xref: /netbsd-src/sys/net/if.c (revision d909946ca08dceb44d7d0f22ec9488679695d976)
1 /*	$NetBSD: if.c,v 1.357 2016/08/01 03:15:30 ozaki-r 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 1.357 2016/08/01 03:15:30 ozaki-r Exp $");
94 
95 #if defined(_KERNEL_OPT)
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 
99 #include "opt_atalk.h"
100 #include "opt_natm.h"
101 #include "opt_wlan.h"
102 #include "opt_net_mpsafe.h"
103 #endif
104 
105 #include <sys/param.h>
106 #include <sys/mbuf.h>
107 #include <sys/systm.h>
108 #include <sys/callout.h>
109 #include <sys/proc.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/domain.h>
113 #include <sys/protosw.h>
114 #include <sys/kernel.h>
115 #include <sys/ioctl.h>
116 #include <sys/sysctl.h>
117 #include <sys/syslog.h>
118 #include <sys/kauth.h>
119 #include <sys/kmem.h>
120 #include <sys/xcall.h>
121 #include <sys/cpu.h>
122 #include <sys/intr.h>
123 
124 #include <net/if.h>
125 #include <net/if_dl.h>
126 #include <net/if_ether.h>
127 #include <net/if_media.h>
128 #include <net80211/ieee80211.h>
129 #include <net80211/ieee80211_ioctl.h>
130 #include <net/if_types.h>
131 #include <net/route.h>
132 #include <net/netisr.h>
133 #include <sys/module.h>
134 #ifdef NETATALK
135 #include <netatalk/at_extern.h>
136 #include <netatalk/at.h>
137 #endif
138 #include <net/pfil.h>
139 #include <netinet/in.h>
140 #include <netinet/in_var.h>
141 #ifndef IPSEC
142 #include <netinet/ip_encap.h>
143 #endif
144 
145 #ifdef INET6
146 #include <netinet6/in6_var.h>
147 #include <netinet6/nd6.h>
148 #endif
149 
150 #include "ether.h"
151 #include "fddi.h"
152 #include "token.h"
153 
154 #include "carp.h"
155 #if NCARP > 0
156 #include <netinet/ip_carp.h>
157 #endif
158 
159 #include <compat/sys/sockio.h>
160 #include <compat/sys/socket.h>
161 
162 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
163 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
164 
165 /*
166  * Global list of interfaces.
167  */
168 /* DEPRECATED. Remove it once kvm(3) users disappeared */
169 struct ifnet_head		ifnet_list;
170 
171 struct pslist_head		ifnet_pslist;
172 static ifnet_t **		ifindex2ifnet = NULL;
173 static u_int			if_index = 1;
174 static size_t			if_indexlim = 0;
175 static uint64_t			index_gen;
176 /* Mutex to protect the above objects. */
177 kmutex_t			ifnet_mtx __cacheline_aligned;
178 struct psref_class		*ifnet_psref_class __read_mostly;
179 static pserialize_t		ifnet_psz;
180 
181 static kmutex_t			if_clone_mtx;
182 
183 struct ifnet *lo0ifp;
184 int	ifqmaxlen = IFQ_MAXLEN;
185 
186 struct psref_class		*ifa_psref_class __read_mostly;
187 
188 static int	if_rt_walktree(struct rtentry *, void *);
189 
190 static struct if_clone *if_clone_lookup(const char *, int *);
191 
192 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
193 static int if_cloners_count;
194 
195 /* Packet filtering hook for interfaces. */
196 pfil_head_t *	if_pfil;
197 
198 static kauth_listener_t if_listener;
199 
200 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
201 static void if_detach_queues(struct ifnet *, struct ifqueue *);
202 static void sysctl_sndq_setup(struct sysctllog **, const char *,
203     struct ifaltq *);
204 static void if_slowtimo(void *);
205 static void if_free_sadl(struct ifnet *);
206 static void if_attachdomain1(struct ifnet *);
207 static int ifconf(u_long, void *);
208 static int if_transmit(struct ifnet *, struct mbuf *);
209 static int if_clone_create(const char *);
210 static int if_clone_destroy(const char *);
211 static void if_link_state_change_si(void *);
212 
213 struct if_percpuq {
214 	struct ifnet	*ipq_ifp;
215 	void		*ipq_si;
216 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
217 };
218 
219 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
220 
221 static void if_percpuq_drops(void *, void *, struct cpu_info *);
222 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
223 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
224     struct if_percpuq *);
225 
226 #if defined(INET) || defined(INET6)
227 static void sysctl_net_pktq_setup(struct sysctllog **, int);
228 #endif
229 
230 static int
231 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
232     void *arg0, void *arg1, void *arg2, void *arg3)
233 {
234 	int result;
235 	enum kauth_network_req req;
236 
237 	result = KAUTH_RESULT_DEFER;
238 	req = (enum kauth_network_req)arg1;
239 
240 	if (action != KAUTH_NETWORK_INTERFACE)
241 		return result;
242 
243 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
244 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
245 		result = KAUTH_RESULT_ALLOW;
246 
247 	return result;
248 }
249 
250 /*
251  * Network interface utility routines.
252  *
253  * Routines with ifa_ifwith* names take sockaddr *'s as
254  * parameters.
255  */
256 void
257 ifinit(void)
258 {
259 #if defined(INET)
260 	sysctl_net_pktq_setup(NULL, PF_INET);
261 #endif
262 #ifdef INET6
263 	if (in6_present)
264 		sysctl_net_pktq_setup(NULL, PF_INET6);
265 #endif
266 
267 #if (defined(INET) || defined(INET6)) && !defined(IPSEC)
268 	encapinit();
269 #endif
270 
271 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
272 	    if_listener_cb, NULL);
273 
274 	/* interfaces are available, inform socket code */
275 	ifioctl = doifioctl;
276 }
277 
278 /*
279  * XXX Initialization before configure().
280  * XXX hack to get pfil_add_hook working in autoconf.
281  */
282 void
283 ifinit1(void)
284 {
285 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
286 
287 	TAILQ_INIT(&ifnet_list);
288 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
289 	ifnet_psz = pserialize_create();
290 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
291 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
292 	PSLIST_INIT(&ifnet_pslist);
293 
294 	if_indexlim = 8;
295 
296 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
297 	KASSERT(if_pfil != NULL);
298 
299 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
300 	etherinit();
301 #endif
302 }
303 
304 ifnet_t *
305 if_alloc(u_char type)
306 {
307 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
308 }
309 
310 void
311 if_free(ifnet_t *ifp)
312 {
313 	kmem_free(ifp, sizeof(ifnet_t));
314 }
315 
316 void
317 if_initname(struct ifnet *ifp, const char *name, int unit)
318 {
319 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
320 	    "%s%d", name, unit);
321 }
322 
323 /*
324  * Null routines used while an interface is going away.  These routines
325  * just return an error.
326  */
327 
328 int
329 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
330     const struct sockaddr *so, const struct rtentry *rt)
331 {
332 
333 	return ENXIO;
334 }
335 
336 void
337 if_nullinput(struct ifnet *ifp, struct mbuf *m)
338 {
339 
340 	/* Nothing. */
341 }
342 
343 void
344 if_nullstart(struct ifnet *ifp)
345 {
346 
347 	/* Nothing. */
348 }
349 
350 int
351 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
352 {
353 
354 	return ENXIO;
355 }
356 
357 int
358 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
359 {
360 
361 	return ENXIO;
362 }
363 
364 int
365 if_nullinit(struct ifnet *ifp)
366 {
367 
368 	return ENXIO;
369 }
370 
371 void
372 if_nullstop(struct ifnet *ifp, int disable)
373 {
374 
375 	/* Nothing. */
376 }
377 
378 void
379 if_nullslowtimo(struct ifnet *ifp)
380 {
381 
382 	/* Nothing. */
383 }
384 
385 void
386 if_nulldrain(struct ifnet *ifp)
387 {
388 
389 	/* Nothing. */
390 }
391 
392 void
393 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
394 {
395 	struct ifaddr *ifa;
396 	struct sockaddr_dl *sdl;
397 
398 	ifp->if_addrlen = addrlen;
399 	if_alloc_sadl(ifp);
400 	ifa = ifp->if_dl;
401 	sdl = satosdl(ifa->ifa_addr);
402 
403 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
404 	if (factory) {
405 		ifp->if_hwdl = ifp->if_dl;
406 		ifaref(ifp->if_hwdl);
407 	}
408 	/* TBD routing socket */
409 }
410 
411 struct ifaddr *
412 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
413 {
414 	unsigned socksize, ifasize;
415 	int addrlen, namelen;
416 	struct sockaddr_dl *mask, *sdl;
417 	struct ifaddr *ifa;
418 
419 	namelen = strlen(ifp->if_xname);
420 	addrlen = ifp->if_addrlen;
421 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
422 	ifasize = sizeof(*ifa) + 2 * socksize;
423 	ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
424 
425 	sdl = (struct sockaddr_dl *)(ifa + 1);
426 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
427 
428 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
429 	    ifp->if_xname, namelen, NULL, addrlen);
430 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
431 	memset(&mask->sdl_data[0], 0xff, namelen);
432 	ifa->ifa_rtrequest = link_rtrequest;
433 	ifa->ifa_addr = (struct sockaddr *)sdl;
434 	ifa->ifa_netmask = (struct sockaddr *)mask;
435 	ifa_psref_init(ifa);
436 
437 	*sdlp = sdl;
438 
439 	return ifa;
440 }
441 
442 static void
443 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
444 {
445 	const struct sockaddr_dl *sdl;
446 
447 	ifp->if_dl = ifa;
448 	ifaref(ifa);
449 	sdl = satosdl(ifa->ifa_addr);
450 	ifp->if_sadl = sdl;
451 }
452 
453 /*
454  * Allocate the link level name for the specified interface.  This
455  * is an attachment helper.  It must be called after ifp->if_addrlen
456  * is initialized, which may not be the case when if_attach() is
457  * called.
458  */
459 void
460 if_alloc_sadl(struct ifnet *ifp)
461 {
462 	struct ifaddr *ifa;
463 	const struct sockaddr_dl *sdl;
464 
465 	/*
466 	 * If the interface already has a link name, release it
467 	 * now.  This is useful for interfaces that can change
468 	 * link types, and thus switch link names often.
469 	 */
470 	if (ifp->if_sadl != NULL)
471 		if_free_sadl(ifp);
472 
473 	ifa = if_dl_create(ifp, &sdl);
474 
475 	ifa_insert(ifp, ifa);
476 	if_sadl_setrefs(ifp, ifa);
477 }
478 
479 static void
480 if_deactivate_sadl(struct ifnet *ifp)
481 {
482 	struct ifaddr *ifa;
483 
484 	KASSERT(ifp->if_dl != NULL);
485 
486 	ifa = ifp->if_dl;
487 
488 	ifp->if_sadl = NULL;
489 
490 	ifp->if_dl = NULL;
491 	ifafree(ifa);
492 }
493 
494 void
495 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
496     const struct sockaddr_dl *sdl)
497 {
498 	int s, ss;
499 	struct ifaddr *ifa;
500 	int bound = curlwp_bind();
501 
502 	s = splnet();
503 
504 	if_deactivate_sadl(ifp);
505 
506 	if_sadl_setrefs(ifp, ifa0);
507 
508 	ss = pserialize_read_enter();
509 	IFADDR_READER_FOREACH(ifa, ifp) {
510 		struct psref psref;
511 		ifa_acquire(ifa, &psref);
512 		pserialize_read_exit(ss);
513 
514 		rtinit(ifa, RTM_LLINFO_UPD, 0);
515 
516 		ss = pserialize_read_enter();
517 		ifa_release(ifa, &psref);
518 	}
519 	pserialize_read_exit(ss);
520 
521 	splx(s);
522 	curlwp_bindx(bound);
523 }
524 
525 /*
526  * Free the link level name for the specified interface.  This is
527  * a detach helper.  This is called from if_detach().
528  */
529 static void
530 if_free_sadl(struct ifnet *ifp)
531 {
532 	struct ifaddr *ifa;
533 	int s;
534 
535 	ifa = ifp->if_dl;
536 	if (ifa == NULL) {
537 		KASSERT(ifp->if_sadl == NULL);
538 		return;
539 	}
540 
541 	KASSERT(ifp->if_sadl != NULL);
542 
543 	s = splnet();
544 	rtinit(ifa, RTM_DELETE, 0);
545 	ifa_remove(ifp, ifa);
546 	if_deactivate_sadl(ifp);
547 	if (ifp->if_hwdl == ifa) {
548 		ifafree(ifa);
549 		ifp->if_hwdl = NULL;
550 	}
551 	splx(s);
552 }
553 
554 static void
555 if_getindex(ifnet_t *ifp)
556 {
557 	bool hitlimit = false;
558 
559 	ifp->if_index_gen = index_gen++;
560 
561 	ifp->if_index = if_index;
562 	if (ifindex2ifnet == NULL) {
563 		if_index++;
564 		goto skip;
565 	}
566 	while (if_byindex(ifp->if_index)) {
567 		/*
568 		 * If we hit USHRT_MAX, we skip back to 0 since
569 		 * there are a number of places where the value
570 		 * of if_index or if_index itself is compared
571 		 * to or stored in an unsigned short.  By
572 		 * jumping back, we won't botch those assignments
573 		 * or comparisons.
