xref: /netbsd-src/sys/net/if.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: if.c,v 1.455 2019/05/21 09:18:37 msaitoh 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 1.455 2019/05/21 09:18:37 msaitoh 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 ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit ||
1047 	    ifp->if_transmit == if_transmit ||
1048 	    ALTQ_IS_ENABLED(&ifp->if_snd);
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 	uint64_t xc;
1311 
1312 #ifdef IFAREF_DEBUG
1313 	if_build_ifa_list(ifp);
1314 #endif
1315 	/*
1316 	 * XXX It's kind of lame that we have to have the
1317 	 * XXX socket structure...
1318 	 */
1319 	memset(&so, 0, sizeof(so));
1320 
1321 	s = splnet();
1322 
1323 	sysctl_teardown(&ifp->if_sysctl_log);
1324 	IFNET_LOCK(ifp);
1325 	if_deactivate(ifp);
1326 	IFNET_UNLOCK(ifp);
1327 
1328 	/*
1329 	 * Unlink from the list and wait for all readers to leave
1330 	 * from pserialize read sections.  Note that we can't do
1331 	 * psref_target_destroy here.  See below.
1332 	 */
1333 	IFNET_GLOBAL_LOCK();
1334 	ifindex2ifnet[ifp->if_index] = NULL;
1335 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1336 	IFNET_WRITER_REMOVE(ifp);
1337 	pserialize_perform(ifnet_psz);
1338 	IFNET_GLOBAL_UNLOCK();
1339 
1340 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1341 		ifp->if_slowtimo = NULL;
1342 		callout_halt(ifp->if_slowtimo_ch, NULL);
1343 		callout_destroy(ifp->if_slowtimo_ch);
1344 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1345 	}
1346 	if_deferred_start_destroy(ifp);
1347 
1348 	/*
1349 	 * Do an if_down() to give protocols a chance to do something.
1350 	 */
1351 	if_down_deactivated(ifp);
1352 
1353 #ifdef ALTQ
1354 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1355 		altq_disable(&ifp->if_snd);
1356 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1357 		altq_detach(&ifp->if_snd);
1358 #endif
1359 
1360 #if NCARP > 0
1361 	/* Remove the interface from any carp group it is a part of.  */
1362 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1363 		carp_ifdetach(ifp);
1364 #endif
1365 
1366 	/*
1367 	 * Rip all the addresses off the interface.  This should make
1368 	 * all of the routes go away.
1369 	 *
1370 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1371 	 * from the list, including our "cursor", ifa.  For safety,
1372 	 * and to honor the TAILQ abstraction, I just restart the
1373 	 * loop after each removal.  Note that the loop will exit
1374 	 * when all of the remaining ifaddrs belong to the AF_LINK
1375 	 * family.  I am counting on the historical fact that at
1376 	 * least one pr_usrreq in each address domain removes at
1377 	 * least one ifaddr.
1378 	 */
1379 again:
1380 	/*
1381 	 * At this point, no other one tries to remove ifa in the list,
1382 	 * so we don't need to take a lock or psref.  Avoid using
1383 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1384 	 * violations of pserialize.
1385 	 */
1386 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1387 		family = ifa->ifa_addr->sa_family;
1388 #ifdef IFAREF_DEBUG
1389 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1390 		    ifa, family, ifa->ifa_refcnt);
1391 		if (last_ifa != NULL && ifa == last_ifa)
1392 			panic("if_detach: loop detected");
1393 		last_ifa = ifa;
1394 #endif
1395 		if (family == AF_LINK)
1396 			continue;
1397 		dp = pffinddomain(family);
1398 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1399 		/*
1400 		 * XXX These PURGEIF calls are redundant with the
1401 		 * purge-all-families calls below, but are left in for
1402 		 * now both to make a smaller change, and to avoid
1403 		 * unplanned interactions with clearing of
1404 		 * ifp->if_addrlist.
1405 		 */
1406 		purged = 0;
1407 		for (pr = dp->dom_protosw;
1408 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1409 			so.so_proto = pr;
1410 			if (pr->pr_usrreqs) {
1411 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1412 				purged = 1;
1413 			}
1414 		}
1415 		if (purged == 0) {
1416 			/*
1417 			 * XXX What's really the best thing to do
1418 			 * XXX here?  --thorpej@NetBSD.org
1419 			 */
1420 			printf("if_detach: WARNING: AF %d not purged\n",
1421 			    family);
1422 			ifa_remove(ifp, ifa);
1423 		}
1424 		goto again;
1425 	}
1426 
1427 	if_free_sadl(ifp, 1);
1428 
1429 restart:
1430 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1431 		family = ifa->ifa_addr->sa_family;
1432 		KASSERT(family == AF_LINK);
1433 		ifa_remove(ifp, ifa);
1434 		goto restart;
1435 	}
1436 
1437 	/* Delete stray routes from the routing table. */
1438 	for (i = 0; i <= AF_MAX; i++)
1439 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1440 
1441 	DOMAIN_FOREACH(dp) {
1442 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1443 		{
1444 			void *p = ifp->if_afdata[dp->dom_family];
1445 			if (p) {
1446 				ifp->if_afdata[dp->dom_family] = NULL;
1447 				(*dp->dom_ifdetach)(ifp, p);
1448 			}
1449 		}
1450 
1451 		/*
1452 		 * One would expect multicast memberships (INET and
1453 		 * INET6) on UDP sockets to be purged by the PURGEIF
1454 		 * calls above, but if all addresses were removed from
1455 		 * the interface prior to destruction, the calls will
1456 		 * not be made (e.g. ppp, for which pppd(8) generally
1457 		 * removes addresses before destroying the interface).
1458 		 * Because there is no invariant that multicast
1459 		 * memberships only exist for interfaces with IPv4
1460 		 * addresses, we must call PURGEIF regardless of
1461 		 * addresses.  (Protocols which might store ifnet
1462 		 * pointers are marked with PR_PURGEIF.)
1463 		 */
1464 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1465 			so.so_proto = pr;
1466 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1467 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1468 		}
1469 	}
1470 
1471 	/*
1472 	 * Must be done after the above pr_purgeif because if_psref may be
1473 	 * still used in pr_purgeif.
1474 	 */
1475 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1476 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1477 
1478 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1479 	(void)pfil_head_destroy(ifp->if_pfil);
1480 
1481 	/* Announce that the interface is gone. */
1482 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1483 
1484 	IF_AFDATA_LOCK_DESTROY(ifp);
1485 
1486 	if (if_is_link_state_changeable(ifp)) {
1487 		softint_disestablish(ifp->if_link_si);
1488 		ifp->if_link_si = NULL;
1489 	}
1490 
1491 	/*
1492 	 * remove packets that came from ifp, from software interrupt queues.
1493 	 */
1494 	DOMAIN_FOREACH(dp) {
1495 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1496 			struct ifqueue *iq = dp->dom_ifqueues[i];
1497 			if (iq == NULL)
1498 				break;
1499 			dp->dom_ifqueues[i] = NULL;
1500 			if_detach_queues(ifp, iq);
1501 		}
1502 	}
1503 
1504 	/*
1505 	 * IP queues have to be processed separately: net-queue barrier
1506 	 * ensures that the packets are dequeued while a cross-call will
1507 	 * ensure that the interrupts have completed. FIXME: not quite..
1508 	 */
1509 #ifdef INET
1510 	pktq_barrier(ip_pktq);
1511 #endif
1512 #ifdef INET6
1513 	if (in6_present)
1514 		pktq_barrier(ip6_pktq);
1515 #endif
1516 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1517 	xc_wait(xc);
1518 
1519 	if (ifp->if_percpuq != NULL) {
1520 		if_percpuq_destroy(ifp->if_percpuq);
1521 		ifp->if_percpuq = NULL;
1522 	}
1523 
1524 	mutex_obj_free(ifp->if_ioctl_lock);
1525 	ifp->if_ioctl_lock = NULL;
1526 	mutex_obj_free(ifp->if_snd.ifq_lock);
1527 
1528 	splx(s);
1529 
1530 #ifdef IFAREF_DEBUG
1531 	if_check_and_free_ifa_list(ifp);
1532 #endif
1533 }
1534 
1535 static void
1536 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1537 {
1538 	struct mbuf *m, *prev, *next;
1539 
1540 	prev = NULL;
1541 	for (m = q->ifq_head; m != NULL; m = next) {
1542 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1543 
1544 		next = m->m_nextpkt;
1545 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1546 			prev = m;
1547 			continue;
1548 		}
1549 
1550 		if (prev != NULL)
1551 			prev->m_nextpkt = m->m_nextpkt;
1552 		else
1553 			q->ifq_head = m->m_nextpkt;
1554 		if (q->ifq_tail == m)
1555 			q->ifq_tail = prev;
1556 		q->ifq_len--;
1557 
1558 		m->m_nextpkt = NULL;
1559 		m_freem(m);
1560 		IF_DROP(q);
1561 	}
1562 }
1563 
1564 /*
1565  * Callback for a radix tree walk to delete all references to an
1566  * ifnet.
1567  */
1568 static int
1569 if_delroute_matcher(struct rtentry *rt, void *v)
1570 {
1571 	struct ifnet *ifp = (struct ifnet *)v;
1572 
1573 	if (rt->rt_ifp == ifp)
1574 		return 1;
1575 	else
1576 		return 0;
1577 }
1578 
1579 /*
1580  * Create a clone network interface.
1581  */
1582 static int
1583 if_clone_create(const char *name)
1584 {
1585 	struct if_clone *ifc;
1586 	int unit;
1587 	struct ifnet *ifp;
1588 	struct psref psref;
1589 
1590 	KASSERT(mutex_owned(&if_clone_mtx));
1591 
1592 	ifc = if_clone_lookup(name, &unit);
1593 	if (ifc == NULL)
1594 		return EINVAL;
1595 
1596 	ifp = if_get(name, &psref);
1597 	if (ifp != NULL) {
1598 		if_put(ifp, &psref);
1599 		return EEXIST;
1600 	}
1601 
1602 	return (*ifc->ifc_create)(ifc, unit);
1603 }
1604 
1605 /*
1606  * Destroy a clone network interface.
