;
2377 	/* The following routines may sleep so release the spin mutex */
2378 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2379 
2380 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2381 	if (notify) {
2382 		DOMAIN_FOREACH(dp) {
2383 			if (dp->dom_if_link_state_change != NULL)
2384 				dp->dom_if_link_state_change(ifp,
2385 				    LINK_STATE_DOWN);
2386 		}
2387 	}
2388 
2389 	/* Notify that the link state has changed. */
2390 	rt_ifmsg(ifp);
2391 
2392 #if NCARP > 0
2393 	if (ifp->if_carp)
2394 		carp_carpdev_state(ifp);
2395 #endif
2396 
2397 	DOMAIN_FOREACH(dp) {
2398 		if (dp->dom_if_link_state_change != NULL)
2399 			dp->dom_if_link_state_change(ifp, link_state);
2400 	}
2401 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2402 	splx(s);
2403 }
2404 
2405 /*
2406  * Process the interface link state change queue.
2407  */
2408 static void
2409 if_link_state_change_si(void *arg)
2410 {
2411 	struct ifnet *ifp = arg;
2412 	int s;
2413 	uint8_t state;
2414 	bool schedule;
2415 
2416 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2417 	s = splnet();
2418 
2419 	/* Pop a link state change from the queue and process it. */
2420 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2421 	LQ_POP(ifp->if_link_queue, state);
2422 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2423 
2424 	if_link_state_change_softint(ifp, state);
2425 
2426 	/* If there is a link state change to come, schedule it. */
2427 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2428 	schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET);
2429 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2430 	if (schedule)
2431 		softint_schedule(ifp->if_link_si);
2432 
2433 	splx(s);
2434 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2435 }
2436 
2437 /*
2438  * Default action when installing a local route on a point-to-point
2439  * interface.
2440  */
2441 void
2442 p2p_rtrequest(int req, struct rtentry *rt,
2443     __unused const struct rt_addrinfo *info)
2444 {
2445 	struct ifnet *ifp = rt->rt_ifp;
2446 	struct ifaddr *ifa, *lo0ifa;
2447 	int s = pserialize_read_enter();
2448 
2449 	switch (req) {
2450 	case RTM_ADD:
2451 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2452 			break;
2453 
2454 		rt->rt_ifp = lo0ifp;
2455 
2456 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2457 			break;
2458 
2459 		IFADDR_READER_FOREACH(ifa, ifp) {
2460 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2461 				break;
2462 		}
2463 		if (ifa == NULL)
2464 			break;
2465 
2466 		/*
2467 		 * Ensure lo0 has an address of the same family.
2468 		 */
2469 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2470 			if (lo0ifa->ifa_addr->sa_family ==
2471 			    ifa->ifa_addr->sa_family)
2472 				break;
2473 		}
2474 		if (lo0ifa == NULL)
2475 			break;
2476 
2477 		/*
2478 		 * Make sure to set rt->rt_ifa to the interface
2479 		 * address we are using, otherwise we will have trouble
2480 		 * with source address selection.
2481 		 */
2482 		if (ifa != rt->rt_ifa)
2483 			rt_replace_ifa(rt, ifa);
2484 		break;
2485 	case RTM_DELETE:
2486 	default:
2487 		break;
2488 	}
2489 	pserialize_read_exit(s);
2490 }
2491 
2492 static void
2493 _if_down(struct ifnet *ifp)
2494 {
2495 	struct ifaddr *ifa;
2496 	struct domain *dp;
2497 	int s, bound;
2498 	struct psref psref;
2499 
2500 	ifp->if_flags &= ~IFF_UP;
2501 	nanotime(&ifp->if_lastchange);
2502 
2503 	bound = curlwp_bind();
2504 	s = pserialize_read_enter();
2505 	IFADDR_READER_FOREACH(ifa, ifp) {
2506 		ifa_acquire(ifa, &psref);
2507 		pserialize_read_exit(s);
2508 
2509 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2510 
2511 		s = pserialize_read_enter();
2512 		ifa_release(ifa, &psref);
2513 	}
2514 	pserialize_read_exit(s);
2515 	curlwp_bindx(bound);
2516 
2517 	IFQ_PURGE(&ifp->if_snd);
2518 #if NCARP > 0
2519 	if (ifp->if_carp)
2520 		carp_carpdev_state(ifp);
2521 #endif
2522 	rt_ifmsg(ifp);
2523 	DOMAIN_FOREACH(dp) {
2524 		if (dp->dom_if_down)
2525 			dp->dom_if_down(ifp);
2526 	}
2527 }
2528 
2529 static void
2530 if_down_deactivated(struct ifnet *ifp)
2531 {
2532 
2533 	KASSERT(if_is_deactivated(ifp));
2534 	_if_down(ifp);
2535 }
2536 
2537 void
2538 if_down_locked(struct ifnet *ifp)
2539 {
2540 
2541 	KASSERT(IFNET_LOCKED(ifp));
2542 	_if_down(ifp);
2543 }
2544 
2545 /*
2546  * Mark an interface down and notify protocols of
2547  * the transition.
2548  * NOTE: must be called at splsoftnet or equivalent.
2549  */
2550 void
2551 if_down(struct ifnet *ifp)
2552 {
2553 
2554 	IFNET_LOCK(ifp);
2555 	if_down_locked(ifp);
2556 	IFNET_UNLOCK(ifp);
2557 }
2558 
2559 /*
2560  * Must be called with holding if_ioctl_lock.
2561  */
2562 static void
2563 if_up_locked(struct ifnet *ifp)
2564 {
2565 #ifdef notyet
2566 	struct ifaddr *ifa;
2567 #endif
2568 	struct domain *dp;
2569 
2570 	KASSERT(IFNET_LOCKED(ifp));
2571 
2572 	KASSERT(!if_is_deactivated(ifp));
2573 	ifp->if_flags |= IFF_UP;
2574 	nanotime(&ifp->if_lastchange);
2575 #ifdef notyet
2576 	/* this has no effect on IP, and will kill all ISO connections XXX */
2577 	IFADDR_READER_FOREACH(ifa, ifp)
2578 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2579 #endif
2580 #if NCARP > 0
2581 	if (ifp->if_carp)
2582 		carp_carpdev_state(ifp);
2583 #endif
2584 	rt_ifmsg(ifp);
2585 	DOMAIN_FOREACH(dp) {
2586 		if (dp->dom_if_up)
2587 			dp->dom_if_up(ifp);
2588 	}
2589 }
2590 
2591 /*
2592  * Handle interface slowtimo timer routine.  Called
2593  * from softclock, we decrement timer (if set) and
2594  * call the appropriate interface routine on expiration.