574 		 */
575 		if (++if_index == 0) {
576 			if_index = 1;
577 		} else if (if_index == USHRT_MAX) {
578 			/*
579 			 * However, if we have to jump back to
580 			 * zero *twice* without finding an empty
581 			 * slot in ifindex2ifnet[], then there
582 			 * there are too many (>65535) interfaces.
583 			 */
584 			if (hitlimit) {
585 				panic("too many interfaces");
586 			}
587 			hitlimit = true;
588 			if_index = 1;
589 		}
590 		ifp->if_index = if_index;
591 	}
592 skip:
593 	/*
594 	 * ifindex2ifnet is indexed by if_index. Since if_index will
595 	 * grow dynamically, it should grow too.
596 	 */
597 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
598 		size_t m, n, oldlim;
599 		void *q;
600 
601 		oldlim = if_indexlim;
602 		while (ifp->if_index >= if_indexlim)
603 			if_indexlim <<= 1;
604 
605 		/* grow ifindex2ifnet */
606 		m = oldlim * sizeof(struct ifnet *);
607 		n = if_indexlim * sizeof(struct ifnet *);
608 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
609 		if (ifindex2ifnet != NULL) {
610 			memcpy(q, ifindex2ifnet, m);
611 			free(ifindex2ifnet, M_IFADDR);
612 		}
613 		ifindex2ifnet = (struct ifnet **)q;
614 	}
615 	ifindex2ifnet[ifp->if_index] = ifp;
616 }
617 
618 /*
619  * Initialize an interface and assign an index for it.
620  *
621  * It must be called prior to a device specific attach routine
622  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
623  * and be followed by if_register:
624  *
625  *     if_initialize(ifp);
626  *     ether_ifattach(ifp, enaddr);
627  *     if_register(ifp);
628  */
629 void
630 if_initialize(ifnet_t *ifp)
631 {
632 	KASSERT(if_indexlim > 0);
633 	TAILQ_INIT(&ifp->if_addrlist);
634 
635 	/*
636 	 * Link level name is allocated later by a separate call to
637 	 * if_alloc_sadl().
638 	 */
639 
640 	if (ifp->if_snd.ifq_maxlen == 0)
641 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
642 
643 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
644 
645 	ifp->if_link_state = LINK_STATE_UNKNOWN;
646 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
647 
648 	ifp->if_capenable = 0;
649 	ifp->if_csum_flags_tx = 0;
650 	ifp->if_csum_flags_rx = 0;
651 
652 #ifdef ALTQ
653 	ifp->if_snd.altq_type = 0;
654 	ifp->if_snd.altq_disc = NULL;
655 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
656 	ifp->if_snd.altq_tbr  = NULL;
657 	ifp->if_snd.altq_ifp  = ifp;
658 #endif
659 
660 	ifp->if_snd.ifq_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET);
661 
662 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
663 	(void)pfil_run_hooks(if_pfil,
664 	    (struct mbuf **)PFIL_IFNET_ATTACH, ifp, PFIL_IFNET);
665 
666 	IF_AFDATA_LOCK_INIT(ifp);
667 
668 	if (if_is_link_state_changeable(ifp)) {
669 		ifp->if_link_si = softint_establish(SOFTINT_NET,
670 		    if_link_state_change_si, ifp);
671 		if (ifp->if_link_si == NULL)
672 			panic("%s: softint_establish() failed", __func__);
673 	}
674 
675 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
676 	PSLIST_INIT(&ifp->if_addr_pslist);
677 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
678 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
679 
680 	IFNET_LOCK();
681 	if_getindex(ifp);
682 	IFNET_UNLOCK();
683 }
684 
685 /*
686  * Register an interface to the list of "active" interfaces.
687  */
688 void
689 if_register(ifnet_t *ifp)
690 {
691 	/*
692 	 * If the driver has not supplied its own if_ioctl, then
693 	 * supply the default.
694 	 */
695 	if (ifp->if_ioctl == NULL)
696 		ifp->if_ioctl = ifioctl_common;
697 
698 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
699 
700 	if (!STAILQ_EMPTY(&domains))
701 		if_attachdomain1(ifp);
702 
703 	/* Announce the interface. */
704 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
705 
706 	if (ifp->if_slowtimo != NULL) {
707 		ifp->if_slowtimo_ch =
708 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
709 		callout_init(ifp->if_slowtimo_ch, 0);
710 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
711 		if_slowtimo(ifp);
712 	}
713 
714 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
715 		ifp->if_transmit = if_transmit;
716 
717 	IFNET_LOCK();
718 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
719 	IFNET_WRITER_INSERT_TAIL(ifp);
720 	IFNET_UNLOCK();
721 }
722 
723 /*
724  * The if_percpuq framework
725  *
726  * It allows network device drivers to execute the network stack
727  * in softint (so called softint-based if_input). It utilizes
728  * softint and percpu ifqueue. It doesn't distribute any packets
729  * between CPUs, unlike pktqueue(9).
730  *
731  * Currently we support two options for device drivers to apply the framework:
732  * - Use it implicitly with less changes
733  *   - If you use if_attach in driver's _attach function and if_input in
734  *     driver's Rx interrupt handler, a packet is queued and a softint handles
735  *     the packet implicitly
736  * - Use it explicitly in each driver (recommended)
737  *   - You can use if_percpuq_* directly in your driver
738  *   - In this case, you need to allocate struct if_percpuq in driver's softc
739  *   - See wm(4) as a reference implementation
740  */
741 
742 static void
743 if_percpuq_softint(void *arg)
744 {
745 	struct if_percpuq *ipq = arg;
746 	struct ifnet *ifp = ipq->ipq_ifp;
747 	struct mbuf *m;
748 
749 	while ((m = if_percpuq_dequeue(ipq)) != NULL)
750 		ifp->_if_input(ifp, m);
751 }
752 
753 static void
754 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
755 {
756 	struct ifqueue *const ifq = p;
757 
758 	memset(ifq, 0, sizeof(*ifq));
759 	ifq->ifq_maxlen = IFQ_MAXLEN;
760 }
761 
762 struct if_percpuq *
763 if_percpuq_create(struct ifnet *ifp)
764 {
765 	struct if_percpuq *ipq;
766 
767 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
768 	if (ipq == NULL)
769 		panic("kmem_zalloc failed");
770 
771 	ipq->ipq_ifp = ifp;
772 	ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
773 	    if_percpuq_softint, ipq);
774 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
775 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
776 
777 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
778 
779 	return ipq;
780 }
781 
782 static struct mbuf *
783 if_percpuq_dequeue(struct if_percpuq *ipq)
784 {
785 	struct mbuf *m;
786 	struct ifqueue *ifq;
787 	int s;
788 
789 	s = splnet();
790 	ifq = percpu_getref(ipq->ipq_ifqs);
791 	IF_DEQUEUE(ifq, m);
792 	percpu_putref(ipq->ipq_ifqs);
793 	splx(s);
794 
795 	return m;
796 }
797 
798 static void
799 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
800 {
801 	struct ifqueue *const ifq = p;
802 
803 	IF_PURGE(ifq);
804 }
805 
806 void
807 if_percpuq_destroy(struct if_percpuq *ipq)
808 {
809 
810 	/* if_detach may already destroy it */
811 	if (ipq == NULL)
812 		return;
813 
814 	softint_disestablish(ipq->ipq_si);
815 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
816 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
817 }
818 
819 void
820 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
821 {
822 	struct ifqueue *ifq;
823 	int s;
824 
825 	KASSERT(ipq != NULL);
826 
827 	s = splnet();
828 	ifq = percpu_getref(ipq->ipq_ifqs);
829 	if (IF_QFULL(ifq)) {
830 		IF_DROP(ifq);
831 		percpu_putref(ipq->ipq_ifqs);
832 		m_freem(m);
833 		goto out;
834 	}
835 	IF_ENQUEUE(ifq, m);
836 	percpu_putref(ipq->ipq_ifqs);
837 
838 	softint_schedule(ipq->ipq_si);
839 out:
840 	splx(s);
841 }
842 
843 static void
844 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
845 {
846 	struct ifqueue *const ifq = p;
847 	int *sum = arg;
848 
849 	*sum += ifq->ifq_drops;
850 }
851 
852 static int
853 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
854 {
855 	struct sysctlnode node;
856 	struct if_percpuq *ipq;
857 	int sum = 0;
858 	int error;
859 
860 	node = *rnode;
861 	ipq = node.sysctl_data;
862 
863 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
864 
865 	node.sysctl_data = &sum;
866 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
867 	if (error != 0 || newp == NULL)
868 		return error;
869 
870 	return 0;
871 }
872 
873 static void
874 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
875     struct if_percpuq *ipq)
876 {
877 	const struct sysctlnode *cnode, *rnode;
878 
879 	if (sysctl_createv(clog, 0, NULL, &rnode,
880 		       CTLFLAG_PERMANENT,
881 		       CTLTYPE_NODE, "interfaces",
882 		       SYSCTL_DESCR("Per-interface controls"),
883 		       NULL, 0, NULL, 0,
884 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
885 		goto bad;
886 
887 	if (sysctl_createv(clog, 0, &rnode, &rnode,
888 		       CTLFLAG_PERMANENT,
889 		       CTLTYPE_NODE, ifname,
890 		       SYSCTL_DESCR("Interface controls"),
891 		       NULL, 0, NULL, 0,
892 		       CTL_CREATE, CTL_EOL) != 0)
893 		goto bad;
894 
895 	if (sysctl_createv(clog, 0, &rnode, &rnode,
896 		       CTLFLAG_PERMANENT,
897 		       CTLTYPE_NODE, "rcvq",
898 		       SYSCTL_DESCR("Interface input queue controls"),
899 		       NULL, 0, NULL, 0,
900 		       CTL_CREATE, CTL_EOL) != 0)
901 		goto bad;
902 
903 #ifdef NOTYET
904 	/* XXX Should show each per-CPU queue length? */
905 	if (sysctl_createv(clog, 0, &rnode, &rnode,
906 		       CTLFLAG_PERMANENT,
907 		       CTLTYPE_INT, "len",
908 		       SYSCTL_DESCR("Current input queue length"),
909 		       sysctl_percpuq_len, 0, NULL, 0,
910 		       CTL_CREATE, CTL_EOL) != 0)
911 		goto bad;
912 
913 	if (sysctl_createv(clog, 0, &rnode, &cnode,
914 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
915 		       CTLTYPE_INT, "maxlen",
916 		       SYSCTL_DESCR("Maximum allowed input queue length"),
917 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
918 		       CTL_CREATE, CTL_EOL) != 0)
919 		goto bad;
920 #endif
921 
922 	if (sysctl_createv(clog, 0, &rnode, &cnode,
923 		       CTLFLAG_PERMANENT,
924 		       CTLTYPE_INT, "drops",
925 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
926 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
927 		       CTL_CREATE, CTL_EOL) != 0)
928 		goto bad;
929 
930 	return;
931 bad:
932 	printf("%s: could not attach sysctl nodes\n", ifname);
933 	return;
934 }
935 
936 
937 /*
938  * The common interface input routine that is called by device drivers,
939  * which should be used only when the driver's rx handler already runs
940  * in softint.
941  */
942 void
943 if_input(struct ifnet *ifp, struct mbuf *m)
944 {
945 
946 	KASSERT(ifp->if_percpuq == NULL);
947 	KASSERT(!cpu_intr_p());
948 
949 	ifp->_if_input(ifp, m);
950 }
951 
952 /*
953  * DEPRECATED. Use if_initialize and if_register instead.
954  * See the above comment of if_initialize.
955  *
956  * Note that it implicitly enables if_percpuq to make drivers easy to
957  * migrate softint-based if_input without much changes. If you don't
958  * want to enable it, use if_initialize instead.
