xref: /netbsd-src/sys/net/if.c (revision 8feb0f0b7eaff0608f8350bbfa3098827b4bb91b)
1 /*	$NetBSD: if.c,v 1.528 2022/11/25 06:18:42 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 1.528 2022/11/25 06:18:42 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 #include <sys/msan.h>
125 #include <sys/hook.h>
126 
127 #include <net/if.h>
128 #include <net/if_dl.h>
129 #include <net/if_ether.h>
130 #include <net/if_media.h>
131 #include <net80211/ieee80211.h>
132 #include <net80211/ieee80211_ioctl.h>
133 #include <net/if_types.h>
134 #include <net/route.h>
135 #include <sys/module.h>
136 #ifdef NETATALK
137 #include <netatalk/at_extern.h>
138 #include <netatalk/at.h>
139 #endif
140 #include <net/pfil.h>
141 #include <netinet/in.h>
142 #include <netinet/in_var.h>
143 #include <netinet/ip_encap.h>
144 #include <net/bpf.h>
145 
146 #ifdef INET6
147 #include <netinet6/in6_var.h>
148 #include <netinet6/nd6.h>
149 #endif
150 
151 #include "ether.h"
152 
153 #include "bridge.h"
154 #if NBRIDGE > 0
155 #include <net/if_bridgevar.h>
156 #endif
157 
158 #include "carp.h"
159 #if NCARP > 0
160 #include <netinet/ip_carp.h>
161 #endif
162 
163 #include <compat/sys/sockio.h>
164 
165 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
166 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
167 
168 /*
169  * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex
170  * for each ifnet.  It doesn't matter because:
171  * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on
172  *   ifq_lock don't happen
173  * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock
174  *   because if_snd, if_link_state_change and if_link_state_change_process
175  *   are all called with KERNEL_LOCK
176  */
177 #define IF_LINK_STATE_CHANGE_LOCK(ifp)		\
178 	mutex_enter((ifp)->if_snd.ifq_lock)
179 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp)	\
180 	mutex_exit((ifp)->if_snd.ifq_lock)
181 
182 /*
183  * Global list of interfaces.
184  */
185 /* DEPRECATED. Remove it once kvm(3) users disappeared */
186 struct ifnet_head		ifnet_list;
187 
188 struct pslist_head		ifnet_pslist;
189 static ifnet_t **		ifindex2ifnet = NULL;
190 static u_int			if_index = 1;
191 static size_t			if_indexlim = 0;
192 static uint64_t			index_gen;
193 /* Mutex to protect the above objects. */
194 kmutex_t			ifnet_mtx __cacheline_aligned;
195 static struct psref_class	*ifnet_psref_class __read_mostly;
196 static pserialize_t		ifnet_psz;
197 static struct workqueue		*ifnet_link_state_wq __read_mostly;
198 
199 static struct workqueue		*if_slowtimo_wq __read_mostly;
200 
201 static kmutex_t			if_clone_mtx;
202 
203 struct ifnet *lo0ifp;
204 int	ifqmaxlen = IFQ_MAXLEN;
205 
206 struct psref_class		*ifa_psref_class __read_mostly;
207 
208 static int	if_delroute_matcher(struct rtentry *, void *);
209 
210 static bool if_is_unit(const char *);
211 static struct if_clone *if_clone_lookup(const char *, int *);
212 
213 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
214 static int if_cloners_count;
215 
216 /* Packet filtering hook for interfaces. */
217 pfil_head_t *			if_pfil __read_mostly;
218 
219 static kauth_listener_t if_listener;
220 
221 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
222 static void sysctl_sndq_setup(struct sysctllog **, const char *,
223     struct ifaltq *);
224 static void if_slowtimo_intr(void *);
225 static void if_slowtimo_work(struct work *, void *);
226 static int sysctl_if_watchdog(SYSCTLFN_PROTO);
227 static void sysctl_watchdog_setup(struct ifnet *);
228 static void if_attachdomain1(struct ifnet *);
229 static int ifconf(u_long, void *);
230 static int if_transmit(struct ifnet *, struct mbuf *);
231 static int if_clone_create(const char *);
232 static int if_clone_destroy(const char *);
233 static void if_link_state_change_work(struct work *, void *);
234 static void if_up_locked(struct ifnet *);
235 static void _if_down(struct ifnet *);
236 static void if_down_deactivated(struct ifnet *);
237 
238 struct if_percpuq {
239 	struct ifnet	*ipq_ifp;
240 	void		*ipq_si;
241 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
242 };
243 
244 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
245 
246 static void if_percpuq_drops(void *, void *, struct cpu_info *);
247 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
248 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
249     struct if_percpuq *);
250 
251 struct if_deferred_start {
252 	struct ifnet	*ids_ifp;
253 	void		(*ids_if_start)(struct ifnet *);
254 	void		*ids_si;
255 };
256 
257 static void if_deferred_start_softint(void *);
258 static void if_deferred_start_common(struct ifnet *);
259 static void if_deferred_start_destroy(struct ifnet *);
260 
261 struct if_slowtimo_data {
262 	kmutex_t		isd_lock;
263 	struct callout		isd_ch;
264 	struct work		isd_work;
265 	struct ifnet		*isd_ifp;
266 	bool			isd_queued;
267 	bool			isd_dying;
268 	bool			isd_trigger;
269 };
270 
271 /*
272  * Hook for if_vlan - needed by if_agr
273  */
274 struct if_vlan_vlan_input_hook_t if_vlan_vlan_input_hook;
275 
276 static void if_sysctl_setup(struct sysctllog **);
277 
278 static int
279 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
280     void *arg0, void *arg1, void *arg2, void *arg3)
281 {
282 	int result;
283 	enum kauth_network_req req;
284 
285 	result = KAUTH_RESULT_DEFER;
286 	req = (enum kauth_network_req)(uintptr_t)arg1;
287 
288 	if (action != KAUTH_NETWORK_INTERFACE)
289 		return result;
290 
291 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
292 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
293 		result = KAUTH_RESULT_ALLOW;
294 
295 	return result;
296 }
297 
298 /*
299  * Network interface utility routines.
300  *
301  * Routines with ifa_ifwith* names take sockaddr *'s as
302  * parameters.
303  */
304 void
305 ifinit(void)
306 {
307 
308 #if (defined(INET) || defined(INET6))
309 	encapinit();
310 #endif
311 
312 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
313 	    if_listener_cb, NULL);
314 
315 	/* interfaces are available, inform socket code */
316 	ifioctl = doifioctl;
317 }
318 
319 /*
320  * XXX Initialization before configure().
321  * XXX hack to get pfil_add_hook working in autoconf.
322  */
323 void
324 ifinit1(void)
325 {
326 	int error __diagused;
327 
328 #ifdef NET_MPSAFE
329 	printf("NET_MPSAFE enabled\n");
330 #endif
331 
332 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
333 
334 	TAILQ_INIT(&ifnet_list);
335 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
336 	ifnet_psz = pserialize_create();
337 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
338 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
339 	error = workqueue_create(&ifnet_link_state_wq, "iflnkst",
340 	    if_link_state_change_work, NULL, PRI_SOFTNET, IPL_NET,
341 	    WQ_MPSAFE);
342 	KASSERT(error == 0);
343 	PSLIST_INIT(&ifnet_pslist);
344 
345 	error = workqueue_create(&if_slowtimo_wq, "ifwdog",
346 	    if_slowtimo_work, NULL, PRI_SOFTNET, IPL_SOFTCLOCK, WQ_MPSAFE);
347 	KASSERTMSG(error == 0, "error=%d", error);
348 
349 	if_indexlim = 8;
350 
351 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
352 	KASSERT(if_pfil != NULL);
353 
354 #if NETHER > 0 || defined(NETATALK) || defined(WLAN)
355 	etherinit();
356 #endif
357 }
358 
359 /* XXX must be after domaininit() */
360 void
361 ifinit_post(void)
362 {
363 
364 	if_sysctl_setup(NULL);
365 }
366 
367 ifnet_t *
368 if_alloc(u_char type)
369 {
370 
371 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
372 }
373 
374 void
375 if_free(ifnet_t *ifp)
376 {
377 
378 	kmem_free(ifp, sizeof(ifnet_t));
379 }
380 
381 void
382 if_initname(struct ifnet *ifp, const char *name, int unit)
383 {
384 
385 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
386 	    "%s%d", name, unit);
387 }
388 
389 /*
390  * Null routines used while an interface is going away.  These routines
391  * just return an error.
392  */
393 
394 int
395 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
396     const struct sockaddr *so, const struct rtentry *rt)
397 {
398 
399 	return ENXIO;
400 }
401 
402 void
403 if_nullinput(struct ifnet *ifp, struct mbuf *m)
404 {
405 
406 	/* Nothing. */
407 }
408 
409 void
410 if_nullstart(struct ifnet *ifp)
411 {
412 
413 	/* Nothing. */
414 }
415 
416 int
417 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
418 {
419 
420 	m_freem(m);
421 	return ENXIO;
422 }
423 
424 int
425 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
426 {
427 
428 	return ENXIO;
429 }
430 
431 int
432 if_nullinit(struct ifnet *ifp)
433 {
434 
435 	return ENXIO;
436 }
437 
438 void
439 if_nullstop(struct ifnet *ifp, int disable)
440 {
441 
442 	/* Nothing. */
443 }
444 
445 void
446 if_nullslowtimo(struct ifnet *ifp)
447 {
448 
449 	/* Nothing. */
450 }
451 
452 void
453 if_nulldrain(struct ifnet *ifp)
454 {
455 
456 	/* Nothing. */
457 }
458 
459 void
460 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
461 {
462 	struct ifaddr *ifa;
463 	struct sockaddr_dl *sdl;
464 
465 	ifp->if_addrlen = addrlen;
466 	if_alloc_sadl(ifp);
467 	ifa = ifp->if_dl;
468 	sdl = satosdl(ifa->ifa_addr);
469 
470 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
471 	if (factory) {
472 		KASSERT(ifp->if_hwdl == NULL);
473 		ifp->if_hwdl = ifp->if_dl;
474 		ifaref(ifp->if_hwdl);
475 	}
476 	/* TBD routing socket */
477 }
478 
479 struct ifaddr *
480 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
481 {
482 	unsigned socksize, ifasize;
483 	int addrlen, namelen;
484 	struct sockaddr_dl *mask, *sdl;
485 	struct ifaddr *ifa;
486 
487 	namelen = strlen(ifp->if_xname);
488 	addrlen = ifp->if_addrlen;
489 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen),
490 	    sizeof(long));
491 	ifasize = sizeof(*ifa) + 2 * socksize;
492 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
493 
494 	sdl = (struct sockaddr_dl *)(ifa + 1);
495 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
496 
497 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
498 	    ifp->if_xname, namelen, NULL, addrlen);
499 	mask->sdl_family = AF_LINK;
500 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
501 	memset(&mask->sdl_data[0], 0xff, namelen);
502 	ifa->ifa_rtrequest = link_rtrequest;
503 	ifa->ifa_addr = (struct sockaddr *)sdl;
504 	ifa->ifa_netmask = (struct sockaddr *)mask;
505 	ifa_psref_init(ifa);
506 
507 	*sdlp = sdl;
508 
509 	return ifa;
510 }
511 
512 static void
513 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
514 {
515 	const struct sockaddr_dl *sdl;
516 
517 	ifp->if_dl = ifa;
518 	ifaref(ifa);
519 	sdl = satosdl(ifa->ifa_addr);
520 	ifp->if_sadl = sdl;
521 }
522 
523 /*
524  * Allocate the link level name for the specified interface.  This
525  * is an attachment helper.  It must be called after ifp->if_addrlen
526  * is initialized, which may not be the case when if_attach() is
527  * called.
528  */
529 void
530 if_alloc_sadl(struct ifnet *ifp)
531 {
532 	struct ifaddr *ifa;
533 	const struct sockaddr_dl *sdl;
534 
535 	/*
536 	 * If the interface already has a link name, release it
537 	 * now.  This is useful for interfaces that can change
538 	 * link types, and thus switch link names often.
539 	 */
540 	if (ifp->if_sadl != NULL)
541 		if_free_sadl(ifp, 0);
542 
543 	ifa = if_dl_create(ifp, &sdl);
544 
545 	ifa_insert(ifp, ifa);
546 	if_sadl_setrefs(ifp, ifa);
547 }
548 
549 static void
550 if_deactivate_sadl(struct ifnet *ifp)
551 {
552 	struct ifaddr *ifa;
553 
554 	KASSERT(ifp->if_dl != NULL);
555 
556 	ifa = ifp->if_dl;
557 
558 	ifp->if_sadl = NULL;
559 
560 	ifp->if_dl = NULL;
561 	ifafree(ifa);
562 }
563 
564 static void
565 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa)
566 {
567 	struct ifaddr *old;
568 
569 	KASSERT(ifp->if_dl != NULL);
570 
571 	old = ifp->if_dl;
572 
573 	ifaref(ifa);
574 	/* XXX Update if_dl and if_sadl atomically */
575 	ifp->if_dl = ifa;
576 	ifp->if_sadl = satosdl(ifa->ifa_addr);
577 
578 	ifafree(old);
579 }
580 
581 void
582 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
583     const struct sockaddr_dl *sdl)
584 {
585 	struct ifaddr *ifa;
586 	const int bound = curlwp_bind();
587 
588 	KASSERT(ifa_held(ifa0));
589 
590 	const int s = splsoftnet();
591 
592 	if_replace_sadl(ifp, ifa0);
593 
594 	int ss = pserialize_read_enter();
595 	IFADDR_READER_FOREACH(ifa, ifp) {
596 		struct psref psref;
597 		ifa_acquire(ifa, &psref);
598 		pserialize_read_exit(ss);
599 
600 		rtinit(ifa, RTM_LLINFO_UPD, 0);
601 
602 		ss = pserialize_read_enter();
603 		ifa_release(ifa, &psref);
604 	}
605 	pserialize_read_exit(ss);
606 
607 	splx(s);
608 	curlwp_bindx(bound);
609 }
610 
611 /*
612  * Free the link level name for the specified interface.  This is
613  * a detach helper.  This is called from if_detach().
614  */
615 void
616 if_free_sadl(struct ifnet *ifp, int factory)
617 {
618 	struct ifaddr *ifa;
619 
620 	if (factory && ifp->if_hwdl != NULL) {
621 		ifa = ifp->if_hwdl;
622 		ifp->if_hwdl = NULL;
623 		ifafree(ifa);
624 	}
625 
626 	ifa = ifp->if_dl;
627 	if (ifa == NULL) {
628 		KASSERT(ifp->if_sadl == NULL);
629 		return;
630 	}
631 
632 	KASSERT(ifp->if_sadl != NULL);
633 
634 	const int s = splsoftnet();
635 	KASSERT(ifa->ifa_addr->sa_family == AF_LINK);
636 	ifa_remove(ifp, ifa);
637 	if_deactivate_sadl(ifp);
638 	splx(s);
639 }
640 
641 static void
642 if_getindex(ifnet_t *ifp)
643 {
644 	bool hitlimit = false;
645 	char xnamebuf[HOOKNAMSIZ];
646 
647 	ifp->if_index_gen = index_gen++;
648 	snprintf(xnamebuf, sizeof(xnamebuf), "%s-lshk", ifp->if_xname);
649 	ifp->if_linkstate_hooks = simplehook_create(IPL_NET,
650 	    xnamebuf);
651 
652 	ifp->if_index = if_index;
653 	if (ifindex2ifnet == NULL) {
654 		if_index++;
655 		goto skip;
656 	}
657 	while (if_byindex(ifp->if_index)) {
658 		/*
659 		 * If we hit USHRT_MAX, we skip back to 0 since
660 		 * there are a number of places where the value
661 		 * of if_index or if_index itself is compared
662 		 * to or stored in an unsigned short.  By
663 		 * jumping back, we won't botch those assignments
664 		 * or comparisons.
