xref: /netbsd-src/sys/net/if.c (revision e670fd5c413e99c2f6a37901bb21c537fcd322d2)
1 /*	$NetBSD: if.c,v 1.487 2021/07/01 22:08:13 blymn 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 1.487 2021/07/01 22:08:13 blymn 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 void
706 if_initialize(ifnet_t *ifp)
707 {
708 
709 	KASSERT(if_indexlim > 0);
710 	TAILQ_INIT(&ifp->if_addrlist);
711 
712 	/*
713 	 * Link level name is allocated later by a separate call to
714 	 * if_alloc_sadl().
715 	 */
716 
717 	if (ifp->if_snd.ifq_maxlen == 0)
718 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
719 
720 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
721 
722 	ifp->if_link_state = LINK_STATE_UNKNOWN;
723 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
724 	ifp->if_link_scheduled = false;
725 
726 	ifp->if_capenable = 0;
727 	ifp->if_csum_flags_tx = 0;
728 	ifp->if_csum_flags_rx = 0;
729 
730 #ifdef ALTQ
731 	ifp->if_snd.altq_type = 0;
732 	ifp->if_snd.altq_disc = NULL;
733 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
734 	ifp->if_snd.altq_tbr  = NULL;
735 	ifp->if_snd.altq_ifp  = ifp;
736 #endif
737 
738 	IFQ_LOCK_INIT(&ifp->if_snd);
739 
740 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
741 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
742 
743 	IF_AFDATA_LOCK_INIT(ifp);
744 
745 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
746 	PSLIST_INIT(&ifp->if_addr_pslist);
747 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
748 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
749 	LIST_INIT(&ifp->if_multiaddrs);
750 	if_stats_init(ifp);
751 
752 	IFNET_GLOBAL_LOCK();
753 	if_getindex(ifp);
754 	IFNET_GLOBAL_UNLOCK();
755 }
756 
757 /*
758  * Register an interface to the list of "active" interfaces.
759  */
760 void
761 if_register(ifnet_t *ifp)
762 {
763 	/*
764 	 * If the driver has not supplied its own if_ioctl or if_stop,
765 	 * then supply the default.
766 	 */
767 	if (ifp->if_ioctl == NULL)
768 		ifp->if_ioctl = ifioctl_common;
769 	if (ifp->if_stop == NULL)
770 		ifp->if_stop = if_nullstop;
771 
772 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
773 
774 	if (!STAILQ_EMPTY(&domains))
775 		if_attachdomain1(ifp);
776 
777 	/* Announce the interface. */
778 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
779 
780 	if (ifp->if_slowtimo != NULL) {
781 		ifp->if_slowtimo_ch =
782 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
783 		callout_init(ifp->if_slowtimo_ch, 0);
784 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
785 		if_slowtimo(ifp);
786 	}
787 
788 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
789 		ifp->if_transmit = if_transmit;
790 
791 	IFNET_GLOBAL_LOCK();
792 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
793 	IFNET_WRITER_INSERT_TAIL(ifp);
794 	IFNET_GLOBAL_UNLOCK();
795 }
796 
797 /*
798  * The if_percpuq framework
799  *
800  * It allows network device drivers to execute the network stack
801  * in softint (so called softint-based if_input). It utilizes
802  * softint and percpu ifqueue. It doesn't distribute any packets
803  * between CPUs, unlike pktqueue(9).
804  *
805  * Currently we support two options for device drivers to apply the framework:
806  * - Use it implicitly with less changes
807  *   - If you use if_attach in driver's _attach function and if_input in
808  *     driver's Rx interrupt handler, a packet is queued and a softint handles
809  *     the packet implicitly
810  * - Use it explicitly in each driver (recommended)
811  *   - You can use if_percpuq_* directly in your driver
812  *   - In this case, you need to allocate struct if_percpuq in driver's softc
813  *   - See wm(4) as a reference implementation
814  */
815 
816 static void
817 if_percpuq_softint(void *arg)
818 {
819 	struct if_percpuq *ipq = arg;
820 	struct ifnet *ifp = ipq->ipq_ifp;
821 	struct mbuf *m;
822 
823 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
824 		if_statinc(ifp, if_ipackets);
825 		bpf_mtap(ifp, m, BPF_D_IN);
826 
827 		ifp->_if_input(ifp, m);
828 	}
829 }
830 
831 static void
832 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
833 {
834 	struct ifqueue *const ifq = p;
835 
836 	memset(ifq, 0, sizeof(*ifq));
837 	ifq->ifq_maxlen = IFQ_MAXLEN;
838 }
839 
840 struct if_percpuq *
841 if_percpuq_create(struct ifnet *ifp)
842 {
843 	struct if_percpuq *ipq;
844 	u_int flags = SOFTINT_NET;
845 
846 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
847 
848 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
849 	ipq->ipq_ifp = ifp;
850 	ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq);
851 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
852 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
853 
854 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
855 
856 	return ipq;
857 }
858 
859 static struct mbuf *
860 if_percpuq_dequeue(struct if_percpuq *ipq)
861 {
862 	struct mbuf *m;
863 	struct ifqueue *ifq;
864 	int s;
865 
866 	s = splnet();
867 	ifq = percpu_getref(ipq->ipq_ifqs);
868 	IF_DEQUEUE(ifq, m);
869 	percpu_putref(ipq->ipq_ifqs);
870 	splx(s);
871 
872 	return m;
873 }
874 
875 static void
876 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
877 {
878 	struct ifqueue *const ifq = p;
879 
880 	IF_PURGE(ifq);
881 }
882 
883 void
884 if_percpuq_destroy(struct if_percpuq *ipq)
885 {
886 
887 	/* if_detach may already destroy it */
888 	if (ipq == NULL)
889 		return;
890 
891 	softint_disestablish(ipq->ipq_si);
892 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
893 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
894 	kmem_free(ipq, sizeof(*ipq));
895 }
896 
897 void
898 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
899 {
900 	struct ifqueue *ifq;
901 	int s;
902 
903 	KASSERT(ipq != NULL);
904 
905 	s = splnet();
906 	ifq = percpu_getref(ipq->ipq_ifqs);
907 	if (IF_QFULL(ifq)) {
908 		IF_DROP(ifq);
909 		percpu_putref(ipq->ipq_ifqs);
910 		m_freem(m);
911 		goto out;
912 	}
913 	IF_ENQUEUE(ifq, m);
914 	percpu_putref(ipq->ipq_ifqs);
915 
916 	softint_schedule(ipq->ipq_si);
917 out:
918 	splx(s);
919 }
920 
921 static void
922 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
923 {
924 	struct ifqueue *const ifq = p;
925 	int *sum = arg;
926 
927 	*sum += ifq->ifq_drops;
928 }
929 
930 static int
931 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
932 {
933 	struct sysctlnode node;
934 	struct if_percpuq *ipq;
935 	int sum = 0;
936 	int error;
937 
938 	node = *rnode;
939 	ipq = node.sysctl_data;
940 
941 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
942 
943 	node.sysctl_data = &sum;
944 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
945 	if (error != 0 || newp == NULL)
946 		return error;
947 
948 	return 0;
949 }
950 
951 static void
952 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
953     struct if_percpuq *ipq)
954 {
955 	const struct sysctlnode *cnode, *rnode;
956 
957 	if (sysctl_createv(clog, 0, NULL, &rnode,
958 		       CTLFLAG_PERMANENT,
959 		       CTLTYPE_NODE, "interfaces",
960 		       SYSCTL_DESCR("Per-interface controls"),
961 		       NULL, 0, NULL, 0,
962 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
963 		goto bad;
964 
965 	if (sysctl_createv(clog, 0, &rnode, &rnode,
966 		       CTLFLAG_PERMANENT,
967 		       CTLTYPE_NODE, ifname,
968 		       SYSCTL_DESCR("Interface controls"),
969 		       NULL, 0, NULL, 0,
970 		       CTL_CREATE, CTL_EOL) != 0)
971 		goto bad;
972 
973 	if (sysctl_createv(clog, 0, &rnode, &rnode,
974 		       CTLFLAG_PERMANENT,
975 		       CTLTYPE_NODE, "rcvq",
976 		       SYSCTL_DESCR("Interface input queue controls"),
977 		       NULL, 0, NULL, 0,
978 		       CTL_CREATE, CTL_EOL) != 0)
979 		goto bad;
980 
981 #ifdef NOTYET
982 	/* XXX Should show each per-CPU queue length? */
983 	if (sysctl_createv(clog, 0, &rnode, &rnode,
984 		       CTLFLAG_PERMANENT,
985 		       CTLTYPE_INT, "len",
986 		       SYSCTL_DESCR("Current input queue length"),
987 		       sysctl_percpuq_len, 0, NULL, 0,
988 		       CTL_CREATE, CTL_EOL) != 0)
989 		goto bad;
990 
991 	if (sysctl_createv(clog, 0, &rnode, &cnode,
992 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
993 		       CTLTYPE_INT, "maxlen",
994 		       SYSCTL_DESCR("Maximum allowed input queue length"),
995 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
996 		       CTL_CREATE, CTL_EOL) != 0)
997 		goto bad;
998 #endif
999 
1000 	if (sysctl_createv(clog, 0, &rnode, &cnode,
1001 		       CTLFLAG_PERMANENT,
1002 		       CTLTYPE_INT, "drops",
1003 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
1004 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
1005 		       CTL_CREATE, CTL_EOL) != 0)
1006 		goto bad;
1007 
1008 	return;
1009 bad:
1010 	printf("%s: could not attach sysctl nodes\n", ifname);
1011 	return;
1012 }
1013 
1014 /*
1015  * The deferred if_start framework
1016  *
1017  * The common APIs to defer if_start to softint when if_start is requested
1018  * from a device driver running in hardware interrupt context.
1019  */
1020 /*
1021  * Call ifp->if_start (or equivalent) in a dedicated softint for
1022  * deferred if_start.
1023  */
1024 static void
1025 if_deferred_start_softint(void *arg)
1026 {
1027 	struct if_deferred_start *ids = arg;
1028 	struct ifnet *ifp = ids->ids_ifp;
1029 
1030 	ids->ids_if_start(ifp);
1031 }
1032 
1033 /*
1034  * The default callback function for deferred if_start.
1035  */
1036 static void
1037 if_deferred_start_common(struct ifnet *ifp)
1038 {
1039 	int s;
1040 
1041 	s = splnet();
1042 	if_start_lock(ifp);
1043 	splx(s);
1044 }
1045 
1046 static inline bool
1047 if_snd_is_used(struct ifnet *ifp)
1048 {
1049 
1050 	return ALTQ_IS_ENABLED(&ifp->if_snd) ||
1051 		ifp->if_transmit == if_transmit ||
1052 		ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit;
1053 }
1054 
1055 /*
1056  * Schedule deferred if_start.
1057  */
1058 void
1059 if_schedule_deferred_start(struct ifnet *ifp)
1060 {
1061 
1062 	KASSERT(ifp->if_deferred_start != NULL);
1063 
1064 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1065 		return;
1066 
1067 	softint_schedule(ifp->if_deferred_start->ids_si);
1068 }
1069 
1070 /*
1071  * Create an instance of deferred if_start. A driver should call the function
1072  * only if the driver needs deferred if_start. Drivers can setup their own
1073  * deferred if_start function via 2nd argument.
1074  */
1075 void
1076 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1077 {
1078 	struct if_deferred_start *ids;
1079 	u_int flags = SOFTINT_NET;
1080 
1081 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
1082 
1083 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1084 	ids->ids_ifp = ifp;
1085 	ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids);
1086 	if (func != NULL)
1087 		ids->ids_if_start = func;
1088 	else
1089 		ids->ids_if_start = if_deferred_start_common;
1090 
1091 	ifp->if_deferred_start = ids;
1092 }
1093 
1094 static void
1095 if_deferred_start_destroy(struct ifnet *ifp)
1096 {
1097 
1098 	if (ifp->if_deferred_start == NULL)
1099 		return;
1100 
1101 	softint_disestablish(ifp->if_deferred_start->ids_si);
1102 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1103 	ifp->if_deferred_start = NULL;
1104 }
1105 
1106 /*
1107  * The common interface input routine that is called by device drivers,
1108  * which should be used only when the driver's rx handler already runs
1109  * in softint.
1110  */
1111 void
1112 if_input(struct ifnet *ifp, struct mbuf *m)
1113 {
1114 
1115 	KASSERT(ifp->if_percpuq == NULL);
1116 	KASSERT(!cpu_intr_p());
1117 
1118 	if_statinc(ifp, if_ipackets);
1119 	bpf_mtap(ifp, m, BPF_D_IN);
1120 
1121 	ifp->_if_input(ifp, m);
1122 }
1123 
1124 /*
1125  * DEPRECATED. Use if_initialize and if_register instead.
1126  * See the above comment of if_initialize.
1127  *
1128  * Note that it implicitly enables if_percpuq to make drivers easy to
1129  * migrate softint-based if_input without much changes. If you don't
1130  * want to enable it, use if_initialize instead.
