xref: /netbsd-src/sys/net/if.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: if.c,v 1.430 2018/07/09 14:54:01 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by William Studenmund and Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the project nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 /*
62  * Copyright (c) 1980, 1986, 1993
63  *	The Regents of the University of California.  All rights reserved.
64  *
65  * Redistribution and use in source and binary forms, with or without
66  * modification, are permitted provided that the following conditions
67  * are met:
68  * 1. Redistributions of source code must retain the above copyright
69  *    notice, this list of conditions and the following disclaimer.
70  * 2. Redistributions in binary form must reproduce the above copyright
71  *    notice, this list of conditions and the following disclaimer in the
72  *    documentation and/or other materials provided with the distribution.
73  * 3. Neither the name of the University nor the names of its contributors
74  *    may be used to endorse or promote products derived from this software
75  *    without specific prior written permission.
76  *
77  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
78  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
79  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
80  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
81  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
82  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
83  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
84  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
85  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
86  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
87  * SUCH DAMAGE.
88  *
89  *	@(#)if.c	8.5 (Berkeley) 1/9/95
90  */
91 
92 #include <sys/cdefs.h>
93 __KERNEL_RCSID(0, "$NetBSD: if.c,v 1.430 2018/07/09 14:54:01 christos Exp $");
94 
95 #if defined(_KERNEL_OPT)
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 #include "opt_atalk.h"
99 #include "opt_natm.h"
100 #include "opt_wlan.h"
101 #include "opt_net_mpsafe.h"
102 #include "opt_mrouting.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 *, int);
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 static void if_up_locked(struct ifnet *);
213 static void _if_down(struct ifnet *);
214 static void if_down_deactivated(struct ifnet *);
215 
216 struct if_percpuq {
217 	struct ifnet	*ipq_ifp;
218 	void		*ipq_si;
219 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
220 };
221 
222 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
223 
224 static void if_percpuq_drops(void *, void *, struct cpu_info *);
225 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
226 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
227     struct if_percpuq *);
228 
229 struct if_deferred_start {
230 	struct ifnet	*ids_ifp;
231 	void		(*ids_if_start)(struct ifnet *);
232 	void		*ids_si;
233 };
234 
235 static void if_deferred_start_softint(void *);
236 static void if_deferred_start_common(struct ifnet *);
237 static void if_deferred_start_destroy(struct ifnet *);
238 
239 #if defined(INET) || defined(INET6)
240 static void sysctl_net_pktq_setup(struct sysctllog **, int);
241 #endif
242 
243 static void if_sysctl_setup(struct sysctllog **);
244 
245 /* Compatibility vector functions */
246 u_long (*vec_compat_cvtcmd)(u_long) = NULL;
247 int (*vec_compat_ifioctl)(struct socket *, u_long, u_long, void *,
248 	struct lwp *) = NULL;
249 int (*vec_compat_ifconf)(struct lwp *, u_long, void *) = (void *)enosys;
250 int (*vec_compat_ifdatareq)(struct lwp *, u_long, void *) = (void *)enosys;
251 
252 static int
253 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
254     void *arg0, void *arg1, void *arg2, void *arg3)
255 {
256 	int result;
257 	enum kauth_network_req req;
258 
259 	result = KAUTH_RESULT_DEFER;
260 	req = (enum kauth_network_req)arg1;
261 
262 	if (action != KAUTH_NETWORK_INTERFACE)
263 		return result;
264 
265 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
266 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
267 		result = KAUTH_RESULT_ALLOW;
268 
269 	return result;
270 }
271 
272 /*
273  * Network interface utility routines.
274  *
275  * Routines with ifa_ifwith* names take sockaddr *'s as
276  * parameters.
277  */
278 void
279 ifinit(void)
280 {
281 
282 #if (defined(INET) || defined(INET6))
283 	encapinit();
284 #endif
285 
286 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
287 	    if_listener_cb, NULL);
288 
289 	/* interfaces are available, inform socket code */
290 	ifioctl = doifioctl;
291 }
292 
293 /*
294  * XXX Initialization before configure().
295  * XXX hack to get pfil_add_hook working in autoconf.
296  */
297 void
298 ifinit1(void)
299 {
300 
301 #ifdef NET_MPSAFE
302 	printf("NET_MPSAFE enabled\n");
303 #endif
304 
305 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
306 
307 	TAILQ_INIT(&ifnet_list);
308 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
309 	ifnet_psz = pserialize_create();
310 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
311 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
312 	PSLIST_INIT(&ifnet_pslist);
313 
314 	if_indexlim = 8;
315 
316 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
317 	KASSERT(if_pfil != NULL);
318 
319 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
320 	etherinit();
321 #endif
322 }
323 
324 /* XXX must be after domaininit() */
325 void
326 ifinit_post(void)
327 {
328 
329 	if_sysctl_setup(NULL);
330 }
331 
332 ifnet_t *
333 if_alloc(u_char type)
334 {
335 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
336 }
337 
338 void
339 if_free(ifnet_t *ifp)
340 {
341 	kmem_free(ifp, sizeof(ifnet_t));
342 }
343 
344 void
345 if_initname(struct ifnet *ifp, const char *name, int unit)
346 {
347 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
348 	    "%s%d", name, unit);
349 }
350 
351 /*
352  * Null routines used while an interface is going away.  These routines
353  * just return an error.
354  */
355 
356 int
357 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
358     const struct sockaddr *so, const struct rtentry *rt)
359 {
360 
361 	return ENXIO;
362 }
363 
364 void
365 if_nullinput(struct ifnet *ifp, struct mbuf *m)
366 {
367 
368 	/* Nothing. */
369 }
370 
371 void
372 if_nullstart(struct ifnet *ifp)
373 {
374 
375 	/* Nothing. */
376 }
377 
378 int
379 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
380 {
381 
382 	m_freem(m);
383 	return ENXIO;
384 }
385 
386 int
387 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
388 {
389 
390 	return ENXIO;
391 }
392 
393 int
394 if_nullinit(struct ifnet *ifp)
395 {
396 
397 	return ENXIO;
398 }
399 
400 void
401 if_nullstop(struct ifnet *ifp, int disable)
402 {
403 
404 	/* Nothing. */
405 }
406 
407 void
408 if_nullslowtimo(struct ifnet *ifp)
409 {
410 
411 	/* Nothing. */
412 }
413 
414 void
415 if_nulldrain(struct ifnet *ifp)
416 {
417 
418 	/* Nothing. */
419 }
420 
421 void
422 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
423 {
424 	struct ifaddr *ifa;
425 	struct sockaddr_dl *sdl;
426 
427 	ifp->if_addrlen = addrlen;
428 	if_alloc_sadl(ifp);
429 	ifa = ifp->if_dl;
430 	sdl = satosdl(ifa->ifa_addr);
431 
432 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
433 	if (factory) {
434 		KASSERT(ifp->if_hwdl == NULL);
435 		ifp->if_hwdl = ifp->if_dl;
436 		ifaref(ifp->if_hwdl);
437 	}
438 	/* TBD routing socket */
439 }
440 
441 struct ifaddr *
442 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
443 {
444 	unsigned socksize, ifasize;
445 	int addrlen, namelen;
446 	struct sockaddr_dl *mask, *sdl;
447 	struct ifaddr *ifa;
448 
449 	namelen = strlen(ifp->if_xname);
450 	addrlen = ifp->if_addrlen;
451 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
452 	ifasize = sizeof(*ifa) + 2 * socksize;
453 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
454 
455 	sdl = (struct sockaddr_dl *)(ifa + 1);
456 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
457 
458 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
459 	    ifp->if_xname, namelen, NULL, addrlen);
460 	mask->sdl_family = AF_LINK;
461 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
462 	memset(&mask->sdl_data[0], 0xff, namelen);
463 	ifa->ifa_rtrequest = link_rtrequest;
464 	ifa->ifa_addr = (struct sockaddr *)sdl;
465 	ifa->ifa_netmask = (struct sockaddr *)mask;
466 	ifa_psref_init(ifa);
467 
468 	*sdlp = sdl;
469 
470 	return ifa;
471 }
472 
473 static void
474 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
475 {
476 	const struct sockaddr_dl *sdl;
477 
478 	ifp->if_dl = ifa;
479 	ifaref(ifa);
480 	sdl = satosdl(ifa->ifa_addr);
481 	ifp->if_sadl = sdl;
482 }
483 
484 /*
485  * Allocate the link level name for the specified interface.  This
486  * is an attachment helper.  It must be called after ifp->if_addrlen
487  * is initialized, which may not be the case when if_attach() is
488  * called.
