xref: /netbsd-src/sys/net/if.c (revision 53d1339bf7f9c7367b35a9e1ebe693f9b047a47b)
1 /*	$NetBSD: if.c,v 1.485 2021/05/17 04:07:43 yamaguchi 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 1.485 2021/05/17 04:07:43 yamaguchi 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 
126 #include <net/if.h>
127 #include <net/if_dl.h>
128 #include <net/if_ether.h>
129 #include <net/if_media.h>
130 #include <net80211/ieee80211.h>
131 #include <net80211/ieee80211_ioctl.h>
132 #include <net/if_types.h>
133 #include <net/route.h>
134 #include <net/netisr.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 "lagg.h"
164 #if NLAGG > 0
165 #include <net/lagg/if_laggvar.h>
166 #endif
167 
168 #include <compat/sys/sockio.h>
169 
170 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
171 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
172 
173 /*
174  * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex
175  * for each ifnet.  It doesn't matter because:
176  * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on
177  *   ifq_lock don't happen
178  * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock
179  *   because if_snd, if_link_state_change and if_link_state_change_process
180  *   are all called with KERNEL_LOCK
181  */
182 #define IF_LINK_STATE_CHANGE_LOCK(ifp)		\
183 	mutex_enter((ifp)->if_snd.ifq_lock)
184 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp)	\
185 	mutex_exit((ifp)->if_snd.ifq_lock)
186 
187 /*
188  * Global list of interfaces.
189  */
190 /* DEPRECATED. Remove it once kvm(3) users disappeared */
191 struct ifnet_head		ifnet_list;
192 
193 struct pslist_head		ifnet_pslist;
194 static ifnet_t **		ifindex2ifnet = NULL;
195 static u_int			if_index = 1;
196 static size_t			if_indexlim = 0;
197 static uint64_t			index_gen;
198 /* Mutex to protect the above objects. */
199 kmutex_t			ifnet_mtx __cacheline_aligned;
200 static struct psref_class	*ifnet_psref_class __read_mostly;
201 static pserialize_t		ifnet_psz;
202 static struct workqueue		*ifnet_link_state_wq __read_mostly;
203 
204 static kmutex_t			if_clone_mtx;
205 
206 struct ifnet *lo0ifp;
207 int	ifqmaxlen = IFQ_MAXLEN;
208 
209 struct psref_class		*ifa_psref_class __read_mostly;
210 
211 static int	if_delroute_matcher(struct rtentry *, void *);
212 
213 static bool if_is_unit(const char *);
214 static struct if_clone *if_clone_lookup(const char *, int *);
215 
216 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
217 static int if_cloners_count;
218 
219 /* Packet filtering hook for interfaces. */
220 pfil_head_t *			if_pfil __read_mostly;
221 
222 static kauth_listener_t if_listener;
223 
224 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
225 static void if_detach_queues(struct ifnet *, struct ifqueue *);
226 static void sysctl_sndq_setup(struct sysctllog **, const char *,
227     struct ifaltq *);
228 static void if_slowtimo(void *);
229 static void if_attachdomain1(struct ifnet *);
230 static int ifconf(u_long, void *);
231 static int if_transmit(struct ifnet *, struct mbuf *);
232 static int if_clone_create(const char *);
233 static int if_clone_destroy(const char *);
234 static void if_link_state_change_work(struct work *, void *);
235 static void if_up_locked(struct ifnet *);
236 static void _if_down(struct ifnet *);
237 static void if_down_deactivated(struct ifnet *);
238 
239 struct if_percpuq {
240 	struct ifnet	*ipq_ifp;
241 	void		*ipq_si;
242 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
243 };
244 
245 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
246 
247 static void if_percpuq_drops(void *, void *, struct cpu_info *);
248 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
249 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
250     struct if_percpuq *);
251 
252 struct if_deferred_start {
253 	struct ifnet	*ids_ifp;
254 	void		(*ids_if_start)(struct ifnet *);
255 	void		*ids_si;
256 };
257 
258 static void if_deferred_start_softint(void *);
259 static void if_deferred_start_common(struct ifnet *);
260 static void if_deferred_start_destroy(struct ifnet *);
261 
262 #if defined(INET) || defined(INET6)
263 static void sysctl_net_pktq_setup(struct sysctllog **, int);
264 #endif
265 
266 /*
267  * Hook for if_vlan - needed by if_agr
268  */
269 struct if_vlan_vlan_input_hook_t if_vlan_vlan_input_hook;
270 
271 static void if_sysctl_setup(struct sysctllog **);
272 
273 static int
274 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
275     void *arg0, void *arg1, void *arg2, void *arg3)
276 {
277 	int result;
278 	enum kauth_network_req req;
279 
280 	result = KAUTH_RESULT_DEFER;
281 	req = (enum kauth_network_req)(uintptr_t)arg1;
282 
283 	if (action != KAUTH_NETWORK_INTERFACE)
284 		return result;
285 
286 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
287 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
288 		result = KAUTH_RESULT_ALLOW;
289 
290 	return result;
291 }
292 
293 /*
294  * Network interface utility routines.
295  *
296  * Routines with ifa_ifwith* names take sockaddr *'s as
297  * parameters.
298  */
299 void
300 ifinit(void)
301 {
302 
303 #if (defined(INET) || defined(INET6))
304 	encapinit();
305 #endif
306 
307 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
308 	    if_listener_cb, NULL);
309 
310 	/* interfaces are available, inform socket code */
311 	ifioctl = doifioctl;
312 }
313 
314 /*
315  * XXX Initialization before configure().
316  * XXX hack to get pfil_add_hook working in autoconf.
317  */
318 void
319 ifinit1(void)
320 {
321 	int error __diagused;
322 
323 #ifdef NET_MPSAFE
324 	printf("NET_MPSAFE enabled\n");
325 #endif
326 
327 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
328 
329 	TAILQ_INIT(&ifnet_list);
330 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
331 	ifnet_psz = pserialize_create();
332 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
333 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
334 	error = workqueue_create(&ifnet_link_state_wq, "iflnkst",
335 	    if_link_state_change_work, NULL, PRI_SOFTNET, IPL_NET,
336 	    WQ_MPSAFE);
337 	KASSERT(error == 0);
338 	PSLIST_INIT(&ifnet_pslist);
339 
340 	if_indexlim = 8;
341 
342 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
343 	KASSERT(if_pfil != NULL);
344 
345 #if NETHER > 0 || defined(NETATALK) || defined(WLAN)
346 	etherinit();
347 #endif
348 }
349 
350 /* XXX must be after domaininit() */
351 void
352 ifinit_post(void)
353 {
354 
355 	if_sysctl_setup(NULL);
356 }
357 
358 ifnet_t *
359 if_alloc(u_char type)
360 {
361 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
362 }
363 
364 void
365 if_free(ifnet_t *ifp)
366 {
367 	kmem_free(ifp, sizeof(ifnet_t));
368 }
369 
370 void
371 if_initname(struct ifnet *ifp, const char *name, int unit)
372 {
373 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
374 	    "%s%d", name, unit);
375 }
376 
377 /*
378  * Null routines used while an interface is going away.  These routines
379  * just return an error.
380  */
381 
382 int
383 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
384     const struct sockaddr *so, const struct rtentry *rt)
385 {
386 
387 	return ENXIO;
388 }
389 
390 void
391 if_nullinput(struct ifnet *ifp, struct mbuf *m)
392 {
393 
394 	/* Nothing. */
395 }
396 
397 void
398 if_nullstart(struct ifnet *ifp)
399 {
400 
401 	/* Nothing. */
402 }
403 
404 int
405 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
406 {
407 
408 	m_freem(m);
409 	return ENXIO;
410 }
411 
412 int
413 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
414 {
415 
416 	return ENXIO;
417 }
418 
419 int
420 if_nullinit(struct ifnet *ifp)
421 {
422 
423 	return ENXIO;
424 }
425 
426 void
427 if_nullstop(struct ifnet *ifp, int disable)
428 {
429 
430 	/* Nothing. */
431 }
432 
433 void
434 if_nullslowtimo(struct ifnet *ifp)
435 {
436 
437 	/* Nothing. */
438 }
439 
440 void
441 if_nulldrain(struct ifnet *ifp)
442 {
443 
444 	/* Nothing. */
445 }
446 
447 void
448 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
449 {
450 	struct ifaddr *ifa;
451 	struct sockaddr_dl *sdl;
452 
453 	ifp->if_addrlen = addrlen;
454 	if_alloc_sadl(ifp);
455 	ifa = ifp->if_dl;
456 	sdl = satosdl(ifa->ifa_addr);
457 
458 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
459 	if (factory) {
460 		KASSERT(ifp->if_hwdl == NULL);
461 		ifp->if_hwdl = ifp->if_dl;
462 		ifaref(ifp->if_hwdl);
463 	}
464 	/* TBD routing socket */
465 }
466 
467 struct ifaddr *
468 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
469 {
470 	unsigned socksize, ifasize;
471 	int addrlen, namelen;
472 	struct sockaddr_dl *mask, *sdl;
473 	struct ifaddr *ifa;
474 
475 	namelen = strlen(ifp->if_xname);
476 	addrlen = ifp->if_addrlen;
477 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
478 	ifasize = sizeof(*ifa) + 2 * socksize;
479 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
480 
481 	sdl = (struct sockaddr_dl *)(ifa + 1);
482 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
483 
484 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
485 	    ifp->if_xname, namelen, NULL, addrlen);
486 	mask->sdl_family = AF_LINK;
487 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
488 	memset(&mask->sdl_data[0], 0xff, namelen);
489 	ifa->ifa_rtrequest = link_rtrequest;
490 	ifa->ifa_addr = (struct sockaddr *)sdl;
491 	ifa->ifa_netmask = (struct sockaddr *)mask;
492 	ifa_psref_init(ifa);
493 
494 	*sdlp = sdl;
495 
496 	return ifa;
497 }
498 
499 static void
500 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
501 {
502 	const struct sockaddr_dl *sdl;
503 
504 	ifp->if_dl = ifa;
505 	ifaref(ifa);
506 	sdl = satosdl(ifa->ifa_addr);
507 	ifp->if_sadl = sdl;
508 }
509 
510 /*
511  * Allocate the link level name for the specified interface.  This
512  * is an attachment helper.  It must be called after ifp->if_addrlen
513  * is initialized, which may not be the case when if_attach() is
514  * called.
515  */
516 void
517 if_alloc_sadl(struct ifnet *ifp)
518 {
519 	struct ifaddr *ifa;
520 	const struct sockaddr_dl *sdl;
521 
522 	/*
523 	 * If the interface already has a link name, release it
524 	 * now.  This is useful for interfaces that can change
525 	 * link types, and thus switch link names often.
526 	 */
527 	if (ifp->if_sadl != NULL)
528 		if_free_sadl(ifp, 0);
529 
530 	ifa = if_dl_create(ifp, &sdl);
531 
532 	ifa_insert(ifp, ifa);
533 	if_sadl_setrefs(ifp, ifa);
534 }
535 
536 static void
537 if_deactivate_sadl(struct ifnet *ifp)
538 {
539 	struct ifaddr *ifa;
540 
541 	KASSERT(ifp->if_dl != NULL);
542 
543 	ifa = ifp->if_dl;
544 
545 	ifp->if_sadl = NULL;
546 
547 	ifp->if_dl = NULL;
548 	ifafree(ifa);
549 }
550 
551 static void
552 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa)
553 {
554 	struct ifaddr *old;
555 
556 	KASSERT(ifp->if_dl != NULL);
557 
558 	old = ifp->if_dl;
559 
560 	ifaref(ifa);
561 	/* XXX Update if_dl and if_sadl atomically */
562 	ifp->if_dl = ifa;
563 	ifp->if_sadl = satosdl(ifa->ifa_addr);
564 
565 	ifafree(old);
566 }
567 
568 void
569 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
570     const struct sockaddr_dl *sdl)
571 {
572 	int s, ss;
573 	struct ifaddr *ifa;
574 	int bound = curlwp_bind();
575 
576 	KASSERT(ifa_held(ifa0));
577 
578 	s = splsoftnet();
579 
580 	if_replace_sadl(ifp, ifa0);
581 
582 	ss = pserialize_read_enter();
583 	IFADDR_READER_FOREACH(ifa, ifp) {
584 		struct psref psref;
585 		ifa_acquire(ifa, &psref);
586 		pserialize_read_exit(ss);
587 
588 		rtinit(ifa, RTM_LLINFO_UPD, 0);
589 
590 		ss = pserialize_read_enter();
591 		ifa_release(ifa, &psref);
592 	}
593 	pserialize_read_exit(ss);
594 
595 	splx(s);
596 	curlwp_bindx(bound);
597 }
598 
599 /*
600  * Free the link level name for the specified interface.  This is
601  * a detach helper.  This is called from if_detach().
602  */
603 void
604 if_free_sadl(struct ifnet *ifp, int factory)
605 {
606 	struct ifaddr *ifa;
607 	int s;
608 
609 	if (factory && ifp->if_hwdl != NULL) {
610 		ifa = ifp->if_hwdl;
611 		ifp->if_hwdl = NULL;
612 		ifafree(ifa);
613 	}
614 
615 	ifa = ifp->if_dl;
616 	if (ifa == NULL) {
617 		KASSERT(ifp->if_sadl == NULL);
618 		return;
619 	}
620 
621 	KASSERT(ifp->if_sadl != NULL);
622 
623 	s = splsoftnet();
624 	KASSERT(ifa->ifa_addr->sa_family == AF_LINK);
625 	ifa_remove(ifp, ifa);
626 	if_deactivate_sadl(ifp);
627 	splx(s);
628 }
629 
630 static void
631 if_getindex(ifnet_t *ifp)
632 {
633 	bool hitlimit = false;
634 
635 	ifp->if_index_gen = index_gen++;
636 
637 	ifp->if_index = if_index;
638 	if (ifindex2ifnet == NULL) {
639 		if_index++;
640 		goto skip;
641 	}
642 	while (if_byindex(ifp->if_index)) {
643 		/*
644 		 * If we hit USHRT_MAX, we skip back to 0 since
645 		 * there are a number of places where the value
646 		 * of if_index or if_index itself is compared
647 		 * to or stored in an unsigned short.  By
648 		 * jumping back, we won't botch those assignments
649 		 * or comparisons.
650 		 */
651 		if (++if_index == 0) {
652 			if_index = 1;
653 		} else if (if_index == USHRT_MAX) {
654 			/*
655 			 * However, if we have to jump back to
656 			 * zero *twice* without finding an empty
657 			 * slot in ifindex2ifnet[], then there
658 			 * there are too many (>65535) interfaces.
659 			 */
660 			if (hitlimit) {
661 				panic("too many interfaces");
662 			}
663 			hitlimit = true;
664 			if_index = 1;
665 		}
666 		ifp->if_index = if_index;
667 	}
668 skip:
669 	/*
670 	 * ifindex2ifnet is indexed by if_index. Since if_index will
671 	 * grow dynamically, it should grow too.
