xref: /netbsd-src/sys/net/if.c (revision 1580a27b92f58fcdcb23fdfbc04a7c2b54a0b7c8)
1 /*	$NetBSD: if.c,v 1.397 2017/11/17 07:37:12 ozaki-r Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by William Studenmund and Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the project nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 /*
62  * Copyright (c) 1980, 1986, 1993
63  *	The Regents of the University of California.  All rights reserved.
64  *
65  * Redistribution and use in source and binary forms, with or without
66  * modification, are permitted provided that the following conditions
67  * are met:
68  * 1. Redistributions of source code must retain the above copyright
69  *    notice, this list of conditions and the following disclaimer.
70  * 2. Redistributions in binary form must reproduce the above copyright
71  *    notice, this list of conditions and the following disclaimer in the
72  *    documentation and/or other materials provided with the distribution.
73  * 3. Neither the name of the University nor the names of its contributors
74  *    may be used to endorse or promote products derived from this software
75  *    without specific prior written permission.
76  *
77  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
78  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
79  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
80  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
81  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
82  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
83  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
84  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
85  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
86  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
87  * SUCH DAMAGE.
88  *
89  *	@(#)if.c	8.5 (Berkeley) 1/9/95
90  */
91 
92 #include <sys/cdefs.h>
93 __KERNEL_RCSID(0, "$NetBSD: if.c,v 1.397 2017/11/17 07:37:12 ozaki-r Exp $");
94 
95 #if defined(_KERNEL_OPT)
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 
99 #include "opt_atalk.h"
100 #include "opt_natm.h"
101 #include "opt_wlan.h"
102 #include "opt_net_mpsafe.h"
103 #endif
104 
105 #include <sys/param.h>
106 #include <sys/mbuf.h>
107 #include <sys/systm.h>
108 #include <sys/callout.h>
109 #include <sys/proc.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/domain.h>
113 #include <sys/protosw.h>
114 #include <sys/kernel.h>
115 #include <sys/ioctl.h>
116 #include <sys/sysctl.h>
117 #include <sys/syslog.h>
118 #include <sys/kauth.h>
119 #include <sys/kmem.h>
120 #include <sys/xcall.h>
121 #include <sys/cpu.h>
122 #include <sys/intr.h>
123 
124 #include <net/if.h>
125 #include <net/if_dl.h>
126 #include <net/if_ether.h>
127 #include <net/if_media.h>
128 #include <net80211/ieee80211.h>
129 #include <net80211/ieee80211_ioctl.h>
130 #include <net/if_types.h>
131 #include <net/route.h>
132 #include <net/netisr.h>
133 #include <sys/module.h>
134 #ifdef NETATALK
135 #include <netatalk/at_extern.h>
136 #include <netatalk/at.h>
137 #endif
138 #include <net/pfil.h>
139 #include <netinet/in.h>
140 #include <netinet/in_var.h>
141 #include <netinet/ip_encap.h>
142 #include <net/bpf.h>
143 
144 #ifdef INET6
145 #include <netinet6/in6_var.h>
146 #include <netinet6/nd6.h>
147 #endif
148 
149 #include "ether.h"
150 #include "fddi.h"
151 #include "token.h"
152 
153 #include "carp.h"
154 #if NCARP > 0
155 #include <netinet/ip_carp.h>
156 #endif
157 
158 #include <compat/sys/sockio.h>
159 #include <compat/sys/socket.h>
160 
161 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
162 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
163 
164 /*
165  * Global list of interfaces.
166  */
167 /* DEPRECATED. Remove it once kvm(3) users disappeared */
168 struct ifnet_head		ifnet_list;
169 
170 struct pslist_head		ifnet_pslist;
171 static ifnet_t **		ifindex2ifnet = NULL;
172 static u_int			if_index = 1;
173 static size_t			if_indexlim = 0;
174 static uint64_t			index_gen;
175 /* Mutex to protect the above objects. */
176 kmutex_t			ifnet_mtx __cacheline_aligned;
177 static struct psref_class	*ifnet_psref_class __read_mostly;
178 static pserialize_t		ifnet_psz;
179 
180 static kmutex_t			if_clone_mtx;
181 
182 struct ifnet *lo0ifp;
183 int	ifqmaxlen = IFQ_MAXLEN;
184 
185 struct psref_class		*ifa_psref_class __read_mostly;
186 
187 static int	if_delroute_matcher(struct rtentry *, void *);
188 
189 static bool if_is_unit(const char *);
190 static struct if_clone *if_clone_lookup(const char *, int *);
191 
192 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
193 static int if_cloners_count;
194 
195 /* Packet filtering hook for interfaces. */
196 pfil_head_t *			if_pfil __read_mostly;
197 
198 static kauth_listener_t if_listener;
199 
200 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
201 static void if_detach_queues(struct ifnet *, struct ifqueue *);
202 static void sysctl_sndq_setup(struct sysctllog **, const char *,
203     struct ifaltq *);
204 static void if_slowtimo(void *);
205 static void if_free_sadl(struct ifnet *);
206 static void if_attachdomain1(struct ifnet *);
207 static int ifconf(u_long, void *);
208 static int if_transmit(struct ifnet *, struct mbuf *);
209 static int if_clone_create(const char *);
210 static int if_clone_destroy(const char *);
211 static void if_link_state_change_si(void *);
212 
213 struct if_percpuq {
214 	struct ifnet	*ipq_ifp;
215 	void		*ipq_si;
216 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
217 };
218 
219 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
220 
221 static void if_percpuq_drops(void *, void *, struct cpu_info *);
222 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
223 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
224     struct if_percpuq *);
225 
226 struct if_deferred_start {
227 	struct ifnet	*ids_ifp;
228 	void		(*ids_if_start)(struct ifnet *);
229 	void		*ids_si;
230 };
231 
232 static void if_deferred_start_softint(void *);
233 static void if_deferred_start_common(struct ifnet *);
234 static void if_deferred_start_destroy(struct ifnet *);
235 
236 #if defined(INET) || defined(INET6)
237 static void sysctl_net_pktq_setup(struct sysctllog **, int);
238 #endif
239 
240 static void if_sysctl_setup(struct sysctllog **);
241 
242 /*
243  * Pointer to stub or real compat_cvtcmd() depending on presence of
244  * the compat module
245  */
246 u_long stub_compat_cvtcmd(u_long);
247 u_long (*vec_compat_cvtcmd)(u_long) = stub_compat_cvtcmd;
248 
249 /* Similarly, pointer to compat_ifioctl() if it is present */
250 
251 int (*vec_compat_ifioctl)(struct socket *, u_long, u_long, void *,
252 	struct lwp *) = NULL;
253 
254 /* The stub version of compat_cvtcmd() */
255 u_long stub_compat_cvtcmd(u_long cmd)
256 {
257 
258 	return cmd;
259 }
260 
261 static int
262 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
263     void *arg0, void *arg1, void *arg2, void *arg3)
264 {
265 	int result;
266 	enum kauth_network_req req;
267 
268 	result = KAUTH_RESULT_DEFER;
269 	req = (enum kauth_network_req)arg1;
270 
271 	if (action != KAUTH_NETWORK_INTERFACE)
272 		return result;
273 
274 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
275 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
276 		result = KAUTH_RESULT_ALLOW;
277 
278 	return result;
279 }
280 
281 /*
282  * Network interface utility routines.
283  *
284  * Routines with ifa_ifwith* names take sockaddr *'s as
285  * parameters.
286  */
287 void
288 ifinit(void)
289 {
290 
291 	if_sysctl_setup(NULL);
292 
293 #if (defined(INET) || defined(INET6))
294 	encapinit();
295 #endif
296 
297 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
298 	    if_listener_cb, NULL);
299 
300 	/* interfaces are available, inform socket code */
301 	ifioctl = doifioctl;
302 }
303 
304 /*
305  * XXX Initialization before configure().
306  * XXX hack to get pfil_add_hook working in autoconf.
307  */
308 void
309 ifinit1(void)
310 {
311 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
312 
313 	TAILQ_INIT(&ifnet_list);
314 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
315 	ifnet_psz = pserialize_create();
316 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
317 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
318 	PSLIST_INIT(&ifnet_pslist);
319 
320 	if_indexlim = 8;
321 
322 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
323 	KASSERT(if_pfil != NULL);
324 
325 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
326 	etherinit();
327 #endif
328 }
329 
330 ifnet_t *
331 if_alloc(u_char type)
332 {
333 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
334 }
335 
336 void
337 if_free(ifnet_t *ifp)
338 {
339 	kmem_free(ifp, sizeof(ifnet_t));
340 }
341 
342 void
343 if_initname(struct ifnet *ifp, const char *name, int unit)
344 {
345 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
346 	    "%s%d", name, unit);
347 }
348 
349 /*
350  * Null routines used while an interface is going away.  These routines
351  * just return an error.
352  */
353 
354 int
355 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
356     const struct sockaddr *so, const struct rtentry *rt)
357 {
358 
359 	return ENXIO;
360 }
361 
362 void
363 if_nullinput(struct ifnet *ifp, struct mbuf *m)
364 {
365 
366 	/* Nothing. */
367 }
368 
369 void
370 if_nullstart(struct ifnet *ifp)
371 {
372 
373 	/* Nothing. */
374 }
375 
376 int
377 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
378 {
379 
380 	m_freem(m);
381 	return ENXIO;
382 }
383 
384 int
385 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
386 {
387 
388 	return ENXIO;
389 }
390 
391 int
392 if_nullinit(struct ifnet *ifp)
393 {
394 
395 	return ENXIO;
396 }
397 
398 void
399 if_nullstop(struct ifnet *ifp, int disable)
400 {
401 
402 	/* Nothing. */
403 }
404 
405 void
406 if_nullslowtimo(struct ifnet *ifp)
407 {
408 
409 	/* Nothing. */
410 }
411 
412 void
413 if_nulldrain(struct ifnet *ifp)
414 {
415 
416 	/* Nothing. */
417 }
418 
419 void
420 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
421 {
422 	struct ifaddr *ifa;
423 	struct sockaddr_dl *sdl;
424 
425 	ifp->if_addrlen = addrlen;
426 	if_alloc_sadl(ifp);
427 	ifa = ifp->if_dl;
428 	sdl = satosdl(ifa->ifa_addr);
429 
430 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
431 	if (factory) {
432 		ifp->if_hwdl = ifp->if_dl;
433 		ifaref(ifp->if_hwdl);
434 	}
435 	/* TBD routing socket */
436 }
437 
438 struct ifaddr *
439 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
440 {
441 	unsigned socksize, ifasize;
442 	int addrlen, namelen;
443 	struct sockaddr_dl *mask, *sdl;
444 	struct ifaddr *ifa;
445 
446 	namelen = strlen(ifp->if_xname);
447 	addrlen = ifp->if_addrlen;
448 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
449 	ifasize = sizeof(*ifa) + 2 * socksize;
450 	ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
451 
452 	sdl = (struct sockaddr_dl *)(ifa + 1);
453 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
454 
455 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
456 	    ifp->if_xname, namelen, NULL, addrlen);
457 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
458 	memset(&mask->sdl_data[0], 0xff, namelen);
459 	ifa->ifa_rtrequest = link_rtrequest;
460 	ifa->ifa_addr = (struct sockaddr *)sdl;
461 	ifa->ifa_netmask = (struct sockaddr *)mask;
462 	ifa_psref_init(ifa);
463 
464 	*sdlp = sdl;
465 
466 	return ifa;
467 }
468 
469 static void
470 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
471 {
472 	const struct sockaddr_dl *sdl;
473 
474 	ifp->if_dl = ifa;
475 	ifaref(ifa);
476 	sdl = satosdl(ifa->ifa_addr);
477 	ifp->if_sadl = sdl;
478 }
479 
480 /*
481  * Allocate the link level name for the specified interface.  This
482  * is an attachment helper.  It must be called after ifp->if_addrlen
483  * is initialized, which may not be the case when if_attach() is
484  * called.
485  */
486 void
487 if_alloc_sadl(struct ifnet *ifp)
488 {
489 	struct ifaddr *ifa;
490 	const struct sockaddr_dl *sdl;
491 
492 	/*
493 	 * If the interface already has a link name, release it
494 	 * now.  This is useful for interfaces that can change
495 	 * link types, and thus switch link names often.
496 	 */
497 	if (ifp->if_sadl != NULL)
498 		if_free_sadl(ifp);
499 
500 	ifa = if_dl_create(ifp, &sdl);
501 
502 	ifa_insert(ifp, ifa);
503 	if_sadl_setrefs(ifp, ifa);
504 }
505 
506 static void
507 if_deactivate_sadl(struct ifnet *ifp)
508 {
509 	struct ifaddr *ifa;
510 
511 	KASSERT(ifp->if_dl != NULL);
512 
513 	ifa = ifp->if_dl;
514 
515 	ifp->if_sadl = NULL;
516 
517 	ifp->if_dl = NULL;
518 	ifafree(ifa);
519 }
520 
521 static void
522 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa)
523 {
524 	struct ifaddr *old;
525 
526 	KASSERT(ifp->if_dl != NULL);
527 
528 	old = ifp->if_dl;
529 
530 	ifaref(ifa);
531 	/* XXX Update if_dl and if_sadl atomically */
532 	ifp->if_dl = ifa;
533 	ifp->if_sadl = satosdl(ifa->ifa_addr);
534 
535 	ifafree(old);
536 }
537 
538 void
539 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
540     const struct sockaddr_dl *sdl)
541 {
542 	int s, ss;
543 	struct ifaddr *ifa;
544 	int bound = curlwp_bind();
545 
546 	KASSERT(ifa_held(ifa0));
547 
548 	s = splsoftnet();
549 
550 	if_replace_sadl(ifp, ifa0);
551 
552 	ss = pserialize_read_enter();
553 	IFADDR_READER_FOREACH(ifa, ifp) {
554 		struct psref psref;
555 		ifa_acquire(ifa, &psref);
556 		pserialize_read_exit(ss);
557 
558 		rtinit(ifa, RTM_LLINFO_UPD, 0);
559 
560 		ss = pserialize_read_enter();
561 		ifa_release(ifa, &psref);
562 	}
563 	pserialize_read_exit(ss);
564 
565 	splx(s);
566 	curlwp_bindx(bound);
567 }
568 
569 /*
570  * Free the link level name for the specified interface.  This is
571  * a detach helper.  This is called from if_detach().