665 		 */
666 		if (++if_index == 0) {
667 			if_index = 1;
668 		} else if (if_index == USHRT_MAX) {
669 			/*
670 			 * However, if we have to jump back to
671 			 * zero *twice* without finding an empty
672 			 * slot in ifindex2ifnet[], then there
673 			 * there are too many (>65535) interfaces.
674 			 */
675 			if (hitlimit)
676 				panic("too many interfaces");
677 			hitlimit = true;
678 			if_index = 1;
679 		}
680 		ifp->if_index = if_index;
681 	}
682 skip:
683 	/*
684 	 * ifindex2ifnet is indexed by if_index. Since if_index will
685 	 * grow dynamically, it should grow too.
686 	 */
687 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
688 		size_t m, n, oldlim;
689 		void *q;
690 
691 		oldlim = if_indexlim;
692 		while (ifp->if_index >= if_indexlim)
693 			if_indexlim <<= 1;
694 
695 		/* grow ifindex2ifnet */
696 		m = oldlim * sizeof(struct ifnet *);
697 		n = if_indexlim * sizeof(struct ifnet *);
698 		q = malloc(n, M_IFADDR, M_WAITOK | M_ZERO);
699 		if (ifindex2ifnet != NULL) {
700 			memcpy(q, ifindex2ifnet, m);
701 			free(ifindex2ifnet, M_IFADDR);
702 		}
703 		ifindex2ifnet = (struct ifnet **)q;
704 	}
705 	ifindex2ifnet[ifp->if_index] = ifp;
706 }
707 
708 /*
709  * Initialize an interface and assign an index for it.
710  *
711  * It must be called prior to a device specific attach routine
712  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
713  * and be followed by if_register:
714  *
715  *     if_initialize(ifp);
716  *     ether_ifattach(ifp, enaddr);
717  *     if_register(ifp);
718  */
719 void
720 if_initialize(ifnet_t *ifp)
721 {
722 
723 	KASSERT(if_indexlim > 0);
724 	TAILQ_INIT(&ifp->if_addrlist);
725 
726 	/*
727 	 * Link level name is allocated later by a separate call to
728 	 * if_alloc_sadl().
729 	 */
730 
731 	if (ifp->if_snd.ifq_maxlen == 0)
732 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
733 
734 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
735 
736 	ifp->if_link_state = LINK_STATE_UNKNOWN;
737 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
738 	ifp->if_link_scheduled = false;
739 
740 	ifp->if_capenable = 0;
741 	ifp->if_csum_flags_tx = 0;
742 	ifp->if_csum_flags_rx = 0;
743 
744 #ifdef ALTQ
745 	ifp->if_snd.altq_type = 0;
746 	ifp->if_snd.altq_disc = NULL;
747 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
748 	ifp->if_snd.altq_tbr  = NULL;
749 	ifp->if_snd.altq_ifp  = ifp;
750 #endif
751 
752 	IFQ_LOCK_INIT(&ifp->if_snd);
753 
754 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
755 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
756 
757 	IF_AFDATA_LOCK_INIT(ifp);
758 
759 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
760 	PSLIST_INIT(&ifp->if_addr_pslist);
761 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
762 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
763 	LIST_INIT(&ifp->if_multiaddrs);
764 	if_stats_init(ifp);
765 
766 	IFNET_GLOBAL_LOCK();
767 	if_getindex(ifp);
768 	IFNET_GLOBAL_UNLOCK();
769 }
770 
771 /*
772  * Register an interface to the list of "active" interfaces.
773  */
774 void
775 if_register(ifnet_t *ifp)
776 {
777 	/*
778 	 * If the driver has not supplied its own if_ioctl or if_stop,
779 	 * then supply the default.
780 	 */
781 	if (ifp->if_ioctl == NULL)
782 		ifp->if_ioctl = ifioctl_common;
783 	if (ifp->if_stop == NULL)
784 		ifp->if_stop = if_nullstop;
785 
786 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
787 
788 	if (!STAILQ_EMPTY(&domains))
789 		if_attachdomain1(ifp);
790 
791 	/* Announce the interface. */
792 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
793 
794 	if (ifp->if_slowtimo != NULL) {
795 		struct if_slowtimo_data *isd;
796 
797 		isd = kmem_zalloc(sizeof(*isd), KM_SLEEP);
798 		mutex_init(&isd->isd_lock, MUTEX_DEFAULT, IPL_SOFTCLOCK);
799 		callout_init(&isd->isd_ch, CALLOUT_MPSAFE);
800 		callout_setfunc(&isd->isd_ch, if_slowtimo_intr, ifp);
801 		isd->isd_ifp = ifp;
802 
803 		ifp->if_slowtimo_data = isd;
804 
805 		if_slowtimo_intr(ifp);
806 
807 		sysctl_watchdog_setup(ifp);
808 	}
809 
810 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
811 		ifp->if_transmit = if_transmit;
812 
813 	IFNET_GLOBAL_LOCK();
814 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
815 	IFNET_WRITER_INSERT_TAIL(ifp);
816 	IFNET_GLOBAL_UNLOCK();
817 }
818 
819 /*
820  * The if_percpuq framework
821  *
822  * It allows network device drivers to execute the network stack
823  * in softint (so called softint-based if_input). It utilizes
824  * softint and percpu ifqueue. It doesn't distribute any packets
825  * between CPUs, unlike pktqueue(9).
826  *
827  * Currently we support two options for device drivers to apply the framework:
828  * - Use it implicitly with less changes
829  *   - If you use if_attach in driver's _attach function and if_input in
830  *     driver's Rx interrupt handler, a packet is queued and a softint handles
831  *     the packet implicitly
832  * - Use it explicitly in each driver (recommended)
833  *   - You can use if_percpuq_* directly in your driver
834  *   - In this case, you need to allocate struct if_percpuq in driver's softc
835  *   - See wm(4) as a reference implementation
836  */
837 
838 static void
839 if_percpuq_softint(void *arg)
840 {
841 	struct if_percpuq *ipq = arg;
842 	struct ifnet *ifp = ipq->ipq_ifp;
843 	struct mbuf *m;
844 
845 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
846 		if_statinc(ifp, if_ipackets);
847 		bpf_mtap(ifp, m, BPF_D_IN);
848 
849 		ifp->_if_input(ifp, m);
850 	}
851 }
852 
853 static void
854 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
855 {
856 	struct ifqueue *const ifq = p;
857 
858 	memset(ifq, 0, sizeof(*ifq));
859 	ifq->ifq_maxlen = IFQ_MAXLEN;
860 }
861 
862 struct if_percpuq *
863 if_percpuq_create(struct ifnet *ifp)
864 {
865 	struct if_percpuq *ipq;
866 	u_int flags = SOFTINT_NET;
867 
868 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
869 
870 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
871 	ipq->ipq_ifp = ifp;
872 	ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq);
873 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
874 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
875 
876 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
877 
878 	return ipq;
879 }
880 
881 static struct mbuf *
882 if_percpuq_dequeue(struct if_percpuq *ipq)
883 {
884 	struct mbuf *m;
885 	struct ifqueue *ifq;
886 
887 	const int s = splnet();
888 	ifq = percpu_getref(ipq->ipq_ifqs);
889 	IF_DEQUEUE(ifq, m);
890 	percpu_putref(ipq->ipq_ifqs);
891 	splx(s);
892 
893 	return m;
894 }
895 
896 static void
897 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
898 {
899 	struct ifqueue *const ifq = p;
900 
901 	IF_PURGE(ifq);
902 }
903 
904 void
905 if_percpuq_destroy(struct if_percpuq *ipq)
906 {
907 
908 	/* if_detach may already destroy it */
909 	if (ipq == NULL)
910 		return;
911 
912 	softint_disestablish(ipq->ipq_si);
913 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
914 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
915 	kmem_free(ipq, sizeof(*ipq));
916 }
917 
918 void
919 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
920 {
921 	struct ifqueue *ifq;
922 
923 	KASSERT(ipq != NULL);
924 
925 	const int s = splnet();
926 	ifq = percpu_getref(ipq->ipq_ifqs);
927 	if (IF_QFULL(ifq)) {
928 		IF_DROP(ifq);
929 		percpu_putref(ipq->ipq_ifqs);
930 		m_freem(m);
931 		goto out;
932 	}
933 	IF_ENQUEUE(ifq, m);
934 	percpu_putref(ipq->ipq_ifqs);
935 
936 	softint_schedule(ipq->ipq_si);
937 out:
938 	splx(s);
939 }
940 
941 static void
942 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
943 {
944 	struct ifqueue *const ifq = p;
945 	uint64_t *sum = arg;
946 
947 	*sum += ifq->ifq_drops;
948 }
949 
950 static int
951 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
952 {
953 	struct sysctlnode node;
954 	struct if_percpuq *ipq;
955 	uint64_t sum = 0;
956 	int error;
957 
958 	node = *rnode;
959 	ipq = node.sysctl_data;
960 
961 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
962 
963 	node.sysctl_data = &sum;
964 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
965 	if (error != 0 || newp == NULL)
966 		return error;
967 
968 	return 0;
969 }
970 
971 static void
972 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
973     struct if_percpuq *ipq)
974 {
975 	const struct sysctlnode *cnode, *rnode;
976 
977 	if (sysctl_createv(clog, 0, NULL, &rnode,
978 		       CTLFLAG_PERMANENT,
979 		       CTLTYPE_NODE, "interfaces",
980 		       SYSCTL_DESCR("Per-interface controls"),
981 		       NULL, 0, NULL, 0,
982 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
983 		goto bad;
984 
985 	if (sysctl_createv(clog, 0, &rnode, &rnode,
986 		       CTLFLAG_PERMANENT,
987 		       CTLTYPE_NODE, ifname,
988 		       SYSCTL_DESCR("Interface controls"),
989 		       NULL, 0, NULL, 0,
990 		       CTL_CREATE, CTL_EOL) != 0)
991 		goto bad;
992 
993 	if (sysctl_createv(clog, 0, &rnode, &rnode,
994 		       CTLFLAG_PERMANENT,
995 		       CTLTYPE_NODE, "rcvq",
996 		       SYSCTL_DESCR("Interface input queue controls"),
997 		       NULL, 0, NULL, 0,
998 		       CTL_CREATE, CTL_EOL) != 0)
999 		goto bad;
1000 
1001 #ifdef NOTYET
1002 	/* XXX Should show each per-CPU queue length? */
1003 	if (sysctl_createv(clog, 0, &rnode, &rnode,
1004 		       CTLFLAG_PERMANENT,
1005 		       CTLTYPE_INT, "len",
1006 		       SYSCTL_DESCR("Current input queue length"),
1007 		       sysctl_percpuq_len, 0, NULL, 0,
1008 		       CTL_CREATE, CTL_EOL) != 0)
1009 		goto bad;
1010 
1011 	if (sysctl_createv(clog, 0, &rnode, &cnode,
1012 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1013 		       CTLTYPE_INT, "maxlen",
1014 		       SYSCTL_DESCR("Maximum allowed input queue length"),
1015 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
1016 		       CTL_CREATE, CTL_EOL) != 0)
1017 		goto bad;
1018 #endif
1019 
1020 	if (sysctl_createv(clog, 0, &rnode, &cnode,
1021 		       CTLFLAG_PERMANENT,
1022 		       CTLTYPE_QUAD, "drops",
1023 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
1024 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
1025 		       CTL_CREATE, CTL_EOL) != 0)
1026 		goto bad;
1027 
1028 	return;
1029 bad:
1030 	printf("%s: could not attach sysctl nodes\n", ifname);
1031 	return;
1032 }
1033 
1034 /*
1035  * The deferred if_start framework
1036  *
1037  * The common APIs to defer if_start to softint when if_start is requested
1038  * from a device driver running in hardware interrupt context.
1039  */
1040 /*
1041  * Call ifp->if_start (or equivalent) in a dedicated softint for
1042  * deferred if_start.
1043  */
1044 static void
1045 if_deferred_start_softint(void *arg)
1046 {
1047 	struct if_deferred_start *ids = arg;
1048 	struct ifnet *ifp = ids->ids_ifp;
1049 
1050 	ids->ids_if_start(ifp);
1051 }
1052 
1053 /*
1054  * The default callback function for deferred if_start.
1055  */
1056 static void
1057 if_deferred_start_common(struct ifnet *ifp)
1058 {
1059 	const int s = splnet();
1060 	if_start_lock(ifp);
1061 	splx(s);
1062 }
1063 
1064 static inline bool
1065 if_snd_is_used(struct ifnet *ifp)
1066 {
1067 
1068 	return ALTQ_IS_ENABLED(&ifp->if_snd) ||
1069 	    ifp->if_transmit == if_transmit ||
1070 	    ifp->if_transmit == NULL ||
1071 	    ifp->if_transmit == if_nulltransmit;
1072 }
1073 
1074 /*
1075  * Schedule deferred if_start.
1076  */
1077 void
1078 if_schedule_deferred_start(struct ifnet *ifp)
1079 {
1080 
1081 	KASSERT(ifp->if_deferred_start != NULL);
1082 
1083 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1084 		return;
1085 
1086 	softint_schedule(ifp->if_deferred_start->ids_si);
1087 }
1088 
1089 /*
1090  * Create an instance of deferred if_start. A driver should call the function
1091  * only if the driver needs deferred if_start. Drivers can setup their own
1092  * deferred if_start function via 2nd argument.
1093  */
1094 void
1095 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1096 {
1097 	struct if_deferred_start *ids;
1098 	u_int flags = SOFTINT_NET;
1099 
1100 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
1101 
1102 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1103 	ids->ids_ifp = ifp;
1104 	ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids);
1105 	if (func != NULL)
1106 		ids->ids_if_start = func;
1107 	else
1108 		ids->ids_if_start = if_deferred_start_common;
1109 
1110 	ifp->if_deferred_start = ids;
1111 }
1112 
1113 static void
1114 if_deferred_start_destroy(struct ifnet *ifp)
1115 {
1116 
1117 	if (ifp->if_deferred_start == NULL)
1118 		return;
1119 
1120 	softint_disestablish(ifp->if_deferred_start->ids_si);
1121 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1122 	ifp->if_deferred_start = NULL;
1123 }
1124 
1125 /*
1126  * The common interface input routine that is called by device drivers,
1127  * which should be used only when the driver's rx handler already runs
1128  * in softint.
1129  */
1130 void
1131 if_input(struct ifnet *ifp, struct mbuf *m)
1132 {
1133 
1134 	KASSERT(ifp->if_percpuq == NULL);
1135 	KASSERT(!cpu_intr_p());
1136 
1137 	if_statinc(ifp, if_ipackets);
1138 	bpf_mtap(ifp, m, BPF_D_IN);
1139 
1140 	ifp->_if_input(ifp, m);
1141 }
1142 
1143 /*
1144  * DEPRECATED. Use if_initialize and if_register instead.
1145  * See the above comment of if_initialize.
1146  *
1147  * Note that it implicitly enables if_percpuq to make drivers easy to
1148  * migrate softint-based if_input without much changes. If you don't
1149  * want to enable it, use if_initialize instead.