959  */
960 void
961 if_attach(ifnet_t *ifp)
962 {
963 
964 	if_initialize(ifp);
965 	ifp->if_percpuq = if_percpuq_create(ifp);
966 	if_register(ifp);
967 }
968 
969 void
970 if_attachdomain(void)
971 {
972 	struct ifnet *ifp;
973 	int s;
974 	int bound = curlwp_bind();
975 
976 	s = pserialize_read_enter();
977 	IFNET_READER_FOREACH(ifp) {
978 		struct psref psref;
979 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
980 		pserialize_read_exit(s);
981 		if_attachdomain1(ifp);
982 		s = pserialize_read_enter();
983 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
984 	}
985 	pserialize_read_exit(s);
986 	curlwp_bindx(bound);
987 }
988 
989 static void
990 if_attachdomain1(struct ifnet *ifp)
991 {
992 	struct domain *dp;
993 	int s;
994 
995 	s = splnet();
996 
997 	/* address family dependent data region */
998 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
999 	DOMAIN_FOREACH(dp) {
1000 		if (dp->dom_ifattach != NULL)
1001 			ifp->if_afdata[dp->dom_family] =
1002 			    (*dp->dom_ifattach)(ifp);
1003 	}
1004 
1005 	splx(s);
1006 }
1007 
1008 /*
1009  * Deactivate an interface.  This points all of the procedure
1010  * handles at error stubs.  May be called from interrupt context.
1011  */
1012 void
1013 if_deactivate(struct ifnet *ifp)
1014 {
1015 	int s;
1016 
1017 	s = splnet();
1018 
1019 	ifp->if_output	 = if_nulloutput;
1020 	ifp->_if_input	 = if_nullinput;
1021 	ifp->if_start	 = if_nullstart;
1022 	ifp->if_transmit = if_nulltransmit;
1023 	ifp->if_ioctl	 = if_nullioctl;
1024 	ifp->if_init	 = if_nullinit;
1025 	ifp->if_stop	 = if_nullstop;
1026 	ifp->if_slowtimo = if_nullslowtimo;
1027 	ifp->if_drain	 = if_nulldrain;
1028 
1029 	/* No more packets may be enqueued. */
1030 	ifp->if_snd.ifq_maxlen = 0;
1031 
1032 	splx(s);
1033 }
1034 
1035 bool
1036 if_is_deactivated(struct ifnet *ifp)
1037 {
1038 
1039 	return ifp->if_output == if_nulloutput;
1040 }
1041 
1042 void
1043 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1044 {
1045 	struct ifaddr *ifa, *nifa;
1046 	int s;
1047 
1048 	s = pserialize_read_enter();
1049 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1050 		nifa = IFADDR_READER_NEXT(ifa);
1051 		if (ifa->ifa_addr->sa_family != family)
1052 			continue;
1053 		pserialize_read_exit(s);
1054 
1055 		(*purgeaddr)(ifa);
1056 
1057 		s = pserialize_read_enter();
1058 	}
1059 	pserialize_read_exit(s);
1060 }
1061 
1062 #ifdef IFAREF_DEBUG
1063 static struct ifaddr **ifa_list;
1064 static int ifa_list_size;
1065 
1066 /* Depends on only one if_attach runs at once */
1067 static void
1068 if_build_ifa_list(struct ifnet *ifp)
1069 {
1070 	struct ifaddr *ifa;
1071 	int i;
1072 
1073 	KASSERT(ifa_list == NULL);
1074 	KASSERT(ifa_list_size == 0);
1075 
1076 	IFADDR_READER_FOREACH(ifa, ifp)
1077 		ifa_list_size++;
1078 
1079 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1080 	if (ifa_list == NULL)
1081 		return;
1082 
1083 	i = 0;
1084 	IFADDR_READER_FOREACH(ifa, ifp) {
1085 		ifa_list[i++] = ifa;
1086 		ifaref(ifa);
1087 	}
1088 }
1089 
1090 static void
1091 if_check_and_free_ifa_list(struct ifnet *ifp)
1092 {
1093 	int i;
1094 	struct ifaddr *ifa;
1095 
1096 	if (ifa_list == NULL)
1097 		return;
1098 
1099 	for (i = 0; i < ifa_list_size; i++) {
1100 		char buf[64];
1101 
1102 		ifa = ifa_list[i];
1103 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1104 		if (ifa->ifa_refcnt > 1) {
1105 			log(LOG_WARNING,
1106 			    "ifa(%s) still referenced (refcnt=%d)\n",
1107 			    buf, ifa->ifa_refcnt - 1);
1108 		} else
1109 			log(LOG_DEBUG,
1110 			    "ifa(%s) not referenced (refcnt=%d)\n",
1111 			    buf, ifa->ifa_refcnt - 1);
1112 		ifafree(ifa);
1113 	}
1114 
1115 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1116 	ifa_list = NULL;
1117 	ifa_list_size = 0;
1118 }
1119 #endif
1120 
1121 /*
1122  * Detach an interface from the list of "active" interfaces,
1123  * freeing any resources as we go along.
1124  *
1125  * NOTE: This routine must be called with a valid thread context,
1126  * as it may block.
1127  */
1128 void
1129 if_detach(struct ifnet *ifp)
1130 {
1131 	struct socket so;
1132 	struct ifaddr *ifa;
1133 #ifdef IFAREF_DEBUG
1134 	struct ifaddr *last_ifa = NULL;
1135 #endif
1136 	struct domain *dp;
1137 	const struct protosw *pr;
1138 	int s, i, family, purged;
1139 	uint64_t xc;
1140 
1141 #ifdef IFAREF_DEBUG
1142 	if_build_ifa_list(ifp);
1143 #endif
1144 	/*
1145 	 * XXX It's kind of lame that we have to have the
1146 	 * XXX socket structure...
1147 	 */
1148 	memset(&so, 0, sizeof(so));
1149 
1150 	s = splnet();
1151 
1152 	sysctl_teardown(&ifp->if_sysctl_log);
1153 	mutex_enter(ifp->if_ioctl_lock);
1154 	if_deactivate(ifp);
1155 	mutex_exit(ifp->if_ioctl_lock);
1156 
1157 	IFNET_LOCK();
1158 	ifindex2ifnet[ifp->if_index] = NULL;
1159 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1160 	IFNET_WRITER_REMOVE(ifp);
1161 	pserialize_perform(ifnet_psz);
1162 	IFNET_UNLOCK();
1163 
1164 	/* Wait for all readers to drain before freeing.  */
1165 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1166 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1167 
1168 	mutex_obj_free(ifp->if_ioctl_lock);
1169 	ifp->if_ioctl_lock = NULL;
1170 
1171 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1172 		ifp->if_slowtimo = NULL;
1173 		callout_halt(ifp->if_slowtimo_ch, NULL);
1174 		callout_destroy(ifp->if_slowtimo_ch);
1175 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1176 	}
1177 
1178 	/*
1179 	 * Do an if_down() to give protocols a chance to do something.
1180 	 */
1181 	if_down(ifp);
1182 
1183 #ifdef ALTQ
1184 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1185 		altq_disable(&ifp->if_snd);
1186 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1187 		altq_detach(&ifp->if_snd);
1188 #endif
1189 
1190 	mutex_obj_free(ifp->if_snd.ifq_lock);
1191 
1192 #if NCARP > 0
1193 	/* Remove the interface from any carp group it is a part of.  */
1194 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1195 		carp_ifdetach(ifp);
1196 #endif
1197 
1198 	/*
1199 	 * Rip all the addresses off the interface.  This should make
1200 	 * all of the routes go away.
1201 	 *
1202 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1203 	 * from the list, including our "cursor", ifa.  For safety,
1204 	 * and to honor the TAILQ abstraction, I just restart the
1205 	 * loop after each removal.  Note that the loop will exit
1206 	 * when all of the remaining ifaddrs belong to the AF_LINK
1207 	 * family.  I am counting on the historical fact that at
1208 	 * least one pr_usrreq in each address domain removes at
1209 	 * least one ifaddr.
1210 	 */
1211 again:
1212 	/*
1213 	 * At this point, no other one tries to remove ifa in the list,
1214 	 * so we don't need to take a lock or psref.
1215 	 */
1216 	IFADDR_READER_FOREACH(ifa, ifp) {
1217 		family = ifa->ifa_addr->sa_family;
1218 #ifdef IFAREF_DEBUG
1219 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1220 		    ifa, family, ifa->ifa_refcnt);
1221 		if (last_ifa != NULL && ifa == last_ifa)
1222 			panic("if_detach: loop detected");
1223 		last_ifa = ifa;
1224 #endif
1225 		if (family == AF_LINK)
1226 			continue;
1227 		dp = pffinddomain(family);
1228 #ifdef DIAGNOSTIC
1229 		if (dp == NULL)
1230 			panic("if_detach: no domain for AF %d",
1231 			    family);
1232 #endif
1233 		/*
1234 		 * XXX These PURGEIF calls are redundant with the
1235 		 * purge-all-families calls below, but are left in for
1236 		 * now both to make a smaller change, and to avoid
1237 		 * unplanned interactions with clearing of
1238 		 * ifp->if_addrlist.
1239 		 */
1240 		purged = 0;
1241 		for (pr = dp->dom_protosw;
1242 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1243 			so.so_proto = pr;
1244 			if (pr->pr_usrreqs) {
1245 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1246 				purged = 1;
1247 			}
1248 		}
1249 		if (purged == 0) {
1250 			/*
1251 			 * XXX What's really the best thing to do
1252 			 * XXX here?  --thorpej@NetBSD.org
1253 			 */
1254 			printf("if_detach: WARNING: AF %d not purged\n",
1255 			    family);
1256 			ifa_remove(ifp, ifa);
1257 		}
1258 		goto again;
1259 	}
1260 
1261 	if_free_sadl(ifp);
1262 
1263 	/* Walk the routing table looking for stragglers. */
1264 	for (i = 0; i <= AF_MAX; i++) {
1265 		while (rt_walktree(i, if_rt_walktree, ifp) == ERESTART)
1266 			continue;
1267 	}
1268 
1269 	DOMAIN_FOREACH(dp) {
1270 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1271 		{
1272 			void *p = ifp->if_afdata[dp->dom_family];
1273 			if (p) {
1274 				ifp->if_afdata[dp->dom_family] = NULL;
1275 				(*dp->dom_ifdetach)(ifp, p);
1276 			}
1277 		}
1278 
1279 		/*
1280 		 * One would expect multicast memberships (INET and
1281 		 * INET6) on UDP sockets to be purged by the PURGEIF
1282 		 * calls above, but if all addresses were removed from
1283 		 * the interface prior to destruction, the calls will
1284 		 * not be made (e.g. ppp, for which pppd(8) generally
1285 		 * removes addresses before destroying the interface).
1286 		 * Because there is no invariant that multicast
1287 		 * memberships only exist for interfaces with IPv4
1288 		 * addresses, we must call PURGEIF regardless of
1289 		 * addresses.  (Protocols which might store ifnet
1290 		 * pointers are marked with PR_PURGEIF.)
1291 		 */
1292 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1293 			so.so_proto = pr;
1294 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1295 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1296 		}
1297 	}
1298 
1299 	(void)pfil_run_hooks(if_pfil,
1300 	    (struct mbuf **)PFIL_IFNET_DETACH, ifp, PFIL_IFNET);
1301 	(void)pfil_head_destroy(ifp->if_pfil);
1302 
1303 	/* Announce that the interface is gone. */
1304 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1305 
1306 	IF_AFDATA_LOCK_DESTROY(ifp);
1307 
1308 	if (if_is_link_state_changeable(ifp)) {
1309 		softint_disestablish(ifp->if_link_si);
1310 		ifp->if_link_si = NULL;
1311 	}
1312 
1313 	/*
1314 	 * remove packets that came from ifp, from software interrupt queues.
1315 	 */
1316 	DOMAIN_FOREACH(dp) {
1317 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1318 			struct ifqueue *iq = dp->dom_ifqueues[i];
1319 			if (iq == NULL)
1320 				break;
1321 			dp->dom_ifqueues[i] = NULL;
1322 			if_detach_queues(ifp, iq);
1323 		}
1324 	}
1325 
1326 	/*
1327 	 * IP queues have to be processed separately: net-queue barrier
1328 	 * ensures that the packets are dequeued while a cross-call will
1329 	 * ensure that the interrupts have completed. FIXME: not quite..
1330 	 */
1331 #ifdef INET
1332 	pktq_barrier(ip_pktq);
1333 #endif
1334 #ifdef INET6
1335 	if (in6_present)
1336 		pktq_barrier(ip6_pktq);
1337 #endif
1338 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1339 	xc_wait(xc);
1340 
1341 	if (ifp->if_percpuq != NULL) {
1342 		if_percpuq_destroy(ifp->if_percpuq);
1343 		ifp->if_percpuq = NULL;
1344 	}
1345 
1346 	splx(s);
1347 
1348 #ifdef IFAREF_DEBUG
1349 	if_check_and_free_ifa_list(ifp);
1350 #endif
1351 }
1352 
1353 static void
1354 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1355 {
1356 	struct mbuf *m, *prev, *next;
1357 
1358 	prev = NULL;
1359 	for (m = q->ifq_head; m != NULL; m = next) {
1360 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1361 
1362 		next = m->m_nextpkt;
1363 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1364 			prev = m;
1365 			continue;
1366 		}
1367 
1368 		if (prev != NULL)
1369 			prev->m_nextpkt = m->m_nextpkt;
1370 		else
1371 			q->ifq_head = m->m_nextpkt;
1372 		if (q->ifq_tail == m)
1373 			q->ifq_tail = prev;
1374 		q->ifq_len--;
1375 
1376 		m->m_nextpkt = NULL;
1377 		m_freem(m);
1378 		IF_DROP(q);
1379 	}
1380 }
1381 
1382 /*
1383  * Callback for a radix tree walk to delete all references to an
1384  * ifnet.