572  */
573 static void
574 if_free_sadl(struct ifnet *ifp)
575 {
576 	struct ifaddr *ifa;
577 	int s;
578 
579 	ifa = ifp->if_dl;
580 	if (ifa == NULL) {
581 		KASSERT(ifp->if_sadl == NULL);
582 		return;
583 	}
584 
585 	KASSERT(ifp->if_sadl != NULL);
586 
587 	s = splsoftnet();
588 	rtinit(ifa, RTM_DELETE, 0);
589 	ifa_remove(ifp, ifa);
590 	if_deactivate_sadl(ifp);
591 	if (ifp->if_hwdl == ifa) {
592 		ifafree(ifa);
593 		ifp->if_hwdl = NULL;
594 	}
595 	splx(s);
596 }
597 
598 static void
599 if_getindex(ifnet_t *ifp)
600 {
601 	bool hitlimit = false;
602 
603 	ifp->if_index_gen = index_gen++;
604 
605 	ifp->if_index = if_index;
606 	if (ifindex2ifnet == NULL) {
607 		if_index++;
608 		goto skip;
609 	}
610 	while (if_byindex(ifp->if_index)) {
611 		/*
612 		 * If we hit USHRT_MAX, we skip back to 0 since
613 		 * there are a number of places where the value
614 		 * of if_index or if_index itself is compared
615 		 * to or stored in an unsigned short.  By
616 		 * jumping back, we won't botch those assignments
617 		 * or comparisons.
618 		 */
619 		if (++if_index == 0) {
620 			if_index = 1;
621 		} else if (if_index == USHRT_MAX) {
622 			/*
623 			 * However, if we have to jump back to
624 			 * zero *twice* without finding an empty
625 			 * slot in ifindex2ifnet[], then there
626 			 * there are too many (>65535) interfaces.
627 			 */
628 			if (hitlimit) {
629 				panic("too many interfaces");
630 			}
631 			hitlimit = true;
632 			if_index = 1;
633 		}
634 		ifp->if_index = if_index;
635 	}
636 skip:
637 	/*
638 	 * ifindex2ifnet is indexed by if_index. Since if_index will
639 	 * grow dynamically, it should grow too.
640 	 */
641 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
642 		size_t m, n, oldlim;
643 		void *q;
644 
645 		oldlim = if_indexlim;
646 		while (ifp->if_index >= if_indexlim)
647 			if_indexlim <<= 1;
648 
649 		/* grow ifindex2ifnet */
650 		m = oldlim * sizeof(struct ifnet *);
651 		n = if_indexlim * sizeof(struct ifnet *);
652 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
653 		if (ifindex2ifnet != NULL) {
654 			memcpy(q, ifindex2ifnet, m);
655 			free(ifindex2ifnet, M_IFADDR);
656 		}
657 		ifindex2ifnet = (struct ifnet **)q;
658 	}
659 	ifindex2ifnet[ifp->if_index] = ifp;
660 }
661 
662 /*
663  * Initialize an interface and assign an index for it.
664  *
665  * It must be called prior to a device specific attach routine
666  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
667  * and be followed by if_register:
668  *
669  *     if_initialize(ifp);
670  *     ether_ifattach(ifp, enaddr);
671  *     if_register(ifp);
672  */
673 int
674 if_initialize(ifnet_t *ifp)
675 {
676 	int rv = 0;
677 
678 	KASSERT(if_indexlim > 0);
679 	TAILQ_INIT(&ifp->if_addrlist);
680 
681 	/*
682 	 * Link level name is allocated later by a separate call to
683 	 * if_alloc_sadl().
684 	 */
685 
686 	if (ifp->if_snd.ifq_maxlen == 0)
687 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
688 
689 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
690 
691 	ifp->if_link_state = LINK_STATE_UNKNOWN;
692 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
693 
694 	ifp->if_capenable = 0;
695 	ifp->if_csum_flags_tx = 0;
696 	ifp->if_csum_flags_rx = 0;
697 
698 #ifdef ALTQ
699 	ifp->if_snd.altq_type = 0;
700 	ifp->if_snd.altq_disc = NULL;
701 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
702 	ifp->if_snd.altq_tbr  = NULL;
703 	ifp->if_snd.altq_ifp  = ifp;
704 #endif
705 
706 	IFQ_LOCK_INIT(&ifp->if_snd);
707 
708 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
709 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
710 
711 	IF_AFDATA_LOCK_INIT(ifp);
712 
713 	if (if_is_link_state_changeable(ifp)) {
714 		ifp->if_link_si = softint_establish(SOFTINT_NET,
715 		    if_link_state_change_si, ifp);
716 		if (ifp->if_link_si == NULL) {
717 			rv = ENOMEM;
718 			goto fail;
719 		}
720 	}
721 
722 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
723 	PSLIST_INIT(&ifp->if_addr_pslist);
724 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
725 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
726 	LIST_INIT(&ifp->if_multiaddrs);
727 
728 	IFNET_LOCK();
729 	if_getindex(ifp);
730 	IFNET_UNLOCK();
731 
732 	return 0;
733 
734 fail:
735 	IF_AFDATA_LOCK_DESTROY(ifp);
736 
737 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
738 	(void)pfil_head_destroy(ifp->if_pfil);
739 
740 	IFQ_LOCK_DESTROY(&ifp->if_snd);
741 
742 	return rv;
743 }
744 
745 /*
746  * Register an interface to the list of "active" interfaces.
747  */
748 void
749 if_register(ifnet_t *ifp)
750 {
751 	/*
752 	 * If the driver has not supplied its own if_ioctl, then
753 	 * supply the default.
754 	 */
755 	if (ifp->if_ioctl == NULL)
756 		ifp->if_ioctl = ifioctl_common;
757 
758 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
759 
760 	if (!STAILQ_EMPTY(&domains))
761 		if_attachdomain1(ifp);
762 
763 	/* Announce the interface. */
764 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
765 
766 	if (ifp->if_slowtimo != NULL) {
767 		ifp->if_slowtimo_ch =
768 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
769 		callout_init(ifp->if_slowtimo_ch, 0);
770 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
771 		if_slowtimo(ifp);
772 	}
773 
774 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
775 		ifp->if_transmit = if_transmit;
776 
777 	IFNET_LOCK();
778 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
779 	IFNET_WRITER_INSERT_TAIL(ifp);
780 	IFNET_UNLOCK();
781 }
782 
783 /*
784  * The if_percpuq framework
785  *
786  * It allows network device drivers to execute the network stack
787  * in softint (so called softint-based if_input). It utilizes
788  * softint and percpu ifqueue. It doesn't distribute any packets
789  * between CPUs, unlike pktqueue(9).
790  *
791  * Currently we support two options for device drivers to apply the framework:
792  * - Use it implicitly with less changes
793  *   - If you use if_attach in driver's _attach function and if_input in
794  *     driver's Rx interrupt handler, a packet is queued and a softint handles
795  *     the packet implicitly
796  * - Use it explicitly in each driver (recommended)
797  *   - You can use if_percpuq_* directly in your driver
798  *   - In this case, you need to allocate struct if_percpuq in driver's softc
799  *   - See wm(4) as a reference implementation
800  */
801 
802 static void
803 if_percpuq_softint(void *arg)
804 {
805 	struct if_percpuq *ipq = arg;
806 	struct ifnet *ifp = ipq->ipq_ifp;
807 	struct mbuf *m;
808 
809 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
810 		ifp->if_ipackets++;
811 		bpf_mtap(ifp, m);
812 
813 		ifp->_if_input(ifp, m);
814 	}
815 }
816 
817 static void
818 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
819 {
820 	struct ifqueue *const ifq = p;
821 
822 	memset(ifq, 0, sizeof(*ifq));
823 	ifq->ifq_maxlen = IFQ_MAXLEN;
824 }
825 
826 struct if_percpuq *
827 if_percpuq_create(struct ifnet *ifp)
828 {
829 	struct if_percpuq *ipq;
830 
831 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
832 	ipq->ipq_ifp = ifp;
833 	ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
834 	    if_percpuq_softint, ipq);
835 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
836 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
837 
838 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
839 
840 	return ipq;
841 }
842 
843 static struct mbuf *
844 if_percpuq_dequeue(struct if_percpuq *ipq)
845 {
846 	struct mbuf *m;
847 	struct ifqueue *ifq;
848 	int s;
849 
850 	s = splnet();
851 	ifq = percpu_getref(ipq->ipq_ifqs);
852 	IF_DEQUEUE(ifq, m);
853 	percpu_putref(ipq->ipq_ifqs);
854 	splx(s);
855 
856 	return m;
857 }
858 
859 static void
860 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
861 {
862 	struct ifqueue *const ifq = p;
863 
864 	IF_PURGE(ifq);
865 }
866 
867 void
868 if_percpuq_destroy(struct if_percpuq *ipq)
869 {
870 
871 	/* if_detach may already destroy it */
872 	if (ipq == NULL)
873 		return;
874 
875 	softint_disestablish(ipq->ipq_si);
876 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
877 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
878 	kmem_free(ipq, sizeof(*ipq));
879 }
880 
881 void
882 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
883 {
884 	struct ifqueue *ifq;
885 	int s;
886 
887 	KASSERT(ipq != NULL);
888 
889 	s = splnet();
890 	ifq = percpu_getref(ipq->ipq_ifqs);
891 	if (IF_QFULL(ifq)) {
892 		IF_DROP(ifq);
893 		percpu_putref(ipq->ipq_ifqs);
894 		m_freem(m);
895 		goto out;
896 	}
897 	IF_ENQUEUE(ifq, m);
898 	percpu_putref(ipq->ipq_ifqs);
899 
900 	softint_schedule(ipq->ipq_si);
901 out:
902 	splx(s);
903 }
904 
905 static void
906 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
907 {
908 	struct ifqueue *const ifq = p;
909 	int *sum = arg;
910 
911 	*sum += ifq->ifq_drops;
912 }
913 
914 static int
915 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
916 {
917 	struct sysctlnode node;
918 	struct if_percpuq *ipq;
919 	int sum = 0;
920 	int error;
921 
922 	node = *rnode;
923 	ipq = node.sysctl_data;
924 
925 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
926 
927 	node.sysctl_data = &sum;
928 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
929 	if (error != 0 || newp == NULL)
930 		return error;
931 
932 	return 0;
933 }
934 
935 static void
936 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
937     struct if_percpuq *ipq)
938 {
939 	const struct sysctlnode *cnode, *rnode;
940 
941 	if (sysctl_createv(clog, 0, NULL, &rnode,
942 		       CTLFLAG_PERMANENT,
943 		       CTLTYPE_NODE, "interfaces",
944 		       SYSCTL_DESCR("Per-interface controls"),
945 		       NULL, 0, NULL, 0,
946 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
947 		goto bad;
948 
949 	if (sysctl_createv(clog, 0, &rnode, &rnode,
950 		       CTLFLAG_PERMANENT,
951 		       CTLTYPE_NODE, ifname,
952 		       SYSCTL_DESCR("Interface controls"),
953 		       NULL, 0, NULL, 0,
954 		       CTL_CREATE, CTL_EOL) != 0)
955 		goto bad;
956 
957 	if (sysctl_createv(clog, 0, &rnode, &rnode,
958 		       CTLFLAG_PERMANENT,
959 		       CTLTYPE_NODE, "rcvq",
960 		       SYSCTL_DESCR("Interface input queue controls"),
961 		       NULL, 0, NULL, 0,
962 		       CTL_CREATE, CTL_EOL) != 0)
963 		goto bad;
964 
965 #ifdef NOTYET
966 	/* XXX Should show each per-CPU queue length? */
967 	if (sysctl_createv(clog, 0, &rnode, &rnode,
968 		       CTLFLAG_PERMANENT,
969 		       CTLTYPE_INT, "len",
970 		       SYSCTL_DESCR("Current input queue length"),
971 		       sysctl_percpuq_len, 0, NULL, 0,
972 		       CTL_CREATE, CTL_EOL) != 0)
973 		goto bad;
974 
975 	if (sysctl_createv(clog, 0, &rnode, &cnode,
976 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
977 		       CTLTYPE_INT, "maxlen",
978 		       SYSCTL_DESCR("Maximum allowed input queue length"),
979 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
980 		       CTL_CREATE, CTL_EOL) != 0)
981 		goto bad;
982 #endif
983 
984 	if (sysctl_createv(clog, 0, &rnode, &cnode,
985 		       CTLFLAG_PERMANENT,
986 		       CTLTYPE_INT, "drops",
987 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
988 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
989 		       CTL_CREATE, CTL_EOL) != 0)
990 		goto bad;
991 
992 	return;
993 bad:
994 	printf("%s: could not attach sysctl nodes\n", ifname);
995 	return;
996 }
997 
998 /*
999  * The deferred if_start framework
1000  *
1001  * The common APIs to defer if_start to softint when if_start is requested
1002  * from a device driver running in hardware interrupt context.