1607  */
1608 static int
1609 if_clone_destroy(const char *name)
1610 {
1611 	struct if_clone *ifc;
1612 	struct ifnet *ifp;
1613 	struct psref psref;
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 	ifp->if_ioctl = if_nullioctl;
1631 	IFNET_UNLOCK(ifp);
1632 
1633 	/*
1634 	 * We cannot call ifc_destroy with holding ifp.
1635 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1636 	 */
1637 	if_put(ifp, &psref);
1638 
1639 	return (*ifc->ifc_destroy)(ifp);
1640 }
1641 
1642 static bool
1643 if_is_unit(const char *name)
1644 {
1645 
1646 	while (*name != '\0') {
1647 		if (*name < '0' || *name > '9')
1648 			return false;
1649 		name++;
1650 	}
1651 
1652 	return true;
1653 }
1654 
1655 /*
1656  * Look up a network interface cloner.
1657  */
1658 static struct if_clone *
1659 if_clone_lookup(const char *name, int *unitp)
1660 {
1661 	struct if_clone *ifc;
1662 	const char *cp;
1663 	char *dp, ifname[IFNAMSIZ + 3];
1664 	int unit;
1665 
1666 	KASSERT(mutex_owned(&if_clone_mtx));
1667 
1668 	strcpy(ifname, "if_");
1669 	/* separate interface name from unit */
1670 	/* TODO: search unit number from backward */
1671 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1672 	    *cp && !if_is_unit(cp);)
1673 		*dp++ = *cp++;
1674 
1675 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1676 		return NULL;	/* No name or unit number */
1677 	*dp++ = '\0';
1678 
1679 again:
1680 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1681 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1682 			break;
1683 	}
1684 
1685 	if (ifc == NULL) {
1686 		int error;
1687 		if (*ifname == '\0')
1688 			return NULL;
1689 		mutex_exit(&if_clone_mtx);
1690 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1691 		mutex_enter(&if_clone_mtx);
1692 		if (error)
1693 			return NULL;
1694 		*ifname = '\0';
1695 		goto again;
1696 	}
1697 
1698 	unit = 0;
1699 	while (cp - name < IFNAMSIZ && *cp) {
1700 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1701 			/* Bogus unit number. */
1702 			return NULL;
1703 		}
1704 		unit = (unit * 10) + (*cp++ - '0');
1705 	}
1706 
1707 	if (unitp != NULL)
1708 		*unitp = unit;
1709 	return ifc;
1710 }
1711 
1712 /*
1713  * Register a network interface cloner.
1714  */
1715 void
1716 if_clone_attach(struct if_clone *ifc)
1717 {
1718 
1719 	mutex_enter(&if_clone_mtx);
1720 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1721 	if_cloners_count++;
1722 	mutex_exit(&if_clone_mtx);
1723 }
1724 
1725 /*
1726  * Unregister a network interface cloner.
1727  */
1728 void
1729 if_clone_detach(struct if_clone *ifc)
1730 {
1731 
1732 	mutex_enter(&if_clone_mtx);
1733 	LIST_REMOVE(ifc, ifc_list);
1734 	if_cloners_count--;
1735 	mutex_exit(&if_clone_mtx);
1736 }
1737 
1738 /*
1739  * Provide list of interface cloners to userspace.
1740  */
1741 int
1742 if_clone_list(int buf_count, char *buffer, int *total)
1743 {
1744 	char outbuf[IFNAMSIZ], *dst;
1745 	struct if_clone *ifc;
1746 	int count, error = 0;
1747 
1748 	mutex_enter(&if_clone_mtx);
1749 	*total = if_cloners_count;
1750 	if ((dst = buffer) == NULL) {
1751 		/* Just asking how many there are. */
1752 		goto out;
1753 	}
1754 
1755 	if (buf_count < 0) {
1756 		error = EINVAL;
1757 		goto out;
1758 	}
1759 
1760 	count = (if_cloners_count < buf_count) ?
1761 	    if_cloners_count : buf_count;
1762 
1763 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1764 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1765 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1766 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1767 			error = ENAMETOOLONG;
1768 			goto out;
1769 		}
1770 		error = copyout(outbuf, dst, sizeof(outbuf));
1771 		if (error != 0)
1772 			break;
1773 	}
1774 
1775 out:
1776 	mutex_exit(&if_clone_mtx);
1777 	return error;
1778 }
1779 
1780 void
1781 ifa_psref_init(struct ifaddr *ifa)
1782 {
1783 
1784 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1785 }
1786 
1787 void
1788 ifaref(struct ifaddr *ifa)
1789 {
1790 
1791 	atomic_inc_uint(&ifa->ifa_refcnt);
1792 }
1793 
1794 void
1795 ifafree(struct ifaddr *ifa)
1796 {
1797 	KASSERT(ifa != NULL);
1798 	KASSERT(ifa->ifa_refcnt > 0);
1799 
1800 	if (atomic_dec_uint_nv(&ifa->ifa_refcnt) == 0) {
1801 		free(ifa, M_IFADDR);
1802 	}
1803 }
1804 
1805 bool
1806 ifa_is_destroying(struct ifaddr *ifa)
1807 {
1808 
1809 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1810 }
1811 
1812 void
1813 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1814 {
1815 
1816 	ifa->ifa_ifp = ifp;
1817 
1818 	/*
1819 	 * Check MP-safety for IFEF_MPSAFE drivers.
1820 	 * Check !IFF_RUNNING for initialization routines that normally don't
1821 	 * take IFNET_LOCK but it's safe because there is no competitor.
1822 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1823 	 * if_stop.
1824 	 */
1825 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1826 	    IFNET_LOCKED(ifp));
1827 
1828 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1829 	IFADDR_ENTRY_INIT(ifa);
1830 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1831 
1832 	ifaref(ifa);
1833 }
1834 
1835 void
1836 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1837 {
1838 
1839 	KASSERT(ifa->ifa_ifp == ifp);
1840 	/*
1841 	 * Check MP-safety for IFEF_MPSAFE drivers.
1842 	 * if_is_deactivated indicates ifa_remove is called form if_detach
1843 	 * where is safe even if IFNET_LOCK isn't held.
1844 	 */
1845 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp));
1846 
1847 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1848 	IFADDR_WRITER_REMOVE(ifa);
1849 #ifdef NET_MPSAFE
1850 	IFNET_GLOBAL_LOCK();
1851 	pserialize_perform(ifnet_psz);
1852 	IFNET_GLOBAL_UNLOCK();
1853 #endif
1854 
1855 #ifdef NET_MPSAFE
1856 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1857 #endif
1858 	IFADDR_ENTRY_DESTROY(ifa);
1859 	ifafree(ifa);
1860 }
1861 
1862 void
1863 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1864 {
1865 
1866 	PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
1867 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1868 }
1869 
1870 void
1871 ifa_release(struct ifaddr *ifa, struct psref *psref)
1872 {
1873 
1874 	if (ifa == NULL)
1875 		return;
1876 
1877 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1878 }
1879 
1880 bool
1881 ifa_held(struct ifaddr *ifa)
1882 {
1883 
1884 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1885 }
1886 
1887 static inline int
1888 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1889 {
1890 	return sockaddr_cmp(sa1, sa2) == 0;
1891 }
1892 
1893 /*
1894  * Locate an interface based on a complete address.
1895  */
1896 /*ARGSUSED*/
1897 struct ifaddr *
1898 ifa_ifwithaddr(const struct sockaddr *addr)
1899 {
1900 	struct ifnet *ifp;
1901 	struct ifaddr *ifa;
1902 
1903 	IFNET_READER_FOREACH(ifp) {
1904 		if (if_is_deactivated(ifp))
1905 			continue;
1906 		IFADDR_READER_FOREACH(ifa, ifp) {
1907 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1908 				continue;
1909 			if (equal(addr, ifa->ifa_addr))
1910 				return ifa;
1911 			if ((ifp->if_flags & IFF_BROADCAST) &&
1912 			    ifa->ifa_broadaddr &&
1913 			    /* IP6 doesn't have broadcast */
1914 			    ifa->ifa_broadaddr->sa_len != 0 &&
1915 			    equal(ifa->ifa_broadaddr, addr))
1916 				return ifa;
1917 		}
1918 	}
1919 	return NULL;
1920 }
1921 
1922 struct ifaddr *
1923 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1924 {
1925 	struct ifaddr *ifa;
1926 	int s = pserialize_read_enter();
1927 
1928 	ifa = ifa_ifwithaddr(addr);
1929 	if (ifa != NULL)
1930 		ifa_acquire(ifa, psref);
1931 	pserialize_read_exit(s);
1932 
1933 	return ifa;
1934 }
1935 
1936 /*
1937  * Locate the point to point interface with a given destination address.
1938  */
1939 /*ARGSUSED*/
1940 struct ifaddr *
1941 ifa_ifwithdstaddr(const struct sockaddr *addr)
1942 {
1943 	struct ifnet *ifp;
1944 	struct ifaddr *ifa;
1945 
1946 	IFNET_READER_FOREACH(ifp) {
1947 		if (if_is_deactivated(ifp))
1948 			continue;
1949 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1950 			continue;
1951 		IFADDR_READER_FOREACH(ifa, ifp) {
1952 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1953 			    ifa->ifa_dstaddr == NULL)
1954 				continue;
1955 			if (equal(addr, ifa->ifa_dstaddr))
1956 				return ifa;
1957 		}
1958 	}
1959 
1960 	return NULL;
1961 }
1962 
1963 struct ifaddr *
1964 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1965 {
1966 	struct ifaddr *ifa;
1967 	int s;
1968 
1969 	s = pserialize_read_enter();
1970 	ifa = ifa_ifwithdstaddr(addr);
1971 	if (ifa != NULL)
1972 		ifa_acquire(ifa, psref);
1973 	pserialize_read_exit(s);
1974 
1975 	return ifa;
1976 }
1977 
1978 /*
1979  * Find an interface on a specific network.  If many, choice
1980  * is most specific found.