672 	 */
673 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
674 		size_t m, n, oldlim;
675 		void *q;
676 
677 		oldlim = if_indexlim;
678 		while (ifp->if_index >= if_indexlim)
679 			if_indexlim <<= 1;
680 
681 		/* grow ifindex2ifnet */
682 		m = oldlim * sizeof(struct ifnet *);
683 		n = if_indexlim * sizeof(struct ifnet *);
684 		q = malloc(n, M_IFADDR, M_WAITOK | M_ZERO);
685 		if (ifindex2ifnet != NULL) {
686 			memcpy(q, ifindex2ifnet, m);
687 			free(ifindex2ifnet, M_IFADDR);
688 		}
689 		ifindex2ifnet = (struct ifnet **)q;
690 	}
691 	ifindex2ifnet[ifp->if_index] = ifp;
692 }
693 
694 /*
695  * Initialize an interface and assign an index for it.
696  *
697  * It must be called prior to a device specific attach routine
698  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
699  * and be followed by if_register:
700  *
701  *     if_initialize(ifp);
702  *     ether_ifattach(ifp, enaddr);
703  *     if_register(ifp);
704  */
705 int
706 if_initialize(ifnet_t *ifp)
707 {
708 	int rv = 0;
709 
710 	KASSERT(if_indexlim > 0);
711 	TAILQ_INIT(&ifp->if_addrlist);
712 
713 	/*
714 	 * Link level name is allocated later by a separate call to
715 	 * if_alloc_sadl().
716 	 */
717 
718 	if (ifp->if_snd.ifq_maxlen == 0)
719 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
720 
721 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
722 
723 	ifp->if_link_state = LINK_STATE_UNKNOWN;
724 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
725 	ifp->if_link_scheduled = false;
726 
727 	ifp->if_capenable = 0;
728 	ifp->if_csum_flags_tx = 0;
729 	ifp->if_csum_flags_rx = 0;
730 
731 #ifdef ALTQ
732 	ifp->if_snd.altq_type = 0;
733 	ifp->if_snd.altq_disc = NULL;
734 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
735 	ifp->if_snd.altq_tbr  = NULL;
736 	ifp->if_snd.altq_ifp  = ifp;
737 #endif
738 
739 	IFQ_LOCK_INIT(&ifp->if_snd);
740 
741 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
742 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
743 
744 	IF_AFDATA_LOCK_INIT(ifp);
745 
746 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
747 	PSLIST_INIT(&ifp->if_addr_pslist);
748 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
749 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
750 	LIST_INIT(&ifp->if_multiaddrs);
751 	if ((rv = if_stats_init(ifp)) != 0) {
752 		goto fail;
753 	}
754 
755 	IFNET_GLOBAL_LOCK();
756 	if_getindex(ifp);
757 	IFNET_GLOBAL_UNLOCK();
758 
759 	return 0;
760 
761 fail:
762 	IF_AFDATA_LOCK_DESTROY(ifp);
763 
764 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
765 	(void)pfil_head_destroy(ifp->if_pfil);
766 
767 	IFQ_LOCK_DESTROY(&ifp->if_snd);
768 
769 	return rv;
770 }
771 
772 /*
773  * Register an interface to the list of "active" interfaces.
774  */
775 void
776 if_register(ifnet_t *ifp)
777 {
778 	/*
779 	 * If the driver has not supplied its own if_ioctl, then
780 	 * supply the default.
781 	 */
782 	if (ifp->if_ioctl == NULL)
783 		ifp->if_ioctl = ifioctl_common;
784 
785 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
786 
787 	if (!STAILQ_EMPTY(&domains))
788 		if_attachdomain1(ifp);
789 
790 	/* Announce the interface. */
791 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
792 
793 	if (ifp->if_slowtimo != NULL) {
794 		ifp->if_slowtimo_ch =
795 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
796 		callout_init(ifp->if_slowtimo_ch, 0);
797 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
798 		if_slowtimo(ifp);
799 	}
800 
801 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
802 		ifp->if_transmit = if_transmit;
803 
804 	IFNET_GLOBAL_LOCK();
805 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
806 	IFNET_WRITER_INSERT_TAIL(ifp);
807 	IFNET_GLOBAL_UNLOCK();
808 }
809 
810 /*
811  * The if_percpuq framework
812  *
813  * It allows network device drivers to execute the network stack
814  * in softint (so called softint-based if_input). It utilizes
815  * softint and percpu ifqueue. It doesn't distribute any packets
816  * between CPUs, unlike pktqueue(9).
817  *
818  * Currently we support two options for device drivers to apply the framework:
819  * - Use it implicitly with less changes
820  *   - If you use if_attach in driver's _attach function and if_input in
821  *     driver's Rx interrupt handler, a packet is queued and a softint handles
822  *     the packet implicitly
823  * - Use it explicitly in each driver (recommended)
824  *   - You can use if_percpuq_* directly in your driver
825  *   - In this case, you need to allocate struct if_percpuq in driver's softc
826  *   - See wm(4) as a reference implementation
827  */
828 
829 static void
830 if_percpuq_softint(void *arg)
831 {
832 	struct if_percpuq *ipq = arg;
833 	struct ifnet *ifp = ipq->ipq_ifp;
834 	struct mbuf *m;
835 
836 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
837 		if_statinc(ifp, if_ipackets);
838 		bpf_mtap(ifp, m, BPF_D_IN);
839 
840 		ifp->_if_input(ifp, m);
841 	}
842 }
843 
844 static void
845 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
846 {
847 	struct ifqueue *const ifq = p;
848 
849 	memset(ifq, 0, sizeof(*ifq));
850 	ifq->ifq_maxlen = IFQ_MAXLEN;
851 }
852 
853 struct if_percpuq *
854 if_percpuq_create(struct ifnet *ifp)
855 {
856 	struct if_percpuq *ipq;
857 	u_int flags = SOFTINT_NET;
858 
859 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
860 
861 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
862 	ipq->ipq_ifp = ifp;
863 	ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq);
864 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
865 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
866 
867 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
868 
869 	return ipq;
870 }
871 
872 static struct mbuf *
873 if_percpuq_dequeue(struct if_percpuq *ipq)
874 {
875 	struct mbuf *m;
876 	struct ifqueue *ifq;
877 	int s;
878 
879 	s = splnet();
880 	ifq = percpu_getref(ipq->ipq_ifqs);
881 	IF_DEQUEUE(ifq, m);
882 	percpu_putref(ipq->ipq_ifqs);
883 	splx(s);
884 
885 	return m;
886 }
887 
888 static void
889 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
890 {
891 	struct ifqueue *const ifq = p;
892 
893 	IF_PURGE(ifq);
894 }
895 
896 void
897 if_percpuq_destroy(struct if_percpuq *ipq)
898 {
899 
900 	/* if_detach may already destroy it */
901 	if (ipq == NULL)
902 		return;
903 
904 	softint_disestablish(ipq->ipq_si);
905 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
906 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
907 	kmem_free(ipq, sizeof(*ipq));
908 }
909 
910 void
911 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
912 {
913 	struct ifqueue *ifq;
914 	int s;
915 
916 	KASSERT(ipq != NULL);
917 
918 	s = splnet();
919 	ifq = percpu_getref(ipq->ipq_ifqs);
920 	if (IF_QFULL(ifq)) {
921 		IF_DROP(ifq);
922 		percpu_putref(ipq->ipq_ifqs);
923 		m_freem(m);
924 		goto out;
925 	}
926 	IF_ENQUEUE(ifq, m);
927 	percpu_putref(ipq->ipq_ifqs);
928 
929 	softint_schedule(ipq->ipq_si);
930 out:
931 	splx(s);
932 }
933 
934 static void
935 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
936 {
937 	struct ifqueue *const ifq = p;
938 	int *sum = arg;
939 
940 	*sum += ifq->ifq_drops;
941 }
942 
943 static int
944 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
945 {
946 	struct sysctlnode node;
947 	struct if_percpuq *ipq;
948 	int sum = 0;
949 	int error;
950 
951 	node = *rnode;
952 	ipq = node.sysctl_data;
953 
954 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
955 
956 	node.sysctl_data = &sum;
957 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
958 	if (error != 0 || newp == NULL)
959 		return error;
960 
961 	return 0;
962 }
963 
964 static void
965 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
966     struct if_percpuq *ipq)
967 {
968 	const struct sysctlnode *cnode, *rnode;
969 
970 	if (sysctl_createv(clog, 0, NULL, &rnode,
971 		       CTLFLAG_PERMANENT,
972 		       CTLTYPE_NODE, "interfaces",
973 		       SYSCTL_DESCR("Per-interface controls"),
974 		       NULL, 0, NULL, 0,
975 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
976 		goto bad;
977 
978 	if (sysctl_createv(clog, 0, &rnode, &rnode,
979 		       CTLFLAG_PERMANENT,
980 		       CTLTYPE_NODE, ifname,
981 		       SYSCTL_DESCR("Interface controls"),
982 		       NULL, 0, NULL, 0,
983 		       CTL_CREATE, CTL_EOL) != 0)
984 		goto bad;
985 
986 	if (sysctl_createv(clog, 0, &rnode, &rnode,
987 		       CTLFLAG_PERMANENT,
988 		       CTLTYPE_NODE, "rcvq",
989 		       SYSCTL_DESCR("Interface input queue controls"),
990 		       NULL, 0, NULL, 0,
991 		       CTL_CREATE, CTL_EOL) != 0)
992 		goto bad;
993 
994 #ifdef NOTYET
995 	/* XXX Should show each per-CPU queue length? */
996 	if (sysctl_createv(clog, 0, &rnode, &rnode,
997 		       CTLFLAG_PERMANENT,
998 		       CTLTYPE_INT, "len",
999 		       SYSCTL_DESCR("Current input queue length"),
1000 		       sysctl_percpuq_len, 0, NULL, 0,
1001 		       CTL_CREATE, CTL_EOL) != 0)
1002 		goto bad;
1003 
1004 	if (sysctl_createv(clog, 0, &rnode, &cnode,
1005 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1006 		       CTLTYPE_INT, "maxlen",
1007 		       SYSCTL_DESCR("Maximum allowed input queue length"),
1008 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
1009 		       CTL_CREATE, CTL_EOL) != 0)
1010 		goto bad;
1011 #endif
1012 
1013 	if (sysctl_createv(clog, 0, &rnode, &cnode,
1014 		       CTLFLAG_PERMANENT,
1015 		       CTLTYPE_INT, "drops",
1016 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
1017 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
1018 		       CTL_CREATE, CTL_EOL) != 0)
1019 		goto bad;
1020 
1021 	return;
1022 bad:
1023 	printf("%s: could not attach sysctl nodes\n", ifname);
1024 	return;
1025 }
1026 
1027 /*
1028  * The deferred if_start framework
1029  *
1030  * The common APIs to defer if_start to softint when if_start is requested
1031  * from a device driver running in hardware interrupt context.
1032  */
1033 /*
1034  * Call ifp->if_start (or equivalent) in a dedicated softint for
1035  * deferred if_start.
1036  */
1037 static void
1038 if_deferred_start_softint(void *arg)
1039 {
1040 	struct if_deferred_start *ids = arg;
1041 	struct ifnet *ifp = ids->ids_ifp;
1042 
1043 	ids->ids_if_start(ifp);
1044 }
1045 
1046 /*
1047  * The default callback function for deferred if_start.
1048  */
1049 static void
1050 if_deferred_start_common(struct ifnet *ifp)
1051 {
1052 	int s;
1053 
1054 	s = splnet();
1055 	if_start_lock(ifp);
1056 	splx(s);
1057 }
1058 
1059 static inline bool
1060 if_snd_is_used(struct ifnet *ifp)
1061 {
1062 
1063 	return ALTQ_IS_ENABLED(&ifp->if_snd) ||
1064 		ifp->if_transmit == if_transmit ||
1065 		ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit;
1066 }
1067 
1068 /*
1069  * Schedule deferred if_start.
1070  */
1071 void
1072 if_schedule_deferred_start(struct ifnet *ifp)
1073 {
1074 
1075 	KASSERT(ifp->if_deferred_start != NULL);
1076 
1077 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1078 		return;
1079 
1080 	softint_schedule(ifp->if_deferred_start->ids_si);
1081 }
1082 
1083 /*
1084  * Create an instance of deferred if_start. A driver should call the function
1085  * only if the driver needs deferred if_start. Drivers can setup their own
1086  * deferred if_start function via 2nd argument.
1087  */
1088 void
1089 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1090 {
1091 	struct if_deferred_start *ids;
1092 	u_int flags = SOFTINT_NET;
1093 
1094 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
1095 
1096 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1097 	ids->ids_ifp = ifp;
1098 	ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids);
1099 	if (func != NULL)
1100 		ids->ids_if_start = func;
1101 	else
1102 		ids->ids_if_start = if_deferred_start_common;
1103 
1104 	ifp->if_deferred_start = ids;
1105 }
1106 
1107 static void
1108 if_deferred_start_destroy(struct ifnet *ifp)
1109 {
1110 
1111 	if (ifp->if_deferred_start == NULL)
1112 		return;
1113 
1114 	softint_disestablish(ifp->if_deferred_start->ids_si);
1115 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1116 	ifp->if_deferred_start = NULL;
1117 }
1118 
1119 /*
1120  * The common interface input routine that is called by device drivers,
1121  * which should be used only when the driver's rx handler already runs
1122  * in softint.
1123  */
1124 void
1125 if_input(struct ifnet *ifp, struct mbuf *m)
1126 {
1127 
1128 	KASSERT(ifp->if_percpuq == NULL);
1129 	KASSERT(!cpu_intr_p());
1130 
1131 	if_statinc(ifp, if_ipackets);
1132 	bpf_mtap(ifp, m, BPF_D_IN);
1133 
1134 	ifp->_if_input(ifp, m);
1135 }
1136 
1137 /*
1138  * DEPRECATED. Use if_initialize and if_register instead.
1139  * See the above comment of if_initialize.
1140  *
1141  * Note that it implicitly enables if_percpuq to make drivers easy to
1142  * migrate softint-based if_input without much changes. If you don't
1143  * want to enable it, use if_initialize instead.
1144  */
1145 int
1146 if_attach(ifnet_t *ifp)
1147 {
1148 	int rv;
1149 
1150 	rv = if_initialize(ifp);
1151 	if (rv != 0)
1152 		return rv;
1153 
1154 	ifp->if_percpuq = if_percpuq_create(ifp);
1155 	if_register(ifp);
1156 
1157 	return 0;
1158 }
1159 
1160 void
1161 if_attachdomain(void)
1162 {
1163 	struct ifnet *ifp;
1164 	int s;
1165 	int bound = curlwp_bind();
1166 
1167 	s = pserialize_read_enter();
1168 	IFNET_READER_FOREACH(ifp) {
1169 		struct psref psref;
1170 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1171 		pserialize_read_exit(s);
1172 		if_attachdomain1(ifp);
1173 		s = pserialize_read_enter();
1174 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1175 	}
1176 	pserialize_read_exit(s);
1177 	curlwp_bindx(bound);
1178 }
1179 
1180 static void
1181 if_attachdomain1(struct ifnet *ifp)
1182 {
1183 	struct domain *dp;
1184 	int s;
1185 
1186 	s = splsoftnet();
1187 
1188 	/* address family dependent data region */
1189 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1190 	DOMAIN_FOREACH(dp) {
1191 		if (dp->dom_ifattach != NULL)
1192 			ifp->if_afdata[dp->dom_family] =
1193 			    (*dp->dom_ifattach)(ifp);
1194 	}
1195 
1196 	splx(s);
1197 }
1198 
1199 /*
1200  * Deactivate an interface.  This points all of the procedure
1201  * handles at error stubs.  May be called from interrupt context.