1131  */
1132 void
1133 if_attach(ifnet_t *ifp)
1134 {
1135 
1136 	if_initialize(ifp);
1137 	ifp->if_percpuq = if_percpuq_create(ifp);
1138 	if_register(ifp);
1139 }
1140 
1141 void
1142 if_attachdomain(void)
1143 {
1144 	struct ifnet *ifp;
1145 	int s;
1146 	int bound = curlwp_bind();
1147 
1148 	s = pserialize_read_enter();
1149 	IFNET_READER_FOREACH(ifp) {
1150 		struct psref psref;
1151 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1152 		pserialize_read_exit(s);
1153 		if_attachdomain1(ifp);
1154 		s = pserialize_read_enter();
1155 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1156 	}
1157 	pserialize_read_exit(s);
1158 	curlwp_bindx(bound);
1159 }
1160 
1161 static void
1162 if_attachdomain1(struct ifnet *ifp)
1163 {
1164 	struct domain *dp;
1165 	int s;
1166 
1167 	s = splsoftnet();
1168 
1169 	/* address family dependent data region */
1170 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1171 	DOMAIN_FOREACH(dp) {
1172 		if (dp->dom_ifattach != NULL)
1173 			ifp->if_afdata[dp->dom_family] =
1174 			    (*dp->dom_ifattach)(ifp);
1175 	}
1176 
1177 	splx(s);
1178 }
1179 
1180 /*
1181  * Deactivate an interface.  This points all of the procedure
1182  * handles at error stubs.  May be called from interrupt context.
1183  */
1184 void
1185 if_deactivate(struct ifnet *ifp)
1186 {
1187 	int s;
1188 
1189 	s = splsoftnet();
1190 
1191 	ifp->if_output	 = if_nulloutput;
1192 	ifp->_if_input	 = if_nullinput;
1193 	ifp->if_start	 = if_nullstart;
1194 	ifp->if_transmit = if_nulltransmit;
1195 	ifp->if_ioctl	 = if_nullioctl;
1196 	ifp->if_init	 = if_nullinit;
1197 	ifp->if_stop	 = if_nullstop;
1198 	ifp->if_slowtimo = if_nullslowtimo;
1199 	ifp->if_drain	 = if_nulldrain;
1200 
1201 	ifp->if_link_state_changed = NULL;
1202 
1203 	/* No more packets may be enqueued. */
1204 	ifp->if_snd.ifq_maxlen = 0;
1205 
1206 	splx(s);
1207 }
1208 
1209 bool
1210 if_is_deactivated(const struct ifnet *ifp)
1211 {
1212 
1213 	return ifp->if_output == if_nulloutput;
1214 }
1215 
1216 void
1217 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1218 {
1219 	struct ifaddr *ifa, *nifa;
1220 	int s;
1221 
1222 	s = pserialize_read_enter();
1223 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1224 		nifa = IFADDR_READER_NEXT(ifa);
1225 		if (ifa->ifa_addr->sa_family != family)
1226 			continue;
1227 		pserialize_read_exit(s);
1228 
1229 		(*purgeaddr)(ifa);
1230 
1231 		s = pserialize_read_enter();
1232 	}
1233 	pserialize_read_exit(s);
1234 }
1235 
1236 #ifdef IFAREF_DEBUG
1237 static struct ifaddr **ifa_list;
1238 static int ifa_list_size;
1239 
1240 /* Depends on only one if_attach runs at once */
1241 static void
1242 if_build_ifa_list(struct ifnet *ifp)
1243 {
1244 	struct ifaddr *ifa;
1245 	int i;
1246 
1247 	KASSERT(ifa_list == NULL);
1248 	KASSERT(ifa_list_size == 0);
1249 
1250 	IFADDR_READER_FOREACH(ifa, ifp)
1251 		ifa_list_size++;
1252 
1253 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1254 	i = 0;
1255 	IFADDR_READER_FOREACH(ifa, ifp) {
1256 		ifa_list[i++] = ifa;
1257 		ifaref(ifa);
1258 	}
1259 }
1260 
1261 static void
1262 if_check_and_free_ifa_list(struct ifnet *ifp)
1263 {
1264 	int i;
1265 	struct ifaddr *ifa;
1266 
1267 	if (ifa_list == NULL)
1268 		return;
1269 
1270 	for (i = 0; i < ifa_list_size; i++) {
1271 		char buf[64];
1272 
1273 		ifa = ifa_list[i];
1274 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1275 		if (ifa->ifa_refcnt > 1) {
1276 			log(LOG_WARNING,
1277 			    "ifa(%s) still referenced (refcnt=%d)\n",
1278 			    buf, ifa->ifa_refcnt - 1);
1279 		} else
1280 			log(LOG_DEBUG,
1281 			    "ifa(%s) not referenced (refcnt=%d)\n",
1282 			    buf, ifa->ifa_refcnt - 1);
1283 		ifafree(ifa);
1284 	}
1285 
1286 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1287 	ifa_list = NULL;
1288 	ifa_list_size = 0;
1289 }
1290 #endif
1291 
1292 /*
1293  * Detach an interface from the list of "active" interfaces,
1294  * freeing any resources as we go along.
1295  *
1296  * NOTE: This routine must be called with a valid thread context,
1297  * as it may block.
1298  */
1299 void
1300 if_detach(struct ifnet *ifp)
1301 {
1302 	struct socket so;
1303 	struct ifaddr *ifa;
1304 #ifdef IFAREF_DEBUG
1305 	struct ifaddr *last_ifa = NULL;
1306 #endif
1307 	struct domain *dp;
1308 	const struct protosw *pr;
1309 	int s, i, family, purged;
1310 
1311 #ifdef IFAREF_DEBUG
1312 	if_build_ifa_list(ifp);
1313 #endif
1314 	/*
1315 	 * XXX It's kind of lame that we have to have the
1316 	 * XXX socket structure...
1317 	 */
1318 	memset(&so, 0, sizeof(so));
1319 
1320 	s = splnet();
1321 
1322 	sysctl_teardown(&ifp->if_sysctl_log);
1323 
1324 	IFNET_LOCK(ifp);
1325 
1326 	/*
1327 	 * Unset all queued link states and pretend a
1328 	 * link state change is scheduled.
1329 	 * This stops any more link state changes occuring for this
1330 	 * interface while it's being detached so it's safe
1331 	 * to drain the workqueue.
1332 	 */
1333 	IF_LINK_STATE_CHANGE_LOCK(ifp);
1334 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
1335 	ifp->if_link_scheduled = true;
1336 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
1337 	workqueue_wait(ifnet_link_state_wq, &ifp->if_link_work);
1338 
1339 	if_deactivate(ifp);
1340 	IFNET_UNLOCK(ifp);
1341 
1342 	/*
1343 	 * Unlink from the list and wait for all readers to leave
1344 	 * from pserialize read sections.  Note that we can't do
1345 	 * psref_target_destroy here.  See below.
1346 	 */
1347 	IFNET_GLOBAL_LOCK();
1348 	ifindex2ifnet[ifp->if_index] = NULL;
1349 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1350 	IFNET_WRITER_REMOVE(ifp);
1351 	pserialize_perform(ifnet_psz);
1352 	IFNET_GLOBAL_UNLOCK();
1353 
1354 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1355 		ifp->if_slowtimo = NULL;
1356 		callout_halt(ifp->if_slowtimo_ch, NULL);
1357 		callout_destroy(ifp->if_slowtimo_ch);
1358 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1359 	}
1360 	if_deferred_start_destroy(ifp);
1361 
1362 	/*
1363 	 * Do an if_down() to give protocols a chance to do something.
1364 	 */
1365 	if_down_deactivated(ifp);
1366 
1367 #ifdef ALTQ
1368 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1369 		altq_disable(&ifp->if_snd);
1370 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1371 		altq_detach(&ifp->if_snd);
1372 #endif
1373 
1374 #if NCARP > 0
1375 	/* Remove the interface from any carp group it is a part of.  */
1376 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1377 		carp_ifdetach(ifp);
1378 #endif
1379 
1380 	/*
1381 	 * Rip all the addresses off the interface.  This should make
1382 	 * all of the routes go away.
1383 	 *
1384 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1385 	 * from the list, including our "cursor", ifa.  For safety,
1386 	 * and to honor the TAILQ abstraction, I just restart the
1387 	 * loop after each removal.  Note that the loop will exit
1388 	 * when all of the remaining ifaddrs belong to the AF_LINK
1389 	 * family.  I am counting on the historical fact that at
1390 	 * least one pr_usrreq in each address domain removes at
1391 	 * least one ifaddr.
1392 	 */
1393 again:
1394 	/*
1395 	 * At this point, no other one tries to remove ifa in the list,
1396 	 * so we don't need to take a lock or psref.  Avoid using
1397 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1398 	 * violations of pserialize.
1399 	 */
1400 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1401 		family = ifa->ifa_addr->sa_family;
1402 #ifdef IFAREF_DEBUG
1403 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1404 		    ifa, family, ifa->ifa_refcnt);
1405 		if (last_ifa != NULL && ifa == last_ifa)
1406 			panic("if_detach: loop detected");
1407 		last_ifa = ifa;
1408 #endif
1409 		if (family == AF_LINK)
1410 			continue;
1411 		dp = pffinddomain(family);
1412 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1413 		/*
1414 		 * XXX These PURGEIF calls are redundant with the
1415 		 * purge-all-families calls below, but are left in for
1416 		 * now both to make a smaller change, and to avoid
1417 		 * unplanned interactions with clearing of
1418 		 * ifp->if_addrlist.
1419 		 */
1420 		purged = 0;
1421 		for (pr = dp->dom_protosw;
1422 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1423 			so.so_proto = pr;
1424 			if (pr->pr_usrreqs) {
1425 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1426 				purged = 1;
1427 			}
1428 		}
1429 		if (purged == 0) {
1430 			/*
1431 			 * XXX What's really the best thing to do
1432 			 * XXX here?  --thorpej@NetBSD.org
1433 			 */
1434 			printf("if_detach: WARNING: AF %d not purged\n",
1435 			    family);
1436 			ifa_remove(ifp, ifa);
1437 		}
1438 		goto again;
1439 	}
1440 
1441 	if_free_sadl(ifp, 1);
1442 
1443 restart:
1444 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1445 		family = ifa->ifa_addr->sa_family;
1446 		KASSERT(family == AF_LINK);
1447 		ifa_remove(ifp, ifa);
1448 		goto restart;
1449 	}
1450 
1451 	/* Delete stray routes from the routing table. */
1452 	for (i = 0; i <= AF_MAX; i++)
1453 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1454 
1455 	DOMAIN_FOREACH(dp) {
1456 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1457 		{
1458 			void *p = ifp->if_afdata[dp->dom_family];
1459 			if (p) {
1460 				ifp->if_afdata[dp->dom_family] = NULL;
1461 				(*dp->dom_ifdetach)(ifp, p);
1462 			}
1463 		}
1464 
1465 		/*
1466 		 * One would expect multicast memberships (INET and
1467 		 * INET6) on UDP sockets to be purged by the PURGEIF
1468 		 * calls above, but if all addresses were removed from
1469 		 * the interface prior to destruction, the calls will
1470 		 * not be made (e.g. ppp, for which pppd(8) generally
1471 		 * removes addresses before destroying the interface).
1472 		 * Because there is no invariant that multicast
1473 		 * memberships only exist for interfaces with IPv4
1474 		 * addresses, we must call PURGEIF regardless of
1475 		 * addresses.  (Protocols which might store ifnet
1476 		 * pointers are marked with PR_PURGEIF.)
1477 		 */
1478 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1479 			so.so_proto = pr;
1480 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1481 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1482 		}
1483 	}
1484 
1485 	/*
1486 	 * Must be done after the above pr_purgeif because if_psref may be
1487 	 * still used in pr_purgeif.
1488 	 */
1489 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1490 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1491 
1492 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1493 	(void)pfil_head_destroy(ifp->if_pfil);
1494 
1495 	/* Announce that the interface is gone. */
1496 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1497 
1498 	IF_AFDATA_LOCK_DESTROY(ifp);
1499 
1500 	/*
1501 	 * remove packets that came from ifp, from software interrupt queues.
1502 	 */
1503 	DOMAIN_FOREACH(dp) {
1504 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1505 			struct ifqueue *iq = dp->dom_ifqueues[i];
1506 			if (iq == NULL)
1507 				break;
1508 			dp->dom_ifqueues[i] = NULL;
1509 			if_detach_queues(ifp, iq);
1510 		}
1511 	}
1512 
1513 	/*
1514 	 * IP queues have to be processed separately: net-queue barrier
1515 	 * ensures that the packets are dequeued while a cross-call will
1516 	 * ensure that the interrupts have completed. FIXME: not quite..