1003  */
1004 /*
1005  * Call ifp->if_start (or equivalent) in a dedicated softint for
1006  * deferred if_start.
1007  */
1008 static void
1009 if_deferred_start_softint(void *arg)
1010 {
1011 	struct if_deferred_start *ids = arg;
1012 	struct ifnet *ifp = ids->ids_ifp;
1013 
1014 	ids->ids_if_start(ifp);
1015 }
1016 
1017 /*
1018  * The default callback function for deferred if_start.
1019  */
1020 static void
1021 if_deferred_start_common(struct ifnet *ifp)
1022 {
1023 	int s;
1024 
1025 	s = splnet();
1026 	if_start_lock(ifp);
1027 	splx(s);
1028 }
1029 
1030 static inline bool
1031 if_snd_is_used(struct ifnet *ifp)
1032 {
1033 
1034 	return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit ||
1035 	    ALTQ_IS_ENABLED(&ifp->if_snd);
1036 }
1037 
1038 /*
1039  * Schedule deferred if_start.
1040  */
1041 void
1042 if_schedule_deferred_start(struct ifnet *ifp)
1043 {
1044 
1045 	KASSERT(ifp->if_deferred_start != NULL);
1046 
1047 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1048 		return;
1049 
1050 	softint_schedule(ifp->if_deferred_start->ids_si);
1051 }
1052 
1053 /*
1054  * Create an instance of deferred if_start. A driver should call the function
1055  * only if the driver needs deferred if_start. Drivers can setup their own
1056  * deferred if_start function via 2nd argument.
1057  */
1058 void
1059 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1060 {
1061 	struct if_deferred_start *ids;
1062 
1063 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1064 	ids->ids_ifp = ifp;
1065 	ids->ids_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
1066 	    if_deferred_start_softint, ids);
1067 	if (func != NULL)
1068 		ids->ids_if_start = func;
1069 	else
1070 		ids->ids_if_start = if_deferred_start_common;
1071 
1072 	ifp->if_deferred_start = ids;
1073 }
1074 
1075 static void
1076 if_deferred_start_destroy(struct ifnet *ifp)
1077 {
1078 
1079 	if (ifp->if_deferred_start == NULL)
1080 		return;
1081 
1082 	softint_disestablish(ifp->if_deferred_start->ids_si);
1083 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1084 	ifp->if_deferred_start = NULL;
1085 }
1086 
1087 /*
1088  * The common interface input routine that is called by device drivers,
1089  * which should be used only when the driver's rx handler already runs
1090  * in softint.
1091  */
1092 void
1093 if_input(struct ifnet *ifp, struct mbuf *m)
1094 {
1095 
1096 	KASSERT(ifp->if_percpuq == NULL);
1097 	KASSERT(!cpu_intr_p());
1098 
1099 	ifp->if_ipackets++;
1100 	bpf_mtap(ifp, m);
1101 
1102 	ifp->_if_input(ifp, m);
1103 }
1104 
1105 /*
1106  * DEPRECATED. Use if_initialize and if_register instead.
1107  * See the above comment of if_initialize.
1108  *
1109  * Note that it implicitly enables if_percpuq to make drivers easy to
1110  * migrate softint-based if_input without much changes. If you don't
1111  * want to enable it, use if_initialize instead.
1112  */
1113 int
1114 if_attach(ifnet_t *ifp)
1115 {
1116 	int rv;
1117 
1118 	rv = if_initialize(ifp);
1119 	if (rv != 0)
1120 		return rv;
1121 
1122 	ifp->if_percpuq = if_percpuq_create(ifp);
1123 	if_register(ifp);
1124 
1125 	return 0;
1126 }
1127 
1128 void
1129 if_attachdomain(void)
1130 {
1131 	struct ifnet *ifp;
1132 	int s;
1133 	int bound = curlwp_bind();
1134 
1135 	s = pserialize_read_enter();
1136 	IFNET_READER_FOREACH(ifp) {
1137 		struct psref psref;
1138 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1139 		pserialize_read_exit(s);
1140 		if_attachdomain1(ifp);
1141 		s = pserialize_read_enter();
1142 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1143 	}
1144 	pserialize_read_exit(s);
1145 	curlwp_bindx(bound);
1146 }
1147 
1148 static void
1149 if_attachdomain1(struct ifnet *ifp)
1150 {
1151 	struct domain *dp;
1152 	int s;
1153 
1154 	s = splsoftnet();
1155 
1156 	/* address family dependent data region */
1157 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1158 	DOMAIN_FOREACH(dp) {
1159 		if (dp->dom_ifattach != NULL)
1160 			ifp->if_afdata[dp->dom_family] =
1161 			    (*dp->dom_ifattach)(ifp);
1162 	}
1163 
1164 	splx(s);
1165 }
1166 
1167 /*
1168  * Deactivate an interface.  This points all of the procedure
1169  * handles at error stubs.  May be called from interrupt context.
1170  */
1171 void
1172 if_deactivate(struct ifnet *ifp)
1173 {
1174 	int s;
1175 
1176 	s = splsoftnet();
1177 
1178 	ifp->if_output	 = if_nulloutput;
1179 	ifp->_if_input	 = if_nullinput;
1180 	ifp->if_start	 = if_nullstart;
1181 	ifp->if_transmit = if_nulltransmit;
1182 	ifp->if_ioctl	 = if_nullioctl;
1183 	ifp->if_init	 = if_nullinit;
1184 	ifp->if_stop	 = if_nullstop;
1185 	ifp->if_slowtimo = if_nullslowtimo;
1186 	ifp->if_drain	 = if_nulldrain;
1187 
1188 	/* No more packets may be enqueued. */
1189 	ifp->if_snd.ifq_maxlen = 0;
1190 
1191 	splx(s);
1192 }
1193 
1194 bool
1195 if_is_deactivated(const struct ifnet *ifp)
1196 {
1197 
1198 	return ifp->if_output == if_nulloutput;
1199 }
1200 
1201 void
1202 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1203 {
1204 	struct ifaddr *ifa, *nifa;
1205 	int s;
1206 
1207 	s = pserialize_read_enter();
1208 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1209 		nifa = IFADDR_READER_NEXT(ifa);
1210 		if (ifa->ifa_addr->sa_family != family)
1211 			continue;
1212 		pserialize_read_exit(s);
1213 
1214 		(*purgeaddr)(ifa);
1215 
1216 		s = pserialize_read_enter();
1217 	}
1218 	pserialize_read_exit(s);
1219 }
1220 
1221 #ifdef IFAREF_DEBUG
1222 static struct ifaddr **ifa_list;
1223 static int ifa_list_size;
1224 
1225 /* Depends on only one if_attach runs at once */
1226 static void
1227 if_build_ifa_list(struct ifnet *ifp)
1228 {
1229 	struct ifaddr *ifa;
1230 	int i;
1231 
1232 	KASSERT(ifa_list == NULL);
1233 	KASSERT(ifa_list_size == 0);
1234 
1235 	IFADDR_READER_FOREACH(ifa, ifp)
1236 		ifa_list_size++;
1237 
1238 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1239 	i = 0;
1240 	IFADDR_READER_FOREACH(ifa, ifp) {
1241 		ifa_list[i++] = ifa;
1242 		ifaref(ifa);
1243 	}
1244 }
1245 
1246 static void
1247 if_check_and_free_ifa_list(struct ifnet *ifp)
1248 {
1249 	int i;
1250 	struct ifaddr *ifa;
1251 
1252 	if (ifa_list == NULL)
1253 		return;
1254 
1255 	for (i = 0; i < ifa_list_size; i++) {
1256 		char buf[64];
1257 
1258 		ifa = ifa_list[i];
1259 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1260 		if (ifa->ifa_refcnt > 1) {
1261 			log(LOG_WARNING,
1262 			    "ifa(%s) still referenced (refcnt=%d)\n",
1263 			    buf, ifa->ifa_refcnt - 1);
1264 		} else
1265 			log(LOG_DEBUG,
1266 			    "ifa(%s) not referenced (refcnt=%d)\n",
1267 			    buf, ifa->ifa_refcnt - 1);
1268 		ifafree(ifa);
1269 	}
1270 
1271 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1272 	ifa_list = NULL;
1273 	ifa_list_size = 0;
1274 }
1275 #endif
1276 
1277 /*
1278  * Detach an interface from the list of "active" interfaces,
1279  * freeing any resources as we go along.
1280  *
1281  * NOTE: This routine must be called with a valid thread context,
1282  * as it may block.
1283  */
1284 void
1285 if_detach(struct ifnet *ifp)
1286 {
1287 	struct socket so;
1288 	struct ifaddr *ifa;
1289 #ifdef IFAREF_DEBUG
1290 	struct ifaddr *last_ifa = NULL;
1291 #endif
1292 	struct domain *dp;
1293 	const struct protosw *pr;
1294 	int s, i, family, purged;
1295 	uint64_t xc;
1296 
1297 #ifdef IFAREF_DEBUG
1298 	if_build_ifa_list(ifp);
1299 #endif
1300 	/*
1301 	 * XXX It's kind of lame that we have to have the
1302 	 * XXX socket structure...
1303 	 */
1304 	memset(&so, 0, sizeof(so));
1305 
1306 	s = splnet();
1307 
1308 	sysctl_teardown(&ifp->if_sysctl_log);
1309 	mutex_enter(ifp->if_ioctl_lock);
1310 	if_deactivate(ifp);
1311 	mutex_exit(ifp->if_ioctl_lock);
1312 
1313 	IFNET_LOCK();
1314 	ifindex2ifnet[ifp->if_index] = NULL;
1315 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1316 	IFNET_WRITER_REMOVE(ifp);
1317 	pserialize_perform(ifnet_psz);
1318 	IFNET_UNLOCK();
1319 
1320 	/* Wait for all readers to drain before freeing.  */
1321 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1322 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1323 
1324 	mutex_obj_free(ifp->if_ioctl_lock);
1325 	ifp->if_ioctl_lock = NULL;
1326 
1327 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1328 		ifp->if_slowtimo = NULL;
1329 		callout_halt(ifp->if_slowtimo_ch, NULL);
1330 		callout_destroy(ifp->if_slowtimo_ch);
1331 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1332 	}
1333 	if_deferred_start_destroy(ifp);
1334 
1335 	/*
1336 	 * Do an if_down() to give protocols a chance to do something.
1337 	 */
1338 	if_down(ifp);
1339 
1340 #ifdef ALTQ
1341 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1342 		altq_disable(&ifp->if_snd);
1343 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1344 		altq_detach(&ifp->if_snd);
1345 #endif
1346 
1347 	mutex_obj_free(ifp->if_snd.ifq_lock);
1348 
1349 #if NCARP > 0
1350 	/* Remove the interface from any carp group it is a part of.  */
1351 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1352 		carp_ifdetach(ifp);
1353 #endif
1354 
1355 	/*
1356 	 * Rip all the addresses off the interface.  This should make
1357 	 * all of the routes go away.
1358 	 *
1359 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1360 	 * from the list, including our "cursor", ifa.  For safety,
1361 	 * and to honor the TAILQ abstraction, I just restart the
1362 	 * loop after each removal.  Note that the loop will exit
1363 	 * when all of the remaining ifaddrs belong to the AF_LINK
1364 	 * family.  I am counting on the historical fact that at
1365 	 * least one pr_usrreq in each address domain removes at
1366 	 * least one ifaddr.
1367 	 */
1368 again:
1369 	/*
1370 	 * At this point, no other one tries to remove ifa in the list,
1371 	 * so we don't need to take a lock or psref.
1372 	 */
1373 	IFADDR_READER_FOREACH(ifa, ifp) {
1374 		family = ifa->ifa_addr->sa_family;
1375 #ifdef IFAREF_DEBUG
1376 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1377 		    ifa, family, ifa->ifa_refcnt);
1378 		if (last_ifa != NULL && ifa == last_ifa)
1379 			panic("if_detach: loop detected");
1380 		last_ifa = ifa;
1381 #endif
1382 		if (family == AF_LINK)
1383 			continue;
1384 		dp = pffinddomain(family);
1385 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1386 		/*
1387 		 * XXX These PURGEIF calls are redundant with the
1388 		 * purge-all-families calls below, but are left in for
1389 		 * now both to make a smaller change, and to avoid
1390 		 * unplanned interactions with clearing of
1391 		 * ifp->if_addrlist.
1392 		 */
1393 		purged = 0;
1394 		for (pr = dp->dom_protosw;
1395 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1396 			so.so_proto = pr;
1397 			if (pr->pr_usrreqs) {
1398 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1399 				purged = 1;
1400 			}
1401 		}
1402 		if (purged == 0) {
1403 			/*
1404 			 * XXX What's really the best thing to do
1405 			 * XXX here?  --thorpej@NetBSD.org
1406 			 */
1407 			printf("if_detach: WARNING: AF %d not purged\n",
1408 			    family);
1409 			ifa_remove(ifp, ifa);
1410 		}
1411 		goto again;
1412 	}
1413 
1414 	if_free_sadl(ifp);
1415 
1416 	/* Delete stray routes from the routing table. */
1417 	for (i = 0; i <= AF_MAX; i++)
1418 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1419 
1420 	DOMAIN_FOREACH(dp) {
1421 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1422 		{
1423 			void *p = ifp->if_afdata[dp->dom_family];
1424 			if (p) {
1425 				ifp->if_afdata[dp->dom_family] = NULL;
1426 				(*dp->dom_ifdetach)(ifp, p);
1427 			}
1428 		}
1429 
1430 		/*
1431 		 * One would expect multicast memberships (INET and
1432 		 * INET6) on UDP sockets to be purged by the PURGEIF
1433 		 * calls above, but if all addresses were removed from
1434 		 * the interface prior to destruction, the calls will
1435 		 * not be made (e.g. ppp, for which pppd(8) generally
1436 		 * removes addresses before destroying the interface).
