xref: /netbsd-src/sys/net/if.c (revision 479d8f7d843cc1b22d497efdf1f27a50ee8418d4)
1 /*	$NetBSD: if.c,v 1.369 2016/12/26 23:21:49 christos 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 1.369 2016/12/26 23:21:49 christos 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 #include <net/bpf.h>
145 
146 #ifdef INET6
147 #include <netinet6/in6_var.h>
148 #include <netinet6/nd6.h>
149 #endif
150 
151 #include "ether.h"
152 #include "fddi.h"
153 #include "token.h"
154 
155 #include "carp.h"
156 #if NCARP > 0
157 #include <netinet/ip_carp.h>
158 #endif
159 
160 #include <compat/sys/sockio.h>
161 #include <compat/sys/socket.h>
162 
163 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
164 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
165 
166 /*
167  * Global list of interfaces.
168  */
169 /* DEPRECATED. Remove it once kvm(3) users disappeared */
170 struct ifnet_head		ifnet_list;
171 
172 struct pslist_head		ifnet_pslist;
173 static ifnet_t **		ifindex2ifnet = NULL;
174 static u_int			if_index = 1;
175 static size_t			if_indexlim = 0;
176 static uint64_t			index_gen;
177 /* Mutex to protect the above objects. */
178 kmutex_t			ifnet_mtx __cacheline_aligned;
179 struct psref_class		*ifnet_psref_class __read_mostly;
180 static pserialize_t		ifnet_psz;
181 
182 static kmutex_t			if_clone_mtx;
183 
184 struct ifnet *lo0ifp;
185 int	ifqmaxlen = IFQ_MAXLEN;
186 
187 struct psref_class		*ifa_psref_class __read_mostly;
188 
189 static int	if_delroute_matcher(struct rtentry *, void *);
190 
191 static struct if_clone *if_clone_lookup(const char *, int *);
192 
193 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
194 static int if_cloners_count;
195 
196 /* Packet filtering hook for interfaces. */
197 pfil_head_t *			if_pfil __read_mostly;
198 
199 static kauth_listener_t if_listener;
200 
201 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
202 static void if_detach_queues(struct ifnet *, struct ifqueue *);
203 static void sysctl_sndq_setup(struct sysctllog **, const char *,
204     struct ifaltq *);
205 static void if_slowtimo(void *);
206 static void if_free_sadl(struct ifnet *);
207 static void if_attachdomain1(struct ifnet *);
208 static int ifconf(u_long, void *);
209 static int if_transmit(struct ifnet *, struct mbuf *);
210 static int if_clone_create(const char *);
211 static int if_clone_destroy(const char *);
212 static void if_link_state_change_si(void *);
213 
214 struct if_percpuq {
215 	struct ifnet	*ipq_ifp;
216 	void		*ipq_si;
217 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
218 };
219 
220 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
221 
222 static void if_percpuq_drops(void *, void *, struct cpu_info *);
223 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
224 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
225     struct if_percpuq *);
226 
227 struct if_deferred_start {
228 	struct ifnet	*ids_ifp;
229 	void		(*ids_if_start)(struct ifnet *);
230 	void		*ids_si;
231 };
232 
233 static void if_deferred_start_softint(void *);
234 static void if_deferred_start_common(struct ifnet *);
235 static void if_deferred_start_destroy(struct ifnet *);
236 
237 #if defined(INET) || defined(INET6)
238 static void sysctl_net_pktq_setup(struct sysctllog **, int);
239 #endif
240 
241 /*
242  * Pointer to stub or real compat_cvtcmd() depending on presence of
243  * the compat module
244  */
245 u_long stub_compat_cvtcmd(u_long);
246 u_long (*vec_compat_cvtcmd)(u_long) = stub_compat_cvtcmd;
247 
248 /* Similarly, pointer to compat_ifioctl() if it is present */
249 
250 int (*vec_compat_ifioctl)(struct socket *, u_long, u_long, void *,
251 	struct lwp *) = NULL;
252 
253 /* The stub version of compat_cvtcmd() */
254 u_long stub_compat_cvtcmd(u_long cmd)
255 {
256 
257 	return cmd;
258 }
259 
260 static int
261 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
262     void *arg0, void *arg1, void *arg2, void *arg3)
263 {
264 	int result;
265 	enum kauth_network_req req;
266 
267 	result = KAUTH_RESULT_DEFER;
268 	req = (enum kauth_network_req)arg1;
269 
270 	if (action != KAUTH_NETWORK_INTERFACE)
271 		return result;
272 
273 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
274 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
275 		result = KAUTH_RESULT_ALLOW;
276 
277 	return result;
278 }
279 
280 /*
281  * Network interface utility routines.
282  *
283  * Routines with ifa_ifwith* names take sockaddr *'s as
284  * parameters.
285  */
286 void
287 ifinit(void)
288 {
289 #if defined(INET)
290 	sysctl_net_pktq_setup(NULL, PF_INET);
291 #endif
292 #ifdef INET6
293 	if (in6_present)
294 		sysctl_net_pktq_setup(NULL, PF_INET6);
295 #endif
296 
297 #if (defined(INET) || defined(INET6)) && !defined(IPSEC)
298 	encapinit();
299 #endif
300 
301 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
302 	    if_listener_cb, NULL);
303 
304 	/* interfaces are available, inform socket code */
305 	ifioctl = doifioctl;
306 }
307 
308 /*
309  * XXX Initialization before configure().
310  * XXX hack to get pfil_add_hook working in autoconf.
311  */
312 void
313 ifinit1(void)
314 {
315 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
316 
317 	TAILQ_INIT(&ifnet_list);
318 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
319 	ifnet_psz = pserialize_create();
320 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
321 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
322 	PSLIST_INIT(&ifnet_pslist);
323 
324 	if_indexlim = 8;
325 
326 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
327 	KASSERT(if_pfil != NULL);
328 
329 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
330 	etherinit();
331 #endif
332 }
333 
334 ifnet_t *
335 if_alloc(u_char type)
336 {
337 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
338 }
339 
340 void
341 if_free(ifnet_t *ifp)
342 {
343 	kmem_free(ifp, sizeof(ifnet_t));
344 }
345 
346 void
347 if_initname(struct ifnet *ifp, const char *name, int unit)
348 {
349 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
350 	    "%s%d", name, unit);
351 }
352 
353 /*
354  * Null routines used while an interface is going away.  These routines
355  * just return an error.
356  */
357 
358 int
359 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
360     const struct sockaddr *so, const struct rtentry *rt)
361 {
362 
363 	return ENXIO;
364 }
365 
366 void
367 if_nullinput(struct ifnet *ifp, struct mbuf *m)
368 {
369 
370 	/* Nothing. */
371 }
372 
373 void
374 if_nullstart(struct ifnet *ifp)
375 {
376 
377 	/* Nothing. */
378 }
379 
380 int
381 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
382 {
383 
384 	return ENXIO;
385 }
386 
387 int
388 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
389 {
390 
391 	return ENXIO;
392 }
393 
394 int
395 if_nullinit(struct ifnet *ifp)
396 {
397 
398 	return ENXIO;
399 }
400 
401 void
402 if_nullstop(struct ifnet *ifp, int disable)
403 {
404 
405 	/* Nothing. */
406 }
407 
408 void
409 if_nullslowtimo(struct ifnet *ifp)
410 {
411 
412 	/* Nothing. */
413 }
414 
415 void
416 if_nulldrain(struct ifnet *ifp)
417 {
418 
419 	/* Nothing. */
420 }
421 
422 void
423 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
424 {
425 	struct ifaddr *ifa;
426 	struct sockaddr_dl *sdl;
427 
428 	ifp->if_addrlen = addrlen;
429 	if_alloc_sadl(ifp);
430 	ifa = ifp->if_dl;
431 	sdl = satosdl(ifa->ifa_addr);
432 
433 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
434 	if (factory) {
435 		ifp->if_hwdl = ifp->if_dl;
436 		ifaref(ifp->if_hwdl);
437 	}
438 	/* TBD routing socket */
439 }
440 
441 struct ifaddr *
442 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
443 {
444 	unsigned socksize, ifasize;
445 	int addrlen, namelen;
446 	struct sockaddr_dl *mask, *sdl;
447 	struct ifaddr *ifa;
448 
449 	namelen = strlen(ifp->if_xname);
450 	addrlen = ifp->if_addrlen;
451 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
452 	ifasize = sizeof(*ifa) + 2 * socksize;
453 	ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
454 
455 	sdl = (struct sockaddr_dl *)(ifa + 1);
456 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
457 
458 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
459 	    ifp->if_xname, namelen, NULL, addrlen);
460 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
461 	memset(&mask->sdl_data[0], 0xff, namelen);
462 	ifa->ifa_rtrequest = link_rtrequest;
463 	ifa->ifa_addr = (struct sockaddr *)sdl;
464 	ifa->ifa_netmask = (struct sockaddr *)mask;
465 	ifa_psref_init(ifa);
466 
467 	*sdlp = sdl;
468 
469 	return ifa;
470 }
471 
472 static void
473 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
474 {
475 	const struct sockaddr_dl *sdl;
476 
477 	ifp->if_dl = ifa;
478 	ifaref(ifa);
479 	sdl = satosdl(ifa->ifa_addr);
480 	ifp->if_sadl = sdl;
481 }
482 
483 /*
484  * Allocate the link level name for the specified interface.  This
485  * is an attachment helper.  It must be called after ifp->if_addrlen
486  * is initialized, which may not be the case when if_attach() is
487  * called.
488  */
489 void
490 if_alloc_sadl(struct ifnet *ifp)
491 {
492 	struct ifaddr *ifa;
493 	const struct sockaddr_dl *sdl;
494 
495 	/*
496 	 * If the interface already has a link name, release it
497 	 * now.  This is useful for interfaces that can change
498 	 * link types, and thus switch link names often.
499 	 */
500 	if (ifp->if_sadl != NULL)
501 		if_free_sadl(ifp);
502 
503 	ifa = if_dl_create(ifp, &sdl);
504 
505 	ifa_insert(ifp, ifa);
506 	if_sadl_setrefs(ifp, ifa);
507 }
508 
509 static void
510 if_deactivate_sadl(struct ifnet *ifp)
511 {
512 	struct ifaddr *ifa;
513 
514 	KASSERT(ifp->if_dl != NULL);
515 
516 	ifa = ifp->if_dl;
517 
518 	ifp->if_sadl = NULL;
519 
520 	ifp->if_dl = NULL;
521 	ifafree(ifa);
522 }
523 
524 void
525 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
526     const struct sockaddr_dl *sdl)
527 {
528 	int s, ss;
529 	struct ifaddr *ifa;
530 	int bound = curlwp_bind();
531 
532 	s = splnet();
533 
534 	if_deactivate_sadl(ifp);
535 
536 	if_sadl_setrefs(ifp, ifa0);
537 
538 	ss = pserialize_read_enter();
539 	IFADDR_READER_FOREACH(ifa, ifp) {
540 		struct psref psref;
541 		ifa_acquire(ifa, &psref);
542 		pserialize_read_exit(ss);
543 
544 		rtinit(ifa, RTM_LLINFO_UPD, 0);
545 
546 		ss = pserialize_read_enter();
547 		ifa_release(ifa, &psref);
548 	}
549 	pserialize_read_exit(ss);
550 
551 	splx(s);
552 	curlwp_bindx(bound);
553 }
554 
555 /*
556  * Free the link level name for the specified interface.  This is
557  * a detach helper.  This is called from if_detach().
558  */
559 static void
560 if_free_sadl(struct ifnet *ifp)
561 {
562 	struct ifaddr *ifa;
563 	int s;
564 
565 	ifa = ifp->if_dl;
566 	if (ifa == NULL) {
567 		KASSERT(ifp->if_sadl == NULL);
568 		return;
569 	}
570 
571 	KASSERT(ifp->if_sadl != NULL);
572 
573 	s = splnet();
574 	rtinit(ifa, RTM_DELETE, 0);
575 	ifa_remove(ifp, ifa);
576 	if_deactivate_sadl(ifp);
577 	if (ifp->if_hwdl == ifa) {
578 		ifafree(ifa);
579 		ifp->if_hwdl = NULL;
580 	}
581 	splx(s);
582 }
583 
584 static void
585 if_getindex(ifnet_t *ifp)
586 {
587 	bool hitlimit = false;
588 
589 	ifp->if_index_gen = index_gen++;
590 
591 	ifp->if_index = if_index;
592 	if (ifindex2ifnet == NULL) {
593 		if_index++;
594 		goto skip;
595 	}
596 	while (if_byindex(ifp->if_index)) {
597 		/*
598 		 * If we hit USHRT_MAX, we skip back to 0 since
599 		 * there are a number of places where the value
600 		 * of if_index or if_index itself is compared
601 		 * to or stored in an unsigned short.  By
602 		 * jumping back, we won't botch those assignments
603 		 * or comparisons.