489  */
490 void
491 if_alloc_sadl(struct ifnet *ifp)
492 {
493 	struct ifaddr *ifa;
494 	const struct sockaddr_dl *sdl;
495 
496 	/*
497 	 * If the interface already has a link name, release it
498 	 * now.  This is useful for interfaces that can change
499 	 * link types, and thus switch link names often.
500 	 */
501 	if (ifp->if_sadl != NULL)
502 		if_free_sadl(ifp, 0);
503 
504 	ifa = if_dl_create(ifp, &sdl);
505 
506 	ifa_insert(ifp, ifa);
507 	if_sadl_setrefs(ifp, ifa);
508 }
509 
510 static void
511 if_deactivate_sadl(struct ifnet *ifp)
512 {
513 	struct ifaddr *ifa;
514 
515 	KASSERT(ifp->if_dl != NULL);
516 
517 	ifa = ifp->if_dl;
518 
519 	ifp->if_sadl = NULL;
520 
521 	ifp->if_dl = NULL;
522 	ifafree(ifa);
523 }
524 
525 static void
526 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa)
527 {
528 	struct ifaddr *old;
529 
530 	KASSERT(ifp->if_dl != NULL);
531 
532 	old = ifp->if_dl;
533 
534 	ifaref(ifa);
535 	/* XXX Update if_dl and if_sadl atomically */
536 	ifp->if_dl = ifa;
537 	ifp->if_sadl = satosdl(ifa->ifa_addr);
538 
539 	ifafree(old);
540 }
541 
542 void
543 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
544     const struct sockaddr_dl *sdl)
545 {
546 	int s, ss;
547 	struct ifaddr *ifa;
548 	int bound = curlwp_bind();
549 
550 	KASSERT(ifa_held(ifa0));
551 
552 	s = splsoftnet();
553 
554 	if_replace_sadl(ifp, ifa0);
555 
556 	ss = pserialize_read_enter();
557 	IFADDR_READER_FOREACH(ifa, ifp) {
558 		struct psref psref;
559 		ifa_acquire(ifa, &psref);
560 		pserialize_read_exit(ss);
561 
562 		rtinit(ifa, RTM_LLINFO_UPD, 0);
563 
564 		ss = pserialize_read_enter();
565 		ifa_release(ifa, &psref);
566 	}
567 	pserialize_read_exit(ss);
568 
569 	splx(s);
570 	curlwp_bindx(bound);
571 }
572 
573 /*
574  * Free the link level name for the specified interface.  This is
575  * a detach helper.  This is called from if_detach().
576  */
577 static void
578 if_free_sadl(struct ifnet *ifp, int factory)
579 {
580 	struct ifaddr *ifa;
581 	int s;
582 
583 	if (factory && ifp->if_hwdl != NULL) {
584 		ifa = ifp->if_hwdl;
585 		ifp->if_hwdl = NULL;
586 		ifafree(ifa);
587 	}
588 
589 	ifa = ifp->if_dl;
590 	if (ifa == NULL) {
591 		KASSERT(ifp->if_sadl == NULL);
592 		return;
593 	}
594 
595 	KASSERT(ifp->if_sadl != NULL);
596 
597 	s = splsoftnet();
598 	KASSERT(ifa->ifa_addr->sa_family == AF_LINK);
599 	ifa_remove(ifp, ifa);
600 	if_deactivate_sadl(ifp);
601 	splx(s);
602 }
603 
604 static void
605 if_getindex(ifnet_t *ifp)
606 {
607 	bool hitlimit = false;
608 
609 	ifp->if_index_gen = index_gen++;
610 
611 	ifp->if_index = if_index;
612 	if (ifindex2ifnet == NULL) {
613 		if_index++;
614 		goto skip;
615 	}
616 	while (if_byindex(ifp->if_index)) {
617 		/*
618 		 * If we hit USHRT_MAX, we skip back to 0 since
619 		 * there are a number of places where the value
620 		 * of if_index or if_index itself is compared
621 		 * to or stored in an unsigned short.  By
622 		 * jumping back, we won't botch those assignments
623 		 * or comparisons.
624 		 */
625 		if (++if_index == 0) {
626 			if_index = 1;
627 		} else if (if_index == USHRT_MAX) {
628 			/*
629 			 * However, if we have to jump back to
630 			 * zero *twice* without finding an empty
631 			 * slot in ifindex2ifnet[], then there
632 			 * there are too many (>65535) interfaces.
633 			 */
634 			if (hitlimit) {
635 				panic("too many interfaces");
636 			}
637 			hitlimit = true;
638 			if_index = 1;
639 		}
640 		ifp->if_index = if_index;
641 	}
642 skip:
643 	/*
644 	 * ifindex2ifnet is indexed by if_index. Since if_index will
645 	 * grow dynamically, it should grow too.
646 	 */
647 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
648 		size_t m, n, oldlim;
649 		void *q;
650 
651 		oldlim = if_indexlim;
652 		while (ifp->if_index >= if_indexlim)
653 			if_indexlim <<= 1;
654 
655 		/* grow ifindex2ifnet */
656 		m = oldlim * sizeof(struct ifnet *);
657 		n = if_indexlim * sizeof(struct ifnet *);
658 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
659 		if (ifindex2ifnet != NULL) {
660 			memcpy(q, ifindex2ifnet, m);
661 			free(ifindex2ifnet, M_IFADDR);
662 		}
663 		ifindex2ifnet = (struct ifnet **)q;
664 	}
665 	ifindex2ifnet[ifp->if_index] = ifp;
666 }
667 
668 /*
669  * Initialize an interface and assign an index for it.
670  *
671  * It must be called prior to a device specific attach routine
672  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
673  * and be followed by if_register:
674  *
675  *     if_initialize(ifp);
676  *     ether_ifattach(ifp, enaddr);
677  *     if_register(ifp);
678  */
679 int
680 if_initialize(ifnet_t *ifp)
681 {
682 	int rv = 0;
683 
684 	KASSERT(if_indexlim > 0);
685 	TAILQ_INIT(&ifp->if_addrlist);
686 
687 	/*
688 	 * Link level name is allocated later by a separate call to
689 	 * if_alloc_sadl().
690 	 */
691 
692 	if (ifp->if_snd.ifq_maxlen == 0)
693 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
694 
695 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
696 
697 	ifp->if_link_state = LINK_STATE_UNKNOWN;
698 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
699 
700 	ifp->if_capenable = 0;
701 	ifp->if_csum_flags_tx = 0;
702 	ifp->if_csum_flags_rx = 0;
703 
704 #ifdef ALTQ
705 	ifp->if_snd.altq_type = 0;
706 	ifp->if_snd.altq_disc = NULL;
707 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
708 	ifp->if_snd.altq_tbr  = NULL;
709 	ifp->if_snd.altq_ifp  = ifp;
710 #endif
711 
712 	IFQ_LOCK_INIT(&ifp->if_snd);
713 
714 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
715 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
716 
717 	IF_AFDATA_LOCK_INIT(ifp);
718 
719 	if (if_is_link_state_changeable(ifp)) {
720 		u_int flags = SOFTINT_NET;
721 		flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
722 		ifp->if_link_si = softint_establish(flags,
723 		    if_link_state_change_si, ifp);
724 		if (ifp->if_link_si == NULL) {
725 			rv = ENOMEM;
726 			goto fail;
727 		}
728 	}
729 
730 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
731 	PSLIST_INIT(&ifp->if_addr_pslist);
732 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
733 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
734 	LIST_INIT(&ifp->if_multiaddrs);
735 
736 	IFNET_GLOBAL_LOCK();
737 	if_getindex(ifp);
738 	IFNET_GLOBAL_UNLOCK();
739 
740 	return 0;
741 
742 fail:
743 	IF_AFDATA_LOCK_DESTROY(ifp);
744 
745 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
746 	(void)pfil_head_destroy(ifp->if_pfil);
747 
748 	IFQ_LOCK_DESTROY(&ifp->if_snd);
749 
750 	return rv;
751 }
752 
753 /*
754  * Register an interface to the list of "active" interfaces.