1437 		 * Because there is no invariant that multicast
1438 		 * memberships only exist for interfaces with IPv4
1439 		 * addresses, we must call PURGEIF regardless of
1440 		 * addresses.  (Protocols which might store ifnet
1441 		 * pointers are marked with PR_PURGEIF.)
1442 		 */
1443 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1444 			so.so_proto = pr;
1445 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1446 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1447 		}
1448 	}
1449 
1450 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1451 	(void)pfil_head_destroy(ifp->if_pfil);
1452 
1453 	/* Announce that the interface is gone. */
1454 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1455 
1456 	IF_AFDATA_LOCK_DESTROY(ifp);
1457 
1458 	if (if_is_link_state_changeable(ifp)) {
1459 		softint_disestablish(ifp->if_link_si);
1460 		ifp->if_link_si = NULL;
1461 	}
1462 
1463 	/*
1464 	 * remove packets that came from ifp, from software interrupt queues.
1465 	 */
1466 	DOMAIN_FOREACH(dp) {
1467 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1468 			struct ifqueue *iq = dp->dom_ifqueues[i];
1469 			if (iq == NULL)
1470 				break;
1471 			dp->dom_ifqueues[i] = NULL;
1472 			if_detach_queues(ifp, iq);
1473 		}
1474 	}
1475 
1476 	/*
1477 	 * IP queues have to be processed separately: net-queue barrier
1478 	 * ensures that the packets are dequeued while a cross-call will
1479 	 * ensure that the interrupts have completed. FIXME: not quite..
1480 	 */
1481 #ifdef INET
1482 	pktq_barrier(ip_pktq);
1483 #endif
1484 #ifdef INET6
1485 	if (in6_present)
1486 		pktq_barrier(ip6_pktq);
1487 #endif
1488 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1489 	xc_wait(xc);
1490 
1491 	if (ifp->if_percpuq != NULL) {
1492 		if_percpuq_destroy(ifp->if_percpuq);
1493 		ifp->if_percpuq = NULL;
1494 	}
1495 
1496 	splx(s);
1497 
1498 #ifdef IFAREF_DEBUG
1499 	if_check_and_free_ifa_list(ifp);
1500 #endif
1501 }
1502 
1503 static void
1504 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1505 {
1506 	struct mbuf *m, *prev, *next;
1507 
1508 	prev = NULL;
1509 	for (m = q->ifq_head; m != NULL; m = next) {
1510 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1511 
1512 		next = m->m_nextpkt;
1513 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1514 			prev = m;
1515 			continue;
1516 		}
1517 
1518 		if (prev != NULL)
1519 			prev->m_nextpkt = m->m_nextpkt;
1520 		else
1521 			q->ifq_head = m->m_nextpkt;
1522 		if (q->ifq_tail == m)
1523 			q->ifq_tail = prev;
1524 		q->ifq_len--;
1525 
1526 		m->m_nextpkt = NULL;
1527 		m_freem(m);
1528 		IF_DROP(q);
1529 	}
1530 }
1531 
1532 /*
1533  * Callback for a radix tree walk to delete all references to an
1534  * ifnet.
1535  */
1536 static int
1537 if_delroute_matcher(struct rtentry *rt, void *v)
1538 {
1539 	struct ifnet *ifp = (struct ifnet *)v;
1540 
1541 	if (rt->rt_ifp == ifp)
1542 		return 1;
1543 	else
1544 		return 0;
1545 }
1546 
1547 /*
1548  * Create a clone network interface.
1549  */
1550 static int
1551 if_clone_create(const char *name)
1552 {
1553 	struct if_clone *ifc;
1554 	int unit;
1555 	struct ifnet *ifp;
1556 	struct psref psref;
1557 
1558 	KASSERT(mutex_owned(&if_clone_mtx));
1559 
1560 	ifc = if_clone_lookup(name, &unit);
1561 	if (ifc == NULL)
1562 		return EINVAL;
1563 
1564 	ifp = if_get(name, &psref);
1565 	if (ifp != NULL) {
1566 		if_put(ifp, &psref);
1567 		return EEXIST;
1568 	}
1569 
1570 	return (*ifc->ifc_create)(ifc, unit);
1571 }
1572 
1573 /*
1574  * Destroy a clone network interface.
1575  */
1576 static int
1577 if_clone_destroy(const char *name)
1578 {
1579 	struct if_clone *ifc;
1580 	struct ifnet *ifp;
1581 	struct psref psref;
1582 
1583 	KASSERT(mutex_owned(&if_clone_mtx));
1584 
1585 	ifc = if_clone_lookup(name, NULL);
1586 	if (ifc == NULL)
1587 		return EINVAL;
1588 
1589 	if (ifc->ifc_destroy == NULL)
1590 		return EOPNOTSUPP;
1591 
1592 	ifp = if_get(name, &psref);
1593 	if (ifp == NULL)
1594 		return ENXIO;
1595 
1596 	/* We have to disable ioctls here */
1597 	mutex_enter(ifp->if_ioctl_lock);
1598 	ifp->if_ioctl = if_nullioctl;
1599 	mutex_exit(ifp->if_ioctl_lock);
1600 
1601 	/*
1602 	 * We cannot call ifc_destroy with holding ifp.
1603 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1604 	 */
1605 	if_put(ifp, &psref);
1606 
1607 	return (*ifc->ifc_destroy)(ifp);
1608 }
1609 
1610 static bool
1611 if_is_unit(const char *name)
1612 {
1613 
1614 	while(*name != '\0') {
1615 		if (*name < '0' || *name > '9')
1616 			return false;
1617 		name++;
1618 	}
1619 
1620 	return true;
1621 }
1622 
1623 /*
1624  * Look up a network interface cloner.
1625  */
1626 static struct if_clone *
1627 if_clone_lookup(const char *name, int *unitp)
1628 {
1629 	struct if_clone *ifc;
1630 	const char *cp;
1631 	char *dp, ifname[IFNAMSIZ + 3];
1632 	int unit;
1633 
1634 	KASSERT(mutex_owned(&if_clone_mtx));
1635 
1636 	strcpy(ifname, "if_");
1637 	/* separate interface name from unit */
1638 	/* TODO: search unit number from backward */
1639 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1640 	    *cp && !if_is_unit(cp);)
1641 		*dp++ = *cp++;
1642 
1643 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1644 		return NULL;	/* No name or unit number */
1645 	*dp++ = '\0';
1646 
1647 again:
1648 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1649 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1650 			break;
1651 	}
1652 
1653 	if (ifc == NULL) {
1654 		int error;
1655 		if (*ifname == '\0')
1656 			return NULL;
1657 		mutex_exit(&if_clone_mtx);
1658 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1659 		mutex_enter(&if_clone_mtx);
1660 		if (error)
1661 			return NULL;
1662 		*ifname = '\0';
1663 		goto again;
1664 	}
1665 
1666 	unit = 0;
1667 	while (cp - name < IFNAMSIZ && *cp) {
1668 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1669 			/* Bogus unit number. */
1670 			return NULL;
1671 		}
1672 		unit = (unit * 10) + (*cp++ - '0');
1673 	}
1674 
1675 	if (unitp != NULL)
1676 		*unitp = unit;
1677 	return ifc;
1678 }
1679 
1680 /*
1681  * Register a network interface cloner.
1682  */
1683 void
1684 if_clone_attach(struct if_clone *ifc)
1685 {
1686 
1687 	mutex_enter(&if_clone_mtx);
1688 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1689 	if_cloners_count++;
1690 	mutex_exit(&if_clone_mtx);
1691 }
1692 
1693 /*
1694  * Unregister a network interface cloner.
1695  */
1696 void
1697 if_clone_detach(struct if_clone *ifc)
1698 {
1699 
1700 	mutex_enter(&if_clone_mtx);
1701 	LIST_REMOVE(ifc, ifc_list);
1702 	if_cloners_count--;
1703 	mutex_exit(&if_clone_mtx);
1704 }
1705 
1706 /*
1707  * Provide list of interface cloners to userspace.
1708  */
1709 int
1710 if_clone_list(int buf_count, char *buffer, int *total)
1711 {
1712 	char outbuf[IFNAMSIZ], *dst;
1713 	struct if_clone *ifc;
1714 	int count, error = 0;
1715 
1716 	mutex_enter(&if_clone_mtx);
1717 	*total = if_cloners_count;
1718 	if ((dst = buffer) == NULL) {
1719 		/* Just asking how many there are. */
1720 		goto out;
1721 	}
1722 
1723 	if (buf_count < 0) {
1724 		error = EINVAL;
1725 		goto out;
1726 	}
1727 
1728 	count = (if_cloners_count < buf_count) ?
1729 	    if_cloners_count : buf_count;
1730 
1731 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1732 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1733 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1734 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1735 			error = ENAMETOOLONG;
1736 			goto out;
1737 		}
1738 		error = copyout(outbuf, dst, sizeof(outbuf));
1739 		if (error != 0)
1740 			break;
1741 	}
1742 
1743 out:
1744 	mutex_exit(&if_clone_mtx);
1745 	return error;
1746 }
1747 
1748 void
1749 ifa_psref_init(struct ifaddr *ifa)
1750 {
1751 
1752 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1753 }
1754 
1755 void
1756 ifaref(struct ifaddr *ifa)
1757 {
1758 	KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING));
1759 	ifa->ifa_refcnt++;
1760 }
1761 
1762 void
1763 ifafree(struct ifaddr *ifa)
1764 {
1765 	KASSERT(ifa != NULL);
1766 	KASSERT(ifa->ifa_refcnt > 0);
1767 
1768 	if (--ifa->ifa_refcnt == 0) {
1769 		free(ifa, M_IFADDR);
1770 	}
1771 }
1772 
1773 bool
1774 ifa_is_destroying(struct ifaddr *ifa)
1775 {
1776 
1777 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1778 }
1779 
1780 void
1781 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1782 {
1783 
1784 	ifa->ifa_ifp = ifp;
1785 
1786 	IFNET_LOCK();
1787 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1788 	IFADDR_ENTRY_INIT(ifa);
1789 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1790 	IFNET_UNLOCK();
1791 
1792 	ifaref(ifa);
1793 }
1794 
1795 void
1796 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1797 {
1798 
1799 	KASSERT(ifa->ifa_ifp == ifp);
1800 
1801 	IFNET_LOCK();
1802 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1803 	IFADDR_WRITER_REMOVE(ifa);
1804 #ifdef NET_MPSAFE
1805 	pserialize_perform(ifnet_psz);
1806 #endif
1807 	IFNET_UNLOCK();
1808 
1809 #ifdef NET_MPSAFE
1810 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1811 #endif
1812 	IFADDR_ENTRY_DESTROY(ifa);
1813 	ifafree(ifa);
1814 }
1815 
1816 void
1817 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1818 {
1819 
1820 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1821 }
1822 
1823 void
1824 ifa_release(struct ifaddr *ifa, struct psref *psref)
1825 {
1826 
1827 	if (ifa == NULL)
1828 		return;
1829 
1830 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1831 }
1832 
1833 bool
1834 ifa_held(struct ifaddr *ifa)
1835 {
1836 
1837 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1838 }
1839 
1840 static inline int
1841 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1842 {
1843 	return sockaddr_cmp(sa1, sa2) == 0;
1844 }
1845 
1846 /*
1847  * Locate an interface based on a complete address.
1848  */
1849 /*ARGSUSED*/
1850 struct ifaddr *
1851 ifa_ifwithaddr(const struct sockaddr *addr)
1852 {
1853 	struct ifnet *ifp;
1854 	struct ifaddr *ifa;
1855 
1856 	IFNET_READER_FOREACH(ifp) {
1857 		if (if_is_deactivated(ifp))
1858 			continue;
1859 		IFADDR_READER_FOREACH(ifa, ifp) {
1860 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1861 				continue;
1862 			if (equal(addr, ifa->ifa_addr))
1863 				return ifa;
1864 			if ((ifp->if_flags & IFF_BROADCAST) &&
1865 			    ifa->ifa_broadaddr &&
1866 			    /* IP6 doesn't have broadcast */
1867 			    ifa->ifa_broadaddr->sa_len != 0 &&
1868 			    equal(ifa->ifa_broadaddr, addr))
1869 				return ifa;
1870 		}
1871 	}
1872 	return NULL;
1873 }
1874 
1875 struct ifaddr *
1876 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1877 {
1878 	struct ifaddr *ifa;
1879 	int s = pserialize_read_enter();
1880 
1881 	ifa = ifa_ifwithaddr(addr);
1882 	if (ifa != NULL)
1883 		ifa_acquire(ifa, psref);
1884 	pserialize_read_exit(s);
1885 
1886 	return ifa;
1887 }
1888 
1889 /*
1890  * Locate the point to point interface with a given destination address.