604 		 */
605 		if (++if_index == 0) {
606 			if_index = 1;
607 		} else if (if_index == USHRT_MAX) {
608 			/*
609 			 * However, if we have to jump back to
610 			 * zero *twice* without finding an empty
611 			 * slot in ifindex2ifnet[], then there
612 			 * there are too many (>65535) interfaces.
613 			 */
614 			if (hitlimit) {
615 				panic("too many interfaces");
616 			}
617 			hitlimit = true;
618 			if_index = 1;
619 		}
620 		ifp->if_index = if_index;
621 	}
622 skip:
623 	/*
624 	 * ifindex2ifnet is indexed by if_index. Since if_index will
625 	 * grow dynamically, it should grow too.
626 	 */
627 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
628 		size_t m, n, oldlim;
629 		void *q;
630 
631 		oldlim = if_indexlim;
632 		while (ifp->if_index >= if_indexlim)
633 			if_indexlim <<= 1;
634 
635 		/* grow ifindex2ifnet */
636 		m = oldlim * sizeof(struct ifnet *);
637 		n = if_indexlim * sizeof(struct ifnet *);
638 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
639 		if (ifindex2ifnet != NULL) {
640 			memcpy(q, ifindex2ifnet, m);
641 			free(ifindex2ifnet, M_IFADDR);
642 		}
643 		ifindex2ifnet = (struct ifnet **)q;
644 	}
645 	ifindex2ifnet[ifp->if_index] = ifp;
646 }
647 
648 /*
649  * Initialize an interface and assign an index for it.
650  *
651  * It must be called prior to a device specific attach routine
652  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
653  * and be followed by if_register:
654  *
655  *     if_initialize(ifp);
656  *     ether_ifattach(ifp, enaddr);
657  *     if_register(ifp);
658  */
659 void
660 if_initialize(ifnet_t *ifp)
661 {
662 	KASSERT(if_indexlim > 0);
663 	TAILQ_INIT(&ifp->if_addrlist);
664 
665 	/*
666 	 * Link level name is allocated later by a separate call to
667 	 * if_alloc_sadl().
668 	 */
669 
670 	if (ifp->if_snd.ifq_maxlen == 0)
671 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
672 
673 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
674 
675 	ifp->if_link_state = LINK_STATE_UNKNOWN;
676 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
677 
678 	ifp->if_capenable = 0;
679 	ifp->if_csum_flags_tx = 0;
680 	ifp->if_csum_flags_rx = 0;
681 
682 #ifdef ALTQ
683 	ifp->if_snd.altq_type = 0;
684 	ifp->if_snd.altq_disc = NULL;
685 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
686 	ifp->if_snd.altq_tbr  = NULL;
687 	ifp->if_snd.altq_ifp  = ifp;
688 #endif
689 
690 	IFQ_LOCK_INIT(&ifp->if_snd);
691 
692 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
693 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
694 
695 	IF_AFDATA_LOCK_INIT(ifp);
696 
697 	if (if_is_link_state_changeable(ifp)) {
698 		ifp->if_link_si = softint_establish(SOFTINT_NET,
699 		    if_link_state_change_si, ifp);
700 		if (ifp->if_link_si == NULL)
701 			panic("%s: softint_establish() failed", __func__);
702 	}
703 
704 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
705 	PSLIST_INIT(&ifp->if_addr_pslist);
706 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
707 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
708 
709 	IFNET_LOCK();
710 	if_getindex(ifp);
711 	IFNET_UNLOCK();
712 }
713 
714 /*
715  * Register an interface to the list of "active" interfaces.
716  */
717 void
718 if_register(ifnet_t *ifp)
719 {
720 	/*
721 	 * If the driver has not supplied its own if_ioctl, then
722 	 * supply the default.
723 	 */
724 	if (ifp->if_ioctl == NULL)
725 		ifp->if_ioctl = ifioctl_common;
726 
727 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
728 
729 	if (!STAILQ_EMPTY(&domains))
730 		if_attachdomain1(ifp);
731 
732 	/* Announce the interface. */
733 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
734 
735 	if (ifp->if_slowtimo != NULL) {
736 		ifp->if_slowtimo_ch =
737 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
738 		callout_init(ifp->if_slowtimo_ch, 0);
739 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
740 		if_slowtimo(ifp);
741 	}
742 
743 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
744 		ifp->if_transmit = if_transmit;
745 
746 	IFNET_LOCK();
747 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
748 	IFNET_WRITER_INSERT_TAIL(ifp);
749 	IFNET_UNLOCK();
750 }
751 
752 /*
753  * The if_percpuq framework
754  *
755  * It allows network device drivers to execute the network stack
756  * in softint (so called softint-based if_input). It utilizes
757  * softint and percpu ifqueue. It doesn't distribute any packets
758  * between CPUs, unlike pktqueue(9).
759  *
760  * Currently we support two options for device drivers to apply the framework:
761  * - Use it implicitly with less changes
762  *   - If you use if_attach in driver's _attach function and if_input in
763  *     driver's Rx interrupt handler, a packet is queued and a softint handles
764  *     the packet implicitly
765  * - Use it explicitly in each driver (recommended)
766  *   - You can use if_percpuq_* directly in your driver
767  *   - In this case, you need to allocate struct if_percpuq in driver's softc
768  *   - See wm(4) as a reference implementation
769  */
770 
771 static void
772 if_percpuq_softint(void *arg)
773 {
774 	struct if_percpuq *ipq = arg;
775 	struct ifnet *ifp = ipq->ipq_ifp;
776 	struct mbuf *m;
777 
778 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
779 		ifp->if_ipackets++;
780 #ifndef NET_MPSAFE
781 		KERNEL_LOCK(1, NULL);
782 #endif
783 		bpf_mtap(ifp, m);
784 #ifndef NET_MPSAFE
785 		KERNEL_UNLOCK_ONE(NULL);
786 #endif
787 
788 		ifp->_if_input(ifp, m);
789 	}
790 }
791 
792 static void
793 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
794 {
795 	struct ifqueue *const ifq = p;
796 
797 	memset(ifq, 0, sizeof(*ifq));
798 	ifq->ifq_maxlen = IFQ_MAXLEN;
799 }
800 
801 struct if_percpuq *
802 if_percpuq_create(struct ifnet *ifp)
803 {
804 	struct if_percpuq *ipq;
805 
806 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
807 	if (ipq == NULL)
808 		panic("kmem_zalloc failed");
809 
810 	ipq->ipq_ifp = ifp;
811 	ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
812 	    if_percpuq_softint, ipq);
813 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
814 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
815 
816 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
817 
818 	return ipq;
819 }
820 
821 static struct mbuf *
822 if_percpuq_dequeue(struct if_percpuq *ipq)
823 {
824 	struct mbuf *m;
825 	struct ifqueue *ifq;
826 	int s;
827 
828 	s = splnet();
829 	ifq = percpu_getref(ipq->ipq_ifqs);
830 	IF_DEQUEUE(ifq, m);
831 	percpu_putref(ipq->ipq_ifqs);
832 	splx(s);
833 
834 	return m;
835 }
836 
837 static void
838 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
839 {
840 	struct ifqueue *const ifq = p;
841 
842 	IF_PURGE(ifq);
843 }
844 
845 void
846 if_percpuq_destroy(struct if_percpuq *ipq)
847 {
848 
849 	/* if_detach may already destroy it */
850 	if (ipq == NULL)
851 		return;
852 
853 	softint_disestablish(ipq->ipq_si);
854 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
855 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
856 	kmem_free(ipq, sizeof(*ipq));
857 }
858 
859 void
860 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
861 {
862 	struct ifqueue *ifq;
863 	int s;
864 
865 	KASSERT(ipq != NULL);
866 
867 	s = splnet();
868 	ifq = percpu_getref(ipq->ipq_ifqs);
869 	if (IF_QFULL(ifq)) {
870 		IF_DROP(ifq);
871 		percpu_putref(ipq->ipq_ifqs);
872 		m_freem(m);
873 		goto out;
874 	}
875 	IF_ENQUEUE(ifq, m);
876 	percpu_putref(ipq->ipq_ifqs);
877 
878 	softint_schedule(ipq->ipq_si);
879 out:
880 	splx(s);
881 }
882 
883 static void
884 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
885 {
886 	struct ifqueue *const ifq = p;
887 	int *sum = arg;
888 
889 	*sum += ifq->ifq_drops;
890 }
891 
892 static int
893 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
894 {
895 	struct sysctlnode node;
896 	struct if_percpuq *ipq;
897 	int sum = 0;
898 	int error;
899 
900 	node = *rnode;
901 	ipq = node.sysctl_data;
902 
903 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
904 
905 	node.sysctl_data = &sum;
906 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
907 	if (error != 0 || newp == NULL)
908 		return error;
909 
910 	return 0;
911 }
912 
913 static void
914 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
915     struct if_percpuq *ipq)
916 {
917 	const struct sysctlnode *cnode, *rnode;
918 
919 	if (sysctl_createv(clog, 0, NULL, &rnode,
920 		       CTLFLAG_PERMANENT,
921 		       CTLTYPE_NODE, "interfaces",
922 		       SYSCTL_DESCR("Per-interface controls"),
923 		       NULL, 0, NULL, 0,
924 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
925 		goto bad;
926 
927 	if (sysctl_createv(clog, 0, &rnode, &rnode,
928 		       CTLFLAG_PERMANENT,
929 		       CTLTYPE_NODE, ifname,
930 		       SYSCTL_DESCR("Interface controls"),
931 		       NULL, 0, NULL, 0,
932 		       CTL_CREATE, CTL_EOL) != 0)
933 		goto bad;
934 
935 	if (sysctl_createv(clog, 0, &rnode, &rnode,
936 		       CTLFLAG_PERMANENT,
937 		       CTLTYPE_NODE, "rcvq",
938 		       SYSCTL_DESCR("Interface input queue controls"),
939 		       NULL, 0, NULL, 0,
940 		       CTL_CREATE, CTL_EOL) != 0)
941 		goto bad;
942 
943 #ifdef NOTYET
944 	/* XXX Should show each per-CPU queue length? */
945 	if (sysctl_createv(clog, 0, &rnode, &rnode,
946 		       CTLFLAG_PERMANENT,
947 		       CTLTYPE_INT, "len",
948 		       SYSCTL_DESCR("Current input queue length"),
949 		       sysctl_percpuq_len, 0, NULL, 0,
950 		       CTL_CREATE, CTL_EOL) != 0)
951 		goto bad;
952 
953 	if (sysctl_createv(clog, 0, &rnode, &cnode,
954 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
955 		       CTLTYPE_INT, "maxlen",
956 		       SYSCTL_DESCR("Maximum allowed input queue length"),
957 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
958 		       CTL_CREATE, CTL_EOL) != 0)
959 		goto bad;
960 #endif
961 
962 	if (sysctl_createv(clog, 0, &rnode, &cnode,
963 		       CTLFLAG_PERMANENT,
964 		       CTLTYPE_INT, "drops",
965 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
966 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
967 		       CTL_CREATE, CTL_EOL) != 0)
968 		goto bad;
969 
970 	return;
971 bad:
972 	printf("%s: could not attach sysctl nodes\n", ifname);
973 	return;
974 }
975 
976 /*
977  * The deferred if_start framework
978  *
979  * The common APIs to defer if_start to softint when if_start is requested
980  * from a device driver running in hardware interrupt context.
981  */
982 /*
983  * Call ifp->if_start (or equivalent) in a dedicated softint for
984  * deferred if_start.
985  */
986 static void
987 if_deferred_start_softint(void *arg)
988 {
989 	struct if_deferred_start *ids = arg;
990 	struct ifnet *ifp = ids->ids_ifp;
991 
992 	ids->ids_if_start(ifp);
993 }
994 
995 /*
996  * The default callback function for deferred if_start.
997  */
998 static void
999 if_deferred_start_common(struct ifnet *ifp)
1000 {
1001 
1002 	if_start_lock(ifp);
1003 }
1004 
1005 static inline bool
1006 if_snd_is_used(struct ifnet *ifp)
1007 {
1008 
1009 	return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit ||
1010 	    ALTQ_IS_ENABLED(&ifp->if_snd);
1011 }
1012 
1013 /*
1014  * Schedule deferred if_start.
1015  */
1016 void
1017 if_schedule_deferred_start(struct ifnet *ifp)
1018 {
1019 
1020 	KASSERT(ifp->if_deferred_start != NULL);
1021 
1022 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1023 		return;
1024 
1025 	softint_schedule(ifp->if_deferred_start->ids_si);
1026 }
1027 
1028 /*
1029  * Create an instance of deferred if_start. A driver should call the function
1030  * only if the driver needs deferred if_start. Drivers can setup their own
1031  * deferred if_start function via 2nd argument.