1202  */
1203 void
1204 if_deactivate(struct ifnet *ifp)
1205 {
1206 	int s;
1207 
1208 	s = splsoftnet();
1209 
1210 	ifp->if_output	 = if_nulloutput;
1211 	ifp->_if_input	 = if_nullinput;
1212 	ifp->if_start	 = if_nullstart;
1213 	ifp->if_transmit = if_nulltransmit;
1214 	ifp->if_ioctl	 = if_nullioctl;
1215 	ifp->if_init	 = if_nullinit;
1216 	ifp->if_stop	 = if_nullstop;
1217 	ifp->if_slowtimo = if_nullslowtimo;
1218 	ifp->if_drain	 = if_nulldrain;
1219 
1220 	ifp->if_link_state_changed = NULL;
1221 
1222 	/* No more packets may be enqueued. */
1223 	ifp->if_snd.ifq_maxlen = 0;
1224 
1225 	splx(s);
1226 }
1227 
1228 bool
1229 if_is_deactivated(const struct ifnet *ifp)
1230 {
1231 
1232 	return ifp->if_output == if_nulloutput;
1233 }
1234 
1235 void
1236 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1237 {
1238 	struct ifaddr *ifa, *nifa;
1239 	int s;
1240 
1241 	s = pserialize_read_enter();
1242 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1243 		nifa = IFADDR_READER_NEXT(ifa);
1244 		if (ifa->ifa_addr->sa_family != family)
1245 			continue;
1246 		pserialize_read_exit(s);
1247 
1248 		(*purgeaddr)(ifa);
1249 
1250 		s = pserialize_read_enter();
1251 	}
1252 	pserialize_read_exit(s);
1253 }
1254 
1255 #ifdef IFAREF_DEBUG
1256 static struct ifaddr **ifa_list;
1257 static int ifa_list_size;
1258 
1259 /* Depends on only one if_attach runs at once */
1260 static void
1261 if_build_ifa_list(struct ifnet *ifp)
1262 {
1263 	struct ifaddr *ifa;
1264 	int i;
1265 
1266 	KASSERT(ifa_list == NULL);
1267 	KASSERT(ifa_list_size == 0);
1268 
1269 	IFADDR_READER_FOREACH(ifa, ifp)
1270 		ifa_list_size++;
1271 
1272 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1273 	i = 0;
1274 	IFADDR_READER_FOREACH(ifa, ifp) {
1275 		ifa_list[i++] = ifa;
1276 		ifaref(ifa);
1277 	}
1278 }
1279 
1280 static void
1281 if_check_and_free_ifa_list(struct ifnet *ifp)
1282 {
1283 	int i;
1284 	struct ifaddr *ifa;
1285 
1286 	if (ifa_list == NULL)
1287 		return;
1288 
1289 	for (i = 0; i < ifa_list_size; i++) {
1290 		char buf[64];
1291 
1292 		ifa = ifa_list[i];
1293 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1294 		if (ifa->ifa_refcnt > 1) {
1295 			log(LOG_WARNING,
1296 			    "ifa(%s) still referenced (refcnt=%d)\n",
1297 			    buf, ifa->ifa_refcnt - 1);
1298 		} else
1299 			log(LOG_DEBUG,
1300 			    "ifa(%s) not referenced (refcnt=%d)\n",
1301 			    buf, ifa->ifa_refcnt - 1);
1302 		ifafree(ifa);
1303 	}
1304 
1305 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1306 	ifa_list = NULL;
1307 	ifa_list_size = 0;
1308 }
1309 #endif
1310 
1311 /*
1312  * Detach an interface from the list of "active" interfaces,
1313  * freeing any resources as we go along.
1314  *
1315  * NOTE: This routine must be called with a valid thread context,
1316  * as it may block.
1317  */
1318 void
1319 if_detach(struct ifnet *ifp)
1320 {
1321 	struct socket so;
1322 	struct ifaddr *ifa;
1323 #ifdef IFAREF_DEBUG
1324 	struct ifaddr *last_ifa = NULL;
1325 #endif
1326 	struct domain *dp;
1327 	const struct protosw *pr;
1328 	int s, i, family, purged;
1329 
1330 #ifdef IFAREF_DEBUG
1331 	if_build_ifa_list(ifp);
1332 #endif
1333 	/*
1334 	 * XXX It's kind of lame that we have to have the
1335 	 * XXX socket structure...
1336 	 */
1337 	memset(&so, 0, sizeof(so));
1338 
1339 	s = splnet();
1340 
1341 	sysctl_teardown(&ifp->if_sysctl_log);
1342 
1343 	IFNET_LOCK(ifp);
1344 
1345 	/*
1346 	 * Unset all queued link states and pretend a
1347 	 * link state change is scheduled.
1348 	 * This stops any more link state changes occuring for this
1349 	 * interface while it's being detached so it's safe
1350 	 * to drain the workqueue.
1351 	 */
1352 	IF_LINK_STATE_CHANGE_LOCK(ifp);
1353 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
1354 	ifp->if_link_scheduled = true;
1355 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
1356 	workqueue_wait(ifnet_link_state_wq, &ifp->if_link_work);
1357 
1358 	if_deactivate(ifp);
1359 	IFNET_UNLOCK(ifp);
1360 
1361 	/*
1362 	 * Unlink from the list and wait for all readers to leave
1363 	 * from pserialize read sections.  Note that we can't do
1364 	 * psref_target_destroy here.  See below.
1365 	 */
1366 	IFNET_GLOBAL_LOCK();
1367 	ifindex2ifnet[ifp->if_index] = NULL;
1368 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1369 	IFNET_WRITER_REMOVE(ifp);
1370 	pserialize_perform(ifnet_psz);
1371 	IFNET_GLOBAL_UNLOCK();
1372 
1373 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1374 		ifp->if_slowtimo = NULL;
1375 		callout_halt(ifp->if_slowtimo_ch, NULL);
1376 		callout_destroy(ifp->if_slowtimo_ch);
1377 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1378 	}
1379 	if_deferred_start_destroy(ifp);
1380 
1381 	/*
1382 	 * Do an if_down() to give protocols a chance to do something.
1383 	 */
1384 	if_down_deactivated(ifp);
1385 
1386 #ifdef ALTQ
1387 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1388 		altq_disable(&ifp->if_snd);
1389 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1390 		altq_detach(&ifp->if_snd);
1391 #endif
1392 
1393 #if NCARP > 0
1394 	/* Remove the interface from any carp group it is a part of.  */
1395 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1396 		carp_ifdetach(ifp);
1397 #endif
1398 
1399 	/*
1400 	 * Rip all the addresses off the interface.  This should make
1401 	 * all of the routes go away.
1402 	 *
1403 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1404 	 * from the list, including our "cursor", ifa.  For safety,
1405 	 * and to honor the TAILQ abstraction, I just restart the
1406 	 * loop after each removal.  Note that the loop will exit
1407 	 * when all of the remaining ifaddrs belong to the AF_LINK
1408 	 * family.  I am counting on the historical fact that at
1409 	 * least one pr_usrreq in each address domain removes at
1410 	 * least one ifaddr.
1411 	 */
1412 again:
1413 	/*
1414 	 * At this point, no other one tries to remove ifa in the list,
1415 	 * so we don't need to take a lock or psref.  Avoid using
1416 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1417 	 * violations of pserialize.
1418 	 */
1419 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1420 		family = ifa->ifa_addr->sa_family;
1421 #ifdef IFAREF_DEBUG
1422 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1423 		    ifa, family, ifa->ifa_refcnt);
1424 		if (last_ifa != NULL && ifa == last_ifa)
1425 			panic("if_detach: loop detected");
1426 		last_ifa = ifa;
1427 #endif
1428 		if (family == AF_LINK)
1429 			continue;
1430 		dp = pffinddomain(family);
1431 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1432 		/*
1433 		 * XXX These PURGEIF calls are redundant with the
1434 		 * purge-all-families calls below, but are left in for
1435 		 * now both to make a smaller change, and to avoid
1436 		 * unplanned interactions with clearing of
1437 		 * ifp->if_addrlist.
1438 		 */
1439 		purged = 0;
1440 		for (pr = dp->dom_protosw;
1441 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1442 			so.so_proto = pr;
1443 			if (pr->pr_usrreqs) {
1444 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1445 				purged = 1;
1446 			}
1447 		}
1448 		if (purged == 0) {
1449 			/*
1450 			 * XXX What's really the best thing to do
1451 			 * XXX here?  --thorpej@NetBSD.org
1452 			 */
1453 			printf("if_detach: WARNING: AF %d not purged\n",
1454 			    family);
1455 			ifa_remove(ifp, ifa);
1456 		}
1457 		goto again;
1458 	}
1459 
1460 	if_free_sadl(ifp, 1);
1461 
1462 restart:
1463 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1464 		family = ifa->ifa_addr->sa_family;
1465 		KASSERT(family == AF_LINK);
1466 		ifa_remove(ifp, ifa);
1467 		goto restart;
1468 	}
1469 
1470 	/* Delete stray routes from the routing table. */
1471 	for (i = 0; i <= AF_MAX; i++)
1472 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1473 
1474 	DOMAIN_FOREACH(dp) {
1475 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1476 		{
1477 			void *p = ifp->if_afdata[dp->dom_family];
1478 			if (p) {
1479 				ifp->if_afdata[dp->dom_family] = NULL;
1480 				(*dp->dom_ifdetach)(ifp, p);
1481 			}
1482 		}
1483 
1484 		/*
1485 		 * One would expect multicast memberships (INET and
1486 		 * INET6) on UDP sockets to be purged by the PURGEIF
1487 		 * calls above, but if all addresses were removed from
1488 		 * the interface prior to destruction, the calls will
1489 		 * not be made (e.g. ppp, for which pppd(8) generally
1490 		 * removes addresses before destroying the interface).
1491 		 * Because there is no invariant that multicast
1492 		 * memberships only exist for interfaces with IPv4
1493 		 * addresses, we must call PURGEIF regardless of
1494 		 * addresses.  (Protocols which might store ifnet
1495 		 * pointers are marked with PR_PURGEIF.)
1496 		 */
1497 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1498 			so.so_proto = pr;
1499 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1500 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1501 		}
1502 	}
1503 
1504 	/*
1505 	 * Must be done after the above pr_purgeif because if_psref may be
1506 	 * still used in pr_purgeif.
1507 	 */
1508 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1509 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1510 
1511 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1512 	(void)pfil_head_destroy(ifp->if_pfil);
1513 
1514 	/* Announce that the interface is gone. */
1515 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1516 
1517 	IF_AFDATA_LOCK_DESTROY(ifp);
1518 
1519 	/*
1520 	 * remove packets that came from ifp, from software interrupt queues.
1521 	 */
1522 	DOMAIN_FOREACH(dp) {
1523 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1524 			struct ifqueue *iq = dp->dom_ifqueues[i];
1525 			if (iq == NULL)
1526 				break;
1527 			dp->dom_ifqueues[i] = NULL;
1528 			if_detach_queues(ifp, iq);
1529 		}
1530 	}
1531 
1532 	/*
1533 	 * IP queues have to be processed separately: net-queue barrier
1534 	 * ensures that the packets are dequeued while a cross-call will
1535 	 * ensure that the interrupts have completed. FIXME: not quite..
1536 	 */
1537 #ifdef INET
1538 	pktq_barrier(ip_pktq);
1539 #endif
1540 #ifdef INET6
1541 	if (in6_present)
1542 		pktq_barrier(ip6_pktq);
1543 #endif
1544 	xc_barrier(0);
1545 
1546 	if (ifp->if_percpuq != NULL) {
1547 		if_percpuq_destroy(ifp->if_percpuq);
1548 		ifp->if_percpuq = NULL;
1549 	}
1550 
1551 	mutex_obj_free(ifp->if_ioctl_lock);
1552 	ifp->if_ioctl_lock = NULL;
1553 	mutex_obj_free(ifp->if_snd.ifq_lock);
1554 	if_stats_fini(ifp);
1555 
1556 	splx(s);
1557 
1558 #ifdef IFAREF_DEBUG
1559 	if_check_and_free_ifa_list(ifp);
1560 #endif
1561 }
1562 
1563 static void
1564 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1565 {
1566 	struct mbuf *m, *prev, *next;
1567 
1568 	prev = NULL;
1569 	for (m = q->ifq_head; m != NULL; m = next) {
1570 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1571 
1572 		next = m->m_nextpkt;
1573 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1574 			prev = m;
1575 			continue;
1576 		}
1577 
1578 		if (prev != NULL)
1579 			prev->m_nextpkt = m->m_nextpkt;
1580 		else
1581 			q->ifq_head = m->m_nextpkt;
1582 		if (q->ifq_tail == m)
1583 			q->ifq_tail = prev;
1584 		q->ifq_len--;
1585 
1586 		m->m_nextpkt = NULL;
1587 		m_freem(m);
1588 		IF_DROP(q);
1589 	}
1590 }
1591 
1592 /*
1593  * Callback for a radix tree walk to delete all references to an
1594  * ifnet.
1595  */
1596 static int
1597 if_delroute_matcher(struct rtentry *rt, void *v)
1598 {
1599 	struct ifnet *ifp = (struct ifnet *)v;
1600 
1601 	if (rt->rt_ifp == ifp)
1602 		return 1;
1603 	else
1604 		return 0;
1605 }
1606 
1607 /*
1608  * Create a clone network interface.
1609  */
1610 static int
1611 if_clone_create(const char *name)
1612 {
1613 	struct if_clone *ifc;
1614 	int unit;
1615 	struct ifnet *ifp;
1616 	struct psref psref;
1617 
1618 	KASSERT(mutex_owned(&if_clone_mtx));
1619 
1620 	ifc = if_clone_lookup(name, &unit);
1621 	if (ifc == NULL)
1622 		return EINVAL;
1623 
1624 	ifp = if_get(name, &psref);
1625 	if (ifp != NULL) {
1626 		if_put(ifp, &psref);
1627 		return EEXIST;
1628 	}
1629 
1630 	return (*ifc->ifc_create)(ifc, unit);
1631 }
1632 
1633 /*
1634  * Destroy a clone network interface.