1891  */
1892 /*ARGSUSED*/
1893 struct ifaddr *
1894 ifa_ifwithdstaddr(const struct sockaddr *addr)
1895 {
1896 	struct ifnet *ifp;
1897 	struct ifaddr *ifa;
1898 
1899 	IFNET_READER_FOREACH(ifp) {
1900 		if (if_is_deactivated(ifp))
1901 			continue;
1902 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1903 			continue;
1904 		IFADDR_READER_FOREACH(ifa, ifp) {
1905 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1906 			    ifa->ifa_dstaddr == NULL)
1907 				continue;
1908 			if (equal(addr, ifa->ifa_dstaddr))
1909 				return ifa;
1910 		}
1911 	}
1912 
1913 	return NULL;
1914 }
1915 
1916 struct ifaddr *
1917 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1918 {
1919 	struct ifaddr *ifa;
1920 	int s;
1921 
1922 	s = pserialize_read_enter();
1923 	ifa = ifa_ifwithdstaddr(addr);
1924 	if (ifa != NULL)
1925 		ifa_acquire(ifa, psref);
1926 	pserialize_read_exit(s);
1927 
1928 	return ifa;
1929 }
1930 
1931 /*
1932  * Find an interface on a specific network.  If many, choice
1933  * is most specific found.
1934  */
1935 struct ifaddr *
1936 ifa_ifwithnet(const struct sockaddr *addr)
1937 {
1938 	struct ifnet *ifp;
1939 	struct ifaddr *ifa, *ifa_maybe = NULL;
1940 	const struct sockaddr_dl *sdl;
1941 	u_int af = addr->sa_family;
1942 	const char *addr_data = addr->sa_data, *cplim;
1943 
1944 	if (af == AF_LINK) {
1945 		sdl = satocsdl(addr);
1946 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1947 		    ifindex2ifnet[sdl->sdl_index] &&
1948 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1949 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1950 		}
1951 	}
1952 #ifdef NETATALK
1953 	if (af == AF_APPLETALK) {
1954 		const struct sockaddr_at *sat, *sat2;
1955 		sat = (const struct sockaddr_at *)addr;
1956 		IFNET_READER_FOREACH(ifp) {
1957 			if (if_is_deactivated(ifp))
1958 				continue;
1959 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1960 			if (ifa == NULL)
1961 				continue;
1962 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1963 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1964 				return ifa; /* exact match */
1965 			if (ifa_maybe == NULL) {
1966 				/* else keep the if with the right range */
1967 				ifa_maybe = ifa;
1968 			}
1969 		}
1970 		return ifa_maybe;
1971 	}
1972 #endif
1973 	IFNET_READER_FOREACH(ifp) {
1974 		if (if_is_deactivated(ifp))
1975 			continue;
1976 		IFADDR_READER_FOREACH(ifa, ifp) {
1977 			const char *cp, *cp2, *cp3;
1978 
1979 			if (ifa->ifa_addr->sa_family != af ||
1980 			    ifa->ifa_netmask == NULL)
1981  next:				continue;
1982 			cp = addr_data;
1983 			cp2 = ifa->ifa_addr->sa_data;
1984 			cp3 = ifa->ifa_netmask->sa_data;
1985 			cplim = (const char *)ifa->ifa_netmask +
1986 			    ifa->ifa_netmask->sa_len;
1987 			while (cp3 < cplim) {
1988 				if ((*cp++ ^ *cp2++) & *cp3++) {
1989 					/* want to continue for() loop */
1990 					goto next;
1991 				}
1992 			}
1993 			if (ifa_maybe == NULL ||
1994 			    rt_refines(ifa->ifa_netmask,
1995 			               ifa_maybe->ifa_netmask))
1996 				ifa_maybe = ifa;
1997 		}
1998 	}
1999 	return ifa_maybe;
2000 }
2001 
2002 struct ifaddr *
2003 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2004 {
2005 	struct ifaddr *ifa;
2006 	int s;
2007 
2008 	s = pserialize_read_enter();
2009 	ifa = ifa_ifwithnet(addr);
2010 	if (ifa != NULL)
2011 		ifa_acquire(ifa, psref);
2012 	pserialize_read_exit(s);
2013 
2014 	return ifa;
2015 }
2016 
2017 /*
2018  * Find the interface of the addresss.
2019  */
2020 struct ifaddr *
2021 ifa_ifwithladdr(const struct sockaddr *addr)
2022 {
2023 	struct ifaddr *ia;
2024 
2025 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2026 	    (ia = ifa_ifwithnet(addr)))
2027 		return ia;
2028 	return NULL;
2029 }
2030 
2031 struct ifaddr *
2032 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2033 {
2034 	struct ifaddr *ifa;
2035 	int s;
2036 
2037 	s = pserialize_read_enter();
2038 	ifa = ifa_ifwithladdr(addr);
2039 	if (ifa != NULL)
2040 		ifa_acquire(ifa, psref);
2041 	pserialize_read_exit(s);
2042 
2043 	return ifa;
2044 }
2045 
2046 /*
2047  * Find an interface using a specific address family
2048  */
2049 struct ifaddr *
2050 ifa_ifwithaf(int af)
2051 {
2052 	struct ifnet *ifp;
2053 	struct ifaddr *ifa = NULL;
2054 	int s;
2055 
2056 	s = pserialize_read_enter();
2057 	IFNET_READER_FOREACH(ifp) {
2058 		if (if_is_deactivated(ifp))
2059 			continue;
2060 		IFADDR_READER_FOREACH(ifa, ifp) {
2061 			if (ifa->ifa_addr->sa_family == af)
2062 				goto out;
2063 		}
2064 	}
2065 out:
2066 	pserialize_read_exit(s);
2067 	return ifa;
2068 }
2069 
2070 /*
2071  * Find an interface address specific to an interface best matching
2072  * a given address.
2073  */
2074 struct ifaddr *
2075 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2076 {
2077 	struct ifaddr *ifa;
2078 	const char *cp, *cp2, *cp3;
2079 	const char *cplim;
2080 	struct ifaddr *ifa_maybe = 0;
2081 	u_int af = addr->sa_family;
2082 
2083 	if (if_is_deactivated(ifp))
2084 		return NULL;
2085 
2086 	if (af >= AF_MAX)
2087 		return NULL;
2088 
2089 	IFADDR_READER_FOREACH(ifa, ifp) {
2090 		if (ifa->ifa_addr->sa_family != af)
2091 			continue;
2092 		ifa_maybe = ifa;
2093 		if (ifa->ifa_netmask == NULL) {
2094 			if (equal(addr, ifa->ifa_addr) ||
2095 			    (ifa->ifa_dstaddr &&
2096 			     equal(addr, ifa->ifa_dstaddr)))
2097 				return ifa;
2098 			continue;
2099 		}
2100 		cp = addr->sa_data;
2101 		cp2 = ifa->ifa_addr->sa_data;
2102 		cp3 = ifa->ifa_netmask->sa_data;
2103 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2104 		for (; cp3 < cplim; cp3++) {
2105 			if ((*cp++ ^ *cp2++) & *cp3)
2106 				break;
2107 		}
2108 		if (cp3 == cplim)
2109 			return ifa;
2110 	}
2111 	return ifa_maybe;
2112 }
2113 
2114 struct ifaddr *
2115 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2116     struct psref *psref)
2117 {
2118 	struct ifaddr *ifa;
2119 	int s;
2120 
2121 	s = pserialize_read_enter();
2122 	ifa = ifaof_ifpforaddr(addr, ifp);
2123 	if (ifa != NULL)
2124 		ifa_acquire(ifa, psref);
2125 	pserialize_read_exit(s);
2126 
2127 	return ifa;
2128 }
2129 
2130 /*
2131  * Default action when installing a route with a Link Level gateway.
2132  * Lookup an appropriate real ifa to point to.
2133  * This should be moved to /sys/net/link.c eventually.
2134  */
2135 void
2136 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2137 {
2138 	struct ifaddr *ifa;
2139 	const struct sockaddr *dst;
2140 	struct ifnet *ifp;
2141 	struct psref psref;
2142 
2143 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2144 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
2145 		return;
2146 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2147 		rt_replace_ifa(rt, ifa);
2148 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2149 			ifa->ifa_rtrequest(cmd, rt, info);
2150 		ifa_release(ifa, &psref);
2151 	}
2152 }
2153 
2154 /*
2155  * bitmask macros to manage a densely packed link_state change queue.
2156  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2157  * LINK_STATE_UP(2) we need 2 bits for each state change.
2158  * As a state change to store is 0, treat all bits set as an unset item.
2159  */
2160 #define LQ_ITEM_BITS		2
2161 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2162 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2163 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2164 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2165 #define LQ_STORE(q, i, v)						      \
2166 	do {								      \
2167 		(q) &= ~LQ_MASK((i));					      \
2168 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2169 	} while (0 /* CONSTCOND */)
2170 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2171 #define LQ_POP(q, v)							      \
2172 	do {								      \
2173 		(v) = LQ_ITEM((q), 0);					      \
2174 		(q) >>= LQ_ITEM_BITS;					      \
2175 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2176 	} while (0 /* CONSTCOND */)
2177 #define LQ_PUSH(q, v)							      \
2178 	do {								      \
2179 		(q) >>= LQ_ITEM_BITS;					      \
2180 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2181 	} while (0 /* CONSTCOND */)
2182 #define LQ_FIND_UNSET(q, i)						      \
2183 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2184 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2185 			break;						      \
2186 	}
2187 /*
2188  * Handle a change in the interface link state and
2189  * queue notifications.
2190  */
2191 void
2192 if_link_state_change(struct ifnet *ifp, int link_state)
2193 {
2194 	int s, idx;
2195 
2196 	KASSERTMSG(if_is_link_state_changeable(ifp),
2197 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2198 	    ifp->if_xname, ifp->if_extflags);
2199 
2200 	/* Ensure change is to a valid state */
2201 	switch (link_state) {
2202 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2203 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2204 	case LINK_STATE_UP:
2205 		break;
2206 	default:
2207 #ifdef DEBUG
2208 		printf("%s: invalid link state %d\n",
2209 		    ifp->if_xname, link_state);
2210 #endif
2211 		return;
2212 	}
2213 
2214 	s = splnet();
2215 
2216 	/* Find the last unset event in the queue. */
2217 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2218 
2219 	/*
2220 	 * Ensure link_state doesn't match the last event in the queue.
2221 	 * ifp->if_link_state is not checked and set here because
2222 	 * that would present an inconsistent picture to the system.
2223 	 */
2224 	if (idx != 0 &&
2225 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2226 		goto out;
2227 
2228 	/* Handle queue overflow. */
2229 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2230 		uint8_t lost;
2231 
2232 		/*
2233 		 * The DOWN state must be protected from being pushed off
2234 		 * the queue to ensure that userland will always be
2235 		 * in a sane state.
2236 		 * Because DOWN is protected, there is no need to protect
2237 		 * UNKNOWN.
2238 		 * It should be invalid to change from any other state to
2239 		 * UNKNOWN anyway ...
2240 		 */
2241 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2242 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2243 		if (lost == LINK_STATE_DOWN) {
2244 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2245 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2246 		}
2247 		printf("%s: lost link state change %s\n",
2248 		    ifp->if_xname,
2249 		    lost == LINK_STATE_UP ? "UP" :
2250 		    lost == LINK_STATE_DOWN ? "DOWN" :
2251 		    "UNKNOWN");
2252 	} else
2253 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2254 
2255 	softint_schedule(ifp->if_link_si);
2256 
2257 out:
2258 	splx(s);
2259 }
2260 
2261 /*
2262  * Handle interface link state change notifications.
2263  */
2264 void
2265 if_link_state_change_softint(struct ifnet *ifp, int link_state)
2266 {
2267 	struct domain *dp;
2268 	int s = splnet();
2269 
2270 	KASSERT(!cpu_intr_p());
2271 
2272 	/* Ensure the change is still valid. */
2273 	if (ifp->if_link_state == link_state) {
2274 		splx(s);
2275 		return;
2276 	}
2277 
2278 #ifdef DEBUG
2279 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2280 		link_state == LINK_STATE_UP ? "UP" :
2281 		link_state == LINK_STATE_DOWN ? "DOWN" :
2282 		"UNKNOWN",
2283 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2284 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2285 		"UNKNOWN");
2286 #endif
2287 
2288 	/*
2289 	 * When going from UNKNOWN to UP, we need to mark existing
2290 	 * addresses as tentative and restart DAD as we may have
2291 	 * erroneously not found a duplicate.
2292 	 *
2293 	 * This needs to happen before rt_ifmsg to avoid a race where
2294 	 * listeners would have an address and expect it to work right
2295 	 * away.
2296 	 */
2297 	if (link_state == LINK_STATE_UP &&
2298 	    ifp->if_link_state == LINK_STATE_UNKNOWN)
2299 	{
2300 		DOMAIN_FOREACH(dp) {
2301 			if (dp->dom_if_link_state_change != NULL)
2302 				dp->dom_if_link_state_change(ifp,
2303 				    LINK_STATE_DOWN);
2304 		}
2305 	}
2306 
2307 	ifp->if_link_state = link_state;
2308 
2309 	/* Notify that the link state has changed. */
2310 	rt_ifmsg(ifp);
2311 
2312 #if NCARP > 0
2313 	if (ifp->if_carp)
2314 		carp_carpdev_state(ifp);
2315 #endif
2316 
2317 	DOMAIN_FOREACH(dp) {
2318 		if (dp->dom_if_link_state_change != NULL)
2319 			dp->dom_if_link_state_change(ifp, link_state);
2320 	}
2321 	splx(s);
2322 }
2323 
2324 /*
2325  * Process the interface link state change queue.