755  */
756 void
757 if_register(ifnet_t *ifp)
758 {
759 	/*
760 	 * If the driver has not supplied its own if_ioctl, then
761 	 * supply the default.
762 	 */
763 	if (ifp->if_ioctl == NULL)
764 		ifp->if_ioctl = ifioctl_common;
765 
766 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
767 
768 	if (!STAILQ_EMPTY(&domains))
769 		if_attachdomain1(ifp);
770 
771 	/* Announce the interface. */
772 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
773 
774 	if (ifp->if_slowtimo != NULL) {
775 		ifp->if_slowtimo_ch =
776 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
777 		callout_init(ifp->if_slowtimo_ch, 0);
778 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
779 		if_slowtimo(ifp);
780 	}
781 
782 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
783 		ifp->if_transmit = if_transmit;
784 
785 	IFNET_GLOBAL_LOCK();
786 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
787 	IFNET_WRITER_INSERT_TAIL(ifp);
788 	IFNET_GLOBAL_UNLOCK();
789 }
790 
791 /*
792  * The if_percpuq framework
793  *
794  * It allows network device drivers to execute the network stack
795  * in softint (so called softint-based if_input). It utilizes
796  * softint and percpu ifqueue. It doesn't distribute any packets
797  * between CPUs, unlike pktqueue(9).
798  *
799  * Currently we support two options for device drivers to apply the framework:
800  * - Use it implicitly with less changes
801  *   - If you use if_attach in driver's _attach function and if_input in
802  *     driver's Rx interrupt handler, a packet is queued and a softint handles
803  *     the packet implicitly
804  * - Use it explicitly in each driver (recommended)
805  *   - You can use if_percpuq_* directly in your driver
806  *   - In this case, you need to allocate struct if_percpuq in driver's softc
807  *   - See wm(4) as a reference implementation
808  */
809 
810 static void
811 if_percpuq_softint(void *arg)
812 {
813 	struct if_percpuq *ipq = arg;
814 	struct ifnet *ifp = ipq->ipq_ifp;
815 	struct mbuf *m;
816 
817 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
818 		ifp->if_ipackets++;
819 		bpf_mtap(ifp, m, BPF_D_IN);
820 
821 		ifp->_if_input(ifp, m);
822 	}
823 }
824 
825 static void
826 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
827 {
828 	struct ifqueue *const ifq = p;
829 
830 	memset(ifq, 0, sizeof(*ifq));
831 	ifq->ifq_maxlen = IFQ_MAXLEN;
832 }
833 
834 struct if_percpuq *
835 if_percpuq_create(struct ifnet *ifp)
836 {
837 	struct if_percpuq *ipq;
838 	u_int flags = SOFTINT_NET;
839 
840 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
841 
842 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
843 	ipq->ipq_ifp = ifp;
844 	ipq->ipq_si = softint_establish(flags, if_percpuq_softint, ipq);
845 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
846 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
847 
848 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
849 
850 	return ipq;
851 }
852 
853 static struct mbuf *
854 if_percpuq_dequeue(struct if_percpuq *ipq)
855 {
856 	struct mbuf *m;
857 	struct ifqueue *ifq;
858 	int s;
859 
860 	s = splnet();
861 	ifq = percpu_getref(ipq->ipq_ifqs);
862 	IF_DEQUEUE(ifq, m);
863 	percpu_putref(ipq->ipq_ifqs);
864 	splx(s);
865 
866 	return m;
867 }
868 
869 static void
870 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
871 {
872 	struct ifqueue *const ifq = p;
873 
874 	IF_PURGE(ifq);
875 }
876 
877 void
878 if_percpuq_destroy(struct if_percpuq *ipq)
879 {
880 
881 	/* if_detach may already destroy it */
882 	if (ipq == NULL)
883 		return;
884 
885 	softint_disestablish(ipq->ipq_si);
886 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
887 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
888 	kmem_free(ipq, sizeof(*ipq));
889 }
890 
891 void
892 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
893 {
894 	struct ifqueue *ifq;
895 	int s;
896 
897 	KASSERT(ipq != NULL);
898 
899 	s = splnet();
900 	ifq = percpu_getref(ipq->ipq_ifqs);
901 	if (IF_QFULL(ifq)) {
902 		IF_DROP(ifq);
903 		percpu_putref(ipq->ipq_ifqs);
904 		m_freem(m);
905 		goto out;
906 	}
907 	IF_ENQUEUE(ifq, m);
908 	percpu_putref(ipq->ipq_ifqs);
909 
910 	softint_schedule(ipq->ipq_si);
911 out:
912 	splx(s);
913 }
914 
915 static void
916 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
917 {
918 	struct ifqueue *const ifq = p;
919 	int *sum = arg;
920 
921 	*sum += ifq->ifq_drops;
922 }
923 
924 static int
925 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
926 {
927 	struct sysctlnode node;
928 	struct if_percpuq *ipq;
929 	int sum = 0;
930 	int error;
931 
932 	node = *rnode;
933 	ipq = node.sysctl_data;
934 
935 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
936 
937 	node.sysctl_data = &sum;
938 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
939 	if (error != 0 || newp == NULL)
940 		return error;
941 
942 	return 0;
943 }
944 
945 static void
946 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
947     struct if_percpuq *ipq)
948 {
949 	const struct sysctlnode *cnode, *rnode;
950 
951 	if (sysctl_createv(clog, 0, NULL, &rnode,
952 		       CTLFLAG_PERMANENT,
953 		       CTLTYPE_NODE, "interfaces",
954 		       SYSCTL_DESCR("Per-interface controls"),
955 		       NULL, 0, NULL, 0,
956 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
957 		goto bad;
958 
959 	if (sysctl_createv(clog, 0, &rnode, &rnode,
960 		       CTLFLAG_PERMANENT,
961 		       CTLTYPE_NODE, ifname,
962 		       SYSCTL_DESCR("Interface controls"),
963 		       NULL, 0, NULL, 0,
964 		       CTL_CREATE, CTL_EOL) != 0)
965 		goto bad;
966 
967 	if (sysctl_createv(clog, 0, &rnode, &rnode,
968 		       CTLFLAG_PERMANENT,
969 		       CTLTYPE_NODE, "rcvq",
970 		       SYSCTL_DESCR("Interface input queue controls"),
971 		       NULL, 0, NULL, 0,
972 		       CTL_CREATE, CTL_EOL) != 0)
973 		goto bad;
974 
975 #ifdef NOTYET
976 	/* XXX Should show each per-CPU queue length? */
977 	if (sysctl_createv(clog, 0, &rnode, &rnode,
978 		       CTLFLAG_PERMANENT,
979 		       CTLTYPE_INT, "len",
980 		       SYSCTL_DESCR("Current input queue length"),
981 		       sysctl_percpuq_len, 0, NULL, 0,
982 		       CTL_CREATE, CTL_EOL) != 0)
983 		goto bad;
984 
985 	if (sysctl_createv(clog, 0, &rnode, &cnode,
986 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
987 		       CTLTYPE_INT, "maxlen",
988 		       SYSCTL_DESCR("Maximum allowed input queue length"),
989 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
990 		       CTL_CREATE, CTL_EOL) != 0)
991 		goto bad;
992 #endif
993 
994 	if (sysctl_createv(clog, 0, &rnode, &cnode,
995 		       CTLFLAG_PERMANENT,
996 		       CTLTYPE_INT, "drops",
997 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
998 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
999 		       CTL_CREATE, CTL_EOL) != 0)
1000 		goto bad;
1001 
1002 	return;
1003 bad:
1004 	printf("%s: could not attach sysctl nodes\n", ifname);
1005 	return;
1006 }
1007 
1008 /*
1009  * The deferred if_start framework
1010  *
1011  * The common APIs to defer if_start to softint when if_start is requested
1012  * from a device driver running in hardware interrupt context.