1517 	 */
1518 #ifdef INET
1519 	pktq_barrier(ip_pktq);
1520 #endif
1521 #ifdef INET6
1522 	if (in6_present)
1523 		pktq_barrier(ip6_pktq);
1524 #endif
1525 	xc_barrier(0);
1526 
1527 	if (ifp->if_percpuq != NULL) {
1528 		if_percpuq_destroy(ifp->if_percpuq);
1529 		ifp->if_percpuq = NULL;
1530 	}
1531 
1532 	mutex_obj_free(ifp->if_ioctl_lock);
1533 	ifp->if_ioctl_lock = NULL;
1534 	mutex_obj_free(ifp->if_snd.ifq_lock);
1535 	if_stats_fini(ifp);
1536 
1537 	splx(s);
1538 
1539 #ifdef IFAREF_DEBUG
1540 	if_check_and_free_ifa_list(ifp);
1541 #endif
1542 }
1543 
1544 static void
1545 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1546 {
1547 	struct mbuf *m, *prev, *next;
1548 
1549 	prev = NULL;
1550 	for (m = q->ifq_head; m != NULL; m = next) {
1551 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1552 
1553 		next = m->m_nextpkt;
1554 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1555 			prev = m;
1556 			continue;
1557 		}
1558 
1559 		if (prev != NULL)
1560 			prev->m_nextpkt = m->m_nextpkt;
1561 		else
1562 			q->ifq_head = m->m_nextpkt;
1563 		if (q->ifq_tail == m)
1564 			q->ifq_tail = prev;
1565 		q->ifq_len--;
1566 
1567 		m->m_nextpkt = NULL;
1568 		m_freem(m);
1569 		IF_DROP(q);
1570 	}
1571 }
1572 
1573 /*
1574  * Callback for a radix tree walk to delete all references to an
1575  * ifnet.
1576  */
1577 static int
1578 if_delroute_matcher(struct rtentry *rt, void *v)
1579 {
1580 	struct ifnet *ifp = (struct ifnet *)v;
1581 
1582 	if (rt->rt_ifp == ifp)
1583 		return 1;
1584 	else
1585 		return 0;
1586 }
1587 
1588 /*
1589  * Create a clone network interface.
1590  */
1591 static int
1592 if_clone_create(const char *name)
1593 {
1594 	struct if_clone *ifc;
1595 	int unit;
1596 	struct ifnet *ifp;
1597 	struct psref psref;
1598 
1599 	KASSERT(mutex_owned(&if_clone_mtx));
1600 
1601 	ifc = if_clone_lookup(name, &unit);
1602 	if (ifc == NULL)
1603 		return EINVAL;
1604 
1605 	ifp = if_get(name, &psref);
1606 	if (ifp != NULL) {
1607 		if_put(ifp, &psref);
1608 		return EEXIST;
1609 	}
1610 
1611 	return (*ifc->ifc_create)(ifc, unit);
1612 }
1613 
1614 /*
1615  * Destroy a clone network interface.
1616  */
1617 static int
1618 if_clone_destroy(const char *name)
1619 {
1620 	struct if_clone *ifc;
1621 	struct ifnet *ifp;
1622 	struct psref psref;
1623 	int error;
1624 	int (*if_ioctl)(struct ifnet *, u_long, void *);
1625 
1626 	KASSERT(mutex_owned(&if_clone_mtx));
1627 
1628 	ifc = if_clone_lookup(name, NULL);
1629 	if (ifc == NULL)
1630 		return EINVAL;
1631 
1632 	if (ifc->ifc_destroy == NULL)
1633 		return EOPNOTSUPP;
1634 
1635 	ifp = if_get(name, &psref);
1636 	if (ifp == NULL)
1637 		return ENXIO;
1638 
1639 	/* We have to disable ioctls here */
1640 	IFNET_LOCK(ifp);
1641 	if_ioctl = ifp->if_ioctl;
1642 	ifp->if_ioctl = if_nullioctl;
1643 	IFNET_UNLOCK(ifp);
1644 
1645 	/*
1646 	 * We cannot call ifc_destroy with holding ifp.
1647 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1648 	 */
1649 	if_put(ifp, &psref);
1650 
1651 	error = (*ifc->ifc_destroy)(ifp);
1652 
1653 	if (error != 0) {
1654 		/* We have to restore if_ioctl on error */
1655 		IFNET_LOCK(ifp);
1656 		ifp->if_ioctl = if_ioctl;
1657 		IFNET_UNLOCK(ifp);
1658 	}
1659 
1660 	return error;
1661 }
1662 
1663 static bool
1664 if_is_unit(const char *name)
1665 {
1666 
1667 	while (*name != '\0') {
1668 		if (*name < '0' || *name > '9')
1669 			return false;
1670 		name++;
1671 	}
1672 
1673 	return true;
1674 }
1675 
1676 /*
1677  * Look up a network interface cloner.
1678  */
1679 static struct if_clone *
1680 if_clone_lookup(const char *name, int *unitp)
1681 {
1682 	struct if_clone *ifc;
1683 	const char *cp;
1684 	char *dp, ifname[IFNAMSIZ + 3];
1685 	int unit;
1686 
1687 	KASSERT(mutex_owned(&if_clone_mtx));
1688 
1689 	strcpy(ifname, "if_");
1690 	/* separate interface name from unit */
1691 	/* TODO: search unit number from backward */
1692 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1693 	    *cp && !if_is_unit(cp);)
1694 		*dp++ = *cp++;
1695 
1696 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1697 		return NULL;	/* No name or unit number */
1698 	*dp++ = '\0';
1699 
1700 again:
1701 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1702 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1703 			break;
1704 	}
1705 
1706 	if (ifc == NULL) {
1707 		int error;
1708 		if (*ifname == '\0')
1709 			return NULL;
1710 		mutex_exit(&if_clone_mtx);
1711 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1712 		mutex_enter(&if_clone_mtx);
1713 		if (error)
1714 			return NULL;
1715 		*ifname = '\0';
1716 		goto again;
1717 	}
1718 
1719 	unit = 0;
1720 	while (cp - name < IFNAMSIZ && *cp) {
1721 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1722 			/* Bogus unit number. */
1723 			return NULL;
1724 		}
1725 		unit = (unit * 10) + (*cp++ - '0');
1726 	}
1727 
1728 	if (unitp != NULL)
1729 		*unitp = unit;
1730 	return ifc;
1731 }
1732 
1733 /*
1734  * Register a network interface cloner.
1735  */
1736 void
1737 if_clone_attach(struct if_clone *ifc)
1738 {
1739 
1740 	mutex_enter(&if_clone_mtx);
1741 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1742 	if_cloners_count++;
1743 	mutex_exit(&if_clone_mtx);
1744 }
1745 
1746 /*
1747  * Unregister a network interface cloner.
1748  */
1749 void
1750 if_clone_detach(struct if_clone *ifc)
1751 {
1752 
1753 	mutex_enter(&if_clone_mtx);
1754 	LIST_REMOVE(ifc, ifc_list);
1755 	if_cloners_count--;
1756 	mutex_exit(&if_clone_mtx);
1757 }
1758 
1759 /*
1760  * Provide list of interface cloners to userspace.
1761  */
1762 int
1763 if_clone_list(int buf_count, char *buffer, int *total)
1764 {
1765 	char outbuf[IFNAMSIZ], *dst;
1766 	struct if_clone *ifc;
1767 	int count, error = 0;
1768 
1769 	mutex_enter(&if_clone_mtx);
1770 	*total = if_cloners_count;
1771 	if ((dst = buffer) == NULL) {
1772 		/* Just asking how many there are. */
1773 		goto out;
1774 	}
1775 
1776 	if (buf_count < 0) {
1777 		error = EINVAL;
1778 		goto out;
1779 	}
1780 
1781 	count = (if_cloners_count < buf_count) ?
1782 	    if_cloners_count : buf_count;
1783 
1784 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1785 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1786 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1787 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1788 			error = ENAMETOOLONG;
1789 			goto out;
1790 		}
1791 		error = copyout(outbuf, dst, sizeof(outbuf));
1792 		if (error != 0)
1793 			break;
1794 	}
1795 
1796 out:
1797 	mutex_exit(&if_clone_mtx);
1798 	return error;
1799 }
1800 
1801 void
1802 ifa_psref_init(struct ifaddr *ifa)
1803 {
1804 
1805 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1806 }
1807 
1808 void
1809 ifaref(struct ifaddr *ifa)
1810 {
1811 
1812 	atomic_inc_uint(&ifa->ifa_refcnt);
1813 }
1814 
1815 void
1816 ifafree(struct ifaddr *ifa)
1817 {
1818 	KASSERT(ifa != NULL);
1819 	KASSERTMSG(ifa->ifa_refcnt > 0, "ifa_refcnt=%d", ifa->ifa_refcnt);
1820 
1821 	if (atomic_dec_uint_nv(&ifa->ifa_refcnt) == 0) {
1822 		free(ifa, M_IFADDR);
1823 	}
1824 }
1825 
1826 bool
1827 ifa_is_destroying(struct ifaddr *ifa)
1828 {
1829 
1830 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1831 }
1832 
1833 void
1834 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1835 {
1836 
1837 	ifa->ifa_ifp = ifp;
1838 
1839 	/*
1840 	 * Check MP-safety for IFEF_MPSAFE drivers.
1841 	 * Check !IFF_RUNNING for initialization routines that normally don't
1842 	 * take IFNET_LOCK but it's safe because there is no competitor.
1843 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1844 	 * if_stop.
1845 	 */
1846 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1847 	    IFNET_LOCKED(ifp));
1848 
1849 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1850 	IFADDR_ENTRY_INIT(ifa);
1851 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1852 
1853 	ifaref(ifa);
1854 }
1855 
1856 void
1857 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1858 {
1859 
1860 	KASSERT(ifa->ifa_ifp == ifp);
1861 	/*
1862 	 * Check MP-safety for IFEF_MPSAFE drivers.
1863 	 * if_is_deactivated indicates ifa_remove is called form if_detach
1864 	 * where is safe even if IFNET_LOCK isn't held.
1865 	 */
1866 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp));
1867 
1868 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1869 	IFADDR_WRITER_REMOVE(ifa);
1870 #ifdef NET_MPSAFE
1871 	IFNET_GLOBAL_LOCK();
1872 	pserialize_perform(ifnet_psz);
1873 	IFNET_GLOBAL_UNLOCK();
1874 #endif
1875 
1876 #ifdef NET_MPSAFE
1877 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1878 #endif
1879 	IFADDR_ENTRY_DESTROY(ifa);
1880 	ifafree(ifa);
1881 }
1882 
1883 void
1884 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1885 {
1886 
1887 	PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
1888 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1889 }
1890 
1891 void
1892 ifa_release(struct ifaddr *ifa, struct psref *psref)
1893 {
1894 
1895 	if (ifa == NULL)
1896 		return;
1897 
1898 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1899 }
1900 
1901 bool
1902 ifa_held(struct ifaddr *ifa)
1903 {
1904 
1905 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1906 }
1907 
1908 static inline int
1909 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1910 {
1911 	return sockaddr_cmp(sa1, sa2) == 0;
1912 }
1913 
1914 /*
1915  * Locate an interface based on a complete address.
1916  */
1917 /*ARGSUSED*/
1918 struct ifaddr *
1919 ifa_ifwithaddr(const struct sockaddr *addr)
1920 {
1921 	struct ifnet *ifp;
1922 	struct ifaddr *ifa;
1923 
1924 	IFNET_READER_FOREACH(ifp) {
1925 		if (if_is_deactivated(ifp))
1926 			continue;
1927 		IFADDR_READER_FOREACH(ifa, ifp) {
1928 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1929 				continue;
1930 			if (equal(addr, ifa->ifa_addr))
1931 				return ifa;
1932 			if ((ifp->if_flags & IFF_BROADCAST) &&
1933 			    ifa->ifa_broadaddr &&
1934 			    /* IP6 doesn't have broadcast */
1935 			    ifa->ifa_broadaddr->sa_len != 0 &&
1936 			    equal(ifa->ifa_broadaddr, addr))
1937 				return ifa;
1938 		}
1939 	}
1940 	return NULL;
1941 }
1942 
1943 struct ifaddr *
1944 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1945 {
1946 	struct ifaddr *ifa;
1947 	int s = pserialize_read_enter();
1948 
1949 	ifa = ifa_ifwithaddr(addr);
1950 	if (ifa != NULL)
1951 		ifa_acquire(ifa, psref);
1952 	pserialize_read_exit(s);
1953 
1954 	return ifa;
1955 }
1956 
1957 /*
1958  * Locate the point to point interface with a given destination address.
1959  */
1960 /*ARGSUSED*/
1961 struct ifaddr *
1962 ifa_ifwithdstaddr(const struct sockaddr *addr)
1963 {
1964 	struct ifnet *ifp;
1965 	struct ifaddr *ifa;
1966 
1967 	IFNET_READER_FOREACH(ifp) {
1968 		if (if_is_deactivated(ifp))
1969 			continue;
1970 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1971 			continue;
1972 		IFADDR_READER_FOREACH(ifa, ifp) {
1973 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1974 			    ifa->ifa_dstaddr == NULL)
1975 				continue;
1976 			if (equal(addr, ifa->ifa_dstaddr))
1977 				return ifa;
1978 		}
1979 	}
1980 
1981 	return NULL;
1982 }
1983 
1984 struct ifaddr *
1985 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1986 {
1987 	struct ifaddr *ifa;
1988 	int s;
1989 
1990 	s = pserialize_read_enter();
1991 	ifa = ifa_ifwithdstaddr(addr);
1992 	if (ifa != NULL)
1993 		ifa_acquire(ifa, psref);
1994 	pserialize_read_exit(s);
1995 
1996 	return ifa;
1997 }
1998 
1999 /*
2000  * Find an interface on a specific network.  If many, choice
2001  * is most specific found.