2595  */
2596 static void
2597 if_slowtimo(void *arg)
2598 {
2599 	void (*slowtimo)(struct ifnet *);
2600 	struct ifnet *ifp = arg;
2601 	int s;
2602 
2603 	slowtimo = ifp->if_slowtimo;
2604 	if (__predict_false(slowtimo == NULL))
2605 		return;
2606 
2607 	s = splnet();
2608 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2609 		(*slowtimo)(ifp);
2610 
2611 	splx(s);
2612 
2613 	if (__predict_true(ifp->if_slowtimo != NULL))
2614 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2615 }
2616 
2617 /*
2618  * Mark an interface up and notify protocols of
2619  * the transition.
2620  * NOTE: must be called at splsoftnet or equivalent.
2621  */
2622 void
2623 if_up(struct ifnet *ifp)
2624 {
2625 
2626 	IFNET_LOCK(ifp);
2627 	if_up_locked(ifp);
2628 	IFNET_UNLOCK(ifp);
2629 }
2630 
2631 /*
2632  * Set/clear promiscuous mode on interface ifp based on the truth value
2633  * of pswitch.  The calls are reference counted so that only the first
2634  * "on" request actually has an effect, as does the final "off" request.
2635  * Results are undefined if the "off" and "on" requests are not matched.
2636  */
2637 int
2638 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2639 {
2640 	int pcount, ret = 0;
2641 	u_short nflags;
2642 
2643 	KASSERT(IFNET_LOCKED(ifp));
2644 
2645 	pcount = ifp->if_pcount;
2646 	if (pswitch) {
2647 		/*
2648 		 * Allow the device to be "placed" into promiscuous
2649 		 * mode even if it is not configured up.  It will
2650 		 * consult IFF_PROMISC when it is brought up.
2651 		 */
2652 		if (ifp->if_pcount++ != 0)
2653 			goto out;
2654 		nflags = ifp->if_flags | IFF_PROMISC;
2655 	} else {
2656 		if (--ifp->if_pcount > 0)
2657 			goto out;
2658 		nflags = ifp->if_flags & ~IFF_PROMISC;
2659 	}
2660 	ret = if_flags_set(ifp, nflags);
2661 	/* Restore interface state if not successful. */
2662 	if (ret != 0) {
2663 		ifp->if_pcount = pcount;
2664 	}
2665 out:
2666 	return ret;
2667 }
2668 
2669 int
2670 ifpromisc(struct ifnet *ifp, int pswitch)
2671 {
2672 	int e;
2673 
2674 	IFNET_LOCK(ifp);
2675 	e = ifpromisc_locked(ifp, pswitch);
2676 	IFNET_UNLOCK(ifp);
2677 
2678 	return e;
2679 }
2680 
2681 /*
2682  * Map interface name to
2683  * interface structure pointer.
2684  */
2685 struct ifnet *
2686 ifunit(const char *name)
2687 {
2688 	struct ifnet *ifp;
2689 	const char *cp = name;
2690 	u_int unit = 0;
2691 	u_int i;
2692 	int s;
2693 
2694 	/*
2695 	 * If the entire name is a number, treat it as an ifindex.
2696 	 */
2697 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2698 		unit = unit * 10 + (*cp - '0');
2699 	}
2700 
2701 	/*
2702 	 * If the number took all of the name, then it's a valid ifindex.
2703 	 */
2704 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2705 		return if_byindex(unit);
2706 
2707 	ifp = NULL;
2708 	s = pserialize_read_enter();
2709 	IFNET_READER_FOREACH(ifp) {
2710 		if (if_is_deactivated(ifp))
2711 			continue;
2712 	 	if (strcmp(ifp->if_xname, name) == 0)
2713 			goto out;
2714 	}
2715 out:
2716 	pserialize_read_exit(s);
2717 	return ifp;
2718 }
2719 
2720 /*
2721  * Get a reference of an ifnet object by an interface name.
2722  * The returned reference is protected by psref(9). The caller
2723  * must release a returned reference by if_put after use.
2724  */
2725 struct ifnet *
2726 if_get(const char *name, struct psref *psref)
2727 {
2728 	struct ifnet *ifp;
2729 	const char *cp = name;
2730 	u_int unit = 0;
2731 	u_int i;
2732 	int s;
2733 
2734 	/*
2735 	 * If the entire name is a number, treat it as an ifindex.
2736 	 */
2737 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2738 		unit = unit * 10 + (*cp - '0');
2739 	}
2740 
2741 	/*
2742 	 * If the number took all of the name, then it's a valid ifindex.
2743 	 */
2744 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2745 		return if_get_byindex(unit, psref);
2746 
2747 	ifp = NULL;
2748 	s = pserialize_read_enter();
2749 	IFNET_READER_FOREACH(ifp) {
2750 		if (if_is_deactivated(ifp))
2751 			continue;
2752 		if (strcmp(ifp->if_xname, name) == 0) {
2753 			PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2754 			psref_acquire(psref, &ifp->if_psref,
2755 			    ifnet_psref_class);
2756 			goto out;
2757 		}
2758 	}
2759 out:
2760 	pserialize_read_exit(s);
2761 	return ifp;
2762 }
2763 
2764 /*
2765  * Release a reference of an ifnet object given by if_get, if_get_byindex
2766  * or if_get_bylla.
2767  */
2768 void
2769 if_put(const struct ifnet *ifp, struct psref *psref)
2770 {
2771 
2772 	if (ifp == NULL)
2773 		return;
2774 
2775 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2776 }
2777 
2778 /*
2779  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2780  * should be used.
2781  */
2782 ifnet_t *
2783 _if_byindex(u_int idx)
2784 {
2785 
2786 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2787 }
2788 
2789 /*
2790  * Return ifp having idx. Return NULL if not found or the found ifp is
2791  * already deactivated.
2792  */
2793 ifnet_t *
2794 if_byindex(u_int idx)
2795 {
2796 	ifnet_t *ifp;
2797 
2798 	ifp = _if_byindex(idx);
2799 	if (ifp != NULL && if_is_deactivated(ifp))
2800 		ifp = NULL;
2801 	return ifp;
2802 }
2803 
2804 /*
2805  * Get a reference of an ifnet object by an interface index.