1385  */
1386 static int
1387 if_rt_walktree(struct rtentry *rt, void *v)
1388 {
1389 	struct ifnet *ifp = (struct ifnet *)v;
1390 	int error;
1391 	struct rtentry *retrt;
1392 
1393 	if (rt->rt_ifp != ifp)
1394 		return 0;
1395 
1396 	/* Delete the entry. */
1397 	error = rtrequest(RTM_DELETE, rt_getkey(rt), rt->rt_gateway,
1398 	    rt_mask(rt), rt->rt_flags, &retrt);
1399 	if (error == 0) {
1400 		KASSERT(retrt == rt);
1401 		KASSERT((retrt->rt_flags & RTF_UP) == 0);
1402 		retrt->rt_ifp = NULL;
1403 		rtfree(retrt);
1404 	} else {
1405 		printf("%s: warning: unable to delete rtentry @ %p, "
1406 		    "error = %d\n", ifp->if_xname, rt, error);
1407 	}
1408 	return ERESTART;
1409 }
1410 
1411 /*
1412  * Create a clone network interface.
1413  */
1414 static int
1415 if_clone_create(const char *name)
1416 {
1417 	struct if_clone *ifc;
1418 	int unit;
1419 	struct ifnet *ifp;
1420 	struct psref psref;
1421 
1422 	ifc = if_clone_lookup(name, &unit);
1423 	if (ifc == NULL)
1424 		return EINVAL;
1425 
1426 	ifp = if_get(name, &psref);
1427 	if (ifp != NULL) {
1428 		if_put(ifp, &psref);
1429 		return EEXIST;
1430 	}
1431 
1432 	return (*ifc->ifc_create)(ifc, unit);
1433 }
1434 
1435 /*
1436  * Destroy a clone network interface.
1437  */
1438 static int
1439 if_clone_destroy(const char *name)
1440 {
1441 	struct if_clone *ifc;
1442 	struct ifnet *ifp;
1443 	struct psref psref;
1444 
1445 	ifc = if_clone_lookup(name, NULL);
1446 	if (ifc == NULL)
1447 		return EINVAL;
1448 
1449 	if (ifc->ifc_destroy == NULL)
1450 		return EOPNOTSUPP;
1451 
1452 	ifp = if_get(name, &psref);
1453 	if (ifp == NULL)
1454 		return ENXIO;
1455 
1456 	/* We have to disable ioctls here */
1457 	mutex_enter(ifp->if_ioctl_lock);
1458 	ifp->if_ioctl = if_nullioctl;
1459 	mutex_exit(ifp->if_ioctl_lock);
1460 
1461 	/*
1462 	 * We cannot call ifc_destroy with holding ifp.
1463 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1464 	 */
1465 	if_put(ifp, &psref);
1466 
1467 	return (*ifc->ifc_destroy)(ifp);
1468 }
1469 
1470 /*
1471  * Look up a network interface cloner.
1472  */
1473 static struct if_clone *
1474 if_clone_lookup(const char *name, int *unitp)
1475 {
1476 	struct if_clone *ifc;
1477 	const char *cp;
1478 	char *dp, ifname[IFNAMSIZ + 3];
1479 	int unit;
1480 
1481 	strcpy(ifname, "if_");
1482 	/* separate interface name from unit */
1483 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1484 	    *cp && (*cp < '0' || *cp > '9');)
1485 		*dp++ = *cp++;
1486 
1487 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1488 		return NULL;	/* No name or unit number */
1489 	*dp++ = '\0';
1490 
1491 again:
1492 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1493 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1494 			break;
1495 	}
1496 
1497 	if (ifc == NULL) {
1498 		if (*ifname == '\0' ||
1499 		    module_autoload(ifname, MODULE_CLASS_DRIVER))
1500 			return NULL;
1501 		*ifname = '\0';
1502 		goto again;
1503 	}
1504 
1505 	unit = 0;
1506 	while (cp - name < IFNAMSIZ && *cp) {
1507 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1508 			/* Bogus unit number. */
1509 			return NULL;
1510 		}
1511 		unit = (unit * 10) + (*cp++ - '0');
1512 	}
1513 
1514 	if (unitp != NULL)
1515 		*unitp = unit;
1516 	return ifc;
1517 }
1518 
1519 /*
1520  * Register a network interface cloner.
1521  */
1522 void
1523 if_clone_attach(struct if_clone *ifc)
1524 {
1525 
1526 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1527 	if_cloners_count++;
1528 }
1529 
1530 /*
1531  * Unregister a network interface cloner.
1532  */
1533 void
1534 if_clone_detach(struct if_clone *ifc)
1535 {
1536 
1537 	LIST_REMOVE(ifc, ifc_list);
1538 	if_cloners_count--;
1539 }
1540 
1541 /*
1542  * Provide list of interface cloners to userspace.
1543  */
1544 int
1545 if_clone_list(int buf_count, char *buffer, int *total)
1546 {
1547 	char outbuf[IFNAMSIZ], *dst;
1548 	struct if_clone *ifc;
1549 	int count, error = 0;
1550 
1551 	*total = if_cloners_count;
1552 	if ((dst = buffer) == NULL) {
1553 		/* Just asking how many there are. */
1554 		return 0;
1555 	}
1556 
1557 	if (buf_count < 0)
1558 		return EINVAL;
1559 
1560 	count = (if_cloners_count < buf_count) ?
1561 	    if_cloners_count : buf_count;
1562 
1563 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1564 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1565 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1566 		if (outbuf[sizeof(outbuf) - 1] != '\0')
1567 			return ENAMETOOLONG;
1568 		error = copyout(outbuf, dst, sizeof(outbuf));
1569 		if (error != 0)
1570 			break;
1571 	}
1572 
1573 	return error;
1574 }
1575 
1576 void
1577 ifa_psref_init(struct ifaddr *ifa)
1578 {
1579 
1580 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1581 }
1582 
1583 void
1584 ifaref(struct ifaddr *ifa)
1585 {
1586 	ifa->ifa_refcnt++;
1587 }
1588 
1589 void
1590 ifafree(struct ifaddr *ifa)
1591 {
1592 	KASSERT(ifa != NULL);
1593 	KASSERT(ifa->ifa_refcnt > 0);
1594 
1595 	if (--ifa->ifa_refcnt == 0) {
1596 		free(ifa, M_IFADDR);
1597 	}
1598 }
1599 
1600 void
1601 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1602 {
1603 
1604 	ifa->ifa_ifp = ifp;
1605 
1606 	IFNET_LOCK();
1607 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1608 	IFADDR_ENTRY_INIT(ifa);
1609 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1610 	IFNET_UNLOCK();
1611 
1612 	ifaref(ifa);
1613 }
1614 
1615 void
1616 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1617 {
1618 
1619 	KASSERT(ifa->ifa_ifp == ifp);
1620 
1621 	IFNET_LOCK();
1622 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1623 	IFADDR_WRITER_REMOVE(ifa);
1624 	IFADDR_ENTRY_DESTROY(ifa);
1625 #if notyet
1626 	pserialize_perform(ifnet_psz);
1627 #endif
1628 	IFNET_UNLOCK();
1629 
1630 #if notyet
1631 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1632 #endif
1633 	ifafree(ifa);
1634 }
1635 
1636 void
1637 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1638 {
1639 
1640 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1641 }
1642 
1643 void
1644 ifa_release(struct ifaddr *ifa, struct psref *psref)
1645 {
1646 
1647 	if (ifa == NULL)
1648 		return;
1649 
1650 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1651 }
1652 
1653 bool
1654 ifa_held(struct ifaddr *ifa)
1655 {
1656 
1657 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1658 }
1659 
1660 static inline int
1661 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1662 {
1663 	return sockaddr_cmp(sa1, sa2) == 0;
1664 }
1665 
1666 /*
1667  * Locate an interface based on a complete address.
1668  */
1669 /*ARGSUSED*/
1670 struct ifaddr *
1671 ifa_ifwithaddr(const struct sockaddr *addr)
1672 {
1673 	struct ifnet *ifp;
1674 	struct ifaddr *ifa;
1675 
1676 	IFNET_READER_FOREACH(ifp) {
1677 		if (if_is_deactivated(ifp))
1678 			continue;
1679 		IFADDR_READER_FOREACH(ifa, ifp) {
1680 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1681 				continue;
1682 			if (equal(addr, ifa->ifa_addr))
1683 				return ifa;
1684 			if ((ifp->if_flags & IFF_BROADCAST) &&
1685 			    ifa->ifa_broadaddr &&
1686 			    /* IP6 doesn't have broadcast */
1687 			    ifa->ifa_broadaddr->sa_len != 0 &&
1688 			    equal(ifa->ifa_broadaddr, addr))
1689 				return ifa;
1690 		}
1691 	}
1692 	return NULL;
1693 }
1694 
1695 struct ifaddr *
1696 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1697 {
1698 	struct ifaddr *ifa;
1699 	int s = pserialize_read_enter();
1700 
1701 	ifa = ifa_ifwithaddr(addr);
1702 	if (ifa != NULL)
1703 		ifa_acquire(ifa, psref);
1704 	pserialize_read_exit(s);
1705 
1706 	return ifa;
1707 }
1708 
1709 /*
1710  * Locate the point to point interface with a given destination address.
1711  */
1712 /*ARGSUSED*/
1713 struct ifaddr *
1714 ifa_ifwithdstaddr(const struct sockaddr *addr)
1715 {
1716 	struct ifnet *ifp;
1717 	struct ifaddr *ifa;
1718 
1719 	IFNET_READER_FOREACH(ifp) {
1720 		if (if_is_deactivated(ifp))
1721 			continue;
1722 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1723 			continue;
1724 		IFADDR_READER_FOREACH(ifa, ifp) {
1725 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1726 			    ifa->ifa_dstaddr == NULL)
1727 				continue;
1728 			if (equal(addr, ifa->ifa_dstaddr))
1729 				return ifa;
1730 		}
1731 	}
1732 
1733 	return NULL;
1734 }
1735 
1736 struct ifaddr *
1737 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1738 {
1739 	struct ifaddr *ifa;
1740 	int s;
1741 
1742 	s = pserialize_read_enter();
1743 	ifa = ifa_ifwithdstaddr(addr);
1744 	if (ifa != NULL)
1745 		ifa_acquire(ifa, psref);
1746 	pserialize_read_exit(s);
1747 
1748 	return ifa;
1749 }
1750 
1751 /*
1752  * Find an interface on a specific network.  If many, choice
1753  * is most specific found.
1754  */
1755 struct ifaddr *
1756 ifa_ifwithnet(const struct sockaddr *addr)
1757 {
1758 	struct ifnet *ifp;
1759 	struct ifaddr *ifa, *ifa_maybe = NULL;
1760 	const struct sockaddr_dl *sdl;
1761 	u_int af = addr->sa_family;
1762 	const char *addr_data = addr->sa_data, *cplim;
1763 
1764 	if (af == AF_LINK) {
1765 		sdl = satocsdl(addr);
1766 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1767 		    ifindex2ifnet[sdl->sdl_index] &&
1768 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1769 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1770 		}
1771 	}
1772 #ifdef NETATALK
1773 	if (af == AF_APPLETALK) {
1774 		const struct sockaddr_at *sat, *sat2;
1775 		sat = (const struct sockaddr_at *)addr;
1776 		IFNET_READER_FOREACH(ifp) {
1777 			if (if_is_deactivated(ifp))
1778 				continue;
1779 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1780 			if (ifa == NULL)
1781 				continue;
1782 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1783 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1784 				return ifa; /* exact match */
1785 			if (ifa_maybe == NULL) {
1786 				/* else keep the if with the right range */
1787 				ifa_maybe = ifa;
1788 			}
1789 		}
1790 		return ifa_maybe;
1791 	}
1792 #endif
1793 	IFNET_READER_FOREACH(ifp) {
1794 		if (if_is_deactivated(ifp))
1795 			continue;
1796 		IFADDR_READER_FOREACH(ifa, ifp) {
1797 			const char *cp, *cp2, *cp3;
1798 
1799 			if (ifa->ifa_addr->sa_family != af ||
1800 			    ifa->ifa_netmask == NULL)
1801  next:				continue;
1802 			cp = addr_data;
1803 			cp2 = ifa->ifa_addr->sa_data;
1804 			cp3 = ifa->ifa_netmask->sa_data;
1805 			cplim = (const char *)ifa->ifa_netmask +
1806 			    ifa->ifa_netmask->sa_len;
1807 			while (cp3 < cplim) {
1808 				if ((*cp++ ^ *cp2++) & *cp3++) {
1809 					/* want to continue for() loop */
1810 					goto next;
1811 				}
1812 			}
1813 			if (ifa_maybe == NULL ||
1814 			    rt_refines(ifa->ifa_netmask,
1815 			               ifa_maybe->ifa_netmask))
1816 				ifa_maybe = ifa;
1817 		}
1818 	}
1819 	return ifa_maybe;
1820 }
1821 
1822 struct ifaddr *
1823 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
1824 {
1825 	struct ifaddr *ifa;
1826 	int s;
1827 
1828 	s = pserialize_read_enter();
1829 	ifa = ifa_ifwithnet(addr);
1830 	if (ifa != NULL)
1831 		ifa_acquire(ifa, psref);
1832 	pserialize_read_exit(s);
1833 
1834 	return ifa;
1835 }
1836 
1837 /*
1838  * Find the interface of the addresss.