1150  */
1151 void
1152 if_attach(ifnet_t *ifp)
1153 {
1154 
1155 	if_initialize(ifp);
1156 	ifp->if_percpuq = if_percpuq_create(ifp);
1157 	if_register(ifp);
1158 }
1159 
1160 void
1161 if_attachdomain(void)
1162 {
1163 	struct ifnet *ifp;
1164 	const int bound = curlwp_bind();
1165 
1166 	int s = pserialize_read_enter();
1167 	IFNET_READER_FOREACH(ifp) {
1168 		struct psref psref;
1169 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1170 		pserialize_read_exit(s);
1171 		if_attachdomain1(ifp);
1172 		s = pserialize_read_enter();
1173 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1174 	}
1175 	pserialize_read_exit(s);
1176 	curlwp_bindx(bound);
1177 }
1178 
1179 static void
1180 if_attachdomain1(struct ifnet *ifp)
1181 {
1182 	struct domain *dp;
1183 	const int s = splsoftnet();
1184 
1185 	/* address family dependent data region */
1186 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1187 	DOMAIN_FOREACH(dp) {
1188 		if (dp->dom_ifattach != NULL)
1189 			ifp->if_afdata[dp->dom_family] =
1190 			    (*dp->dom_ifattach)(ifp);
1191 	}
1192 
1193 	splx(s);
1194 }
1195 
1196 /*
1197  * Deactivate an interface.  This points all of the procedure
1198  * handles at error stubs.  May be called from interrupt context.
1199  */
1200 void
1201 if_deactivate(struct ifnet *ifp)
1202 {
1203 	const int s = splsoftnet();
1204 
1205 	ifp->if_output	 = if_nulloutput;
1206 	ifp->_if_input	 = if_nullinput;
1207 	ifp->if_start	 = if_nullstart;
1208 	ifp->if_transmit = if_nulltransmit;
1209 	ifp->if_ioctl	 = if_nullioctl;
1210 	ifp->if_init	 = if_nullinit;
1211 	ifp->if_stop	 = if_nullstop;
1212 	if (ifp->if_slowtimo)
1213 		ifp->if_slowtimo = if_nullslowtimo;
1214 	ifp->if_drain	 = if_nulldrain;
1215 
1216 	/* No more packets may be enqueued. */
1217 	ifp->if_snd.ifq_maxlen = 0;
1218 
1219 	splx(s);
1220 }
1221 
1222 bool
1223 if_is_deactivated(const struct ifnet *ifp)
1224 {
1225 
1226 	return ifp->if_output == if_nulloutput;
1227 }
1228 
1229 void
1230 if_purgeaddrs(struct ifnet *ifp, int family,
1231     void (*purgeaddr)(struct ifaddr *))
1232 {
1233 	struct ifaddr *ifa, *nifa;
1234 	int s;
1235 
1236 	s = pserialize_read_enter();
1237 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1238 		nifa = IFADDR_READER_NEXT(ifa);
1239 		if (ifa->ifa_addr->sa_family != family)
1240 			continue;
1241 		pserialize_read_exit(s);
1242 
1243 		(*purgeaddr)(ifa);
1244 
1245 		s = pserialize_read_enter();
1246 	}
1247 	pserialize_read_exit(s);
1248 }
1249 
1250 #ifdef IFAREF_DEBUG
1251 static struct ifaddr **ifa_list;
1252 static int ifa_list_size;
1253 
1254 /* Depends on only one if_attach runs at once */
1255 static void
1256 if_build_ifa_list(struct ifnet *ifp)
1257 {
1258 	struct ifaddr *ifa;
1259 	int i;
1260 
1261 	KASSERT(ifa_list == NULL);
1262 	KASSERT(ifa_list_size == 0);
1263 
1264 	IFADDR_READER_FOREACH(ifa, ifp)
1265 		ifa_list_size++;
1266 
1267 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1268 	i = 0;
1269 	IFADDR_READER_FOREACH(ifa, ifp) {
1270 		ifa_list[i++] = ifa;
1271 		ifaref(ifa);
1272 	}
1273 }
1274 
1275 static void
1276 if_check_and_free_ifa_list(struct ifnet *ifp)
1277 {
1278 	int i;
1279 	struct ifaddr *ifa;
1280 
1281 	if (ifa_list == NULL)
1282 		return;
1283 
1284 	for (i = 0; i < ifa_list_size; i++) {
1285 		char buf[64];
1286 
1287 		ifa = ifa_list[i];
1288 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1289 		if (ifa->ifa_refcnt > 1) {
1290 			log(LOG_WARNING,
1291 			    "ifa(%s) still referenced (refcnt=%d)\n",
1292 			    buf, ifa->ifa_refcnt - 1);
1293 		} else
1294 			log(LOG_DEBUG,
1295 			    "ifa(%s) not referenced (refcnt=%d)\n",
1296 			    buf, ifa->ifa_refcnt - 1);
1297 		ifafree(ifa);
1298 	}
1299 
1300 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1301 	ifa_list = NULL;
1302 	ifa_list_size = 0;
1303 }
1304 #endif
1305 
1306 /*
1307  * Detach an interface from the list of "active" interfaces,
1308  * freeing any resources as we go along.
1309  *
1310  * NOTE: This routine must be called with a valid thread context,
1311  * as it may block.
1312  */
1313 void
1314 if_detach(struct ifnet *ifp)
1315 {
1316 	struct socket so;
1317 	struct ifaddr *ifa;
1318 #ifdef IFAREF_DEBUG
1319 	struct ifaddr *last_ifa = NULL;
1320 #endif
1321 	struct domain *dp;
1322 	const struct protosw *pr;
1323 	int i, family, purged;
1324 
1325 #ifdef IFAREF_DEBUG
1326 	if_build_ifa_list(ifp);
1327 #endif
1328 	/*
1329 	 * XXX It's kind of lame that we have to have the
1330 	 * XXX socket structure...
1331 	 */
1332 	memset(&so, 0, sizeof(so));
1333 
1334 	const int s = splnet();
1335 
1336 	sysctl_teardown(&ifp->if_sysctl_log);
1337 
1338 	IFNET_LOCK(ifp);
1339 
1340 	/*
1341 	 * Unset all queued link states and pretend a
1342 	 * link state change is scheduled.
1343 	 * This stops any more link state changes occurring for this
1344 	 * interface while it's being detached so it's safe
1345 	 * to drain the workqueue.
1346 	 */
1347 	IF_LINK_STATE_CHANGE_LOCK(ifp);
1348 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
1349 	ifp->if_link_scheduled = true;
1350 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
1351 	workqueue_wait(ifnet_link_state_wq, &ifp->if_link_work);
1352 
1353 	if_deactivate(ifp);
1354 	IFNET_UNLOCK(ifp);
1355 
1356 	/*
1357 	 * Unlink from the list and wait for all readers to leave
1358 	 * from pserialize read sections.  Note that we can't do
1359 	 * psref_target_destroy here.  See below.
1360 	 */
1361 	IFNET_GLOBAL_LOCK();
1362 	ifindex2ifnet[ifp->if_index] = NULL;
1363 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1364 	IFNET_WRITER_REMOVE(ifp);
1365 	pserialize_perform(ifnet_psz);
1366 	IFNET_GLOBAL_UNLOCK();
1367 
1368 	if (ifp->if_slowtimo != NULL) {
1369 		struct if_slowtimo_data *isd = ifp->if_slowtimo_data;
1370 
1371 		mutex_enter(&isd->isd_lock);
1372 		isd->isd_dying = true;
1373 		mutex_exit(&isd->isd_lock);
1374 		callout_halt(&isd->isd_ch, NULL);
1375 		workqueue_wait(if_slowtimo_wq, &isd->isd_work);
1376 		callout_destroy(&isd->isd_ch);
1377 		mutex_destroy(&isd->isd_lock);
1378 		kmem_free(isd, sizeof(*isd));
1379 
1380 		ifp->if_slowtimo_data = NULL; /* paraonia */
1381 		ifp->if_slowtimo = NULL;      /* paranoia */
1382 	}
1383 	if_deferred_start_destroy(ifp);
1384 
1385 	/*
1386 	 * Do an if_down() to give protocols a chance to do something.
1387 	 */
1388 	if_down_deactivated(ifp);
1389 
1390 #ifdef ALTQ
1391 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1392 		altq_disable(&ifp->if_snd);
1393 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1394 		altq_detach(&ifp->if_snd);
1395 #endif
1396 
1397 #if NCARP > 0
1398 	/* Remove the interface from any carp group it is a part of.  */
1399 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1400 		carp_ifdetach(ifp);
1401 #endif
1402 
1403 	/*
1404 	 * Ensure that all packets on protocol input pktqueues have been
1405 	 * processed, or, at least, removed from the queues.
1406 	 *
1407 	 * A cross-call will ensure that the interrupts have completed.
1408 	 * FIXME: not quite..
1409 	 */
1410 	pktq_ifdetach();
1411 	xc_barrier(0);
1412 
1413 	/*
1414 	 * Rip all the addresses off the interface.  This should make
1415 	 * all of the routes go away.
1416 	 *
1417 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1418 	 * from the list, including our "cursor", ifa.  For safety,
1419 	 * and to honor the TAILQ abstraction, I just restart the
1420 	 * loop after each removal.  Note that the loop will exit
1421 	 * when all of the remaining ifaddrs belong to the AF_LINK
1422 	 * family.  I am counting on the historical fact that at
1423 	 * least one pr_usrreq in each address domain removes at
1424 	 * least one ifaddr.
1425 	 */
1426 again:
1427 	/*
1428 	 * At this point, no other one tries to remove ifa in the list,
1429 	 * so we don't need to take a lock or psref.  Avoid using
1430 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1431 	 * violations of pserialize.
1432 	 */
1433 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1434 		family = ifa->ifa_addr->sa_family;
1435 #ifdef IFAREF_DEBUG
1436 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1437 		    ifa, family, ifa->ifa_refcnt);
1438 		if (last_ifa != NULL && ifa == last_ifa)
1439 			panic("if_detach: loop detected");
1440 		last_ifa = ifa;
1441 #endif
1442 		if (family == AF_LINK)
1443 			continue;
1444 		dp = pffinddomain(family);
1445 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1446 		/*
1447 		 * XXX These PURGEIF calls are redundant with the
1448 		 * purge-all-families calls below, but are left in for
1449 		 * now both to make a smaller change, and to avoid
1450 		 * unplanned interactions with clearing of
1451 		 * ifp->if_addrlist.
1452 		 */
1453 		purged = 0;
1454 		for (pr = dp->dom_protosw;
1455 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1456 			so.so_proto = pr;
1457 			if (pr->pr_usrreqs) {
1458 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1459 				purged = 1;
1460 			}
1461 		}
1462 		if (purged == 0) {
1463 			/*
1464 			 * XXX What's really the best thing to do
1465 			 * XXX here?  --thorpej@NetBSD.org
1466 			 */
1467 			printf("if_detach: WARNING: AF %d not purged\n",
1468 			    family);
1469 			ifa_remove(ifp, ifa);
1470 		}
1471 		goto again;
1472 	}
1473 
1474 	if_free_sadl(ifp, 1);
1475 
1476 restart:
1477 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1478 		family = ifa->ifa_addr->sa_family;
1479 		KASSERT(family == AF_LINK);
1480 		ifa_remove(ifp, ifa);
1481 		goto restart;
1482 	}
1483 
1484 	/* Delete stray routes from the routing table. */
1485 	for (i = 0; i <= AF_MAX; i++)
1486 		rt_delete_matched_entries(i, if_delroute_matcher, ifp, false);
1487 
1488 	DOMAIN_FOREACH(dp) {
1489 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1490 		{
1491 			void *p = ifp->if_afdata[dp->dom_family];
1492 			if (p) {
1493 				ifp->if_afdata[dp->dom_family] = NULL;
1494 				(*dp->dom_ifdetach)(ifp, p);
1495 			}
1496 		}
1497 
1498 		/*
1499 		 * One would expect multicast memberships (INET and
1500 		 * INET6) on UDP sockets to be purged by the PURGEIF
1501 		 * calls above, but if all addresses were removed from
1502 		 * the interface prior to destruction, the calls will
1503 		 * not be made (e.g. ppp, for which pppd(8) generally
1504 		 * removes addresses before destroying the interface).
1505 		 * Because there is no invariant that multicast
1506 		 * memberships only exist for interfaces with IPv4
1507 		 * addresses, we must call PURGEIF regardless of
1508 		 * addresses.  (Protocols which might store ifnet
1509 		 * pointers are marked with PR_PURGEIF.)
1510 		 */
1511 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
1512 		{
1513 			so.so_proto = pr;
1514 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1515 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1516 		}
1517 	}
1518 
1519 	/*
1520 	 * Must be done after the above pr_purgeif because if_psref may be
1521 	 * still used in pr_purgeif.
1522 	 */
1523 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1524 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1525 
1526 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1527 	(void)pfil_head_destroy(ifp->if_pfil);
1528 
1529 	/* Announce that the interface is gone. */
1530 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1531 
1532 	IF_AFDATA_LOCK_DESTROY(ifp);
1533 
1534 	if (ifp->if_percpuq != NULL) {
1535 		if_percpuq_destroy(ifp->if_percpuq);
1536 		ifp->if_percpuq = NULL;
1537 	}
1538 
1539 	mutex_obj_free(ifp->if_ioctl_lock);
1540 	ifp->if_ioctl_lock = NULL;
1541 	mutex_obj_free(ifp->if_snd.ifq_lock);
1542 	if_stats_fini(ifp);
1543 	KASSERT(!simplehook_has_hooks(ifp->if_linkstate_hooks));
1544 	simplehook_destroy(ifp->if_linkstate_hooks);
1545 
1546 	splx(s);
1547 
1548 #ifdef IFAREF_DEBUG
1549 	if_check_and_free_ifa_list(ifp);
1550 #endif
1551 }
1552 
1553 /*
1554  * Callback for a radix tree walk to delete all references to an
1555  * ifnet.
1556  */
1557 static int
1558 if_delroute_matcher(struct rtentry *rt, void *v)
1559 {
1560 	struct ifnet *ifp = (struct ifnet *)v;
1561 
1562 	if (rt->rt_ifp == ifp)
1563 		return 1;
1564 	else
1565 		return 0;
1566 }
1567 
1568 /*
1569  * Create a clone network interface.
1570  */
1571 static int
1572 if_clone_create(const char *name)
1573 {
1574 	struct if_clone *ifc;
1575 	struct ifnet *ifp;
1576 	struct psref psref;
1577 	int unit;
1578 
1579 	KASSERT(mutex_owned(&if_clone_mtx));
1580 
1581 	ifc = if_clone_lookup(name, &unit);
1582 	if (ifc == NULL)
1583 		return EINVAL;
1584 
1585 	ifp = if_get(name, &psref);
1586 	if (ifp != NULL) {
1587 		if_put(ifp, &psref);
1588 		return EEXIST;
1589 	}
1590 
1591 	return (*ifc->ifc_create)(ifc, unit);
1592 }
1593 
1594 /*
1595  * Destroy a clone network interface.
1596  */
1597 static int
1598 if_clone_destroy(const char *name)
1599 {
1600 	struct if_clone *ifc;
1601 	struct ifnet *ifp;
1602 	struct psref psref;
1603 	int error;
1604 	int (*if_ioctlfn)(struct ifnet *, u_long, void *);
1605 
1606 	KASSERT(mutex_owned(&if_clone_mtx));
1607 
1608 	ifc = if_clone_lookup(name, NULL);
1609 	if (ifc == NULL)
1610 		return EINVAL;
1611 
1612 	if (ifc->ifc_destroy == NULL)
1613 		return EOPNOTSUPP;
1614 
1615 	ifp = if_get(name, &psref);
1616 	if (ifp == NULL)
1617 		return ENXIO;
1618 
1619 	/* We have to disable ioctls here */
1620 	IFNET_LOCK(ifp);
1621 	if_ioctlfn = ifp->if_ioctl;
1622 	ifp->if_ioctl = if_nullioctl;
1623 	IFNET_UNLOCK(ifp);
1624 
1625 	/*
1626 	 * We cannot call ifc_destroy with holding ifp.