1635  */
1636 static int
1637 if_clone_destroy(const char *name)
1638 {
1639 	struct if_clone *ifc;
1640 	struct ifnet *ifp;
1641 	struct psref psref;
1642 	int error;
1643 	int (*if_ioctl)(struct ifnet *, u_long, void *);
1644 
1645 	KASSERT(mutex_owned(&if_clone_mtx));
1646 
1647 	ifc = if_clone_lookup(name, NULL);
1648 	if (ifc == NULL)
1649 		return EINVAL;
1650 
1651 	if (ifc->ifc_destroy == NULL)
1652 		return EOPNOTSUPP;
1653 
1654 	ifp = if_get(name, &psref);
1655 	if (ifp == NULL)
1656 		return ENXIO;
1657 
1658 	/* We have to disable ioctls here */
1659 	IFNET_LOCK(ifp);
1660 	if_ioctl = ifp->if_ioctl;
1661 	ifp->if_ioctl = if_nullioctl;
1662 	IFNET_UNLOCK(ifp);
1663 
1664 	/*
1665 	 * We cannot call ifc_destroy with holding ifp.
1666 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1667 	 */
1668 	if_put(ifp, &psref);
1669 
1670 	error = (*ifc->ifc_destroy)(ifp);
1671 
1672 	if (error != 0) {
1673 		/* We have to restore if_ioctl on error */
1674 		IFNET_LOCK(ifp);
1675 		ifp->if_ioctl = if_ioctl;
1676 		IFNET_UNLOCK(ifp);
1677 	}
1678 
1679 	return error;
1680 }
1681 
1682 static bool
1683 if_is_unit(const char *name)
1684 {
1685 
1686 	while (*name != '\0') {
1687 		if (*name < '0' || *name > '9')
1688 			return false;
1689 		name++;
1690 	}
1691 
1692 	return true;
1693 }
1694 
1695 /*
1696  * Look up a network interface cloner.
1697  */
1698 static struct if_clone *
1699 if_clone_lookup(const char *name, int *unitp)
1700 {
1701 	struct if_clone *ifc;
1702 	const char *cp;
1703 	char *dp, ifname[IFNAMSIZ + 3];
1704 	int unit;
1705 
1706 	KASSERT(mutex_owned(&if_clone_mtx));
1707 
1708 	strcpy(ifname, "if_");
1709 	/* separate interface name from unit */
1710 	/* TODO: search unit number from backward */
1711 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1712 	    *cp && !if_is_unit(cp);)
1713 		*dp++ = *cp++;
1714 
1715 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1716 		return NULL;	/* No name or unit number */
1717 	*dp++ = '\0';
1718 
1719 again:
1720 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1721 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1722 			break;
1723 	}
1724 
1725 	if (ifc == NULL) {
1726 		int error;
1727 		if (*ifname == '\0')
1728 			return NULL;
1729 		mutex_exit(&if_clone_mtx);
1730 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1731 		mutex_enter(&if_clone_mtx);
1732 		if (error)
1733 			return NULL;
1734 		*ifname = '\0';
1735 		goto again;
1736 	}
1737 
1738 	unit = 0;
1739 	while (cp - name < IFNAMSIZ && *cp) {
1740 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1741 			/* Bogus unit number. */
1742 			return NULL;
1743 		}
1744 		unit = (unit * 10) + (*cp++ - '0');
1745 	}
1746 
1747 	if (unitp != NULL)
1748 		*unitp = unit;
1749 	return ifc;
1750 }
1751 
1752 /*
1753  * Register a network interface cloner.
1754  */
1755 void
1756 if_clone_attach(struct if_clone *ifc)
1757 {
1758 
1759 	mutex_enter(&if_clone_mtx);
1760 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1761 	if_cloners_count++;
1762 	mutex_exit(&if_clone_mtx);
1763 }
1764 
1765 /*
1766  * Unregister a network interface cloner.
1767  */
1768 void
1769 if_clone_detach(struct if_clone *ifc)
1770 {
1771 
1772 	mutex_enter(&if_clone_mtx);
1773 	LIST_REMOVE(ifc, ifc_list);
1774 	if_cloners_count--;
1775 	mutex_exit(&if_clone_mtx);
1776 }
1777 
1778 /*
1779  * Provide list of interface cloners to userspace.
1780  */
1781 int
1782 if_clone_list(int buf_count, char *buffer, int *total)
1783 {
1784 	char outbuf[IFNAMSIZ], *dst;
1785 	struct if_clone *ifc;
1786 	int count, error = 0;
1787 
1788 	mutex_enter(&if_clone_mtx);
1789 	*total = if_cloners_count;
1790 	if ((dst = buffer) == NULL) {
1791 		/* Just asking how many there are. */
1792 		goto out;
1793 	}
1794 
1795 	if (buf_count < 0) {
1796 		error = EINVAL;
1797 		goto out;
1798 	}
1799 
1800 	count = (if_cloners_count < buf_count) ?
1801 	    if_cloners_count : buf_count;
1802 
1803 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1804 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1805 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1806 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1807 			error = ENAMETOOLONG;
1808 			goto out;
1809 		}
1810 		error = copyout(outbuf, dst, sizeof(outbuf));
1811 		if (error != 0)
1812 			break;
1813 	}
1814 
1815 out:
1816 	mutex_exit(&if_clone_mtx);
1817 	return error;
1818 }
1819 
1820 void
1821 ifa_psref_init(struct ifaddr *ifa)
1822 {
1823 
1824 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1825 }
1826 
1827 void
1828 ifaref(struct ifaddr *ifa)
1829 {
1830 
1831 	atomic_inc_uint(&ifa->ifa_refcnt);
1832 }
1833 
1834 void
1835 ifafree(struct ifaddr *ifa)
1836 {
1837 	KASSERT(ifa != NULL);
1838 	KASSERTMSG(ifa->ifa_refcnt > 0, "ifa_refcnt=%d", ifa->ifa_refcnt);
1839 
1840 	if (atomic_dec_uint_nv(&ifa->ifa_refcnt) == 0) {
1841 		free(ifa, M_IFADDR);
1842 	}
1843 }
1844 
1845 bool
1846 ifa_is_destroying(struct ifaddr *ifa)
1847 {
1848 
1849 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1850 }
1851 
1852 void
1853 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1854 {
1855 
1856 	ifa->ifa_ifp = ifp;
1857 
1858 	/*
1859 	 * Check MP-safety for IFEF_MPSAFE drivers.
1860 	 * Check !IFF_RUNNING for initialization routines that normally don't
1861 	 * take IFNET_LOCK but it's safe because there is no competitor.
1862 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1863 	 * if_stop.
1864 	 */
1865 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1866 	    IFNET_LOCKED(ifp));
1867 
1868 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1869 	IFADDR_ENTRY_INIT(ifa);
1870 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1871 
1872 	ifaref(ifa);
1873 }
1874 
1875 void
1876 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1877 {
1878 
1879 	KASSERT(ifa->ifa_ifp == ifp);
1880 	/*
1881 	 * Check MP-safety for IFEF_MPSAFE drivers.
1882 	 * if_is_deactivated indicates ifa_remove is called form if_detach
1883 	 * where is safe even if IFNET_LOCK isn't held.
1884 	 */
1885 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp));
1886 
1887 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1888 	IFADDR_WRITER_REMOVE(ifa);
1889 #ifdef NET_MPSAFE
1890 	IFNET_GLOBAL_LOCK();
1891 	pserialize_perform(ifnet_psz);
1892 	IFNET_GLOBAL_UNLOCK();
1893 #endif
1894 
1895 #ifdef NET_MPSAFE
1896 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1897 #endif
1898 	IFADDR_ENTRY_DESTROY(ifa);
1899 	ifafree(ifa);
1900 }
1901 
1902 void
1903 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1904 {
1905 
1906 	PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
1907 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1908 }
1909 
1910 void
1911 ifa_release(struct ifaddr *ifa, struct psref *psref)
1912 {
1913 
1914 	if (ifa == NULL)
1915 		return;
1916 
1917 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1918 }
1919 
1920 bool
1921 ifa_held(struct ifaddr *ifa)
1922 {
1923 
1924 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1925 }
1926 
1927 static inline int
1928 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1929 {
1930 	return sockaddr_cmp(sa1, sa2) == 0;
1931 }
1932 
1933 /*
1934  * Locate an interface based on a complete address.
1935  */
1936 /*ARGSUSED*/
1937 struct ifaddr *
1938 ifa_ifwithaddr(const struct sockaddr *addr)
1939 {
1940 	struct ifnet *ifp;
1941 	struct ifaddr *ifa;
1942 
1943 	IFNET_READER_FOREACH(ifp) {
1944 		if (if_is_deactivated(ifp))
1945 			continue;
1946 		IFADDR_READER_FOREACH(ifa, ifp) {
1947 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1948 				continue;
1949 			if (equal(addr, ifa->ifa_addr))
1950 				return ifa;
1951 			if ((ifp->if_flags & IFF_BROADCAST) &&
1952 			    ifa->ifa_broadaddr &&
1953 			    /* IP6 doesn't have broadcast */
1954 			    ifa->ifa_broadaddr->sa_len != 0 &&
1955 			    equal(ifa->ifa_broadaddr, addr))
1956 				return ifa;
1957 		}
1958 	}
1959 	return NULL;
1960 }
1961 
1962 struct ifaddr *
1963 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1964 {
1965 	struct ifaddr *ifa;
1966 	int s = pserialize_read_enter();
1967 
1968 	ifa = ifa_ifwithaddr(addr);
1969 	if (ifa != NULL)
1970 		ifa_acquire(ifa, psref);
1971 	pserialize_read_exit(s);
1972 
1973 	return ifa;
1974 }
1975 
1976 /*
1977  * Locate the point to point interface with a given destination address.
1978  */
1979 /*ARGSUSED*/
1980 struct ifaddr *
1981 ifa_ifwithdstaddr(const struct sockaddr *addr)
1982 {
1983 	struct ifnet *ifp;
1984 	struct ifaddr *ifa;
1985 
1986 	IFNET_READER_FOREACH(ifp) {
1987 		if (if_is_deactivated(ifp))
1988 			continue;
1989 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1990 			continue;
1991 		IFADDR_READER_FOREACH(ifa, ifp) {
1992 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1993 			    ifa->ifa_dstaddr == NULL)
1994 				continue;
1995 			if (equal(addr, ifa->ifa_dstaddr))
1996 				return ifa;
1997 		}
1998 	}
1999 
2000 	return NULL;
2001 }
2002 
2003 struct ifaddr *
2004 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
2005 {
2006 	struct ifaddr *ifa;
2007 	int s;
2008 
2009 	s = pserialize_read_enter();
2010 	ifa = ifa_ifwithdstaddr(addr);
2011 	if (ifa != NULL)
2012 		ifa_acquire(ifa, psref);
2013 	pserialize_read_exit(s);
2014 
2015 	return ifa;
2016 }
2017 
2018 /*
2019  * Find an interface on a specific network.  If many, choice
2020  * is most specific found.
2021  */
2022 struct ifaddr *
2023 ifa_ifwithnet(const struct sockaddr *addr)
2024 {
2025 	struct ifnet *ifp;
2026 	struct ifaddr *ifa, *ifa_maybe = NULL;
2027 	const struct sockaddr_dl *sdl;
2028 	u_int af = addr->sa_family;
2029 	const char *addr_data = addr->sa_data, *cplim;
2030 
2031 	if (af == AF_LINK) {
2032 		sdl = satocsdl(addr);
2033 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
2034 		    ifindex2ifnet[sdl->sdl_index] &&
2035 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
2036 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
2037 		}
2038 	}
2039 #ifdef NETATALK
2040 	if (af == AF_APPLETALK) {
2041 		const struct sockaddr_at *sat, *sat2;
2042 		sat = (const struct sockaddr_at *)addr;
2043 		IFNET_READER_FOREACH(ifp) {
2044 			if (if_is_deactivated(ifp))
2045 				continue;
2046 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
2047 			if (ifa == NULL)
2048 				continue;
2049 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
2050 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
2051 				return ifa; /* exact match */
2052 			if (ifa_maybe == NULL) {
2053 				/* else keep the if with the right range */
2054 				ifa_maybe = ifa;
2055 			}
2056 		}
2057 		return ifa_maybe;
2058 	}
2059 #endif
2060 	IFNET_READER_FOREACH(ifp) {
2061 		if (if_is_deactivated(ifp))
2062 			continue;
2063 		IFADDR_READER_FOREACH(ifa, ifp) {
2064 			const char *cp, *cp2, *cp3;
2065 
2066 			if (ifa->ifa_addr->sa_family != af ||
2067 			    ifa->ifa_netmask == NULL)
2068  next:				continue;
2069 			cp = addr_data;
2070 			cp2 = ifa->ifa_addr->sa_data;
2071 			cp3 = ifa->ifa_netmask->sa_data;
2072 			cplim = (const char *)ifa->ifa_netmask +
2073 			    ifa->ifa_netmask->sa_len;
2074 			while (cp3 < cplim) {
2075 				if ((*cp++ ^ *cp2++) & *cp3++) {
2076 					/* want to continue for() loop */
2077 					goto next;
2078 				}
2079 			}
2080 			if (ifa_maybe == NULL ||
2081 			    rt_refines(ifa->ifa_netmask,
2082 			               ifa_maybe->ifa_netmask))
2083 				ifa_maybe = ifa;
2084 		}
2085 	}
2086 	return ifa_maybe;
2087 }
2088 
2089 struct ifaddr *
2090 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2091 {
2092 	struct ifaddr *ifa;
2093 	int s;
2094 
2095 	s = pserialize_read_enter();
2096 	ifa = ifa_ifwithnet(addr);
2097 	if (ifa != NULL)
2098 		ifa_acquire(ifa, psref);
2099 	pserialize_read_exit(s);
2100 
2101 	return ifa;
2102 }
2103 
2104 /*
2105  * Find the interface of the addresss.
2106  */
2107 struct ifaddr *
2108 ifa_ifwithladdr(const struct sockaddr *addr)
2109 {
2110 	struct ifaddr *ia;
2111 
2112 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2113 	    (ia = ifa_ifwithnet(addr)))
2114 		return ia;
2115 	return NULL;
2116 }
2117 
2118 struct ifaddr *
2119 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2120 {
2121 	struct ifaddr *ifa;
2122 	int s;
2123 
2124 	s = pserialize_read_enter();
2125 	ifa = ifa_ifwithladdr(addr);
2126 	if (ifa != NULL)
2127 		ifa_acquire(ifa, psref);
2128 	pserialize_read_exit(s);
2129 
2130 	return ifa;
2131 }
2132 
2133 /*
2134  * Find an interface using a specific address family
2135  */
2136 struct ifaddr *
2137 ifa_ifwithaf(int af)
2138 {
2139 	struct ifnet *ifp;
2140 	struct ifaddr *ifa = NULL;
2141 	int s;
2142 
2143 	s = pserialize_read_enter();
2144 	IFNET_READER_FOREACH(ifp) {
2145 		if (if_is_deactivated(ifp))
2146 			continue;
2147 		IFADDR_READER_FOREACH(ifa, ifp) {
2148 			if (ifa->ifa_addr->sa_family == af)
2149 				goto out;
2150 		}
2151 	}
2152 out:
2153 	pserialize_read_exit(s);
2154 	return ifa;
2155 }
2156 
2157 /*
2158  * Find an interface address specific to an interface best matching
2159  * a given address.