1032  */
1033 void
1034 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1035 {
1036 	struct if_deferred_start *ids;
1037 
1038 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1039 	if (ids == NULL)
1040 		panic("kmem_zalloc failed");
1041 
1042 	ids->ids_ifp = ifp;
1043 	ids->ids_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
1044 	    if_deferred_start_softint, ids);
1045 	if (func != NULL)
1046 		ids->ids_if_start = func;
1047 	else
1048 		ids->ids_if_start = if_deferred_start_common;
1049 
1050 	ifp->if_deferred_start = ids;
1051 }
1052 
1053 static void
1054 if_deferred_start_destroy(struct ifnet *ifp)
1055 {
1056 
1057 	if (ifp->if_deferred_start == NULL)
1058 		return;
1059 
1060 	softint_disestablish(ifp->if_deferred_start->ids_si);
1061 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1062 	ifp->if_deferred_start = NULL;
1063 }
1064 
1065 /*
1066  * The common interface input routine that is called by device drivers,
1067  * which should be used only when the driver's rx handler already runs
1068  * in softint.
1069  */
1070 void
1071 if_input(struct ifnet *ifp, struct mbuf *m)
1072 {
1073 
1074 	KASSERT(ifp->if_percpuq == NULL);
1075 	KASSERT(!cpu_intr_p());
1076 
1077 	ifp->if_ipackets++;
1078 #ifndef NET_MPSAFE
1079 	KERNEL_LOCK(1, NULL);
1080 #endif
1081 	bpf_mtap(ifp, m);
1082 #ifndef NET_MPSAFE
1083 	KERNEL_UNLOCK_ONE(NULL);
1084 #endif
1085 
1086 	ifp->_if_input(ifp, m);
1087 }
1088 
1089 /*
1090  * DEPRECATED. Use if_initialize and if_register instead.
1091  * See the above comment of if_initialize.
1092  *
1093  * Note that it implicitly enables if_percpuq to make drivers easy to
1094  * migrate softint-based if_input without much changes. If you don't
1095  * want to enable it, use if_initialize instead.
1096  */
1097 void
1098 if_attach(ifnet_t *ifp)
1099 {
1100 
1101 	if_initialize(ifp);
1102 	ifp->if_percpuq = if_percpuq_create(ifp);
1103 	if_register(ifp);
1104 }
1105 
1106 void
1107 if_attachdomain(void)
1108 {
1109 	struct ifnet *ifp;
1110 	int s;
1111 	int bound = curlwp_bind();
1112 
1113 	s = pserialize_read_enter();
1114 	IFNET_READER_FOREACH(ifp) {
1115 		struct psref psref;
1116 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1117 		pserialize_read_exit(s);
1118 		if_attachdomain1(ifp);
1119 		s = pserialize_read_enter();
1120 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1121 	}
1122 	pserialize_read_exit(s);
1123 	curlwp_bindx(bound);
1124 }
1125 
1126 static void
1127 if_attachdomain1(struct ifnet *ifp)
1128 {
1129 	struct domain *dp;
1130 	int s;
1131 
1132 	s = splnet();
1133 
1134 	/* address family dependent data region */
1135 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1136 	DOMAIN_FOREACH(dp) {
1137 		if (dp->dom_ifattach != NULL)
1138 			ifp->if_afdata[dp->dom_family] =
1139 			    (*dp->dom_ifattach)(ifp);
1140 	}
1141 
1142 	splx(s);
1143 }
1144 
1145 /*
1146  * Deactivate an interface.  This points all of the procedure
1147  * handles at error stubs.  May be called from interrupt context.
1148  */
1149 void
1150 if_deactivate(struct ifnet *ifp)
1151 {
1152 	int s;
1153 
1154 	s = splnet();
1155 
1156 	ifp->if_output	 = if_nulloutput;
1157 	ifp->_if_input	 = if_nullinput;
1158 	ifp->if_start	 = if_nullstart;
1159 	ifp->if_transmit = if_nulltransmit;
1160 	ifp->if_ioctl	 = if_nullioctl;
1161 	ifp->if_init	 = if_nullinit;
1162 	ifp->if_stop	 = if_nullstop;
1163 	ifp->if_slowtimo = if_nullslowtimo;
1164 	ifp->if_drain	 = if_nulldrain;
1165 
1166 	/* No more packets may be enqueued. */
1167 	ifp->if_snd.ifq_maxlen = 0;
1168 
1169 	splx(s);
1170 }
1171 
1172 bool
1173 if_is_deactivated(const struct ifnet *ifp)
1174 {
1175 
1176 	return ifp->if_output == if_nulloutput;
1177 }
1178 
1179 void
1180 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1181 {
1182 	struct ifaddr *ifa, *nifa;
1183 	int s;
1184 
1185 	s = pserialize_read_enter();
1186 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1187 		nifa = IFADDR_READER_NEXT(ifa);
1188 		if (ifa->ifa_addr->sa_family != family)
1189 			continue;
1190 		pserialize_read_exit(s);
1191 
1192 		(*purgeaddr)(ifa);
1193 
1194 		s = pserialize_read_enter();
1195 	}
1196 	pserialize_read_exit(s);
1197 }
1198 
1199 #ifdef IFAREF_DEBUG
1200 static struct ifaddr **ifa_list;
1201 static int ifa_list_size;
1202 
1203 /* Depends on only one if_attach runs at once */
1204 static void
1205 if_build_ifa_list(struct ifnet *ifp)
1206 {
1207 	struct ifaddr *ifa;
1208 	int i;
1209 
1210 	KASSERT(ifa_list == NULL);
1211 	KASSERT(ifa_list_size == 0);
1212 
1213 	IFADDR_READER_FOREACH(ifa, ifp)
1214 		ifa_list_size++;
1215 
1216 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1217 	if (ifa_list == NULL)
1218 		return;
1219 
1220 	i = 0;
1221 	IFADDR_READER_FOREACH(ifa, ifp) {
1222 		ifa_list[i++] = ifa;
1223 		ifaref(ifa);
1224 	}
1225 }
1226 
1227 static void
1228 if_check_and_free_ifa_list(struct ifnet *ifp)
1229 {
1230 	int i;
1231 	struct ifaddr *ifa;
1232 
1233 	if (ifa_list == NULL)
1234 		return;
1235 
1236 	for (i = 0; i < ifa_list_size; i++) {
1237 		char buf[64];
1238 
1239 		ifa = ifa_list[i];
1240 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1241 		if (ifa->ifa_refcnt > 1) {
1242 			log(LOG_WARNING,
1243 			    "ifa(%s) still referenced (refcnt=%d)\n",
1244 			    buf, ifa->ifa_refcnt - 1);
1245 		} else
1246 			log(LOG_DEBUG,
1247 			    "ifa(%s) not referenced (refcnt=%d)\n",
1248 			    buf, ifa->ifa_refcnt - 1);
1249 		ifafree(ifa);
1250 	}
1251 
1252 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1253 	ifa_list = NULL;
1254 	ifa_list_size = 0;
1255 }
1256 #endif
1257 
1258 /*
1259  * Detach an interface from the list of "active" interfaces,
1260  * freeing any resources as we go along.
1261  *
1262  * NOTE: This routine must be called with a valid thread context,
1263  * as it may block.
1264  */
1265 void
1266 if_detach(struct ifnet *ifp)
1267 {
1268 	struct socket so;
1269 	struct ifaddr *ifa;
1270 #ifdef IFAREF_DEBUG
1271 	struct ifaddr *last_ifa = NULL;
1272 #endif
1273 	struct domain *dp;
1274 	const struct protosw *pr;
1275 	int s, i, family, purged;
1276 	uint64_t xc;
1277 
1278 #ifdef IFAREF_DEBUG
1279 	if_build_ifa_list(ifp);
1280 #endif
1281 	/*
1282 	 * XXX It's kind of lame that we have to have the
1283 	 * XXX socket structure...
1284 	 */
1285 	memset(&so, 0, sizeof(so));
1286 
1287 	s = splnet();
1288 
1289 	sysctl_teardown(&ifp->if_sysctl_log);
1290 	mutex_enter(ifp->if_ioctl_lock);
1291 	if_deactivate(ifp);
1292 	mutex_exit(ifp->if_ioctl_lock);
1293 
1294 	IFNET_LOCK();
1295 	ifindex2ifnet[ifp->if_index] = NULL;
1296 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1297 	IFNET_WRITER_REMOVE(ifp);
1298 	pserialize_perform(ifnet_psz);
1299 	IFNET_UNLOCK();
1300 
1301 	/* Wait for all readers to drain before freeing.  */
1302 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1303 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1304 
1305 	mutex_obj_free(ifp->if_ioctl_lock);
1306 	ifp->if_ioctl_lock = NULL;
1307 
1308 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1309 		ifp->if_slowtimo = NULL;
1310 		callout_halt(ifp->if_slowtimo_ch, NULL);
1311 		callout_destroy(ifp->if_slowtimo_ch);
1312 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1313 	}
1314 	if_deferred_start_destroy(ifp);
1315 
1316 	/*
1317 	 * Do an if_down() to give protocols a chance to do something.
1318 	 */
1319 	if_down(ifp);
1320 
1321 #ifdef ALTQ
1322 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1323 		altq_disable(&ifp->if_snd);
1324 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1325 		altq_detach(&ifp->if_snd);
1326 #endif
1327 
1328 	mutex_obj_free(ifp->if_snd.ifq_lock);
1329 
1330 #if NCARP > 0
1331 	/* Remove the interface from any carp group it is a part of.  */
1332 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1333 		carp_ifdetach(ifp);
1334 #endif
1335 
1336 	/*
1337 	 * Rip all the addresses off the interface.  This should make
1338 	 * all of the routes go away.
1339 	 *
1340 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1341 	 * from the list, including our "cursor", ifa.  For safety,
1342 	 * and to honor the TAILQ abstraction, I just restart the
1343 	 * loop after each removal.  Note that the loop will exit
1344 	 * when all of the remaining ifaddrs belong to the AF_LINK
1345 	 * family.  I am counting on the historical fact that at
1346 	 * least one pr_usrreq in each address domain removes at
1347 	 * least one ifaddr.
1348 	 */
1349 again:
1350 	/*
1351 	 * At this point, no other one tries to remove ifa in the list,
1352 	 * so we don't need to take a lock or psref.
1353 	 */
1354 	IFADDR_READER_FOREACH(ifa, ifp) {
1355 		family = ifa->ifa_addr->sa_family;
1356 #ifdef IFAREF_DEBUG
1357 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1358 		    ifa, family, ifa->ifa_refcnt);
1359 		if (last_ifa != NULL && ifa == last_ifa)
1360 			panic("if_detach: loop detected");
1361 		last_ifa = ifa;
1362 #endif
1363 		if (family == AF_LINK)
1364 			continue;
1365 		dp = pffinddomain(family);
1366 #ifdef DIAGNOSTIC
1367 		if (dp == NULL)
1368 			panic("if_detach: no domain for AF %d",
1369 			    family);
1370 #endif
1371 		/*
1372 		 * XXX These PURGEIF calls are redundant with the
1373 		 * purge-all-families calls below, but are left in for
1374 		 * now both to make a smaller change, and to avoid
1375 		 * unplanned interactions with clearing of
1376 		 * ifp->if_addrlist.
1377 		 */
1378 		purged = 0;
1379 		for (pr = dp->dom_protosw;
1380 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1381 			so.so_proto = pr;
1382 			if (pr->pr_usrreqs) {
1383 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1384 				purged = 1;
1385 			}
1386 		}
1387 		if (purged == 0) {
1388 			/*
1389 			 * XXX What's really the best thing to do
1390 			 * XXX here?  --thorpej@NetBSD.org
1391 			 */
1392 			printf("if_detach: WARNING: AF %d not purged\n",
1393 			    family);
1394 			ifa_remove(ifp, ifa);
1395 		}
1396 		goto again;
1397 	}
1398 
1399 	if_free_sadl(ifp);
1400 
1401 	/* Delete stray routes from the routing table. */
1402 	for (i = 0; i <= AF_MAX; i++)
1403 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1404 
1405 	DOMAIN_FOREACH(dp) {
1406 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1407 		{
1408 			void *p = ifp->if_afdata[dp->dom_family];
1409 			if (p) {
1410 				ifp->if_afdata[dp->dom_family] = NULL;
1411 				(*dp->dom_ifdetach)(ifp, p);
1412 			}
1413 		}
1414 
1415 		/*
1416 		 * One would expect multicast memberships (INET and
1417 		 * INET6) on UDP sockets to be purged by the PURGEIF
1418 		 * calls above, but if all addresses were removed from
1419 		 * the interface prior to destruction, the calls will
1420 		 * not be made (e.g. ppp, for which pppd(8) generally
1421 		 * removes addresses before destroying the interface).
1422 		 * Because there is no invariant that multicast
1423 		 * memberships only exist for interfaces with IPv4
1424 		 * addresses, we must call PURGEIF regardless of
1425 		 * addresses.  (Protocols which might store ifnet
1426 		 * pointers are marked with PR_PURGEIF.)
1427 		 */
1428 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1429 			so.so_proto = pr;
1430 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1431 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1432 		}
1433 	}
1434 
1435 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1436 	(void)pfil_head_destroy(ifp->if_pfil);
1437 
1438 	/* Announce that the interface is gone. */
1439 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1440 
1441 	IF_AFDATA_LOCK_DESTROY(ifp);
1442 
1443 	if (if_is_link_state_changeable(ifp)) {
1444 		softint_disestablish(ifp->if_link_si);
1445 		ifp->if_link_si = NULL;
1446 	}
1447 
1448 	/*
1449 	 * remove packets that came from ifp, from software interrupt queues.