2806  * The returned reference is protected by psref(9). The caller
2807  * must release a returned reference by if_put after use.
2808  */
2809 ifnet_t *
2810 if_get_byindex(u_int idx, struct psref *psref)
2811 {
2812 	ifnet_t *ifp;
2813 	int s;
2814 
2815 	s = pserialize_read_enter();
2816 	ifp = if_byindex(idx);
2817 	if (__predict_true(ifp != NULL)) {
2818 		PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2819 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2820 	}
2821 	pserialize_read_exit(s);
2822 
2823 	return ifp;
2824 }
2825 
2826 ifnet_t *
2827 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2828 {
2829 	ifnet_t *ifp;
2830 	int s;
2831 
2832 	s = pserialize_read_enter();
2833 	IFNET_READER_FOREACH(ifp) {
2834 		if (if_is_deactivated(ifp))
2835 			continue;
2836 		if (ifp->if_addrlen != lla_len)
2837 			continue;
2838 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2839 			psref_acquire(psref, &ifp->if_psref,
2840 			    ifnet_psref_class);
2841 			break;
2842 		}
2843 	}
2844 	pserialize_read_exit(s);
2845 
2846 	return ifp;
2847 }
2848 
2849 /*
2850  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2851  * for example using pserialize or the ifp is already held or some other
2852  * object is held which guarantes the ifp to not be freed indirectly.
2853  */
2854 void
2855 if_acquire(struct ifnet *ifp, struct psref *psref)
2856 {
2857 
2858 	KASSERT(ifp->if_index != 0);
2859 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2860 }
2861 
2862 bool
2863 if_held(struct ifnet *ifp)
2864 {
2865 
2866 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2867 }
2868 
2869 /*
2870  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4.
2871  * Check the tunnel nesting count.
2872  * Return > 0, if tunnel nesting count is more than limit.
2873  * Return 0, if tunnel nesting count is equal or less than limit.
2874  */
2875 int
2876 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
2877 {
2878 	struct m_tag *mtag;
2879 	int *count;
2880 
2881 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO);
2882 	if (mtag != NULL) {
2883 		count = (int *)(mtag + 1);
2884 		if (++(*count) > limit) {
2885 			log(LOG_NOTICE,
2886 			    "%s: recursively called too many times(%d)\n",
2887 			    ifp->if_xname, *count);
2888 			return EIO;
2889 		}
2890 	} else {
2891 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
2892 		    M_NOWAIT);
2893 		if (mtag != NULL) {
2894 			m_tag_prepend(m, mtag);
2895 			count = (int *)(mtag + 1);
2896 			*count = 0;
2897 		} else {
2898 			log(LOG_DEBUG,
2899 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
2900 			    ifp->if_xname);
2901 		}
2902 	}
2903 
2904 	return 0;
2905 }
2906 
2907 static void
2908 if_tunnel_ro_init_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2909 {
2910 	struct tunnel_ro *tro = p;
2911 
2912 	tro->tr_ro = kmem_zalloc(sizeof(*tro->tr_ro), KM_SLEEP);
2913 	tro->tr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2914 }
2915 
2916 percpu_t *
2917 if_tunnel_alloc_ro_percpu(void)
2918 {
2919 	percpu_t *ro_percpu;
2920 
2921 	ro_percpu = percpu_alloc(sizeof(struct tunnel_ro));
2922 	percpu_foreach(ro_percpu, if_tunnel_ro_init_pc, NULL);
2923 
2924 	return ro_percpu;
2925 }
2926 
2927 static void
2928 if_tunnel_ro_fini_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2929 {
2930 	struct tunnel_ro *tro = p;
2931 
2932 	rtcache_free(tro->tr_ro);
2933 	kmem_free(tro->tr_ro, sizeof(*tro->tr_ro));
2934 
2935 	mutex_obj_free(tro->tr_lock);
2936 }
2937 
2938 void
2939 if_tunnel_free_ro_percpu(percpu_t *ro_percpu)
2940 {
2941 
2942 	percpu_foreach(ro_percpu, if_tunnel_ro_fini_pc, NULL);
2943 	percpu_free(ro_percpu, sizeof(struct tunnel_ro));
2944 }
2945 
2946 
2947 static void
2948 if_tunnel_rtcache_free_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2949 {
2950 	struct tunnel_ro *tro = p;
2951 
2952 	mutex_enter(tro->tr_lock);
2953 	rtcache_free(tro->tr_ro);
2954 	mutex_exit(tro->tr_lock);
2955 }
2956 
2957 void if_tunnel_ro_percpu_rtcache_free(percpu_t *ro_percpu)
2958 {
2959 
2960 	percpu_foreach(ro_percpu, if_tunnel_rtcache_free_pc, NULL);
2961 }
2962 
2963 
2964 /* common */
2965 int
2966 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2967 {
2968 	int s;
2969 	struct ifreq *ifr;
2970 	struct ifcapreq *ifcr;
2971 	struct ifdatareq *ifdr;
2972 	unsigned short flags;
2973 
2974 	switch (cmd) {
2975 	case SIOCSIFCAP:
2976 		ifcr = data;
2977 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2978 			return EINVAL;
2979 
2980 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2981 			return 0;
2982 
2983 		ifp->if_capenable = ifcr->ifcr_capenable;
2984 
2985 		/* Pre-compute the checksum flags mask. */
2986 		ifp->if_csum_flags_tx = 0;
2987 		ifp->if_csum_flags_rx = 0;
2988 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx)
2989 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2990 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
2991 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2992 
2993 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx)
2994 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2995 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
2996 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2997 
2998 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx)
2999 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
3000 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
3001 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
3002 
3003 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx)
3004 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
3005 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
3006 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
3007 
3008 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx)
3009 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
3010 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
3011 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
3012 
3013 		if (ifp->if_capenable & IFCAP_TSOv4)
3014 			ifp->if_csum_flags_tx |= M_CSUM_TSOv4;
3015 		if (ifp->if_capenable & IFCAP_TSOv6)
3016 			ifp->if_csum_flags_tx |= M_CSUM_TSOv6;
3017 
3018 #if NBRIDGE > 0
3019 		if (ifp->if_bridge != NULL)
3020 			bridge_calc_csum_flags(ifp->if_bridge);
3021 #endif
3022 
3023 		if (ifp->if_flags & IFF_UP)
3024 			return ENETRESET;
3025 		return 0;
3026 	case SIOCSIFFLAGS:
3027 		ifr = data;
3028 		/*
3029 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
3030 		 * and if_down aren't MP-safe yet, so we must hold the lock.