2160  */
2161 struct ifaddr *
2162 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2163 {
2164 	struct ifaddr *ifa;
2165 	const char *cp, *cp2, *cp3;
2166 	const char *cplim;
2167 	struct ifaddr *ifa_maybe = 0;
2168 	u_int af = addr->sa_family;
2169 
2170 	if (if_is_deactivated(ifp))
2171 		return NULL;
2172 
2173 	if (af >= AF_MAX)
2174 		return NULL;
2175 
2176 	IFADDR_READER_FOREACH(ifa, ifp) {
2177 		if (ifa->ifa_addr->sa_family != af)
2178 			continue;
2179 		ifa_maybe = ifa;
2180 		if (ifa->ifa_netmask == NULL) {
2181 			if (equal(addr, ifa->ifa_addr) ||
2182 			    (ifa->ifa_dstaddr &&
2183 			     equal(addr, ifa->ifa_dstaddr)))
2184 				return ifa;
2185 			continue;
2186 		}
2187 		cp = addr->sa_data;
2188 		cp2 = ifa->ifa_addr->sa_data;
2189 		cp3 = ifa->ifa_netmask->sa_data;
2190 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2191 		for (; cp3 < cplim; cp3++) {
2192 			if ((*cp++ ^ *cp2++) & *cp3)
2193 				break;
2194 		}
2195 		if (cp3 == cplim)
2196 			return ifa;
2197 	}
2198 	return ifa_maybe;
2199 }
2200 
2201 struct ifaddr *
2202 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2203     struct psref *psref)
2204 {
2205 	struct ifaddr *ifa;
2206 	int s;
2207 
2208 	s = pserialize_read_enter();
2209 	ifa = ifaof_ifpforaddr(addr, ifp);
2210 	if (ifa != NULL)
2211 		ifa_acquire(ifa, psref);
2212 	pserialize_read_exit(s);
2213 
2214 	return ifa;
2215 }
2216 
2217 /*
2218  * Default action when installing a route with a Link Level gateway.
2219  * Lookup an appropriate real ifa to point to.
2220  * This should be moved to /sys/net/link.c eventually.
2221  */
2222 void
2223 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2224 {
2225 	struct ifaddr *ifa;
2226 	const struct sockaddr *dst;
2227 	struct ifnet *ifp;
2228 	struct psref psref;
2229 
2230 	if (cmd != RTM_ADD || ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2231 		return;
2232 	ifp = rt->rt_ifa->ifa_ifp;
2233 	dst = rt_getkey(rt);
2234 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2235 		rt_replace_ifa(rt, ifa);
2236 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2237 			ifa->ifa_rtrequest(cmd, rt, info);
2238 		ifa_release(ifa, &psref);
2239 	}
2240 }
2241 
2242 /*
2243  * bitmask macros to manage a densely packed link_state change queue.
2244  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2245  * LINK_STATE_UP(2) we need 2 bits for each state change.
2246  * As a state change to store is 0, treat all bits set as an unset item.
2247  */
2248 #define LQ_ITEM_BITS		2
2249 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2250 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2251 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2252 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2253 #define LQ_STORE(q, i, v)						      \
2254 	do {								      \
2255 		(q) &= ~LQ_MASK((i));					      \
2256 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2257 	} while (0 /* CONSTCOND */)
2258 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2259 #define LQ_POP(q, v)							      \
2260 	do {								      \
2261 		(v) = LQ_ITEM((q), 0);					      \
2262 		(q) >>= LQ_ITEM_BITS;					      \
2263 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2264 	} while (0 /* CONSTCOND */)
2265 #define LQ_PUSH(q, v)							      \
2266 	do {								      \
2267 		(q) >>= LQ_ITEM_BITS;					      \
2268 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2269 	} while (0 /* CONSTCOND */)
2270 #define LQ_FIND_UNSET(q, i)						      \
2271 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2272 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2273 			break;						      \
2274 	}
2275 
2276 /*
2277  * Handle a change in the interface link state and
2278  * queue notifications.
2279  */
2280 void
2281 if_link_state_change(struct ifnet *ifp, int link_state)
2282 {
2283 	int idx;
2284 
2285 	/* Ensure change is to a valid state */
2286 	switch (link_state) {
2287 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2288 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2289 	case LINK_STATE_UP:
2290 		break;
2291 	default:
2292 #ifdef DEBUG
2293 		printf("%s: invalid link state %d\n",
2294 		    ifp->if_xname, link_state);
2295 #endif
2296 		return;
2297 	}
2298 
2299 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2300 
2301 	/* Find the last unset event in the queue. */
2302 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2303 
2304 	if (idx == 0) {
2305 		/*
2306 		 * There is no queue of link state changes.
2307 		 * As we have the lock we can safely compare against the
2308 		 * current link state and return if the same.
2309 		 * Otherwise, if scheduled is true then the interface is being
2310 		 * detached and the queue is being drained so we need
2311 		 * to avoid queuing more work.
2312 		 */
2313 		 if (ifp->if_link_state == link_state || ifp->if_link_scheduled)
2314 			goto out;
2315 	} else {
2316 		/* Ensure link_state doesn't match the last queued state. */
2317 		if (LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2318 			goto out;
2319 	}
2320 
2321 	/* Handle queue overflow. */
2322 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2323 		uint8_t lost;
2324 
2325 		/*
2326 		 * The DOWN state must be protected from being pushed off
2327 		 * the queue to ensure that userland will always be
2328 		 * in a sane state.
2329 		 * Because DOWN is protected, there is no need to protect
2330 		 * UNKNOWN.
2331 		 * It should be invalid to change from any other state to
2332 		 * UNKNOWN anyway ...
2333 		 */
2334 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2335 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2336 		if (lost == LINK_STATE_DOWN) {
2337 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2338 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2339 		}
2340 		printf("%s: lost link state change %s\n",
2341 		    ifp->if_xname,
2342 		    lost == LINK_STATE_UP ? "UP" :
2343 		    lost == LINK_STATE_DOWN ? "DOWN" :
2344 		    "UNKNOWN");
2345 	} else
2346 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2347 
2348 	if (ifp->if_link_scheduled)
2349 		goto out;
2350 
2351 	ifp->if_link_scheduled = true;
2352 	workqueue_enqueue(ifnet_link_state_wq, &ifp->if_link_work, NULL);
2353 
2354 out:
2355 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2356 }
2357 
2358 /*
2359  * Handle interface link state change notifications.
2360  */
2361 static void
2362 if_link_state_change_process(struct ifnet *ifp, int link_state)
2363 {
2364 	struct domain *dp;
2365 	int s = splnet();
2366 	bool notify;
2367 
2368 	KASSERT(!cpu_intr_p());
2369 
2370 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2371 
2372 	/* Ensure the change is still valid. */
2373 	if (ifp->if_link_state == link_state) {
2374 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2375 		splx(s);
2376 		return;
2377 	}
2378 
2379 #ifdef DEBUG
2380 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2381 		link_state == LINK_STATE_UP ? "UP" :
2382 		link_state == LINK_STATE_DOWN ? "DOWN" :
2383 		"UNKNOWN",
2384 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2385 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2386 		"UNKNOWN");
2387 #endif
2388 
2389 	/*
2390 	 * When going from UNKNOWN to UP, we need to mark existing
2391 	 * addresses as tentative and restart DAD as we may have
2392 	 * erroneously not found a duplicate.
2393 	 *
2394 	 * This needs to happen before rt_ifmsg to avoid a race where
2395 	 * listeners would have an address and expect it to work right
2396 	 * away.
2397 	 */
2398 	notify = (link_state == LINK_STATE_UP &&
2399 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2400 	ifp->if_link_state = link_state;
2401 	/* The following routines may sleep so release the spin mutex */
2402 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2403 
2404 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2405 	if (notify) {
2406 		DOMAIN_FOREACH(dp) {
2407 			if (dp->dom_if_link_state_change != NULL)
2408 				dp->dom_if_link_state_change(ifp,
2409 				    LINK_STATE_DOWN);
2410 		}
2411 	}
2412 
2413 	/* Notify that the link state has changed. */
2414 	rt_ifmsg(ifp);
2415 
2416 #if NCARP > 0
2417 	if (ifp->if_carp)
2418 		carp_carpdev_state(ifp);
2419 #endif
2420 
2421 	if (ifp->if_link_state_changed != NULL)
2422 		ifp->if_link_state_changed(ifp, link_state);
2423 
2424 #if NBRIDGE > 0
2425 	if (ifp->if_bridge != NULL)
2426 		bridge_calc_link_state(ifp->if_bridge);
2427 #endif
2428 
2429 #if NLAGG > 0
2430 	if (ifp->if_lagg != NULL)
2431 		lagg_linkstate_changed(ifp);
2432 #endif
2433 
2434 	DOMAIN_FOREACH(dp) {
2435 		if (dp->dom_if_link_state_change != NULL)
2436 			dp->dom_if_link_state_change(ifp, link_state);
2437 	}
2438 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2439 	splx(s);
2440 }
2441 
2442 /*
2443  * Process the interface link state change queue.
2444  */
2445 static void
2446 if_link_state_change_work(struct work *work, void *arg)
2447 {
2448 	struct ifnet *ifp = container_of(work, struct ifnet, if_link_work);
2449 	int s;
2450 	uint8_t state;
2451 
2452 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2453 	s = splnet();
2454 
2455 	/* Pop a link state change from the queue and process it.
2456 	 * If there is nothing to process then if_detach() has been called.
2457 	 * We keep if_link_scheduled = true so the queue can safely drain
2458 	 * without more work being queued. */
2459 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2460 	LQ_POP(ifp->if_link_queue, state);
2461 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2462 	if (state == LINK_STATE_UNSET)
2463 		goto out;
2464 
2465 	if_link_state_change_process(ifp, state);
2466 
2467 	/* If there is a link state change to come, schedule it. */
2468 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2469 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET) {
2470 		ifp->if_link_scheduled = true;
2471 		workqueue_enqueue(ifnet_link_state_wq, &ifp->if_link_work, NULL);
2472 	} else
2473 		ifp->if_link_scheduled = false;
2474 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2475 
2476 out:
2477 	splx(s);
2478 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2479 }
2480 
2481 /*
2482  * Used to mark addresses on an interface as DETATCHED or TENTATIVE
2483  * and thus start Duplicate Address Detection without changing the
2484  * real link state.
2485  */
2486 void
2487 if_domain_link_state_change(struct ifnet *ifp, int link_state)
2488 {
2489 	struct domain *dp;
2490 	int s = splnet();
2491 
2492 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2493 
2494 	DOMAIN_FOREACH(dp) {
2495 		if (dp->dom_if_link_state_change != NULL)
2496 			dp->dom_if_link_state_change(ifp, link_state);
2497 	}
2498 
2499 	splx(s);
2500 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2501 }
2502 
2503 /*
2504  * Default action when installing a local route on a point-to-point
2505  * interface.
2506  */
2507 void
2508 p2p_rtrequest(int req, struct rtentry *rt,
2509     __unused const struct rt_addrinfo *info)
2510 {
2511 	struct ifnet *ifp = rt->rt_ifp;
2512 	struct ifaddr *ifa, *lo0ifa;
2513 	int s = pserialize_read_enter();
2514 
2515 	switch (req) {
2516 	case RTM_ADD:
2517 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2518 			break;
2519 
2520 		rt->rt_ifp = lo0ifp;
2521 
2522 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2523 			break;
2524 
2525 		IFADDR_READER_FOREACH(ifa, ifp) {
2526 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2527 				break;
2528 		}
2529 		if (ifa == NULL)
2530 			break;
2531 
2532 		/*
2533 		 * Ensure lo0 has an address of the same family.
2534 		 */
2535 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2536 			if (lo0ifa->ifa_addr->sa_family ==
2537 			    ifa->ifa_addr->sa_family)
2538 				break;
2539 		}
2540 		if (lo0ifa == NULL)
2541 			break;
2542 
2543 		/*
2544 		 * Make sure to set rt->rt_ifa to the interface
2545 		 * address we are using, otherwise we will have trouble
2546 		 * with source address selection.
2547 		 */
2548 		if (ifa != rt->rt_ifa)
2549 			rt_replace_ifa(rt, ifa);
2550 		break;
2551 	case RTM_DELETE:
2552 	default:
2553 		break;
2554 	}
2555 	pserialize_read_exit(s);
2556 }
2557 
2558 static void
2559 _if_down(struct ifnet *ifp)
2560 {
2561 	struct ifaddr *ifa;
2562 	struct domain *dp;
2563 	int s, bound;
2564 	struct psref psref;
2565 
2566 	ifp->if_flags &= ~IFF_UP;
2567 	nanotime(&ifp->if_lastchange);
2568 
2569 	bound = curlwp_bind();
2570 	s = pserialize_read_enter();
2571 	IFADDR_READER_FOREACH(ifa, ifp) {
2572 		ifa_acquire(ifa, &psref);
2573 		pserialize_read_exit(s);
2574 
2575 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2576 
2577 		s = pserialize_read_enter();
2578 		ifa_release(ifa, &psref);
2579 	}
2580 	pserialize_read_exit(s);
2581 	curlwp_bindx(bound);
2582 
2583 	IFQ_PURGE(&ifp->if_snd);
2584 #if NCARP > 0
2585 	if (ifp->if_carp)
2586 		carp_carpdev_state(ifp);
2587 #endif
2588 	rt_ifmsg(ifp);
2589 	DOMAIN_FOREACH(dp) {
2590 		if (dp->dom_if_down)
2591 			dp->dom_if_down(ifp);
2592 	}
2593 }
2594 
2595 static void
2596 if_down_deactivated(struct ifnet *ifp)
2597 {
2598 
2599 	KASSERT(if_is_deactivated(ifp));
2600 	_if_down(ifp);
2601 }
2602 
2603 void
2604 if_down_locked(struct ifnet *ifp)
2605 {
2606 
2607 	KASSERT(IFNET_LOCKED(ifp));
2608 	_if_down(ifp);
2609 }
2610 
2611 /*
2612  * Mark an interface down and notify protocols of
2613  * the transition.
2614  * NOTE: must be called at splsoftnet or equivalent.
2615  */
2616 void
2617 if_down(struct ifnet *ifp)
2618 {
2619 
2620 	IFNET_LOCK(ifp);
2621 	if_down_locked(ifp);
2622 	IFNET_UNLOCK(ifp);
2623 }
2624 
2625 /*
2626  * Must be called with holding if_ioctl_lock.