1981  */
1982 struct ifaddr *
1983 ifa_ifwithnet(const struct sockaddr *addr)
1984 {
1985 	struct ifnet *ifp;
1986 	struct ifaddr *ifa, *ifa_maybe = NULL;
1987 	const struct sockaddr_dl *sdl;
1988 	u_int af = addr->sa_family;
1989 	const char *addr_data = addr->sa_data, *cplim;
1990 
1991 	if (af == AF_LINK) {
1992 		sdl = satocsdl(addr);
1993 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1994 		    ifindex2ifnet[sdl->sdl_index] &&
1995 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1996 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1997 		}
1998 	}
1999 #ifdef NETATALK
2000 	if (af == AF_APPLETALK) {
2001 		const struct sockaddr_at *sat, *sat2;
2002 		sat = (const struct sockaddr_at *)addr;
2003 		IFNET_READER_FOREACH(ifp) {
2004 			if (if_is_deactivated(ifp))
2005 				continue;
2006 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
2007 			if (ifa == NULL)
2008 				continue;
2009 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
2010 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
2011 				return ifa; /* exact match */
2012 			if (ifa_maybe == NULL) {
2013 				/* else keep the if with the right range */
2014 				ifa_maybe = ifa;
2015 			}
2016 		}
2017 		return ifa_maybe;
2018 	}
2019 #endif
2020 	IFNET_READER_FOREACH(ifp) {
2021 		if (if_is_deactivated(ifp))
2022 			continue;
2023 		IFADDR_READER_FOREACH(ifa, ifp) {
2024 			const char *cp, *cp2, *cp3;
2025 
2026 			if (ifa->ifa_addr->sa_family != af ||
2027 			    ifa->ifa_netmask == NULL)
2028  next:				continue;
2029 			cp = addr_data;
2030 			cp2 = ifa->ifa_addr->sa_data;
2031 			cp3 = ifa->ifa_netmask->sa_data;
2032 			cplim = (const char *)ifa->ifa_netmask +
2033 			    ifa->ifa_netmask->sa_len;
2034 			while (cp3 < cplim) {
2035 				if ((*cp++ ^ *cp2++) & *cp3++) {
2036 					/* want to continue for() loop */
2037 					goto next;
2038 				}
2039 			}
2040 			if (ifa_maybe == NULL ||
2041 			    rt_refines(ifa->ifa_netmask,
2042 			               ifa_maybe->ifa_netmask))
2043 				ifa_maybe = ifa;
2044 		}
2045 	}
2046 	return ifa_maybe;
2047 }
2048 
2049 struct ifaddr *
2050 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2051 {
2052 	struct ifaddr *ifa;
2053 	int s;
2054 
2055 	s = pserialize_read_enter();
2056 	ifa = ifa_ifwithnet(addr);
2057 	if (ifa != NULL)
2058 		ifa_acquire(ifa, psref);
2059 	pserialize_read_exit(s);
2060 
2061 	return ifa;
2062 }
2063 
2064 /*
2065  * Find the interface of the addresss.
2066  */
2067 struct ifaddr *
2068 ifa_ifwithladdr(const struct sockaddr *addr)
2069 {
2070 	struct ifaddr *ia;
2071 
2072 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2073 	    (ia = ifa_ifwithnet(addr)))
2074 		return ia;
2075 	return NULL;
2076 }
2077 
2078 struct ifaddr *
2079 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2080 {
2081 	struct ifaddr *ifa;
2082 	int s;
2083 
2084 	s = pserialize_read_enter();
2085 	ifa = ifa_ifwithladdr(addr);
2086 	if (ifa != NULL)
2087 		ifa_acquire(ifa, psref);
2088 	pserialize_read_exit(s);
2089 
2090 	return ifa;
2091 }
2092 
2093 /*
2094  * Find an interface using a specific address family
2095  */
2096 struct ifaddr *
2097 ifa_ifwithaf(int af)
2098 {
2099 	struct ifnet *ifp;
2100 	struct ifaddr *ifa = NULL;
2101 	int s;
2102 
2103 	s = pserialize_read_enter();
2104 	IFNET_READER_FOREACH(ifp) {
2105 		if (if_is_deactivated(ifp))
2106 			continue;
2107 		IFADDR_READER_FOREACH(ifa, ifp) {
2108 			if (ifa->ifa_addr->sa_family == af)
2109 				goto out;
2110 		}
2111 	}
2112 out:
2113 	pserialize_read_exit(s);
2114 	return ifa;
2115 }
2116 
2117 /*
2118  * Find an interface address specific to an interface best matching
2119  * a given address.
2120  */
2121 struct ifaddr *
2122 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2123 {
2124 	struct ifaddr *ifa;
2125 	const char *cp, *cp2, *cp3;
2126 	const char *cplim;
2127 	struct ifaddr *ifa_maybe = 0;
2128 	u_int af = addr->sa_family;
2129 
2130 	if (if_is_deactivated(ifp))
2131 		return NULL;
2132 
2133 	if (af >= AF_MAX)
2134 		return NULL;
2135 
2136 	IFADDR_READER_FOREACH(ifa, ifp) {
2137 		if (ifa->ifa_addr->sa_family != af)
2138 			continue;
2139 		ifa_maybe = ifa;
2140 		if (ifa->ifa_netmask == NULL) {
2141 			if (equal(addr, ifa->ifa_addr) ||
2142 			    (ifa->ifa_dstaddr &&
2143 			     equal(addr, ifa->ifa_dstaddr)))
2144 				return ifa;
2145 			continue;
2146 		}
2147 		cp = addr->sa_data;
2148 		cp2 = ifa->ifa_addr->sa_data;
2149 		cp3 = ifa->ifa_netmask->sa_data;
2150 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2151 		for (; cp3 < cplim; cp3++) {
2152 			if ((*cp++ ^ *cp2++) & *cp3)
2153 				break;
2154 		}
2155 		if (cp3 == cplim)
2156 			return ifa;
2157 	}
2158 	return ifa_maybe;
2159 }
2160 
2161 struct ifaddr *
2162 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2163     struct psref *psref)
2164 {
2165 	struct ifaddr *ifa;
2166 	int s;
2167 
2168 	s = pserialize_read_enter();
2169 	ifa = ifaof_ifpforaddr(addr, ifp);
2170 	if (ifa != NULL)
2171 		ifa_acquire(ifa, psref);
2172 	pserialize_read_exit(s);
2173 
2174 	return ifa;
2175 }
2176 
2177 /*
2178  * Default action when installing a route with a Link Level gateway.
2179  * Lookup an appropriate real ifa to point to.
2180  * This should be moved to /sys/net/link.c eventually.
2181  */
2182 void
2183 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2184 {
2185 	struct ifaddr *ifa;
2186 	const struct sockaddr *dst;
2187 	struct ifnet *ifp;
2188 	struct psref psref;
2189 
2190 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2191 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL ||
2192 	    ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2193 		return;
2194 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2195 		rt_replace_ifa(rt, ifa);
2196 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2197 			ifa->ifa_rtrequest(cmd, rt, info);
2198 		ifa_release(ifa, &psref);
2199 	}
2200 }
2201 
2202 /*
2203  * bitmask macros to manage a densely packed link_state change queue.
2204  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2205  * LINK_STATE_UP(2) we need 2 bits for each state change.
2206  * As a state change to store is 0, treat all bits set as an unset item.
2207  */
2208 #define LQ_ITEM_BITS		2
2209 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2210 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2211 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2212 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2213 #define LQ_STORE(q, i, v)						      \
2214 	do {								      \
2215 		(q) &= ~LQ_MASK((i));					      \
2216 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2217 	} while (0 /* CONSTCOND */)
2218 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2219 #define LQ_POP(q, v)							      \
2220 	do {								      \
2221 		(v) = LQ_ITEM((q), 0);					      \
2222 		(q) >>= LQ_ITEM_BITS;					      \
2223 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2224 	} while (0 /* CONSTCOND */)
2225 #define LQ_PUSH(q, v)							      \
2226 	do {								      \
2227 		(q) >>= LQ_ITEM_BITS;					      \
2228 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2229 	} while (0 /* CONSTCOND */)
2230 #define LQ_FIND_UNSET(q, i)						      \
2231 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2232 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2233 			break;						      \
2234 	}
2235 
2236 /*
2237  * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex
2238  * for each ifnet.  It doesn't matter because:
2239  * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on
2240  *   ifq_lock don't happen
2241  * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock
2242  *   because if_snd, if_link_state_change and if_link_state_change_softint
2243  *   are all called with KERNEL_LOCK
2244  */
2245 #define IF_LINK_STATE_CHANGE_LOCK(ifp)		\
2246 	mutex_enter((ifp)->if_snd.ifq_lock)
2247 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp)	\
2248 	mutex_exit((ifp)->if_snd.ifq_lock)
2249 
2250 /*
2251  * Handle a change in the interface link state and
2252  * queue notifications.