1627 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1628 	 */
1629 	if_put(ifp, &psref);
1630 
1631 	error = (*ifc->ifc_destroy)(ifp);
1632 
1633 	if (error != 0) {
1634 		/* We have to restore if_ioctl on error */
1635 		IFNET_LOCK(ifp);
1636 		ifp->if_ioctl = if_ioctlfn;
1637 		IFNET_UNLOCK(ifp);
1638 	}
1639 
1640 	return error;
1641 }
1642 
1643 static bool
1644 if_is_unit(const char *name)
1645 {
1646 
1647 	while (*name != '\0') {
1648 		if (*name < '0' || *name > '9')
1649 			return false;
1650 		name++;
1651 	}
1652 
1653 	return true;
1654 }
1655 
1656 /*
1657  * Look up a network interface cloner.
1658  */
1659 static struct if_clone *
1660 if_clone_lookup(const char *name, int *unitp)
1661 {
1662 	struct if_clone *ifc;
1663 	const char *cp;
1664 	char *dp, ifname[IFNAMSIZ + 3];
1665 	int unit;
1666 
1667 	KASSERT(mutex_owned(&if_clone_mtx));
1668 
1669 	strcpy(ifname, "if_");
1670 	/* separate interface name from unit */
1671 	/* TODO: search unit number from backward */
1672 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1673 	    *cp && !if_is_unit(cp);)
1674 		*dp++ = *cp++;
1675 
1676 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1677 		return NULL;	/* No name or unit number */
1678 	*dp++ = '\0';
1679 
1680 again:
1681 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1682 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1683 			break;
1684 	}
1685 
1686 	if (ifc == NULL) {
1687 		int error;
1688 		if (*ifname == '\0')
1689 			return NULL;
1690 		mutex_exit(&if_clone_mtx);
1691 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1692 		mutex_enter(&if_clone_mtx);
1693 		if (error)
1694 			return NULL;
1695 		*ifname = '\0';
1696 		goto again;
1697 	}
1698 
1699 	unit = 0;
1700 	while (cp - name < IFNAMSIZ && *cp) {
1701 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1702 			/* Bogus unit number. */
1703 			return NULL;
1704 		}
1705 		unit = (unit * 10) + (*cp++ - '0');
1706 	}
1707 
1708 	if (unitp != NULL)
1709 		*unitp = unit;
1710 	return ifc;
1711 }
1712 
1713 /*
1714  * Register a network interface cloner.
1715  */
1716 void
1717 if_clone_attach(struct if_clone *ifc)
1718 {
1719 
1720 	mutex_enter(&if_clone_mtx);
1721 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1722 	if_cloners_count++;
1723 	mutex_exit(&if_clone_mtx);
1724 }
1725 
1726 /*
1727  * Unregister a network interface cloner.
1728  */
1729 void
1730 if_clone_detach(struct if_clone *ifc)
1731 {
1732 
1733 	mutex_enter(&if_clone_mtx);
1734 	LIST_REMOVE(ifc, ifc_list);
1735 	if_cloners_count--;
1736 	mutex_exit(&if_clone_mtx);
1737 }
1738 
1739 /*
1740  * Provide list of interface cloners to userspace.
1741  */
1742 int
1743 if_clone_list(int buf_count, char *buffer, int *total)
1744 {
1745 	char outbuf[IFNAMSIZ], *dst;
1746 	struct if_clone *ifc;
1747 	int count, error = 0;
1748 
1749 	mutex_enter(&if_clone_mtx);
1750 	*total = if_cloners_count;
1751 	if ((dst = buffer) == NULL) {
1752 		/* Just asking how many there are. */
1753 		goto out;
1754 	}
1755 
1756 	if (buf_count < 0) {
1757 		error = EINVAL;
1758 		goto out;
1759 	}
1760 
1761 	count = (if_cloners_count < buf_count) ? 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 	KASSERTMSG(ifa->ifa_refcnt > 0, "ifa_refcnt=%d", ifa->ifa_refcnt);
1799 
1800 #ifndef __HAVE_ATOMIC_AS_MEMBAR
1801 	membar_release();
1802 #endif
1803 	if (atomic_dec_uint_nv(&ifa->ifa_refcnt) != 0)
1804 		return;
1805 #ifndef __HAVE_ATOMIC_AS_MEMBAR
1806 	membar_acquire();
1807 #endif
1808 	free(ifa, M_IFADDR);
1809 }
1810 
1811 bool
1812 ifa_is_destroying(struct ifaddr *ifa)
1813 {
1814 
1815 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1816 }
1817 
1818 void
1819 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1820 {
1821 
1822 	ifa->ifa_ifp = ifp;
1823 
1824 	/*
1825 	 * Check MP-safety for IFEF_MPSAFE drivers.
1826 	 * Check !IFF_RUNNING for initialization routines that normally don't
1827 	 * take IFNET_LOCK but it's safe because there is no competitor.
1828 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1829 	 * if_stop.
1830 	 */
1831 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1832 	    IFNET_LOCKED(ifp));
1833 
1834 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1835 	IFADDR_ENTRY_INIT(ifa);
1836 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1837 
1838 	ifaref(ifa);
1839 }
1840 
1841 void
1842 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1843 {
1844 
1845 	KASSERT(ifa->ifa_ifp == ifp);
1846 	/*
1847 	 * Check MP-safety for IFEF_MPSAFE drivers.
1848 	 * if_is_deactivated indicates ifa_remove is called from if_detach
1849 	 * where it is safe even if IFNET_LOCK isn't held.
1850 	 */
1851 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) ||
1852 	    IFNET_LOCKED(ifp));
1853 
1854 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1855 	IFADDR_WRITER_REMOVE(ifa);
1856 #ifdef NET_MPSAFE
1857 	IFNET_GLOBAL_LOCK();
1858 	pserialize_perform(ifnet_psz);
1859 	IFNET_GLOBAL_UNLOCK();
1860 #endif
1861 
1862 #ifdef NET_MPSAFE
1863 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1864 #endif
1865 	IFADDR_ENTRY_DESTROY(ifa);
1866 	ifafree(ifa);
1867 }
1868 
1869 void
1870 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1871 {
1872 
1873 	PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
1874 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1875 }
1876 
1877 void
1878 ifa_release(struct ifaddr *ifa, struct psref *psref)
1879 {
1880 
1881 	if (ifa == NULL)
1882 		return;
1883 
1884 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1885 }
1886 
1887 bool
1888 ifa_held(struct ifaddr *ifa)
1889 {
1890 
1891 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1892 }
1893 
1894 static inline int
1895 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1896 {
1897 
1898 	return sockaddr_cmp(sa1, sa2) == 0;
1899 }
1900 
1901 /*
1902  * Locate an interface based on a complete address.
1903  */
1904 /*ARGSUSED*/
1905 struct ifaddr *
1906 ifa_ifwithaddr(const struct sockaddr *addr)
1907 {
1908 	struct ifnet *ifp;
1909 	struct ifaddr *ifa;
1910 
1911 	IFNET_READER_FOREACH(ifp) {
1912 		if (if_is_deactivated(ifp))
1913 			continue;
1914 		IFADDR_READER_FOREACH(ifa, ifp) {
1915 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1916 				continue;
1917 			if (equal(addr, ifa->ifa_addr))
1918 				return ifa;
1919 			if ((ifp->if_flags & IFF_BROADCAST) &&
1920 			    ifa->ifa_broadaddr &&
1921 			    /* IP6 doesn't have broadcast */
1922 			    ifa->ifa_broadaddr->sa_len != 0 &&
1923 			    equal(ifa->ifa_broadaddr, addr))
1924 				return ifa;
1925 		}
1926 	}
1927 	return NULL;
1928 }
1929 
1930 struct ifaddr *
1931 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1932 {
1933 	struct ifaddr *ifa;
1934 	int s = pserialize_read_enter();
1935 
1936 	ifa = ifa_ifwithaddr(addr);
1937 	if (ifa != NULL)
1938 		ifa_acquire(ifa, psref);
1939 	pserialize_read_exit(s);
1940 
1941 	return ifa;
1942 }
1943 
1944 /*
1945  * Locate the point to point interface with a given destination address.
1946  */
1947 /*ARGSUSED*/
1948 struct ifaddr *
1949 ifa_ifwithdstaddr(const struct sockaddr *addr)
1950 {
1951 	struct ifnet *ifp;
1952 	struct ifaddr *ifa;
1953 
1954 	IFNET_READER_FOREACH(ifp) {
1955 		if (if_is_deactivated(ifp))
1956 			continue;
1957 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1958 			continue;
1959 		IFADDR_READER_FOREACH(ifa, ifp) {
1960 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1961 			    ifa->ifa_dstaddr == NULL)
1962 				continue;
1963 			if (equal(addr, ifa->ifa_dstaddr))
1964 				return ifa;
1965 		}
1966 	}
1967 
1968 	return NULL;
1969 }
1970 
1971 struct ifaddr *
1972 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1973 {
1974 	struct ifaddr *ifa;
1975 	int s;
1976 
1977 	s = pserialize_read_enter();
1978 	ifa = ifa_ifwithdstaddr(addr);
1979 	if (ifa != NULL)
1980 		ifa_acquire(ifa, psref);
1981 	pserialize_read_exit(s);
1982 
1983 	return ifa;
1984 }
1985 
1986 /*
1987  * Find an interface on a specific network.  If many, choice
1988  * is most specific found.
1989  */
1990 struct ifaddr *
1991 ifa_ifwithnet(const struct sockaddr *addr)
1992 {
1993 	struct ifnet *ifp;
1994 	struct ifaddr *ifa, *ifa_maybe = NULL;
1995 	const struct sockaddr_dl *sdl;
1996 	u_int af = addr->sa_family;
1997 	const char *addr_data = addr->sa_data, *cplim;
1998 
1999 	if (af == AF_LINK) {
2000 		sdl = satocsdl(addr);
2001 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
2002 		    ifindex2ifnet[sdl->sdl_index] &&
2003 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
2004 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
2005 		}
2006 	}
2007 #ifdef NETATALK
2008 	if (af == AF_APPLETALK) {
2009 		const struct sockaddr_at *sat, *sat2;
2010 		sat = (const struct sockaddr_at *)addr;
2011 		IFNET_READER_FOREACH(ifp) {
2012 			if (if_is_deactivated(ifp))
2013 				continue;
2014 			ifa = at_ifawithnet((const struct sockaddr_at *)addr,
2015 			    ifp);
2016 			if (ifa == NULL)
2017 				continue;
2018 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
2019 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
2020 				return ifa; /* exact match */
2021 			if (ifa_maybe == NULL) {
2022 				/* else keep the if with the right range */
2023 				ifa_maybe = ifa;
2024 			}
2025 		}
2026 		return ifa_maybe;
2027 	}
2028 #endif
2029 	IFNET_READER_FOREACH(ifp) {
2030 		if (if_is_deactivated(ifp))
2031 			continue;
2032 		IFADDR_READER_FOREACH(ifa, ifp) {
2033 			const char *cp, *cp2, *cp3;
2034 
2035 			if (ifa->ifa_addr->sa_family != af ||
2036 			    ifa->ifa_netmask == NULL)
2037  next:				continue;
2038 			cp = addr_data;
2039 			cp2 = ifa->ifa_addr->sa_data;
2040 			cp3 = ifa->ifa_netmask->sa_data;
2041 			cplim = (const char *)ifa->ifa_netmask +
2042 			    ifa->ifa_netmask->sa_len;
2043 			while (cp3 < cplim) {
2044 				if ((*cp++ ^ *cp2++) & *cp3++) {
2045 					/* want to continue for() loop */
2046 					goto next;
2047 				}
2048 			}
2049 			if (ifa_maybe == NULL ||
2050 			    rt_refines(ifa->ifa_netmask,
2051 				       ifa_maybe->ifa_netmask))
2052 				ifa_maybe = ifa;
2053 		}
2054 	}
2055 	return ifa_maybe;
2056 }
2057 
2058 struct ifaddr *
2059 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2060 {
2061 	struct ifaddr *ifa;
2062 	int s;
2063 
2064 	s = pserialize_read_enter();
2065 	ifa = ifa_ifwithnet(addr);
2066 	if (ifa != NULL)
2067 		ifa_acquire(ifa, psref);
2068 	pserialize_read_exit(s);
2069 
2070 	return ifa;
2071 }
2072 
2073 /*
2074  * Find the interface of the address.
2075  */
2076 struct ifaddr *
2077 ifa_ifwithladdr(const struct sockaddr *addr)
2078 {
2079 	struct ifaddr *ia;
2080 
2081 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2082 	    (ia = ifa_ifwithnet(addr)))
2083 		return ia;
2084 	return NULL;
2085 }
2086 
2087 struct ifaddr *
2088 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2089 {
2090 	struct ifaddr *ifa;
2091 	int s;
2092 
2093 	s = pserialize_read_enter();
2094 	ifa = ifa_ifwithladdr(addr);
2095 	if (ifa != NULL)
2096 		ifa_acquire(ifa, psref);
2097 	pserialize_read_exit(s);
2098 
2099 	return ifa;
2100 }
2101 
2102 /*
2103  * Find an interface using a specific address family
2104  */
2105 struct ifaddr *
2106 ifa_ifwithaf(int af)
2107 {
2108 	struct ifnet *ifp;
2109 	struct ifaddr *ifa = NULL;
2110 	int s;
2111 
2112 	s = pserialize_read_enter();
2113 	IFNET_READER_FOREACH(ifp) {
2114 		if (if_is_deactivated(ifp))
2115 			continue;
2116 		IFADDR_READER_FOREACH(ifa, ifp) {
2117 			if (ifa->ifa_addr->sa_family == af)
2118 				goto out;
2119 		}
2120 	}
2121 out:
2122 	pserialize_read_exit(s);
2123 	return ifa;
2124 }
2125 
2126 /*
2127  * Find an interface address specific to an interface best matching
2128  * a given address.
2129  */
2130 struct ifaddr *
2131 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2132 {
2133 	struct ifaddr *ifa;
2134 	const char *cp, *cp2, *cp3;
2135 	const char *cplim;
2136 	struct ifaddr *ifa_maybe = 0;
2137 	u_int af = addr->sa_family;
2138 
2139 	if (if_is_deactivated(ifp))
2140 		return NULL;
2141 
2142 	if (af >= AF_MAX)
2143 		return NULL;
2144 
2145 	IFADDR_READER_FOREACH(ifa, ifp) {
2146 		if (ifa->ifa_addr->sa_family != af)
2147 			continue;
2148 		ifa_maybe = ifa;
2149 		if (ifa->ifa_netmask == NULL) {
2150 			if (equal(addr, ifa->ifa_addr) ||
2151 			    (ifa->ifa_dstaddr &&
2152 			     equal(addr, ifa->ifa_dstaddr)))
2153 				return ifa;
2154 			continue;
2155 		}
2156 		cp = addr->sa_data;
2157 		cp2 = ifa->ifa_addr->sa_data;
2158 		cp3 = ifa->ifa_netmask->sa_data;
2159 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2160 		for (; cp3 < cplim; cp3++) {
2161 			if ((*cp++ ^ *cp2++) & *cp3)
2162 				break;
2163 		}
2164 		if (cp3 == cplim)
2165 			return ifa;
2166 	}
2167 	return ifa_maybe;
2168 }
2169 
2170 struct ifaddr *
2171 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2172     struct psref *psref)
2173 {
2174 	struct ifaddr *ifa;
2175 	int s;
2176 
2177 	s = pserialize_read_enter();
2178 	ifa = ifaof_ifpforaddr(addr, ifp);
2179 	if (ifa != NULL)
2180 		ifa_acquire(ifa, psref);
2181 	pserialize_read_exit(s);
2182 
2183 	return ifa;
2184 }
2185 
2186 /*
2187  * Default action when installing a route with a Link Level gateway.