2002  */
2003 struct ifaddr *
2004 ifa_ifwithnet(const struct sockaddr *addr)
2005 {
2006 	struct ifnet *ifp;
2007 	struct ifaddr *ifa, *ifa_maybe = NULL;
2008 	const struct sockaddr_dl *sdl;
2009 	u_int af = addr->sa_family;
2010 	const char *addr_data = addr->sa_data, *cplim;
2011 
2012 	if (af == AF_LINK) {
2013 		sdl = satocsdl(addr);
2014 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
2015 		    ifindex2ifnet[sdl->sdl_index] &&
2016 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
2017 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
2018 		}
2019 	}
2020 #ifdef NETATALK
2021 	if (af == AF_APPLETALK) {
2022 		const struct sockaddr_at *sat, *sat2;
2023 		sat = (const struct sockaddr_at *)addr;
2024 		IFNET_READER_FOREACH(ifp) {
2025 			if (if_is_deactivated(ifp))
2026 				continue;
2027 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
2028 			if (ifa == NULL)
2029 				continue;
2030 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
2031 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
2032 				return ifa; /* exact match */
2033 			if (ifa_maybe == NULL) {
2034 				/* else keep the if with the right range */
2035 				ifa_maybe = ifa;
2036 			}
2037 		}
2038 		return ifa_maybe;
2039 	}
2040 #endif
2041 	IFNET_READER_FOREACH(ifp) {
2042 		if (if_is_deactivated(ifp))
2043 			continue;
2044 		IFADDR_READER_FOREACH(ifa, ifp) {
2045 			const char *cp, *cp2, *cp3;
2046 
2047 			if (ifa->ifa_addr->sa_family != af ||
2048 			    ifa->ifa_netmask == NULL)
2049  next:				continue;
2050 			cp = addr_data;
2051 			cp2 = ifa->ifa_addr->sa_data;
2052 			cp3 = ifa->ifa_netmask->sa_data;
2053 			cplim = (const char *)ifa->ifa_netmask +
2054 			    ifa->ifa_netmask->sa_len;
2055 			while (cp3 < cplim) {
2056 				if ((*cp++ ^ *cp2++) & *cp3++) {
2057 					/* want to continue for() loop */
2058 					goto next;
2059 				}
2060 			}
2061 			if (ifa_maybe == NULL ||
2062 			    rt_refines(ifa->ifa_netmask,
2063 			               ifa_maybe->ifa_netmask))
2064 				ifa_maybe = ifa;
2065 		}
2066 	}
2067 	return ifa_maybe;
2068 }
2069 
2070 struct ifaddr *
2071 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2072 {
2073 	struct ifaddr *ifa;
2074 	int s;
2075 
2076 	s = pserialize_read_enter();
2077 	ifa = ifa_ifwithnet(addr);
2078 	if (ifa != NULL)
2079 		ifa_acquire(ifa, psref);
2080 	pserialize_read_exit(s);
2081 
2082 	return ifa;
2083 }
2084 
2085 /*
2086  * Find the interface of the addresss.
2087  */
2088 struct ifaddr *
2089 ifa_ifwithladdr(const struct sockaddr *addr)
2090 {
2091 	struct ifaddr *ia;
2092 
2093 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2094 	    (ia = ifa_ifwithnet(addr)))
2095 		return ia;
2096 	return NULL;
2097 }
2098 
2099 struct ifaddr *
2100 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2101 {
2102 	struct ifaddr *ifa;
2103 	int s;
2104 
2105 	s = pserialize_read_enter();
2106 	ifa = ifa_ifwithladdr(addr);
2107 	if (ifa != NULL)
2108 		ifa_acquire(ifa, psref);
2109 	pserialize_read_exit(s);
2110 
2111 	return ifa;
2112 }
2113 
2114 /*
2115  * Find an interface using a specific address family
2116  */
2117 struct ifaddr *
2118 ifa_ifwithaf(int af)
2119 {
2120 	struct ifnet *ifp;
2121 	struct ifaddr *ifa = NULL;
2122 	int s;
2123 
2124 	s = pserialize_read_enter();
2125 	IFNET_READER_FOREACH(ifp) {
2126 		if (if_is_deactivated(ifp))
2127 			continue;
2128 		IFADDR_READER_FOREACH(ifa, ifp) {
2129 			if (ifa->ifa_addr->sa_family == af)
2130 				goto out;
2131 		}
2132 	}
2133 out:
2134 	pserialize_read_exit(s);
2135 	return ifa;
2136 }
2137 
2138 /*
2139  * Find an interface address specific to an interface best matching
2140  * a given address.
2141  */
2142 struct ifaddr *
2143 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2144 {
2145 	struct ifaddr *ifa;
2146 	const char *cp, *cp2, *cp3;
2147 	const char *cplim;
2148 	struct ifaddr *ifa_maybe = 0;
2149 	u_int af = addr->sa_family;
2150 
2151 	if (if_is_deactivated(ifp))
2152 		return NULL;
2153 
2154 	if (af >= AF_MAX)
2155 		return NULL;
2156 
2157 	IFADDR_READER_FOREACH(ifa, ifp) {
2158 		if (ifa->ifa_addr->sa_family != af)
2159 			continue;
2160 		ifa_maybe = ifa;
2161 		if (ifa->ifa_netmask == NULL) {
2162 			if (equal(addr, ifa->ifa_addr) ||
2163 			    (ifa->ifa_dstaddr &&
2164 			     equal(addr, ifa->ifa_dstaddr)))
2165 				return ifa;
2166 			continue;
2167 		}
2168 		cp = addr->sa_data;
2169 		cp2 = ifa->ifa_addr->sa_data;
2170 		cp3 = ifa->ifa_netmask->sa_data;
2171 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2172 		for (; cp3 < cplim; cp3++) {
2173 			if ((*cp++ ^ *cp2++) & *cp3)
2174 				break;
2175 		}
2176 		if (cp3 == cplim)
2177 			return ifa;
2178 	}
2179 	return ifa_maybe;
2180 }
2181 
2182 struct ifaddr *
2183 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2184     struct psref *psref)
2185 {
2186 	struct ifaddr *ifa;
2187 	int s;
2188 
2189 	s = pserialize_read_enter();
2190 	ifa = ifaof_ifpforaddr(addr, ifp);
2191 	if (ifa != NULL)
2192 		ifa_acquire(ifa, psref);
2193 	pserialize_read_exit(s);
2194 
2195 	return ifa;
2196 }
2197 
2198 /*
2199  * Default action when installing a route with a Link Level gateway.
2200  * Lookup an appropriate real ifa to point to.
2201  * This should be moved to /sys/net/link.c eventually.
2202  */
2203 void
2204 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2205 {
2206 	struct ifaddr *ifa;
2207 	const struct sockaddr *dst;
2208 	struct ifnet *ifp;
2209 	struct psref psref;
2210 
2211 	if (cmd != RTM_ADD || ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2212 		return;
2213 	ifp = rt->rt_ifa->ifa_ifp;
2214 	dst = rt_getkey(rt);
2215 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2216 		rt_replace_ifa(rt, ifa);
2217 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2218 			ifa->ifa_rtrequest(cmd, rt, info);
2219 		ifa_release(ifa, &psref);
2220 	}
2221 }
2222 
2223 /*
2224  * bitmask macros to manage a densely packed link_state change queue.
2225  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2226  * LINK_STATE_UP(2) we need 2 bits for each state change.
2227  * As a state change to store is 0, treat all bits set as an unset item.
2228  */
2229 #define LQ_ITEM_BITS		2
2230 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2231 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2232 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2233 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2234 #define LQ_STORE(q, i, v)						      \
2235 	do {								      \
2236 		(q) &= ~LQ_MASK((i));					      \
2237 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2238 	} while (0 /* CONSTCOND */)
2239 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2240 #define LQ_POP(q, v)							      \
2241 	do {								      \
2242 		(v) = LQ_ITEM((q), 0);					      \
2243 		(q) >>= LQ_ITEM_BITS;					      \
2244 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2245 	} while (0 /* CONSTCOND */)
2246 #define LQ_PUSH(q, v)							      \
2247 	do {								      \
2248 		(q) >>= LQ_ITEM_BITS;					      \
2249 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2250 	} while (0 /* CONSTCOND */)
2251 #define LQ_FIND_UNSET(q, i)						      \
2252 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2253 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2254 			break;						      \
2255 	}
2256 
2257 /*
2258  * Handle a change in the interface link state and
2259  * queue notifications.
2260  */
2261 void
2262 if_link_state_change(struct ifnet *ifp, int link_state)
2263 {
2264 	int idx;
2265 
2266 	/* Ensure change is to a valid state */
2267 	switch (link_state) {
2268 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2269 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2270 	case LINK_STATE_UP:
2271 		break;
2272 	default:
2273 #ifdef DEBUG
2274 		printf("%s: invalid link state %d\n",
2275 		    ifp->if_xname, link_state);
2276 #endif
2277 		return;
2278 	}
2279 
2280 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2281 
2282 	/* Find the last unset event in the queue. */
2283 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2284 
2285 	if (idx == 0) {
2286 		/*
2287 		 * There is no queue of link state changes.
2288 		 * As we have the lock we can safely compare against the
2289 		 * current link state and return if the same.
2290 		 * Otherwise, if scheduled is true then the interface is being
2291 		 * detached and the queue is being drained so we need
2292 		 * to avoid queuing more work.
2293 		 */
2294 		 if (ifp->if_link_state == link_state || ifp->if_link_scheduled)
2295 			goto out;
2296 	} else {
2297 		/* Ensure link_state doesn't match the last queued state. */
2298 		if (LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2299 			goto out;
2300 	}
2301 
2302 	/* Handle queue overflow. */
2303 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2304 		uint8_t lost;
2305 
2306 		/*
2307 		 * The DOWN state must be protected from being pushed off
2308 		 * the queue to ensure that userland will always be
2309 		 * in a sane state.
2310 		 * Because DOWN is protected, there is no need to protect
2311 		 * UNKNOWN.
2312 		 * It should be invalid to change from any other state to
2313 		 * UNKNOWN anyway ...
2314 		 */
2315 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2316 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2317 		if (lost == LINK_STATE_DOWN) {
2318 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2319 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2320 		}
2321 		printf("%s: lost link state change %s\n",
2322 		    ifp->if_xname,
2323 		    lost == LINK_STATE_UP ? "UP" :
2324 		    lost == LINK_STATE_DOWN ? "DOWN" :
2325 		    "UNKNOWN");
2326 	} else
2327 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2328 
2329 	if (ifp->if_link_scheduled)
2330 		goto out;
2331 
2332 	ifp->if_link_scheduled = true;
2333 	workqueue_enqueue(ifnet_link_state_wq, &ifp->if_link_work, NULL);
2334 
2335 out:
2336 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2337 }
2338 
2339 /*
2340  * Handle interface link state change notifications.
2341  */
2342 static void
2343 if_link_state_change_process(struct ifnet *ifp, int link_state)
2344 {
2345 	struct domain *dp;
2346 	int s = splnet();
2347 	bool notify;
2348 
2349 	KASSERT(!cpu_intr_p());
2350 
2351 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2352 
2353 	/* Ensure the change is still valid. */
2354 	if (ifp->if_link_state == link_state) {
2355 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2356 		splx(s);
2357 		return;
2358 	}
2359 
2360 #ifdef DEBUG
2361 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2362 		link_state == LINK_STATE_UP ? "UP" :
2363 		link_state == LINK_STATE_DOWN ? "DOWN" :
2364 		"UNKNOWN",
2365 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2366 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2367 		"UNKNOWN");
2368 #endif
2369 
2370 	/*
2371 	 * When going from UNKNOWN to UP, we need to mark existing
2372 	 * addresses as tentative and restart DAD as we may have
2373 	 * erroneously not found a duplicate.
2374 	 *
2375 	 * This needs to happen before rt_ifmsg to avoid a race where
2376 	 * listeners would have an address and expect it to work right
2377 	 * away.