1450 	 */
1451 	DOMAIN_FOREACH(dp) {
1452 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1453 			struct ifqueue *iq = dp->dom_ifqueues[i];
1454 			if (iq == NULL)
1455 				break;
1456 			dp->dom_ifqueues[i] = NULL;
1457 			if_detach_queues(ifp, iq);
1458 		}
1459 	}
1460 
1461 	/*
1462 	 * IP queues have to be processed separately: net-queue barrier
1463 	 * ensures that the packets are dequeued while a cross-call will
1464 	 * ensure that the interrupts have completed. FIXME: not quite..
1465 	 */
1466 #ifdef INET
1467 	pktq_barrier(ip_pktq);
1468 #endif
1469 #ifdef INET6
1470 	if (in6_present)
1471 		pktq_barrier(ip6_pktq);
1472 #endif
1473 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1474 	xc_wait(xc);
1475 
1476 	if (ifp->if_percpuq != NULL) {
1477 		if_percpuq_destroy(ifp->if_percpuq);
1478 		ifp->if_percpuq = NULL;
1479 	}
1480 
1481 	splx(s);
1482 
1483 #ifdef IFAREF_DEBUG
1484 	if_check_and_free_ifa_list(ifp);
1485 #endif
1486 }
1487 
1488 static void
1489 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1490 {
1491 	struct mbuf *m, *prev, *next;
1492 
1493 	prev = NULL;
1494 	for (m = q->ifq_head; m != NULL; m = next) {
1495 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1496 
1497 		next = m->m_nextpkt;
1498 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1499 			prev = m;
1500 			continue;
1501 		}
1502 
1503 		if (prev != NULL)
1504 			prev->m_nextpkt = m->m_nextpkt;
1505 		else
1506 			q->ifq_head = m->m_nextpkt;
1507 		if (q->ifq_tail == m)
1508 			q->ifq_tail = prev;
1509 		q->ifq_len--;
1510 
1511 		m->m_nextpkt = NULL;
1512 		m_freem(m);
1513 		IF_DROP(q);
1514 	}
1515 }
1516 
1517 /*
1518  * Callback for a radix tree walk to delete all references to an
1519  * ifnet.
1520  */
1521 static int
1522 if_delroute_matcher(struct rtentry *rt, void *v)
1523 {
1524 	struct ifnet *ifp = (struct ifnet *)v;
1525 
1526 	if (rt->rt_ifp == ifp)
1527 		return 1;
1528 	else
1529 		return 0;
1530 }
1531 
1532 /*
1533  * Create a clone network interface.
1534  */
1535 static int
1536 if_clone_create(const char *name)
1537 {
1538 	struct if_clone *ifc;
1539 	int unit;
1540 	struct ifnet *ifp;
1541 	struct psref psref;
1542 
1543 	ifc = if_clone_lookup(name, &unit);
1544 	if (ifc == NULL)
1545 		return EINVAL;
1546 
1547 	ifp = if_get(name, &psref);
1548 	if (ifp != NULL) {
1549 		if_put(ifp, &psref);
1550 		return EEXIST;
1551 	}
1552 
1553 	return (*ifc->ifc_create)(ifc, unit);
1554 }
1555 
1556 /*
1557  * Destroy a clone network interface.
1558  */
1559 static int
1560 if_clone_destroy(const char *name)
1561 {
1562 	struct if_clone *ifc;
1563 	struct ifnet *ifp;
1564 	struct psref psref;
1565 
1566 	ifc = if_clone_lookup(name, NULL);
1567 	if (ifc == NULL)
1568 		return EINVAL;
1569 
1570 	if (ifc->ifc_destroy == NULL)
1571 		return EOPNOTSUPP;
1572 
1573 	ifp = if_get(name, &psref);
1574 	if (ifp == NULL)
1575 		return ENXIO;
1576 
1577 	/* We have to disable ioctls here */
1578 	mutex_enter(ifp->if_ioctl_lock);
1579 	ifp->if_ioctl = if_nullioctl;
1580 	mutex_exit(ifp->if_ioctl_lock);
1581 
1582 	/*
1583 	 * We cannot call ifc_destroy with holding ifp.
1584 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1585 	 */
1586 	if_put(ifp, &psref);
1587 
1588 	return (*ifc->ifc_destroy)(ifp);
1589 }
1590 
1591 /*
1592  * Look up a network interface cloner.
1593  */
1594 static struct if_clone *
1595 if_clone_lookup(const char *name, int *unitp)
1596 {
1597 	struct if_clone *ifc;
1598 	const char *cp;
1599 	char *dp, ifname[IFNAMSIZ + 3];
1600 	int unit;
1601 
1602 	strcpy(ifname, "if_");
1603 	/* separate interface name from unit */
1604 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1605 	    *cp && (*cp < '0' || *cp > '9');)
1606 		*dp++ = *cp++;
1607 
1608 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1609 		return NULL;	/* No name or unit number */
1610 	*dp++ = '\0';
1611 
1612 again:
1613 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1614 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1615 			break;
1616 	}
1617 
1618 	if (ifc == NULL) {
1619 		if (*ifname == '\0' ||
1620 		    module_autoload(ifname, MODULE_CLASS_DRIVER))
1621 			return NULL;
1622 		*ifname = '\0';
1623 		goto again;
1624 	}
1625 
1626 	unit = 0;
1627 	while (cp - name < IFNAMSIZ && *cp) {
1628 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1629 			/* Bogus unit number. */
1630 			return NULL;
1631 		}
1632 		unit = (unit * 10) + (*cp++ - '0');
1633 	}
1634 
1635 	if (unitp != NULL)
1636 		*unitp = unit;
1637 	return ifc;
1638 }
1639 
1640 /*
1641  * Register a network interface cloner.
1642  */
1643 void
1644 if_clone_attach(struct if_clone *ifc)
1645 {
1646 
1647 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1648 	if_cloners_count++;
1649 }
1650 
1651 /*
1652  * Unregister a network interface cloner.
1653  */
1654 void
1655 if_clone_detach(struct if_clone *ifc)
1656 {
1657 
1658 	LIST_REMOVE(ifc, ifc_list);
1659 	if_cloners_count--;
1660 }
1661 
1662 /*
1663  * Provide list of interface cloners to userspace.
1664  */
1665 int
1666 if_clone_list(int buf_count, char *buffer, int *total)
1667 {
1668 	char outbuf[IFNAMSIZ], *dst;
1669 	struct if_clone *ifc;
1670 	int count, error = 0;
1671 
1672 	*total = if_cloners_count;
1673 	if ((dst = buffer) == NULL) {
1674 		/* Just asking how many there are. */
1675 		return 0;
1676 	}
1677 
1678 	if (buf_count < 0)
1679 		return EINVAL;
1680 
1681 	count = (if_cloners_count < buf_count) ?
1682 	    if_cloners_count : buf_count;
1683 
1684 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1685 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1686 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1687 		if (outbuf[sizeof(outbuf) - 1] != '\0')
1688 			return ENAMETOOLONG;
1689 		error = copyout(outbuf, dst, sizeof(outbuf));
1690 		if (error != 0)
1691 			break;
1692 	}
1693 
1694 	return error;
1695 }
1696 
1697 void
1698 ifa_psref_init(struct ifaddr *ifa)
1699 {
1700 
1701 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1702 }
1703 
1704 void
1705 ifaref(struct ifaddr *ifa)
1706 {
1707 	KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING));
1708 	ifa->ifa_refcnt++;
1709 }
1710 
1711 void
1712 ifafree(struct ifaddr *ifa)
1713 {
1714 	KASSERT(ifa != NULL);
1715 	KASSERT(ifa->ifa_refcnt > 0);
1716 
1717 	if (--ifa->ifa_refcnt == 0) {
1718 		free(ifa, M_IFADDR);
1719 	}
1720 }
1721 
1722 bool
1723 ifa_is_destroying(struct ifaddr *ifa)
1724 {
1725 
1726 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1727 }
1728 
1729 void
1730 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1731 {
1732 
1733 	ifa->ifa_ifp = ifp;
1734 
1735 	IFNET_LOCK();
1736 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1737 	IFADDR_ENTRY_INIT(ifa);
1738 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1739 	IFNET_UNLOCK();
1740 
1741 	ifaref(ifa);
1742 }
1743 
1744 void
1745 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1746 {
1747 
1748 	KASSERT(ifa->ifa_ifp == ifp);
1749 
1750 	IFNET_LOCK();
1751 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1752 	IFADDR_WRITER_REMOVE(ifa);
1753 #ifdef NET_MPSAFE
1754 	pserialize_perform(ifnet_psz);
1755 #endif
1756 	IFNET_UNLOCK();
1757 
1758 #ifdef NET_MPSAFE
1759 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1760 #endif
1761 	IFADDR_ENTRY_DESTROY(ifa);
1762 	ifafree(ifa);
1763 }
1764 
1765 void
1766 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1767 {
1768 
1769 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1770 }
1771 
1772 void
1773 ifa_release(struct ifaddr *ifa, struct psref *psref)
1774 {
1775 
1776 	if (ifa == NULL)
1777 		return;
1778 
1779 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1780 }
1781 
1782 bool
1783 ifa_held(struct ifaddr *ifa)
1784 {
1785 
1786 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1787 }
1788 
1789 static inline int
1790 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1791 {
1792 	return sockaddr_cmp(sa1, sa2) == 0;
1793 }
1794 
1795 /*
1796  * Locate an interface based on a complete address.
1797  */
1798 /*ARGSUSED*/
1799 struct ifaddr *
1800 ifa_ifwithaddr(const struct sockaddr *addr)
1801 {
1802 	struct ifnet *ifp;
1803 	struct ifaddr *ifa;
1804 
1805 	IFNET_READER_FOREACH(ifp) {
1806 		if (if_is_deactivated(ifp))
1807 			continue;
1808 		IFADDR_READER_FOREACH(ifa, ifp) {
1809 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1810 				continue;
1811 			if (equal(addr, ifa->ifa_addr))
1812 				return ifa;
1813 			if ((ifp->if_flags & IFF_BROADCAST) &&
1814 			    ifa->ifa_broadaddr &&
1815 			    /* IP6 doesn't have broadcast */
1816 			    ifa->ifa_broadaddr->sa_len != 0 &&
1817 			    equal(ifa->ifa_broadaddr, addr))
1818 				return ifa;
1819 		}
1820 	}
1821 	return NULL;
1822 }
1823 
1824 struct ifaddr *
1825 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1826 {
1827 	struct ifaddr *ifa;
1828 	int s = pserialize_read_enter();
1829 
1830 	ifa = ifa_ifwithaddr(addr);
1831 	if (ifa != NULL)
1832 		ifa_acquire(ifa, psref);
1833 	pserialize_read_exit(s);
1834 
1835 	return ifa;
1836 }
1837 
1838 /*
1839  * Locate the point to point interface with a given destination address.
1840  */
1841 /*ARGSUSED*/
1842 struct ifaddr *
1843 ifa_ifwithdstaddr(const struct sockaddr *addr)
1844 {
1845 	struct ifnet *ifp;
1846 	struct ifaddr *ifa;
1847 
1848 	IFNET_READER_FOREACH(ifp) {
1849 		if (if_is_deactivated(ifp))
1850 			continue;
1851 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1852 			continue;
1853 		IFADDR_READER_FOREACH(ifa, ifp) {
1854 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1855 			    ifa->ifa_dstaddr == NULL)
1856 				continue;
1857 			if (equal(addr, ifa->ifa_dstaddr))
1858 				return ifa;
1859 		}
1860 	}
1861 
1862 	return NULL;
1863 }
1864 
1865 struct ifaddr *
1866 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1867 {
1868 	struct ifaddr *ifa;
1869 	int s;
1870 
1871 	s = pserialize_read_enter();
1872 	ifa = ifa_ifwithdstaddr(addr);
1873 	if (ifa != NULL)
1874 		ifa_acquire(ifa, psref);
1875 	pserialize_read_exit(s);
1876 
1877 	return ifa;
1878 }
1879 
1880 /*
1881  * Find an interface on a specific network.  If many, choice
1882  * is most specific found.