1839  */
1840 struct ifaddr *
1841 ifa_ifwithladdr(const struct sockaddr *addr)
1842 {
1843 	struct ifaddr *ia;
1844 
1845 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
1846 	    (ia = ifa_ifwithnet(addr)))
1847 		return ia;
1848 	return NULL;
1849 }
1850 
1851 struct ifaddr *
1852 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
1853 {
1854 	struct ifaddr *ifa;
1855 	int s;
1856 
1857 	s = pserialize_read_enter();
1858 	ifa = ifa_ifwithladdr(addr);
1859 	if (ifa != NULL)
1860 		ifa_acquire(ifa, psref);
1861 	pserialize_read_exit(s);
1862 
1863 	return ifa;
1864 }
1865 
1866 /*
1867  * Find an interface using a specific address family
1868  */
1869 struct ifaddr *
1870 ifa_ifwithaf(int af)
1871 {
1872 	struct ifnet *ifp;
1873 	struct ifaddr *ifa = NULL;
1874 	int s;
1875 
1876 	s = pserialize_read_enter();
1877 	IFNET_READER_FOREACH(ifp) {
1878 		if (if_is_deactivated(ifp))
1879 			continue;
1880 		IFADDR_READER_FOREACH(ifa, ifp) {
1881 			if (ifa->ifa_addr->sa_family == af)
1882 				goto out;
1883 		}
1884 	}
1885 out:
1886 	pserialize_read_exit(s);
1887 	return ifa;
1888 }
1889 
1890 /*
1891  * Find an interface address specific to an interface best matching
1892  * a given address.
1893  */
1894 struct ifaddr *
1895 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
1896 {
1897 	struct ifaddr *ifa;
1898 	const char *cp, *cp2, *cp3;
1899 	const char *cplim;
1900 	struct ifaddr *ifa_maybe = 0;
1901 	u_int af = addr->sa_family;
1902 
1903 	if (if_is_deactivated(ifp))
1904 		return NULL;
1905 
1906 	if (af >= AF_MAX)
1907 		return NULL;
1908 
1909 	IFADDR_READER_FOREACH(ifa, ifp) {
1910 		if (ifa->ifa_addr->sa_family != af)
1911 			continue;
1912 		ifa_maybe = ifa;
1913 		if (ifa->ifa_netmask == NULL) {
1914 			if (equal(addr, ifa->ifa_addr) ||
1915 			    (ifa->ifa_dstaddr &&
1916 			     equal(addr, ifa->ifa_dstaddr)))
1917 				return ifa;
1918 			continue;
1919 		}
1920 		cp = addr->sa_data;
1921 		cp2 = ifa->ifa_addr->sa_data;
1922 		cp3 = ifa->ifa_netmask->sa_data;
1923 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1924 		for (; cp3 < cplim; cp3++) {
1925 			if ((*cp++ ^ *cp2++) & *cp3)
1926 				break;
1927 		}
1928 		if (cp3 == cplim)
1929 			return ifa;
1930 	}
1931 	return ifa_maybe;
1932 }
1933 
1934 struct ifaddr *
1935 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
1936     struct psref *psref)
1937 {
1938 	struct ifaddr *ifa;
1939 	int s;
1940 
1941 	s = pserialize_read_enter();
1942 	ifa = ifaof_ifpforaddr(addr, ifp);
1943 	if (ifa != NULL)
1944 		ifa_acquire(ifa, psref);
1945 	pserialize_read_exit(s);
1946 
1947 	return ifa;
1948 }
1949 
1950 /*
1951  * Default action when installing a route with a Link Level gateway.
1952  * Lookup an appropriate real ifa to point to.
1953  * This should be moved to /sys/net/link.c eventually.
1954  */
1955 void
1956 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
1957 {
1958 	struct ifaddr *ifa;
1959 	const struct sockaddr *dst;
1960 	struct ifnet *ifp;
1961 	struct psref psref;
1962 
1963 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1964 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
1965 		return;
1966 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
1967 		rt_replace_ifa(rt, ifa);
1968 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1969 			ifa->ifa_rtrequest(cmd, rt, info);
1970 		ifa_release(ifa, &psref);
1971 	}
1972 }
1973 
1974 /*
1975  * bitmask macros to manage a densely packed link_state change queue.
1976  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
1977  * LINK_STATE_UP(2) we need 2 bits for each state change.
1978  * As a state change to store is 0, treat all bits set as an unset item.
1979  */
1980 #define LQ_ITEM_BITS		2
1981 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
1982 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
1983 #define LINK_STATE_UNSET	LQ_ITEM_MASK
1984 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
1985 #define LQ_STORE(q, i, v)						      \
1986 	do {								      \
1987 		(q) &= ~LQ_MASK((i));					      \
1988 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
1989 	} while (0 /* CONSTCOND */)
1990 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
1991 #define LQ_POP(q, v)							      \
1992 	do {								      \
1993 		(v) = LQ_ITEM((q), 0);					      \
1994 		(q) >>= LQ_ITEM_BITS;					      \
1995 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
1996 	} while (0 /* CONSTCOND */)
1997 #define LQ_PUSH(q, v)							      \
1998 	do {								      \
1999 		(q) >>= LQ_ITEM_BITS;					      \
2000 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2001 	} while (0 /* CONSTCOND */)
2002 #define LQ_FIND_UNSET(q, i)						      \
2003 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2004 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2005 			break;						      \
2006 	}
2007 /*
2008  * Handle a change in the interface link state and
2009  * queue notifications.
2010  */
2011 void
2012 if_link_state_change(struct ifnet *ifp, int link_state)
2013 {
2014 	int s, idx;
2015 
2016 	KASSERTMSG(if_is_link_state_changeable(ifp),
2017 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2018 	    ifp->if_xname, ifp->if_extflags);
2019 
2020 	/* Ensure change is to a valid state */
2021 	switch (link_state) {
2022 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2023 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2024 	case LINK_STATE_UP:
2025 		break;
2026 	default:
2027 #ifdef DEBUG
2028 		printf("%s: invalid link state %d\n",
2029 		    ifp->if_xname, link_state);
2030 #endif
2031 		return;
2032 	}
2033 
2034 	s = splnet();
2035 
2036 	/* Find the last unset event in the queue. */
2037 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2038 
2039 	/*
2040 	 * Ensure link_state doesn't match the last event in the queue.
2041 	 * ifp->if_link_state is not checked and set here because
2042 	 * that would present an inconsistent picture to the system.
2043 	 */
2044 	if (idx != 0 &&
2045 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2046 		goto out;
2047 
2048 	/* Handle queue overflow. */
2049 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2050 		uint8_t lost;
2051 
2052 		/*
2053 		 * The DOWN state must be protected from being pushed off
2054 		 * the queue to ensure that userland will always be
2055 		 * in a sane state.
2056 		 * Because DOWN is protected, there is no need to protect
2057 		 * UNKNOWN.
2058 		 * It should be invalid to change from any other state to
2059 		 * UNKNOWN anyway ...
2060 		 */
2061 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2062 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2063 		if (lost == LINK_STATE_DOWN) {
2064 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2065 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2066 		}
2067 		printf("%s: lost link state change %s\n",
2068 		    ifp->if_xname,
2069 		    lost == LINK_STATE_UP ? "UP" :
2070 		    lost == LINK_STATE_DOWN ? "DOWN" :
2071 		    "UNKNOWN");
2072 	} else
2073 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2074 
2075 	softint_schedule(ifp->if_link_si);
2076 
2077 out:
2078 	splx(s);
2079 }
2080 
2081 /*
2082  * Handle interface link state change notifications.
2083  * Must be called at splnet().
2084  */
2085 static void
2086 if_link_state_change0(struct ifnet *ifp, int link_state)
2087 {
2088 	struct domain *dp;
2089 
2090 	/* Ensure the change is still valid. */
2091 	if (ifp->if_link_state == link_state)
2092 		return;
2093 
2094 #ifdef DEBUG
2095 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2096 		link_state == LINK_STATE_UP ? "UP" :
2097 		link_state == LINK_STATE_DOWN ? "DOWN" :
2098 		"UNKNOWN",
2099 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2100 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2101 		"UNKNOWN");
2102 #endif
2103 
2104 	/*
2105 	 * When going from UNKNOWN to UP, we need to mark existing
2106 	 * addresses as tentative and restart DAD as we may have
2107 	 * erroneously not found a duplicate.
2108 	 *
2109 	 * This needs to happen before rt_ifmsg to avoid a race where
2110 	 * listeners would have an address and expect it to work right
2111 	 * away.
2112 	 */
2113 	if (link_state == LINK_STATE_UP &&
2114 	    ifp->if_link_state == LINK_STATE_UNKNOWN)
2115 	{
2116 		DOMAIN_FOREACH(dp) {
2117 			if (dp->dom_if_link_state_change != NULL)
2118 				dp->dom_if_link_state_change(ifp,
2119 				    LINK_STATE_DOWN);
2120 		}
2121 	}
2122 
2123 	ifp->if_link_state = link_state;
2124 
2125 	/* Notify that the link state has changed. */
2126 	rt_ifmsg(ifp);
2127 
2128 #if NCARP > 0
2129 	if (ifp->if_carp)
2130 		carp_carpdev_state(ifp);
2131 #endif
2132 
2133 	DOMAIN_FOREACH(dp) {
2134 		if (dp->dom_if_link_state_change != NULL)
2135 			dp->dom_if_link_state_change(ifp, link_state);
2136 	}
2137 }
2138 
2139 /*
2140  * Process the interface link state change queue.
2141  */
2142 static void
2143 if_link_state_change_si(void *arg)
2144 {
2145 	struct ifnet *ifp = arg;
2146 	int s;
2147 	uint8_t state;
2148 
2149 	s = splnet();
2150 
2151 	/* Pop a link state change from the queue and process it. */
2152 	LQ_POP(ifp->if_link_queue, state);
2153 	if_link_state_change0(ifp, state);
2154 
2155 	/* If there is a link state change to come, schedule it. */
2156 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET)
2157 		softint_schedule(ifp->if_link_si);
2158 
2159 	splx(s);
2160 }
2161 
2162 /*
2163  * Default action when installing a local route on a point-to-point
2164  * interface.
2165  */
2166 void
2167 p2p_rtrequest(int req, struct rtentry *rt,
2168     __unused const struct rt_addrinfo *info)
2169 {
2170 	struct ifnet *ifp = rt->rt_ifp;
2171 	struct ifaddr *ifa, *lo0ifa;
2172 	int s = pserialize_read_enter();
2173 
2174 	switch (req) {
2175 	case RTM_ADD:
2176 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2177 			break;
2178 
2179 		rt->rt_ifp = lo0ifp;
2180 
2181 		IFADDR_READER_FOREACH(ifa, ifp) {
2182 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2183 				break;
2184 		}
2185 		if (ifa == NULL)
2186 			break;
2187 
2188 		/*
2189 		 * Ensure lo0 has an address of the same family.
2190 		 */
2191 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2192 			if (lo0ifa->ifa_addr->sa_family ==
2193 			    ifa->ifa_addr->sa_family)
2194 				break;
2195 		}
2196 		if (lo0ifa == NULL)
2197 			break;
2198 
2199 		/*
2200 		 * Make sure to set rt->rt_ifa to the interface
2201 		 * address we are using, otherwise we will have trouble
2202 		 * with source address selection.
2203 		 */
2204 		if (ifa != rt->rt_ifa)
2205 			rt_replace_ifa(rt, ifa);
2206 		break;
2207 	case RTM_DELETE:
2208 	default:
2209 		break;
2210 	}
2211 	pserialize_read_exit(s);
2212 }
2213 
2214 /*
2215  * Mark an interface down and notify protocols of
2216  * the transition.