2378 	 */
2379 	notify = (link_state == LINK_STATE_UP &&
2380 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2381 	ifp->if_link_state = link_state;
2382 	/* The following routines may sleep so release the spin mutex */
2383 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2384 
2385 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2386 	if (notify) {
2387 		DOMAIN_FOREACH(dp) {
2388 			if (dp->dom_if_link_state_change != NULL)
2389 				dp->dom_if_link_state_change(ifp,
2390 				    LINK_STATE_DOWN);
2391 		}
2392 	}
2393 
2394 	/* Notify that the link state has changed. */
2395 	rt_ifmsg(ifp);
2396 
2397 #if NCARP > 0
2398 	if (ifp->if_carp)
2399 		carp_carpdev_state(ifp);
2400 #endif
2401 
2402 	if (ifp->if_link_state_changed != NULL)
2403 		ifp->if_link_state_changed(ifp, link_state);
2404 
2405 #if NBRIDGE > 0
2406 	if (ifp->if_bridge != NULL)
2407 		bridge_calc_link_state(ifp->if_bridge);
2408 #endif
2409 
2410 #if NLAGG > 0
2411 	if (ifp->if_lagg != NULL)
2412 		lagg_linkstate_changed(ifp);
2413 #endif
2414 
2415 	DOMAIN_FOREACH(dp) {
2416 		if (dp->dom_if_link_state_change != NULL)
2417 			dp->dom_if_link_state_change(ifp, link_state);
2418 	}
2419 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2420 	splx(s);
2421 }
2422 
2423 /*
2424  * Process the interface link state change queue.
2425  */
2426 static void
2427 if_link_state_change_work(struct work *work, void *arg)
2428 {
2429 	struct ifnet *ifp = container_of(work, struct ifnet, if_link_work);
2430 	int s;
2431 	uint8_t state;
2432 
2433 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2434 	s = splnet();
2435 
2436 	/* Pop a link state change from the queue and process it.
2437 	 * If there is nothing to process then if_detach() has been called.
2438 	 * We keep if_link_scheduled = true so the queue can safely drain
2439 	 * without more work being queued. */
2440 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2441 	LQ_POP(ifp->if_link_queue, state);
2442 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2443 	if (state == LINK_STATE_UNSET)
2444 		goto out;
2445 
2446 	if_link_state_change_process(ifp, state);
2447 
2448 	/* If there is a link state change to come, schedule it. */
2449 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2450 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET) {
2451 		ifp->if_link_scheduled = true;
2452 		workqueue_enqueue(ifnet_link_state_wq, &ifp->if_link_work, NULL);
2453 	} else
2454 		ifp->if_link_scheduled = false;
2455 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2456 
2457 out:
2458 	splx(s);
2459 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2460 }
2461 
2462 /*
2463  * Used to mark addresses on an interface as DETATCHED or TENTATIVE
2464  * and thus start Duplicate Address Detection without changing the
2465  * real link state.
2466  */
2467 void
2468 if_domain_link_state_change(struct ifnet *ifp, int link_state)
2469 {
2470 	struct domain *dp;
2471 	int s = splnet();
2472 
2473 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2474 
2475 	DOMAIN_FOREACH(dp) {
2476 		if (dp->dom_if_link_state_change != NULL)
2477 			dp->dom_if_link_state_change(ifp, link_state);
2478 	}
2479 
2480 	splx(s);
2481 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2482 }
2483 
2484 /*
2485  * Default action when installing a local route on a point-to-point
2486  * interface.
2487  */
2488 void
2489 p2p_rtrequest(int req, struct rtentry *rt,
2490     __unused const struct rt_addrinfo *info)
2491 {
2492 	struct ifnet *ifp = rt->rt_ifp;
2493 	struct ifaddr *ifa, *lo0ifa;
2494 	int s = pserialize_read_enter();
2495 
2496 	switch (req) {
2497 	case RTM_ADD:
2498 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2499 			break;
2500 
2501 		rt->rt_ifp = lo0ifp;
2502 
2503 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA))
2504 			break;
2505 
2506 		IFADDR_READER_FOREACH(ifa, ifp) {
2507 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2508 				break;
2509 		}
2510 		if (ifa == NULL)
2511 			break;
2512 
2513 		/*
2514 		 * Ensure lo0 has an address of the same family.
2515 		 */
2516 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2517 			if (lo0ifa->ifa_addr->sa_family ==
2518 			    ifa->ifa_addr->sa_family)
2519 				break;
2520 		}
2521 		if (lo0ifa == NULL)
2522 			break;
2523 
2524 		/*
2525 		 * Make sure to set rt->rt_ifa to the interface
2526 		 * address we are using, otherwise we will have trouble
2527 		 * with source address selection.
2528 		 */
2529 		if (ifa != rt->rt_ifa)
2530 			rt_replace_ifa(rt, ifa);
2531 		break;
2532 	case RTM_DELETE:
2533 	default:
2534 		break;
2535 	}
2536 	pserialize_read_exit(s);
2537 }
2538 
2539 static void
2540 _if_down(struct ifnet *ifp)
2541 {
2542 	struct ifaddr *ifa;
2543 	struct domain *dp;
2544 	int s, bound;
2545 	struct psref psref;
2546 
2547 	ifp->if_flags &= ~IFF_UP;
2548 	nanotime(&ifp->if_lastchange);
2549 
2550 	bound = curlwp_bind();
2551 	s = pserialize_read_enter();
2552 	IFADDR_READER_FOREACH(ifa, ifp) {
2553 		ifa_acquire(ifa, &psref);
2554 		pserialize_read_exit(s);
2555 
2556 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2557 
2558 		s = pserialize_read_enter();
2559 		ifa_release(ifa, &psref);
2560 	}
2561 	pserialize_read_exit(s);
2562 	curlwp_bindx(bound);
2563 
2564 	IFQ_PURGE(&ifp->if_snd);
2565 #if NCARP > 0
2566 	if (ifp->if_carp)
2567 		carp_carpdev_state(ifp);
2568 #endif
2569 	rt_ifmsg(ifp);
2570 	DOMAIN_FOREACH(dp) {
2571 		if (dp->dom_if_down)
2572 			dp->dom_if_down(ifp);
2573 	}
2574 }
2575 
2576 static void
2577 if_down_deactivated(struct ifnet *ifp)
2578 {
2579 
2580 	KASSERT(if_is_deactivated(ifp));
2581 	_if_down(ifp);
2582 }
2583 
2584 void
2585 if_down_locked(struct ifnet *ifp)
2586 {
2587 
2588 	KASSERT(IFNET_LOCKED(ifp));
2589 	_if_down(ifp);
2590 }
2591 
2592 /*
2593  * Mark an interface down and notify protocols of
2594  * the transition.
2595  * NOTE: must be called at splsoftnet or equivalent.
2596  */
2597 void
2598 if_down(struct ifnet *ifp)
2599 {
2600 
2601 	IFNET_LOCK(ifp);
2602 	if_down_locked(ifp);
2603 	IFNET_UNLOCK(ifp);
2604 }
2605 
2606 /*
2607  * Must be called with holding if_ioctl_lock.
2608  */
2609 static void
2610 if_up_locked(struct ifnet *ifp)
2611 {
2612 #ifdef notyet
2613 	struct ifaddr *ifa;
2614 #endif
2615 	struct domain *dp;
2616 
2617 	KASSERT(IFNET_LOCKED(ifp));
2618 
2619 	KASSERT(!if_is_deactivated(ifp));
2620 	ifp->if_flags |= IFF_UP;
2621 	nanotime(&ifp->if_lastchange);
2622 #ifdef notyet
2623 	/* this has no effect on IP, and will kill all ISO connections XXX */
2624 	IFADDR_READER_FOREACH(ifa, ifp)
2625 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2626 #endif
2627 #if NCARP > 0
2628 	if (ifp->if_carp)
2629 		carp_carpdev_state(ifp);
2630 #endif
2631 	rt_ifmsg(ifp);
2632 	DOMAIN_FOREACH(dp) {
2633 		if (dp->dom_if_up)
2634 			dp->dom_if_up(ifp);
2635 	}
2636 }
2637 
2638 /*
2639  * Handle interface slowtimo timer routine.  Called
2640  * from softclock, we decrement timer (if set) and
2641  * call the appropriate interface routine on expiration.
2642  */
2643 static void
2644 if_slowtimo(void *arg)
2645 {
2646 	void (*slowtimo)(struct ifnet *);
2647 	struct ifnet *ifp = arg;
2648 	int s;
2649 
2650 	slowtimo = ifp->if_slowtimo;
2651 	if (__predict_false(slowtimo == NULL))
2652 		return;
2653 
2654 	s = splnet();
2655 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2656 		(*slowtimo)(ifp);
2657 
2658 	splx(s);
2659 
2660 	if (__predict_true(ifp->if_slowtimo != NULL))
2661 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2662 }
2663 
2664 /*
2665  * Mark an interface up and notify protocols of
2666  * the transition.
2667  * NOTE: must be called at splsoftnet or equivalent.
2668  */
2669 void
2670 if_up(struct ifnet *ifp)
2671 {
2672 
2673 	IFNET_LOCK(ifp);
2674 	if_up_locked(ifp);
2675 	IFNET_UNLOCK(ifp);
2676 }
2677 
2678 /*
2679  * Set/clear promiscuous mode on interface ifp based on the truth value
2680  * of pswitch.  The calls are reference counted so that only the first
2681  * "on" request actually has an effect, as does the final "off" request.
2682  * Results are undefined if the "off" and "on" requests are not matched.
2683  */
2684 int
2685 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2686 {
2687 	int pcount, ret = 0;
2688 	u_short nflags;
2689 
2690 	KASSERT(IFNET_LOCKED(ifp));
2691 
2692 	pcount = ifp->if_pcount;
2693 	if (pswitch) {
2694 		/*
2695 		 * Allow the device to be "placed" into promiscuous
2696 		 * mode even if it is not configured up.  It will
2697 		 * consult IFF_PROMISC when it is brought up.
2698 		 */
2699 		if (ifp->if_pcount++ != 0)
2700 			goto out;
2701 		nflags = ifp->if_flags | IFF_PROMISC;
2702 	} else {
2703 		if (--ifp->if_pcount > 0)
2704 			goto out;
2705 		nflags = ifp->if_flags & ~IFF_PROMISC;
2706 	}
2707 	ret = if_flags_set(ifp, nflags);
2708 	/* Restore interface state if not successful. */
2709 	if (ret != 0) {
2710 		ifp->if_pcount = pcount;
2711 	}
2712 out:
2713 	return ret;
2714 }
2715 
2716 int
2717 ifpromisc(struct ifnet *ifp, int pswitch)
2718 {
2719 	int e;
2720 
2721 	IFNET_LOCK(ifp);
2722 	e = ifpromisc_locked(ifp, pswitch);
2723 	IFNET_UNLOCK(ifp);
2724 
2725 	return e;
2726 }
2727 
2728 /*
2729  * Map interface name to
2730  * interface structure pointer.
2731  */
2732 struct ifnet *
2733 ifunit(const char *name)
2734 {
2735 	struct ifnet *ifp;
2736 	const char *cp = name;
2737 	u_int unit = 0;
2738 	u_int i;
2739 	int s;
2740 
2741 	/*
2742 	 * If the entire name is a number, treat it as an ifindex.
2743 	 */
2744 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2745 		unit = unit * 10 + (*cp - '0');
2746 	}
2747 
2748 	/*
2749 	 * If the number took all of the name, then it's a valid ifindex.
2750 	 */
2751 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2752 		return if_byindex(unit);
2753 
2754 	ifp = NULL;
2755 	s = pserialize_read_enter();
2756 	IFNET_READER_FOREACH(ifp) {
2757 		if (if_is_deactivated(ifp))
2758 			continue;
2759 	 	if (strcmp(ifp->if_xname, name) == 0)
2760 			goto out;
2761 	}
2762 out:
2763 	pserialize_read_exit(s);
2764 	return ifp;
2765 }
2766 
2767 /*
2768  * Get a reference of an ifnet object by an interface name.
2769  * The returned reference is protected by psref(9). The caller
2770  * must release a returned reference by if_put after use.
2771  */
2772 struct ifnet *
2773 if_get(const char *name, struct psref *psref)
2774 {
2775 	struct ifnet *ifp;
2776 	const char *cp = name;
2777 	u_int unit = 0;
2778 	u_int i;
2779 	int s;
2780 
2781 	/*
2782 	 * If the entire name is a number, treat it as an ifindex.
2783 	 */
2784 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2785 		unit = unit * 10 + (*cp - '0');
2786 	}
2787 
2788 	/*
2789 	 * If the number took all of the name, then it's a valid ifindex.
2790 	 */
2791 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2792 		return if_get_byindex(unit, psref);
2793 
2794 	ifp = NULL;
2795 	s = pserialize_read_enter();
2796 	IFNET_READER_FOREACH(ifp) {
2797 		if (if_is_deactivated(ifp))
2798 			continue;
2799 		if (strcmp(ifp->if_xname, name) == 0) {
2800 			PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2801 			psref_acquire(psref, &ifp->if_psref,
2802 			    ifnet_psref_class);
2803 			goto out;
2804 		}
2805 	}
2806 out:
2807 	pserialize_read_exit(s);
2808 	return ifp;
2809 }
2810 
2811 /*
2812  * Release a reference of an ifnet object given by if_get, if_get_byindex
2813  * or if_get_bylla.