1883  */
1884 struct ifaddr *
1885 ifa_ifwithnet(const struct sockaddr *addr)
1886 {
1887 	struct ifnet *ifp;
1888 	struct ifaddr *ifa, *ifa_maybe = NULL;
1889 	const struct sockaddr_dl *sdl;
1890 	u_int af = addr->sa_family;
1891 	const char *addr_data = addr->sa_data, *cplim;
1892 
1893 	if (af == AF_LINK) {
1894 		sdl = satocsdl(addr);
1895 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1896 		    ifindex2ifnet[sdl->sdl_index] &&
1897 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1898 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1899 		}
1900 	}
1901 #ifdef NETATALK
1902 	if (af == AF_APPLETALK) {
1903 		const struct sockaddr_at *sat, *sat2;
1904 		sat = (const struct sockaddr_at *)addr;
1905 		IFNET_READER_FOREACH(ifp) {
1906 			if (if_is_deactivated(ifp))
1907 				continue;
1908 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1909 			if (ifa == NULL)
1910 				continue;
1911 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1912 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1913 				return ifa; /* exact match */
1914 			if (ifa_maybe == NULL) {
1915 				/* else keep the if with the right range */
1916 				ifa_maybe = ifa;
1917 			}
1918 		}
1919 		return ifa_maybe;
1920 	}
1921 #endif
1922 	IFNET_READER_FOREACH(ifp) {
1923 		if (if_is_deactivated(ifp))
1924 			continue;
1925 		IFADDR_READER_FOREACH(ifa, ifp) {
1926 			const char *cp, *cp2, *cp3;
1927 
1928 			if (ifa->ifa_addr->sa_family != af ||
1929 			    ifa->ifa_netmask == NULL)
1930  next:				continue;
1931 			cp = addr_data;
1932 			cp2 = ifa->ifa_addr->sa_data;
1933 			cp3 = ifa->ifa_netmask->sa_data;
1934 			cplim = (const char *)ifa->ifa_netmask +
1935 			    ifa->ifa_netmask->sa_len;
1936 			while (cp3 < cplim) {
1937 				if ((*cp++ ^ *cp2++) & *cp3++) {
1938 					/* want to continue for() loop */
1939 					goto next;
1940 				}
1941 			}
1942 			if (ifa_maybe == NULL ||
1943 			    rt_refines(ifa->ifa_netmask,
1944 			               ifa_maybe->ifa_netmask))
1945 				ifa_maybe = ifa;
1946 		}
1947 	}
1948 	return ifa_maybe;
1949 }
1950 
1951 struct ifaddr *
1952 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
1953 {
1954 	struct ifaddr *ifa;
1955 	int s;
1956 
1957 	s = pserialize_read_enter();
1958 	ifa = ifa_ifwithnet(addr);
1959 	if (ifa != NULL)
1960 		ifa_acquire(ifa, psref);
1961 	pserialize_read_exit(s);
1962 
1963 	return ifa;
1964 }
1965 
1966 /*
1967  * Find the interface of the addresss.
1968  */
1969 struct ifaddr *
1970 ifa_ifwithladdr(const struct sockaddr *addr)
1971 {
1972 	struct ifaddr *ia;
1973 
1974 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
1975 	    (ia = ifa_ifwithnet(addr)))
1976 		return ia;
1977 	return NULL;
1978 }
1979 
1980 struct ifaddr *
1981 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
1982 {
1983 	struct ifaddr *ifa;
1984 	int s;
1985 
1986 	s = pserialize_read_enter();
1987 	ifa = ifa_ifwithladdr(addr);
1988 	if (ifa != NULL)
1989 		ifa_acquire(ifa, psref);
1990 	pserialize_read_exit(s);
1991 
1992 	return ifa;
1993 }
1994 
1995 /*
1996  * Find an interface using a specific address family
1997  */
1998 struct ifaddr *
1999 ifa_ifwithaf(int af)
2000 {
2001 	struct ifnet *ifp;
2002 	struct ifaddr *ifa = NULL;
2003 	int s;
2004 
2005 	s = pserialize_read_enter();
2006 	IFNET_READER_FOREACH(ifp) {
2007 		if (if_is_deactivated(ifp))
2008 			continue;
2009 		IFADDR_READER_FOREACH(ifa, ifp) {
2010 			if (ifa->ifa_addr->sa_family == af)
2011 				goto out;
2012 		}
2013 	}
2014 out:
2015 	pserialize_read_exit(s);
2016 	return ifa;
2017 }
2018 
2019 /*
2020  * Find an interface address specific to an interface best matching
2021  * a given address.
2022  */
2023 struct ifaddr *
2024 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2025 {
2026 	struct ifaddr *ifa;
2027 	const char *cp, *cp2, *cp3;
2028 	const char *cplim;
2029 	struct ifaddr *ifa_maybe = 0;
2030 	u_int af = addr->sa_family;
2031 
2032 	if (if_is_deactivated(ifp))
2033 		return NULL;
2034 
2035 	if (af >= AF_MAX)
2036 		return NULL;
2037 
2038 	IFADDR_READER_FOREACH(ifa, ifp) {
2039 		if (ifa->ifa_addr->sa_family != af)
2040 			continue;
2041 		ifa_maybe = ifa;
2042 		if (ifa->ifa_netmask == NULL) {
2043 			if (equal(addr, ifa->ifa_addr) ||
2044 			    (ifa->ifa_dstaddr &&
2045 			     equal(addr, ifa->ifa_dstaddr)))
2046 				return ifa;
2047 			continue;
2048 		}
2049 		cp = addr->sa_data;
2050 		cp2 = ifa->ifa_addr->sa_data;
2051 		cp3 = ifa->ifa_netmask->sa_data;
2052 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2053 		for (; cp3 < cplim; cp3++) {
2054 			if ((*cp++ ^ *cp2++) & *cp3)
2055 				break;
2056 		}
2057 		if (cp3 == cplim)
2058 			return ifa;
2059 	}
2060 	return ifa_maybe;
2061 }
2062 
2063 struct ifaddr *
2064 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2065     struct psref *psref)
2066 {
2067 	struct ifaddr *ifa;
2068 	int s;
2069 
2070 	s = pserialize_read_enter();
2071 	ifa = ifaof_ifpforaddr(addr, ifp);
2072 	if (ifa != NULL)
2073 		ifa_acquire(ifa, psref);
2074 	pserialize_read_exit(s);
2075 
2076 	return ifa;
2077 }
2078 
2079 /*
2080  * Default action when installing a route with a Link Level gateway.
2081  * Lookup an appropriate real ifa to point to.
2082  * This should be moved to /sys/net/link.c eventually.
2083  */
2084 void
2085 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2086 {
2087 	struct ifaddr *ifa;
2088 	const struct sockaddr *dst;
2089 	struct ifnet *ifp;
2090 	struct psref psref;
2091 
2092 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2093 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
2094 		return;
2095 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2096 		rt_replace_ifa(rt, ifa);
2097 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2098 			ifa->ifa_rtrequest(cmd, rt, info);
2099 		ifa_release(ifa, &psref);
2100 	}
2101 }
2102 
2103 /*
2104  * bitmask macros to manage a densely packed link_state change queue.
2105  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2106  * LINK_STATE_UP(2) we need 2 bits for each state change.
2107  * As a state change to store is 0, treat all bits set as an unset item.
2108  */
2109 #define LQ_ITEM_BITS		2
2110 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2111 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2112 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2113 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2114 #define LQ_STORE(q, i, v)						      \
2115 	do {								      \
2116 		(q) &= ~LQ_MASK((i));					      \
2117 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2118 	} while (0 /* CONSTCOND */)
2119 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2120 #define LQ_POP(q, v)							      \
2121 	do {								      \
2122 		(v) = LQ_ITEM((q), 0);					      \
2123 		(q) >>= LQ_ITEM_BITS;					      \
2124 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2125 	} while (0 /* CONSTCOND */)
2126 #define LQ_PUSH(q, v)							      \
2127 	do {								      \
2128 		(q) >>= LQ_ITEM_BITS;					      \
2129 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2130 	} while (0 /* CONSTCOND */)
2131 #define LQ_FIND_UNSET(q, i)						      \
2132 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2133 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2134 			break;						      \
2135 	}
2136 /*
2137  * Handle a change in the interface link state and
2138  * queue notifications.
2139  */
2140 void
2141 if_link_state_change(struct ifnet *ifp, int link_state)
2142 {
2143 	int s, idx;
2144 
2145 	KASSERTMSG(if_is_link_state_changeable(ifp),
2146 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2147 	    ifp->if_xname, ifp->if_extflags);
2148 
2149 	/* Ensure change is to a valid state */
2150 	switch (link_state) {
2151 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2152 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2153 	case LINK_STATE_UP:
2154 		break;
2155 	default:
2156 #ifdef DEBUG
2157 		printf("%s: invalid link state %d\n",
2158 		    ifp->if_xname, link_state);
2159 #endif
2160 		return;
2161 	}
2162 
2163 	s = splnet();
2164 
2165 	/* Find the last unset event in the queue. */
2166 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2167 
2168 	/*
2169 	 * Ensure link_state doesn't match the last event in the queue.
2170 	 * ifp->if_link_state is not checked and set here because
2171 	 * that would present an inconsistent picture to the system.
2172 	 */
2173 	if (idx != 0 &&
2174 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2175 		goto out;
2176 
2177 	/* Handle queue overflow. */
2178 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2179 		uint8_t lost;
2180 
2181 		/*
2182 		 * The DOWN state must be protected from being pushed off
2183 		 * the queue to ensure that userland will always be
2184 		 * in a sane state.
2185 		 * Because DOWN is protected, there is no need to protect
2186 		 * UNKNOWN.
2187 		 * It should be invalid to change from any other state to
2188 		 * UNKNOWN anyway ...
2189 		 */
2190 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2191 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2192 		if (lost == LINK_STATE_DOWN) {
2193 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2194 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2195 		}
2196 		printf("%s: lost link state change %s\n",
2197 		    ifp->if_xname,
2198 		    lost == LINK_STATE_UP ? "UP" :
2199 		    lost == LINK_STATE_DOWN ? "DOWN" :
2200 		    "UNKNOWN");
2201 	} else
2202 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2203 
2204 	softint_schedule(ifp->if_link_si);
2205 
2206 out:
2207 	splx(s);
2208 }
2209 
2210 /*
2211  * Handle interface link state change notifications.
2212  * Must be called at splnet().
2213  */
2214 static void
2215 if_link_state_change0(struct ifnet *ifp, int link_state)
2216 {
2217 	struct domain *dp;
2218 
2219 	/* Ensure the change is still valid. */
2220 	if (ifp->if_link_state == link_state)
2221 		return;
2222 
2223 #ifdef DEBUG
2224 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2225 		link_state == LINK_STATE_UP ? "UP" :
2226 		link_state == LINK_STATE_DOWN ? "DOWN" :
2227 		"UNKNOWN",
2228 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2229 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2230 		"UNKNOWN");
2231 #endif
2232 
2233 	/*
2234 	 * When going from UNKNOWN to UP, we need to mark existing
2235 	 * addresses as tentative and restart DAD as we may have
2236 	 * erroneously not found a duplicate.
2237 	 *
2238 	 * This needs to happen before rt_ifmsg to avoid a race where
2239 	 * listeners would have an address and expect it to work right
2240 	 * away.
2241 	 */
2242 	if (link_state == LINK_STATE_UP &&
2243 	    ifp->if_link_state == LINK_STATE_UNKNOWN)
2244 	{
2245 		DOMAIN_FOREACH(dp) {
2246 			if (dp->dom_if_link_state_change != NULL)
2247 				dp->dom_if_link_state_change(ifp,
2248 				    LINK_STATE_DOWN);
2249 		}
2250 	}
2251 
2252 	ifp->if_link_state = link_state;
2253 
2254 	/* Notify that the link state has changed. */
2255 	rt_ifmsg(ifp);
2256 
2257 #if NCARP > 0
2258 	if (ifp->if_carp)
2259 		carp_carpdev_state(ifp);
2260 #endif
2261 
2262 	DOMAIN_FOREACH(dp) {
2263 		if (dp->dom_if_link_state_change != NULL)
2264 			dp->dom_if_link_state_change(ifp, link_state);
2265 	}
2266 }
2267 
2268 /*
2269  * Process the interface link state change queue.
2270  */
2271 static void
2272 if_link_state_change_si(void *arg)
2273 {
2274 	struct ifnet *ifp = arg;
2275 	int s;
2276 	uint8_t state;
2277 
2278 	s = splnet();
2279 
2280 	/* Pop a link state change from the queue and process it. */
2281 	LQ_POP(ifp->if_link_queue, state);
2282 	if_link_state_change0(ifp, state);
2283 
2284 	/* If there is a link state change to come, schedule it. */
2285 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET)
2286 		softint_schedule(ifp->if_link_si);
2287 
2288 	splx(s);
2289 }
2290 
2291 /*
2292  * Default action when installing a local route on a point-to-point
2293  * interface.
2294  */
2295 void
2296 p2p_rtrequest(int req, struct rtentry *rt,
2297     __unused const struct rt_addrinfo *info)
2298 {
2299 	struct ifnet *ifp = rt->rt_ifp;
2300 	struct ifaddr *ifa, *lo0ifa;
2301 	int s = pserialize_read_enter();
2302 
2303 	switch (req) {
2304 	case RTM_ADD:
2305 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2306 			break;
2307 
2308 		rt->rt_ifp = lo0ifp;
2309 
2310 		IFADDR_READER_FOREACH(ifa, ifp) {
2311 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2312 				break;
2313 		}
2314 		if (ifa == NULL)
2315 			break;
2316 
2317 		/*
2318 		 * Ensure lo0 has an address of the same family.