2326  */
2327 static void
2328 if_link_state_change_si(void *arg)
2329 {
2330 	struct ifnet *ifp = arg;
2331 	int s;
2332 	uint8_t state;
2333 
2334 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2335 	s = splnet();
2336 
2337 	/* Pop a link state change from the queue and process it. */
2338 	LQ_POP(ifp->if_link_queue, state);
2339 	if_link_state_change_softint(ifp, state);
2340 
2341 	/* If there is a link state change to come, schedule it. */
2342 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET)
2343 		softint_schedule(ifp->if_link_si);
2344 
2345 	splx(s);
2346 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2347 }
2348 
2349 /*
2350  * Default action when installing a local route on a point-to-point
2351  * interface.
2352  */
2353 void
2354 p2p_rtrequest(int req, struct rtentry *rt,
2355     __unused const struct rt_addrinfo *info)
2356 {
2357 	struct ifnet *ifp = rt->rt_ifp;
2358 	struct ifaddr *ifa, *lo0ifa;
2359 	int s = pserialize_read_enter();
2360 
2361 	switch (req) {
2362 	case RTM_ADD:
2363 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2364 			break;
2365 
2366 		rt->rt_ifp = lo0ifp;
2367 
2368 		IFADDR_READER_FOREACH(ifa, ifp) {
2369 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2370 				break;
2371 		}
2372 		if (ifa == NULL)
2373 			break;
2374 
2375 		/*
2376 		 * Ensure lo0 has an address of the same family.
2377 		 */
2378 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2379 			if (lo0ifa->ifa_addr->sa_family ==
2380 			    ifa->ifa_addr->sa_family)
2381 				break;
2382 		}
2383 		if (lo0ifa == NULL)
2384 			break;
2385 
2386 		/*
2387 		 * Make sure to set rt->rt_ifa to the interface
2388 		 * address we are using, otherwise we will have trouble
2389 		 * with source address selection.
2390 		 */
2391 		if (ifa != rt->rt_ifa)
2392 			rt_replace_ifa(rt, ifa);
2393 		break;
2394 	case RTM_DELETE:
2395 	default:
2396 		break;
2397 	}
2398 	pserialize_read_exit(s);
2399 }
2400 
2401 /*
2402  * Mark an interface down and notify protocols of
2403  * the transition.
2404  * NOTE: must be called at splsoftnet or equivalent.
2405  */
2406 void
2407 if_down(struct ifnet *ifp)
2408 {
2409 	struct ifaddr *ifa;
2410 	struct domain *dp;
2411 	int s, bound;
2412 	struct psref psref;
2413 
2414 	ifp->if_flags &= ~IFF_UP;
2415 	nanotime(&ifp->if_lastchange);
2416 
2417 	bound = curlwp_bind();
2418 	s = pserialize_read_enter();
2419 	IFADDR_READER_FOREACH(ifa, ifp) {
2420 		ifa_acquire(ifa, &psref);
2421 		pserialize_read_exit(s);
2422 
2423 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2424 
2425 		s = pserialize_read_enter();
2426 		ifa_release(ifa, &psref);
2427 	}
2428 	pserialize_read_exit(s);
2429 	curlwp_bindx(bound);
2430 
2431 	IFQ_PURGE(&ifp->if_snd);
2432 #if NCARP > 0
2433 	if (ifp->if_carp)
2434 		carp_carpdev_state(ifp);
2435 #endif
2436 	rt_ifmsg(ifp);
2437 	DOMAIN_FOREACH(dp) {
2438 		if (dp->dom_if_down)
2439 			dp->dom_if_down(ifp);
2440 	}
2441 }
2442 
2443 /*
2444  * Mark an interface up and notify protocols of
2445  * the transition.
2446  * NOTE: must be called at splsoftnet or equivalent.
2447  */
2448 void
2449 if_up(struct ifnet *ifp)
2450 {
2451 #ifdef notyet
2452 	struct ifaddr *ifa;
2453 #endif
2454 	struct domain *dp;
2455 
2456 	ifp->if_flags |= IFF_UP;
2457 	nanotime(&ifp->if_lastchange);
2458 #ifdef notyet
2459 	/* this has no effect on IP, and will kill all ISO connections XXX */
2460 	IFADDR_READER_FOREACH(ifa, ifp)
2461 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2462 #endif
2463 #if NCARP > 0
2464 	if (ifp->if_carp)
2465 		carp_carpdev_state(ifp);
2466 #endif
2467 	rt_ifmsg(ifp);
2468 	DOMAIN_FOREACH(dp) {
2469 		if (dp->dom_if_up)
2470 			dp->dom_if_up(ifp);
2471 	}
2472 }
2473 
2474 /*
2475  * Handle interface slowtimo timer routine.  Called
2476  * from softclock, we decrement timer (if set) and
2477  * call the appropriate interface routine on expiration.
2478  */
2479 static void
2480 if_slowtimo(void *arg)
2481 {
2482 	void (*slowtimo)(struct ifnet *);
2483 	struct ifnet *ifp = arg;
2484 	int s;
2485 
2486 	slowtimo = ifp->if_slowtimo;
2487 	if (__predict_false(slowtimo == NULL))
2488 		return;
2489 
2490 	s = splnet();
2491 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2492 		(*slowtimo)(ifp);
2493 
2494 	splx(s);
2495 
2496 	if (__predict_true(ifp->if_slowtimo != NULL))
2497 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2498 }
2499 
2500 /*
2501  * Set/clear promiscuous mode on interface ifp based on the truth value
2502  * of pswitch.  The calls are reference counted so that only the first
2503  * "on" request actually has an effect, as does the final "off" request.
2504  * Results are undefined if the "off" and "on" requests are not matched.
2505  */
2506 int
2507 ifpromisc(struct ifnet *ifp, int pswitch)
2508 {
2509 	int pcount, ret;
2510 	short nflags;
2511 
2512 	pcount = ifp->if_pcount;
2513 	if (pswitch) {
2514 		/*
2515 		 * Allow the device to be "placed" into promiscuous
2516 		 * mode even if it is not configured up.  It will
2517 		 * consult IFF_PROMISC when it is brought up.
2518 		 */
2519 		if (ifp->if_pcount++ != 0)
2520 			return 0;
2521 		nflags = ifp->if_flags | IFF_PROMISC;
2522 	} else {
2523 		if (--ifp->if_pcount > 0)
2524 			return 0;
2525 		nflags = ifp->if_flags & ~IFF_PROMISC;
2526 	}
2527 	ret = if_flags_set(ifp, nflags);
2528 	/* Restore interface state if not successful. */
2529 	if (ret != 0) {
2530 		ifp->if_pcount = pcount;
2531 	}
2532 	return ret;
2533 }
2534 
2535 /*
2536  * Map interface name to
2537  * interface structure pointer.
2538  */
2539 struct ifnet *
2540 ifunit(const char *name)
2541 {
2542 	struct ifnet *ifp;
2543 	const char *cp = name;
2544 	u_int unit = 0;
2545 	u_int i;
2546 	int s;
2547 
2548 	/*
2549 	 * If the entire name is a number, treat it as an ifindex.
2550 	 */
2551 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2552 		unit = unit * 10 + (*cp - '0');
2553 	}
2554 
2555 	/*
2556 	 * If the number took all of the name, then it's a valid ifindex.
2557 	 */
2558 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2559 		return if_byindex(unit);
2560 
2561 	ifp = NULL;
2562 	s = pserialize_read_enter();
2563 	IFNET_READER_FOREACH(ifp) {
2564 		if (if_is_deactivated(ifp))
2565 			continue;
2566 	 	if (strcmp(ifp->if_xname, name) == 0)
2567 			goto out;
2568 	}
2569 out:
2570 	pserialize_read_exit(s);
2571 	return ifp;
2572 }
2573 
2574 /*
2575  * Get a reference of an ifnet object by an interface name.
2576  * The returned reference is protected by psref(9). The caller
2577  * must release a returned reference by if_put after use.
2578  */
2579 struct ifnet *
2580 if_get(const char *name, struct psref *psref)
2581 {
2582 	struct ifnet *ifp;
2583 	const char *cp = name;
2584 	u_int unit = 0;
2585 	u_int i;
2586 	int s;
2587 
2588 	/*
2589 	 * If the entire name is a number, treat it as an ifindex.
2590 	 */
2591 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2592 		unit = unit * 10 + (*cp - '0');
2593 	}
2594 
2595 	/*
2596 	 * If the number took all of the name, then it's a valid ifindex.
2597 	 */
2598 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2599 		return if_get_byindex(unit, psref);
2600 
2601 	ifp = NULL;
2602 	s = pserialize_read_enter();
2603 	IFNET_READER_FOREACH(ifp) {
2604 		if (if_is_deactivated(ifp))
2605 			continue;
2606 		if (strcmp(ifp->if_xname, name) == 0) {
2607 			psref_acquire(psref, &ifp->if_psref,
2608 			    ifnet_psref_class);
2609 			goto out;
2610 		}
2611 	}
2612 out:
2613 	pserialize_read_exit(s);
2614 	return ifp;
2615 }
2616 
2617 /*
2618  * Release a reference of an ifnet object given by if_get, if_get_byindex
2619  * or if_get_bylla.
2620  */
2621 void
2622 if_put(const struct ifnet *ifp, struct psref *psref)
2623 {
2624 
2625 	if (ifp == NULL)
2626 		return;
2627 
2628 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2629 }
2630 
2631 ifnet_t *
2632 if_byindex(u_int idx)
2633 {
2634 	ifnet_t *ifp;
2635 
2636 	ifp = (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2637 	if (ifp != NULL && if_is_deactivated(ifp))
2638 		ifp = NULL;
2639 	return ifp;
2640 }
2641 
2642 /*
2643  * Get a reference of an ifnet object by an interface index.
2644  * The returned reference is protected by psref(9). The caller
2645  * must release a returned reference by if_put after use.
2646  */
2647 ifnet_t *
2648 if_get_byindex(u_int idx, struct psref *psref)
2649 {
2650 	ifnet_t *ifp;
2651 	int s;
2652 
2653 	s = pserialize_read_enter();
2654 	ifp = if_byindex(idx);
2655 	if (__predict_true(ifp != NULL))
2656 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2657 	pserialize_read_exit(s);
2658 
2659 	return ifp;
2660 }
2661 
2662 ifnet_t *
2663 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2664 {
2665 	ifnet_t *ifp;
2666 	int s;
2667 
2668 	s = pserialize_read_enter();
2669 	IFNET_READER_FOREACH(ifp) {
2670 		if (if_is_deactivated(ifp))
2671 			continue;
2672 		if (ifp->if_addrlen != lla_len)
2673 			continue;
2674 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2675 			psref_acquire(psref, &ifp->if_psref,
2676 			    ifnet_psref_class);
2677 			break;
2678 		}
2679 	}
2680 	pserialize_read_exit(s);
2681 
2682 	return ifp;
2683 }
2684 
2685 /*
2686  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2687  * for example using pserialize or the ifp is already held or some other
2688  * object is held which guarantes the ifp to not be freed indirectly.
2689  */
2690 void
2691 if_acquire(struct ifnet *ifp, struct psref *psref)
2692 {
2693 
2694 	KASSERT(ifp->if_index != 0);
2695 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2696 }
2697 
2698 bool
2699 if_held(struct ifnet *ifp)
2700 {
2701 
2702 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2703 }
2704 
2705 
2706 /* common */
2707 int
2708 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2709 {
2710 	int s;
2711 	struct ifreq *ifr;
2712 	struct ifcapreq *ifcr;
2713 	struct ifdatareq *ifdr;
2714 
2715 	switch (cmd) {
2716 	case SIOCSIFCAP:
2717 		ifcr = data;
2718 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2719 			return EINVAL;
2720 
2721 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2722 			return 0;
2723 
2724 		ifp->if_capenable = ifcr->ifcr_capenable;
2725 
2726 		/* Pre-compute the checksum flags mask. */
2727 		ifp->if_csum_flags_tx = 0;
2728 		ifp->if_csum_flags_rx = 0;
2729 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2730 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2731 		}
2732 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2733 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2734 		}
2735 
2736 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2737 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2738 		}
2739 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2740 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2741 		}
2742 
2743 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2744 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2745 		}
2746 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2747 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2748 		}
2749 
2750 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2751 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2752 		}
2753 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2754 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2755 		}
2756 
2757 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2758 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2759 		}
2760 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2761 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2762 		}
2763 		if (ifp->if_flags & IFF_UP)
2764 			return ENETRESET;
2765 		return 0;
2766 	case SIOCSIFFLAGS:
2767 		ifr = data;
2768 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2769 			s = splsoftnet();
2770 			if_down(ifp);
2771 			splx(s);
2772 		}
2773 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2774 			s = splsoftnet();
2775 			if_up(ifp);
2776 			splx(s);
2777 		}
2778 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2779 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2780 		break;
2781 	case SIOCGIFFLAGS:
2782 		ifr = data;
2783 		ifr->ifr_flags = ifp->if_flags;
2784 		break;
2785 
2786 	case SIOCGIFMETRIC:
2787 		ifr = data;
2788 		ifr->ifr_metric = ifp->if_metric;
2789 		break;
2790 
2791 	case SIOCGIFMTU:
2792 		ifr = data;
2793 		ifr->ifr_mtu = ifp->if_mtu;
2794 		break;
2795 
2796 	case SIOCGIFDLT:
2797 		ifr = data;
2798 		ifr->ifr_dlt = ifp->if_dlt;
2799 		break;
2800 
2801 	case SIOCGIFCAP:
2802 		ifcr = data;
2803 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2804 		ifcr->ifcr_capenable = ifp->if_capenable;
2805 		break;
2806 
2807 	case SIOCSIFMETRIC:
2808 		ifr = data;
2809 		ifp->if_metric = ifr->ifr_metric;
2810 		break;
2811 
2812 	case SIOCGIFDATA:
2813 		ifdr = data;
2814 		ifdr->ifdr_data = ifp->if_data;
2815 		break;
2816 
2817 	case SIOCGIFINDEX:
2818 		ifr = data;
2819 		ifr->ifr_index = ifp->if_index;
2820 		break;
2821 
2822 	case SIOCZIFDATA:
2823 		ifdr = data;
2824 		ifdr->ifdr_data = ifp->if_data;
2825 		/*
2826 		 * Assumes that the volatile counters that can be
2827 		 * zero'ed are at the end of if_data.