2814  */
2815 void
2816 if_put(const struct ifnet *ifp, struct psref *psref)
2817 {
2818 
2819 	if (ifp == NULL)
2820 		return;
2821 
2822 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2823 }
2824 
2825 /*
2826  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2827  * should be used.
2828  */
2829 ifnet_t *
2830 _if_byindex(u_int idx)
2831 {
2832 
2833 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2834 }
2835 
2836 /*
2837  * Return ifp having idx. Return NULL if not found or the found ifp is
2838  * already deactivated.
2839  */
2840 ifnet_t *
2841 if_byindex(u_int idx)
2842 {
2843 	ifnet_t *ifp;
2844 
2845 	ifp = _if_byindex(idx);
2846 	if (ifp != NULL && if_is_deactivated(ifp))
2847 		ifp = NULL;
2848 	return ifp;
2849 }
2850 
2851 /*
2852  * Get a reference of an ifnet object by an interface index.
2853  * The returned reference is protected by psref(9). The caller
2854  * must release a returned reference by if_put after use.
2855  */
2856 ifnet_t *
2857 if_get_byindex(u_int idx, struct psref *psref)
2858 {
2859 	ifnet_t *ifp;
2860 	int s;
2861 
2862 	s = pserialize_read_enter();
2863 	ifp = if_byindex(idx);
2864 	if (__predict_true(ifp != NULL)) {
2865 		PSREF_DEBUG_FILL_RETURN_ADDRESS(psref);
2866 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2867 	}
2868 	pserialize_read_exit(s);
2869 
2870 	return ifp;
2871 }
2872 
2873 ifnet_t *
2874 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2875 {
2876 	ifnet_t *ifp;
2877 	int s;
2878 
2879 	s = pserialize_read_enter();
2880 	IFNET_READER_FOREACH(ifp) {
2881 		if (if_is_deactivated(ifp))
2882 			continue;
2883 		if (ifp->if_addrlen != lla_len)
2884 			continue;
2885 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2886 			psref_acquire(psref, &ifp->if_psref,
2887 			    ifnet_psref_class);
2888 			break;
2889 		}
2890 	}
2891 	pserialize_read_exit(s);
2892 
2893 	return ifp;
2894 }
2895 
2896 /*
2897  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2898  * for example using pserialize or the ifp is already held or some other
2899  * object is held which guarantes the ifp to not be freed indirectly.
2900  */
2901 void
2902 if_acquire(struct ifnet *ifp, struct psref *psref)
2903 {
2904 
2905 	KASSERT(ifp->if_index != 0);
2906 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2907 }
2908 
2909 bool
2910 if_held(struct ifnet *ifp)
2911 {
2912 
2913 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2914 }
2915 
2916 /*
2917  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4.
2918  * Check the tunnel nesting count.
2919  * Return > 0, if tunnel nesting count is more than limit.
2920  * Return 0, if tunnel nesting count is equal or less than limit.
2921  */
2922 int
2923 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
2924 {
2925 	struct m_tag *mtag;
2926 	int *count;
2927 
2928 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO);
2929 	if (mtag != NULL) {
2930 		count = (int *)(mtag + 1);
2931 		if (++(*count) > limit) {
2932 			log(LOG_NOTICE,
2933 			    "%s: recursively called too many times(%d)\n",
2934 			    ifp->if_xname, *count);
2935 			return EIO;
2936 		}
2937 	} else {
2938 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
2939 		    M_NOWAIT);
2940 		if (mtag != NULL) {
2941 			m_tag_prepend(m, mtag);
2942 			count = (int *)(mtag + 1);
2943 			*count = 0;
2944 		} else {
2945 			log(LOG_DEBUG,
2946 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
2947 			    ifp->if_xname);
2948 		}
2949 	}
2950 
2951 	return 0;
2952 }
2953 
2954 static void
2955 if_tunnel_ro_init_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2956 {
2957 	struct tunnel_ro *tro = p;
2958 
2959 	tro->tr_ro = kmem_zalloc(sizeof(*tro->tr_ro), KM_SLEEP);
2960 	tro->tr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2961 }
2962 
2963 static void
2964 if_tunnel_ro_fini_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2965 {
2966 	struct tunnel_ro *tro = p;
2967 
2968 	rtcache_free(tro->tr_ro);
2969 	kmem_free(tro->tr_ro, sizeof(*tro->tr_ro));
2970 
2971 	mutex_obj_free(tro->tr_lock);
2972 }
2973 
2974 percpu_t *
2975 if_tunnel_alloc_ro_percpu(void)
2976 {
2977 
2978 	return percpu_create(sizeof(struct tunnel_ro),
2979 	    if_tunnel_ro_init_pc, if_tunnel_ro_fini_pc, NULL);
2980 }
2981 
2982 void
2983 if_tunnel_free_ro_percpu(percpu_t *ro_percpu)
2984 {
2985 
2986 	percpu_free(ro_percpu, sizeof(struct tunnel_ro));
2987 }
2988 
2989 
2990 static void
2991 if_tunnel_rtcache_free_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
2992 {
2993 	struct tunnel_ro *tro = p;
2994 
2995 	mutex_enter(tro->tr_lock);
2996 	rtcache_free(tro->tr_ro);
2997 	mutex_exit(tro->tr_lock);
2998 }
2999 
3000 void if_tunnel_ro_percpu_rtcache_free(percpu_t *ro_percpu)
3001 {
3002 
3003 	percpu_foreach(ro_percpu, if_tunnel_rtcache_free_pc, NULL);
3004 }
3005 
3006 void
3007 if_export_if_data(ifnet_t * const ifp, struct if_data *ifi, bool zero_stats)
3008 {
3009 
3010 	/* Collet the volatile stats first; this zeros *ifi. */
3011 	if_stats_to_if_data(ifp, ifi, zero_stats);
3012 
3013 	ifi->ifi_type = ifp->if_type;
3014 	ifi->ifi_addrlen = ifp->if_addrlen;
3015 	ifi->ifi_hdrlen = ifp->if_hdrlen;
3016 	ifi->ifi_link_state = ifp->if_link_state;
3017 	ifi->ifi_mtu = ifp->if_mtu;
3018 	ifi->ifi_metric = ifp->if_metric;
3019 	ifi->ifi_baudrate = ifp->if_baudrate;
3020 	ifi->ifi_lastchange = ifp->if_lastchange;
3021 }
3022 
3023 /* common */
3024 int
3025 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
3026 {
3027 	int s;
3028 	struct ifreq *ifr;
3029 	struct ifcapreq *ifcr;
3030 	struct ifdatareq *ifdr;
3031 	unsigned short flags;
3032 	char *descr;
3033 	int error;
3034 
3035 	switch (cmd) {
3036 	case SIOCSIFCAP:
3037 		ifcr = data;
3038 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
3039 			return EINVAL;
3040 
3041 		if (ifcr->ifcr_capenable == ifp->if_capenable)
3042 			return 0;
3043 
3044 		ifp->if_capenable = ifcr->ifcr_capenable;
3045 
3046 		/* Pre-compute the checksum flags mask. */
3047 		ifp->if_csum_flags_tx = 0;
3048 		ifp->if_csum_flags_rx = 0;
3049 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx)
3050 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
3051 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
3052 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
3053 
3054 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx)
3055 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
3056 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
3057 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
3058 
3059 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx)
3060 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
3061 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
3062 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
3063 
3064 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx)
3065 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
3066 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
3067 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
3068 
3069 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx)
3070 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
3071 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
3072 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
3073 
3074 		if (ifp->if_capenable & IFCAP_TSOv4)
3075 			ifp->if_csum_flags_tx |= M_CSUM_TSOv4;
3076 		if (ifp->if_capenable & IFCAP_TSOv6)
3077 			ifp->if_csum_flags_tx |= M_CSUM_TSOv6;
3078 
3079 #if NBRIDGE > 0
3080 		if (ifp->if_bridge != NULL)
3081 			bridge_calc_csum_flags(ifp->if_bridge);
3082 #endif
3083 
3084 		if (ifp->if_flags & IFF_UP)
3085 			return ENETRESET;
3086 		return 0;
3087 	case SIOCSIFFLAGS:
3088 		ifr = data;
3089 		/*
3090 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
3091 		 * and if_down aren't MP-safe yet, so we must hold the lock.
3092 		 */
3093 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3094 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
3095 			s = splsoftnet();
3096 			if_down_locked(ifp);
3097 			splx(s);
3098 		}
3099 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
3100 			s = splsoftnet();
3101 			if_up_locked(ifp);
3102 			splx(s);
3103 		}
3104 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3105 		flags = (ifp->if_flags & IFF_CANTCHANGE) |
3106 		    (ifr->ifr_flags &~ IFF_CANTCHANGE);
3107 		if (ifp->if_flags != flags) {
3108 			ifp->if_flags = flags;
3109 			/* Notify that the flags have changed. */
3110 			rt_ifmsg(ifp);
3111 		}
3112 		break;
3113 	case SIOCGIFFLAGS:
3114 		ifr = data;
3115 		ifr->ifr_flags = ifp->if_flags;
3116 		break;
3117 
3118 	case SIOCGIFMETRIC:
3119 		ifr = data;
3120 		ifr->ifr_metric = ifp->if_metric;
3121 		break;
3122 
3123 	case SIOCGIFMTU:
3124 		ifr = data;
3125 		ifr->ifr_mtu = ifp->if_mtu;
3126 		break;
3127 
3128 	case SIOCGIFDLT:
3129 		ifr = data;
3130 		ifr->ifr_dlt = ifp->if_dlt;
3131 		break;
3132 
3133 	case SIOCGIFCAP:
3134 		ifcr = data;
3135 		ifcr->ifcr_capabilities = ifp->if_capabilities;
3136 		ifcr->ifcr_capenable = ifp->if_capenable;
3137 		break;
3138 
3139 	case SIOCSIFMETRIC:
3140 		ifr = data;
3141 		ifp->if_metric = ifr->ifr_metric;
3142 		break;
3143 
3144 	case SIOCGIFDATA:
3145 		ifdr = data;
3146 		if_export_if_data(ifp, &ifdr->ifdr_data, false);
3147 		break;
3148 
3149 	case SIOCGIFINDEX:
3150 		ifr = data;
3151 		ifr->ifr_index = ifp->if_index;
3152 		break;
3153 
3154 	case SIOCZIFDATA:
3155 		ifdr = data;
3156 		if_export_if_data(ifp, &ifdr->ifdr_data, true);
3157 		getnanotime(&ifp->if_lastchange);
3158 		break;
3159 	case SIOCSIFMTU:
3160 		ifr = data;
3161 		if (ifp->if_mtu == ifr->ifr_mtu)
3162 			break;
3163 		ifp->if_mtu = ifr->ifr_mtu;
3164 		return ENETRESET;
3165 	case SIOCSIFDESCR:
3166 		error = kauth_authorize_network(curlwp->l_cred,
3167 		    KAUTH_NETWORK_INTERFACE,
3168 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, KAUTH_ARG(cmd),
3169 		    NULL);
3170 		if (error)
3171 			return error;
3172 
3173 		ifr = data;
3174 
3175 		if (ifr->ifr_buflen > IFDESCRSIZE)
3176 			return ENAMETOOLONG;
3177 
3178 		if (ifr->ifr_buf == NULL || ifr->ifr_buflen == 0) {
3179 			/* unset description */
3180 			descr = NULL;
3181 		} else {
3182 			descr = kmem_zalloc(IFDESCRSIZE, KM_SLEEP);
3183 			/*
3184 			 * copy (IFDESCRSIZE - 1) bytes to ensure
3185 			 * terminating nul
3186 			 */
3187 			error = copyin(ifr->ifr_buf, descr, IFDESCRSIZE - 1);
3188 			if (error) {
3189 				kmem_free(descr, IFDESCRSIZE);
3190 				return error;
3191 			}
3192 		}
3193 
3194 		if (ifp->if_description != NULL)
3195 			kmem_free(ifp->if_description, IFDESCRSIZE);
3196 
3197 		ifp->if_description = descr;
3198 		break;
3199 
3200  	case SIOCGIFDESCR:
3201 		ifr = data;
3202 		descr = ifp->if_description;
3203 
3204 		if (descr == NULL)
3205 			return ENOMSG;
3206 
3207 		if (ifr->ifr_buflen < IFDESCRSIZE)
3208 			return EINVAL;
3209 
3210 		error = copyout(descr, ifr->ifr_buf, IFDESCRSIZE);
3211 		if (error)
3212 			return error;
3213  		break;
3214 
3215 	default:
3216 		return ENOTTY;
3217 	}
3218 	return 0;
3219 }
3220 
3221 int
3222 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3223 {
3224 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3225 	struct ifaddr *ifa;
3226 	const struct sockaddr *any, *sa;
3227 	union {
3228 		struct sockaddr sa;
3229 		struct sockaddr_storage ss;
3230 	} u, v;
3231 	int s, error = 0;
3232 
3233 	switch (cmd) {
3234 	case SIOCSIFADDRPREF:
3235 		error = kauth_authorize_network(curlwp->l_cred,
3236 		    KAUTH_NETWORK_INTERFACE,
3237 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, KAUTH_ARG(cmd),
3238 		    NULL);
3239 		if (error)
3240 			return error;
3241 		break;
3242 	case SIOCGIFADDRPREF:
3243 		break;
3244 	default:
3245 		return EOPNOTSUPP;
3246 	}
3247 
3248 	/* sanity checks */
3249 	if (data == NULL || ifp == NULL) {
3250 		panic("invalid argument to %s", __func__);
3251 		/*NOTREACHED*/
3252 	}
3253 
3254 	/* address must be specified on ADD and DELETE */
3255 	sa = sstocsa(&ifap->ifap_addr);
3256 	if (sa->sa_family != sofamily(so))
3257 		return EINVAL;
3258 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3259 		return EINVAL;
3260 
3261 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3262 
3263 	s = pserialize_read_enter();
3264 	IFADDR_READER_FOREACH(ifa, ifp) {
3265 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3266 			continue;
3267 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3268 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3269 			break;
3270 	}
3271 	if (ifa == NULL) {
3272 		error = EADDRNOTAVAIL;
3273 		goto out;
3274 	}
3275 
3276 	switch (cmd) {
3277 	case SIOCSIFADDRPREF:
3278 		ifa->ifa_preference = ifap->ifap_preference;
3279 		goto out;
3280 	case SIOCGIFADDRPREF:
3281 		/* fill in the if_laddrreq structure */
3282 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3283 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3284 		ifap->ifap_preference = ifa->ifa_preference;
3285 		goto out;
3286 	default:
3287 		error = EOPNOTSUPP;
3288 	}
3289 out:
3290 	pserialize_read_exit(s);
3291 	return error;
3292 }
3293 
3294 /*
3295  * Interface ioctls.