1013  */
1014 /*
1015  * Call ifp->if_start (or equivalent) in a dedicated softint for
1016  * deferred if_start.
1017  */
1018 static void
1019 if_deferred_start_softint(void *arg)
1020 {
1021 	struct if_deferred_start *ids = arg;
1022 	struct ifnet *ifp = ids->ids_ifp;
1023 
1024 	ids->ids_if_start(ifp);
1025 }
1026 
1027 /*
1028  * The default callback function for deferred if_start.
1029  */
1030 static void
1031 if_deferred_start_common(struct ifnet *ifp)
1032 {
1033 	int s;
1034 
1035 	s = splnet();
1036 	if_start_lock(ifp);
1037 	splx(s);
1038 }
1039 
1040 static inline bool
1041 if_snd_is_used(struct ifnet *ifp)
1042 {
1043 
1044 	return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit ||
1045 	    ALTQ_IS_ENABLED(&ifp->if_snd);
1046 }
1047 
1048 /*
1049  * Schedule deferred if_start.
1050  */
1051 void
1052 if_schedule_deferred_start(struct ifnet *ifp)
1053 {
1054 
1055 	KASSERT(ifp->if_deferred_start != NULL);
1056 
1057 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1058 		return;
1059 
1060 	softint_schedule(ifp->if_deferred_start->ids_si);
1061 }
1062 
1063 /*
1064  * Create an instance of deferred if_start. A driver should call the function
1065  * only if the driver needs deferred if_start. Drivers can setup their own
1066  * deferred if_start function via 2nd argument.
1067  */
1068 void
1069 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1070 {
1071 	struct if_deferred_start *ids;
1072 	u_int flags = SOFTINT_NET;
1073 
1074 	flags |= if_is_mpsafe(ifp) ? SOFTINT_MPSAFE : 0;
1075 
1076 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1077 	ids->ids_ifp = ifp;
1078 	ids->ids_si = softint_establish(flags, if_deferred_start_softint, ids);
1079 	if (func != NULL)
1080 		ids->ids_if_start = func;
1081 	else
1082 		ids->ids_if_start = if_deferred_start_common;
1083 
1084 	ifp->if_deferred_start = ids;
1085 }
1086 
1087 static void
1088 if_deferred_start_destroy(struct ifnet *ifp)
1089 {
1090 
1091 	if (ifp->if_deferred_start == NULL)
1092 		return;
1093 
1094 	softint_disestablish(ifp->if_deferred_start->ids_si);
1095 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1096 	ifp->if_deferred_start = NULL;
1097 }
1098 
1099 /*
1100  * The common interface input routine that is called by device drivers,
1101  * which should be used only when the driver's rx handler already runs
1102  * in softint.
1103  */
1104 void
1105 if_input(struct ifnet *ifp, struct mbuf *m)
1106 {
1107 
1108 	KASSERT(ifp->if_percpuq == NULL);
1109 	KASSERT(!cpu_intr_p());
1110 
1111 	ifp->if_ipackets++;
1112 	bpf_mtap(ifp, m, BPF_D_IN);
1113 
1114 	ifp->_if_input(ifp, m);
1115 }
1116 
1117 /*
1118  * DEPRECATED. Use if_initialize and if_register instead.
1119  * See the above comment of if_initialize.
1120  *
1121  * Note that it implicitly enables if_percpuq to make drivers easy to
1122  * migrate softint-based if_input without much changes. If you don't
1123  * want to enable it, use if_initialize instead.
1124  */
1125 int
1126 if_attach(ifnet_t *ifp)
1127 {
1128 	int rv;
1129 
1130 	rv = if_initialize(ifp);
1131 	if (rv != 0)
1132 		return rv;
1133 
1134 	ifp->if_percpuq = if_percpuq_create(ifp);
1135 	if_register(ifp);
1136 
1137 	return 0;
1138 }
1139 
1140 void
1141 if_attachdomain(void)
1142 {
1143 	struct ifnet *ifp;
1144 	int s;
1145 	int bound = curlwp_bind();
1146 
1147 	s = pserialize_read_enter();
1148 	IFNET_READER_FOREACH(ifp) {
1149 		struct psref psref;
1150 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1151 		pserialize_read_exit(s);
1152 		if_attachdomain1(ifp);
1153 		s = pserialize_read_enter();
1154 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1155 	}
1156 	pserialize_read_exit(s);
1157 	curlwp_bindx(bound);
1158 }
1159 
1160 static void
1161 if_attachdomain1(struct ifnet *ifp)
1162 {
1163 	struct domain *dp;
1164 	int s;
1165 
1166 	s = splsoftnet();
1167 
1168 	/* address family dependent data region */
1169 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1170 	DOMAIN_FOREACH(dp) {
1171 		if (dp->dom_ifattach != NULL)
1172 			ifp->if_afdata[dp->dom_family] =
1173 			    (*dp->dom_ifattach)(ifp);
1174 	}
1175 
1176 	splx(s);
1177 }
1178 
1179 /*
1180  * Deactivate an interface.  This points all of the procedure
1181  * handles at error stubs.  May be called from interrupt context.
1182  */
1183 void
1184 if_deactivate(struct ifnet *ifp)
1185 {
1186 	int s;
1187 
1188 	s = splsoftnet();
1189 
1190 	ifp->if_output	 = if_nulloutput;
1191 	ifp->_if_input	 = if_nullinput;
1192 	ifp->if_start	 = if_nullstart;
1193 	ifp->if_transmit = if_nulltransmit;
1194 	ifp->if_ioctl	 = if_nullioctl;
1195 	ifp->if_init	 = if_nullinit;
1196 	ifp->if_stop	 = if_nullstop;
1197 	ifp->if_slowtimo = if_nullslowtimo;
1198 	ifp->if_drain	 = if_nulldrain;
1199 
1200 	/* No more packets may be enqueued. */
1201 	ifp->if_snd.ifq_maxlen = 0;
1202 
1203 	splx(s);
1204 }
1205 
1206 bool
1207 if_is_deactivated(const struct ifnet *ifp)
1208 {
1209 
1210 	return ifp->if_output == if_nulloutput;
1211 }
1212 
1213 void
1214 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1215 {
1216 	struct ifaddr *ifa, *nifa;
1217 	int s;
1218 
1219 	s = pserialize_read_enter();
1220 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1221 		nifa = IFADDR_READER_NEXT(ifa);
1222 		if (ifa->ifa_addr->sa_family != family)
1223 			continue;
1224 		pserialize_read_exit(s);
1225 
1226 		(*purgeaddr)(ifa);
1227 
1228 		s = pserialize_read_enter();
1229 	}
1230 	pserialize_read_exit(s);
1231 }
1232 
1233 #ifdef IFAREF_DEBUG
1234 static struct ifaddr **ifa_list;
1235 static int ifa_list_size;
1236 
1237 /* Depends on only one if_attach runs at once */
1238 static void
1239 if_build_ifa_list(struct ifnet *ifp)
1240 {
1241 	struct ifaddr *ifa;
1242 	int i;
1243 
1244 	KASSERT(ifa_list == NULL);
1245 	KASSERT(ifa_list_size == 0);
1246 
1247 	IFADDR_READER_FOREACH(ifa, ifp)
1248 		ifa_list_size++;
1249 
1250 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1251 	i = 0;
1252 	IFADDR_READER_FOREACH(ifa, ifp) {
1253 		ifa_list[i++] = ifa;
1254 		ifaref(ifa);
1255 	}
1256 }
1257 
1258 static void
1259 if_check_and_free_ifa_list(struct ifnet *ifp)
1260 {
1261 	int i;
1262 	struct ifaddr *ifa;
1263 
1264 	if (ifa_list == NULL)
1265 		return;
1266 
1267 	for (i = 0; i < ifa_list_size; i++) {
1268 		char buf[64];
1269 
1270 		ifa = ifa_list[i];
1271 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1272 		if (ifa->ifa_refcnt > 1) {
1273 			log(LOG_WARNING,
1274 			    "ifa(%s) still referenced (refcnt=%d)\n",
1275 			    buf, ifa->ifa_refcnt - 1);
1276 		} else
1277 			log(LOG_DEBUG,
1278 			    "ifa(%s) not referenced (refcnt=%d)\n",
1279 			    buf, ifa->ifa_refcnt - 1);
1280 		ifafree(ifa);
1281 	}
1282 
1283 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1284 	ifa_list = NULL;
1285 	ifa_list_size = 0;
1286 }
1287 #endif
1288 
1289 /*
1290  * Detach an interface from the list of "active" interfaces,
1291  * freeing any resources as we go along.