2319 		 */
2320 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2321 			if (lo0ifa->ifa_addr->sa_family ==
2322 			    ifa->ifa_addr->sa_family)
2323 				break;
2324 		}
2325 		if (lo0ifa == NULL)
2326 			break;
2327 
2328 		/*
2329 		 * Make sure to set rt->rt_ifa to the interface
2330 		 * address we are using, otherwise we will have trouble
2331 		 * with source address selection.
2332 		 */
2333 		if (ifa != rt->rt_ifa)
2334 			rt_replace_ifa(rt, ifa);
2335 		break;
2336 	case RTM_DELETE:
2337 	default:
2338 		break;
2339 	}
2340 	pserialize_read_exit(s);
2341 }
2342 
2343 /*
2344  * Mark an interface down and notify protocols of
2345  * the transition.
2346  * NOTE: must be called at splsoftnet or equivalent.
2347  */
2348 void
2349 if_down(struct ifnet *ifp)
2350 {
2351 	struct ifaddr *ifa;
2352 	struct domain *dp;
2353 	int s, bound;
2354 	struct psref psref;
2355 
2356 	ifp->if_flags &= ~IFF_UP;
2357 	nanotime(&ifp->if_lastchange);
2358 
2359 	bound = curlwp_bind();
2360 	s = pserialize_read_enter();
2361 	IFADDR_READER_FOREACH(ifa, ifp) {
2362 		ifa_acquire(ifa, &psref);
2363 		pserialize_read_exit(s);
2364 
2365 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2366 
2367 		s = pserialize_read_enter();
2368 		ifa_release(ifa, &psref);
2369 	}
2370 	pserialize_read_exit(s);
2371 	curlwp_bindx(bound);
2372 
2373 	IFQ_PURGE(&ifp->if_snd);
2374 #if NCARP > 0
2375 	if (ifp->if_carp)
2376 		carp_carpdev_state(ifp);
2377 #endif
2378 	rt_ifmsg(ifp);
2379 	DOMAIN_FOREACH(dp) {
2380 		if (dp->dom_if_down)
2381 			dp->dom_if_down(ifp);
2382 	}
2383 }
2384 
2385 /*
2386  * Mark an interface up and notify protocols of
2387  * the transition.
2388  * NOTE: must be called at splsoftnet or equivalent.
2389  */
2390 void
2391 if_up(struct ifnet *ifp)
2392 {
2393 #ifdef notyet
2394 	struct ifaddr *ifa;
2395 #endif
2396 	struct domain *dp;
2397 
2398 	ifp->if_flags |= IFF_UP;
2399 	nanotime(&ifp->if_lastchange);
2400 #ifdef notyet
2401 	/* this has no effect on IP, and will kill all ISO connections XXX */
2402 	IFADDR_READER_FOREACH(ifa, ifp)
2403 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2404 #endif
2405 #if NCARP > 0
2406 	if (ifp->if_carp)
2407 		carp_carpdev_state(ifp);
2408 #endif
2409 	rt_ifmsg(ifp);
2410 	DOMAIN_FOREACH(dp) {
2411 		if (dp->dom_if_up)
2412 			dp->dom_if_up(ifp);
2413 	}
2414 }
2415 
2416 /*
2417  * Handle interface slowtimo timer routine.  Called
2418  * from softclock, we decrement timer (if set) and
2419  * call the appropriate interface routine on expiration.
2420  */
2421 static void
2422 if_slowtimo(void *arg)
2423 {
2424 	void (*slowtimo)(struct ifnet *);
2425 	struct ifnet *ifp = arg;
2426 	int s;
2427 
2428 	slowtimo = ifp->if_slowtimo;
2429 	if (__predict_false(slowtimo == NULL))
2430 		return;
2431 
2432 	s = splnet();
2433 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2434 		(*slowtimo)(ifp);
2435 
2436 	splx(s);
2437 
2438 	if (__predict_true(ifp->if_slowtimo != NULL))
2439 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2440 }
2441 
2442 /*
2443  * Set/clear promiscuous mode on interface ifp based on the truth value
2444  * of pswitch.  The calls are reference counted so that only the first
2445  * "on" request actually has an effect, as does the final "off" request.
2446  * Results are undefined if the "off" and "on" requests are not matched.
2447  */
2448 int
2449 ifpromisc(struct ifnet *ifp, int pswitch)
2450 {
2451 	int pcount, ret;
2452 	short nflags;
2453 
2454 	pcount = ifp->if_pcount;
2455 	if (pswitch) {
2456 		/*
2457 		 * Allow the device to be "placed" into promiscuous
2458 		 * mode even if it is not configured up.  It will
2459 		 * consult IFF_PROMISC when it is brought up.
2460 		 */
2461 		if (ifp->if_pcount++ != 0)
2462 			return 0;
2463 		nflags = ifp->if_flags | IFF_PROMISC;
2464 	} else {
2465 		if (--ifp->if_pcount > 0)
2466 			return 0;
2467 		nflags = ifp->if_flags & ~IFF_PROMISC;
2468 	}
2469 	ret = if_flags_set(ifp, nflags);
2470 	/* Restore interface state if not successful. */
2471 	if (ret != 0) {
2472 		ifp->if_pcount = pcount;
2473 	}
2474 	return ret;
2475 }
2476 
2477 /*
2478  * Map interface name to
2479  * interface structure pointer.
2480  */
2481 struct ifnet *
2482 ifunit(const char *name)
2483 {
2484 	struct ifnet *ifp;
2485 	const char *cp = name;
2486 	u_int unit = 0;
2487 	u_int i;
2488 	int s;
2489 
2490 	/*
2491 	 * If the entire name is a number, treat it as an ifindex.
2492 	 */
2493 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2494 		unit = unit * 10 + (*cp - '0');
2495 	}
2496 
2497 	/*
2498 	 * If the number took all of the name, then it's a valid ifindex.
2499 	 */
2500 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2501 		if (unit >= if_indexlim)
2502 			return NULL;
2503 		ifp = ifindex2ifnet[unit];
2504 		if (ifp == NULL || if_is_deactivated(ifp))
2505 			return NULL;
2506 		return ifp;
2507 	}
2508 
2509 	ifp = NULL;
2510 	s = pserialize_read_enter();
2511 	IFNET_READER_FOREACH(ifp) {
2512 		if (if_is_deactivated(ifp))
2513 			continue;
2514 	 	if (strcmp(ifp->if_xname, name) == 0)
2515 			goto out;
2516 	}
2517 out:
2518 	pserialize_read_exit(s);
2519 	return ifp;
2520 }
2521 
2522 /*
2523  * Get a reference of an ifnet object by an interface name.
2524  * The returned reference is protected by psref(9). The caller
2525  * must release a returned reference by if_put after use.
2526  */
2527 struct ifnet *
2528 if_get(const char *name, struct psref *psref)
2529 {
2530 	struct ifnet *ifp;
2531 	const char *cp = name;
2532 	u_int unit = 0;
2533 	u_int i;
2534 	int s;
2535 
2536 	/*
2537 	 * If the entire name is a number, treat it as an ifindex.
2538 	 */
2539 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2540 		unit = unit * 10 + (*cp - '0');
2541 	}
2542 
2543 	/*
2544 	 * If the number took all of the name, then it's a valid ifindex.
2545 	 */
2546 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2547 		if (unit >= if_indexlim)
2548 			return NULL;
2549 		ifp = ifindex2ifnet[unit];
2550 		if (ifp == NULL || if_is_deactivated(ifp))
2551 			return NULL;
2552 		return ifp;
2553 	}
2554 
2555 	ifp = NULL;
2556 	s = pserialize_read_enter();
2557 	IFNET_READER_FOREACH(ifp) {
2558 		if (if_is_deactivated(ifp))
2559 			continue;
2560 		if (strcmp(ifp->if_xname, name) == 0) {
2561 			psref_acquire(psref, &ifp->if_psref,
2562 			    ifnet_psref_class);
2563 			goto out;
2564 		}
2565 	}
2566 out:
2567 	pserialize_read_exit(s);
2568 	return ifp;
2569 }
2570 
2571 /*
2572  * Release a reference of an ifnet object given by if_get or
2573  * if_get_byindex.
2574  */
2575 void
2576 if_put(const struct ifnet *ifp, struct psref *psref)
2577 {
2578 
2579 	if (ifp == NULL)
2580 		return;
2581 
2582 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2583 }
2584 
2585 ifnet_t *
2586 if_byindex(u_int idx)
2587 {
2588 	ifnet_t *ifp;
2589 
2590 	ifp = (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL;
2591 	if (ifp != NULL && if_is_deactivated(ifp))
2592 		ifp = NULL;
2593 	return ifp;
2594 }
2595 
2596 /*
2597  * Get a reference of an ifnet object by an interface index.
2598  * The returned reference is protected by psref(9). The caller
2599  * must release a returned reference by if_put after use.
2600  */
2601 ifnet_t *
2602 if_get_byindex(u_int idx, struct psref *psref)
2603 {
2604 	ifnet_t *ifp;
2605 	int s;
2606 
2607 	s = pserialize_read_enter();
2608 	ifp = (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2609 	if (ifp != NULL && if_is_deactivated(ifp))
2610 		ifp = NULL;
2611 	if (__predict_true(ifp != NULL))
2612 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2613 	pserialize_read_exit(s);
2614 
2615 	return ifp;
2616 }
2617 
2618 /*
2619  * XXX it's safe only if the passed ifp is guaranteed to not be freed,
2620  * for example the ifp is already held or some other object is held which
2621  * guarantes the ifp to not be freed indirectly.
2622  */
2623 void
2624 if_acquire_NOMPSAFE(struct ifnet *ifp, struct psref *psref)
2625 {
2626 
2627 	KASSERT(ifp->if_index != 0);
2628 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2629 }
2630 
2631 bool
2632 if_held(struct ifnet *ifp)
2633 {
2634 
2635 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2636 }
2637 
2638 
2639 /* common */
2640 int
2641 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2642 {
2643 	int s;
2644 	struct ifreq *ifr;
2645 	struct ifcapreq *ifcr;
2646 	struct ifdatareq *ifdr;
2647 
2648 	switch (cmd) {
2649 	case SIOCSIFCAP:
2650 		ifcr = data;
2651 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2652 			return EINVAL;
2653 
2654 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2655 			return 0;
2656 
2657 		ifp->if_capenable = ifcr->ifcr_capenable;
2658 
2659 		/* Pre-compute the checksum flags mask. */
2660 		ifp->if_csum_flags_tx = 0;
2661 		ifp->if_csum_flags_rx = 0;
2662 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2663 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2664 		}
2665 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2666 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2667 		}
2668 
2669 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2670 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2671 		}
2672 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2673 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2674 		}
2675 
2676 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2677 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2678 		}
2679 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2680 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2681 		}
2682 
2683 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2684 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2685 		}
2686 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2687 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2688 		}
2689 
2690 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2691 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2692 		}
2693 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2694 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2695 		}
2696 		if (ifp->if_flags & IFF_UP)
2697 			return ENETRESET;
2698 		return 0;
2699 	case SIOCSIFFLAGS:
2700 		ifr = data;
2701 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2702 			s = splnet();
2703 			if_down(ifp);
2704 			splx(s);
2705 		}
2706 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2707 			s = splnet();
2708 			if_up(ifp);
2709 			splx(s);
2710 		}
2711 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2712 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2713 		break;
2714 	case SIOCGIFFLAGS:
2715 		ifr = data;
2716 		ifr->ifr_flags = ifp->if_flags;
2717 		break;
2718 
2719 	case SIOCGIFMETRIC:
2720 		ifr = data;
2721 		ifr->ifr_metric = ifp->if_metric;
2722 		break;
2723 
2724 	case SIOCGIFMTU:
2725 		ifr = data;
2726 		ifr->ifr_mtu = ifp->if_mtu;
2727 		break;
2728 
2729 	case SIOCGIFDLT:
2730 		ifr = data;
2731 		ifr->ifr_dlt = ifp->if_dlt;
2732 		break;
2733 
2734 	case SIOCGIFCAP:
2735 		ifcr = data;
2736 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2737 		ifcr->ifcr_capenable = ifp->if_capenable;
2738 		break;
2739 
2740 	case SIOCSIFMETRIC:
2741 		ifr = data;
2742 		ifp->if_metric = ifr->ifr_metric;
2743 		break;
2744 
2745 	case SIOCGIFDATA:
2746 		ifdr = data;
2747 		ifdr->ifdr_data = ifp->if_data;
2748 		break;
2749 
2750 	case SIOCGIFINDEX:
2751 		ifr = data;
2752 		ifr->ifr_index = ifp->if_index;
2753 		break;
2754 
2755 	case SIOCZIFDATA:
2756 		ifdr = data;
2757 		ifdr->ifdr_data = ifp->if_data;
2758 		/*
2759 		 * Assumes that the volatile counters that can be
2760 		 * zero'ed are at the end of if_data.