2253  */
2254 void
2255 if_link_state_change(struct ifnet *ifp, int link_state)
2256 {
2257 	int idx;
2258 
2259 	KASSERTMSG(if_is_link_state_changeable(ifp),
2260 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2261 	    ifp->if_xname, ifp->if_extflags);
2262 
2263 	/* Ensure change is to a valid state */
2264 	switch (link_state) {
2265 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2266 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2267 	case LINK_STATE_UP:
2268 		break;
2269 	default:
2270 #ifdef DEBUG
2271 		printf("%s: invalid link state %d\n",
2272 		    ifp->if_xname, link_state);
2273 #endif
2274 		return;
2275 	}
2276 
2277 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2278 
2279 	/* Find the last unset event in the queue. */
2280 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2281 
2282 	/*
2283 	 * Ensure link_state doesn't match the last event in the queue.
2284 	 * ifp->if_link_state is not checked and set here because
2285 	 * that would present an inconsistent picture to the system.
2286 	 */
2287 	if (idx != 0 &&
2288 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2289 		goto out;
2290 
2291 	/* Handle queue overflow. */
2292 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2293 		uint8_t lost;
2294 
2295 		/*
2296 		 * The DOWN state must be protected from being pushed off
2297 		 * the queue to ensure that userland will always be
2298 		 * in a sane state.
2299 		 * Because DOWN is protected, there is no need to protect
2300 		 * UNKNOWN.
2301 		 * It should be invalid to change from any other state to
2302 		 * UNKNOWN anyway ...
2303 		 */
2304 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2305 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2306 		if (lost == LINK_STATE_DOWN) {
2307 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2308 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2309 		}
2310 		printf("%s: lost link state change %s\n",
2311 		    ifp->if_xname,
2312 		    lost == LINK_STATE_UP ? "UP" :
2313 		    lost == LINK_STATE_DOWN ? "DOWN" :
2314 		    "UNKNOWN");
2315 	} else
2316 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2317 
2318 	softint_schedule(ifp->if_link_si);
2319 
2320 out:
2321 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2322 }
2323 
2324 /*
2325  * Handle interface link state change notifications.
2326  */
2327 void
2328 if_link_state_change_softint(struct ifnet *ifp, int link_state)
2329 {
2330 	struct domain *dp;
2331 	int s = splnet();
2332 	bool notify;
2333 
2334 	KASSERT(!cpu_intr_p());
2335 
2336 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2337 
2338 	/* Ensure the change is still valid. */
2339 	if (ifp->if_link_state == link_state) {
2340 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2341 		splx(s);
2342 		return;
2343 	}
2344 
2345 #ifdef DEBUG
2346 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2347 		link_state == LINK_STATE_UP ? "UP" :
2348 		link_state == LINK_STATE_DOWN ? "DOWN" :
2349 		"UNKNOWN",
2350 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2351 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2352 		"UNKNOWN");
2353 #endif
2354 
2355 	/*
2356 	 * When going from UNKNOWN to UP, we need to mark existing
2357 	 * addresses as tentative and restart DAD as we may have
2358 	 * erroneously not found a duplicate.
2359 	 *
2360 	 * This needs to happen before rt_ifmsg to avoid a race where
2361 	 * listeners would have an address and expect it to work right
2362 	 * away.
2363 	 */
2364 	notify = (link_state == LINK_STATE_UP &&
2365 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2366 	ifp->if_link_state = link_state;
2367 	/* The following routines may sleep so release the spin mutex */
2368 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2369 
2370 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2371 	if (notify) {
2372 		DOMAIN_FOREACH(dp) {
2373 			if (dp->dom_if_link_state_change != NULL)
2374 				dp->dom_if_link_state_change(ifp,
2375 				    LINK_STATE_DOWN);
2376 		}
2377 	}
2378 
2379 	/* Notify that the link state has changed. */
2380 	rt_ifmsg(ifp);
2381 
2382 #if NCARP > 0
2383 	if (ifp->if_carp)
2384 		carp_carpdev_state(ifp);
2385 #endif
2386 
2387 	DOMAIN_FOREACH(dp) {
2388 		if (dp->dom_if_link_state_change != NULL)
2389 			dp->dom_if_link_state_change(ifp, link_state);
2390 	}
2391 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2392 	splx(s);
2393 }
2394 
2395 /*
2396  * Process the interface link state change queue.
2397  */
2398 static void
2399 if_link_state_change_si(void *arg)
2400 {
2401 	struct ifnet *ifp = arg;
2402 	int s;
2403 	uint8_t state;
2404 	bool schedule;
2405 
2406 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2407 	s = splnet();
2408 
2409 	/* Pop a link state change from the queue and process it. */
2410 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2411 	LQ_POP(ifp->if_link_queue, state);
2412 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2413 
2414 	if_link_state_change_softint(ifp, state);
2415 
2416 	/* If there is a link state change to come, schedule it. */
2417 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2418 	schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET);
2419 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2420 	if (schedule)
2421 		softint_schedule(ifp->if_link_si);
2422 
2423 	splx(s);
2424 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2425 }
2426 
2427 /*
2428  * Default action when installing a local route on a point-to-point
2429  * interface.
2430  */
2431 void
2432 p2p_rtrequest(int req, struct rtentry *rt,
2433     __unused const struct rt_addrinfo *info)
2434 {
2435 	struct ifnet *ifp = rt->rt_ifp;
2436 	struct ifaddr *ifa, *lo0ifa;
2437 	int s = pserialize_read_enter();
2438 
2439 	switch (req) {
2440 	case RTM_ADD:
2441 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2442 			break;
2443 
2444 		rt->rt_ifp = lo0ifp;
2445 
2446 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2447 			break;
2448 
2449 		IFADDR_READER_FOREACH(ifa, ifp) {
2450 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2451 				break;
2452 		}
2453 		if (ifa == NULL)
2454 			break;
2455 
2456 		/*
2457 		 * Ensure lo0 has an address of the same family.
2458 		 */
2459 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2460 			if (lo0ifa->ifa_addr->sa_family ==
2461 			    ifa->ifa_addr->sa_family)
2462 				break;
2463 		}
2464 		if (lo0ifa == NULL)
2465 			break;
2466 
2467 		/*
2468 		 * Make sure to set rt->rt_ifa to the interface
2469 		 * address we are using, otherwise we will have trouble
2470 		 * with source address selection.
2471 		 */
2472 		if (ifa != rt->rt_ifa)
2473 			rt_replace_ifa(rt, ifa);
2474 		break;
2475 	case RTM_DELETE:
2476 	default:
2477 		break;
2478 	}
2479 	pserialize_read_exit(s);
2480 }
2481 
2482 static void
2483 _if_down(struct ifnet *ifp)
2484 {
2485 	struct ifaddr *ifa;
2486 	struct domain *dp;
2487 	int s, bound;
2488 	struct psref psref;
2489 
2490 	ifp->if_flags &= ~IFF_UP;
2491 	nanotime(&ifp->if_lastchange);
2492 
2493 	bound = curlwp_bind();
2494 	s = pserialize_read_enter();
2495 	IFADDR_READER_FOREACH(ifa, ifp) {
2496 		ifa_acquire(ifa, &psref);
2497 		pserialize_read_exit(s);
2498 
2499 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2500 
2501 		s = pserialize_read_enter();
2502 		ifa_release(ifa, &psref);
2503 	}
2504 	pserialize_read_exit(s);
2505 	curlwp_bindx(bound);
2506 
2507 	IFQ_PURGE(&ifp->if_snd);
2508 #if NCARP > 0
2509 	if (ifp->if_carp)
2510 		carp_carpdev_state(ifp);
2511 #endif
2512 	rt_ifmsg(ifp);
2513 	DOMAIN_FOREACH(dp) {
2514 		if (dp->dom_if_down)
2515 			dp->dom_if_down(ifp);
2516 	}
2517 }
2518 
2519 static void
2520 if_down_deactivated(struct ifnet *ifp)
2521 {
2522 
2523 	KASSERT(if_is_deactivated(ifp));
2524 	_if_down(ifp);
2525 }
2526 
2527 void
2528 if_down_locked(struct ifnet *ifp)
2529 {
2530 
2531 	KASSERT(IFNET_LOCKED(ifp));
2532 	_if_down(ifp);
2533 }
2534 
2535 /*
2536  * Mark an interface down and notify protocols of
2537  * the transition.
2538  * NOTE: must be called at splsoftnet or equivalent.
2539  */
2540 void
2541 if_down(struct ifnet *ifp)
2542 {
2543 
2544 	IFNET_LOCK(ifp);
2545 	if_down_locked(ifp);
2546 	IFNET_UNLOCK(ifp);
2547 }
2548 
2549 /*
2550  * Must be called with holding if_ioctl_lock.
2551  */
2552 static void
2553 if_up_locked(struct ifnet *ifp)
2554 {
2555 #ifdef notyet
2556 	struct ifaddr *ifa;
2557 #endif
2558 	struct domain *dp;
2559 
2560 	KASSERT(IFNET_LOCKED(ifp));
2561 
2562 	KASSERT(!if_is_deactivated(ifp));
2563 	ifp->if_flags |= IFF_UP;
2564 	nanotime(&ifp->if_lastchange);
2565 #ifdef notyet
2566 	/* this has no effect on IP, and will kill all ISO connections XXX */
2567 	IFADDR_READER_FOREACH(ifa, ifp)
2568 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2569 #endif
2570 #if NCARP > 0
2571 	if (ifp->if_carp)
2572 		carp_carpdev_state(ifp);
2573 #endif
2574 	rt_ifmsg(ifp);
2575 	DOMAIN_FOREACH(dp) {
2576 		if (dp->dom_if_up)
2577 			dp->dom_if_up(ifp);
2578 	}
2579 }
2580 
2581 /*
2582  * Handle interface slowtimo timer routine.  Called
2583  * from softclock, we decrement timer (if set) and
2584  * call the appropriate interface routine on expiration.