2217  * NOTE: must be called at splsoftnet or equivalent.
2218  */
2219 void
2220 if_down(struct ifnet *ifp)
2221 {
2222 	struct ifaddr *ifa;
2223 	struct domain *dp;
2224 	int s, bound;
2225 	struct psref psref;
2226 
2227 	ifp->if_flags &= ~IFF_UP;
2228 	nanotime(&ifp->if_lastchange);
2229 
2230 	bound = curlwp_bind();
2231 	s = pserialize_read_enter();
2232 	IFADDR_READER_FOREACH(ifa, ifp) {
2233 		ifa_acquire(ifa, &psref);
2234 		pserialize_read_exit(s);
2235 
2236 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2237 
2238 		s = pserialize_read_enter();
2239 		ifa_release(ifa, &psref);
2240 	}
2241 	pserialize_read_exit(s);
2242 	curlwp_bindx(bound);
2243 
2244 	IFQ_PURGE(&ifp->if_snd);
2245 #if NCARP > 0
2246 	if (ifp->if_carp)
2247 		carp_carpdev_state(ifp);
2248 #endif
2249 	rt_ifmsg(ifp);
2250 	DOMAIN_FOREACH(dp) {
2251 		if (dp->dom_if_down)
2252 			dp->dom_if_down(ifp);
2253 	}
2254 }
2255 
2256 /*
2257  * Mark an interface up and notify protocols of
2258  * the transition.
2259  * NOTE: must be called at splsoftnet or equivalent.
2260  */
2261 void
2262 if_up(struct ifnet *ifp)
2263 {
2264 #ifdef notyet
2265 	struct ifaddr *ifa;
2266 #endif
2267 	struct domain *dp;
2268 
2269 	ifp->if_flags |= IFF_UP;
2270 	nanotime(&ifp->if_lastchange);
2271 #ifdef notyet
2272 	/* this has no effect on IP, and will kill all ISO connections XXX */
2273 	IFADDR_READER_FOREACH(ifa, ifp)
2274 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2275 #endif
2276 #if NCARP > 0
2277 	if (ifp->if_carp)
2278 		carp_carpdev_state(ifp);
2279 #endif
2280 	rt_ifmsg(ifp);
2281 	DOMAIN_FOREACH(dp) {
2282 		if (dp->dom_if_up)
2283 			dp->dom_if_up(ifp);
2284 	}
2285 }
2286 
2287 /*
2288  * Handle interface slowtimo timer routine.  Called
2289  * from softclock, we decrement timer (if set) and
2290  * call the appropriate interface routine on expiration.
2291  */
2292 static void
2293 if_slowtimo(void *arg)
2294 {
2295 	void (*slowtimo)(struct ifnet *);
2296 	struct ifnet *ifp = arg;
2297 	int s;
2298 
2299 	slowtimo = ifp->if_slowtimo;
2300 	if (__predict_false(slowtimo == NULL))
2301 		return;
2302 
2303 	s = splnet();
2304 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2305 		(*slowtimo)(ifp);
2306 
2307 	splx(s);
2308 
2309 	if (__predict_true(ifp->if_slowtimo != NULL))
2310 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2311 }
2312 
2313 /*
2314  * Set/clear promiscuous mode on interface ifp based on the truth value
2315  * of pswitch.  The calls are reference counted so that only the first
2316  * "on" request actually has an effect, as does the final "off" request.
2317  * Results are undefined if the "off" and "on" requests are not matched.
2318  */
2319 int
2320 ifpromisc(struct ifnet *ifp, int pswitch)
2321 {
2322 	int pcount, ret;
2323 	short nflags;
2324 
2325 	pcount = ifp->if_pcount;
2326 	if (pswitch) {
2327 		/*
2328 		 * Allow the device to be "placed" into promiscuous
2329 		 * mode even if it is not configured up.  It will
2330 		 * consult IFF_PROMISC when it is brought up.
2331 		 */
2332 		if (ifp->if_pcount++ != 0)
2333 			return 0;
2334 		nflags = ifp->if_flags | IFF_PROMISC;
2335 	} else {
2336 		if (--ifp->if_pcount > 0)
2337 			return 0;
2338 		nflags = ifp->if_flags & ~IFF_PROMISC;
2339 	}
2340 	ret = if_flags_set(ifp, nflags);
2341 	/* Restore interface state if not successful. */
2342 	if (ret != 0) {
2343 		ifp->if_pcount = pcount;
2344 	}
2345 	return ret;
2346 }
2347 
2348 /*
2349  * Map interface name to
2350  * interface structure pointer.
2351  */
2352 struct ifnet *
2353 ifunit(const char *name)
2354 {
2355 	struct ifnet *ifp;
2356 	const char *cp = name;
2357 	u_int unit = 0;
2358 	u_int i;
2359 	int s;
2360 
2361 	/*
2362 	 * If the entire name is a number, treat it as an ifindex.
2363 	 */
2364 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2365 		unit = unit * 10 + (*cp - '0');
2366 	}
2367 
2368 	/*
2369 	 * If the number took all of the name, then it's a valid ifindex.
2370 	 */
2371 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2372 		if (unit >= if_indexlim)
2373 			return NULL;
2374 		ifp = ifindex2ifnet[unit];
2375 		if (ifp == NULL || if_is_deactivated(ifp))
2376 			return NULL;
2377 		return ifp;
2378 	}
2379 
2380 	ifp = NULL;
2381 	s = pserialize_read_enter();
2382 	IFNET_READER_FOREACH(ifp) {
2383 		if (if_is_deactivated(ifp))
2384 			continue;
2385 	 	if (strcmp(ifp->if_xname, name) == 0)
2386 			goto out;
2387 	}
2388 out:
2389 	pserialize_read_exit(s);
2390 	return ifp;
2391 }
2392 
2393 /*
2394  * Get a reference of an ifnet object by an interface name.
2395  * The returned reference is protected by psref(9). The caller
2396  * must release a returned reference by if_put after use.
2397  */
2398 struct ifnet *
2399 if_get(const char *name, struct psref *psref)
2400 {
2401 	struct ifnet *ifp;
2402 	const char *cp = name;
2403 	u_int unit = 0;
2404 	u_int i;
2405 	int s;
2406 
2407 	/*
2408 	 * If the entire name is a number, treat it as an ifindex.
2409 	 */
2410 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2411 		unit = unit * 10 + (*cp - '0');
2412 	}
2413 
2414 	/*
2415 	 * If the number took all of the name, then it's a valid ifindex.
2416 	 */
2417 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2418 		if (unit >= if_indexlim)
2419 			return NULL;
2420 		ifp = ifindex2ifnet[unit];
2421 		if (ifp == NULL || if_is_deactivated(ifp))
2422 			return NULL;
2423 		return ifp;
2424 	}
2425 
2426 	ifp = NULL;
2427 	s = pserialize_read_enter();
2428 	IFNET_READER_FOREACH(ifp) {
2429 		if (if_is_deactivated(ifp))
2430 			continue;
2431 		if (strcmp(ifp->if_xname, name) == 0) {
2432 			psref_acquire(psref, &ifp->if_psref,
2433 			    ifnet_psref_class);
2434 			goto out;
2435 		}
2436 	}
2437 out:
2438 	pserialize_read_exit(s);
2439 	return ifp;
2440 }
2441 
2442 /*
2443  * Release a reference of an ifnet object given by if_get or
2444  * if_get_byindex.
2445  */
2446 void
2447 if_put(const struct ifnet *ifp, struct psref *psref)
2448 {
2449 
2450 	if (ifp == NULL)
2451 		return;
2452 
2453 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2454 }
2455 
2456 ifnet_t *
2457 if_byindex(u_int idx)
2458 {
2459 	ifnet_t *ifp;
2460 
2461 	ifp = (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL;
2462 	if (ifp != NULL && if_is_deactivated(ifp))
2463 		ifp = NULL;
2464 	return ifp;
2465 }
2466 
2467 /*
2468  * Get a reference of an ifnet object by an interface index.
2469  * The returned reference is protected by psref(9). The caller
2470  * must release a returned reference by if_put after use.
2471  */
2472 ifnet_t *
2473 if_get_byindex(u_int idx, struct psref *psref)
2474 {
2475 	ifnet_t *ifp;
2476 	int s;
2477 
2478 	s = pserialize_read_enter();
2479 	ifp = (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2480 	if (ifp != NULL && if_is_deactivated(ifp))
2481 		ifp = NULL;
2482 	if (__predict_true(ifp != NULL))
2483 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2484 	pserialize_read_exit(s);
2485 
2486 	return ifp;
2487 }
2488 
2489 /*
2490  * XXX it's safe only if the passed ifp is guaranteed to not be freed,
2491  * for example the ifp is already held or some other object is held which
2492  * guarantes the ifp to not be freed indirectly.
2493  */
2494 void
2495 if_acquire_NOMPSAFE(struct ifnet *ifp, struct psref *psref)
2496 {
2497 
2498 	KASSERT(ifp->if_index != 0);
2499 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2500 }
2501 
2502 bool
2503 if_held(struct ifnet *ifp)
2504 {
2505 
2506 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2507 }
2508 
2509 
2510 /* common */
2511 int
2512 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2513 {
2514 	int s;
2515 	struct ifreq *ifr;
2516 	struct ifcapreq *ifcr;
2517 	struct ifdatareq *ifdr;
2518 
2519 	switch (cmd) {
2520 	case SIOCSIFCAP:
2521 		ifcr = data;
2522 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2523 			return EINVAL;
2524 
2525 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2526 			return 0;
2527 
2528 		ifp->if_capenable = ifcr->ifcr_capenable;
2529 
2530 		/* Pre-compute the checksum flags mask. */
2531 		ifp->if_csum_flags_tx = 0;
2532 		ifp->if_csum_flags_rx = 0;
2533 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2534 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2535 		}
2536 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2537 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2538 		}
2539 
2540 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2541 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2542 		}
2543 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2544 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2545 		}
2546 
2547 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2548 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2549 		}
2550 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2551 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2552 		}
2553 
2554 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2555 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2556 		}
2557 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2558 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2559 		}
2560 
2561 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2562 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2563 		}
2564 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2565 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2566 		}
2567 		if (ifp->if_flags & IFF_UP)
2568 			return ENETRESET;
2569 		return 0;
2570 	case SIOCSIFFLAGS:
2571 		ifr = data;
2572 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2573 			s = splnet();
2574 			if_down(ifp);
2575 			splx(s);
2576 		}
2577 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2578 			s = splnet();
2579 			if_up(ifp);
2580 			splx(s);
2581 		}
2582 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2583 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2584 		break;
2585 	case SIOCGIFFLAGS:
2586 		ifr = data;
2587 		ifr->ifr_flags = ifp->if_flags;
2588 		break;
2589 
2590 	case SIOCGIFMETRIC:
2591 		ifr = data;
2592 		ifr->ifr_metric = ifp->if_metric;
2593 		break;
2594 
2595 	case SIOCGIFMTU:
2596 		ifr = data;
2597 		ifr->ifr_mtu = ifp->if_mtu;
2598 		break;
2599 
2600 	case SIOCGIFDLT:
2601 		ifr = data;
2602 		ifr->ifr_dlt = ifp->if_dlt;
2603 		break;
2604 
2605 	case SIOCGIFCAP:
2606 		ifcr = data;
2607 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2608 		ifcr->ifcr_capenable = ifp->if_capenable;
2609 		break;
2610 
2611 	case SIOCSIFMETRIC:
2612 		ifr = data;
2613 		ifp->if_metric = ifr->ifr_metric;
2614 		break;
2615 
2616 	case SIOCGIFDATA:
2617 		ifdr = data;
2618 		ifdr->ifdr_data = ifp->if_data;
2619 		break;
2620 
2621 	case SIOCGIFINDEX:
2622 		ifr = data;
2623 		ifr->ifr_index = ifp->if_index;
2624 		break;
2625 
2626 	case SIOCZIFDATA:
2627 		ifdr = data;
2628 		ifdr->ifdr_data = ifp->if_data;
2629 		/*
2630 		 * Assumes that the volatile counters that can be
2631 		 * zero'ed are at the end of if_data.
2632 		 */
2633 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
2634 		    offsetof(struct if_data, ifi_ipackets));
2635 		/*
2636 		 * The memset() clears to the bottm of if_data. In the area,
2637 		 * if_lastchange is included. Please be careful if new entry
2638 		 * will be added into if_data or rewite this.
2639 		 *
2640 		 * And also, update if_lastchnage.