3031 		 */
3032 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3033 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
3034 			s = splsoftnet();
3035 			if_down_locked(ifp);
3036 			splx(s);
3037 		}
3038 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
3039 			s = splsoftnet();
3040 			if_up_locked(ifp);
3041 			splx(s);
3042 		}
3043 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3044 		flags = (ifp->if_flags & IFF_CANTCHANGE) |
3045 		    (ifr->ifr_flags &~ IFF_CANTCHANGE);
3046 		if (ifp->if_flags != flags) {
3047 			ifp->if_flags = flags;
3048 			/* Notify that the flags have changed. */
3049 			rt_ifmsg(ifp);
3050 		}
3051 		break;
3052 	case SIOCGIFFLAGS:
3053 		ifr = data;
3054 		ifr->ifr_flags = ifp->if_flags;
3055 		break;
3056 
3057 	case SIOCGIFMETRIC:
3058 		ifr = data;
3059 		ifr->ifr_metric = ifp->if_metric;
3060 		break;
3061 
3062 	case SIOCGIFMTU:
3063 		ifr = data;
3064 		ifr->ifr_mtu = ifp->if_mtu;
3065 		break;
3066 
3067 	case SIOCGIFDLT:
3068 		ifr = data;
3069 		ifr->ifr_dlt = ifp->if_dlt;
3070 		break;
3071 
3072 	case SIOCGIFCAP:
3073 		ifcr = data;
3074 		ifcr->ifcr_capabilities = ifp->if_capabilities;
3075 		ifcr->ifcr_capenable = ifp->if_capenable;
3076 		break;
3077 
3078 	case SIOCSIFMETRIC:
3079 		ifr = data;
3080 		ifp->if_metric = ifr->ifr_metric;
3081 		break;
3082 
3083 	case SIOCGIFDATA:
3084 		ifdr = data;
3085 		ifdr->ifdr_data = ifp->if_data;
3086 		break;
3087 
3088 	case SIOCGIFINDEX:
3089 		ifr = data;
3090 		ifr->ifr_index = ifp->if_index;
3091 		break;
3092 
3093 	case SIOCZIFDATA:
3094 		ifdr = data;
3095 		ifdr->ifdr_data = ifp->if_data;
3096 		/*
3097 		 * Assumes that the volatile counters that can be
3098 		 * zero'ed are at the end of if_data.
3099 		 */
3100 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
3101 		    offsetof(struct if_data, ifi_ipackets));
3102 		/*
3103 		 * The memset() clears to the bottm of if_data. In the area,
3104 		 * if_lastchange is included. Please be careful if new entry
3105 		 * will be added into if_data or rewite this.
3106 		 *
3107 		 * And also, update if_lastchnage.
3108 		 */
3109 		getnanotime(&ifp->if_lastchange);
3110 		break;
3111 	case SIOCSIFMTU:
3112 		ifr = data;
3113 		if (ifp->if_mtu == ifr->ifr_mtu)
3114 			break;
3115 		ifp->if_mtu = ifr->ifr_mtu;
3116 		/*
3117 		 * If the link MTU changed, do network layer specific procedure.
3118 		 */
3119 #ifdef INET6
3120 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3121 		if (in6_present)
3122 			nd6_setmtu(ifp);
3123 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3124 #endif
3125 		return ENETRESET;
3126 	case SIOCSIFDESCR:
3127 		{
3128 			char *descrbuf;
3129 
3130  			ifr = data;
3131 
3132  			if (ifr->ifr_buflen > IFDESCRSIZE)
3133 				return ENAMETOOLONG;
3134 
3135  			if (ifr->ifr_buf == NULL || ifr->ifr_buflen == 0) {
3136 				/* unset description */
3137 				descrbuf = NULL;
3138 			} else {
3139 				int error;
3140 
3141  				descrbuf = kmem_zalloc(IFDESCRSIZE, KM_SLEEP);
3142 				/* copy (IFDESCRSIZE - 1) bytes to ensure terminating nul */
3143 				error = copyin(ifr->ifr_buf, descrbuf, IFDESCRSIZE - 1);
3144 				if (error) {
3145 					kmem_free(descrbuf, IFDESCRSIZE);
3146 					return error;
3147 				}
3148 			}
3149 
3150  			if (ifp->if_description != NULL)
3151 				kmem_free(ifp->if_description, IFDESCRSIZE);
3152 
3153  			ifp->if_description = descrbuf;
3154 		}
3155 		break;
3156 
3157  	case SIOCGIFDESCR:
3158 		{
3159 			char *descr;
3160 
3161  			ifr = data;
3162 			descr = ifp->if_description;
3163 
3164  			if (descr == NULL)
3165 				return ENOMSG;
3166 
3167  			if (ifr->ifr_buflen < IFDESCRSIZE)
3168 				return EINVAL;
3169 			else {
3170 				int error;
3171 				error = copyout(descr, ifr->ifr_buf, IFDESCRSIZE);
3172 				if (error)
3173 					return error;
3174 			}
3175 		}
3176  		break;
3177 
3178 	default:
3179 		return ENOTTY;
3180 	}
3181 	return 0;
3182 }
3183 
3184 int
3185 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3186 {
3187 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3188 	struct ifaddr *ifa;
3189 	const struct sockaddr *any, *sa;
3190 	union {
3191 		struct sockaddr sa;
3192 		struct sockaddr_storage ss;
3193 	} u, v;
3194 	int s, error = 0;
3195 
3196 	switch (cmd) {
3197 	case SIOCSIFADDRPREF:
3198 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
3199 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
3200 		    NULL) != 0)
3201 			return EPERM;
3202 	case SIOCGIFADDRPREF:
3203 		break;
3204 	default:
3205 		return EOPNOTSUPP;
3206 	}
3207 
3208 	/* sanity checks */
3209 	if (data == NULL || ifp == NULL) {
3210 		panic("invalid argument to %s", __func__);
3211 		/*NOTREACHED*/
3212 	}
3213 
3214 	/* address must be specified on ADD and DELETE */
3215 	sa = sstocsa(&ifap->ifap_addr);
3216 	if (sa->sa_family != sofamily(so))
3217 		return EINVAL;
3218 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3219 		return EINVAL;
3220 
3221 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3222 
3223 	s = pserialize_read_enter();
3224 	IFADDR_READER_FOREACH(ifa, ifp) {
3225 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3226 			continue;
3227 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3228 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3229 			break;
3230 	}
3231 	if (ifa == NULL) {
3232 		error = EADDRNOTAVAIL;
3233 		goto out;
3234 	}
3235 
3236 	switch (cmd) {
3237 	case SIOCSIFADDRPREF:
3238 		ifa->ifa_preference = ifap->ifap_preference;
3239 		goto out;
3240 	case SIOCGIFADDRPREF:
3241 		/* fill in the if_laddrreq structure */
3242 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3243 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3244 		ifap->ifap_preference = ifa->ifa_preference;
3245 		goto out;
3246 	default:
3247 		error = EOPNOTSUPP;
3248 	}
3249 out:
3250 	pserialize_read_exit(s);
3251 	return error;
3252 }
3253 
3254 /*
3255  * Interface ioctls.