2585  */
2586 static void
2587 if_slowtimo(void *arg)
2588 {
2589 	void (*slowtimo)(struct ifnet *);
2590 	struct ifnet *ifp = arg;
2591 	int s;
2592 
2593 	slowtimo = ifp->if_slowtimo;
2594 	if (__predict_false(slowtimo == NULL))
2595 		return;
2596 
2597 	s = splnet();
2598 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2599 		(*slowtimo)(ifp);
2600 
2601 	splx(s);
2602 
2603 	if (__predict_true(ifp->if_slowtimo != NULL))
2604 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2605 }
2606 
2607 /*
2608  * Mark an interface up and notify protocols of
2609  * the transition.
2610  * NOTE: must be called at splsoftnet or equivalent.
2611  */
2612 void
2613 if_up(struct ifnet *ifp)
2614 {
2615 
2616 	IFNET_LOCK(ifp);
2617 	if_up_locked(ifp);
2618 	IFNET_UNLOCK(ifp);
2619 }
2620 
2621 /*
2622  * Set/clear promiscuous mode on interface ifp based on the truth value
2623  * of pswitch.  The calls are reference counted so that only the first
2624  * "on" request actually has an effect, as does the final "off" request.
2625  * Results are undefined if the "off" and "on" requests are not matched.
2626  */
2627 int
2628 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2629 {
2630 	int pcount, ret = 0;
2631 	short nflags;
2632 
2633 	KASSERT(IFNET_LOCKED(ifp));
2634 
2635 	pcount = ifp->if_pcount;
2636 	if (pswitch) {
2637 		/*
2638 		 * Allow the device to be "placed" into promiscuous
2639 		 * mode even if it is not configured up.  It will
2640 		 * consult IFF_PROMISC when it is brought up.
2641 		 */
2642 		if (ifp->if_pcount++ != 0)
2643 			goto out;
2644 		nflags = ifp->if_flags | IFF_PROMISC;
2645 	} else {
2646 		if (--ifp->if_pcount > 0)
2647 			goto out;
2648 		nflags = ifp->if_flags & ~IFF_PROMISC;
2649 	}
2650 	ret = if_flags_set(ifp, nflags);
2651 	/* Restore interface state if not successful. */
2652 	if (ret != 0) {
2653 		ifp->if_pcount = pcount;
2654 	}
2655 out:
2656 	return ret;
2657 }
2658 
2659 int
2660 ifpromisc(struct ifnet *ifp, int pswitch)
2661 {
2662 	int e;
2663 
2664 	IFNET_LOCK(ifp);
2665 	e = ifpromisc_locked(ifp, pswitch);
2666 	IFNET_UNLOCK(ifp);
2667 
2668 	return e;
2669 }
2670 
2671 /*
2672  * Map interface name to
2673  * interface structure pointer.
2674  */
2675 struct ifnet *
2676 ifunit(const char *name)
2677 {
2678 	struct ifnet *ifp;
2679 	const char *cp = name;
2680 	u_int unit = 0;
2681 	u_int i;
2682 	int s;
2683 
2684 	/*
2685 	 * If the entire name is a number, treat it as an ifindex.
2686 	 */
2687 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2688 		unit = unit * 10 + (*cp - '0');
2689 	}
2690 
2691 	/*
2692 	 * If the number took all of the name, then it's a valid ifindex.
2693 	 */
2694 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2695 		return if_byindex(unit);
2696 
2697 	ifp = NULL;
2698 	s = pserialize_read_enter();
2699 	IFNET_READER_FOREACH(ifp) {
2700 		if (if_is_deactivated(ifp))
2701 			continue;
2702 	 	if (strcmp(ifp->if_xname, name) == 0)
2703 			goto out;
2704 	}
2705 out:
2706 	pserialize_read_exit(s);
2707 	return ifp;
2708 }
2709 
2710 /*
2711  * Get a reference of an ifnet object by an interface name.
2712  * The returned reference is protected by psref(9). The caller
2713  * must release a returned reference by if_put after use.
2714  */
2715 struct ifnet *
2716 if_get(const char *name, struct psref *psref)
2717 {
2718 	struct ifnet *ifp;
2719 	const char *cp = name;
2720 	u_int unit = 0;
2721 	u_int i;
2722 	int s;
2723 
2724 	/*
2725 	 * If the entire name is a number, treat it as an ifindex.
2726 	 */
2727 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2728 		unit = unit * 10 + (*cp - '0');
2729 	}
2730 
2731 	/*
2732 	 * If the number took all of the name, then it's a valid ifindex.
2733 	 */
2734 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2735 		return if_get_byindex(unit, psref);
2736 
2737 	ifp = NULL;
2738 	s = pserialize_read_enter();
2739 	IFNET_READER_FOREACH(ifp) {
2740 		if (if_is_deactivated(ifp))
2741 			continue;
2742 		if (strcmp(ifp->if_xname, name) == 0) {
2743 			PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2744 			psref_acquire(psref, &ifp->if_psref,
2745 			    ifnet_psref_class);
2746 			goto out;
2747 		}
2748 	}
2749 out:
2750 	pserialize_read_exit(s);
2751 	return ifp;
2752 }
2753 
2754 /*
2755  * Release a reference of an ifnet object given by if_get, if_get_byindex
2756  * or if_get_bylla.
2757  */
2758 void
2759 if_put(const struct ifnet *ifp, struct psref *psref)
2760 {
2761 
2762 	if (ifp == NULL)
2763 		return;
2764 
2765 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2766 }
2767 
2768 /*
2769  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2770  * should be used.
2771  */
2772 ifnet_t *
2773 _if_byindex(u_int idx)
2774 {
2775 
2776 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2777 }
2778 
2779 /*
2780  * Return ifp having idx. Return NULL if not found or the found ifp is
2781  * already deactivated.
2782  */
2783 ifnet_t *
2784 if_byindex(u_int idx)
2785 {
2786 	ifnet_t *ifp;
2787 
2788 	ifp = _if_byindex(idx);
2789 	if (ifp != NULL && if_is_deactivated(ifp))
2790 		ifp = NULL;
2791 	return ifp;
2792 }
2793 
2794 /*
2795  * Get a reference of an ifnet object by an interface index.
2796  * The returned reference is protected by psref(9). The caller
2797  * must release a returned reference by if_put after use.
2798  */
2799 ifnet_t *
2800 if_get_byindex(u_int idx, struct psref *psref)
2801 {
2802 	ifnet_t *ifp;
2803 	int s;
2804 
2805 	s = pserialize_read_enter();
2806 	ifp = if_byindex(idx);
2807 	if (__predict_true(ifp != NULL)) {
2808 		PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2809 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2810 	}
2811 	pserialize_read_exit(s);
2812 
2813 	return ifp;
2814 }
2815 
2816 ifnet_t *
2817 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2818 {
2819 	ifnet_t *ifp;
2820 	int s;
2821 
2822 	s = pserialize_read_enter();
2823 	IFNET_READER_FOREACH(ifp) {
2824 		if (if_is_deactivated(ifp))
2825 			continue;
2826 		if (ifp->if_addrlen != lla_len)
2827 			continue;
2828 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2829 			psref_acquire(psref, &ifp->if_psref,
2830 			    ifnet_psref_class);
2831 			break;
2832 		}
2833 	}
2834 	pserialize_read_exit(s);
2835 
2836 	return ifp;
2837 }
2838 
2839 /*
2840  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2841  * for example using pserialize or the ifp is already held or some other
2842  * object is held which guarantes the ifp to not be freed indirectly.
2843  */
2844 void
2845 if_acquire(struct ifnet *ifp, struct psref *psref)
2846 {
2847 
2848 	KASSERT(ifp->if_index != 0);
2849 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2850 }
2851 
2852 bool
2853 if_held(struct ifnet *ifp)
2854 {
2855 
2856 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2857 }
2858 
2859 /*
2860  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4.
2861  * Check the tunnel nesting count.
2862  * Return > 0, if tunnel nesting count is more than limit.
2863  * Return 0, if tunnel nesting count is equal or less than limit.
2864  */
2865 int
2866 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
2867 {
2868 	struct m_tag *mtag;
2869 	int *count;
2870 
2871 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO);
2872 	if (mtag != NULL) {
2873 		count = (int *)(mtag + 1);
2874 		if (++(*count) > limit) {
2875 			log(LOG_NOTICE,
2876 			    "%s: recursively called too many times(%d)\n",
2877 			    ifp->if_xname, *count);
2878 			return EIO;
2879 		}
2880 	} else {
2881 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
2882 		    M_NOWAIT);
2883 		if (mtag != NULL) {
2884 			m_tag_prepend(m, mtag);
2885 			count = (int *)(mtag + 1);
2886 			*count = 0;
2887 		} else {
2888 			log(LOG_DEBUG,
2889 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
2890 			    ifp->if_xname);
2891 		}
2892 	}
2893 
2894 	return 0;
2895 }
2896 
2897 /* common */
2898 int
2899 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2900 {
2901 	int s;
2902 	struct ifreq *ifr;
2903 	struct ifcapreq *ifcr;
2904 	struct ifdatareq *ifdr;
2905 
2906 	switch (cmd) {
2907 	case SIOCSIFCAP:
2908 		ifcr = data;
2909 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2910 			return EINVAL;
2911 
2912 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2913 			return 0;
2914 
2915 		ifp->if_capenable = ifcr->ifcr_capenable;
2916 
2917 		/* Pre-compute the checksum flags mask. */
2918 		ifp->if_csum_flags_tx = 0;
2919 		ifp->if_csum_flags_rx = 0;
2920 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx)
2921 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2922 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
2923 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2924 
2925 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx)
2926 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2927 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
2928 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2929 
2930 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx)
2931 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2932 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
2933 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2934 
2935 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx)
2936 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2937 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
2938 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2939 
2940 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx)
2941 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2942 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
2943 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2944 
2945 		if (ifp->if_capenable & IFCAP_TSOv4)
2946 			ifp->if_csum_flags_tx |= M_CSUM_TSOv4;
2947 		if (ifp->if_capenable & IFCAP_TSOv6)
2948 			ifp->if_csum_flags_tx |= M_CSUM_TSOv6;
2949 
2950 #if NBRIDGE > 0
2951 		if (ifp->if_bridge != NULL)
2952 			bridge_calc_csum_flags(ifp->if_bridge);
2953 #endif
2954 
2955 		if (ifp->if_flags & IFF_UP)
2956 			return ENETRESET;
2957 		return 0;
2958 	case SIOCSIFFLAGS:
2959 		ifr = data;
2960 		/*
2961 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
2962 		 * and if_down aren't MP-safe yet, so we must hold the lock.