1292  *
1293  * NOTE: This routine must be called with a valid thread context,
1294  * as it may block.
1295  */
1296 void
1297 if_detach(struct ifnet *ifp)
1298 {
1299 	struct socket so;
1300 	struct ifaddr *ifa;
1301 #ifdef IFAREF_DEBUG
1302 	struct ifaddr *last_ifa = NULL;
1303 #endif
1304 	struct domain *dp;
1305 	const struct protosw *pr;
1306 	int s, i, family, purged;
1307 	uint64_t xc;
1308 
1309 #ifdef IFAREF_DEBUG
1310 	if_build_ifa_list(ifp);
1311 #endif
1312 	/*
1313 	 * XXX It's kind of lame that we have to have the
1314 	 * XXX socket structure...
1315 	 */
1316 	memset(&so, 0, sizeof(so));
1317 
1318 	s = splnet();
1319 
1320 	sysctl_teardown(&ifp->if_sysctl_log);
1321 	IFNET_LOCK(ifp);
1322 	if_deactivate(ifp);
1323 	IFNET_UNLOCK(ifp);
1324 
1325 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1326 		ifp->if_slowtimo = NULL;
1327 		callout_halt(ifp->if_slowtimo_ch, NULL);
1328 		callout_destroy(ifp->if_slowtimo_ch);
1329 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1330 	}
1331 	if_deferred_start_destroy(ifp);
1332 
1333 	/*
1334 	 * Do an if_down() to give protocols a chance to do something.
1335 	 */
1336 	if_down_deactivated(ifp);
1337 
1338 #ifdef ALTQ
1339 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1340 		altq_disable(&ifp->if_snd);
1341 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1342 		altq_detach(&ifp->if_snd);
1343 #endif
1344 
1345 #if NCARP > 0
1346 	/* Remove the interface from any carp group it is a part of.  */
1347 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1348 		carp_ifdetach(ifp);
1349 #endif
1350 
1351 	/*
1352 	 * Rip all the addresses off the interface.  This should make
1353 	 * all of the routes go away.
1354 	 *
1355 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1356 	 * from the list, including our "cursor", ifa.  For safety,
1357 	 * and to honor the TAILQ abstraction, I just restart the
1358 	 * loop after each removal.  Note that the loop will exit
1359 	 * when all of the remaining ifaddrs belong to the AF_LINK
1360 	 * family.  I am counting on the historical fact that at
1361 	 * least one pr_usrreq in each address domain removes at
1362 	 * least one ifaddr.
1363 	 */
1364 again:
1365 	/*
1366 	 * At this point, no other one tries to remove ifa in the list,
1367 	 * so we don't need to take a lock or psref.  Avoid using
1368 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1369 	 * violations of pserialize.
1370 	 */
1371 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1372 		family = ifa->ifa_addr->sa_family;
1373 #ifdef IFAREF_DEBUG
1374 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1375 		    ifa, family, ifa->ifa_refcnt);
1376 		if (last_ifa != NULL && ifa == last_ifa)
1377 			panic("if_detach: loop detected");
1378 		last_ifa = ifa;
1379 #endif
1380 		if (family == AF_LINK)
1381 			continue;
1382 		dp = pffinddomain(family);
1383 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1384 		/*
1385 		 * XXX These PURGEIF calls are redundant with the
1386 		 * purge-all-families calls below, but are left in for
1387 		 * now both to make a smaller change, and to avoid
1388 		 * unplanned interactions with clearing of
1389 		 * ifp->if_addrlist.
1390 		 */
1391 		purged = 0;
1392 		for (pr = dp->dom_protosw;
1393 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1394 			so.so_proto = pr;
1395 			if (pr->pr_usrreqs) {
1396 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1397 				purged = 1;
1398 			}
1399 		}
1400 		if (purged == 0) {
1401 			/*
1402 			 * XXX What's really the best thing to do
1403 			 * XXX here?  --thorpej@NetBSD.org
1404 			 */
1405 			printf("if_detach: WARNING: AF %d not purged\n",
1406 			    family);
1407 			ifa_remove(ifp, ifa);
1408 		}
1409 		goto again;
1410 	}
1411 
1412 	if_free_sadl(ifp, 1);
1413 
1414 restart:
1415 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1416 		family = ifa->ifa_addr->sa_family;
1417 		KASSERT(family == AF_LINK);
1418 		ifa_remove(ifp, ifa);
1419 		goto restart;
1420 	}
1421 
1422 	/* Delete stray routes from the routing table. */
1423 	for (i = 0; i <= AF_MAX; i++)
1424 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1425 
1426 	DOMAIN_FOREACH(dp) {
1427 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1428 		{
1429 			void *p = ifp->if_afdata[dp->dom_family];
1430 			if (p) {
1431 				ifp->if_afdata[dp->dom_family] = NULL;
1432 				(*dp->dom_ifdetach)(ifp, p);
1433 			}
1434 		}
1435 
1436 		/*
1437 		 * One would expect multicast memberships (INET and
1438 		 * INET6) on UDP sockets to be purged by the PURGEIF
1439 		 * calls above, but if all addresses were removed from
1440 		 * the interface prior to destruction, the calls will
1441 		 * not be made (e.g. ppp, for which pppd(8) generally
1442 		 * removes addresses before destroying the interface).
1443 		 * Because there is no invariant that multicast
1444 		 * memberships only exist for interfaces with IPv4
1445 		 * addresses, we must call PURGEIF regardless of
1446 		 * addresses.  (Protocols which might store ifnet
1447 		 * pointers are marked with PR_PURGEIF.)
1448 		 */
1449 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1450 			so.so_proto = pr;
1451 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1452 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1453 		}
1454 	}
1455 
1456 	/* Wait for all readers to drain before freeing.  */
1457 	IFNET_GLOBAL_LOCK();
1458 	ifindex2ifnet[ifp->if_index] = NULL;
1459 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1460 	IFNET_WRITER_REMOVE(ifp);
1461 	pserialize_perform(ifnet_psz);
1462 	IFNET_GLOBAL_UNLOCK();
1463 
1464 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1465 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1466 
1467 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1468 	(void)pfil_head_destroy(ifp->if_pfil);
1469 
1470 	/* Announce that the interface is gone. */
1471 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1472 
1473 	IF_AFDATA_LOCK_DESTROY(ifp);
1474 
1475 	if (if_is_link_state_changeable(ifp)) {
1476 		softint_disestablish(ifp->if_link_si);
1477 		ifp->if_link_si = NULL;
1478 	}
1479 
1480 	/*
1481 	 * remove packets that came from ifp, from software interrupt queues.
1482 	 */
1483 	DOMAIN_FOREACH(dp) {
1484 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1485 			struct ifqueue *iq = dp->dom_ifqueues[i];
1486 			if (iq == NULL)
1487 				break;
1488 			dp->dom_ifqueues[i] = NULL;
1489 			if_detach_queues(ifp, iq);
1490 		}
1491 	}
1492 
1493 	/*
1494 	 * IP queues have to be processed separately: net-queue barrier
1495 	 * ensures that the packets are dequeued while a cross-call will
1496 	 * ensure that the interrupts have completed. FIXME: not quite..