3296  */
3297 static int
3298 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3299 {
3300 	struct ifnet *ifp;
3301 	struct ifreq *ifr;
3302 	int error = 0;
3303 	u_long ocmd = cmd;
3304 	u_short oif_flags;
3305 	struct ifreq ifrb;
3306 	struct oifreq *oifr = NULL;
3307 	int r;
3308 	struct psref psref;
3309 	int bound;
3310 	bool do_if43_post = false;
3311 	bool do_ifm80_post = false;
3312 
3313 	switch (cmd) {
3314 	case SIOCGIFCONF:
3315 		return ifconf(cmd, data);
3316 	case SIOCINITIFADDR:
3317 		return EPERM;
3318 	default:
3319 		MODULE_HOOK_CALL(uipc_syscalls_40_hook, (cmd, data), enosys(),
3320 		    error);
3321 		if (error != ENOSYS)
3322 			return error;
3323 		MODULE_HOOK_CALL(uipc_syscalls_50_hook, (l, cmd, data),
3324 		    enosys(), error);
3325 		if (error != ENOSYS)
3326 			return error;
3327 		error = 0;
3328 		break;
3329 	}
3330 
3331 	ifr = data;
3332 	/* Pre-conversion */
3333 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), error);
3334 	if (cmd != ocmd) {
3335 		oifr = data;
3336 		data = ifr = &ifrb;
3337 		IFREQO2N_43(oifr, ifr);
3338 		do_if43_post = true;
3339 	}
3340 	MODULE_HOOK_CALL(ifmedia_80_pre_hook, (ifr, &cmd, &do_ifm80_post),
3341 	    enosys(), error);
3342 
3343 	switch (cmd) {
3344 	case SIOCIFCREATE:
3345 	case SIOCIFDESTROY:
3346 		bound = curlwp_bind();
3347 		if (l != NULL) {
3348 			ifp = if_get(ifr->ifr_name, &psref);
3349 			error = kauth_authorize_network(l->l_cred,
3350 			    KAUTH_NETWORK_INTERFACE,
3351 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3352 			    KAUTH_ARG(cmd), NULL);
3353 			if (ifp != NULL)
3354 				if_put(ifp, &psref);
3355 			if (error != 0) {
3356 				curlwp_bindx(bound);
3357 				return error;
3358 			}
3359 		}
3360 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3361 		mutex_enter(&if_clone_mtx);
3362 		r = (cmd == SIOCIFCREATE) ?
3363 			if_clone_create(ifr->ifr_name) :
3364 			if_clone_destroy(ifr->ifr_name);
3365 		mutex_exit(&if_clone_mtx);
3366 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3367 		curlwp_bindx(bound);
3368 		return r;
3369 
3370 	case SIOCIFGCLONERS:
3371 		{
3372 			struct if_clonereq *req = (struct if_clonereq *)data;
3373 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3374 			    &req->ifcr_total);
3375 		}
3376 	}
3377 
3378 	bound = curlwp_bind();
3379 	ifp = if_get(ifr->ifr_name, &psref);
3380 	if (ifp == NULL) {
3381 		curlwp_bindx(bound);
3382 		return ENXIO;
3383 	}
3384 
3385 	switch (cmd) {
3386 	case SIOCALIFADDR:
3387 	case SIOCDLIFADDR:
3388 	case SIOCSIFADDRPREF:
3389 	case SIOCSIFFLAGS:
3390 	case SIOCSIFCAP:
3391 	case SIOCSIFMETRIC:
3392 	case SIOCZIFDATA:
3393 	case SIOCSIFMTU:
3394 	case SIOCSIFPHYADDR:
3395 	case SIOCDIFPHYADDR:
3396 #ifdef INET6
3397 	case SIOCSIFPHYADDR_IN6:
3398 #endif
3399 	case SIOCSLIFPHYADDR:
3400 	case SIOCADDMULTI:
3401 	case SIOCDELMULTI:
3402 	case SIOCSETHERCAP:
3403 	case SIOCSIFMEDIA:
3404 	case SIOCSDRVSPEC:
3405 	case SIOCG80211:
3406 	case SIOCS80211:
3407 	case SIOCS80211NWID:
3408 	case SIOCS80211NWKEY:
3409 	case SIOCS80211POWER:
3410 	case SIOCS80211BSSID:
3411 	case SIOCS80211CHANNEL:
3412 	case SIOCSLINKSTR:
3413 		if (l != NULL) {
3414 			error = kauth_authorize_network(l->l_cred,
3415 			    KAUTH_NETWORK_INTERFACE,
3416 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3417 			    KAUTH_ARG(cmd), NULL);
3418 			if (error != 0)
3419 				goto out;
3420 		}
3421 	}
3422 
3423 	oif_flags = ifp->if_flags;
3424 
3425 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3426 	IFNET_LOCK(ifp);
3427 
3428 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3429 	if (error != ENOTTY)
3430 		;
3431 	else if (so->so_proto == NULL)
3432 		error = EOPNOTSUPP;
3433 	else {
3434 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3435 		MODULE_HOOK_CALL(if_ifioctl_43_hook,
3436 			     (so, ocmd, cmd, data, l), enosys(), error);
3437 		if (error == ENOSYS)
3438 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3439 			    cmd, data, ifp);
3440 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3441 	}
3442 
3443 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3444 		if ((ifp->if_flags & IFF_UP) != 0) {
3445 			int s = splsoftnet();
3446 			if_up_locked(ifp);
3447 			splx(s);
3448 		}
3449 	}
3450 
3451 	/* Post-conversion */
3452 	if (do_ifm80_post && (error == 0))
3453 		MODULE_HOOK_CALL(ifmedia_80_post_hook, (ifr, cmd),
3454 		    enosys(), error);
3455 	if (do_if43_post)
3456 		IFREQN2O_43(oifr, ifr);
3457 
3458 	IFNET_UNLOCK(ifp);
3459 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3460 out:
3461 	if_put(ifp, &psref);
3462 	curlwp_bindx(bound);
3463 	return error;
3464 }
3465 
3466 /*
3467  * Return interface configuration
3468  * of system.  List may be used
3469  * in later ioctl's (above) to get
3470  * other information.
3471  *
3472  * Each record is a struct ifreq.  Before the addition of
3473  * sockaddr_storage, the API rule was that sockaddr flavors that did
3474  * not fit would extend beyond the struct ifreq, with the next struct
3475  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3476  * union in struct ifreq includes struct sockaddr_storage, every kind
3477  * of sockaddr must fit.  Thus, there are no longer any overlength
3478  * records.
3479  *
3480  * Records are added to the user buffer if they fit, and ifc_len is
3481  * adjusted to the length that was written.  Thus, the user is only
3482  * assured of getting the complete list if ifc_len on return is at
3483  * least sizeof(struct ifreq) less than it was on entry.
3484  *
3485  * If the user buffer pointer is NULL, this routine copies no data and
3486  * returns the amount of space that would be needed.
3487  *
3488  * Invariants:
3489  * ifrp points to the next part of the user's buffer to be used.  If
3490  * ifrp != NULL, space holds the number of bytes remaining that we may
3491  * write at ifrp.  Otherwise, space holds the number of bytes that
3492  * would have been written had there been adequate space.
3493  */
3494 /*ARGSUSED*/
3495 static int
3496 ifconf(u_long cmd, void *data)
3497 {
3498 	struct ifconf *ifc = (struct ifconf *)data;
3499 	struct ifnet *ifp;
3500 	struct ifaddr *ifa;
3501 	struct ifreq ifr, *ifrp = NULL;
3502 	int space = 0, error = 0;
3503 	const int sz = (int)sizeof(struct ifreq);
3504 	const bool docopy = ifc->ifc_req != NULL;
3505 	int s;
3506 	int bound;
3507 	struct psref psref;
3508 
3509 	if (docopy) {
3510 		if (ifc->ifc_len < 0)
3511 			return EINVAL;
3512 
3513 		space = ifc->ifc_len;
3514 		ifrp = ifc->ifc_req;
3515 	}
3516 	memset(&ifr, 0, sizeof(ifr));
3517 
3518 	bound = curlwp_bind();
3519 	s = pserialize_read_enter();
3520 	IFNET_READER_FOREACH(ifp) {
3521 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3522 		pserialize_read_exit(s);
3523 
3524 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3525 		    sizeof(ifr.ifr_name));
3526 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3527 			error = ENAMETOOLONG;
3528 			goto release_exit;
3529 		}
3530 		if (IFADDR_READER_EMPTY(ifp)) {
3531 			/* Interface with no addresses - send zero sockaddr. */
3532 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3533 			if (!docopy) {
3534 				space += sz;
3535 				goto next;
3536 			}
3537 			if (space >= sz) {
3538 				error = copyout(&ifr, ifrp, sz);
3539 				if (error != 0)
3540 					goto release_exit;
3541 				ifrp++;
3542 				space -= sz;
3543 			}
3544 		}
3545 
3546 		s = pserialize_read_enter();
3547 		IFADDR_READER_FOREACH(ifa, ifp) {
3548 			struct sockaddr *sa = ifa->ifa_addr;
3549 			/* all sockaddrs must fit in sockaddr_storage */
3550 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3551 
3552 			if (!docopy) {
3553 				space += sz;
3554 				continue;
3555 			}
3556 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3557 			pserialize_read_exit(s);
3558 
3559 			if (space >= sz) {
3560 				error = copyout(&ifr, ifrp, sz);
3561 				if (error != 0)
3562 					goto release_exit;
3563 				ifrp++; space -= sz;
3564 			}
3565 			s = pserialize_read_enter();
3566 		}
3567 		pserialize_read_exit(s);
3568 
3569         next:
3570 		s = pserialize_read_enter();
3571 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3572 	}
3573 	pserialize_read_exit(s);
3574 	curlwp_bindx(bound);
3575 
3576 	if (docopy) {
3577 		KASSERT(0 <= space && space <= ifc->ifc_len);
3578 		ifc->ifc_len -= space;
3579 	} else {
3580 		KASSERT(space >= 0);
3581 		ifc->ifc_len = space;
3582 	}
3583 	return (0);
3584 
3585 release_exit:
3586 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3587 	curlwp_bindx(bound);
3588 	return error;
3589 }
3590 
3591 int
3592 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3593 {
3594 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3595 	struct ifreq ifrb;
3596 	struct oifreq *oifr = NULL;
3597 	u_long ocmd = cmd;
3598 	int hook;
3599 
3600 	MODULE_HOOK_CALL(if_cvtcmd_43_hook, (&cmd, ocmd), enosys(), hook);
3601 	if (hook != ENOSYS) {
3602 		if (cmd != ocmd) {
3603 			oifr = (struct oifreq *)(void *)ifr;
3604 			ifr = &ifrb;
3605 			IFREQO2N_43(oifr, ifr);
3606 				len = sizeof(oifr->ifr_addr);
3607 		}
3608 	}
3609 
3610 	if (len < sa->sa_len)
3611 		return EFBIG;
3612 
3613 	memset(&ifr->ifr_addr, 0, len);
3614 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3615 
3616 	if (cmd != ocmd)
3617 		IFREQN2O_43(oifr, ifr);
3618 	return 0;
3619 }
3620 
3621 /*
3622  * wrapper function for the drivers which doesn't have if_transmit().