2188  * Lookup an appropriate real ifa to point to.
2189  * This should be moved to /sys/net/link.c eventually.
2190  */
2191 void
2192 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2193 {
2194 	struct ifaddr *ifa;
2195 	const struct sockaddr *dst;
2196 	struct ifnet *ifp;
2197 	struct psref psref;
2198 
2199 	if (cmd != RTM_ADD || ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2200 		return;
2201 	ifp = rt->rt_ifa->ifa_ifp;
2202 	dst = rt_getkey(rt);
2203 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2204 		rt_replace_ifa(rt, ifa);
2205 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2206 			ifa->ifa_rtrequest(cmd, rt, info);
2207 		ifa_release(ifa, &psref);
2208 	}
2209 }
2210 
2211 /*
2212  * bitmask macros to manage a densely packed link_state change queue.
2213  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2214  * LINK_STATE_UP(2) we need 2 bits for each state change.
2215  * As a state change to store is 0, treat all bits set as an unset item.
2216  */
2217 #define LQ_ITEM_BITS		2
2218 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2219 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2220 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2221 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2222 #define LQ_STORE(q, i, v)						      \
2223 	do {								      \
2224 		(q) &= ~LQ_MASK((i));					      \
2225 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2226 	} while (0 /* CONSTCOND */)
2227 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2228 #define LQ_POP(q, v)							      \
2229 	do {								      \
2230 		(v) = LQ_ITEM((q), 0);					      \
2231 		(q) >>= LQ_ITEM_BITS;					      \
2232 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2233 	} while (0 /* CONSTCOND */)
2234 #define LQ_PUSH(q, v)							      \
2235 	do {								      \
2236 		(q) >>= LQ_ITEM_BITS;					      \
2237 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2238 	} while (0 /* CONSTCOND */)
2239 #define LQ_FIND_UNSET(q, i)						      \
2240 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2241 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2242 			break;						      \
2243 	}
2244 
2245 /*
2246  * Handle a change in the interface link state and
2247  * queue notifications.
2248  */
2249 void
2250 if_link_state_change(struct ifnet *ifp, int link_state)
2251 {
2252 	int idx;
2253 
2254 	/* Ensure change is to a valid state */
2255 	switch (link_state) {
2256 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2257 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2258 	case LINK_STATE_UP:
2259 		break;
2260 	default:
2261 #ifdef DEBUG
2262 		printf("%s: invalid link state %d\n",
2263 		    ifp->if_xname, link_state);
2264 #endif
2265 		return;
2266 	}
2267 
2268 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2269 
2270 	/* Find the last unset event in the queue. */
2271 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2272 
2273 	if (idx == 0) {
2274 		/*
2275 		 * There is no queue of link state changes.
2276 		 * As we have the lock we can safely compare against the
2277 		 * current link state and return if the same.
2278 		 * Otherwise, if scheduled is true then the interface is being
2279 		 * detached and the queue is being drained so we need
2280 		 * to avoid queuing more work.
2281 		 */
2282 		 if (ifp->if_link_state == link_state ||
2283 		     ifp->if_link_scheduled)
2284 			goto out;
2285 	} else {
2286 		/* Ensure link_state doesn't match the last queued state. */
2287 		if (LQ_ITEM(ifp->if_link_queue, idx - 1)
2288 		    == (uint8_t)link_state)
2289 			goto out;
2290 	}
2291 
2292 	/* Handle queue overflow. */
2293 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2294 		uint8_t lost;
2295 
2296 		/*
2297 		 * The DOWN state must be protected from being pushed off
2298 		 * the queue to ensure that userland will always be
2299 		 * in a sane state.
2300 		 * Because DOWN is protected, there is no need to protect
2301 		 * UNKNOWN.
2302 		 * It should be invalid to change from any other state to
2303 		 * UNKNOWN anyway ...
2304 		 */
2305 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2306 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2307 		if (lost == LINK_STATE_DOWN) {
2308 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2309 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2310 		}
2311 		printf("%s: lost link state change %s\n",
2312 		    ifp->if_xname,
2313 		    lost == LINK_STATE_UP ? "UP" :
2314 		    lost == LINK_STATE_DOWN ? "DOWN" :
2315 		    "UNKNOWN");
2316 	} else
2317 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2318 
2319 	if (ifp->if_link_scheduled)
2320 		goto out;
2321 
2322 	ifp->if_link_scheduled = true;
2323 	workqueue_enqueue(ifnet_link_state_wq, &ifp->if_link_work, NULL);
2324 
2325 out:
2326 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2327 }
2328 
2329 /*
2330  * Handle interface link state change notifications.
2331  */
2332 static void
2333 if_link_state_change_process(struct ifnet *ifp, int link_state)
2334 {
2335 	struct domain *dp;
2336 	const int s = splnet();
2337 	bool notify;
2338 
2339 	KASSERT(!cpu_intr_p());
2340 
2341 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2342 
2343 	/* Ensure the change is still valid. */
2344 	if (ifp->if_link_state == link_state) {
2345 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2346 		splx(s);
2347 		return;
2348 	}
2349 
2350 #ifdef DEBUG
2351 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2352 		link_state == LINK_STATE_UP ? "UP" :
2353 		link_state == LINK_STATE_DOWN ? "DOWN" :
2354 		"UNKNOWN",
2355 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2356 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2357 		"UNKNOWN");
2358 #endif
2359 
2360 	/*
2361 	 * When going from UNKNOWN to UP, we need to mark existing
2362 	 * addresses as tentative and restart DAD as we may have
2363 	 * erroneously not found a duplicate.
2364 	 *
2365 	 * This needs to happen before rt_ifmsg to avoid a race where
2366 	 * listeners would have an address and expect it to work right
2367 	 * away.
2368 	 */
2369 	notify = (link_state == LINK_STATE_UP &&
2370 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2371 	ifp->if_link_state = link_state;
2372 	/* The following routines may sleep so release the spin mutex */
2373 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2374 
2375 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2376 	if (notify) {
2377 		DOMAIN_FOREACH(dp) {
2378 			if (dp->dom_if_link_state_change != NULL)
2379 				dp->dom_if_link_state_change(ifp,
2380 				    LINK_STATE_DOWN);
2381 		}
2382 	}
2383 
2384 	/* Notify that the link state has changed. */
2385 	rt_ifmsg(ifp);
2386 
2387 	simplehook_dohooks(ifp->if_linkstate_hooks);
2388 
2389 	DOMAIN_FOREACH(dp) {
2390 		if (dp->dom_if_link_state_change != NULL)
2391 			dp->dom_if_link_state_change(ifp, link_state);
2392 	}
2393 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2394 	splx(s);
2395 }
2396 
2397 /*
2398  * Process the interface link state change queue.
2399  */
2400 static void
2401 if_link_state_change_work(struct work *work, void *arg)
2402 {
2403 	struct ifnet *ifp = container_of(work, struct ifnet, if_link_work);
2404 	uint8_t state;
2405 
2406 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2407 	const int s = splnet();
2408 
2409 	/*
2410 	 * Pop a link state change from the queue and process it.
2411 	 * If there is nothing to process then if_detach() has been called.
2412 	 * We keep if_link_scheduled = true so the queue can safely drain
2413 	 * without more work being queued.
2414 	 */
2415 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2416 	LQ_POP(ifp->if_link_queue, state);
2417 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2418 	if (state == LINK_STATE_UNSET)
2419 		goto out;
2420 
2421 	if_link_state_change_process(ifp, state);
2422 
2423 	/* If there is a link state change to come, schedule it. */
2424 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2425 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET) {
2426 		ifp->if_link_scheduled = true;
2427 		workqueue_enqueue(ifnet_link_state_wq, &ifp->if_link_work,
2428 		    NULL);
2429 	} else
2430 		ifp->if_link_scheduled = false;
2431 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2432 
2433 out:
2434 	splx(s);
2435 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2436 }
2437 
2438 void *
2439 if_linkstate_change_establish(struct ifnet *ifp, void (*fn)(void *), void *arg)
2440 {
2441 	khook_t *hk;
2442 
2443 	hk = simplehook_establish(ifp->if_linkstate_hooks, fn, arg);
2444 
2445 	return (void *)hk;
2446 }
2447 
2448 void
2449 if_linkstate_change_disestablish(struct ifnet *ifp, void *vhook,
2450     kmutex_t *lock)
2451 {
2452 
2453 	simplehook_disestablish(ifp->if_linkstate_hooks, vhook, lock);
2454 }
2455 
2456 /*
2457  * Used to mark addresses on an interface as DETATCHED or TENTATIVE
2458  * and thus start Duplicate Address Detection without changing the
2459  * real link state.
2460  */
2461 void
2462 if_domain_link_state_change(struct ifnet *ifp, int link_state)
2463 {
2464 	struct domain *dp;
2465 
2466 	const int s = splnet();
2467 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2468 
2469 	DOMAIN_FOREACH(dp) {
2470 		if (dp->dom_if_link_state_change != NULL)
2471 			dp->dom_if_link_state_change(ifp, link_state);
2472 	}
2473 
2474 	splx(s);
2475 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2476 }
2477 
2478 /*
2479  * Default action when installing a local route on a point-to-point
2480  * interface.
2481  */
2482 void
2483 p2p_rtrequest(int req, struct rtentry *rt,
2484     __unused const struct rt_addrinfo *info)
2485 {
2486 	struct ifnet *ifp = rt->rt_ifp;
2487 	struct ifaddr *ifa, *lo0ifa;
2488 	int s = pserialize_read_enter();
2489 
2490 	switch (req) {
2491 	case RTM_ADD:
2492 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2493 			break;
2494 
2495 		rt->rt_ifp = lo0ifp;
2496 
2497 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2498 			break;
2499 
2500 		IFADDR_READER_FOREACH(ifa, ifp) {
2501 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2502 				break;
2503 		}
2504 		if (ifa == NULL)
2505 			break;
2506 
2507 		/*
2508 		 * Ensure lo0 has an address of the same family.
2509 		 */
2510 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2511 			if (lo0ifa->ifa_addr->sa_family ==
2512 			    ifa->ifa_addr->sa_family)
2513 				break;
2514 		}
2515 		if (lo0ifa == NULL)
2516 			break;
2517 
2518 		/*
2519 		 * Make sure to set rt->rt_ifa to the interface
2520 		 * address we are using, otherwise we will have trouble
2521 		 * with source address selection.
2522 		 */
2523 		if (ifa != rt->rt_ifa)
2524 			rt_replace_ifa(rt, ifa);
2525 		break;
2526 	case RTM_DELETE:
2527 	default:
2528 		break;
2529 	}
2530 	pserialize_read_exit(s);
2531 }
2532 
2533 static void
2534 _if_down(struct ifnet *ifp)
2535 {
2536 	struct ifaddr *ifa;
2537 	struct domain *dp;
2538 	struct psref psref;
2539 
2540 	ifp->if_flags &= ~IFF_UP;
2541 	nanotime(&ifp->if_lastchange);
2542 
2543 	const int bound = curlwp_bind();
2544 	int s = pserialize_read_enter();
2545 	IFADDR_READER_FOREACH(ifa, ifp) {
2546 		ifa_acquire(ifa, &psref);
2547 		pserialize_read_exit(s);
2548 
2549 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2550 
2551 		s = pserialize_read_enter();
2552 		ifa_release(ifa, &psref);
2553 	}
2554 	pserialize_read_exit(s);
2555 	curlwp_bindx(bound);
2556 
2557 	IFQ_PURGE(&ifp->if_snd);
2558 #if NCARP > 0
2559 	if (ifp->if_carp)
2560 		carp_carpdev_state(ifp);
2561 #endif
2562 	rt_ifmsg(ifp);
2563 	DOMAIN_FOREACH(dp) {
2564 		if (dp->dom_if_down)
2565 			dp->dom_if_down(ifp);
2566 	}
2567 }
2568 
2569 static void
2570 if_down_deactivated(struct ifnet *ifp)
2571 {
2572 
2573 	KASSERT(if_is_deactivated(ifp));
2574 	_if_down(ifp);
2575 }
2576 
2577 void
2578 if_down_locked(struct ifnet *ifp)
2579 {
2580 
2581 	KASSERT(IFNET_LOCKED(ifp));
2582 	_if_down(ifp);
2583 }
2584 
2585 /*
2586  * Mark an interface down and notify protocols of
2587  * the transition.
2588  * NOTE: must be called at splsoftnet or equivalent.
2589  */
2590 void
2591 if_down(struct ifnet *ifp)
2592 {
2593 
2594 	IFNET_LOCK(ifp);
2595 	if_down_locked(ifp);
2596 	IFNET_UNLOCK(ifp);
2597 }
2598 
2599 /*
2600  * Must be called with holding if_ioctl_lock.
2601  */
2602 static void
2603 if_up_locked(struct ifnet *ifp)
2604 {
2605 #ifdef notyet
2606 	struct ifaddr *ifa;
2607 #endif
2608 	struct domain *dp;
2609 
2610 	KASSERT(IFNET_LOCKED(ifp));
2611 
2612 	KASSERT(!if_is_deactivated(ifp));
2613 	ifp->if_flags |= IFF_UP;
2614 	nanotime(&ifp->if_lastchange);
2615 #ifdef notyet
2616 	/* this has no effect on IP, and will kill all ISO connections XXX */
2617 	IFADDR_READER_FOREACH(ifa, ifp)
2618 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2619 #endif
2620 #if NCARP > 0
2621 	if (ifp->if_carp)
2622 		carp_carpdev_state(ifp);
2623 #endif
2624 	rt_ifmsg(ifp);
2625 	DOMAIN_FOREACH(dp) {
2626 		if (dp->dom_if_up)
2627 			dp->dom_if_up(ifp);
2628 	}
2629 }
2630 
2631 /*
2632  * Handle interface slowtimo timer routine.  Called
2633  * from softclock, we decrement timer (if set) and
2634  * call the appropriate interface routine on expiration.