2627  */
2628 static void
2629 if_up_locked(struct ifnet *ifp)
2630 {
2631 #ifdef notyet
2632 	struct ifaddr *ifa;
2633 #endif
2634 	struct domain *dp;
2635 
2636 	KASSERT(IFNET_LOCKED(ifp));
2637 
2638 	KASSERT(!if_is_deactivated(ifp));
2639 	ifp->if_flags |= IFF_UP;
2640 	nanotime(&ifp->if_lastchange);
2641 #ifdef notyet
2642 	/* this has no effect on IP, and will kill all ISO connections XXX */
2643 	IFADDR_READER_FOREACH(ifa, ifp)
2644 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2645 #endif
2646 #if NCARP > 0
2647 	if (ifp->if_carp)
2648 		carp_carpdev_state(ifp);
2649 #endif
2650 	rt_ifmsg(ifp);
2651 	DOMAIN_FOREACH(dp) {
2652 		if (dp->dom_if_up)
2653 			dp->dom_if_up(ifp);
2654 	}
2655 }
2656 
2657 /*
2658  * Handle interface slowtimo timer routine.  Called
2659  * from softclock, we decrement timer (if set) and
2660  * call the appropriate interface routine on expiration.
2661  */
2662 static void
2663 if_slowtimo(void *arg)
2664 {
2665 	void (*slowtimo)(struct ifnet *);
2666 	struct ifnet *ifp = arg;
2667 	int s;
2668 
2669 	slowtimo = ifp->if_slowtimo;
2670 	if (__predict_false(slowtimo == NULL))
2671 		return;
2672 
2673 	s = splnet();
2674 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2675 		(*slowtimo)(ifp);
2676 
2677 	splx(s);
2678 
2679 	if (__predict_true(ifp->if_slowtimo != NULL))
2680 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2681 }
2682 
2683 /*
2684  * Mark an interface up and notify protocols of
2685  * the transition.
2686  * NOTE: must be called at splsoftnet or equivalent.
2687  */
2688 void
2689 if_up(struct ifnet *ifp)
2690 {
2691 
2692 	IFNET_LOCK(ifp);
2693 	if_up_locked(ifp);
2694 	IFNET_UNLOCK(ifp);
2695 }
2696 
2697 /*
2698  * Set/clear promiscuous mode on interface ifp based on the truth value
2699  * of pswitch.  The calls are reference counted so that only the first
2700  * "on" request actually has an effect, as does the final "off" request.
2701  * Results are undefined if the "off" and "on" requests are not matched.
2702  */
2703 int
2704 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2705 {
2706 	int pcount, ret = 0;
2707 	u_short nflags;
2708 
2709 	KASSERT(IFNET_LOCKED(ifp));
2710 
2711 	pcount = ifp->if_pcount;
2712 	if (pswitch) {
2713 		/*
2714 		 * Allow the device to be "placed" into promiscuous
2715 		 * mode even if it is not configured up.  It will
2716 		 * consult IFF_PROMISC when it is brought up.
2717 		 */
2718 		if (ifp->if_pcount++ != 0)
2719 			goto out;
2720 		nflags = ifp->if_flags | IFF_PROMISC;
2721 	} else {
2722 		if (--ifp->if_pcount > 0)
2723 			goto out;
2724 		nflags = ifp->if_flags & ~IFF_PROMISC;
2725 	}
2726 	ret = if_flags_set(ifp, nflags);
2727 	/* Restore interface state if not successful. */
2728 	if (ret != 0) {
2729 		ifp->if_pcount = pcount;
2730 	}
2731 out:
2732 	return ret;
2733 }
2734 
2735 int
2736 ifpromisc(struct ifnet *ifp, int pswitch)
2737 {
2738 	int e;
2739 
2740 	IFNET_LOCK(ifp);
2741 	e = ifpromisc_locked(ifp, pswitch);
2742 	IFNET_UNLOCK(ifp);
2743 
2744 	return e;
2745 }
2746 
2747 /*
2748  * Map interface name to
2749  * interface structure pointer.
2750  */
2751 struct ifnet *
2752 ifunit(const char *name)
2753 {
2754 	struct ifnet *ifp;
2755 	const char *cp = name;
2756 	u_int unit = 0;
2757 	u_int i;
2758 	int s;
2759 
2760 	/*
2761 	 * If the entire name is a number, treat it as an ifindex.
2762 	 */
2763 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2764 		unit = unit * 10 + (*cp - '0');
2765 	}
2766 
2767 	/*
2768 	 * If the number took all of the name, then it's a valid ifindex.
2769 	 */
2770 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2771 		return if_byindex(unit);
2772 
2773 	ifp = NULL;
2774 	s = pserialize_read_enter();
2775 	IFNET_READER_FOREACH(ifp) {
2776 		if (if_is_deactivated(ifp))
2777 			continue;
2778 	 	if (strcmp(ifp->if_xname, name) == 0)
2779 			goto out;
2780 	}
2781 out:
2782 	pserialize_read_exit(s);
2783 	return ifp;
2784 }
2785 
2786 /*
2787  * Get a reference of an ifnet object by an interface name.
2788  * The returned reference is protected by psref(9). The caller
2789  * must release a returned reference by if_put after use.
2790  */
2791 struct ifnet *
2792 if_get(const char *name, struct psref *psref)
2793 {
2794 	struct ifnet *ifp;
2795 	const char *cp = name;
2796 	u_int unit = 0;
2797 	u_int i;
2798 	int s;
2799 
2800 	/*
2801 	 * If the entire name is a number, treat it as an ifindex.
2802 	 */
2803 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2804 		unit = unit * 10 + (*cp - '0');
2805 	}
2806 
2807 	/*
2808 	 * If the number took all of the name, then it's a valid ifindex.
2809 	 */
2810 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2811 		return if_get_byindex(unit, psref);
2812 
2813 	ifp = NULL;
2814 	s = pserialize_read_enter();
2815 	IFNET_READER_FOREACH(ifp) {
2816 		if (if_is_deactivated(ifp))
2817 			continue;
2818 		if (strcmp(ifp->if_xname, name) == 0) {
2819 			PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2820 			psref_acquire(psref, &ifp->if_psref,
2821 			    ifnet_psref_class);
2822 			goto out;
2823 		}
2824 	}
2825 out:
2826 	pserialize_read_exit(s);
2827 	return ifp;
2828 }
2829 
2830 /*
2831  * Release a reference of an ifnet object given by if_get, if_get_byindex
2832  * or if_get_bylla.
2833  */
2834 void
2835 if_put(const struct ifnet *ifp, struct psref *psref)
2836 {
2837 
2838 	if (ifp == NULL)
2839 		return;
2840 
2841 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2842 }
2843 
2844 /*
2845  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2846  * should be used.
2847  */
2848 ifnet_t *
2849 _if_byindex(u_int idx)
2850 {
2851 
2852 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2853 }
2854 
2855 /*
2856  * Return ifp having idx. Return NULL if not found or the found ifp is
2857  * already deactivated.
2858  */
2859 ifnet_t *
2860 if_byindex(u_int idx)
2861 {
2862 	ifnet_t *ifp;
2863 
2864 	ifp = _if_byindex(idx);
2865 	if (ifp != NULL && if_is_deactivated(ifp))
2866 		ifp = NULL;
2867 	return ifp;
2868 }
2869 
2870 /*
2871  * Get a reference of an ifnet object by an interface index.
2872  * The returned reference is protected by psref(9). The caller
2873  * must release a returned reference by if_put after use.
2874  */
2875 ifnet_t *
2876 if_get_byindex(u_int idx, struct psref *psref)
2877 {
2878 	ifnet_t *ifp;
2879 	int s;
2880 
2881 	s = pserialize_read_enter();
2882 	ifp = if_byindex(idx);
2883 	if (__predict_true(ifp != NULL)) {
2884 		PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2885 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2886 	}
2887 	pserialize_read_exit(s);
2888 
2889 	return ifp;
2890 }
2891 
2892 ifnet_t *
2893 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2894 {
2895 	ifnet_t *ifp;
2896 	int s;
2897 
2898 	s = pserialize_read_enter();
2899 	IFNET_READER_FOREACH(ifp) {
2900 		if (if_is_deactivated(ifp))
2901 			continue;
2902 		if (ifp->if_addrlen != lla_len)
2903 			continue;
2904 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2905 			psref_acquire(psref, &ifp->if_psref,
2906 			    ifnet_psref_class);
2907 			break;
2908 		}
2909 	}
2910 	pserialize_read_exit(s);
2911 
2912 	return ifp;
2913 }
2914 
2915 /*
2916  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2917  * for example using pserialize or the ifp is already held or some other
2918  * object is held which guarantes the ifp to not be freed indirectly.
2919  */
2920 void
2921 if_acquire(struct ifnet *ifp, struct psref *psref)
2922 {
2923 
2924 	KASSERT(ifp->if_index != 0);
2925 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2926 }
2927 
2928 bool
2929 if_held(struct ifnet *ifp)
2930 {
2931 
2932 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2933 }
2934 
2935 /*
2936  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4.
2937  * Check the tunnel nesting count.
2938  * Return > 0, if tunnel nesting count is more than limit.
2939  * Return 0, if tunnel nesting count is equal or less than limit.
2940  */
2941 int
2942 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
2943 {
2944 	struct m_tag *mtag;
2945 	int *count;
2946 
2947 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO);
2948 	if (mtag != NULL) {
2949 		count = (int *)(mtag + 1);
2950 		if (++(*count) > limit) {
2951 			log(LOG_NOTICE,
2952 			    "%s: recursively called too many times(%d)\n",
2953 			    ifp->if_xname, *count);
2954 			return EIO;
2955 		}
2956 	} else {
2957 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
2958 		    M_NOWAIT);
2959 		if (mtag != NULL) {
2960 			m_tag_prepend(m, mtag);
2961 			count = (int *)(mtag + 1);
2962 			*count = 0;
2963 		} else {
2964 			log(LOG_DEBUG,
2965 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
2966 			    ifp->if_xname);
2967 		}
2968 	}
2969 
2970 	return 0;
2971 }
2972 
2973 static void
2974 if_tunnel_ro_init_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2975 {
2976 	struct tunnel_ro *tro = p;
2977 
2978 	tro->tr_ro = kmem_zalloc(sizeof(*tro->tr_ro), KM_SLEEP);
2979 	tro->tr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2980 }
2981 
2982 static void
2983 if_tunnel_ro_fini_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2984 {
2985 	struct tunnel_ro *tro = p;
2986 
2987 	rtcache_free(tro->tr_ro);
2988 	kmem_free(tro->tr_ro, sizeof(*tro->tr_ro));
2989 
2990 	mutex_obj_free(tro->tr_lock);
2991 }
2992 
2993 percpu_t *
2994 if_tunnel_alloc_ro_percpu(void)
2995 {
2996 
2997 	return percpu_create(sizeof(struct tunnel_ro),
2998 	    if_tunnel_ro_init_pc, if_tunnel_ro_fini_pc, NULL);
2999 }
3000 
3001 void
3002 if_tunnel_free_ro_percpu(percpu_t *ro_percpu)
3003 {
3004 
3005 	percpu_free(ro_percpu, sizeof(struct tunnel_ro));
3006 }
3007 
3008 
3009 static void
3010 if_tunnel_rtcache_free_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
3011 {
3012 	struct tunnel_ro *tro = p;
3013 
3014 	mutex_enter(tro->tr_lock);
3015 	rtcache_free(tro->tr_ro);
3016 	mutex_exit(tro->tr_lock);
3017 }
3018 
3019 void if_tunnel_ro_percpu_rtcache_free(percpu_t *ro_percpu)
3020 {
3021 
3022 	percpu_foreach(ro_percpu, if_tunnel_rtcache_free_pc, NULL);
3023 }
3024 
3025 void
3026 if_export_if_data(ifnet_t * const ifp, struct if_data *ifi, bool zero_stats)
3027 {
3028 
3029 	/* Collet the volatile stats first; this zeros *ifi. */
3030 	if_stats_to_if_data(ifp, ifi, zero_stats);
3031 
3032 	ifi->ifi_type = ifp->if_type;
3033 	ifi->ifi_addrlen = ifp->if_addrlen;
3034 	ifi->ifi_hdrlen = ifp->if_hdrlen;
3035 	ifi->ifi_link_state = ifp->if_link_state;
3036 	ifi->ifi_mtu = ifp->if_mtu;
3037 	ifi->ifi_metric = ifp->if_metric;
3038 	ifi->ifi_baudrate = ifp->if_baudrate;
3039 	ifi->ifi_lastchange = ifp->if_lastchange;
3040 }
3041 
3042 /* common */
3043 int
3044 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
3045 {
3046 	int s;
3047 	struct ifreq *ifr;
3048 	struct ifcapreq *ifcr;
3049 	struct ifdatareq *ifdr;
3050 	unsigned short flags;
3051 	char *descr;
3052 	int error;
3053 
3054 	switch (cmd) {
3055 	case SIOCSIFCAP:
3056 		ifcr = data;
3057 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
3058 			return EINVAL;
3059 
3060 		if (ifcr->ifcr_capenable == ifp->if_capenable)
3061 			return 0;
3062 
3063 		ifp->if_capenable = ifcr->ifcr_capenable;
3064 
3065 		/* Pre-compute the checksum flags mask. */
3066 		ifp->if_csum_flags_tx = 0;
3067 		ifp->if_csum_flags_rx = 0;
3068 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx)
3069 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
3070 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
3071 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
3072 
3073 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx)
3074 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
3075 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
3076 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
3077 
3078 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx)
3079 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
3080 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
3081 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
3082 
3083 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx)
3084 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
3085 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
3086 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
3087 
3088 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx)
3089 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
3090 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
3091 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
3092 
3093 		if (ifp->if_capenable & IFCAP_TSOv4)
3094 			ifp->if_csum_flags_tx |= M_CSUM_TSOv4;
3095 		if (ifp->if_capenable & IFCAP_TSOv6)
3096 			ifp->if_csum_flags_tx |= M_CSUM_TSOv6;
3097 
3098 #if NBRIDGE > 0
3099 		if (ifp->if_bridge != NULL)
3100 			bridge_calc_csum_flags(ifp->if_bridge);
3101 #endif
3102 
3103 		if (ifp->if_flags & IFF_UP)
3104 			return ENETRESET;
3105 		return 0;
3106 	case SIOCSIFFLAGS:
3107 		ifr = data;
3108 		/*
3109 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
3110 		 * and if_down aren't MP-safe yet, so we must hold the lock.