3256  */
3257 static int
3258 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3259 {
3260 	struct ifnet *ifp;
3261 	struct ifreq *ifr;
3262 	int error = 0;
3263 	u_long ocmd = cmd;
3264 	u_short oif_flags;
3265 	struct ifreq ifrb;
3266 	struct oifreq *oifr = NULL;
3267 	int r;
3268 	struct psref psref;
3269 	int bound;
3270 	bool do_if43_post = false;
3271 	bool do_ifm80_post = false;
3272 
3273 	switch (cmd) {
3274 	case SIOCGIFCONF:
3275 		return ifconf(cmd, data);
3276 	case SIOCINITIFADDR:
3277 		return EPERM;
3278 	default:
3279 		MODULE_HOOK_CALL(uipc_syscalls_40_hook, (cmd, data), enosys(),
3280 		    error);
3281 		if (error != ENOSYS)
3282 			return error;
3283 		MODULE_HOOK_CALL(uipc_syscalls_50_hook, (l, cmd, data),
3284 		    enosys(), error);
3285 		if (error != ENOSYS)
3286 			return error;
3287 		error = 0;
3288 		break;
3289 	}
3290 
3291 	ifr = data;
3292 	/* Pre-conversion */
3293 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), error);
3294 	if (cmd != ocmd) {
3295 		oifr = data;
3296 		data = ifr = &ifrb;
3297 		IFREQO2N_43(oifr, ifr);
3298 		do_if43_post = true;
3299 	}
3300 	MODULE_HOOK_CALL(ifmedia_80_pre_hook, (ifr, &cmd, &do_ifm80_post),
3301 	    enosys(), error);
3302 
3303 	switch (cmd) {
3304 	case SIOCIFCREATE:
3305 	case SIOCIFDESTROY:
3306 		bound = curlwp_bind();
3307 		if (l != NULL) {
3308 			ifp = if_get(ifr->ifr_name, &psref);
3309 			error = kauth_authorize_network(l->l_cred,
3310 			    KAUTH_NETWORK_INTERFACE,
3311 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3312 			    (void *)cmd, NULL);
3313 			if (ifp != NULL)
3314 				if_put(ifp, &psref);
3315 			if (error != 0) {
3316 				curlwp_bindx(bound);
3317 				return error;
3318 			}
3319 		}
3320 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3321 		mutex_enter(&if_clone_mtx);
3322 		r = (cmd == SIOCIFCREATE) ?
3323 			if_clone_create(ifr->ifr_name) :
3324 			if_clone_destroy(ifr->ifr_name);
3325 		mutex_exit(&if_clone_mtx);
3326 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3327 		curlwp_bindx(bound);
3328 		return r;
3329 
3330 	case SIOCIFGCLONERS:
3331 		{
3332 			struct if_clonereq *req = (struct if_clonereq *)data;
3333 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3334 			    &req->ifcr_total);
3335 		}
3336 	}
3337 
3338 	bound = curlwp_bind();
3339 	ifp = if_get(ifr->ifr_name, &psref);
3340 	if (ifp == NULL) {
3341 		curlwp_bindx(bound);
3342 		return ENXIO;
3343 	}
3344 
3345 	switch (cmd) {
3346 	case SIOCALIFADDR:
3347 	case SIOCDLIFADDR:
3348 	case SIOCSIFADDRPREF:
3349 	case SIOCSIFFLAGS:
3350 	case SIOCSIFCAP:
3351 	case SIOCSIFMETRIC:
3352 	case SIOCZIFDATA:
3353 	case SIOCSIFMTU:
3354 	case SIOCSIFPHYADDR:
3355 	case SIOCDIFPHYADDR:
3356 #ifdef INET6
3357 	case SIOCSIFPHYADDR_IN6:
3358 #endif
3359 	case SIOCSLIFPHYADDR:
3360 	case SIOCADDMULTI:
3361 	case SIOCDELMULTI:
3362 	case SIOCSETHERCAP:
3363 	case SIOCSIFMEDIA:
3364 	case SIOCSDRVSPEC:
3365 	case SIOCG80211:
3366 	case SIOCS80211:
3367 	case SIOCS80211NWID:
3368 	case SIOCS80211NWKEY:
3369 	case SIOCS80211POWER:
3370 	case SIOCS80211BSSID:
3371 	case SIOCS80211CHANNEL:
3372 	case SIOCSLINKSTR:
3373 		if (l != NULL) {
3374 			error = kauth_authorize_network(l->l_cred,
3375 			    KAUTH_NETWORK_INTERFACE,
3376 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3377 			    (void *)cmd, NULL);
3378 			if (error != 0)
3379 				goto out;
3380 		}
3381 	}
3382 
3383 	oif_flags = ifp->if_flags;
3384 
3385 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3386 	IFNET_LOCK(ifp);
3387 
3388 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3389 	if (error != ENOTTY)
3390 		;
3391 	else if (so->so_proto == NULL)
3392 		error = EOPNOTSUPP;
3393 	else {
3394 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3395 		MODULE_HOOK_CALL(if_ifioctl_43_hook,
3396 			     (so, ocmd, cmd, data, l), enosys(), error);
3397 		if (error == ENOSYS)
3398 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3399 			    cmd, data, ifp);
3400 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3401 	}
3402 
3403 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3404 		if ((ifp->if_flags & IFF_UP) != 0) {
3405 			int s = splsoftnet();
3406 			if_up_locked(ifp);
3407 			splx(s);
3408 		}
3409 	}
3410 
3411 	/* Post-conversion */
3412 	if (do_ifm80_post && (error == 0))
3413 		MODULE_HOOK_CALL(ifmedia_80_post_hook, (ifr, cmd),
3414 		    enosys(), error);
3415 	if (do_if43_post)
3416 		IFREQN2O_43(oifr, ifr);
3417 
3418 	IFNET_UNLOCK(ifp);
3419 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3420 out:
3421 	if_put(ifp, &psref);
3422 	curlwp_bindx(bound);
3423 	return error;
3424 }
3425 
3426 /*
3427  * Return interface configuration
3428  * of system.  List may be used
3429  * in later ioctl's (above) to get
3430  * other information.