2635  */
2636 static bool
2637 if_slowtimo_countdown(struct ifnet *ifp)
2638 {
2639 	bool fire = false;
2640 	const int s = splnet();
2641 
2642 	KERNEL_LOCK(1, NULL);
2643 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2644 		fire = true;
2645 	KERNEL_UNLOCK_ONE(NULL);
2646 	splx(s);
2647 
2648 	return fire;
2649 }
2650 
2651 static void
2652 if_slowtimo_intr(void *arg)
2653 {
2654 	struct ifnet *ifp = arg;
2655 	struct if_slowtimo_data *isd = ifp->if_slowtimo_data;
2656 
2657 	mutex_enter(&isd->isd_lock);
2658 	if (!isd->isd_dying) {
2659 		if (isd->isd_trigger || if_slowtimo_countdown(ifp)) {
2660 			if (!isd->isd_queued) {
2661 				isd->isd_queued = true;
2662 				workqueue_enqueue(if_slowtimo_wq,
2663 				    &isd->isd_work, NULL);
2664 			}
2665 		} else
2666 			callout_schedule(&isd->isd_ch, hz / IFNET_SLOWHZ);
2667 	}
2668 	mutex_exit(&isd->isd_lock);
2669 }
2670 
2671 static void
2672 if_slowtimo_work(struct work *work, void *arg)
2673 {
2674 	struct if_slowtimo_data *isd =
2675 	    container_of(work, struct if_slowtimo_data, isd_work);
2676 	struct ifnet *ifp = isd->isd_ifp;
2677 	const int s = splnet();
2678 
2679 	KERNEL_LOCK(1, NULL);
2680 	(*ifp->if_slowtimo)(ifp);
2681 	KERNEL_UNLOCK_ONE(NULL);
2682 	splx(s);
2683 
2684 	mutex_enter(&isd->isd_lock);
2685 	if (isd->isd_trigger) {
2686 		isd->isd_trigger = false;
2687 		printf("%s: watchdog triggered\n", ifp->if_xname);
2688 	}
2689 	isd->isd_queued = false;
2690 	if (!isd->isd_dying)
2691 		callout_schedule(&isd->isd_ch, hz / IFNET_SLOWHZ);
2692 	mutex_exit(&isd->isd_lock);
2693 }
2694 
2695 static int
2696 sysctl_if_watchdog(SYSCTLFN_ARGS)
2697 {
2698 	struct sysctlnode node = *rnode;
2699 	struct ifnet *ifp = node.sysctl_data;
2700 	struct if_slowtimo_data *isd = ifp->if_slowtimo_data;
2701 	int arg = 0;
2702 	int error;
2703 
2704 	node.sysctl_data = &arg;
2705 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
2706 	if (error || newp == NULL)
2707 		return error;
2708 	if (arg) {
2709 		mutex_enter(&isd->isd_lock);
2710 		KASSERT(!isd->isd_dying);
2711 		isd->isd_trigger = true;
2712 		callout_schedule(&isd->isd_ch, 0);
2713 		mutex_exit(&isd->isd_lock);
2714 	}
2715 
2716 	return 0;
2717 }
2718 
2719 static void
2720 sysctl_watchdog_setup(struct ifnet *ifp)
2721 {
2722 	struct sysctllog **clog = &ifp->if_sysctl_log;
2723 	const struct sysctlnode *rnode;
2724 
2725 	if (sysctl_createv(clog, 0, NULL, &rnode,
2726 		CTLFLAG_PERMANENT, CTLTYPE_NODE, "interfaces",
2727 		SYSCTL_DESCR("Per-interface controls"),
2728 		NULL, 0, NULL, 0,
2729 		CTL_NET, CTL_CREATE, CTL_EOL) != 0)
2730 		goto bad;
2731 	if (sysctl_createv(clog, 0, &rnode, &rnode,
2732 		CTLFLAG_PERMANENT, CTLTYPE_NODE, ifp->if_xname,
2733 		SYSCTL_DESCR("Interface controls"),
2734 		NULL, 0, NULL, 0,
2735 		CTL_CREATE, CTL_EOL) != 0)
2736 		goto bad;
2737 	if (sysctl_createv(clog, 0, &rnode, &rnode,
2738 		CTLFLAG_PERMANENT, CTLTYPE_NODE, "watchdog",
2739 		SYSCTL_DESCR("Interface watchdog controls"),
2740 		NULL, 0, NULL, 0,
2741 		CTL_CREATE, CTL_EOL) != 0)
2742 		goto bad;
2743 	if (sysctl_createv(clog, 0, &rnode, NULL,
2744 		CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_INT, "trigger",
2745 		SYSCTL_DESCR("Trigger watchdog timeout"),
2746 		sysctl_if_watchdog, 0, (int *)ifp, 0,
2747 		CTL_CREATE, CTL_EOL) != 0)
2748 		goto bad;
2749 
2750 	return;
2751 
2752 bad:
2753 	printf("%s: could not attach sysctl watchdog nodes\n", ifp->if_xname);
2754 }
2755 
2756 /*
2757  * Mark an interface up and notify protocols of
2758  * the transition.
2759  * NOTE: must be called at splsoftnet or equivalent.
2760  */
2761 void
2762 if_up(struct ifnet *ifp)
2763 {
2764 
2765 	IFNET_LOCK(ifp);
2766 	if_up_locked(ifp);
2767 	IFNET_UNLOCK(ifp);
2768 }
2769 
2770 /*
2771  * Set/clear promiscuous mode on interface ifp based on the truth value
2772  * of pswitch.  The calls are reference counted so that only the first
2773  * "on" request actually has an effect, as does the final "off" request.
2774  * Results are undefined if the "off" and "on" requests are not matched.
2775  */
2776 int
2777 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2778 {
2779 	int pcount, ret = 0;
2780 	u_short nflags;
2781 
2782 	KASSERT(IFNET_LOCKED(ifp));
2783 
2784 	pcount = ifp->if_pcount;
2785 	if (pswitch) {
2786 		/*
2787 		 * Allow the device to be "placed" into promiscuous
2788 		 * mode even if it is not configured up.  It will
2789 		 * consult IFF_PROMISC when it is brought up.
2790 		 */
2791 		if (ifp->if_pcount++ != 0)
2792 			goto out;
2793 		nflags = ifp->if_flags | IFF_PROMISC;
2794 	} else {
2795 		if (--ifp->if_pcount > 0)
2796 			goto out;
2797 		nflags = ifp->if_flags & ~IFF_PROMISC;
2798 	}
2799 	ret = if_flags_set(ifp, nflags);
2800 	/* Restore interface state if not successful. */
2801 	if (ret != 0)
2802 		ifp->if_pcount = pcount;
2803 
2804 out:
2805 	return ret;
2806 }
2807 
2808 int
2809 ifpromisc(struct ifnet *ifp, int pswitch)
2810 {
2811 	int e;
2812 
2813 	IFNET_LOCK(ifp);
2814 	e = ifpromisc_locked(ifp, pswitch);
2815 	IFNET_UNLOCK(ifp);
2816 
2817 	return e;
2818 }
2819 
2820 /*
2821  * if_ioctl(ifp, cmd, data)
2822  *
2823  *	Apply an ioctl command to the interface.  Returns 0 on success,
2824  *	nonzero errno(3) number on failure.
2825  *
2826  *	For SIOCADDMULTI/SIOCDELMULTI, caller need not hold locks -- it
2827  *	is the driver's responsibility to take any internal locks.
2828  *	(Kernel logic should generally invoke these only through
2829  *	if_mcast_op.)
2830  *
2831  *	For all other ioctls, caller must hold ifp->if_ioctl_lock,
2832  *	a.k.a. IFNET_LOCK.  May sleep.
2833  */
2834 int
2835 if_ioctl(struct ifnet *ifp, u_long cmd, void *data)
2836 {
2837 
2838 	switch (cmd) {
2839 	case SIOCADDMULTI:
2840 	case SIOCDELMULTI:
2841 		break;
2842 	default:
2843 		KASSERTMSG(IFNET_LOCKED(ifp), "%s", ifp->if_xname);
2844 	}
2845 
2846 	return (*ifp->if_ioctl)(ifp, cmd, data);
2847 }
2848 
2849 /*
2850  * if_init(ifp)
2851  *
2852  *	Prepare the hardware underlying ifp to process packets
2853  *	according to its current configuration.  Returns 0 on success,
2854  *	nonzero errno(3) number on failure.
2855  *
2856  *	May sleep.  Caller must hold ifp->if_ioctl_lock, a.k.a
2857  *	IFNET_LOCK.
2858  */
2859 int
2860 if_init(struct ifnet *ifp)
2861 {
2862 
2863 	KASSERTMSG(IFNET_LOCKED(ifp), "%s", ifp->if_xname);
2864 
2865 	return (*ifp->if_init)(ifp);
2866 }
2867 
2868 /*
2869  * if_stop(ifp, disable)
2870  *
2871  *	Stop the hardware underlying ifp from processing packets.
2872  *
2873  *	If disable is true, ... XXX(?)
2874  *
2875  *	May sleep.  Caller must hold ifp->if_ioctl_lock, a.k.a
2876  *	IFNET_LOCK.
2877  */
2878 void
2879 if_stop(struct ifnet *ifp, int disable)
2880 {
2881 
2882 	KASSERTMSG(IFNET_LOCKED(ifp), "%s", ifp->if_xname);
2883 
2884 	(*ifp->if_stop)(ifp, disable);
2885 }
2886 
2887 /*
2888  * Map interface name to
2889  * interface structure pointer.
2890  */
2891 struct ifnet *
2892 ifunit(const char *name)
2893 {
2894 	struct ifnet *ifp;
2895 	const char *cp = name;
2896 	u_int unit = 0;
2897 	u_int i;
2898 
2899 	/*
2900 	 * If the entire name is a number, treat it as an ifindex.
2901 	 */
2902 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++)
2903 		unit = unit * 10 + (*cp - '0');
2904 
2905 	/*
2906 	 * If the number took all of the name, then it's a valid ifindex.
2907 	 */
2908 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2909 		return if_byindex(unit);
2910 
2911 	ifp = NULL;
2912 	const int s = pserialize_read_enter();
2913 	IFNET_READER_FOREACH(ifp) {
2914 		if (if_is_deactivated(ifp))
2915 			continue;
2916 		if (strcmp(ifp->if_xname, name) == 0)
2917 			goto out;
2918 	}
2919 out:
2920 	pserialize_read_exit(s);
2921 	return ifp;
2922 }
2923 
2924 /*
2925  * Get a reference of an ifnet object by an interface name.
2926  * The returned reference is protected by psref(9). The caller
2927  * must release a returned reference by if_put after use.
2928  */
2929 struct ifnet *
2930 if_get(const char *name, struct psref *psref)
2931 {
2932 	struct ifnet *ifp;
2933 	const char *cp = name;
2934 	u_int unit = 0;
2935 	u_int i;
2936 
2937 	/*
2938 	 * If the entire name is a number, treat it as an ifindex.
2939 	 */
2940 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++)
2941 		unit = unit * 10 + (*cp - '0');
2942 
2943 	/*
2944 	 * If the number took all of the name, then it's a valid ifindex.
2945 	 */
2946 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2947 		return if_get_byindex(unit, psref);
2948 
2949 	ifp = NULL;
2950 	const int s = pserialize_read_enter();
2951 	IFNET_READER_FOREACH(ifp) {
2952 		if (if_is_deactivated(ifp))
2953 			continue;
2954 		if (strcmp(ifp->if_xname, name) == 0) {
2955 			PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2956 			psref_acquire(psref, &ifp->if_psref,
2957 			    ifnet_psref_class);
2958 			goto out;
2959 		}
2960 	}
2961 out:
2962 	pserialize_read_exit(s);
2963 	return ifp;
2964 }
2965 
2966 /*
2967  * Release a reference of an ifnet object given by if_get, if_get_byindex
2968  * or if_get_bylla.
2969  */
2970 void
2971 if_put(const struct ifnet *ifp, struct psref *psref)
2972 {
2973 
2974 	if (ifp == NULL)
2975 		return;
2976 
2977 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2978 }
2979 
2980 /*
2981  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2982  * should be used.
2983  */
2984 ifnet_t *
2985 _if_byindex(u_int idx)
2986 {
2987 
2988 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2989 }
2990 
2991 /*
2992  * Return ifp having idx. Return NULL if not found or the found ifp is
2993  * already deactivated.
2994  */
2995 ifnet_t *
2996 if_byindex(u_int idx)
2997 {
2998 	ifnet_t *ifp;
2999 
3000 	ifp = _if_byindex(idx);
3001 	if (ifp != NULL && if_is_deactivated(ifp))
3002 		ifp = NULL;
3003 	return ifp;
3004 }
3005 
3006 /*
3007  * Get a reference of an ifnet object by an interface index.
3008  * The returned reference is protected by psref(9). The caller
3009  * must release a returned reference by if_put after use.
3010  */
3011 ifnet_t *
3012 if_get_byindex(u_int idx, struct psref *psref)
3013 {
3014 	ifnet_t *ifp;
3015 
3016 	const int s = pserialize_read_enter();
3017 	ifp = if_byindex(idx);
3018 	if (__predict_true(ifp != NULL)) {
3019 		PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
3020 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
3021 	}
3022 	pserialize_read_exit(s);
3023 
3024 	return ifp;
3025 }
3026 
3027 ifnet_t *
3028 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
3029 {
3030 	ifnet_t *ifp;
3031 
3032 	const int s = pserialize_read_enter();
3033 	IFNET_READER_FOREACH(ifp) {
3034 		if (if_is_deactivated(ifp))
3035 			continue;
3036 		if (ifp->if_addrlen != lla_len)
3037 			continue;
3038 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
3039 			psref_acquire(psref, &ifp->if_psref,
3040 			    ifnet_psref_class);
3041 			break;
3042 		}
3043 	}
3044 	pserialize_read_exit(s);
3045 
3046 	return ifp;
3047 }
3048 
3049 /*
3050  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
3051  * for example using pserialize or the ifp is already held or some other
3052  * object is held which guarantes the ifp to not be freed indirectly.
3053  */
3054 void
3055 if_acquire(struct ifnet *ifp, struct psref *psref)
3056 {
3057 
3058 	KASSERT(ifp->if_index != 0);
3059 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
3060 }
3061 
3062 bool
3063 if_held(struct ifnet *ifp)
3064 {
3065 
3066 	return psref_held(&ifp->if_psref, ifnet_psref_class);
3067 }
3068 
3069 /*
3070  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over
3071  * IPv4. Check the tunnel nesting count.
3072  * Return > 0, if tunnel nesting count is more than limit.
3073  * Return 0, if tunnel nesting count is equal or less than limit.
3074  */
3075 int
3076 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
3077 {
3078 	struct m_tag *mtag;
3079 	int *count;
3080 
3081 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO);
3082 	if (mtag != NULL) {
3083 		count = (int *)(mtag + 1);
3084 		if (++(*count) > limit) {
3085 			log(LOG_NOTICE,
3086 			    "%s: recursively called too many times(%d)\n",
3087 			    ifp->if_xname, *count);
3088 			return EIO;
3089 		}
3090 	} else {
3091 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
3092 		    M_NOWAIT);
3093 		if (mtag != NULL) {
3094 			m_tag_prepend(m, mtag);
3095 			count = (int *)(mtag + 1);
3096 			*count = 0;
3097 		} else {
3098 			log(LOG_DEBUG, "%s: m_tag_get() failed, "
3099 			    "recursion calls are not prevented.\n",
3100 			    ifp->if_xname);
3101 		}
3102 	}
3103 
3104 	return 0;
3105 }
3106 
3107 static void
3108 if_tunnel_ro_init_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
3109 {
3110 	struct tunnel_ro *tro = p;
3111 
3112 	tro->tr_ro = kmem_zalloc(sizeof(*tro->tr_ro), KM_SLEEP);
3113 	tro->tr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
3114 }
3115 
3116 static void
3117 if_tunnel_ro_fini_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
3118 {
3119 	struct tunnel_ro *tro = p;
3120 
3121 	rtcache_free(tro->tr_ro);
3122 	kmem_free(tro->tr_ro, sizeof(*tro->tr_ro));
3123 
3124 	mutex_obj_free(tro->tr_lock);
3125 }
3126 
3127 percpu_t *
3128 if_tunnel_alloc_ro_percpu(void)
3129 {
3130 
3131 	return percpu_create(sizeof(struct tunnel_ro),
3132 	    if_tunnel_ro_init_pc, if_tunnel_ro_fini_pc, NULL);
3133 }
3134 
3135 void
3136 if_tunnel_free_ro_percpu(percpu_t *ro_percpu)
3137 {
3138 
3139 	percpu_free(ro_percpu, sizeof(struct tunnel_ro));
3140 }
3141 
3142 
3143 static void
3144 if_tunnel_rtcache_free_pc(void *p, void *arg __unused,
3145     struct cpu_info *ci __unused)
3146 {
3147 	struct tunnel_ro *tro = p;
3148 
3149 	mutex_enter(tro->tr_lock);
3150 	rtcache_free(tro->tr_ro);
3151 	mutex_exit(tro->tr_lock);
3152 }
3153 
3154 void if_tunnel_ro_percpu_rtcache_free(percpu_t *ro_percpu)
3155 {
3156 
3157 	percpu_foreach(ro_percpu, if_tunnel_rtcache_free_pc, NULL);
3158 }
3159 
3160 void
3161 if_export_if_data(ifnet_t * const ifp, struct if_data *ifi, bool zero_stats)
3162 {
3163 
3164 	/* Collect the volatile stats first; this zeros *ifi. */
3165 	if_stats_to_if_data(ifp, ifi, zero_stats);
3166 
3167 	ifi->ifi_type = ifp->if_type;
3168 	ifi->ifi_addrlen = ifp->if_addrlen;
3169 	ifi->ifi_hdrlen = ifp->if_hdrlen;
3170 	ifi->ifi_link_state = ifp->if_link_state;
3171 	ifi->ifi_mtu = ifp->if_mtu;
3172 	ifi->ifi_metric = ifp->if_metric;
3173 	ifi->ifi_baudrate = ifp->if_baudrate;
3174 	ifi->ifi_lastchange = ifp->if_lastchange;
3175 }
3176 
3177 /* common */
3178 int
3179 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
3180 {
3181 	struct ifreq *ifr;
3182 	struct ifcapreq *ifcr;
3183 	struct ifdatareq *ifdr;
3184 	unsigned short flags;
3185 	char *descr;
3186 	int error;
3187 
3188 	switch (cmd) {
3189 	case SIOCSIFCAP:
3190 		ifcr = data;
3191 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
3192 			return EINVAL;
3193 
3194 		if (ifcr->ifcr_capenable == ifp->if_capenable)
3195 			return 0;
3196 
3197 		ifp->if_capenable = ifcr->ifcr_capenable;
3198 
3199 		/* Pre-compute the checksum flags mask. */
3200 		ifp->if_csum_flags_tx = 0;
3201 		ifp->if_csum_flags_rx = 0;
3202 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx)
3203 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
3204 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
3205 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
3206 
3207 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx)
3208 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
3209 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
3210 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
3211 
3212 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx)
3213 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
3214 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
3215 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
3216 
3217 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx)
3218 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
3219 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
3220 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
3221 
3222 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx)
3223 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
3224 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
3225 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
3226 
3227 		if (ifp->if_capenable & IFCAP_TSOv4)
3228 			ifp->if_csum_flags_tx |= M_CSUM_TSOv4;
3229 		if (ifp->if_capenable & IFCAP_TSOv6)
3230 			ifp->if_csum_flags_tx |= M_CSUM_TSOv6;
3231 
3232 #if NBRIDGE > 0
3233 		if (ifp->if_bridge != NULL)
3234 			bridge_calc_csum_flags(ifp->if_bridge);
3235 #endif
3236 
3237 		if (ifp->if_flags & IFF_UP)
3238 			return ENETRESET;
3239 		return 0;
3240 	case SIOCSIFFLAGS:
3241 		ifr = data;
3242 		/*
3243 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
3244 		 * and if_down aren't MP-safe yet, so we must hold the lock.