2963 		 */
2964 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
2965 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2966 			s = splsoftnet();
2967 			if_down_locked(ifp);
2968 			splx(s);
2969 		}
2970 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2971 			s = splsoftnet();
2972 			if_up_locked(ifp);
2973 			splx(s);
2974 		}
2975 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
2976 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2977 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2978 		break;
2979 	case SIOCGIFFLAGS:
2980 		ifr = data;
2981 		ifr->ifr_flags = ifp->if_flags;
2982 		break;
2983 
2984 	case SIOCGIFMETRIC:
2985 		ifr = data;
2986 		ifr->ifr_metric = ifp->if_metric;
2987 		break;
2988 
2989 	case SIOCGIFMTU:
2990 		ifr = data;
2991 		ifr->ifr_mtu = ifp->if_mtu;
2992 		break;
2993 
2994 	case SIOCGIFDLT:
2995 		ifr = data;
2996 		ifr->ifr_dlt = ifp->if_dlt;
2997 		break;
2998 
2999 	case SIOCGIFCAP:
3000 		ifcr = data;
3001 		ifcr->ifcr_capabilities = ifp->if_capabilities;
3002 		ifcr->ifcr_capenable = ifp->if_capenable;
3003 		break;
3004 
3005 	case SIOCSIFMETRIC:
3006 		ifr = data;
3007 		ifp->if_metric = ifr->ifr_metric;
3008 		break;
3009 
3010 	case SIOCGIFDATA:
3011 		ifdr = data;
3012 		ifdr->ifdr_data = ifp->if_data;
3013 		break;
3014 
3015 	case SIOCGIFINDEX:
3016 		ifr = data;
3017 		ifr->ifr_index = ifp->if_index;
3018 		break;
3019 
3020 	case SIOCZIFDATA:
3021 		ifdr = data;
3022 		ifdr->ifdr_data = ifp->if_data;
3023 		/*
3024 		 * Assumes that the volatile counters that can be
3025 		 * zero'ed are at the end of if_data.
3026 		 */
3027 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
3028 		    offsetof(struct if_data, ifi_ipackets));
3029 		/*
3030 		 * The memset() clears to the bottm of if_data. In the area,
3031 		 * if_lastchange is included. Please be careful if new entry
3032 		 * will be added into if_data or rewite this.
3033 		 *
3034 		 * And also, update if_lastchnage.
3035 		 */
3036 		getnanotime(&ifp->if_lastchange);
3037 		break;
3038 	case SIOCSIFMTU:
3039 		ifr = data;
3040 		if (ifp->if_mtu == ifr->ifr_mtu)
3041 			break;
3042 		ifp->if_mtu = ifr->ifr_mtu;
3043 		/*
3044 		 * If the link MTU changed, do network layer specific procedure.
3045 		 */
3046 #ifdef INET6
3047 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3048 		if (in6_present)
3049 			nd6_setmtu(ifp);
3050 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3051 #endif
3052 		return ENETRESET;
3053 	default:
3054 		return ENOTTY;
3055 	}
3056 	return 0;
3057 }
3058 
3059 int
3060 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3061 {
3062 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3063 	struct ifaddr *ifa;
3064 	const struct sockaddr *any, *sa;
3065 	union {
3066 		struct sockaddr sa;
3067 		struct sockaddr_storage ss;
3068 	} u, v;
3069 	int s, error = 0;
3070 
3071 	switch (cmd) {
3072 	case SIOCSIFADDRPREF:
3073 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
3074 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
3075 		    NULL) != 0)
3076 			return EPERM;
3077 	case SIOCGIFADDRPREF:
3078 		break;
3079 	default:
3080 		return EOPNOTSUPP;
3081 	}
3082 
3083 	/* sanity checks */
3084 	if (data == NULL || ifp == NULL) {
3085 		panic("invalid argument to %s", __func__);
3086 		/*NOTREACHED*/
3087 	}
3088 
3089 	/* address must be specified on ADD and DELETE */
3090 	sa = sstocsa(&ifap->ifap_addr);
3091 	if (sa->sa_family != sofamily(so))
3092 		return EINVAL;
3093 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3094 		return EINVAL;
3095 
3096 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3097 
3098 	s = pserialize_read_enter();
3099 	IFADDR_READER_FOREACH(ifa, ifp) {
3100 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3101 			continue;
3102 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3103 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3104 			break;
3105 	}
3106 	if (ifa == NULL) {
3107 		error = EADDRNOTAVAIL;
3108 		goto out;
3109 	}
3110 
3111 	switch (cmd) {
3112 	case SIOCSIFADDRPREF:
3113 		ifa->ifa_preference = ifap->ifap_preference;
3114 		goto out;
3115 	case SIOCGIFADDRPREF:
3116 		/* fill in the if_laddrreq structure */
3117 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3118 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3119 		ifap->ifap_preference = ifa->ifa_preference;
3120 		goto out;
3121 	default:
3122 		error = EOPNOTSUPP;
3123 	}
3124 out:
3125 	pserialize_read_exit(s);
3126 	return error;
3127 }
3128 
3129 /*
3130  * Interface ioctls.
3131  */
3132 static int
3133 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3134 {
3135 	struct ifnet *ifp;
3136 	struct ifreq *ifr;
3137 	int error = 0;
3138 	u_long ocmd = cmd;
3139 	short oif_flags;
3140 	struct ifreq ifrb;
3141 	struct oifreq *oifr = NULL;
3142 	int r;
3143 	struct psref psref;
3144 	int bound;
3145 	bool do_if43_post = false;
3146 	bool do_ifm80_post = false;
3147 
3148 	switch (cmd) {
3149 	case SIOCGIFCONF:
3150 		return ifconf(cmd, data);
3151 	case SIOCINITIFADDR:
3152 		return EPERM;
3153 	default:
3154 		MODULE_HOOK_CALL(uipc_syscalls_40_hook, (cmd, data), enosys(),
3155 		    error);
3156 		if (error != ENOSYS)
3157 			return error;
3158 		MODULE_HOOK_CALL(uipc_syscalls_50_hook, (l, cmd, data),
3159 		    enosys(), error);
3160 		if (error != ENOSYS)
3161 			return error;
3162 		error = 0;
3163 		break;
3164 	}
3165 
3166 	ifr = data;
3167 	/* Pre-conversion */
3168 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), error);
3169 	if (cmd != ocmd) {
3170 		oifr = data;
3171 		data = ifr = &ifrb;
3172 		IFREQO2N_43(oifr, ifr);
3173 		do_if43_post = true;
3174 	}
3175 	MODULE_HOOK_CALL(ifmedia_80_pre_hook, (ifr, &cmd, &do_ifm80_post),
3176 	    enosys(), error);
3177 
3178 	switch (cmd) {
3179 	case SIOCIFCREATE:
3180 	case SIOCIFDESTROY:
3181 		bound = curlwp_bind();
3182 		if (l != NULL) {
3183 			ifp = if_get(ifr->ifr_name, &psref);
3184 			error = kauth_authorize_network(l->l_cred,
3185 			    KAUTH_NETWORK_INTERFACE,
3186 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3187 			    (void *)cmd, NULL);
3188 			if (ifp != NULL)
3189 				if_put(ifp, &psref);
3190 			if (error != 0) {
3191 				curlwp_bindx(bound);
3192 				return error;
3193 			}
3194 		}
3195 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3196 		mutex_enter(&if_clone_mtx);
3197 		r = (cmd == SIOCIFCREATE) ?