3111 		 */
3112 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3113 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
3114 			s = splsoftnet();
3115 			if_down_locked(ifp);
3116 			splx(s);
3117 		}
3118 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
3119 			s = splsoftnet();
3120 			if_up_locked(ifp);
3121 			splx(s);
3122 		}
3123 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3124 		flags = (ifp->if_flags & IFF_CANTCHANGE) |
3125 		    (ifr->ifr_flags &~ IFF_CANTCHANGE);
3126 		if (ifp->if_flags != flags) {
3127 			ifp->if_flags = flags;
3128 			/* Notify that the flags have changed. */
3129 			rt_ifmsg(ifp);
3130 		}
3131 		break;
3132 	case SIOCGIFFLAGS:
3133 		ifr = data;
3134 		ifr->ifr_flags = ifp->if_flags;
3135 		break;
3136 
3137 	case SIOCGIFMETRIC:
3138 		ifr = data;
3139 		ifr->ifr_metric = ifp->if_metric;
3140 		break;
3141 
3142 	case SIOCGIFMTU:
3143 		ifr = data;
3144 		ifr->ifr_mtu = ifp->if_mtu;
3145 		break;
3146 
3147 	case SIOCGIFDLT:
3148 		ifr = data;
3149 		ifr->ifr_dlt = ifp->if_dlt;
3150 		break;
3151 
3152 	case SIOCGIFCAP:
3153 		ifcr = data;
3154 		ifcr->ifcr_capabilities = ifp->if_capabilities;
3155 		ifcr->ifcr_capenable = ifp->if_capenable;
3156 		break;
3157 
3158 	case SIOCSIFMETRIC:
3159 		ifr = data;
3160 		ifp->if_metric = ifr->ifr_metric;
3161 		break;
3162 
3163 	case SIOCGIFDATA:
3164 		ifdr = data;
3165 		if_export_if_data(ifp, &ifdr->ifdr_data, false);
3166 		break;
3167 
3168 	case SIOCGIFINDEX:
3169 		ifr = data;
3170 		ifr->ifr_index = ifp->if_index;
3171 		break;
3172 
3173 	case SIOCZIFDATA:
3174 		ifdr = data;
3175 		if_export_if_data(ifp, &ifdr->ifdr_data, true);
3176 		getnanotime(&ifp->if_lastchange);
3177 		break;
3178 	case SIOCSIFMTU:
3179 		ifr = data;
3180 		if (ifp->if_mtu == ifr->ifr_mtu)
3181 			break;
3182 		ifp->if_mtu = ifr->ifr_mtu;
3183 		return ENETRESET;
3184 	case SIOCSIFDESCR:
3185 		error = kauth_authorize_network(curlwp->l_cred,
3186 		    KAUTH_NETWORK_INTERFACE,
3187 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, KAUTH_ARG(cmd),
3188 		    NULL);
3189 		if (error)
3190 			return error;
3191 
3192 		ifr = data;
3193 
3194 		if (ifr->ifr_buflen > IFDESCRSIZE)
3195 			return ENAMETOOLONG;
3196 
3197 		if (ifr->ifr_buf == NULL || ifr->ifr_buflen == 0) {
3198 			/* unset description */
3199 			descr = NULL;
3200 		} else {
3201 			descr = kmem_zalloc(IFDESCRSIZE, KM_SLEEP);
3202 			/*
3203 			 * copy (IFDESCRSIZE - 1) bytes to ensure
3204 			 * terminating nul
3205 			 */
3206 			error = copyin(ifr->ifr_buf, descr, IFDESCRSIZE - 1);
3207 			if (error) {
3208 				kmem_free(descr, IFDESCRSIZE);
3209 				return error;
3210 			}
3211 		}
3212 
3213 		if (ifp->if_description != NULL)
3214 			kmem_free(ifp->if_description, IFDESCRSIZE);
3215 
3216 		ifp->if_description = descr;
3217 		break;
3218 
3219  	case SIOCGIFDESCR:
3220 		ifr = data;
3221 		descr = ifp->if_description;
3222 
3223 		if (descr == NULL)
3224 			return ENOMSG;
3225 
3226 		if (ifr->ifr_buflen < IFDESCRSIZE)
3227 			return EINVAL;
3228 
3229 		error = copyout(descr, ifr->ifr_buf, IFDESCRSIZE);
3230 		if (error)
3231 			return error;
3232  		break;
3233 
3234 	default:
3235 		return ENOTTY;
3236 	}
3237 	return 0;
3238 }
3239 
3240 int
3241 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3242 {
3243 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3244 	struct ifaddr *ifa;
3245 	const struct sockaddr *any, *sa;
3246 	union {
3247 		struct sockaddr sa;
3248 		struct sockaddr_storage ss;
3249 	} u, v;
3250 	int s, error = 0;
3251 
3252 	switch (cmd) {
3253 	case SIOCSIFADDRPREF:
3254 		error = kauth_authorize_network(curlwp->l_cred,
3255 		    KAUTH_NETWORK_INTERFACE,
3256 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, KAUTH_ARG(cmd),
3257 		    NULL);
3258 		if (error)
3259 			return error;
3260 		break;
3261 	case SIOCGIFADDRPREF:
3262 		break;
3263 	default:
3264 		return EOPNOTSUPP;
3265 	}
3266 
3267 	/* sanity checks */
3268 	if (data == NULL || ifp == NULL) {
3269 		panic("invalid argument to %s", __func__);
3270 		/*NOTREACHED*/
3271 	}
3272 
3273 	/* address must be specified on ADD and DELETE */
3274 	sa = sstocsa(&ifap->ifap_addr);
3275 	if (sa->sa_family != sofamily(so))
3276 		return EINVAL;
3277 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3278 		return EINVAL;
3279 
3280 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3281 
3282 	s = pserialize_read_enter();
3283 	IFADDR_READER_FOREACH(ifa, ifp) {
3284 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3285 			continue;
3286 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3287 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3288 			break;
3289 	}
3290 	if (ifa == NULL) {
3291 		error = EADDRNOTAVAIL;
3292 		goto out;
3293 	}
3294 
3295 	switch (cmd) {
3296 	case SIOCSIFADDRPREF:
3297 		ifa->ifa_preference = ifap->ifap_preference;
3298 		goto out;
3299 	case SIOCGIFADDRPREF:
3300 		/* fill in the if_laddrreq structure */
3301 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3302 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3303 		ifap->ifap_preference = ifa->ifa_preference;
3304 		goto out;
3305 	default:
3306 		error = EOPNOTSUPP;
3307 	}
3308 out:
3309 	pserialize_read_exit(s);
3310 	return error;
3311 }
3312 
3313 /*
3314  * Interface ioctls.
3315  */
3316 static int
3317 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3318 {
3319 	struct ifnet *ifp;
3320 	struct ifreq *ifr;
3321 	int error = 0;
3322 	u_long ocmd = cmd;
3323 	u_short oif_flags;
3324 	struct ifreq ifrb;
3325 	struct oifreq *oifr = NULL;
3326 	int r;
3327 	struct psref psref;
3328 	int bound;
3329 	bool do_if43_post = false;
3330 	bool do_ifm80_post = false;
3331 
3332 	switch (cmd) {
3333 	case SIOCGIFCONF:
3334 		return ifconf(cmd, data);
3335 	case SIOCINITIFADDR:
3336 		return EPERM;
3337 	default:
3338 		MODULE_HOOK_CALL(uipc_syscalls_40_hook, (cmd, data), enosys(),
3339 		    error);
3340 		if (error != ENOSYS)
3341 			return error;
3342 		MODULE_HOOK_CALL(uipc_syscalls_50_hook, (l, cmd, data),
3343 		    enosys(), error);
3344 		if (error != ENOSYS)
3345 			return error;
3346 		error = 0;
3347 		break;
3348 	}
3349 
3350 	ifr = data;
3351 	/* Pre-conversion */
3352 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), error);
3353 	if (cmd != ocmd) {
3354 		oifr = data;
3355 		data = ifr = &ifrb;
3356 		IFREQO2N_43(oifr, ifr);
3357 		do_if43_post = true;
3358 	}
3359 	MODULE_HOOK_CALL(ifmedia_80_pre_hook, (ifr, &cmd, &do_ifm80_post),
3360 	    enosys(), error);
3361 
3362 	switch (cmd) {
3363 	case SIOCIFCREATE:
3364 	case SIOCIFDESTROY:
3365 		bound = curlwp_bind();
3366 		if (l != NULL) {
3367 			ifp = if_get(ifr->ifr_name, &psref);
3368 			error = kauth_authorize_network(l->l_cred,
3369 			    KAUTH_NETWORK_INTERFACE,
3370 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3371 			    KAUTH_ARG(cmd), NULL);
3372 			if (ifp != NULL)
3373 				if_put(ifp, &psref);
3374 			if (error != 0) {
3375 				curlwp_bindx(bound);
3376 				return error;
3377 			}
3378 		}
3379 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3380 		mutex_enter(&if_clone_mtx);
3381 		r = (cmd == SIOCIFCREATE) ?
3382 			if_clone_create(ifr->ifr_name) :
3383 			if_clone_destroy(ifr->ifr_name);
3384 		mutex_exit(&if_clone_mtx);
3385 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3386 		curlwp_bindx(bound);
3387 		return r;
3388 
3389 	case SIOCIFGCLONERS:
3390 		{
3391 			struct if_clonereq *req = (struct if_clonereq *)data;
3392 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3393 			    &req->ifcr_total);
3394 		}
3395 	}
3396 
3397 	bound = curlwp_bind();
3398 	ifp = if_get(ifr->ifr_name, &psref);
3399 	if (ifp == NULL) {
3400 		curlwp_bindx(bound);
3401 		return ENXIO;
3402 	}
3403 
3404 	switch (cmd) {
3405 	case SIOCALIFADDR:
3406 	case SIOCDLIFADDR:
3407 	case SIOCSIFADDRPREF:
3408 	case SIOCSIFFLAGS:
3409 	case SIOCSIFCAP:
3410 	case SIOCSIFMETRIC:
3411 	case SIOCZIFDATA:
3412 	case SIOCSIFMTU:
3413 	case SIOCSIFPHYADDR:
3414 	case SIOCDIFPHYADDR:
3415 #ifdef INET6
3416 	case SIOCSIFPHYADDR_IN6:
3417 #endif
3418 	case SIOCSLIFPHYADDR:
3419 	case SIOCADDMULTI:
3420 	case SIOCDELMULTI:
3421 	case SIOCSETHERCAP:
3422 	case SIOCSIFMEDIA:
3423 	case SIOCSDRVSPEC:
3424 	case SIOCG80211:
3425 	case SIOCS80211:
3426 	case SIOCS80211NWID:
3427 	case SIOCS80211NWKEY:
3428 	case SIOCS80211POWER:
3429 	case SIOCS80211BSSID:
3430 	case SIOCS80211CHANNEL:
3431 	case SIOCSLINKSTR:
3432 		if (l != NULL) {
3433 			error = kauth_authorize_network(l->l_cred,
3434 			    KAUTH_NETWORK_INTERFACE,
3435 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3436 			    KAUTH_ARG(cmd), NULL);
3437 			if (error != 0)
3438 				goto out;
3439 		}
3440 	}
3441 
3442 	oif_flags = ifp->if_flags;
3443 
3444 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3445 	IFNET_LOCK(ifp);
3446 
3447 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3448 	if (error != ENOTTY)
3449 		;
3450 	else if (so->so_proto == NULL)
3451 		error = EOPNOTSUPP;
3452 	else {
3453 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3454 		MODULE_HOOK_CALL(if_ifioctl_43_hook,
3455 			     (so, ocmd, cmd, data, l), enosys(), error);
3456 		if (error == ENOSYS)
3457 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3458 			    cmd, data, ifp);
3459 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3460 	}
3461 
3462 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3463 		if ((ifp->if_flags & IFF_UP) != 0) {
3464 			int s = splsoftnet();
3465 			if_up_locked(ifp);
3466 			splx(s);
3467 		}
3468 	}
3469 
3470 	/* Post-conversion */
3471 	if (do_ifm80_post && (error == 0))
3472 		MODULE_HOOK_CALL(ifmedia_80_post_hook, (ifr, cmd),
3473 		    enosys(), error);
3474 	if (do_if43_post)
3475 		IFREQN2O_43(oifr, ifr);
3476 
3477 	IFNET_UNLOCK(ifp);
3478 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3479 out:
3480 	if_put(ifp, &psref);
3481 	curlwp_bindx(bound);
3482 	return error;
3483 }
3484 
3485 /*
3486  * Return interface configuration
3487  * of system.  List may be used
3488  * in later ioctl's (above) to get
3489  * other information.
3490  *
3491  * Each record is a struct ifreq.  Before the addition of
3492  * sockaddr_storage, the API rule was that sockaddr flavors that did
3493  * not fit would extend beyond the struct ifreq, with the next struct
3494  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3495  * union in struct ifreq includes struct sockaddr_storage, every kind
3496  * of sockaddr must fit.  Thus, there are no longer any overlength
3497  * records.
3498  *
3499  * Records are added to the user buffer if they fit, and ifc_len is
3500  * adjusted to the length that was written.  Thus, the user is only
3501  * assured of getting the complete list if ifc_len on return is at
3502  * least sizeof(struct ifreq) less than it was on entry.
3503  *
3504  * If the user buffer pointer is NULL, this routine copies no data and
3505  * returns the amount of space that would be needed.
3506  *
3507  * Invariants:
3508  * ifrp points to the next part of the user's buffer to be used.  If
3509  * ifrp != NULL, space holds the number of bytes remaining that we may
3510  * write at ifrp.  Otherwise, space holds the number of bytes that
3511  * would have been written had there been adequate space.
3512  */
3513 /*ARGSUSED*/
3514 static int
3515 ifconf(u_long cmd, void *data)
3516 {
3517 	struct ifconf *ifc = (struct ifconf *)data;
3518 	struct ifnet *ifp;
3519 	struct ifaddr *ifa;
3520 	struct ifreq ifr, *ifrp = NULL;
3521 	int space = 0, error = 0;
3522 	const int sz = (int)sizeof(struct ifreq);
3523 	const bool docopy = ifc->ifc_req != NULL;
3524 	int s;
3525 	int bound;
3526 	struct psref psref;
3527 
3528 	if (docopy) {
3529 		if (ifc->ifc_len < 0)
3530 			return EINVAL;
3531 
3532 		space = ifc->ifc_len;
3533 		ifrp = ifc->ifc_req;
3534 	}
3535 	memset(&ifr, 0, sizeof(ifr));
3536 
3537 	bound = curlwp_bind();
3538 	s = pserialize_read_enter();
3539 	IFNET_READER_FOREACH(ifp) {
3540 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3541 		pserialize_read_exit(s);
3542 
3543 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3544 		    sizeof(ifr.ifr_name));
3545 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3546 			error = ENAMETOOLONG;
3547 			goto release_exit;
3548 		}
3549 		if (IFADDR_READER_EMPTY(ifp)) {
3550 			/* Interface with no addresses - send zero sockaddr. */
3551 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3552 			if (!docopy) {
3553 				space += sz;
3554 				goto next;
3555 			}
3556 			if (space >= sz) {
3557 				error = copyout(&ifr, ifrp, sz);
3558 				if (error != 0)
3559 					goto release_exit;
3560 				ifrp++;
3561 				space -= sz;
3562 			}
3563 		}
3564 
3565 		s = pserialize_read_enter();
3566 		IFADDR_READER_FOREACH(ifa, ifp) {
3567 			struct sockaddr *sa = ifa->ifa_addr;
3568 			/* all sockaddrs must fit in sockaddr_storage */
3569 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3570 
3571 			if (!docopy) {
3572 				space += sz;
3573 				continue;
3574 			}
3575 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3576 			pserialize_read_exit(s);
3577 
3578 			if (space >= sz) {
3579 				error = copyout(&ifr, ifrp, sz);
3580 				if (error != 0)
3581 					goto release_exit;
3582 				ifrp++; space -= sz;
3583 			}
3584 			s = pserialize_read_enter();
3585 		}
3586 		pserialize_read_exit(s);
3587 
3588         next:
3589 		s = pserialize_read_enter();
3590 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3591 	}
3592 	pserialize_read_exit(s);
3593 	curlwp_bindx(bound);
3594 
3595 	if (docopy) {
3596 		KASSERT(0 <= space && space <= ifc->ifc_len);
3597 		ifc->ifc_len -= space;
3598 	} else {
3599 		KASSERT(space >= 0);
3600 		ifc->ifc_len = space;
3601 	}
3602 	return (0);
3603 
3604 release_exit:
3605 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3606 	curlwp_bindx(bound);
3607 	return error;
3608 }
3609 
3610 int
3611 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3612 {
3613 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3614 	struct ifreq ifrb;
3615 	struct oifreq *oifr = NULL;
3616 	u_long ocmd = cmd;
3617 	int hook;
3618 
3619 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook);
3620 	if (hook != ENOSYS) {
3621 		if (cmd != ocmd) {
3622 			oifr = (struct oifreq *)(void *)ifr;
3623 			ifr = &ifrb;
3624 			IFREQO2N_43(oifr, ifr);
3625 				len = sizeof(oifr->ifr_addr);
3626 		}
3627 	}
3628 
3629 	if (len < sa->sa_len)
3630 		return EFBIG;
3631 
3632 	memset(&ifr->ifr_addr, 0, len);
3633 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3634 
3635 	if (cmd != ocmd)
3636 		IFREQN2O_43(oifr, ifr);
3637 	return 0;
3638 }
3639 
3640 /*
3641  * wrapper function for the drivers which doesn't have if_transmit().