1497 	 */
1498 #ifdef INET
1499 	pktq_barrier(ip_pktq);
1500 #endif
1501 #ifdef INET6
1502 	if (in6_present)
1503 		pktq_barrier(ip6_pktq);
1504 #endif
1505 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1506 	xc_wait(xc);
1507 
1508 	if (ifp->if_percpuq != NULL) {
1509 		if_percpuq_destroy(ifp->if_percpuq);
1510 		ifp->if_percpuq = NULL;
1511 	}
1512 
1513 	mutex_obj_free(ifp->if_ioctl_lock);
1514 	ifp->if_ioctl_lock = NULL;
1515 	mutex_obj_free(ifp->if_snd.ifq_lock);
1516 
1517 	splx(s);
1518 
1519 #ifdef IFAREF_DEBUG
1520 	if_check_and_free_ifa_list(ifp);
1521 #endif
1522 }
1523 
1524 static void
1525 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1526 {
1527 	struct mbuf *m, *prev, *next;
1528 
1529 	prev = NULL;
1530 	for (m = q->ifq_head; m != NULL; m = next) {
1531 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1532 
1533 		next = m->m_nextpkt;
1534 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1535 			prev = m;
1536 			continue;
1537 		}
1538 
1539 		if (prev != NULL)
1540 			prev->m_nextpkt = m->m_nextpkt;
1541 		else
1542 			q->ifq_head = m->m_nextpkt;
1543 		if (q->ifq_tail == m)
1544 			q->ifq_tail = prev;
1545 		q->ifq_len--;
1546 
1547 		m->m_nextpkt = NULL;
1548 		m_freem(m);
1549 		IF_DROP(q);
1550 	}
1551 }
1552 
1553 /*
1554  * Callback for a radix tree walk to delete all references to an
1555  * ifnet.
1556  */
1557 static int
1558 if_delroute_matcher(struct rtentry *rt, void *v)
1559 {
1560 	struct ifnet *ifp = (struct ifnet *)v;
1561 
1562 	if (rt->rt_ifp == ifp)
1563 		return 1;
1564 	else
1565 		return 0;
1566 }
1567 
1568 /*
1569  * Create a clone network interface.
1570  */
1571 static int
1572 if_clone_create(const char *name)
1573 {
1574 	struct if_clone *ifc;
1575 	int unit;
1576 	struct ifnet *ifp;
1577 	struct psref psref;
1578 
1579 	KASSERT(mutex_owned(&if_clone_mtx));
1580 
1581 	ifc = if_clone_lookup(name, &unit);
1582 	if (ifc == NULL)
1583 		return EINVAL;
1584 
1585 	ifp = if_get(name, &psref);
1586 	if (ifp != NULL) {
1587 		if_put(ifp, &psref);
1588 		return EEXIST;
1589 	}
1590 
1591 	return (*ifc->ifc_create)(ifc, unit);
1592 }
1593 
1594 /*
1595  * Destroy a clone network interface.
1596  */
1597 static int
1598 if_clone_destroy(const char *name)
1599 {
1600 	struct if_clone *ifc;
1601 	struct ifnet *ifp;
1602 	struct psref psref;
1603 
1604 	KASSERT(mutex_owned(&if_clone_mtx));
1605 
1606 	ifc = if_clone_lookup(name, NULL);
1607 	if (ifc == NULL)
1608 		return EINVAL;
1609 
1610 	if (ifc->ifc_destroy == NULL)
1611 		return EOPNOTSUPP;
1612 
1613 	ifp = if_get(name, &psref);
1614 	if (ifp == NULL)
1615 		return ENXIO;
1616 
1617 	/* We have to disable ioctls here */
1618 	IFNET_LOCK(ifp);
1619 	ifp->if_ioctl = if_nullioctl;
1620 	IFNET_UNLOCK(ifp);
1621 
1622 	/*
1623 	 * We cannot call ifc_destroy with holding ifp.
1624 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1625 	 */
1626 	if_put(ifp, &psref);
1627 
1628 	return (*ifc->ifc_destroy)(ifp);
1629 }
1630 
1631 static bool
1632 if_is_unit(const char *name)
1633 {
1634 
1635 	while(*name != '\0') {
1636 		if (*name < '0' || *name > '9')
1637 			return false;
1638 		name++;
1639 	}
1640 
1641 	return true;
1642 }
1643 
1644 /*
1645  * Look up a network interface cloner.
1646  */
1647 static struct if_clone *
1648 if_clone_lookup(const char *name, int *unitp)
1649 {
1650 	struct if_clone *ifc;
1651 	const char *cp;
1652 	char *dp, ifname[IFNAMSIZ + 3];
1653 	int unit;
1654 
1655 	KASSERT(mutex_owned(&if_clone_mtx));
1656 
1657 	strcpy(ifname, "if_");
1658 	/* separate interface name from unit */
1659 	/* TODO: search unit number from backward */
1660 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1661 	    *cp && !if_is_unit(cp);)
1662 		*dp++ = *cp++;
1663 
1664 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1665 		return NULL;	/* No name or unit number */
1666 	*dp++ = '\0';
1667 
1668 again:
1669 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1670 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1671 			break;
1672 	}
1673 
1674 	if (ifc == NULL) {
1675 		int error;
1676 		if (*ifname == '\0')
1677 			return NULL;
1678 		mutex_exit(&if_clone_mtx);
1679 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1680 		mutex_enter(&if_clone_mtx);
1681 		if (error)
1682 			return NULL;
1683 		*ifname = '\0';
1684 		goto again;
1685 	}
1686 
1687 	unit = 0;
1688 	while (cp - name < IFNAMSIZ && *cp) {
1689 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1690 			/* Bogus unit number. */
1691 			return NULL;
1692 		}
1693 		unit = (unit * 10) + (*cp++ - '0');
1694 	}
1695 
1696 	if (unitp != NULL)
1697 		*unitp = unit;
1698 	return ifc;
1699 }
1700 
1701 /*
1702  * Register a network interface cloner.
1703  */
1704 void
1705 if_clone_attach(struct if_clone *ifc)
1706 {
1707 
1708 	mutex_enter(&if_clone_mtx);
1709 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1710 	if_cloners_count++;
1711 	mutex_exit(&if_clone_mtx);
1712 }
1713 
1714 /*
1715  * Unregister a network interface cloner.
1716  */
1717 void
1718 if_clone_detach(struct if_clone *ifc)
1719 {
1720 
1721 	mutex_enter(&if_clone_mtx);
1722 	LIST_REMOVE(ifc, ifc_list);
1723 	if_cloners_count--;
1724 	mutex_exit(&if_clone_mtx);
1725 }
1726 
1727 /*
1728  * Provide list of interface cloners to userspace.
1729  */
1730 int
1731 if_clone_list(int buf_count, char *buffer, int *total)
1732 {
1733 	char outbuf[IFNAMSIZ], *dst;
1734 	struct if_clone *ifc;
1735 	int count, error = 0;
1736 
1737 	mutex_enter(&if_clone_mtx);
1738 	*total = if_cloners_count;
1739 	if ((dst = buffer) == NULL) {
1740 		/* Just asking how many there are. */
1741 		goto out;
1742 	}
1743 
1744 	if (buf_count < 0) {
1745 		error = EINVAL;
1746 		goto out;
1747 	}
1748 
1749 	count = (if_cloners_count < buf_count) ?
1750 	    if_cloners_count : buf_count;
1751 
1752 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1753 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1754 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1755 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1756 			error = ENAMETOOLONG;
1757 			goto out;
1758 		}
1759 		error = copyout(outbuf, dst, sizeof(outbuf));
1760 		if (error != 0)
1761 			break;
1762 	}
1763 
1764 out:
1765 	mutex_exit(&if_clone_mtx);
1766 	return error;
1767 }
1768 
1769 void
1770 ifa_psref_init(struct ifaddr *ifa)
1771 {
1772 
1773 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1774 }
1775 
1776 void
1777 ifaref(struct ifaddr *ifa)
1778 {
1779 	KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING));
1780 	ifa->ifa_refcnt++;
1781 }
1782 
1783 void
1784 ifafree(struct ifaddr *ifa)
1785 {
1786 	KASSERT(ifa != NULL);
1787 	KASSERT(ifa->ifa_refcnt > 0);
1788 
1789 	if (--ifa->ifa_refcnt == 0) {
1790 		free(ifa, M_IFADDR);
1791 	}
1792 }
1793 
1794 bool
1795 ifa_is_destroying(struct ifaddr *ifa)
1796 {
1797 
1798 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1799 }
1800 
1801 void
1802 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1803 {
1804 
1805 	ifa->ifa_ifp = ifp;
1806 
1807 	/*
1808 	 * Check MP-safety for IFEF_MPSAFE drivers.