2828 		 */
2829 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
2830 		    offsetof(struct if_data, ifi_ipackets));
2831 		/*
2832 		 * The memset() clears to the bottm of if_data. In the area,
2833 		 * if_lastchange is included. Please be careful if new entry
2834 		 * will be added into if_data or rewite this.
2835 		 *
2836 		 * And also, update if_lastchnage.
2837 		 */
2838 		getnanotime(&ifp->if_lastchange);
2839 		break;
2840 	case SIOCSIFMTU:
2841 		ifr = data;
2842 		if (ifp->if_mtu == ifr->ifr_mtu)
2843 			break;
2844 		ifp->if_mtu = ifr->ifr_mtu;
2845 		/*
2846 		 * If the link MTU changed, do network layer specific procedure.
2847 		 */
2848 #ifdef INET6
2849 		if (in6_present)
2850 			nd6_setmtu(ifp);
2851 #endif
2852 		return ENETRESET;
2853 	default:
2854 		return ENOTTY;
2855 	}
2856 	return 0;
2857 }
2858 
2859 int
2860 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
2861 {
2862 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
2863 	struct ifaddr *ifa;
2864 	const struct sockaddr *any, *sa;
2865 	union {
2866 		struct sockaddr sa;
2867 		struct sockaddr_storage ss;
2868 	} u, v;
2869 	int s, error = 0;
2870 
2871 	switch (cmd) {
2872 	case SIOCSIFADDRPREF:
2873 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
2874 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
2875 		    NULL) != 0)
2876 			return EPERM;
2877 	case SIOCGIFADDRPREF:
2878 		break;
2879 	default:
2880 		return EOPNOTSUPP;
2881 	}
2882 
2883 	/* sanity checks */
2884 	if (data == NULL || ifp == NULL) {
2885 		panic("invalid argument to %s", __func__);
2886 		/*NOTREACHED*/
2887 	}
2888 
2889 	/* address must be specified on ADD and DELETE */
2890 	sa = sstocsa(&ifap->ifap_addr);
2891 	if (sa->sa_family != sofamily(so))
2892 		return EINVAL;
2893 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
2894 		return EINVAL;
2895 
2896 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
2897 
2898 	s = pserialize_read_enter();
2899 	IFADDR_READER_FOREACH(ifa, ifp) {
2900 		if (ifa->ifa_addr->sa_family != sa->sa_family)
2901 			continue;
2902 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
2903 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
2904 			break;
2905 	}
2906 	if (ifa == NULL) {
2907 		error = EADDRNOTAVAIL;
2908 		goto out;
2909 	}
2910 
2911 	switch (cmd) {
2912 	case SIOCSIFADDRPREF:
2913 		ifa->ifa_preference = ifap->ifap_preference;
2914 		goto out;
2915 	case SIOCGIFADDRPREF:
2916 		/* fill in the if_laddrreq structure */
2917 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
2918 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
2919 		ifap->ifap_preference = ifa->ifa_preference;
2920 		goto out;
2921 	default:
2922 		error = EOPNOTSUPP;
2923 	}
2924 out:
2925 	pserialize_read_exit(s);
2926 	return error;
2927 }
2928 
2929 /*
2930  * Interface ioctls.
2931  */
2932 static int
2933 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
2934 {
2935 	struct ifnet *ifp;
2936 	struct ifreq *ifr;
2937 	int error = 0;
2938 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
2939 	u_long ocmd = cmd;
2940 #endif
2941 	short oif_flags;
2942 #ifdef COMPAT_OIFREQ
2943 	struct ifreq ifrb;
2944 	struct oifreq *oifr = NULL;
2945 #endif
2946 	int r;
2947 	struct psref psref;
2948 	int bound;
2949 
2950 	switch (cmd) {
2951 #ifdef COMPAT_OIFREQ
2952 	case OSIOCGIFCONF:
2953 	case OOSIOCGIFCONF:
2954 		return compat_ifconf(cmd, data);
2955 #endif
2956 #ifdef COMPAT_OIFDATA
2957 	case OSIOCGIFDATA:
2958 	case OSIOCZIFDATA:
2959 		return compat_ifdatareq(l, cmd, data);
2960 #endif
2961 	case SIOCGIFCONF:
2962 		return ifconf(cmd, data);
2963 	case SIOCINITIFADDR:
2964 		return EPERM;
2965 	}
2966 
2967 #ifdef COMPAT_OIFREQ
2968 	cmd = (*vec_compat_cvtcmd)(cmd);
2969 	if (cmd != ocmd) {
2970 		oifr = data;
2971 		data = ifr = &ifrb;
2972 		ifreqo2n(oifr, ifr);
2973 	} else
2974 #endif
2975 		ifr = data;
2976 
2977 	switch (cmd) {
2978 	case SIOCIFCREATE:
2979 	case SIOCIFDESTROY:
2980 		bound = curlwp_bind();
2981 		if (l != NULL) {
2982 			ifp = if_get(ifr->ifr_name, &psref);
2983 			error = kauth_authorize_network(l->l_cred,
2984 			    KAUTH_NETWORK_INTERFACE,
2985 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2986 			    (void *)cmd, NULL);
2987 			if (ifp != NULL)
2988 				if_put(ifp, &psref);
2989 			if (error != 0) {
2990 				curlwp_bindx(bound);
2991 				return error;
2992 			}
2993 		}
2994 		mutex_enter(&if_clone_mtx);
2995 		r = (cmd == SIOCIFCREATE) ?
2996 			if_clone_create(ifr->ifr_name) :
2997 			if_clone_destroy(ifr->ifr_name);
2998 		mutex_exit(&if_clone_mtx);
2999 		curlwp_bindx(bound);
3000 		return r;
3001 
3002 	case SIOCIFGCLONERS:
3003 		{
3004 			struct if_clonereq *req = (struct if_clonereq *)data;
3005 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3006 			    &req->ifcr_total);
3007 		}
3008 	}
3009 
3010 	bound = curlwp_bind();
3011 	ifp = if_get(ifr->ifr_name, &psref);
3012 	if (ifp == NULL) {
3013 		curlwp_bindx(bound);
3014 		return ENXIO;
3015 	}
3016 
3017 	switch (cmd) {
3018 	case SIOCALIFADDR:
3019 	case SIOCDLIFADDR:
3020 	case SIOCSIFADDRPREF:
3021 	case SIOCSIFFLAGS:
3022 	case SIOCSIFCAP:
3023 	case SIOCSIFMETRIC:
3024 	case SIOCZIFDATA:
3025 	case SIOCSIFMTU:
3026 	case SIOCSIFPHYADDR:
3027 	case SIOCDIFPHYADDR:
3028 #ifdef INET6
3029 	case SIOCSIFPHYADDR_IN6:
3030 #endif
3031 	case SIOCSLIFPHYADDR:
3032 	case SIOCADDMULTI:
3033 	case SIOCDELMULTI:
3034 	case SIOCSIFMEDIA:
3035 	case SIOCSDRVSPEC:
3036 	case SIOCG80211:
3037 	case SIOCS80211:
3038 	case SIOCS80211NWID:
3039 	case SIOCS80211NWKEY:
3040 	case SIOCS80211POWER:
3041 	case SIOCS80211BSSID:
3042 	case SIOCS80211CHANNEL:
3043 	case SIOCSLINKSTR:
3044 		if (l != NULL) {
3045 			error = kauth_authorize_network(l->l_cred,
3046 			    KAUTH_NETWORK_INTERFACE,
3047 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3048 			    (void *)cmd, NULL);
3049 			if (error != 0)
3050 				goto out;
3051 		}
3052 	}
3053 
3054 	oif_flags = ifp->if_flags;
3055 
3056 	mutex_enter(ifp->if_ioctl_lock);
3057 
3058 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3059 	if (error != ENOTTY)
3060 		;
3061 	else if (so->so_proto == NULL)
3062 		error = EOPNOTSUPP;
3063 	else {
3064 #ifdef COMPAT_OSOCK
3065 		if (vec_compat_ifioctl != NULL)
3066 			error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l);
3067 		else
3068 #endif
3069 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3070 			    cmd, data, ifp);
3071 	}
3072 
3073 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3074 		if ((ifp->if_flags & IFF_UP) != 0) {
3075 			int s = splsoftnet();
3076 			if_up(ifp);
3077 			splx(s);
3078 		}
3079 	}
3080 #ifdef COMPAT_OIFREQ
3081 	if (cmd != ocmd)
3082 		ifreqn2o(oifr, ifr);
3083 #endif
3084 
3085 	mutex_exit(ifp->if_ioctl_lock);
3086 out:
3087 	if_put(ifp, &psref);
3088 	curlwp_bindx(bound);
3089 	return error;
3090 }
3091 
3092 /*
3093  * Return interface configuration
3094  * of system.  List may be used
3095  * in later ioctl's (above) to get
3096  * other information.
3097  *
3098  * Each record is a struct ifreq.  Before the addition of
3099  * sockaddr_storage, the API rule was that sockaddr flavors that did
3100  * not fit would extend beyond the struct ifreq, with the next struct
3101  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3102  * union in struct ifreq includes struct sockaddr_storage, every kind
3103  * of sockaddr must fit.  Thus, there are no longer any overlength
3104  * records.
3105  *
3106  * Records are added to the user buffer if they fit, and ifc_len is
3107  * adjusted to the length that was written.  Thus, the user is only
3108  * assured of getting the complete list if ifc_len on return is at
3109  * least sizeof(struct ifreq) less than it was on entry.
3110  *
3111  * If the user buffer pointer is NULL, this routine copies no data and
3112  * returns the amount of space that would be needed.
3113  *
3114  * Invariants:
3115  * ifrp points to the next part of the user's buffer to be used.  If
3116  * ifrp != NULL, space holds the number of bytes remaining that we may
3117  * write at ifrp.  Otherwise, space holds the number of bytes that
3118  * would have been written had there been adequate space.
3119  */
3120 /*ARGSUSED*/
3121 static int
3122 ifconf(u_long cmd, void *data)
3123 {
3124 	struct ifconf *ifc = (struct ifconf *)data;
3125 	struct ifnet *ifp;
3126 	struct ifaddr *ifa;
3127 	struct ifreq ifr, *ifrp = NULL;
3128 	int space = 0, error = 0;
3129 	const int sz = (int)sizeof(struct ifreq);
3130 	const bool docopy = ifc->ifc_req != NULL;
3131 	int s;
3132 	int bound;
3133 	struct psref psref;
3134 
3135 	if (docopy) {
3136 		space = ifc->ifc_len;
3137 		ifrp = ifc->ifc_req;
3138 	}
3139 
3140 	bound = curlwp_bind();
3141 	s = pserialize_read_enter();
3142 	IFNET_READER_FOREACH(ifp) {
3143 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3144 		pserialize_read_exit(s);
3145 
3146 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3147 		    sizeof(ifr.ifr_name));
3148 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3149 			error = ENAMETOOLONG;
3150 			goto release_exit;
3151 		}
3152 		if (IFADDR_READER_EMPTY(ifp)) {
3153 			/* Interface with no addresses - send zero sockaddr. */
3154 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3155 			if (!docopy) {
3156 				space += sz;
3157 				continue;
3158 			}
3159 			if (space >= sz) {
3160 				error = copyout(&ifr, ifrp, sz);
3161 				if (error != 0)
3162 					goto release_exit;
3163 				ifrp++;
3164 				space -= sz;
3165 			}
3166 		}
3167 
3168 		IFADDR_READER_FOREACH(ifa, ifp) {
3169 			struct sockaddr *sa = ifa->ifa_addr;
3170 			/* all sockaddrs must fit in sockaddr_storage */
3171 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3172 
3173 			if (!docopy) {
3174 				space += sz;
3175 				continue;
3176 			}
3177 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3178 			if (space >= sz) {
3179 				error = copyout(&ifr, ifrp, sz);
3180 				if (error != 0)
3181 					goto release_exit;
3182 				ifrp++; space -= sz;
3183 			}
3184 		}
3185 
3186 		s = pserialize_read_enter();
3187 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3188 	}
3189 	pserialize_read_exit(s);
3190 	curlwp_bindx(bound);
3191 
3192 	if (docopy) {
3193 		KASSERT(0 <= space && space <= ifc->ifc_len);
3194 		ifc->ifc_len -= space;
3195 	} else {
3196 		KASSERT(space >= 0);
3197 		ifc->ifc_len = space;
3198 	}
3199 	return (0);
3200 
3201 release_exit:
3202 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3203 	curlwp_bindx(bound);
3204 	return error;
3205 }
3206 
3207 int
3208 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3209 {
3210 	uint8_t len;
3211 #ifdef COMPAT_OIFREQ
3212 	struct ifreq ifrb;
3213 	struct oifreq *oifr = NULL;
3214 	u_long ocmd = cmd;
3215 	cmd = (*vec_compat_cvtcmd)(cmd);
3216 	if (cmd != ocmd) {
3217 		oifr = (struct oifreq *)(void *)ifr;
3218 		ifr = &ifrb;
3219 		ifreqo2n(oifr, ifr);
3220 		len = sizeof(oifr->ifr_addr);
3221 	} else
3222 #endif
3223 		len = sizeof(ifr->ifr_ifru.ifru_space);
3224 
3225 	if (len < sa->sa_len)
3226 		return EFBIG;
3227 
3228 	memset(&ifr->ifr_addr, 0, len);
3229 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3230 
3231 #ifdef COMPAT_OIFREQ
3232 	if (cmd != ocmd)
3233 		ifreqn2o(oifr, ifr);
3234 #endif
3235 	return 0;
3236 }
3237 
3238 /*
3239  * wrapper function for the drivers which doesn't have if_transmit().