3642  */
3643 static int
3644 if_transmit(struct ifnet *ifp, struct mbuf *m)
3645 {
3646 	int s, error;
3647 	size_t pktlen = m->m_pkthdr.len;
3648 	bool mcast = (m->m_flags & M_MCAST) != 0;
3649 
3650 	s = splnet();
3651 
3652 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3653 	if (error != 0) {
3654 		/* mbuf is already freed */
3655 		goto out;
3656 	}
3657 
3658 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
3659 	if_statadd_ref(nsr, if_obytes, pktlen);
3660 	if (mcast)
3661 		if_statinc_ref(nsr, if_omcasts);
3662 	IF_STAT_PUTREF(ifp);
3663 
3664 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3665 		if_start_lock(ifp);
3666 out:
3667 	splx(s);
3668 
3669 	return error;
3670 }
3671 
3672 int
3673 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3674 {
3675 	int error;
3676 
3677 	kmsan_check_mbuf(m);
3678 
3679 #ifdef ALTQ
3680 	KERNEL_LOCK(1, NULL);
3681 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3682 		error = if_transmit(ifp, m);
3683 		KERNEL_UNLOCK_ONE(NULL);
3684 	} else {
3685 		KERNEL_UNLOCK_ONE(NULL);
3686 		error = (*ifp->if_transmit)(ifp, m);
3687 		/* mbuf is alredy freed */
3688 	}
3689 #else /* !ALTQ */
3690 	error = (*ifp->if_transmit)(ifp, m);
3691 	/* mbuf is alredy freed */
3692 #endif /* !ALTQ */
3693 
3694 	return error;
3695 }
3696 
3697 /*
3698  * Queue message on interface, and start output if interface
3699  * not yet active.
3700  */
3701 int
3702 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3703 {
3704 
3705 	return if_transmit_lock(ifp, m);
3706 }
3707 
3708 /*
3709  * Queue message on interface, possibly using a second fast queue
3710  */
3711 int
3712 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3713 {
3714 	int error = 0;
3715 
3716 	if (ifq != NULL
3717 #ifdef ALTQ
3718 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3719 #endif
3720 	    ) {
3721 		if (IF_QFULL(ifq)) {
3722 			IF_DROP(&ifp->if_snd);
3723 			m_freem(m);
3724 			if (error == 0)
3725 				error = ENOBUFS;
3726 		} else
3727 			IF_ENQUEUE(ifq, m);
3728 	} else
3729 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3730 	if (error != 0) {
3731 		if_statinc(ifp, if_oerrors);
3732 		return error;
3733 	}
3734 	return 0;
3735 }
3736 
3737 int
3738 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3739 {
3740 	int rc;
3741 
3742 	KASSERT(IFNET_LOCKED(ifp));
3743 	if (ifp->if_initaddr != NULL)
3744 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3745 	else if (src ||
3746 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3747 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3748 
3749 	return rc;
3750 }
3751 
3752 int
3753 if_do_dad(struct ifnet *ifp)
3754 {
3755 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3756 		return 0;
3757 
3758 	switch (ifp->if_type) {
3759 	case IFT_FAITH:
3760 		/*
3761 		 * These interfaces do not have the IFF_LOOPBACK flag,
3762 		 * but loop packets back.  We do not have to do DAD on such
3763 		 * interfaces.  We should even omit it, because loop-backed
3764 		 * responses would confuse the DAD procedure.
3765 		 */
3766 		return 0;
3767 	default:
3768 		/*
3769 		 * Our DAD routine requires the interface up and running.
3770 		 * However, some interfaces can be up before the RUNNING
3771 		 * status.  Additionaly, users may try to assign addresses
3772 		 * before the interface becomes up (or running).
3773 		 * We simply skip DAD in such a case as a work around.
3774 		 * XXX: we should rather mark "tentative" on such addresses,
3775 		 * and do DAD after the interface becomes ready.
3776 		 */
3777 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
3778 		    (IFF_UP | IFF_RUNNING))
3779 			return 0;
3780 
3781 		return 1;
3782 	}
3783 }
3784 
3785 int
3786 if_flags_set(ifnet_t *ifp, const u_short flags)
3787 {
3788 	int rc;
3789 
3790 	KASSERT(IFNET_LOCKED(ifp));
3791 
3792 	if (ifp->if_setflags != NULL)
3793 		rc = (*ifp->if_setflags)(ifp, flags);
3794 	else {
3795 		u_short cantflags, chgdflags;
3796 		struct ifreq ifr;
3797 
3798 		chgdflags = ifp->if_flags ^ flags;
3799 		cantflags = chgdflags & IFF_CANTCHANGE;
3800 
3801 		if (cantflags != 0)
3802 			ifp->if_flags ^= cantflags;
3803 
3804                 /* Traditionally, we do not call if_ioctl after
3805                  * setting/clearing only IFF_PROMISC if the interface
3806                  * isn't IFF_UP.  Uphold that tradition.
3807 		 */
3808 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3809 			return 0;
3810 
3811 		memset(&ifr, 0, sizeof(ifr));
3812 
3813 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3814 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3815 
3816 		if (rc != 0 && cantflags != 0)
3817 			ifp->if_flags ^= cantflags;
3818 	}
3819 
3820 	return rc;
3821 }
3822 
3823 int
3824 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3825 {
3826 	int rc;
3827 	struct ifreq ifr;
3828 
3829 	/*
3830 	 * XXX NOMPSAFE - this calls if_ioctl without holding IFNET_LOCK()
3831 	 * in some cases - e.g. when called from vlan/netinet/netinet6 code
3832 	 * directly rather than via doifoictl()
3833 	 */
3834 	ifreq_setaddr(cmd, &ifr, sa);
3835 	rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3836 
3837 	return rc;
3838 }
3839 
3840 static void
3841 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3842     struct ifaltq *ifq)
3843 {
3844 	const struct sysctlnode *cnode, *rnode;
3845 
3846 	if (sysctl_createv(clog, 0, NULL, &rnode,
3847 		       CTLFLAG_PERMANENT,
3848 		       CTLTYPE_NODE, "interfaces",
3849 		       SYSCTL_DESCR("Per-interface controls"),
3850 		       NULL, 0, NULL, 0,
3851 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3852 		goto bad;
3853 
3854 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3855 		       CTLFLAG_PERMANENT,
3856 		       CTLTYPE_NODE, ifname,
3857 		       SYSCTL_DESCR("Interface controls"),
3858 		       NULL, 0, NULL, 0,
3859 		       CTL_CREATE, CTL_EOL) != 0)
3860 		goto bad;
3861 
3862 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3863 		       CTLFLAG_PERMANENT,
3864 		       CTLTYPE_NODE, "sndq",
3865 		       SYSCTL_DESCR("Interface output queue controls"),
3866 		       NULL, 0, NULL, 0,
3867 		       CTL_CREATE, CTL_EOL) != 0)
3868 		goto bad;
3869 
3870 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3871 		       CTLFLAG_PERMANENT,
3872 		       CTLTYPE_INT, "len",
3873 		       SYSCTL_DESCR("Current output queue length"),
3874 		       NULL, 0, &ifq->ifq_len, 0,
3875 		       CTL_CREATE, CTL_EOL) != 0)
3876 		goto bad;
3877 
3878 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3879 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3880 		       CTLTYPE_INT, "maxlen",
3881 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3882 		       NULL, 0, &ifq->ifq_maxlen, 0,
3883 		       CTL_CREATE, CTL_EOL) != 0)
3884 		goto bad;
3885 
3886 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3887 		       CTLFLAG_PERMANENT,
3888 		       CTLTYPE_INT, "drops",
3889 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3890 		       NULL, 0, &ifq->ifq_drops, 0,
3891 		       CTL_CREATE, CTL_EOL) != 0)
3892 		goto bad;
3893 
3894 	return;
3895 bad:
3896 	printf("%s: could not attach sysctl nodes\n", ifname);
3897 	return;
3898 }
3899 
3900 #if defined(INET) || defined(INET6)
3901 
3902 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3903 	static int							\
3904 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3905 	{								\
3906 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3907 	}
3908 
3909 #if defined(INET)
3910 static int
3911 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3912 {
3913 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3914 }
3915 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3916 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3917 #endif
3918 
3919 #if defined(INET6)
3920 static int
3921 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3922 {
3923 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3924 }
3925 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3926 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3927 #endif
3928 
3929 static void
3930 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3931 {
3932 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3933 	const char *pfname = NULL, *ipname = NULL;
3934 	int ipn = 0, qid = 0;
3935 
3936 	switch (pf) {
3937 #if defined(INET)
3938 	case PF_INET:
3939 		len_func = sysctl_net_ip_pktq_items;
3940 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3941 		drops_func = sysctl_net_ip_pktq_drops;
3942 		pfname = "inet", ipn = IPPROTO_IP;
3943 		ipname = "ip", qid = IPCTL_IFQ;
3944 		break;
3945 #endif
3946 #if defined(INET6)
3947 	case PF_INET6:
3948 		len_func = sysctl_net_ip6_pktq_items;
3949 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3950 		drops_func = sysctl_net_ip6_pktq_drops;
3951 		pfname = "inet6", ipn = IPPROTO_IPV6;
3952 		ipname = "ip6", qid = IPV6CTL_IFQ;
3953 		break;
3954 #endif
3955 	default:
3956 		KASSERT(false);
3957 	}
3958 
3959 	sysctl_createv(clog, 0, NULL, NULL,
3960 		       CTLFLAG_PERMANENT,
3961 		       CTLTYPE_NODE, pfname, NULL,
3962 		       NULL, 0, NULL, 0,
3963 		       CTL_NET, pf, CTL_EOL);
3964 	sysctl_createv(clog, 0, NULL, NULL,
3965 		       CTLFLAG_PERMANENT,
3966 		       CTLTYPE_NODE, ipname, NULL,
3967 		       NULL, 0, NULL, 0,
3968 		       CTL_NET, pf, ipn, CTL_EOL);
3969 	sysctl_createv(clog, 0, NULL, NULL,
3970 		       CTLFLAG_PERMANENT,
3971 		       CTLTYPE_NODE, "ifq",
3972 		       SYSCTL_DESCR("Protocol input queue controls"),
3973 		       NULL, 0, NULL, 0,
3974 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3975 
3976 	sysctl_createv(clog, 0, NULL, NULL,
3977 		       CTLFLAG_PERMANENT,
3978 		       CTLTYPE_QUAD, "len",
3979 		       SYSCTL_DESCR("Current input queue length"),
3980 		       len_func, 0, NULL, 0,
3981 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3982 	sysctl_createv(clog, 0, NULL, NULL,
3983 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3984 		       CTLTYPE_INT, "maxlen",
3985 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3986 		       maxlen_func, 0, NULL, 0,
3987 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3988 	sysctl_createv(clog, 0, NULL, NULL,
3989 		       CTLFLAG_PERMANENT,
3990 		       CTLTYPE_QUAD, "drops",
3991 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3992 		       drops_func, 0, NULL, 0,
3993 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3994 }
3995 #endif /* INET || INET6 */
3996 
3997 static int
3998 if_sdl_sysctl(SYSCTLFN_ARGS)
3999 {
4000 	struct ifnet *ifp;
4001 	const struct sockaddr_dl *sdl;
4002 	struct psref psref;
4003 	int error = 0;
4004 	int bound;
4005 
4006 	if (namelen != 1)
4007 		return EINVAL;
4008 
4009 	bound = curlwp_bind();
4010 	ifp = if_get_byindex(name[0], &psref);
4011 	if (ifp == NULL) {
4012 		error = ENODEV;
4013 		goto out0;
4014 	}
4015 
4016 	sdl = ifp->if_sadl;
4017 	if (sdl == NULL) {
4018 		*oldlenp = 0;
4019 		goto out1;
4020 	}
4021 
4022 	if (oldp == NULL) {
4023 		*oldlenp = sdl->sdl_alen;
4024 		goto out1;
4025 	}
4026 
4027 	if (*oldlenp >= sdl->sdl_alen)
4028 		*oldlenp = sdl->sdl_alen;
4029 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
4030 out1:
4031 	if_put(ifp, &psref);
4032 out0:
4033 	curlwp_bindx(bound);
4034 	return error;
4035 }
4036 
4037 static void
4038 if_sysctl_setup(struct sysctllog **clog)
4039 {
4040 	const struct sysctlnode *rnode = NULL;
4041 
4042 	sysctl_createv(clog, 0, NULL, &rnode,
4043 		       CTLFLAG_PERMANENT,
4044 		       CTLTYPE_NODE, "sdl",
4045 		       SYSCTL_DESCR("Get active link-layer address"),
4046 		       if_sdl_sysctl, 0, NULL, 0,
4047 		       CTL_NET, CTL_CREATE, CTL_EOL);
4048 
4049 #if defined(INET)
4050 	sysctl_net_pktq_setup(NULL, PF_INET);
4051 #endif
4052 #ifdef INET6
4053 	if (in6_present)
4054 		sysctl_net_pktq_setup(NULL, PF_INET6);
4055 #endif
4056 }
4057