1809 	 * Check !IFF_RUNNING for initialization routines that normally don't
1810 	 * take IFNET_LOCK but it's safe because there is no competitor.
1811 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1812 	 * if_stop.
1813 	 */
1814 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1815 	    IFNET_LOCKED(ifp));
1816 
1817 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1818 	IFADDR_ENTRY_INIT(ifa);
1819 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1820 
1821 	ifaref(ifa);
1822 }
1823 
1824 void
1825 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1826 {
1827 
1828 	KASSERT(ifa->ifa_ifp == ifp);
1829 	/*
1830 	 * Check MP-safety for IFEF_MPSAFE drivers.
1831 	 * if_is_deactivated indicates ifa_remove is called form if_detach
1832 	 * where is safe even if IFNET_LOCK isn't held.
1833 	 */
1834 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp));
1835 
1836 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1837 	IFADDR_WRITER_REMOVE(ifa);
1838 #ifdef NET_MPSAFE
1839 	IFNET_GLOBAL_LOCK();
1840 	pserialize_perform(ifnet_psz);
1841 	IFNET_GLOBAL_UNLOCK();
1842 #endif
1843 
1844 #ifdef NET_MPSAFE
1845 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1846 #endif
1847 	IFADDR_ENTRY_DESTROY(ifa);
1848 	ifafree(ifa);
1849 }
1850 
1851 void
1852 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1853 {
1854 
1855 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1856 }
1857 
1858 void
1859 ifa_release(struct ifaddr *ifa, struct psref *psref)
1860 {
1861 
1862 	if (ifa == NULL)
1863 		return;
1864 
1865 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1866 }
1867 
1868 bool
1869 ifa_held(struct ifaddr *ifa)
1870 {
1871 
1872 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1873 }
1874 
1875 static inline int
1876 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1877 {
1878 	return sockaddr_cmp(sa1, sa2) == 0;
1879 }
1880 
1881 /*
1882  * Locate an interface based on a complete address.
1883  */
1884 /*ARGSUSED*/
1885 struct ifaddr *
1886 ifa_ifwithaddr(const struct sockaddr *addr)
1887 {
1888 	struct ifnet *ifp;
1889 	struct ifaddr *ifa;
1890 
1891 	IFNET_READER_FOREACH(ifp) {
1892 		if (if_is_deactivated(ifp))
1893 			continue;
1894 		IFADDR_READER_FOREACH(ifa, ifp) {
1895 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1896 				continue;
1897 			if (equal(addr, ifa->ifa_addr))
1898 				return ifa;
1899 			if ((ifp->if_flags & IFF_BROADCAST) &&
1900 			    ifa->ifa_broadaddr &&
1901 			    /* IP6 doesn't have broadcast */
1902 			    ifa->ifa_broadaddr->sa_len != 0 &&
1903 			    equal(ifa->ifa_broadaddr, addr))
1904 				return ifa;
1905 		}
1906 	}
1907 	return NULL;
1908 }
1909 
1910 struct ifaddr *
1911 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1912 {
1913 	struct ifaddr *ifa;
1914 	int s = pserialize_read_enter();
1915 
1916 	ifa = ifa_ifwithaddr(addr);
1917 	if (ifa != NULL)
1918 		ifa_acquire(ifa, psref);
1919 	pserialize_read_exit(s);
1920 
1921 	return ifa;
1922 }
1923 
1924 /*
1925  * Locate the point to point interface with a given destination address.
1926  */
1927 /*ARGSUSED*/
1928 struct ifaddr *
1929 ifa_ifwithdstaddr(const struct sockaddr *addr)
1930 {
1931 	struct ifnet *ifp;
1932 	struct ifaddr *ifa;
1933 
1934 	IFNET_READER_FOREACH(ifp) {
1935 		if (if_is_deactivated(ifp))
1936 			continue;
1937 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1938 			continue;
1939 		IFADDR_READER_FOREACH(ifa, ifp) {
1940 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1941 			    ifa->ifa_dstaddr == NULL)
1942 				continue;
1943 			if (equal(addr, ifa->ifa_dstaddr))
1944 				return ifa;
1945 		}
1946 	}
1947 
1948 	return NULL;
1949 }
1950 
1951 struct ifaddr *
1952 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1953 {
1954 	struct ifaddr *ifa;
1955 	int s;
1956 
1957 	s = pserialize_read_enter();
1958 	ifa = ifa_ifwithdstaddr(addr);
1959 	if (ifa != NULL)
1960 		ifa_acquire(ifa, psref);
1961 	pserialize_read_exit(s);
1962 
1963 	return ifa;
1964 }
1965 
1966 /*
1967  * Find an interface on a specific network.  If many, choice
1968  * is most specific found.
1969  */
1970 struct ifaddr *
1971 ifa_ifwithnet(const struct sockaddr *addr)
1972 {
1973 	struct ifnet *ifp;
1974 	struct ifaddr *ifa, *ifa_maybe = NULL;
1975 	const struct sockaddr_dl *sdl;
1976 	u_int af = addr->sa_family;
1977 	const char *addr_data = addr->sa_data, *cplim;
1978 
1979 	if (af == AF_LINK) {
1980 		sdl = satocsdl(addr);
1981 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1982 		    ifindex2ifnet[sdl->sdl_index] &&
1983 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1984 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1985 		}
1986 	}
1987 #ifdef NETATALK
1988 	if (af == AF_APPLETALK) {
1989 		const struct sockaddr_at *sat, *sat2;
1990 		sat = (const struct sockaddr_at *)addr;
1991 		IFNET_READER_FOREACH(ifp) {
1992 			if (if_is_deactivated(ifp))
1993 				continue;
1994 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1995 			if (ifa == NULL)
1996 				continue;
1997 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1998 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1999 				return ifa; /* exact match */
2000 			if (ifa_maybe == NULL) {
2001 				/* else keep the if with the right range */
2002 				ifa_maybe = ifa;
2003 			}
2004 		}
2005 		return ifa_maybe;
2006 	}
2007 #endif
2008 	IFNET_READER_FOREACH(ifp) {
2009 		if (if_is_deactivated(ifp))
2010 			continue;
2011 		IFADDR_READER_FOREACH(ifa, ifp) {
2012 			const char *cp, *cp2, *cp3;
2013 
2014 			if (ifa->ifa_addr->sa_family != af ||
2015 			    ifa->ifa_netmask == NULL)
2016  next:				continue;
2017 			cp = addr_data;
2018 			cp2 = ifa->ifa_addr->sa_data;
2019 			cp3 = ifa->ifa_netmask->sa_data;
2020 			cplim = (const char *)ifa->ifa_netmask +
2021 			    ifa->ifa_netmask->sa_len;
2022 			while (cp3 < cplim) {
2023 				if ((*cp++ ^ *cp2++) & *cp3++) {
2024 					/* want to continue for() loop */
2025 					goto next;
2026 				}
2027 			}
2028 			if (ifa_maybe == NULL ||
2029 			    rt_refines(ifa->ifa_netmask,
2030 			               ifa_maybe->ifa_netmask))
2031 				ifa_maybe = ifa;
2032 		}
2033 	}
2034 	return ifa_maybe;
2035 }
2036 
2037 struct ifaddr *
2038 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2039 {
2040 	struct ifaddr *ifa;
2041 	int s;
2042 
2043 	s = pserialize_read_enter();
2044 	ifa = ifa_ifwithnet(addr);
2045 	if (ifa != NULL)
2046 		ifa_acquire(ifa, psref);
2047 	pserialize_read_exit(s);
2048 
2049 	return ifa;
2050 }
2051 
2052 /*
2053  * Find the interface of the addresss.