3240  */
3241 static int
3242 if_transmit(struct ifnet *ifp, struct mbuf *m)
3243 {
3244 	int s, error;
3245 	size_t pktlen = m->m_pkthdr.len;
3246 	bool mcast = (m->m_flags & M_MCAST) != 0;
3247 
3248 	s = splnet();
3249 
3250 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3251 	if (error != 0) {
3252 		/* mbuf is already freed */
3253 		goto out;
3254 	}
3255 
3256 	ifp->if_obytes += pktlen;
3257 	if (mcast)
3258 		ifp->if_omcasts++;
3259 
3260 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3261 		if_start_lock(ifp);
3262 out:
3263 	splx(s);
3264 
3265 	return error;
3266 }
3267 
3268 int
3269 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3270 {
3271 	int error;
3272 
3273 #ifdef ALTQ
3274 	KERNEL_LOCK(1, NULL);
3275 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3276 		error = if_transmit(ifp, m);
3277 		KERNEL_UNLOCK_ONE(NULL);
3278 	} else {
3279 		KERNEL_UNLOCK_ONE(NULL);
3280 		error = (*ifp->if_transmit)(ifp, m);
3281 		/* mbuf is alredy freed */
3282 	}
3283 #else /* !ALTQ */
3284 	error = (*ifp->if_transmit)(ifp, m);
3285 	/* mbuf is alredy freed */
3286 #endif /* !ALTQ */
3287 
3288 	return error;
3289 }
3290 
3291 /*
3292  * Queue message on interface, and start output if interface
3293  * not yet active.
3294  */
3295 int
3296 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3297 {
3298 
3299 	return if_transmit_lock(ifp, m);
3300 }
3301 
3302 /*
3303  * Queue message on interface, possibly using a second fast queue
3304  */
3305 int
3306 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3307 {
3308 	int error = 0;
3309 
3310 	if (ifq != NULL
3311 #ifdef ALTQ
3312 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3313 #endif
3314 	    ) {
3315 		if (IF_QFULL(ifq)) {
3316 			IF_DROP(&ifp->if_snd);
3317 			m_freem(m);
3318 			if (error == 0)
3319 				error = ENOBUFS;
3320 		} else
3321 			IF_ENQUEUE(ifq, m);
3322 	} else
3323 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3324 	if (error != 0) {
3325 		++ifp->if_oerrors;
3326 		return error;
3327 	}
3328 	return 0;
3329 }
3330 
3331 int
3332 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3333 {
3334 	int rc;
3335 
3336 	if (ifp->if_initaddr != NULL)
3337 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3338 	else if (src ||
3339 		/* FIXME: may not hold if_ioctl_lock */
3340 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3341 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3342 
3343 	return rc;
3344 }
3345 
3346 int
3347 if_do_dad(struct ifnet *ifp)
3348 {
3349 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3350 		return 0;
3351 
3352 	switch (ifp->if_type) {
3353 	case IFT_FAITH:
3354 		/*
3355 		 * These interfaces do not have the IFF_LOOPBACK flag,
3356 		 * but loop packets back.  We do not have to do DAD on such
3357 		 * interfaces.  We should even omit it, because loop-backed
3358 		 * responses would confuse the DAD procedure.
3359 		 */
3360 		return 0;
3361 	default:
3362 		/*
3363 		 * Our DAD routine requires the interface up and running.
3364 		 * However, some interfaces can be up before the RUNNING
3365 		 * status.  Additionaly, users may try to assign addresses
3366 		 * before the interface becomes up (or running).
3367 		 * We simply skip DAD in such a case as a work around.
3368 		 * XXX: we should rather mark "tentative" on such addresses,
3369 		 * and do DAD after the interface becomes ready.
3370 		 */
3371 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
3372 		    (IFF_UP|IFF_RUNNING))
3373 			return 0;
3374 
3375 		return 1;
3376 	}
3377 }
3378 
3379 int
3380 if_flags_set(ifnet_t *ifp, const short flags)
3381 {
3382 	int rc;
3383 
3384 	if (ifp->if_setflags != NULL)
3385 		rc = (*ifp->if_setflags)(ifp, flags);
3386 	else {
3387 		short cantflags, chgdflags;
3388 		struct ifreq ifr;
3389 
3390 		chgdflags = ifp->if_flags ^ flags;
3391 		cantflags = chgdflags & IFF_CANTCHANGE;
3392 
3393 		if (cantflags != 0)
3394 			ifp->if_flags ^= cantflags;
3395 
3396                 /* Traditionally, we do not call if_ioctl after
3397                  * setting/clearing only IFF_PROMISC if the interface
3398                  * isn't IFF_UP.  Uphold that tradition.
3399 		 */
3400 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3401 			return 0;
3402 
3403 		memset(&ifr, 0, sizeof(ifr));
3404 
3405 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3406 		/* FIXME: may not hold if_ioctl_lock */
3407 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3408 
3409 		if (rc != 0 && cantflags != 0)
3410 			ifp->if_flags ^= cantflags;
3411 	}
3412 
3413 	return rc;
3414 }
3415 
3416 int
3417 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3418 {
3419 	int rc;
3420 	struct ifreq ifr;
3421 
3422 	if (ifp->if_mcastop != NULL)
3423 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3424 	else {
3425 		ifreq_setaddr(cmd, &ifr, sa);
3426 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3427 	}
3428 
3429 	return rc;
3430 }
3431 
3432 static void
3433 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3434     struct ifaltq *ifq)
3435 {
3436 	const struct sysctlnode *cnode, *rnode;
3437 
3438 	if (sysctl_createv(clog, 0, NULL, &rnode,
3439 		       CTLFLAG_PERMANENT,
3440 		       CTLTYPE_NODE, "interfaces",
3441 		       SYSCTL_DESCR("Per-interface controls"),
3442 		       NULL, 0, NULL, 0,
3443 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3444 		goto bad;
3445 
3446 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3447 		       CTLFLAG_PERMANENT,
3448 		       CTLTYPE_NODE, ifname,
3449 		       SYSCTL_DESCR("Interface controls"),
3450 		       NULL, 0, NULL, 0,
3451 		       CTL_CREATE, CTL_EOL) != 0)
3452 		goto bad;
3453 
3454 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3455 		       CTLFLAG_PERMANENT,
3456 		       CTLTYPE_NODE, "sndq",
3457 		       SYSCTL_DESCR("Interface output queue controls"),
3458 		       NULL, 0, NULL, 0,
3459 		       CTL_CREATE, CTL_EOL) != 0)
3460 		goto bad;
3461 
3462 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3463 		       CTLFLAG_PERMANENT,
3464 		       CTLTYPE_INT, "len",
3465 		       SYSCTL_DESCR("Current output queue length"),
3466 		       NULL, 0, &ifq->ifq_len, 0,
3467 		       CTL_CREATE, CTL_EOL) != 0)
3468 		goto bad;
3469 
3470 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3471 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3472 		       CTLTYPE_INT, "maxlen",
3473 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3474 		       NULL, 0, &ifq->ifq_maxlen, 0,
3475 		       CTL_CREATE, CTL_EOL) != 0)
3476 		goto bad;
3477 
3478 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3479 		       CTLFLAG_PERMANENT,
3480 		       CTLTYPE_INT, "drops",
3481 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3482 		       NULL, 0, &ifq->ifq_drops, 0,
3483 		       CTL_CREATE, CTL_EOL) != 0)
3484 		goto bad;
3485 
3486 	return;
3487 bad:
3488 	printf("%s: could not attach sysctl nodes\n", ifname);
3489 	return;
3490 }
3491 
3492 #if defined(INET) || defined(INET6)
3493 
3494 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3495 	static int							\
3496 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3497 	{								\
3498 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3499 	}
3500 
3501 #if defined(INET)
3502 static int
3503 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3504 {
3505 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3506 }
3507 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3508 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3509 #endif
3510 
3511 #if defined(INET6)
3512 static int
3513 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3514 {
3515 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3516 }
3517 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3518 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3519 #endif
3520 
3521 static void
3522 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3523 {
3524 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3525 	const char *pfname = NULL, *ipname = NULL;
3526 	int ipn = 0, qid = 0;
3527 
3528 	switch (pf) {
3529 #if defined(INET)
3530 	case PF_INET:
3531 		len_func = sysctl_net_ip_pktq_items;
3532 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3533 		drops_func = sysctl_net_ip_pktq_drops;
3534 		pfname = "inet", ipn = IPPROTO_IP;
3535 		ipname = "ip", qid = IPCTL_IFQ;
3536 		break;
3537 #endif
3538 #if defined(INET6)
3539 	case PF_INET6:
3540 		len_func = sysctl_net_ip6_pktq_items;
3541 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3542 		drops_func = sysctl_net_ip6_pktq_drops;
3543 		pfname = "inet6", ipn = IPPROTO_IPV6;
3544 		ipname = "ip6", qid = IPV6CTL_IFQ;
3545 		break;
3546 #endif
3547 	default:
3548 		KASSERT(false);
3549 	}
3550 
3551 	sysctl_createv(clog, 0, NULL, NULL,
3552 		       CTLFLAG_PERMANENT,
3553 		       CTLTYPE_NODE, pfname, NULL,
3554 		       NULL, 0, NULL, 0,
3555 		       CTL_NET, pf, CTL_EOL);
3556 	sysctl_createv(clog, 0, NULL, NULL,
3557 		       CTLFLAG_PERMANENT,
3558 		       CTLTYPE_NODE, ipname, NULL,
3559 		       NULL, 0, NULL, 0,
3560 		       CTL_NET, pf, ipn, CTL_EOL);
3561 	sysctl_createv(clog, 0, NULL, NULL,
3562 		       CTLFLAG_PERMANENT,
3563 		       CTLTYPE_NODE, "ifq",
3564 		       SYSCTL_DESCR("Protocol input queue controls"),
3565 		       NULL, 0, NULL, 0,
3566 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3567 
3568 	sysctl_createv(clog, 0, NULL, NULL,
3569 		       CTLFLAG_PERMANENT,
3570 		       CTLTYPE_INT, "len",
3571 		       SYSCTL_DESCR("Current input queue length"),
3572 		       len_func, 0, NULL, 0,
3573 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3574 	sysctl_createv(clog, 0, NULL, NULL,
3575 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3576 		       CTLTYPE_INT, "maxlen",
3577 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3578 		       maxlen_func, 0, NULL, 0,
3579 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3580 	sysctl_createv(clog, 0, NULL, NULL,
3581 		       CTLFLAG_PERMANENT,
3582 		       CTLTYPE_INT, "drops",
3583 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3584 		       drops_func, 0, NULL, 0,
3585 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3586 }
3587 #endif /* INET || INET6 */
3588 
3589 static int
3590 if_sdl_sysctl(SYSCTLFN_ARGS)
3591 {
3592 	struct ifnet *ifp;
3593 	const struct sockaddr_dl *sdl;
3594 	struct psref psref;
3595 	int error = 0;
3596 	int bound;
3597 
3598 	if (namelen != 1)
3599 		return EINVAL;
3600 
3601 	bound = curlwp_bind();
3602 	ifp = if_get_byindex(name[0], &psref);
3603 	if (ifp == NULL) {
3604 		error = ENODEV;
3605 		goto out0;
3606 	}
3607 
3608 	sdl = ifp->if_sadl;
3609 	if (sdl == NULL) {
3610 		*oldlenp = 0;
3611 		goto out1;
3612 	}
3613 
3614 	if (oldp == NULL) {
3615 		*oldlenp = sdl->sdl_alen;
3616 		goto out1;
3617 	}
3618 
3619 	if (*oldlenp >= sdl->sdl_alen)
3620 		*oldlenp = sdl->sdl_alen;
3621 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3622 out1:
3623 	if_put(ifp, &psref);
3624 out0:
3625 	curlwp_bindx(bound);
3626 	return error;
3627 }
3628 
3629 static void
3630 if_sysctl_setup(struct sysctllog **clog)
3631 {
3632 	const struct sysctlnode *rnode = NULL;
3633 
3634 	sysctl_createv(clog, 0, NULL, &rnode,
3635 		       CTLFLAG_PERMANENT,
3636 		       CTLTYPE_NODE, "sdl",
3637 		       SYSCTL_DESCR("Get active link-layer address"),
3638 		       if_sdl_sysctl, 0, NULL, 0,
3639 		       CTL_NET, CTL_CREATE, CTL_EOL);
3640 
3641 #if defined(INET)
3642 	sysctl_net_pktq_setup(NULL, PF_INET);
3643 #endif
3644 #ifdef INET6
3645 	if (in6_present)
3646 		sysctl_net_pktq_setup(NULL, PF_INET6);
3647 #endif
3648 }
3649