2761 		 */
2762 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
2763 		    offsetof(struct if_data, ifi_ipackets));
2764 		/*
2765 		 * The memset() clears to the bottm of if_data. In the area,
2766 		 * if_lastchange is included. Please be careful if new entry
2767 		 * will be added into if_data or rewite this.
2768 		 *
2769 		 * And also, update if_lastchnage.
2770 		 */
2771 		getnanotime(&ifp->if_lastchange);
2772 		break;
2773 	case SIOCSIFMTU:
2774 		ifr = data;
2775 		if (ifp->if_mtu == ifr->ifr_mtu)
2776 			break;
2777 		ifp->if_mtu = ifr->ifr_mtu;
2778 		/*
2779 		 * If the link MTU changed, do network layer specific procedure.
2780 		 */
2781 #ifdef INET6
2782 		if (in6_present)
2783 			nd6_setmtu(ifp);
2784 #endif
2785 		return ENETRESET;
2786 	default:
2787 		return ENOTTY;
2788 	}
2789 	return 0;
2790 }
2791 
2792 int
2793 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
2794 {
2795 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
2796 	struct ifaddr *ifa;
2797 	const struct sockaddr *any, *sa;
2798 	union {
2799 		struct sockaddr sa;
2800 		struct sockaddr_storage ss;
2801 	} u, v;
2802 	int s, error = 0;
2803 
2804 	switch (cmd) {
2805 	case SIOCSIFADDRPREF:
2806 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
2807 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
2808 		    NULL) != 0)
2809 			return EPERM;
2810 	case SIOCGIFADDRPREF:
2811 		break;
2812 	default:
2813 		return EOPNOTSUPP;
2814 	}
2815 
2816 	/* sanity checks */
2817 	if (data == NULL || ifp == NULL) {
2818 		panic("invalid argument to %s", __func__);
2819 		/*NOTREACHED*/
2820 	}
2821 
2822 	/* address must be specified on ADD and DELETE */
2823 	sa = sstocsa(&ifap->ifap_addr);
2824 	if (sa->sa_family != sofamily(so))
2825 		return EINVAL;
2826 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
2827 		return EINVAL;
2828 
2829 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
2830 
2831 	s = pserialize_read_enter();
2832 	IFADDR_READER_FOREACH(ifa, ifp) {
2833 		if (ifa->ifa_addr->sa_family != sa->sa_family)
2834 			continue;
2835 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
2836 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
2837 			break;
2838 	}
2839 	if (ifa == NULL) {
2840 		error = EADDRNOTAVAIL;
2841 		goto out;
2842 	}
2843 
2844 	switch (cmd) {
2845 	case SIOCSIFADDRPREF:
2846 		ifa->ifa_preference = ifap->ifap_preference;
2847 		goto out;
2848 	case SIOCGIFADDRPREF:
2849 		/* fill in the if_laddrreq structure */
2850 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
2851 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
2852 		ifap->ifap_preference = ifa->ifa_preference;
2853 		goto out;
2854 	default:
2855 		error = EOPNOTSUPP;
2856 	}
2857 out:
2858 	pserialize_read_exit(s);
2859 	return error;
2860 }
2861 
2862 /*
2863  * Interface ioctls.
2864  */
2865 static int
2866 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
2867 {
2868 	struct ifnet *ifp;
2869 	struct ifreq *ifr;
2870 	int error = 0;
2871 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
2872 	u_long ocmd = cmd;
2873 #endif
2874 	short oif_flags;
2875 #ifdef COMPAT_OIFREQ
2876 	struct ifreq ifrb;
2877 	struct oifreq *oifr = NULL;
2878 #endif
2879 	int r;
2880 	struct psref psref;
2881 	int bound;
2882 
2883 	switch (cmd) {
2884 #ifdef COMPAT_OIFREQ
2885 	case OSIOCGIFCONF:
2886 	case OOSIOCGIFCONF:
2887 		return compat_ifconf(cmd, data);
2888 #endif
2889 #ifdef COMPAT_OIFDATA
2890 	case OSIOCGIFDATA:
2891 	case OSIOCZIFDATA:
2892 		return compat_ifdatareq(l, cmd, data);
2893 #endif
2894 	case SIOCGIFCONF:
2895 		return ifconf(cmd, data);
2896 	case SIOCINITIFADDR:
2897 		return EPERM;
2898 	}
2899 
2900 #ifdef COMPAT_OIFREQ
2901 	cmd = (*vec_compat_cvtcmd)(cmd);
2902 	if (cmd != ocmd) {
2903 		oifr = data;
2904 		data = ifr = &ifrb;
2905 		ifreqo2n(oifr, ifr);
2906 	} else
2907 #endif
2908 		ifr = data;
2909 
2910 	switch (cmd) {
2911 	case SIOCIFCREATE:
2912 	case SIOCIFDESTROY:
2913 		bound = curlwp_bind();
2914 		if (l != NULL) {
2915 			ifp = if_get(ifr->ifr_name, &psref);
2916 			error = kauth_authorize_network(l->l_cred,
2917 			    KAUTH_NETWORK_INTERFACE,
2918 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2919 			    (void *)cmd, NULL);
2920 			if (ifp != NULL)
2921 				if_put(ifp, &psref);
2922 			if (error != 0) {
2923 				curlwp_bindx(bound);
2924 				return error;
2925 			}
2926 		}
2927 		mutex_enter(&if_clone_mtx);
2928 		r = (cmd == SIOCIFCREATE) ?
2929 			if_clone_create(ifr->ifr_name) :
2930 			if_clone_destroy(ifr->ifr_name);
2931 		mutex_exit(&if_clone_mtx);
2932 		curlwp_bindx(bound);
2933 		return r;
2934 
2935 	case SIOCIFGCLONERS:
2936 		{
2937 			struct if_clonereq *req = (struct if_clonereq *)data;
2938 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
2939 			    &req->ifcr_total);
2940 		}
2941 	}
2942 
2943 	bound = curlwp_bind();
2944 	ifp = if_get(ifr->ifr_name, &psref);
2945 	if (ifp == NULL) {
2946 		curlwp_bindx(bound);
2947 		return ENXIO;
2948 	}
2949 
2950 	switch (cmd) {
2951 	case SIOCALIFADDR:
2952 	case SIOCDLIFADDR:
2953 	case SIOCSIFADDRPREF:
2954 	case SIOCSIFFLAGS:
2955 	case SIOCSIFCAP:
2956 	case SIOCSIFMETRIC:
2957 	case SIOCZIFDATA:
2958 	case SIOCSIFMTU:
2959 	case SIOCSIFPHYADDR:
2960 	case SIOCDIFPHYADDR:
2961 #ifdef INET6
2962 	case SIOCSIFPHYADDR_IN6:
2963 #endif
2964 	case SIOCSLIFPHYADDR:
2965 	case SIOCADDMULTI:
2966 	case SIOCDELMULTI:
2967 	case SIOCSIFMEDIA:
2968 	case SIOCSDRVSPEC:
2969 	case SIOCG80211:
2970 	case SIOCS80211:
2971 	case SIOCS80211NWID:
2972 	case SIOCS80211NWKEY:
2973 	case SIOCS80211POWER:
2974 	case SIOCS80211BSSID:
2975 	case SIOCS80211CHANNEL:
2976 	case SIOCSLINKSTR:
2977 		if (l != NULL) {
2978 			error = kauth_authorize_network(l->l_cred,
2979 			    KAUTH_NETWORK_INTERFACE,
2980 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2981 			    (void *)cmd, NULL);
2982 			if (error != 0)
2983 				goto out;
2984 		}
2985 	}
2986 
2987 	oif_flags = ifp->if_flags;
2988 
2989 	mutex_enter(ifp->if_ioctl_lock);
2990 
2991 	error = (*ifp->if_ioctl)(ifp, cmd, data);
2992 	if (error != ENOTTY)
2993 		;
2994 	else if (so->so_proto == NULL)
2995 		error = EOPNOTSUPP;
2996 	else {
2997 #ifdef COMPAT_OSOCK
2998 		if (vec_compat_ifioctl != NULL)
2999 			error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l);
3000 		else
3001 #endif
3002 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3003 			    cmd, data, ifp);
3004 	}
3005 
3006 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3007 		if ((ifp->if_flags & IFF_UP) != 0) {
3008 			int s = splnet();
3009 			if_up(ifp);
3010 			splx(s);
3011 		}
3012 	}
3013 #ifdef COMPAT_OIFREQ
3014 	if (cmd != ocmd)
3015 		ifreqn2o(oifr, ifr);
3016 #endif
3017 
3018 	mutex_exit(ifp->if_ioctl_lock);
3019 out:
3020 	if_put(ifp, &psref);
3021 	curlwp_bindx(bound);
3022 	return error;
3023 }
3024 
3025 /*
3026  * Return interface configuration
3027  * of system.  List may be used
3028  * in later ioctl's (above) to get
3029  * other information.
3030  *
3031  * Each record is a struct ifreq.  Before the addition of
3032  * sockaddr_storage, the API rule was that sockaddr flavors that did
3033  * not fit would extend beyond the struct ifreq, with the next struct
3034  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3035  * union in struct ifreq includes struct sockaddr_storage, every kind
3036  * of sockaddr must fit.  Thus, there are no longer any overlength
3037  * records.
3038  *
3039  * Records are added to the user buffer if they fit, and ifc_len is
3040  * adjusted to the length that was written.  Thus, the user is only
3041  * assured of getting the complete list if ifc_len on return is at
3042  * least sizeof(struct ifreq) less than it was on entry.
3043  *
3044  * If the user buffer pointer is NULL, this routine copies no data and
3045  * returns the amount of space that would be needed.
3046  *
3047  * Invariants:
3048  * ifrp points to the next part of the user's buffer to be used.  If
3049  * ifrp != NULL, space holds the number of bytes remaining that we may
3050  * write at ifrp.  Otherwise, space holds the number of bytes that
3051  * would have been written had there been adequate space.
3052  */
3053 /*ARGSUSED*/
3054 static int
3055 ifconf(u_long cmd, void *data)
3056 {
3057 	struct ifconf *ifc = (struct ifconf *)data;
3058 	struct ifnet *ifp;
3059 	struct ifaddr *ifa;
3060 	struct ifreq ifr, *ifrp = NULL;
3061 	int space = 0, error = 0;
3062 	const int sz = (int)sizeof(struct ifreq);
3063 	const bool docopy = ifc->ifc_req != NULL;
3064 	int s;
3065 	int bound;
3066 	struct psref psref;
3067 
3068 	if (docopy) {
3069 		space = ifc->ifc_len;
3070 		ifrp = ifc->ifc_req;
3071 	}
3072 
3073 	bound = curlwp_bind();
3074 	s = pserialize_read_enter();
3075 	IFNET_READER_FOREACH(ifp) {
3076 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3077 		pserialize_read_exit(s);
3078 
3079 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3080 		    sizeof(ifr.ifr_name));
3081 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3082 			error = ENAMETOOLONG;
3083 			goto release_exit;
3084 		}
3085 		if (IFADDR_READER_EMPTY(ifp)) {
3086 			/* Interface with no addresses - send zero sockaddr. */
3087 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3088 			if (!docopy) {
3089 				space += sz;
3090 				continue;
3091 			}
3092 			if (space >= sz) {
3093 				error = copyout(&ifr, ifrp, sz);
3094 				if (error != 0)
3095 					goto release_exit;
3096 				ifrp++;
3097 				space -= sz;
3098 			}
3099 		}
3100 
3101 		IFADDR_READER_FOREACH(ifa, ifp) {
3102 			struct sockaddr *sa = ifa->ifa_addr;
3103 			/* all sockaddrs must fit in sockaddr_storage */
3104 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3105 
3106 			if (!docopy) {
3107 				space += sz;
3108 				continue;
3109 			}
3110 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3111 			if (space >= sz) {
3112 				error = copyout(&ifr, ifrp, sz);
3113 				if (error != 0)
3114 					goto release_exit;
3115 				ifrp++; space -= sz;
3116 			}
3117 		}
3118 
3119 		s = pserialize_read_enter();
3120 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3121 	}
3122 	pserialize_read_exit(s);
3123 	curlwp_bindx(bound);
3124 
3125 	if (docopy) {
3126 		KASSERT(0 <= space && space <= ifc->ifc_len);
3127 		ifc->ifc_len -= space;
3128 	} else {
3129 		KASSERT(space >= 0);
3130 		ifc->ifc_len = space;
3131 	}
3132 	return (0);
3133 
3134 release_exit:
3135 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3136 	curlwp_bindx(bound);
3137 	return error;
3138 }
3139 
3140 int
3141 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3142 {
3143 	uint8_t len;
3144 #ifdef COMPAT_OIFREQ
3145 	struct ifreq ifrb;
3146 	struct oifreq *oifr = NULL;
3147 	u_long ocmd = cmd;
3148 	cmd = (*vec_compat_cvtcmd)(cmd);
3149 	if (cmd != ocmd) {
3150 		oifr = (struct oifreq *)(void *)ifr;
3151 		ifr = &ifrb;
3152 		ifreqo2n(oifr, ifr);
3153 		len = sizeof(oifr->ifr_addr);
3154 	} else
3155 #endif
3156 		len = sizeof(ifr->ifr_ifru.ifru_space);
3157 
3158 	if (len < sa->sa_len)
3159 		return EFBIG;
3160 
3161 	memset(&ifr->ifr_addr, 0, len);
3162 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3163 
3164 #ifdef COMPAT_OIFREQ
3165 	if (cmd != ocmd)
3166 		ifreqn2o(oifr, ifr);
3167 #endif
3168 	return 0;
3169 }
3170 
3171 /*
3172  * wrapper function for the drivers which doesn't have if_transmit().