2641 		 */
2642 		getnanotime(&ifp->if_lastchange);
2643 		break;
2644 	case SIOCSIFMTU:
2645 		ifr = data;
2646 		if (ifp->if_mtu == ifr->ifr_mtu)
2647 			break;
2648 		ifp->if_mtu = ifr->ifr_mtu;
2649 		/*
2650 		 * If the link MTU changed, do network layer specific procedure.
2651 		 */
2652 #ifdef INET6
2653 		if (in6_present)
2654 			nd6_setmtu(ifp);
2655 #endif
2656 		return ENETRESET;
2657 	default:
2658 		return ENOTTY;
2659 	}
2660 	return 0;
2661 }
2662 
2663 int
2664 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
2665 {
2666 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
2667 	struct ifaddr *ifa;
2668 	const struct sockaddr *any, *sa;
2669 	union {
2670 		struct sockaddr sa;
2671 		struct sockaddr_storage ss;
2672 	} u, v;
2673 	int s, error = 0;
2674 
2675 	switch (cmd) {
2676 	case SIOCSIFADDRPREF:
2677 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
2678 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
2679 		    NULL) != 0)
2680 			return EPERM;
2681 	case SIOCGIFADDRPREF:
2682 		break;
2683 	default:
2684 		return EOPNOTSUPP;
2685 	}
2686 
2687 	/* sanity checks */
2688 	if (data == NULL || ifp == NULL) {
2689 		panic("invalid argument to %s", __func__);
2690 		/*NOTREACHED*/
2691 	}
2692 
2693 	/* address must be specified on ADD and DELETE */
2694 	sa = sstocsa(&ifap->ifap_addr);
2695 	if (sa->sa_family != sofamily(so))
2696 		return EINVAL;
2697 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
2698 		return EINVAL;
2699 
2700 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
2701 
2702 	s = pserialize_read_enter();
2703 	IFADDR_READER_FOREACH(ifa, ifp) {
2704 		if (ifa->ifa_addr->sa_family != sa->sa_family)
2705 			continue;
2706 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
2707 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
2708 			break;
2709 	}
2710 	if (ifa == NULL) {
2711 		error = EADDRNOTAVAIL;
2712 		goto out;
2713 	}
2714 
2715 	switch (cmd) {
2716 	case SIOCSIFADDRPREF:
2717 		ifa->ifa_preference = ifap->ifap_preference;
2718 		goto out;
2719 	case SIOCGIFADDRPREF:
2720 		/* fill in the if_laddrreq structure */
2721 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
2722 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
2723 		ifap->ifap_preference = ifa->ifa_preference;
2724 		goto out;
2725 	default:
2726 		error = EOPNOTSUPP;
2727 	}
2728 out:
2729 	pserialize_read_exit(s);
2730 	return error;
2731 }
2732 
2733 /*
2734  * Interface ioctls.
2735  */
2736 static int
2737 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
2738 {
2739 	struct ifnet *ifp;
2740 	struct ifreq *ifr;
2741 	int error = 0;
2742 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
2743 	u_long ocmd = cmd;
2744 #endif
2745 	short oif_flags;
2746 #ifdef COMPAT_OIFREQ
2747 	struct ifreq ifrb;
2748 	struct oifreq *oifr = NULL;
2749 #endif
2750 	int r;
2751 	struct psref psref;
2752 	int bound;
2753 
2754 	switch (cmd) {
2755 #ifdef COMPAT_OIFREQ
2756 	case OSIOCGIFCONF:
2757 	case OOSIOCGIFCONF:
2758 		return compat_ifconf(cmd, data);
2759 #endif
2760 #ifdef COMPAT_OIFDATA
2761 	case OSIOCGIFDATA:
2762 	case OSIOCZIFDATA:
2763 		return compat_ifdatareq(l, cmd, data);
2764 #endif
2765 	case SIOCGIFCONF:
2766 		return ifconf(cmd, data);
2767 	case SIOCINITIFADDR:
2768 		return EPERM;
2769 	}
2770 
2771 #ifdef COMPAT_OIFREQ
2772 	cmd = compat_cvtcmd(cmd);
2773 	if (cmd != ocmd) {
2774 		oifr = data;
2775 		data = ifr = &ifrb;
2776 		ifreqo2n(oifr, ifr);
2777 	} else
2778 #endif
2779 		ifr = data;
2780 
2781 	switch (cmd) {
2782 	case SIOCIFCREATE:
2783 	case SIOCIFDESTROY:
2784 		bound = curlwp_bind();
2785 		if (l != NULL) {
2786 			ifp = if_get(ifr->ifr_name, &psref);
2787 			error = kauth_authorize_network(l->l_cred,
2788 			    KAUTH_NETWORK_INTERFACE,
2789 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2790 			    (void *)cmd, NULL);
2791 			if (ifp != NULL)
2792 				if_put(ifp, &psref);
2793 			if (error != 0) {
2794 				curlwp_bindx(bound);
2795 				return error;
2796 			}
2797 		}
2798 		mutex_enter(&if_clone_mtx);
2799 		r = (cmd == SIOCIFCREATE) ?
2800 			if_clone_create(ifr->ifr_name) :
2801 			if_clone_destroy(ifr->ifr_name);
2802 		mutex_exit(&if_clone_mtx);
2803 		curlwp_bindx(bound);
2804 		return r;
2805 
2806 	case SIOCIFGCLONERS:
2807 		{
2808 			struct if_clonereq *req = (struct if_clonereq *)data;
2809 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
2810 			    &req->ifcr_total);
2811 		}
2812 	}
2813 
2814 	bound = curlwp_bind();
2815 	ifp = if_get(ifr->ifr_name, &psref);
2816 	if (ifp == NULL) {
2817 		curlwp_bindx(bound);
2818 		return ENXIO;
2819 	}
2820 
2821 	switch (cmd) {
2822 	case SIOCALIFADDR:
2823 	case SIOCDLIFADDR:
2824 	case SIOCSIFADDRPREF:
2825 	case SIOCSIFFLAGS:
2826 	case SIOCSIFCAP:
2827 	case SIOCSIFMETRIC:
2828 	case SIOCZIFDATA:
2829 	case SIOCSIFMTU:
2830 	case SIOCSIFPHYADDR:
2831 	case SIOCDIFPHYADDR:
2832 #ifdef INET6
2833 	case SIOCSIFPHYADDR_IN6:
2834 #endif
2835 	case SIOCSLIFPHYADDR:
2836 	case SIOCADDMULTI:
2837 	case SIOCDELMULTI:
2838 	case SIOCSIFMEDIA:
2839 	case SIOCSDRVSPEC:
2840 	case SIOCG80211:
2841 	case SIOCS80211:
2842 	case SIOCS80211NWID:
2843 	case SIOCS80211NWKEY:
2844 	case SIOCS80211POWER:
2845 	case SIOCS80211BSSID:
2846 	case SIOCS80211CHANNEL:
2847 	case SIOCSLINKSTR:
2848 		if (l != NULL) {
2849 			error = kauth_authorize_network(l->l_cred,
2850 			    KAUTH_NETWORK_INTERFACE,
2851 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2852 			    (void *)cmd, NULL);
2853 			if (error != 0)
2854 				goto out;
2855 		}
2856 	}
2857 
2858 	oif_flags = ifp->if_flags;
2859 
2860 	mutex_enter(ifp->if_ioctl_lock);
2861 
2862 	error = (*ifp->if_ioctl)(ifp, cmd, data);
2863 	if (error != ENOTTY)
2864 		;
2865 	else if (so->so_proto == NULL)
2866 		error = EOPNOTSUPP;
2867 	else {
2868 #ifdef COMPAT_OSOCK
2869 		error = compat_ifioctl(so, ocmd, cmd, data, l);
2870 #else
2871 		error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
2872 		    cmd, data, ifp);
2873 #endif
2874 	}
2875 
2876 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
2877 		if ((ifp->if_flags & IFF_UP) != 0) {
2878 			int s = splnet();
2879 			if_up(ifp);
2880 			splx(s);
2881 		}
2882 	}
2883 #ifdef COMPAT_OIFREQ
2884 	if (cmd != ocmd)
2885 		ifreqn2o(oifr, ifr);
2886 #endif
2887 
2888 	mutex_exit(ifp->if_ioctl_lock);
2889 out:
2890 	if_put(ifp, &psref);
2891 	curlwp_bindx(bound);
2892 	return error;
2893 }
2894 
2895 /*
2896  * Return interface configuration
2897  * of system.  List may be used
2898  * in later ioctl's (above) to get
2899  * other information.
2900  *
2901  * Each record is a struct ifreq.  Before the addition of
2902  * sockaddr_storage, the API rule was that sockaddr flavors that did
2903  * not fit would extend beyond the struct ifreq, with the next struct
2904  * ifreq starting sa_len beyond the struct sockaddr.  Because the
2905  * union in struct ifreq includes struct sockaddr_storage, every kind
2906  * of sockaddr must fit.  Thus, there are no longer any overlength
2907  * records.
2908  *
2909  * Records are added to the user buffer if they fit, and ifc_len is
2910  * adjusted to the length that was written.  Thus, the user is only
2911  * assured of getting the complete list if ifc_len on return is at
2912  * least sizeof(struct ifreq) less than it was on entry.
2913  *
2914  * If the user buffer pointer is NULL, this routine copies no data and
2915  * returns the amount of space that would be needed.
2916  *
2917  * Invariants:
2918  * ifrp points to the next part of the user's buffer to be used.  If
2919  * ifrp != NULL, space holds the number of bytes remaining that we may
2920  * write at ifrp.  Otherwise, space holds the number of bytes that
2921  * would have been written had there been adequate space.
2922  */
2923 /*ARGSUSED*/
2924 static int
2925 ifconf(u_long cmd, void *data)
2926 {
2927 	struct ifconf *ifc = (struct ifconf *)data;
2928 	struct ifnet *ifp;
2929 	struct ifaddr *ifa;
2930 	struct ifreq ifr, *ifrp = NULL;
2931 	int space = 0, error = 0;
2932 	const int sz = (int)sizeof(struct ifreq);
2933 	const bool docopy = ifc->ifc_req != NULL;
2934 	int s;
2935 	int bound;
2936 	struct psref psref;
2937 
2938 	if (docopy) {
2939 		space = ifc->ifc_len;
2940 		ifrp = ifc->ifc_req;
2941 	}
2942 
2943 	bound = curlwp_bind();
2944 	s = pserialize_read_enter();
2945 	IFNET_READER_FOREACH(ifp) {
2946 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
2947 		pserialize_read_exit(s);
2948 
2949 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
2950 		    sizeof(ifr.ifr_name));
2951 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
2952 			error = ENAMETOOLONG;
2953 			goto release_exit;
2954 		}
2955 		if (IFADDR_READER_EMPTY(ifp)) {
2956 			/* Interface with no addresses - send zero sockaddr. */
2957 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
2958 			if (!docopy) {
2959 				space += sz;
2960 				continue;
2961 			}
2962 			if (space >= sz) {
2963 				error = copyout(&ifr, ifrp, sz);
2964 				if (error != 0)
2965 					goto release_exit;
2966 				ifrp++;
2967 				space -= sz;
2968 			}
2969 		}
2970 
2971 		IFADDR_READER_FOREACH(ifa, ifp) {
2972 			struct sockaddr *sa = ifa->ifa_addr;
2973 			/* all sockaddrs must fit in sockaddr_storage */
2974 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
2975 
2976 			if (!docopy) {
2977 				space += sz;
2978 				continue;
2979 			}
2980 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
2981 			if (space >= sz) {
2982 				error = copyout(&ifr, ifrp, sz);
2983 				if (error != 0)
2984 					goto release_exit;
2985 				ifrp++; space -= sz;
2986 			}
2987 		}
2988 
2989 		s = pserialize_read_enter();
2990 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
2991 	}
2992 	pserialize_read_exit(s);
2993 	curlwp_bindx(bound);
2994 
2995 	if (docopy) {
2996 		KASSERT(0 <= space && space <= ifc->ifc_len);
2997 		ifc->ifc_len -= space;
2998 	} else {
2999 		KASSERT(space >= 0);
3000 		ifc->ifc_len = space;
3001 	}
3002 	return (0);
3003 
3004 release_exit:
3005 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3006 	curlwp_bindx(bound);
3007 	return error;
3008 }
3009 
3010 int
3011 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3012 {
3013 	uint8_t len;
3014 #ifdef COMPAT_OIFREQ
3015 	struct ifreq ifrb;
3016 	struct oifreq *oifr = NULL;
3017 	u_long ocmd = cmd;
3018 	cmd = compat_cvtcmd(cmd);
3019 	if (cmd != ocmd) {
3020 		oifr = (struct oifreq *)(void *)ifr;
3021 		ifr = &ifrb;
3022 		ifreqo2n(oifr, ifr);
3023 		len = sizeof(oifr->ifr_addr);
3024 	} else
3025 #endif
3026 		len = sizeof(ifr->ifr_ifru.ifru_space);
3027 
3028 	if (len < sa->sa_len)
3029 		return EFBIG;
3030 
3031 	memset(&ifr->ifr_addr, 0, len);
3032 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3033 
3034 #ifdef COMPAT_OIFREQ
3035 	if (cmd != ocmd)
3036 		ifreqn2o(oifr, ifr);
3037 #endif
3038 	return 0;
3039 }
3040 
3041 /*
3042  * wrapper function for the drivers which doesn't have if_transmit().