3431  *
3432  * Each record is a struct ifreq.  Before the addition of
3433  * sockaddr_storage, the API rule was that sockaddr flavors that did
3434  * not fit would extend beyond the struct ifreq, with the next struct
3435  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3436  * union in struct ifreq includes struct sockaddr_storage, every kind
3437  * of sockaddr must fit.  Thus, there are no longer any overlength
3438  * records.
3439  *
3440  * Records are added to the user buffer if they fit, and ifc_len is
3441  * adjusted to the length that was written.  Thus, the user is only
3442  * assured of getting the complete list if ifc_len on return is at
3443  * least sizeof(struct ifreq) less than it was on entry.
3444  *
3445  * If the user buffer pointer is NULL, this routine copies no data and
3446  * returns the amount of space that would be needed.
3447  *
3448  * Invariants:
3449  * ifrp points to the next part of the user's buffer to be used.  If
3450  * ifrp != NULL, space holds the number of bytes remaining that we may
3451  * write at ifrp.  Otherwise, space holds the number of bytes that
3452  * would have been written had there been adequate space.
3453  */
3454 /*ARGSUSED*/
3455 static int
3456 ifconf(u_long cmd, void *data)
3457 {
3458 	struct ifconf *ifc = (struct ifconf *)data;
3459 	struct ifnet *ifp;
3460 	struct ifaddr *ifa;
3461 	struct ifreq ifr, *ifrp = NULL;
3462 	int space = 0, error = 0;
3463 	const int sz = (int)sizeof(struct ifreq);
3464 	const bool docopy = ifc->ifc_req != NULL;
3465 	int s;
3466 	int bound;
3467 	struct psref psref;
3468 
3469 	memset(&ifr, 0, sizeof(ifr));
3470 	if (docopy) {
3471 		space = ifc->ifc_len;
3472 		ifrp = ifc->ifc_req;
3473 	}
3474 
3475 	bound = curlwp_bind();
3476 	s = pserialize_read_enter();
3477 	IFNET_READER_FOREACH(ifp) {
3478 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3479 		pserialize_read_exit(s);
3480 
3481 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3482 		    sizeof(ifr.ifr_name));
3483 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3484 			error = ENAMETOOLONG;
3485 			goto release_exit;
3486 		}
3487 		if (IFADDR_READER_EMPTY(ifp)) {
3488 			/* Interface with no addresses - send zero sockaddr. */
3489 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3490 			if (!docopy) {
3491 				space += sz;
3492 				goto next;
3493 			}
3494 			if (space >= sz) {
3495 				error = copyout(&ifr, ifrp, sz);
3496 				if (error != 0)
3497 					goto release_exit;
3498 				ifrp++;
3499 				space -= sz;
3500 			}
3501 		}
3502 
3503 		s = pserialize_read_enter();
3504 		IFADDR_READER_FOREACH(ifa, ifp) {
3505 			struct sockaddr *sa = ifa->ifa_addr;
3506 			/* all sockaddrs must fit in sockaddr_storage */
3507 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3508 
3509 			if (!docopy) {
3510 				space += sz;
3511 				continue;
3512 			}
3513 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3514 			pserialize_read_exit(s);
3515 
3516 			if (space >= sz) {
3517 				error = copyout(&ifr, ifrp, sz);
3518 				if (error != 0)
3519 					goto release_exit;
3520 				ifrp++; space -= sz;
3521 			}
3522 			s = pserialize_read_enter();
3523 		}
3524 		pserialize_read_exit(s);
3525 
3526         next:
3527 		s = pserialize_read_enter();
3528 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3529 	}
3530 	pserialize_read_exit(s);
3531 	curlwp_bindx(bound);
3532 
3533 	if (docopy) {
3534 		KASSERT(0 <= space && space <= ifc->ifc_len);
3535 		ifc->ifc_len -= space;
3536 	} else {
3537 		KASSERT(space >= 0);
3538 		ifc->ifc_len = space;
3539 	}
3540 	return (0);
3541 
3542 release_exit:
3543 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3544 	curlwp_bindx(bound);
3545 	return error;
3546 }
3547 
3548 int
3549 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3550 {
3551 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3552 	struct ifreq ifrb;
3553 	struct oifreq *oifr = NULL;
3554 	u_long ocmd = cmd;
3555 	int hook;
3556 
3557 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook);
3558 	if (hook != ENOSYS) {
3559 		if (cmd != ocmd) {
3560 			oifr = (struct oifreq *)(void *)ifr;
3561 			ifr = &ifrb;
3562 			IFREQO2N_43(oifr, ifr);
3563 				len = sizeof(oifr->ifr_addr);
3564 		}
3565 	}
3566 
3567 	if (len < sa->sa_len)
3568 		return EFBIG;
3569 
3570 	memset(&ifr->ifr_addr, 0, len);
3571 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3572 
3573 	if (cmd != ocmd)
3574 		IFREQN2O_43(oifr, ifr);
3575 	return 0;
3576 }
3577 
3578 /*
3579  * wrapper function for the drivers which doesn't have if_transmit().