3245 		 */
3246 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3247 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
3248 			const int s = splsoftnet();
3249 			if_down_locked(ifp);
3250 			splx(s);
3251 		}
3252 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
3253 			const int s = splsoftnet();
3254 			if_up_locked(ifp);
3255 			splx(s);
3256 		}
3257 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3258 		flags = (ifp->if_flags & IFF_CANTCHANGE) |
3259 		    (ifr->ifr_flags &~ IFF_CANTCHANGE);
3260 		if (ifp->if_flags != flags) {
3261 			ifp->if_flags = flags;
3262 			/* Notify that the flags have changed. */
3263 			rt_ifmsg(ifp);
3264 		}
3265 		break;
3266 	case SIOCGIFFLAGS:
3267 		ifr = data;
3268 		ifr->ifr_flags = ifp->if_flags;
3269 		break;
3270 
3271 	case SIOCGIFMETRIC:
3272 		ifr = data;
3273 		ifr->ifr_metric = ifp->if_metric;
3274 		break;
3275 
3276 	case SIOCGIFMTU:
3277 		ifr = data;
3278 		ifr->ifr_mtu = ifp->if_mtu;
3279 		break;
3280 
3281 	case SIOCGIFDLT:
3282 		ifr = data;
3283 		ifr->ifr_dlt = ifp->if_dlt;
3284 		break;
3285 
3286 	case SIOCGIFCAP:
3287 		ifcr = data;
3288 		ifcr->ifcr_capabilities = ifp->if_capabilities;
3289 		ifcr->ifcr_capenable = ifp->if_capenable;
3290 		break;
3291 
3292 	case SIOCSIFMETRIC:
3293 		ifr = data;
3294 		ifp->if_metric = ifr->ifr_metric;
3295 		break;
3296 
3297 	case SIOCGIFDATA:
3298 		ifdr = data;
3299 		if_export_if_data(ifp, &ifdr->ifdr_data, false);
3300 		break;
3301 
3302 	case SIOCGIFINDEX:
3303 		ifr = data;
3304 		ifr->ifr_index = ifp->if_index;
3305 		break;
3306 
3307 	case SIOCZIFDATA:
3308 		ifdr = data;
3309 		if_export_if_data(ifp, &ifdr->ifdr_data, true);
3310 		getnanotime(&ifp->if_lastchange);
3311 		break;
3312 	case SIOCSIFMTU:
3313 		ifr = data;
3314 		if (ifp->if_mtu == ifr->ifr_mtu)
3315 			break;
3316 		ifp->if_mtu = ifr->ifr_mtu;
3317 		return ENETRESET;
3318 	case SIOCSIFDESCR:
3319 		error = kauth_authorize_network(kauth_cred_get(),
3320 		    KAUTH_NETWORK_INTERFACE,
3321 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, KAUTH_ARG(cmd),
3322 		    NULL);
3323 		if (error)
3324 			return error;
3325 
3326 		ifr = data;
3327 
3328 		if (ifr->ifr_buflen > IFDESCRSIZE)
3329 			return ENAMETOOLONG;
3330 
3331 		if (ifr->ifr_buf == NULL || ifr->ifr_buflen == 0) {
3332 			/* unset description */
3333 			descr = NULL;
3334 		} else {
3335 			descr = kmem_zalloc(IFDESCRSIZE, KM_SLEEP);
3336 			/*
3337 			 * copy (IFDESCRSIZE - 1) bytes to ensure
3338 			 * terminating nul
3339 			 */
3340 			error = copyin(ifr->ifr_buf, descr, IFDESCRSIZE - 1);
3341 			if (error) {
3342 				kmem_free(descr, IFDESCRSIZE);
3343 				return error;
3344 			}
3345 		}
3346 
3347 		if (ifp->if_description != NULL)
3348 			kmem_free(ifp->if_description, IFDESCRSIZE);
3349 
3350 		ifp->if_description = descr;
3351 		break;
3352 
3353 	case SIOCGIFDESCR:
3354 		ifr = data;
3355 		descr = ifp->if_description;
3356 
3357 		if (descr == NULL)
3358 			return ENOMSG;
3359 
3360 		if (ifr->ifr_buflen < IFDESCRSIZE)
3361 			return EINVAL;
3362 
3363 		error = copyout(descr, ifr->ifr_buf, IFDESCRSIZE);
3364 		if (error)
3365 			return error;
3366 		break;
3367 
3368 	default:
3369 		return ENOTTY;
3370 	}
3371 	return 0;
3372 }
3373 
3374 int
3375 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3376 {
3377 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3378 	struct ifaddr *ifa;
3379 	const struct sockaddr *any, *sa;
3380 	union {
3381 		struct sockaddr sa;
3382 		struct sockaddr_storage ss;
3383 	} u, v;
3384 	int s, error = 0;
3385 
3386 	switch (cmd) {
3387 	case SIOCSIFADDRPREF:
3388 		error = kauth_authorize_network(kauth_cred_get(),
3389 		    KAUTH_NETWORK_INTERFACE,
3390 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, KAUTH_ARG(cmd),
3391 		    NULL);
3392 		if (error)
3393 			return error;
3394 		break;
3395 	case SIOCGIFADDRPREF:
3396 		break;
3397 	default:
3398 		return EOPNOTSUPP;
3399 	}
3400 
3401 	/* sanity checks */
3402 	if (data == NULL || ifp == NULL) {
3403 		panic("invalid argument to %s", __func__);
3404 		/*NOTREACHED*/
3405 	}
3406 
3407 	/* address must be specified on ADD and DELETE */
3408 	sa = sstocsa(&ifap->ifap_addr);
3409 	if (sa->sa_family != sofamily(so))
3410 		return EINVAL;
3411 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3412 		return EINVAL;
3413 
3414 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3415 
3416 	s = pserialize_read_enter();
3417 	IFADDR_READER_FOREACH(ifa, ifp) {
3418 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3419 			continue;
3420 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3421 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3422 			break;
3423 	}
3424 	if (ifa == NULL) {
3425 		error = EADDRNOTAVAIL;
3426 		goto out;
3427 	}
3428 
3429 	switch (cmd) {
3430 	case SIOCSIFADDRPREF:
3431 		ifa->ifa_preference = ifap->ifap_preference;
3432 		goto out;
3433 	case SIOCGIFADDRPREF:
3434 		/* fill in the if_laddrreq structure */
3435 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3436 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3437 		ifap->ifap_preference = ifa->ifa_preference;
3438 		goto out;
3439 	default:
3440 		error = EOPNOTSUPP;
3441 	}
3442 out:
3443 	pserialize_read_exit(s);
3444 	return error;
3445 }
3446 
3447 /*
3448  * Interface ioctls.
3449  */
3450 static int
3451 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3452 {
3453 	struct ifnet *ifp;
3454 	struct ifreq *ifr;
3455 	int error = 0;
3456 	u_long ocmd = cmd;
3457 	u_short oif_flags;
3458 	struct ifreq ifrb;
3459 	struct oifreq *oifr = NULL;
3460 	int r;
3461 	struct psref psref;
3462 	bool do_if43_post = false;
3463 	bool do_ifm80_post = false;
3464 
3465 	switch (cmd) {
3466 	case SIOCGIFCONF:
3467 		return ifconf(cmd, data);
3468 	case SIOCINITIFADDR:
3469 		return EPERM;
3470 	default:
3471 		MODULE_HOOK_CALL(uipc_syscalls_40_hook, (cmd, data), enosys(),
3472 		    error);
3473 		if (error != ENOSYS)
3474 			return error;
3475 		MODULE_HOOK_CALL(uipc_syscalls_50_hook, (l, cmd, data),
3476 		    enosys(), error);
3477 		if (error != ENOSYS)
3478 			return error;
3479 		error = 0;
3480 		break;
3481 	}
3482 
3483 	ifr = data;
3484 	/* Pre-conversion */
3485 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), error);
3486 	if (cmd != ocmd) {
3487 		oifr = data;
3488 		data = ifr = &ifrb;
3489 		IFREQO2N_43(oifr, ifr);
3490 		do_if43_post = true;
3491 	}
3492 	MODULE_HOOK_CALL(ifmedia_80_pre_hook, (ifr, &cmd, &do_ifm80_post),
3493 	    enosys(), error);
3494 
3495 	switch (cmd) {
3496 	case SIOCIFCREATE:
3497 	case SIOCIFDESTROY: {
3498 		const int bound = curlwp_bind();
3499 		if (l != NULL) {
3500 			ifp = if_get(ifr->ifr_name, &psref);
3501 			error = kauth_authorize_network(l->l_cred,
3502 			    KAUTH_NETWORK_INTERFACE,
3503 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3504 			    KAUTH_ARG(cmd), NULL);
3505 			if (ifp != NULL)
3506 				if_put(ifp, &psref);
3507 			if (error != 0) {
3508 				curlwp_bindx(bound);
3509 				return error;
3510 			}
3511 		}
3512 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3513 		mutex_enter(&if_clone_mtx);
3514 		r = (cmd == SIOCIFCREATE) ?
3515 			if_clone_create(ifr->ifr_name) :
3516 			if_clone_destroy(ifr->ifr_name);
3517 		mutex_exit(&if_clone_mtx);
3518 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3519 		curlwp_bindx(bound);
3520 		return r;
3521 	    }
3522 	case SIOCIFGCLONERS: {
3523 		struct if_clonereq *req = (struct if_clonereq *)data;
3524 		return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3525 		    &req->ifcr_total);
3526 	    }
3527 	}
3528 
3529 	if ((cmd & IOC_IN) == 0 || IOCPARM_LEN(cmd) < sizeof(ifr->ifr_name))
3530 		return EINVAL;
3531 
3532 	const int bound = curlwp_bind();
3533 	ifp = if_get(ifr->ifr_name, &psref);
3534 	if (ifp == NULL) {
3535 		curlwp_bindx(bound);
3536 		return ENXIO;
3537 	}
3538 
3539 	switch (cmd) {
3540 	case SIOCALIFADDR:
3541 	case SIOCDLIFADDR:
3542 	case SIOCSIFADDRPREF:
3543 	case SIOCSIFFLAGS:
3544 	case SIOCSIFCAP:
3545 	case SIOCSIFMETRIC:
3546 	case SIOCZIFDATA:
3547 	case SIOCSIFMTU:
3548 	case SIOCSIFPHYADDR:
3549 	case SIOCDIFPHYADDR:
3550 #ifdef INET6
3551 	case SIOCSIFPHYADDR_IN6:
3552 #endif
3553 	case SIOCSLIFPHYADDR:
3554 	case SIOCADDMULTI:
3555 	case SIOCDELMULTI:
3556 	case SIOCSETHERCAP:
3557 	case SIOCSIFMEDIA:
3558 	case SIOCSDRVSPEC:
3559 	case SIOCG80211:
3560 	case SIOCS80211:
3561 	case SIOCS80211NWID:
3562 	case SIOCS80211NWKEY:
3563 	case SIOCS80211POWER:
3564 	case SIOCS80211BSSID:
3565 	case SIOCS80211CHANNEL:
3566 	case SIOCSLINKSTR:
3567 		if (l != NULL) {
3568 			error = kauth_authorize_network(l->l_cred,
3569 			    KAUTH_NETWORK_INTERFACE,
3570 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3571 			    KAUTH_ARG(cmd), NULL);
3572 			if (error != 0)
3573 				goto out;
3574 		}
3575 	}
3576 
3577 	oif_flags = ifp->if_flags;
3578 
3579 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3580 	IFNET_LOCK(ifp);
3581 
3582 	error = if_ioctl(ifp, cmd, data);
3583 	if (error != ENOTTY)
3584 		;
3585 	else if (so->so_proto == NULL)
3586 		error = EOPNOTSUPP;
3587 	else {
3588 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3589 		MODULE_HOOK_CALL(if_ifioctl_43_hook,
3590 			     (so, ocmd, cmd, data, l), enosys(), error);
3591 		if (error == ENOSYS)
3592 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3593 			    cmd, data, ifp);
3594 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3595 	}
3596 
3597 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3598 		if ((ifp->if_flags & IFF_UP) != 0) {
3599 			const int s = splsoftnet();
3600 			if_up_locked(ifp);
3601 			splx(s);
3602 		}
3603 	}
3604 
3605 	/* Post-conversion */
3606 	if (do_ifm80_post && (error == 0))
3607 		MODULE_HOOK_CALL(ifmedia_80_post_hook, (ifr, cmd),
3608 		    enosys(), error);
3609 	if (do_if43_post)
3610 		IFREQN2O_43(oifr, ifr);
3611 
3612 	IFNET_UNLOCK(ifp);
3613 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3614 out:
3615 	if_put(ifp, &psref);
3616 	curlwp_bindx(bound);
3617 	return error;
3618 }
3619 
3620 /*
3621  * Return interface configuration
3622  * of system.  List may be used
3623  * in later ioctl's (above) to get
3624  * other information.
3625  *
3626  * Each record is a struct ifreq.  Before the addition of
3627  * sockaddr_storage, the API rule was that sockaddr flavors that did
3628  * not fit would extend beyond the struct ifreq, with the next struct
3629  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3630  * union in struct ifreq includes struct sockaddr_storage, every kind
3631  * of sockaddr must fit.  Thus, there are no longer any overlength
3632  * records.