3198 			if_clone_create(ifr->ifr_name) :
3199 			if_clone_destroy(ifr->ifr_name);
3200 		mutex_exit(&if_clone_mtx);
3201 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3202 		curlwp_bindx(bound);
3203 		return r;
3204 
3205 	case SIOCIFGCLONERS:
3206 		{
3207 			struct if_clonereq *req = (struct if_clonereq *)data;
3208 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3209 			    &req->ifcr_total);
3210 		}
3211 	}
3212 
3213 	bound = curlwp_bind();
3214 	ifp = if_get(ifr->ifr_name, &psref);
3215 	if (ifp == NULL) {
3216 		curlwp_bindx(bound);
3217 		return ENXIO;
3218 	}
3219 
3220 	switch (cmd) {
3221 	case SIOCALIFADDR:
3222 	case SIOCDLIFADDR:
3223 	case SIOCSIFADDRPREF:
3224 	case SIOCSIFFLAGS:
3225 	case SIOCSIFCAP:
3226 	case SIOCSIFMETRIC:
3227 	case SIOCZIFDATA:
3228 	case SIOCSIFMTU:
3229 	case SIOCSIFPHYADDR:
3230 	case SIOCDIFPHYADDR:
3231 #ifdef INET6
3232 	case SIOCSIFPHYADDR_IN6:
3233 #endif
3234 	case SIOCSLIFPHYADDR:
3235 	case SIOCADDMULTI:
3236 	case SIOCDELMULTI:
3237 	case SIOCSETHERCAP:
3238 	case SIOCSIFMEDIA:
3239 	case SIOCSDRVSPEC:
3240 	case SIOCG80211:
3241 	case SIOCS80211:
3242 	case SIOCS80211NWID:
3243 	case SIOCS80211NWKEY:
3244 	case SIOCS80211POWER:
3245 	case SIOCS80211BSSID:
3246 	case SIOCS80211CHANNEL:
3247 	case SIOCSLINKSTR:
3248 		if (l != NULL) {
3249 			error = kauth_authorize_network(l->l_cred,
3250 			    KAUTH_NETWORK_INTERFACE,
3251 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3252 			    (void *)cmd, NULL);
3253 			if (error != 0)
3254 				goto out;
3255 		}
3256 	}
3257 
3258 	oif_flags = ifp->if_flags;
3259 
3260 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3261 	IFNET_LOCK(ifp);
3262 
3263 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3264 	if (error != ENOTTY)
3265 		;
3266 	else if (so->so_proto == NULL)
3267 		error = EOPNOTSUPP;
3268 	else {
3269 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3270 		MODULE_HOOK_CALL(if_ifioctl_43_hook,
3271 			     (so, ocmd, cmd, data, l), enosys(), error);
3272 		if (error == ENOSYS)
3273 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3274 			    cmd, data, ifp);
3275 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3276 	}
3277 
3278 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3279 		if ((ifp->if_flags & IFF_UP) != 0) {
3280 			int s = splsoftnet();
3281 			if_up_locked(ifp);
3282 			splx(s);
3283 		}
3284 	}
3285 
3286 	/* Post-conversion */
3287 	if (do_ifm80_post && (error == 0))
3288 		MODULE_HOOK_CALL(ifmedia_80_post_hook, (ifr, cmd),
3289 		    enosys(), error);
3290 	if (do_if43_post)
3291 		IFREQN2O_43(oifr, ifr);
3292 
3293 	IFNET_UNLOCK(ifp);
3294 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3295 out:
3296 	if_put(ifp, &psref);
3297 	curlwp_bindx(bound);
3298 	return error;
3299 }
3300 
3301 /*
3302  * Return interface configuration
3303  * of system.  List may be used
3304  * in later ioctl's (above) to get
3305  * other information.
3306  *
3307  * Each record is a struct ifreq.  Before the addition of
3308  * sockaddr_storage, the API rule was that sockaddr flavors that did
3309  * not fit would extend beyond the struct ifreq, with the next struct
3310  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3311  * union in struct ifreq includes struct sockaddr_storage, every kind
3312  * of sockaddr must fit.  Thus, there are no longer any overlength
3313  * records.
3314  *
3315  * Records are added to the user buffer if they fit, and ifc_len is
3316  * adjusted to the length that was written.  Thus, the user is only
3317  * assured of getting the complete list if ifc_len on return is at
3318  * least sizeof(struct ifreq) less than it was on entry.
3319  *
3320  * If the user buffer pointer is NULL, this routine copies no data and
3321  * returns the amount of space that would be needed.
3322  *
3323  * Invariants:
3324  * ifrp points to the next part of the user's buffer to be used.  If
3325  * ifrp != NULL, space holds the number of bytes remaining that we may
3326  * write at ifrp.  Otherwise, space holds the number of bytes that
3327  * would have been written had there been adequate space.
3328  */
3329 /*ARGSUSED*/
3330 static int
3331 ifconf(u_long cmd, void *data)
3332 {
3333 	struct ifconf *ifc = (struct ifconf *)data;
3334 	struct ifnet *ifp;
3335 	struct ifaddr *ifa;
3336 	struct ifreq ifr, *ifrp = NULL;
3337 	int space = 0, error = 0;
3338 	const int sz = (int)sizeof(struct ifreq);
3339 	const bool docopy = ifc->ifc_req != NULL;
3340 	int s;
3341 	int bound;
3342 	struct psref psref;
3343 
3344 	memset(&ifr, 0, sizeof(ifr));
3345 	if (docopy) {
3346 		space = ifc->ifc_len;
3347 		ifrp = ifc->ifc_req;
3348 	}
3349 
3350 	bound = curlwp_bind();
3351 	s = pserialize_read_enter();
3352 	IFNET_READER_FOREACH(ifp) {
3353 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3354 		pserialize_read_exit(s);
3355 
3356 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3357 		    sizeof(ifr.ifr_name));
3358 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3359 			error = ENAMETOOLONG;
3360 			goto release_exit;
3361 		}
3362 		if (IFADDR_READER_EMPTY(ifp)) {
3363 			/* Interface with no addresses - send zero sockaddr. */
3364 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3365 			if (!docopy) {
3366 				space += sz;
3367 				goto next;
3368 			}
3369 			if (space >= sz) {
3370 				error = copyout(&ifr, ifrp, sz);
3371 				if (error != 0)
3372 					goto release_exit;
3373 				ifrp++;
3374 				space -= sz;
3375 			}
3376 		}
3377 
3378 		s = pserialize_read_enter();
3379 		IFADDR_READER_FOREACH(ifa, ifp) {
3380 			struct sockaddr *sa = ifa->ifa_addr;
3381 			/* all sockaddrs must fit in sockaddr_storage */
3382 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3383 
3384 			if (!docopy) {
3385 				space += sz;
3386 				continue;
3387 			}
3388 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3389 			pserialize_read_exit(s);
3390 
3391 			if (space >= sz) {
3392 				error = copyout(&ifr, ifrp, sz);
3393 				if (error != 0)
3394 					goto release_exit;
3395 				ifrp++; space -= sz;
3396 			}
3397 			s = pserialize_read_enter();
3398 		}
3399 		pserialize_read_exit(s);
3400 
3401         next:
3402 		s = pserialize_read_enter();
3403 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3404 	}
3405 	pserialize_read_exit(s);
3406 	curlwp_bindx(bound);
3407 
3408 	if (docopy) {
3409 		KASSERT(0 <= space && space <= ifc->ifc_len);
3410 		ifc->ifc_len -= space;
3411 	} else {
3412 		KASSERT(space >= 0);
3413 		ifc->ifc_len = space;
3414 	}
3415 	return (0);
3416 
3417 release_exit:
3418 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3419 	curlwp_bindx(bound);
3420 	return error;
3421 }
3422 
3423 int
3424 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3425 {
3426 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3427 	struct ifreq ifrb;
3428 	struct oifreq *oifr = NULL;
3429 	u_long ocmd = cmd;
3430 	int hook;
3431 
3432 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook);
3433 	if (hook != ENOSYS) {
3434 		if (cmd != ocmd) {
3435 			oifr = (struct oifreq *)(void *)ifr;
3436 			ifr = &ifrb;
3437 			IFREQO2N_43(oifr, ifr);
3438 				len = sizeof(oifr->ifr_addr);
3439 		}
3440 	}
3441 
3442 	if (len < sa->sa_len)
3443 		return EFBIG;
3444 
3445 	memset(&ifr->ifr_addr, 0, len);
3446 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3447 
3448 	if (cmd != ocmd)
3449 		IFREQN2O_43(oifr, ifr);
3450 	return 0;
3451 }
3452 
3453 /*
3454  * wrapper function for the drivers which doesn't have if_transmit().
3455  */
3456 static int
3457 if_transmit(struct ifnet *ifp, struct mbuf *m)
3458 {
3459 	int s, error;
3460 	size_t pktlen = m->m_pkthdr.len;
3461 	bool mcast = (m->m_flags & M_MCAST) != 0;
3462 
3463 	s = splnet();
3464 
3465 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3466 	if (error != 0) {
3467 		/* mbuf is already freed */
3468 		goto out;
3469 	}
3470 
3471 	ifp->if_obytes += pktlen;
3472 	if (mcast)
3473 		ifp->if_omcasts++;
3474 
3475 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3476 		if_start_lock(ifp);
3477 out:
3478 	splx(s);
3479 
3480 	return error;
3481 }
3482 
3483 int
3484 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3485 {
3486 	int error;
3487 
3488 #ifdef ALTQ
3489 	KERNEL_LOCK(1, NULL);
3490 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3491 		error = if_transmit(ifp, m);
3492 		KERNEL_UNLOCK_ONE(NULL);
3493 	} else {
3494 		KERNEL_UNLOCK_ONE(NULL);
3495 		error = (*ifp->if_transmit)(ifp, m);
3496 		/* mbuf is alredy freed */
3497 	}
3498 #else /* !ALTQ */
3499 	error = (*ifp->if_transmit)(ifp, m);
3500 	/* mbuf is alredy freed */
3501 #endif /* !ALTQ */
3502 
3503 	return error;
3504 }
3505 
3506 /*
3507  * Queue message on interface, and start output if interface
3508  * not yet active.
3509  */
3510 int
3511 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3512 {
3513 
3514 	return if_transmit_lock(ifp, m);
3515 }
3516 
3517 /*
3518  * Queue message on interface, possibly using a second fast queue
3519  */
3520 int
3521 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3522 {
3523 	int error = 0;
3524 
3525 	if (ifq != NULL
3526 #ifdef ALTQ
3527 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3528 #endif
3529 	    ) {
3530 		if (IF_QFULL(ifq)) {
3531 			IF_DROP(&ifp->if_snd);
3532 			m_freem(m);
3533 			if (error == 0)
3534 				error = ENOBUFS;
3535 		} else
3536 			IF_ENQUEUE(ifq, m);
3537 	} else
3538 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3539 	if (error != 0) {
3540 		++ifp->if_oerrors;
3541 		return error;
3542 	}
3543 	return 0;
3544 }
3545 
3546 int
3547 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3548 {
3549 	int rc;
3550 
3551 	KASSERT(IFNET_LOCKED(ifp));
3552 	if (ifp->if_initaddr != NULL)
3553 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3554 	else if (src ||
3555 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3556 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3557 
3558 	return rc;
3559 }
3560 
3561 int
3562 if_do_dad(struct ifnet *ifp)
3563 {
3564 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3565 		return 0;
3566 
3567 	switch (ifp->if_type) {
3568 	case IFT_FAITH:
3569 		/*
3570 		 * These interfaces do not have the IFF_LOOPBACK flag,
3571 		 * but loop packets back.  We do not have to do DAD on such
3572 		 * interfaces.  We should even omit it, because loop-backed
3573 		 * responses would confuse the DAD procedure.