3580  */
3581 static int
3582 if_transmit(struct ifnet *ifp, struct mbuf *m)
3583 {
3584 	int s, error;
3585 	size_t pktlen = m->m_pkthdr.len;
3586 	bool mcast = (m->m_flags & M_MCAST) != 0;
3587 
3588 	s = splnet();
3589 
3590 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3591 	if (error != 0) {
3592 		/* mbuf is already freed */
3593 		goto out;
3594 	}
3595 
3596 	ifp->if_obytes += pktlen;
3597 	if (mcast)
3598 		ifp->if_omcasts++;
3599 
3600 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3601 		if_start_lock(ifp);
3602 out:
3603 	splx(s);
3604 
3605 	return error;
3606 }
3607 
3608 int
3609 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3610 {
3611 	int error;
3612 
3613 #ifdef ALTQ
3614 	KERNEL_LOCK(1, NULL);
3615 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3616 		error = if_transmit(ifp, m);
3617 		KERNEL_UNLOCK_ONE(NULL);
3618 	} else {
3619 		KERNEL_UNLOCK_ONE(NULL);
3620 		error = (*ifp->if_transmit)(ifp, m);
3621 		/* mbuf is alredy freed */
3622 	}
3623 #else /* !ALTQ */
3624 	error = (*ifp->if_transmit)(ifp, m);
3625 	/* mbuf is alredy freed */
3626 #endif /* !ALTQ */
3627 
3628 	return error;
3629 }
3630 
3631 /*
3632  * Queue message on interface, and start output if interface
3633  * not yet active.
3634  */
3635 int
3636 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3637 {
3638 
3639 	return if_transmit_lock(ifp, m);
3640 }
3641 
3642 /*
3643  * Queue message on interface, possibly using a second fast queue
3644  */
3645 int
3646 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3647 {
3648 	int error = 0;
3649 
3650 	if (ifq != NULL
3651 #ifdef ALTQ
3652 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3653 #endif
3654 	    ) {
3655 		if (IF_QFULL(ifq)) {
3656 			IF_DROP(&ifp->if_snd);
3657 			m_freem(m);
3658 			if (error == 0)
3659 				error = ENOBUFS;
3660 		} else
3661 			IF_ENQUEUE(ifq, m);
3662 	} else
3663 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3664 	if (error != 0) {
3665 		++ifp->if_oerrors;
3666 		return error;
3667 	}
3668 	return 0;
3669 }
3670 
3671 int
3672 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3673 {
3674 	int rc;
3675 
3676 	KASSERT(IFNET_LOCKED(ifp));
3677 	if (ifp->if_initaddr != NULL)
3678 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3679 	else if (src ||
3680 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3681 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3682 
3683 	return rc;
3684 }
3685 
3686 int
3687 if_do_dad(struct ifnet *ifp)
3688 {
3689 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3690 		return 0;
3691 
3692 	switch (ifp->if_type) {
3693 	case IFT_FAITH:
3694 		/*
3695 		 * These interfaces do not have the IFF_LOOPBACK flag,
3696 		 * but loop packets back.  We do not have to do DAD on such
3697 		 * interfaces.  We should even omit it, because loop-backed
3698 		 * responses would confuse the DAD procedure.
3699 		 */
3700 		return 0;
3701 	default:
3702 		/*
3703 		 * Our DAD routine requires the interface up and running.
3704 		 * However, some interfaces can be up before the RUNNING
3705 		 * status.  Additionaly, users may try to assign addresses
3706 		 * before the interface becomes up (or running).
3707 		 * We simply skip DAD in such a case as a work around.
3708 		 * XXX: we should rather mark "tentative" on such addresses,
3709 		 * and do DAD after the interface becomes ready.
3710 		 */
3711 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
3712 		    (IFF_UP | IFF_RUNNING))
3713 			return 0;
3714 
3715 		return 1;
3716 	}
3717 }
3718 
3719 int
3720 if_flags_set(ifnet_t *ifp, const u_short flags)
3721 {
3722 	int rc;
3723 
3724 	KASSERT(IFNET_LOCKED(ifp));
3725 
3726 	if (ifp->if_setflags != NULL)
3727 		rc = (*ifp->if_setflags)(ifp, flags);
3728 	else {
3729 		u_short cantflags, chgdflags;
3730 		struct ifreq ifr;
3731 
3732 		chgdflags = ifp->if_flags ^ flags;
3733 		cantflags = chgdflags & IFF_CANTCHANGE;
3734 
3735 		if (cantflags != 0)
3736 			ifp->if_flags ^= cantflags;
3737 
3738                 /* Traditionally, we do not call if_ioctl after
3739                  * setting/clearing only IFF_PROMISC if the interface
3740                  * isn't IFF_UP.  Uphold that tradition.