3623  */
3624 static int
3625 if_transmit(struct ifnet *ifp, struct mbuf *m)
3626 {
3627 	int s, error;
3628 	size_t pktlen = m->m_pkthdr.len;
3629 	bool mcast = (m->m_flags & M_MCAST) != 0;
3630 
3631 	s = splnet();
3632 
3633 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3634 	if (error != 0) {
3635 		/* mbuf is already freed */
3636 		goto out;
3637 	}
3638 
3639 	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
3640 	if_statadd_ref(nsr, if_obytes, pktlen);
3641 	if (mcast)
3642 		if_statinc_ref(nsr, if_omcasts);
3643 	IF_STAT_PUTREF(ifp);
3644 
3645 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3646 		if_start_lock(ifp);
3647 out:
3648 	splx(s);
3649 
3650 	return error;
3651 }
3652 
3653 int
3654 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3655 {
3656 	int error;
3657 
3658 	kmsan_check_mbuf(m);
3659 
3660 #ifdef ALTQ
3661 	KERNEL_LOCK(1, NULL);
3662 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3663 		error = if_transmit(ifp, m);
3664 		KERNEL_UNLOCK_ONE(NULL);
3665 	} else {
3666 		KERNEL_UNLOCK_ONE(NULL);
3667 		error = (*ifp->if_transmit)(ifp, m);
3668 		/* mbuf is alredy freed */
3669 	}
3670 #else /* !ALTQ */
3671 	error = (*ifp->if_transmit)(ifp, m);
3672 	/* mbuf is alredy freed */
3673 #endif /* !ALTQ */
3674 
3675 	return error;
3676 }
3677 
3678 /*
3679  * Queue message on interface, and start output if interface
3680  * not yet active.
3681  */
3682 int
3683 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3684 {
3685 
3686 	return if_transmit_lock(ifp, m);
3687 }
3688 
3689 /*
3690  * Queue message on interface, possibly using a second fast queue
3691  */
3692 int
3693 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3694 {
3695 	int error = 0;
3696 
3697 	if (ifq != NULL
3698 #ifdef ALTQ
3699 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3700 #endif
3701 	    ) {
3702 		if (IF_QFULL(ifq)) {
3703 			IF_DROP(&ifp->if_snd);
3704 			m_freem(m);
3705 			if (error == 0)
3706 				error = ENOBUFS;
3707 		} else
3708 			IF_ENQUEUE(ifq, m);
3709 	} else
3710 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3711 	if (error != 0) {
3712 		if_statinc(ifp, if_oerrors);
3713 		return error;
3714 	}
3715 	return 0;
3716 }
3717 
3718 int
3719 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3720 {
3721 	int rc;
3722 
3723 	KASSERT(IFNET_LOCKED(ifp));
3724 	if (ifp->if_initaddr != NULL)
3725 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3726 	else if (src ||
3727 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3728 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3729 
3730 	return rc;
3731 }
3732 
3733 int
3734 if_do_dad(struct ifnet *ifp)
3735 {
3736 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3737 		return 0;
3738 
3739 	switch (ifp->if_type) {
3740 	case IFT_FAITH:
3741 		/*
3742 		 * These interfaces do not have the IFF_LOOPBACK flag,
3743 		 * but loop packets back.  We do not have to do DAD on such
3744 		 * interfaces.  We should even omit it, because loop-backed
3745 		 * responses would confuse the DAD procedure.
3746 		 */
3747 		return 0;
3748 	default:
3749 		/*
3750 		 * Our DAD routine requires the interface up and running.
3751 		 * However, some interfaces can be up before the RUNNING
3752 		 * status.  Additionaly, users may try to assign addresses
3753 		 * before the interface becomes up (or running).
3754 		 * We simply skip DAD in such a case as a work around.
3755 		 * XXX: we should rather mark "tentative" on such addresses,
3756 		 * and do DAD after the interface becomes ready.
3757 		 */
3758 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
3759 		    (IFF_UP | IFF_RUNNING))
3760 			return 0;
3761 
3762 		return 1;
3763 	}
3764 }
3765 
3766 int
3767 if_flags_set(ifnet_t *ifp, const u_short flags)
3768 {
3769 	int rc;
3770 
3771 	KASSERT(IFNET_LOCKED(ifp));
3772 
3773 	if (ifp->if_setflags != NULL)
3774 		rc = (*ifp->if_setflags)(ifp, flags);
3775 	else {
3776 		u_short cantflags, chgdflags;
3777 		struct ifreq ifr;
3778 
3779 		chgdflags = ifp->if_flags ^ flags;
3780 		cantflags = chgdflags & IFF_CANTCHANGE;
3781 
3782 		if (cantflags != 0)
3783 			ifp->if_flags ^= cantflags;
3784 
3785                 /* Traditionally, we do not call if_ioctl after
3786                  * setting/clearing only IFF_PROMISC if the interface
3787                  * isn't IFF_UP.  Uphold that tradition.
3788 		 */
3789 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3790 			return 0;
3791 
3792 		memset(&ifr, 0, sizeof(ifr));
3793 
3794 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3795 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3796 
3797 		if (rc != 0 && cantflags != 0)
3798 			ifp->if_flags ^= cantflags;
3799 	}
3800 
3801 	return rc;
3802 }
3803 
3804 int
3805 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3806 {
3807 	int rc;
3808 	struct ifreq ifr;
3809 
3810 	/*
3811 	 * XXX NOMPSAFE - this calls if_ioctl without holding IFNET_LOCK()
3812 	 * in some cases - e.g. when called from vlan/netinet/netinet6 code
3813 	 * directly rather than via doifoictl()
3814 	 */
3815 	ifreq_setaddr(cmd, &ifr, sa);
3816 	rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3817 
3818 	return rc;
3819 }
3820 
3821 static void
3822 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3823     struct ifaltq *ifq)
3824 {
3825 	const struct sysctlnode *cnode, *rnode;
3826 
3827 	if (sysctl_createv(clog, 0, NULL, &rnode,
3828 		       CTLFLAG_PERMANENT,
3829 		       CTLTYPE_NODE, "interfaces",
3830 		       SYSCTL_DESCR("Per-interface controls"),
3831 		       NULL, 0, NULL, 0,
3832 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3833 		goto bad;
3834 
3835 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3836 		       CTLFLAG_PERMANENT,
3837 		       CTLTYPE_NODE, ifname,
3838 		       SYSCTL_DESCR("Interface controls"),
3839 		       NULL, 0, NULL, 0,
3840 		       CTL_CREATE, CTL_EOL) != 0)
3841 		goto bad;
3842 
3843 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3844 		       CTLFLAG_PERMANENT,
3845 		       CTLTYPE_NODE, "sndq",
3846 		       SYSCTL_DESCR("Interface output queue controls"),
3847 		       NULL, 0, NULL, 0,
3848 		       CTL_CREATE, CTL_EOL) != 0)
3849 		goto bad;
3850 
3851 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3852 		       CTLFLAG_PERMANENT,
3853 		       CTLTYPE_INT, "len",
3854 		       SYSCTL_DESCR("Current output queue length"),
3855 		       NULL, 0, &ifq->ifq_len, 0,
3856 		       CTL_CREATE, CTL_EOL) != 0)
3857 		goto bad;
3858 
3859 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3860 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3861 		       CTLTYPE_INT, "maxlen",
3862 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3863 		       NULL, 0, &ifq->ifq_maxlen, 0,
3864 		       CTL_CREATE, CTL_EOL) != 0)
3865 		goto bad;
3866 
3867 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3868 		       CTLFLAG_PERMANENT,
3869 		       CTLTYPE_INT, "drops",
3870 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3871 		       NULL, 0, &ifq->ifq_drops, 0,
3872 		       CTL_CREATE, CTL_EOL) != 0)
3873 		goto bad;
3874 
3875 	return;
3876 bad:
3877 	printf("%s: could not attach sysctl nodes\n", ifname);
3878 	return;
3879 }
3880 
3881 #if defined(INET) || defined(INET6)
3882 
3883 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3884 	static int							\
3885 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3886 	{								\
3887 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3888 	}
3889 
3890 #if defined(INET)
3891 static int
3892 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3893 {
3894 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3895 }
3896 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3897 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3898 #endif
3899 
3900 #if defined(INET6)
3901 static int
3902 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3903 {
3904 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3905 }
3906 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3907 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3908 #endif
3909 
3910 static void
3911 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3912 {
3913 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3914 	const char *pfname = NULL, *ipname = NULL;
3915 	int ipn = 0, qid = 0;
3916 
3917 	switch (pf) {
3918 #if defined(INET)
3919 	case PF_INET:
3920 		len_func = sysctl_net_ip_pktq_items;
3921 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3922 		drops_func = sysctl_net_ip_pktq_drops;
3923 		pfname = "inet", ipn = IPPROTO_IP;
3924 		ipname = "ip", qid = IPCTL_IFQ;
3925 		break;
3926 #endif
3927 #if defined(INET6)
3928 	case PF_INET6:
3929 		len_func = sysctl_net_ip6_pktq_items;
3930 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3931 		drops_func = sysctl_net_ip6_pktq_drops;
3932 		pfname = "inet6", ipn = IPPROTO_IPV6;
3933 		ipname = "ip6", qid = IPV6CTL_IFQ;
3934 		break;
3935 #endif
3936 	default:
3937 		KASSERT(false);
3938 	}
3939 
3940 	sysctl_createv(clog, 0, NULL, NULL,
3941 		       CTLFLAG_PERMANENT,
3942 		       CTLTYPE_NODE, pfname, NULL,
3943 		       NULL, 0, NULL, 0,
3944 		       CTL_NET, pf, CTL_EOL);
3945 	sysctl_createv(clog, 0, NULL, NULL,
3946 		       CTLFLAG_PERMANENT,
3947 		       CTLTYPE_NODE, ipname, NULL,
3948 		       NULL, 0, NULL, 0,
3949 		       CTL_NET, pf, ipn, CTL_EOL);
3950 	sysctl_createv(clog, 0, NULL, NULL,
3951 		       CTLFLAG_PERMANENT,
3952 		       CTLTYPE_NODE, "ifq",
3953 		       SYSCTL_DESCR("Protocol input queue controls"),
3954 		       NULL, 0, NULL, 0,
3955 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3956 
3957 	sysctl_createv(clog, 0, NULL, NULL,
3958 		       CTLFLAG_PERMANENT,
3959 		       CTLTYPE_QUAD, "len",
3960 		       SYSCTL_DESCR("Current input queue length"),
3961 		       len_func, 0, NULL, 0,
3962 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3963 	sysctl_createv(clog, 0, NULL, NULL,
3964 		       CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
3965 		       CTLTYPE_INT, "maxlen",
3966 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3967 		       maxlen_func, 0, NULL, 0,
3968 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3969 	sysctl_createv(clog, 0, NULL, NULL,
3970 		       CTLFLAG_PERMANENT,
3971 		       CTLTYPE_QUAD, "drops",
3972 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3973 		       drops_func, 0, NULL, 0,
3974 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3975 }
3976 #endif /* INET || INET6 */
3977 
3978 static int
3979 if_sdl_sysctl(SYSCTLFN_ARGS)
3980 {
3981 	struct ifnet *ifp;
3982 	const struct sockaddr_dl *sdl;
3983 	struct psref psref;
3984 	int error = 0;
3985 	int bound;
3986 
3987 	if (namelen != 1)
3988 		return EINVAL;
3989 
3990 	bound = curlwp_bind();
3991 	ifp = if_get_byindex(name[0], &psref);
3992 	if (ifp == NULL) {
3993 		error = ENODEV;
3994 		goto out0;
3995 	}
3996 
3997 	sdl = ifp->if_sadl;
3998 	if (sdl == NULL) {
3999 		*oldlenp = 0;
4000 		goto out1;
4001 	}
4002 
4003 	if (oldp == NULL) {
4004 		*oldlenp = sdl->sdl_alen;
4005 		goto out1;
4006 	}
4007 
4008 	if (*oldlenp >= sdl->sdl_alen)
4009 		*oldlenp = sdl->sdl_alen;
4010 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
4011 out1:
4012 	if_put(ifp, &psref);
4013 out0:
4014 	curlwp_bindx(bound);
4015 	return error;
4016 }
4017 
4018 static void
4019 if_sysctl_setup(struct sysctllog **clog)
4020 {
4021 	const struct sysctlnode *rnode = NULL;
4022 
4023 	sysctl_createv(clog, 0, NULL, &rnode,
4024 		       CTLFLAG_PERMANENT,
4025 		       CTLTYPE_NODE, "sdl",
4026 		       SYSCTL_DESCR("Get active link-layer address"),
4027 		       if_sdl_sysctl, 0, NULL, 0,
4028 		       CTL_NET, CTL_CREATE, CTL_EOL);
4029 
4030 #if defined(INET)
4031 	sysctl_net_pktq_setup(NULL, PF_INET);
4032 #endif
4033 #ifdef INET6
4034 	if (in6_present)
4035 		sysctl_net_pktq_setup(NULL, PF_INET6);
4036 #endif
4037 }
4038