3633  *
3634  * Records are added to the user buffer if they fit, and ifc_len is
3635  * adjusted to the length that was written.  Thus, the user is only
3636  * assured of getting the complete list if ifc_len on return is at
3637  * least sizeof(struct ifreq) less than it was on entry.
3638  *
3639  * If the user buffer pointer is NULL, this routine copies no data and
3640  * returns the amount of space that would be needed.
3641  *
3642  * Invariants:
3643  * ifrp points to the next part of the user's buffer to be used.  If
3644  * ifrp != NULL, space holds the number of bytes remaining that we may
3645  * write at ifrp.  Otherwise, space holds the number of bytes that
3646  * would have been written had there been adequate space.
3647  */
3648 /*ARGSUSED*/
3649 static int
3650 ifconf(u_long cmd, void *data)
3651 {
3652 	struct ifconf *ifc = (struct ifconf *)data;
3653 	struct ifnet *ifp;
3654 	struct ifaddr *ifa;
3655 	struct ifreq ifr, *ifrp = NULL;
3656 	int space = 0, error = 0;
3657 	const int sz = (int)sizeof(struct ifreq);
3658 	const bool docopy = ifc->ifc_req != NULL;
3659 	struct psref psref;
3660 
3661 	if (docopy) {
3662 		if (ifc->ifc_len < 0)
3663 			return EINVAL;
3664 
3665 		space = ifc->ifc_len;
3666 		ifrp = ifc->ifc_req;
3667 	}
3668 	memset(&ifr, 0, sizeof(ifr));
3669 
3670 	const int bound = curlwp_bind();
3671 	int s = pserialize_read_enter();
3672 	IFNET_READER_FOREACH(ifp) {
3673 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3674 		pserialize_read_exit(s);
3675 
3676 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3677 		    sizeof(ifr.ifr_name));
3678 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3679 			error = ENAMETOOLONG;
3680 			goto release_exit;
3681 		}
3682 		if (IFADDR_READER_EMPTY(ifp)) {
3683 			/* Interface with no addresses - send zero sockaddr. */
3684 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3685 			if (!docopy) {
3686 				space += sz;
3687 				goto next;
3688 			}
3689 			if (space >= sz) {
3690 				error = copyout(&ifr, ifrp, sz);
3691 				if (error != 0)
3692 					goto release_exit;
3693 				ifrp++;
3694 				space -= sz;
3695 			}
3696 		}
3697 
3698 		s = pserialize_read_enter();
3699 		IFADDR_READER_FOREACH(ifa, ifp) {
3700 			struct sockaddr *sa = ifa->ifa_addr;
3701 			/* all sockaddrs must fit in sockaddr_storage */
3702 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3703 
3704 			if (!docopy) {
3705 				space += sz;
3706 				continue;
3707 			}
3708 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3709 			pserialize_read_exit(s);
3710 
3711 			if (space >= sz) {
3712 				error = copyout(&ifr, ifrp, sz);
3713 				if (error != 0)
3714 					goto release_exit;
3715 				ifrp++; space -= sz;
3716 			}
3717 			s = pserialize_read_enter();
3718 		}
3719 		pserialize_read_exit(s);
3720 
3721 next:
3722 		s = pserialize_read_enter();
3723 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3724 	}
3725 	pserialize_read_exit(s);
3726 	curlwp_bindx(bound);
3727 
3728 	if (docopy) {
3729 		KASSERT(0 <= space && space <= ifc->ifc_len);
3730 		ifc->ifc_len -= space;
3731 	} else {
3732 		KASSERT(space >= 0);
3733 		ifc->ifc_len = space;
3734 	}
3735 	return 0;
3736 
3737 release_exit:
3738 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3739 	curlwp_bindx(bound);
3740 	return error;
3741 }
3742 
3743 int
3744 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3745 {
3746 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3747 	struct ifreq ifrb;
3748 	struct oifreq *oifr = NULL;
3749 	u_long ocmd = cmd;
3750 	int hook;
3751 
3752 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook);
3753 	if (hook != ENOSYS) {
3754 		if (cmd != ocmd) {
3755 			oifr = (struct oifreq *)(void *)ifr;
3756 			ifr = &ifrb;
3757 			IFREQO2N_43(oifr, ifr);
3758 				len = sizeof(oifr->ifr_addr);
3759 		}
3760 	}
3761 
3762 	if (len < sa->sa_len)
3763 		return EFBIG;
3764 
3765 	memset(&ifr->ifr_addr, 0, len);
3766 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3767 
3768 	if (cmd != ocmd)
3769 		IFREQN2O_43(oifr, ifr);
3770 	return 0;
3771 }
3772 
3773 /*
3774  * wrapper function for the drivers which doesn't have if_transmit().
3775  */
3776 static int
3777 if_transmit(struct ifnet *ifp, struct mbuf *m)
3778 {
3779 	int error;
3780 	size_t pktlen = m->m_pkthdr.len;
3781 	bool mcast = (m->m_flags & M_MCAST) != 0;
3782 
3783 	const int s = splnet();
3784 
3785 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3786 	if (error != 0) {
3787 		/* mbuf is already freed */
3788 		goto out;
3789 	}
3790 
3791 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
3792 	if_statadd_ref(nsr, if_obytes, pktlen);
3793 	if (mcast)
3794 		if_statinc_ref(nsr, if_omcasts);
3795 	IF_STAT_PUTREF(ifp);
3796 
3797 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3798 		if_start_lock(ifp);
3799 out:
3800 	splx(s);
3801 
3802 	return error;
3803 }
3804 
3805 int
3806 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3807 {
3808 	int error;
3809 
3810 	kmsan_check_mbuf(m);
3811 
3812 #ifdef ALTQ
3813 	KERNEL_LOCK(1, NULL);
3814 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3815 		error = if_transmit(ifp, m);
3816 		KERNEL_UNLOCK_ONE(NULL);
3817 	} else {
3818 		KERNEL_UNLOCK_ONE(NULL);
3819 		error = (*ifp->if_transmit)(ifp, m);
3820 		/* mbuf is already freed */
3821 	}
3822 #else /* !ALTQ */
3823 	error = (*ifp->if_transmit)(ifp, m);
3824 	/* mbuf is already freed */
3825 #endif /* !ALTQ */
3826 
3827 	return error;
3828 }
3829 
3830 /*
3831  * Queue message on interface, and start output if interface
3832  * not yet active.
3833  */
3834 int
3835 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3836 {
3837 
3838 	return if_transmit_lock(ifp, m);
3839 }
3840 
3841 /*
3842  * Queue message on interface, possibly using a second fast queue
3843  */
3844 int
3845 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3846 {
3847 	int error = 0;
3848 
3849 	if (ifq != NULL
3850 #ifdef ALTQ
3851 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3852 #endif
3853 	    ) {
3854 		if (IF_QFULL(ifq)) {
3855 			IF_DROP(&ifp->if_snd);
3856 			m_freem(m);
3857 			if (error == 0)
3858 				error = ENOBUFS;
3859 		} else
3860 			IF_ENQUEUE(ifq, m);
3861 	} else
3862 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3863 	if (error != 0) {
3864 		if_statinc(ifp, if_oerrors);
3865 		return error;
3866 	}
3867 	return 0;
3868 }
3869 
3870 int
3871 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3872 {
3873 	int rc;
3874 
3875 	KASSERT(IFNET_LOCKED(ifp));
3876 	if (ifp->if_initaddr != NULL)
3877 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3878 	else if (src || (rc = if_ioctl(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3879 		rc = if_ioctl(ifp, SIOCINITIFADDR, ifa);
3880 
3881 	return rc;
3882 }
3883 
3884 int
3885 if_do_dad(struct ifnet *ifp)
3886 {
3887 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3888 		return 0;
3889 
3890 	switch (ifp->if_type) {
3891 	case IFT_FAITH:
3892 		/*
3893 		 * These interfaces do not have the IFF_LOOPBACK flag,
3894 		 * but loop packets back.  We do not have to do DAD on such
3895 		 * interfaces.  We should even omit it, because loop-backed
3896 		 * responses would confuse the DAD procedure.
3897 		 */
3898 		return 0;
3899 	default:
3900 		/*
3901 		 * Our DAD routine requires the interface up and running.
3902 		 * However, some interfaces can be up before the RUNNING
3903 		 * status.  Additionally, users may try to assign addresses
3904 		 * before the interface becomes up (or running).
3905 		 * We simply skip DAD in such a case as a work around.
3906 		 * XXX: we should rather mark "tentative" on such addresses,
3907 		 * and do DAD after the interface becomes ready.
3908 		 */
3909 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
3910 		    (IFF_UP | IFF_RUNNING))
3911 			return 0;
3912 
3913 		return 1;
3914 	}
3915 }
3916 
3917 /*
3918  * if_flags_set(ifp, flags)
3919  *
3920  *	Ask ifp to change ifp->if_flags to flags, as if with the
3921  *	SIOCSIFFLAGS ioctl command.
3922  *
3923  *	May sleep.  Caller must hold ifp->if_ioctl_lock, a.k.a
3924  *	IFNET_LOCK.
3925  */
3926 int
3927 if_flags_set(ifnet_t *ifp, const u_short flags)
3928 {
3929 	int rc;
3930 
3931 	KASSERT(IFNET_LOCKED(ifp));
3932 
3933 	if (ifp->if_setflags != NULL)
3934 		rc = (*ifp->if_setflags)(ifp, flags);
3935 	else {
3936 		u_short cantflags, chgdflags;
3937 		struct ifreq ifr;
3938 
3939 		chgdflags = ifp->if_flags ^ flags;
3940 		cantflags = chgdflags & IFF_CANTCHANGE;
3941 
3942 		if (cantflags != 0)
3943 			ifp->if_flags ^= cantflags;
3944 
3945 		/*
3946 		 * Traditionally, we do not call if_ioctl after
3947 		 * setting/clearing only IFF_PROMISC if the interface
3948 		 * isn't IFF_UP.  Uphold that tradition.
3949 		 */
3950 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3951 			return 0;
3952 
3953 		memset(&ifr, 0, sizeof(ifr));
3954 
3955 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3956 		rc = if_ioctl(ifp, SIOCSIFFLAGS, &ifr);
3957 
3958 		if (rc != 0 && cantflags != 0)
3959 			ifp->if_flags ^= cantflags;
3960 	}
3961 
3962 	return rc;
3963 }
3964 
3965 /*
3966  * if_mcast_op(ifp, cmd, sa)
3967  *
3968  *	Apply a multicast command, SIOCADDMULTI/SIOCDELMULTI, to the
3969  *	interface.  Returns 0 on success, nonzero errno(3) number on
3970  *	failure.
3971  *
3972  *	May sleep.
3973  *
3974  *	Use this, not if_ioctl, for the multicast commands.
3975  */
3976 int
3977 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3978 {
3979 	int rc;
3980 	struct ifreq ifr;
3981 
3982 	switch (cmd) {
3983 	case SIOCADDMULTI:
3984 	case SIOCDELMULTI:
3985 		break;
3986 	default:
3987 		panic("invalid ifnet multicast command: 0x%lx", cmd);
3988 	}
3989 
3990 	ifreq_setaddr(cmd, &ifr, sa);
3991 	rc = if_ioctl(ifp, cmd, &ifr);
3992 
3993 	return rc;
3994 }
3995 
3996 static void
3997 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3998     struct ifaltq *ifq)
3999 {
4000 	const struct sysctlnode *cnode, *rnode;
4001 
4002 	if (sysctl_createv(clog, 0, NULL, &rnode,
4003 		       CTLFLAG_PERMANENT,
4004 		       CTLTYPE_NODE, "interfaces",
4005 		       SYSCTL_DESCR("Per-interface controls"),
4006 		       NULL, 0, NULL, 0,
4007 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
4008 		goto bad;
4009 
4010 	if (sysctl_createv(clog, 0, &rnode, &rnode,
4011 		       CTLFLAG_PERMANENT,
4012 		       CTLTYPE_NODE, ifname,
4013 		       SYSCTL_DESCR("Interface controls"),
4014 		       NULL, 0, NULL, 0,
4015 		       CTL_CREATE, CTL_EOL) != 0)
4016 		goto bad;
4017 
4018 	if (sysctl_createv(clog, 0, &rnode, &rnode,
4019 		       CTLFLAG_PERMANENT,
4020 		       CTLTYPE_NODE, "sndq",
4021 		       SYSCTL_DESCR("Interface output queue controls"),
4022 		       NULL, 0, NULL, 0,
4023 		       CTL_CREATE, CTL_EOL) != 0)
4024 		goto bad;
4025 
4026 	if (sysctl_createv(clog, 0, &rnode, &cnode,
4027 		       CTLFLAG_PERMANENT,
4028 		       CTLTYPE_INT, "len",
4029 		       SYSCTL_DESCR("Current output queue length"),
4030 		       NULL, 0, &ifq->ifq_len, 0,
4031 		       CTL_CREATE, CTL_EOL) != 0)
4032 		goto bad;
4033 
4034 	if (sysctl_createv(clog, 0, &rnode, &cnode,
4035 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
4036 		       CTLTYPE_INT, "maxlen",
4037 		       SYSCTL_DESCR("Maximum allowed output queue length"),
4038 		       NULL, 0, &ifq->ifq_maxlen, 0,
4039 		       CTL_CREATE, CTL_EOL) != 0)
4040 		goto bad;
4041 
4042 	if (sysctl_createv(clog, 0, &rnode, &cnode,
4043 		       CTLFLAG_PERMANENT,
4044 		       CTLTYPE_QUAD, "drops",
4045 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
4046 		       NULL, 0, &ifq->ifq_drops, 0,
4047 		       CTL_CREATE, CTL_EOL) != 0)
4048 		goto bad;
4049 
4050 	return;
4051 bad:
4052 	printf("%s: could not attach sysctl nodes\n", ifname);
4053 	return;
4054 }
4055 
4056 static int
4057 if_sdl_sysctl(SYSCTLFN_ARGS)
4058 {
4059 	struct ifnet *ifp;
4060 	const struct sockaddr_dl *sdl;
4061 	struct psref psref;
4062 	int error = 0;
4063 
4064 	if (namelen != 1)
4065 		return EINVAL;
4066 
4067 	const int bound = curlwp_bind();
4068 	ifp = if_get_byindex(name[0], &psref);
4069 	if (ifp == NULL) {
4070 		error = ENODEV;
4071 		goto out0;
4072 	}
4073 
4074 	sdl = ifp->if_sadl;
4075 	if (sdl == NULL) {
4076 		*oldlenp = 0;
4077 		goto out1;
4078 	}
4079 
4080 	if (oldp == NULL) {
4081 		*oldlenp = sdl->sdl_alen;
4082 		goto out1;
4083 	}
4084 
4085 	if (*oldlenp >= sdl->sdl_alen)
4086 		*oldlenp = sdl->sdl_alen;
4087 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen],
4088 	    oldp, *oldlenp);
4089 out1:
4090 	if_put(ifp, &psref);
4091 out0:
4092 	curlwp_bindx(bound);
4093 	return error;
4094 }
4095 
4096 static void
4097 if_sysctl_setup(struct sysctllog **clog)
4098 {
4099 	const struct sysctlnode *rnode = NULL;
4100 
4101 	sysctl_createv(clog, 0, NULL, &rnode,
4102 		       CTLFLAG_PERMANENT,
4103 		       CTLTYPE_NODE, "sdl",
4104 		       SYSCTL_DESCR("Get active link-layer address"),
4105 		       if_sdl_sysctl, 0, NULL, 0,
4106 		       CTL_NET, CTL_CREATE, CTL_EOL);
4107 }
4108