3043  */
3044 static int
3045 if_transmit(struct ifnet *ifp, struct mbuf *m)
3046 {
3047 	int s, error;
3048 
3049 	s = splnet();
3050 
3051 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3052 	if (error != 0) {
3053 		/* mbuf is already freed */
3054 		goto out;
3055 	}
3056 
3057 	ifp->if_obytes += m->m_pkthdr.len;;
3058 	if (m->m_flags & M_MCAST)
3059 		ifp->if_omcasts++;
3060 
3061 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3062 		if_start_lock(ifp);
3063 out:
3064 	splx(s);
3065 
3066 	return error;
3067 }
3068 
3069 int
3070 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3071 {
3072 	int error;
3073 
3074 #ifdef ALTQ
3075 	KERNEL_LOCK(1, NULL);
3076 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3077 		error = if_transmit(ifp, m);
3078 		KERNEL_UNLOCK_ONE(NULL);
3079 	} else {
3080 		KERNEL_UNLOCK_ONE(NULL);
3081 		error = (*ifp->if_transmit)(ifp, m);
3082 	}
3083 #else /* !ALTQ */
3084 	error = (*ifp->if_transmit)(ifp, m);
3085 #endif /* !ALTQ */
3086 
3087 	return error;
3088 }
3089 
3090 /*
3091  * Queue message on interface, and start output if interface
3092  * not yet active.
3093  */
3094 int
3095 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3096 {
3097 
3098 	return if_transmit_lock(ifp, m);
3099 }
3100 
3101 /*
3102  * Queue message on interface, possibly using a second fast queue
3103  */
3104 int
3105 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3106 {
3107 	int error = 0;
3108 
3109 	if (ifq != NULL
3110 #ifdef ALTQ
3111 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3112 #endif
3113 	    ) {
3114 		if (IF_QFULL(ifq)) {
3115 			IF_DROP(&ifp->if_snd);
3116 			m_freem(m);
3117 			if (error == 0)
3118 				error = ENOBUFS;
3119 		} else
3120 			IF_ENQUEUE(ifq, m);
3121 	} else
3122 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3123 	if (error != 0) {
3124 		++ifp->if_oerrors;
3125 		return error;
3126 	}
3127 	return 0;
3128 }
3129 
3130 int
3131 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3132 {
3133 	int rc;
3134 
3135 	if (ifp->if_initaddr != NULL)
3136 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3137 	else if (src ||
3138 		/* FIXME: may not hold if_ioctl_lock */
3139 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3140 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3141 
3142 	return rc;
3143 }
3144 
3145 int
3146 if_do_dad(struct ifnet *ifp)
3147 {
3148 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3149 		return 0;
3150 
3151 	switch (ifp->if_type) {
3152 	case IFT_FAITH:
3153 		/*
3154 		 * These interfaces do not have the IFF_LOOPBACK flag,
3155 		 * but loop packets back.  We do not have to do DAD on such
3156 		 * interfaces.  We should even omit it, because loop-backed
3157 		 * responses would confuse the DAD procedure.
3158 		 */
3159 		return 0;
3160 	default:
3161 		/*
3162 		 * Our DAD routine requires the interface up and running.
3163 		 * However, some interfaces can be up before the RUNNING
3164 		 * status.  Additionaly, users may try to assign addresses
3165 		 * before the interface becomes up (or running).
3166 		 * We simply skip DAD in such a case as a work around.
3167 		 * XXX: we should rather mark "tentative" on such addresses,
3168 		 * and do DAD after the interface becomes ready.
3169 		 */
3170 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
3171 		    (IFF_UP|IFF_RUNNING))
3172 			return 0;
3173 
3174 		return 1;
3175 	}
3176 }
3177 
3178 int
3179 if_flags_set(ifnet_t *ifp, const short flags)
3180 {
3181 	int rc;
3182 
3183 	if (ifp->if_setflags != NULL)
3184 		rc = (*ifp->if_setflags)(ifp, flags);
3185 	else {
3186 		short cantflags, chgdflags;
3187 		struct ifreq ifr;
3188 
3189 		chgdflags = ifp->if_flags ^ flags;
3190 		cantflags = chgdflags & IFF_CANTCHANGE;
3191 
3192 		if (cantflags != 0)
3193 			ifp->if_flags ^= cantflags;
3194 
3195                 /* Traditionally, we do not call if_ioctl after
3196                  * setting/clearing only IFF_PROMISC if the interface
3197                  * isn't IFF_UP.  Uphold that tradition.
3198 		 */
3199 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3200 			return 0;
3201 
3202 		memset(&ifr, 0, sizeof(ifr));
3203 
3204 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3205 		/* FIXME: may not hold if_ioctl_lock */
3206 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3207 
3208 		if (rc != 0 && cantflags != 0)
3209 			ifp->if_flags ^= cantflags;
3210 	}
3211 
3212 	return rc;
3213 }
3214 
3215 int
3216 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3217 {
3218 	int rc;
3219 	struct ifreq ifr;
3220 
3221 	if (ifp->if_mcastop != NULL)
3222 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3223 	else {
3224 		ifreq_setaddr(cmd, &ifr, sa);
3225 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3226 	}
3227 
3228 	return rc;
3229 }
3230 
3231 static void
3232 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3233     struct ifaltq *ifq)
3234 {
3235 	const struct sysctlnode *cnode, *rnode;
3236 
3237 	if (sysctl_createv(clog, 0, NULL, &rnode,
3238 		       CTLFLAG_PERMANENT,
3239 		       CTLTYPE_NODE, "interfaces",
3240 		       SYSCTL_DESCR("Per-interface controls"),
3241 		       NULL, 0, NULL, 0,
3242 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3243 		goto bad;
3244 
3245 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3246 		       CTLFLAG_PERMANENT,
3247 		       CTLTYPE_NODE, ifname,
3248 		       SYSCTL_DESCR("Interface controls"),
3249 		       NULL, 0, NULL, 0,
3250 		       CTL_CREATE, CTL_EOL) != 0)
3251 		goto bad;
3252 
3253 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3254 		       CTLFLAG_PERMANENT,
3255 		       CTLTYPE_NODE, "sndq",
3256 		       SYSCTL_DESCR("Interface output queue controls"),
3257 		       NULL, 0, NULL, 0,
3258 		       CTL_CREATE, CTL_EOL) != 0)
3259 		goto bad;
3260 
3261 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3262 		       CTLFLAG_PERMANENT,
3263 		       CTLTYPE_INT, "len",
3264 		       SYSCTL_DESCR("Current output queue length"),
3265 		       NULL, 0, &ifq->ifq_len, 0,
3266 		       CTL_CREATE, CTL_EOL) != 0)
3267 		goto bad;
3268 
3269 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3270 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3271 		       CTLTYPE_INT, "maxlen",
3272 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3273 		       NULL, 0, &ifq->ifq_maxlen, 0,
3274 		       CTL_CREATE, CTL_EOL) != 0)
3275 		goto bad;
3276 
3277 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3278 		       CTLFLAG_PERMANENT,
3279 		       CTLTYPE_INT, "drops",
3280 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3281 		       NULL, 0, &ifq->ifq_drops, 0,
3282 		       CTL_CREATE, CTL_EOL) != 0)
3283 		goto bad;
3284 
3285 	return;
3286 bad:
3287 	printf("%s: could not attach sysctl nodes\n", ifname);
3288 	return;
3289 }
3290 
3291 #if defined(INET) || defined(INET6)
3292 
3293 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3294 	static int							\
3295 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3296 	{								\
3297 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3298 	}
3299 
3300 #if defined(INET)
3301 static int
3302 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3303 {
3304 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3305 }
3306 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3307 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3308 #endif
3309 
3310 #if defined(INET6)
3311 static int
3312 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3313 {
3314 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3315 }
3316 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3317 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3318 #endif
3319 
3320 static void
3321 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3322 {
3323 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3324 	const char *pfname = NULL, *ipname = NULL;
3325 	int ipn = 0, qid = 0;
3326 
3327 	switch (pf) {
3328 #if defined(INET)
3329 	case PF_INET:
3330 		len_func = sysctl_net_ip_pktq_items;
3331 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3332 		drops_func = sysctl_net_ip_pktq_drops;
3333 		pfname = "inet", ipn = IPPROTO_IP;
3334 		ipname = "ip", qid = IPCTL_IFQ;
3335 		break;
3336 #endif
3337 #if defined(INET6)
3338 	case PF_INET6:
3339 		len_func = sysctl_net_ip6_pktq_items;
3340 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3341 		drops_func = sysctl_net_ip6_pktq_drops;
3342 		pfname = "inet6", ipn = IPPROTO_IPV6;
3343 		ipname = "ip6", qid = IPV6CTL_IFQ;
3344 		break;
3345 #endif
3346 	default:
3347 		KASSERT(false);
3348 	}
3349 
3350 	sysctl_createv(clog, 0, NULL, NULL,
3351 		       CTLFLAG_PERMANENT,
3352 		       CTLTYPE_NODE, pfname, NULL,
3353 		       NULL, 0, NULL, 0,
3354 		       CTL_NET, pf, CTL_EOL);
3355 	sysctl_createv(clog, 0, NULL, NULL,
3356 		       CTLFLAG_PERMANENT,
3357 		       CTLTYPE_NODE, ipname, NULL,
3358 		       NULL, 0, NULL, 0,
3359 		       CTL_NET, pf, ipn, CTL_EOL);
3360 	sysctl_createv(clog, 0, NULL, NULL,
3361 		       CTLFLAG_PERMANENT,
3362 		       CTLTYPE_NODE, "ifq",
3363 		       SYSCTL_DESCR("Protocol input queue controls"),
3364 		       NULL, 0, NULL, 0,
3365 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3366 
3367 	sysctl_createv(clog, 0, NULL, NULL,
3368 		       CTLFLAG_PERMANENT,
3369 		       CTLTYPE_INT, "len",
3370 		       SYSCTL_DESCR("Current input queue length"),
3371 		       len_func, 0, NULL, 0,
3372 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3373 	sysctl_createv(clog, 0, NULL, NULL,
3374 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3375 		       CTLTYPE_INT, "maxlen",
3376 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3377 		       maxlen_func, 0, NULL, 0,
3378 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3379 	sysctl_createv(clog, 0, NULL, NULL,
3380 		       CTLFLAG_PERMANENT,
3381 		       CTLTYPE_INT, "drops",
3382 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3383 		       drops_func, 0, NULL, 0,
3384 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3385 }
3386 #endif /* INET || INET6 */
3387 
3388 static int
3389 if_sdl_sysctl(SYSCTLFN_ARGS)
3390 {
3391 	struct ifnet *ifp;
3392 	const struct sockaddr_dl *sdl;
3393 	struct psref psref;
3394 	int error = 0;
3395 	int bound;
3396 
3397 	if (namelen != 1)
3398 		return EINVAL;
3399 
3400 	bound = curlwp_bind();
3401 	ifp = if_get_byindex(name[0], &psref);
3402 	if (ifp == NULL) {
3403 		error = ENODEV;
3404 		goto out0;
3405 	}
3406 
3407 	sdl = ifp->if_sadl;
3408 	if (sdl == NULL) {
3409 		*oldlenp = 0;
3410 		goto out1;
3411 	}
3412 
3413 	if (oldp == NULL) {
3414 		*oldlenp = sdl->sdl_alen;
3415 		goto out1;
3416 	}
3417 
3418 	if (*oldlenp >= sdl->sdl_alen)
3419 		*oldlenp = sdl->sdl_alen;
3420 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3421 out1:
3422 	if_put(ifp, &psref);
3423 out0:
3424 	curlwp_bindx(bound);
3425 	return error;
3426 }
3427 
3428 SYSCTL_SETUP(sysctl_net_sdl_setup, "sysctl net.sdl subtree setup")
3429 {
3430 	const struct sysctlnode *rnode = NULL;
3431 
3432 	sysctl_createv(clog, 0, NULL, &rnode,
3433 		       CTLFLAG_PERMANENT,
3434 		       CTLTYPE_NODE, "sdl",
3435 		       SYSCTL_DESCR("Get active link-layer address"),
3436 		       if_sdl_sysctl, 0, NULL, 0,
3437 		       CTL_NET, CTL_CREATE, CTL_EOL);
3438 }
3439