3574 		 */
3575 		return 0;
3576 	default:
3577 		/*
3578 		 * Our DAD routine requires the interface up and running.
3579 		 * However, some interfaces can be up before the RUNNING
3580 		 * status.  Additionaly, users may try to assign addresses
3581 		 * before the interface becomes up (or running).
3582 		 * We simply skip DAD in such a case as a work around.
3583 		 * XXX: we should rather mark "tentative" on such addresses,
3584 		 * and do DAD after the interface becomes ready.
3585 		 */
3586 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
3587 		    (IFF_UP | IFF_RUNNING))
3588 			return 0;
3589 
3590 		return 1;
3591 	}
3592 }
3593 
3594 int
3595 if_flags_set(ifnet_t *ifp, const short flags)
3596 {
3597 	int rc;
3598 
3599 	KASSERT(IFNET_LOCKED(ifp));
3600 
3601 	if (ifp->if_setflags != NULL)
3602 		rc = (*ifp->if_setflags)(ifp, flags);
3603 	else {
3604 		short cantflags, chgdflags;
3605 		struct ifreq ifr;
3606 
3607 		chgdflags = ifp->if_flags ^ flags;
3608 		cantflags = chgdflags & IFF_CANTCHANGE;
3609 
3610 		if (cantflags != 0)
3611 			ifp->if_flags ^= cantflags;
3612 
3613                 /* Traditionally, we do not call if_ioctl after
3614                  * setting/clearing only IFF_PROMISC if the interface
3615                  * isn't IFF_UP.  Uphold that tradition.
3616 		 */
3617 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3618 			return 0;
3619 
3620 		memset(&ifr, 0, sizeof(ifr));
3621 
3622 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3623 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3624 
3625 		if (rc != 0 && cantflags != 0)
3626 			ifp->if_flags ^= cantflags;
3627 	}
3628 
3629 	return rc;
3630 }
3631 
3632 int
3633 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3634 {
3635 	int rc;
3636 	struct ifreq ifr;
3637 
3638 	if (ifp->if_mcastop != NULL)
3639 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3640 	else {
3641 		ifreq_setaddr(cmd, &ifr, sa);
3642 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3643 	}
3644 
3645 	return rc;
3646 }
3647 
3648 static void
3649 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3650     struct ifaltq *ifq)
3651 {
3652 	const struct sysctlnode *cnode, *rnode;
3653 
3654 	if (sysctl_createv(clog, 0, NULL, &rnode,
3655 		       CTLFLAG_PERMANENT,
3656 		       CTLTYPE_NODE, "interfaces",
3657 		       SYSCTL_DESCR("Per-interface controls"),
3658 		       NULL, 0, NULL, 0,
3659 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3660 		goto bad;
3661 
3662 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3663 		       CTLFLAG_PERMANENT,
3664 		       CTLTYPE_NODE, ifname,
3665 		       SYSCTL_DESCR("Interface controls"),
3666 		       NULL, 0, NULL, 0,
3667 		       CTL_CREATE, CTL_EOL) != 0)
3668 		goto bad;
3669 
3670 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3671 		       CTLFLAG_PERMANENT,
3672 		       CTLTYPE_NODE, "sndq",
3673 		       SYSCTL_DESCR("Interface output queue controls"),
3674 		       NULL, 0, NULL, 0,
3675 		       CTL_CREATE, CTL_EOL) != 0)
3676 		goto bad;
3677 
3678 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3679 		       CTLFLAG_PERMANENT,
3680 		       CTLTYPE_INT, "len",
3681 		       SYSCTL_DESCR("Current output queue length"),
3682 		       NULL, 0, &ifq->ifq_len, 0,
3683 		       CTL_CREATE, CTL_EOL) != 0)
3684 		goto bad;
3685 
3686 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3687 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3688 		       CTLTYPE_INT, "maxlen",
3689 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3690 		       NULL, 0, &ifq->ifq_maxlen, 0,
3691 		       CTL_CREATE, CTL_EOL) != 0)
3692 		goto bad;
3693 
3694 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3695 		       CTLFLAG_PERMANENT,
3696 		       CTLTYPE_INT, "drops",
3697 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3698 		       NULL, 0, &ifq->ifq_drops, 0,
3699 		       CTL_CREATE, CTL_EOL) != 0)
3700 		goto bad;
3701 
3702 	return;
3703 bad:
3704 	printf("%s: could not attach sysctl nodes\n", ifname);
3705 	return;
3706 }
3707 
3708 #if defined(INET) || defined(INET6)
3709 
3710 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3711 	static int							\
3712 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3713 	{								\
3714 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3715 	}
3716 
3717 #if defined(INET)
3718 static int
3719 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3720 {
3721 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3722 }
3723 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3724 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3725 #endif
3726 
3727 #if defined(INET6)
3728 static int
3729 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3730 {
3731 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3732 }
3733 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3734 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3735 #endif
3736 
3737 static void
3738 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3739 {
3740 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3741 	const char *pfname = NULL, *ipname = NULL;
3742 	int ipn = 0, qid = 0;
3743 
3744 	switch (pf) {
3745 #if defined(INET)
3746 	case PF_INET:
3747 		len_func = sysctl_net_ip_pktq_items;
3748 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3749 		drops_func = sysctl_net_ip_pktq_drops;
3750 		pfname = "inet", ipn = IPPROTO_IP;
3751 		ipname = "ip", qid = IPCTL_IFQ;
3752 		break;
3753 #endif
3754 #if defined(INET6)
3755 	case PF_INET6:
3756 		len_func = sysctl_net_ip6_pktq_items;
3757 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3758 		drops_func = sysctl_net_ip6_pktq_drops;
3759 		pfname = "inet6", ipn = IPPROTO_IPV6;
3760 		ipname = "ip6", qid = IPV6CTL_IFQ;
3761 		break;
3762 #endif
3763 	default:
3764 		KASSERT(false);
3765 	}
3766 
3767 	sysctl_createv(clog, 0, NULL, NULL,
3768 		       CTLFLAG_PERMANENT,
3769 		       CTLTYPE_NODE, pfname, NULL,
3770 		       NULL, 0, NULL, 0,
3771 		       CTL_NET, pf, CTL_EOL);
3772 	sysctl_createv(clog, 0, NULL, NULL,
3773 		       CTLFLAG_PERMANENT,
3774 		       CTLTYPE_NODE, ipname, NULL,
3775 		       NULL, 0, NULL, 0,
3776 		       CTL_NET, pf, ipn, CTL_EOL);
3777 	sysctl_createv(clog, 0, NULL, NULL,
3778 		       CTLFLAG_PERMANENT,
3779 		       CTLTYPE_NODE, "ifq",
3780 		       SYSCTL_DESCR("Protocol input queue controls"),
3781 		       NULL, 0, NULL, 0,
3782 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3783 
3784 	sysctl_createv(clog, 0, NULL, NULL,
3785 		       CTLFLAG_PERMANENT,
3786 		       CTLTYPE_QUAD, "len",
3787 		       SYSCTL_DESCR("Current input queue length"),
3788 		       len_func, 0, NULL, 0,
3789 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3790 	sysctl_createv(clog, 0, NULL, NULL,
3791 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3792 		       CTLTYPE_INT, "maxlen",
3793 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3794 		       maxlen_func, 0, NULL, 0,
3795 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3796 	sysctl_createv(clog, 0, NULL, NULL,
3797 		       CTLFLAG_PERMANENT,
3798 		       CTLTYPE_QUAD, "drops",
3799 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3800 		       drops_func, 0, NULL, 0,
3801 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3802 }
3803 #endif /* INET || INET6 */
3804 
3805 static int
3806 if_sdl_sysctl(SYSCTLFN_ARGS)
3807 {
3808 	struct ifnet *ifp;
3809 	const struct sockaddr_dl *sdl;
3810 	struct psref psref;
3811 	int error = 0;
3812 	int bound;
3813 
3814 	if (namelen != 1)
3815 		return EINVAL;
3816 
3817 	bound = curlwp_bind();
3818 	ifp = if_get_byindex(name[0], &psref);
3819 	if (ifp == NULL) {
3820 		error = ENODEV;
3821 		goto out0;
3822 	}
3823 
3824 	sdl = ifp->if_sadl;
3825 	if (sdl == NULL) {
3826 		*oldlenp = 0;
3827 		goto out1;
3828 	}
3829 
3830 	if (oldp == NULL) {
3831 		*oldlenp = sdl->sdl_alen;
3832 		goto out1;
3833 	}
3834 
3835 	if (*oldlenp >= sdl->sdl_alen)
3836 		*oldlenp = sdl->sdl_alen;
3837 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3838 out1:
3839 	if_put(ifp, &psref);
3840 out0:
3841 	curlwp_bindx(bound);
3842 	return error;
3843 }
3844 
3845 static void
3846 if_sysctl_setup(struct sysctllog **clog)
3847 {
3848 	const struct sysctlnode *rnode = NULL;
3849 
3850 	sysctl_createv(clog, 0, NULL, &rnode,
3851 		       CTLFLAG_PERMANENT,
3852 		       CTLTYPE_NODE, "sdl",
3853 		       SYSCTL_DESCR("Get active link-layer address"),
3854 		       if_sdl_sysctl, 0, NULL, 0,
3855 		       CTL_NET, CTL_CREATE, CTL_EOL);
3856 
3857 #if defined(INET)
3858 	sysctl_net_pktq_setup(NULL, PF_INET);
3859 #endif
3860 #ifdef INET6
3861 	if (in6_present)
3862 		sysctl_net_pktq_setup(NULL, PF_INET6);
3863 #endif
3864 }
3865