3173  */
3174 static int
3175 if_transmit(struct ifnet *ifp, struct mbuf *m)
3176 {
3177 	int s, error;
3178 
3179 	s = splnet();
3180 
3181 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3182 	if (error != 0) {
3183 		/* mbuf is already freed */
3184 		goto out;
3185 	}
3186 
3187 	ifp->if_obytes += m->m_pkthdr.len;;
3188 	if (m->m_flags & M_MCAST)
3189 		ifp->if_omcasts++;
3190 
3191 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3192 		if_start_lock(ifp);
3193 out:
3194 	splx(s);
3195 
3196 	return error;
3197 }
3198 
3199 int
3200 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3201 {
3202 	int error;
3203 
3204 #ifdef ALTQ
3205 	KERNEL_LOCK(1, NULL);
3206 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3207 		error = if_transmit(ifp, m);
3208 		KERNEL_UNLOCK_ONE(NULL);
3209 	} else {
3210 		KERNEL_UNLOCK_ONE(NULL);
3211 		error = (*ifp->if_transmit)(ifp, m);
3212 	}
3213 #else /* !ALTQ */
3214 	error = (*ifp->if_transmit)(ifp, m);
3215 #endif /* !ALTQ */
3216 
3217 	return error;
3218 }
3219 
3220 /*
3221  * Queue message on interface, and start output if interface
3222  * not yet active.
3223  */
3224 int
3225 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3226 {
3227 
3228 	return if_transmit_lock(ifp, m);
3229 }
3230 
3231 /*
3232  * Queue message on interface, possibly using a second fast queue
3233  */
3234 int
3235 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3236 {
3237 	int error = 0;
3238 
3239 	if (ifq != NULL
3240 #ifdef ALTQ
3241 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3242 #endif
3243 	    ) {
3244 		if (IF_QFULL(ifq)) {
3245 			IF_DROP(&ifp->if_snd);
3246 			m_freem(m);
3247 			if (error == 0)
3248 				error = ENOBUFS;
3249 		} else
3250 			IF_ENQUEUE(ifq, m);
3251 	} else
3252 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3253 	if (error != 0) {
3254 		++ifp->if_oerrors;
3255 		return error;
3256 	}
3257 	return 0;
3258 }
3259 
3260 int
3261 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3262 {
3263 	int rc;
3264 
3265 	if (ifp->if_initaddr != NULL)
3266 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3267 	else if (src ||
3268 		/* FIXME: may not hold if_ioctl_lock */
3269 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3270 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3271 
3272 	return rc;
3273 }
3274 
3275 int
3276 if_do_dad(struct ifnet *ifp)
3277 {
3278 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3279 		return 0;
3280 
3281 	switch (ifp->if_type) {
3282 	case IFT_FAITH:
3283 		/*
3284 		 * These interfaces do not have the IFF_LOOPBACK flag,
3285 		 * but loop packets back.  We do not have to do DAD on such
3286 		 * interfaces.  We should even omit it, because loop-backed
3287 		 * responses would confuse the DAD procedure.
3288 		 */
3289 		return 0;
3290 	default:
3291 		/*
3292 		 * Our DAD routine requires the interface up and running.
3293 		 * However, some interfaces can be up before the RUNNING
3294 		 * status.  Additionaly, users may try to assign addresses
3295 		 * before the interface becomes up (or running).
3296 		 * We simply skip DAD in such a case as a work around.
3297 		 * XXX: we should rather mark "tentative" on such addresses,
3298 		 * and do DAD after the interface becomes ready.
3299 		 */
3300 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
3301 		    (IFF_UP|IFF_RUNNING))
3302 			return 0;
3303 
3304 		return 1;
3305 	}
3306 }
3307 
3308 int
3309 if_flags_set(ifnet_t *ifp, const short flags)
3310 {
3311 	int rc;
3312 
3313 	if (ifp->if_setflags != NULL)
3314 		rc = (*ifp->if_setflags)(ifp, flags);
3315 	else {
3316 		short cantflags, chgdflags;
3317 		struct ifreq ifr;
3318 
3319 		chgdflags = ifp->if_flags ^ flags;
3320 		cantflags = chgdflags & IFF_CANTCHANGE;
3321 
3322 		if (cantflags != 0)
3323 			ifp->if_flags ^= cantflags;
3324 
3325                 /* Traditionally, we do not call if_ioctl after
3326                  * setting/clearing only IFF_PROMISC if the interface
3327                  * isn't IFF_UP.  Uphold that tradition.
3328 		 */
3329 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3330 			return 0;
3331 
3332 		memset(&ifr, 0, sizeof(ifr));
3333 
3334 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3335 		/* FIXME: may not hold if_ioctl_lock */
3336 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3337 
3338 		if (rc != 0 && cantflags != 0)
3339 			ifp->if_flags ^= cantflags;
3340 	}
3341 
3342 	return rc;
3343 }
3344 
3345 int
3346 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3347 {
3348 	int rc;
3349 	struct ifreq ifr;
3350 
3351 	if (ifp->if_mcastop != NULL)
3352 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3353 	else {
3354 		ifreq_setaddr(cmd, &ifr, sa);
3355 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3356 	}
3357 
3358 	return rc;
3359 }
3360 
3361 static void
3362 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3363     struct ifaltq *ifq)
3364 {
3365 	const struct sysctlnode *cnode, *rnode;
3366 
3367 	if (sysctl_createv(clog, 0, NULL, &rnode,
3368 		       CTLFLAG_PERMANENT,
3369 		       CTLTYPE_NODE, "interfaces",
3370 		       SYSCTL_DESCR("Per-interface controls"),
3371 		       NULL, 0, NULL, 0,
3372 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3373 		goto bad;
3374 
3375 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3376 		       CTLFLAG_PERMANENT,
3377 		       CTLTYPE_NODE, ifname,
3378 		       SYSCTL_DESCR("Interface controls"),
3379 		       NULL, 0, NULL, 0,
3380 		       CTL_CREATE, CTL_EOL) != 0)
3381 		goto bad;
3382 
3383 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3384 		       CTLFLAG_PERMANENT,
3385 		       CTLTYPE_NODE, "sndq",
3386 		       SYSCTL_DESCR("Interface output queue controls"),
3387 		       NULL, 0, NULL, 0,
3388 		       CTL_CREATE, CTL_EOL) != 0)
3389 		goto bad;
3390 
3391 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3392 		       CTLFLAG_PERMANENT,
3393 		       CTLTYPE_INT, "len",
3394 		       SYSCTL_DESCR("Current output queue length"),
3395 		       NULL, 0, &ifq->ifq_len, 0,
3396 		       CTL_CREATE, CTL_EOL) != 0)
3397 		goto bad;
3398 
3399 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3400 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3401 		       CTLTYPE_INT, "maxlen",
3402 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3403 		       NULL, 0, &ifq->ifq_maxlen, 0,
3404 		       CTL_CREATE, CTL_EOL) != 0)
3405 		goto bad;
3406 
3407 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3408 		       CTLFLAG_PERMANENT,
3409 		       CTLTYPE_INT, "drops",
3410 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3411 		       NULL, 0, &ifq->ifq_drops, 0,
3412 		       CTL_CREATE, CTL_EOL) != 0)
3413 		goto bad;
3414 
3415 	return;
3416 bad:
3417 	printf("%s: could not attach sysctl nodes\n", ifname);
3418 	return;
3419 }
3420 
3421 #if defined(INET) || defined(INET6)
3422 
3423 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3424 	static int							\
3425 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3426 	{								\
3427 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3428 	}
3429 
3430 #if defined(INET)
3431 static int
3432 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3433 {
3434 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3435 }
3436 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3437 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3438 #endif
3439 
3440 #if defined(INET6)
3441 static int
3442 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3443 {
3444 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3445 }
3446 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3447 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3448 #endif
3449 
3450 static void
3451 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3452 {
3453 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3454 	const char *pfname = NULL, *ipname = NULL;
3455 	int ipn = 0, qid = 0;
3456 
3457 	switch (pf) {
3458 #if defined(INET)
3459 	case PF_INET:
3460 		len_func = sysctl_net_ip_pktq_items;
3461 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3462 		drops_func = sysctl_net_ip_pktq_drops;
3463 		pfname = "inet", ipn = IPPROTO_IP;
3464 		ipname = "ip", qid = IPCTL_IFQ;
3465 		break;
3466 #endif
3467 #if defined(INET6)
3468 	case PF_INET6:
3469 		len_func = sysctl_net_ip6_pktq_items;
3470 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3471 		drops_func = sysctl_net_ip6_pktq_drops;
3472 		pfname = "inet6", ipn = IPPROTO_IPV6;
3473 		ipname = "ip6", qid = IPV6CTL_IFQ;
3474 		break;
3475 #endif
3476 	default:
3477 		KASSERT(false);
3478 	}
3479 
3480 	sysctl_createv(clog, 0, NULL, NULL,
3481 		       CTLFLAG_PERMANENT,
3482 		       CTLTYPE_NODE, pfname, NULL,
3483 		       NULL, 0, NULL, 0,
3484 		       CTL_NET, pf, CTL_EOL);
3485 	sysctl_createv(clog, 0, NULL, NULL,
3486 		       CTLFLAG_PERMANENT,
3487 		       CTLTYPE_NODE, ipname, NULL,
3488 		       NULL, 0, NULL, 0,
3489 		       CTL_NET, pf, ipn, CTL_EOL);
3490 	sysctl_createv(clog, 0, NULL, NULL,
3491 		       CTLFLAG_PERMANENT,
3492 		       CTLTYPE_NODE, "ifq",
3493 		       SYSCTL_DESCR("Protocol input queue controls"),
3494 		       NULL, 0, NULL, 0,
3495 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3496 
3497 	sysctl_createv(clog, 0, NULL, NULL,
3498 		       CTLFLAG_PERMANENT,
3499 		       CTLTYPE_INT, "len",
3500 		       SYSCTL_DESCR("Current input queue length"),
3501 		       len_func, 0, NULL, 0,
3502 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3503 	sysctl_createv(clog, 0, NULL, NULL,
3504 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3505 		       CTLTYPE_INT, "maxlen",
3506 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3507 		       maxlen_func, 0, NULL, 0,
3508 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3509 	sysctl_createv(clog, 0, NULL, NULL,
3510 		       CTLFLAG_PERMANENT,
3511 		       CTLTYPE_INT, "drops",
3512 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3513 		       drops_func, 0, NULL, 0,
3514 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3515 }
3516 #endif /* INET || INET6 */
3517 
3518 static int
3519 if_sdl_sysctl(SYSCTLFN_ARGS)
3520 {
3521 	struct ifnet *ifp;
3522 	const struct sockaddr_dl *sdl;
3523 	struct psref psref;
3524 	int error = 0;
3525 	int bound;
3526 
3527 	if (namelen != 1)
3528 		return EINVAL;
3529 
3530 	bound = curlwp_bind();
3531 	ifp = if_get_byindex(name[0], &psref);
3532 	if (ifp == NULL) {
3533 		error = ENODEV;
3534 		goto out0;
3535 	}
3536 
3537 	sdl = ifp->if_sadl;
3538 	if (sdl == NULL) {
3539 		*oldlenp = 0;
3540 		goto out1;
3541 	}
3542 
3543 	if (oldp == NULL) {
3544 		*oldlenp = sdl->sdl_alen;
3545 		goto out1;
3546 	}
3547 
3548 	if (*oldlenp >= sdl->sdl_alen)
3549 		*oldlenp = sdl->sdl_alen;
3550 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3551 out1:
3552 	if_put(ifp, &psref);
3553 out0:
3554 	curlwp_bindx(bound);
3555 	return error;
3556 }
3557 
3558 SYSCTL_SETUP(sysctl_net_sdl_setup, "sysctl net.sdl subtree setup")
3559 {
3560 	const struct sysctlnode *rnode = NULL;
3561 
3562 	sysctl_createv(clog, 0, NULL, &rnode,
3563 		       CTLFLAG_PERMANENT,
3564 		       CTLTYPE_NODE, "sdl",
3565 		       SYSCTL_DESCR("Get active link-layer address"),
3566 		       if_sdl_sysctl, 0, NULL, 0,
3567 		       CTL_NET, CTL_CREATE, CTL_EOL);
3568 }
3569