3741 		 */
3742 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3743 			return 0;
3744 
3745 		memset(&ifr, 0, sizeof(ifr));
3746 
3747 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3748 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3749 
3750 		if (rc != 0 && cantflags != 0)
3751 			ifp->if_flags ^= cantflags;
3752 	}
3753 
3754 	return rc;
3755 }
3756 
3757 int
3758 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3759 {
3760 	int rc;
3761 	struct ifreq ifr;
3762 
3763 	if (ifp->if_mcastop != NULL)
3764 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3765 	else {
3766 		ifreq_setaddr(cmd, &ifr, sa);
3767 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3768 	}
3769 
3770 	return rc;
3771 }
3772 
3773 static void
3774 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3775     struct ifaltq *ifq)
3776 {
3777 	const struct sysctlnode *cnode, *rnode;
3778 
3779 	if (sysctl_createv(clog, 0, NULL, &rnode,
3780 		       CTLFLAG_PERMANENT,
3781 		       CTLTYPE_NODE, "interfaces",
3782 		       SYSCTL_DESCR("Per-interface controls"),
3783 		       NULL, 0, NULL, 0,
3784 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3785 		goto bad;
3786 
3787 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3788 		       CTLFLAG_PERMANENT,
3789 		       CTLTYPE_NODE, ifname,
3790 		       SYSCTL_DESCR("Interface controls"),
3791 		       NULL, 0, NULL, 0,
3792 		       CTL_CREATE, CTL_EOL) != 0)
3793 		goto bad;
3794 
3795 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3796 		       CTLFLAG_PERMANENT,
3797 		       CTLTYPE_NODE, "sndq",
3798 		       SYSCTL_DESCR("Interface output queue controls"),
3799 		       NULL, 0, NULL, 0,
3800 		       CTL_CREATE, CTL_EOL) != 0)
3801 		goto bad;
3802 
3803 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3804 		       CTLFLAG_PERMANENT,
3805 		       CTLTYPE_INT, "len",
3806 		       SYSCTL_DESCR("Current output queue length"),
3807 		       NULL, 0, &ifq->ifq_len, 0,
3808 		       CTL_CREATE, CTL_EOL) != 0)
3809 		goto bad;
3810 
3811 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3812 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3813 		       CTLTYPE_INT, "maxlen",
3814 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3815 		       NULL, 0, &ifq->ifq_maxlen, 0,
3816 		       CTL_CREATE, CTL_EOL) != 0)
3817 		goto bad;
3818 
3819 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3820 		       CTLFLAG_PERMANENT,
3821 		       CTLTYPE_INT, "drops",
3822 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3823 		       NULL, 0, &ifq->ifq_drops, 0,
3824 		       CTL_CREATE, CTL_EOL) != 0)
3825 		goto bad;
3826 
3827 	return;
3828 bad:
3829 	printf("%s: could not attach sysctl nodes\n", ifname);
3830 	return;
3831 }
3832 
3833 #if defined(INET) || defined(INET6)
3834 
3835 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3836 	static int							\
3837 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3838 	{								\
3839 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3840 	}
3841 
3842 #if defined(INET)
3843 static int
3844 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3845 {
3846 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3847 }
3848 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3849 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3850 #endif
3851 
3852 #if defined(INET6)
3853 static int
3854 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3855 {
3856 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3857 }
3858 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3859 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3860 #endif
3861 
3862 static void
3863 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3864 {
3865 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3866 	const char *pfname = NULL, *ipname = NULL;
3867 	int ipn = 0, qid = 0;
3868 
3869 	switch (pf) {
3870 #if defined(INET)
3871 	case PF_INET:
3872 		len_func = sysctl_net_ip_pktq_items;
3873 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3874 		drops_func = sysctl_net_ip_pktq_drops;
3875 		pfname = "inet", ipn = IPPROTO_IP;
3876 		ipname = "ip", qid = IPCTL_IFQ;
3877 		break;
3878 #endif
3879 #if defined(INET6)
3880 	case PF_INET6:
3881 		len_func = sysctl_net_ip6_pktq_items;
3882 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3883 		drops_func = sysctl_net_ip6_pktq_drops;
3884 		pfname = "inet6", ipn = IPPROTO_IPV6;
3885 		ipname = "ip6", qid = IPV6CTL_IFQ;
3886 		break;
3887 #endif
3888 	default:
3889 		KASSERT(false);
3890 	}
3891 
3892 	sysctl_createv(clog, 0, NULL, NULL,
3893 		       CTLFLAG_PERMANENT,
3894 		       CTLTYPE_NODE, pfname, NULL,
3895 		       NULL, 0, NULL, 0,
3896 		       CTL_NET, pf, CTL_EOL);
3897 	sysctl_createv(clog, 0, NULL, NULL,
3898 		       CTLFLAG_PERMANENT,
3899 		       CTLTYPE_NODE, ipname, NULL,
3900 		       NULL, 0, NULL, 0,
3901 		       CTL_NET, pf, ipn, CTL_EOL);
3902 	sysctl_createv(clog, 0, NULL, NULL,
3903 		       CTLFLAG_PERMANENT,
3904 		       CTLTYPE_NODE, "ifq",
3905 		       SYSCTL_DESCR("Protocol input queue controls"),
3906 		       NULL, 0, NULL, 0,
3907 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3908 
3909 	sysctl_createv(clog, 0, NULL, NULL,
3910 		       CTLFLAG_PERMANENT,
3911 		       CTLTYPE_QUAD, "len",
3912 		       SYSCTL_DESCR("Current input queue length"),
3913 		       len_func, 0, NULL, 0,
3914 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3915 	sysctl_createv(clog, 0, NULL, NULL,
3916 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3917 		       CTLTYPE_INT, "maxlen",
3918 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3919 		       maxlen_func, 0, NULL, 0,
3920 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3921 	sysctl_createv(clog, 0, NULL, NULL,
3922 		       CTLFLAG_PERMANENT,
3923 		       CTLTYPE_QUAD, "drops",
3924 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3925 		       drops_func, 0, NULL, 0,
3926 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3927 }
3928 #endif /* INET || INET6 */
3929 
3930 static int
3931 if_sdl_sysctl(SYSCTLFN_ARGS)
3932 {
3933 	struct ifnet *ifp;
3934 	const struct sockaddr_dl *sdl;
3935 	struct psref psref;
3936 	int error = 0;
3937 	int bound;
3938 
3939 	if (namelen != 1)
3940 		return EINVAL;
3941 
3942 	bound = curlwp_bind();
3943 	ifp = if_get_byindex(name[0], &psref);
3944 	if (ifp == NULL) {
3945 		error = ENODEV;
3946 		goto out0;
3947 	}
3948 
3949 	sdl = ifp->if_sadl;
3950 	if (sdl == NULL) {
3951 		*oldlenp = 0;
3952 		goto out1;
3953 	}
3954 
3955 	if (oldp == NULL) {
3956 		*oldlenp = sdl->sdl_alen;
3957 		goto out1;
3958 	}
3959 
3960 	if (*oldlenp >= sdl->sdl_alen)
3961 		*oldlenp = sdl->sdl_alen;
3962 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3963 out1:
3964 	if_put(ifp, &psref);
3965 out0:
3966 	curlwp_bindx(bound);
3967 	return error;
3968 }
3969 
3970 static void
3971 if_sysctl_setup(struct sysctllog **clog)
3972 {
3973 	const struct sysctlnode *rnode = NULL;
3974 
3975 	sysctl_createv(clog, 0, NULL, &rnode,
3976 		       CTLFLAG_PERMANENT,
3977 		       CTLTYPE_NODE, "sdl",
3978 		       SYSCTL_DESCR("Get active link-layer address"),
3979 		       if_sdl_sysctl, 0, NULL, 0,
3980 		       CTL_NET, CTL_CREATE, CTL_EOL);
3981 
3982 #if defined(INET)
3983 	sysctl_net_pktq_setup(NULL, PF_INET);
3984 #endif
3985 #ifdef INET6
3986 	if (in6_present)
3987 		sysctl_net_pktq_setup(NULL, PF_INET6);
3988 #endif
3989 }
3990