2054  */
2055 struct ifaddr *
2056 ifa_ifwithladdr(const struct sockaddr *addr)
2057 {
2058 	struct ifaddr *ia;
2059 
2060 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2061 	    (ia = ifa_ifwithnet(addr)))
2062 		return ia;
2063 	return NULL;
2064 }
2065 
2066 struct ifaddr *
2067 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2068 {
2069 	struct ifaddr *ifa;
2070 	int s;
2071 
2072 	s = pserialize_read_enter();
2073 	ifa = ifa_ifwithladdr(addr);
2074 	if (ifa != NULL)
2075 		ifa_acquire(ifa, psref);
2076 	pserialize_read_exit(s);
2077 
2078 	return ifa;
2079 }
2080 
2081 /*
2082  * Find an interface using a specific address family
2083  */
2084 struct ifaddr *
2085 ifa_ifwithaf(int af)
2086 {
2087 	struct ifnet *ifp;
2088 	struct ifaddr *ifa = NULL;
2089 	int s;
2090 
2091 	s = pserialize_read_enter();
2092 	IFNET_READER_FOREACH(ifp) {
2093 		if (if_is_deactivated(ifp))
2094 			continue;
2095 		IFADDR_READER_FOREACH(ifa, ifp) {
2096 			if (ifa->ifa_addr->sa_family == af)
2097 				goto out;
2098 		}
2099 	}
2100 out:
2101 	pserialize_read_exit(s);
2102 	return ifa;
2103 }
2104 
2105 /*
2106  * Find an interface address specific to an interface best matching
2107  * a given address.
2108  */
2109 struct ifaddr *
2110 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2111 {
2112 	struct ifaddr *ifa;
2113 	const char *cp, *cp2, *cp3;
2114 	const char *cplim;
2115 	struct ifaddr *ifa_maybe = 0;
2116 	u_int af = addr->sa_family;
2117 
2118 	if (if_is_deactivated(ifp))
2119 		return NULL;
2120 
2121 	if (af >= AF_MAX)
2122 		return NULL;
2123 
2124 	IFADDR_READER_FOREACH(ifa, ifp) {
2125 		if (ifa->ifa_addr->sa_family != af)
2126 			continue;
2127 		ifa_maybe = ifa;
2128 		if (ifa->ifa_netmask == NULL) {
2129 			if (equal(addr, ifa->ifa_addr) ||
2130 			    (ifa->ifa_dstaddr &&
2131 			     equal(addr, ifa->ifa_dstaddr)))
2132 				return ifa;
2133 			continue;
2134 		}
2135 		cp = addr->sa_data;
2136 		cp2 = ifa->ifa_addr->sa_data;
2137 		cp3 = ifa->ifa_netmask->sa_data;
2138 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2139 		for (; cp3 < cplim; cp3++) {
2140 			if ((*cp++ ^ *cp2++) & *cp3)
2141 				break;
2142 		}
2143 		if (cp3 == cplim)
2144 			return ifa;
2145 	}
2146 	return ifa_maybe;
2147 }
2148 
2149 struct ifaddr *
2150 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2151     struct psref *psref)
2152 {
2153 	struct ifaddr *ifa;
2154 	int s;
2155 
2156 	s = pserialize_read_enter();
2157 	ifa = ifaof_ifpforaddr(addr, ifp);
2158 	if (ifa != NULL)
2159 		ifa_acquire(ifa, psref);
2160 	pserialize_read_exit(s);
2161 
2162 	return ifa;
2163 }
2164 
2165 /*
2166  * Default action when installing a route with a Link Level gateway.
2167  * Lookup an appropriate real ifa to point to.
2168  * This should be moved to /sys/net/link.c eventually.
2169  */
2170 void
2171 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2172 {
2173 	struct ifaddr *ifa;
2174 	const struct sockaddr *dst;
2175 	struct ifnet *ifp;
2176 	struct psref psref;
2177 
2178 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2179 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
2180 		return;
2181 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2182 		rt_replace_ifa(rt, ifa);
2183 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2184 			ifa->ifa_rtrequest(cmd, rt, info);
2185 		ifa_release(ifa, &psref);
2186 	}
2187 }
2188 
2189 /*
2190  * bitmask macros to manage a densely packed link_state change queue.
2191  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2192  * LINK_STATE_UP(2) we need 2 bits for each state change.
2193  * As a state change to store is 0, treat all bits set as an unset item.
2194  */
2195 #define LQ_ITEM_BITS		2
2196 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2197 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2198 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2199 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2200 #define LQ_STORE(q, i, v)						      \
2201 	do {								      \
2202 		(q) &= ~LQ_MASK((i));					      \
2203 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2204 	} while (0 /* CONSTCOND */)
2205 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2206 #define LQ_POP(q, v)							      \
2207 	do {								      \
2208 		(v) = LQ_ITEM((q), 0);					      \
2209 		(q) >>= LQ_ITEM_BITS;					      \
2210 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2211 	} while (0 /* CONSTCOND */)
2212 #define LQ_PUSH(q, v)							      \
2213 	do {								      \
2214 		(q) >>= LQ_ITEM_BITS;					      \
2215 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2216 	} while (0 /* CONSTCOND */)
2217 #define LQ_FIND_UNSET(q, i)						      \
2218 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2219 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2220 			break;						      \
2221 	}
2222 
2223 /*
2224  * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex
2225  * for each ifnet.  It doesn't matter because:
2226  * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on
2227  *   ifq_lock don't happen
2228  * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock
2229  *   because if_snd, if_link_state_change and if_link_state_change_softint
2230  *   are all called with KERNEL_LOCK
2231  */
2232 #define IF_LINK_STATE_CHANGE_LOCK(ifp)		\
2233 	mutex_enter((ifp)->if_snd.ifq_lock)
2234 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp)	\
2235 	mutex_exit((ifp)->if_snd.ifq_lock)
2236 
2237 /*
2238  * Handle a change in the interface link state and
2239  * queue notifications.
2240  */
2241 void
2242 if_link_state_change(struct ifnet *ifp, int link_state)
2243 {
2244 	int idx;
2245 
2246 	KASSERTMSG(if_is_link_state_changeable(ifp),
2247 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2248 	    ifp->if_xname, ifp->if_extflags);
2249 
2250 	/* Ensure change is to a valid state */
2251 	switch (link_state) {
2252 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2253 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2254 	case LINK_STATE_UP:
2255 		break;
2256 	default:
2257 #ifdef DEBUG
2258 		printf("%s: invalid link state %d\n",
2259 		    ifp->if_xname, link_state);
2260 #endif
2261 		return;
2262 	}
2263 
2264 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2265 
2266 	/* Find the last unset event in the queue. */
2267 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2268 
2269 	/*
2270 	 * Ensure link_state doesn't match the last event in the queue.
2271 	 * ifp->if_link_state is not checked and set here because
2272 	 * that would present an inconsistent picture to the system.
2273 	 */
2274 	if (idx != 0 &&
2275 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2276 		goto out;
2277 
2278 	/* Handle queue overflow. */
2279 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2280 		uint8_t lost;
2281 
2282 		/*
2283 		 * The DOWN state must be protected from being pushed off
2284 		 * the queue to ensure that userland will always be
2285 		 * in a sane state.
2286 		 * Because DOWN is protected, there is no need to protect
2287 		 * UNKNOWN.
2288 		 * It should be invalid to change from any other state to
2289 		 * UNKNOWN anyway ...
2290 		 */
2291 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2292 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2293 		if (lost == LINK_STATE_DOWN) {
2294 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2295 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2296 		}
2297 		printf("%s: lost link state change %s\n",
2298 		    ifp->if_xname,
2299 		    lost == LINK_STATE_UP ? "UP" :
2300 		    lost == LINK_STATE_DOWN ? "DOWN" :
2301 		    "UNKNOWN");
2302 	} else
2303 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2304 
2305 	softint_schedule(ifp->if_link_si);
2306 
2307 out:
2308 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2309 }
2310 
2311 /*
2312  * Handle interface link state change notifications.
2313  */
2314 void
2315 if_link_state_change_softint(struct ifnet *ifp, int link_state)
2316 {
2317 	struct domain *dp;
2318 	int s = splnet();
2319 	bool notify;
2320 
2321 	KASSERT(!cpu_intr_p());
2322 
2323 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2324 
2325 	/* Ensure the change is still valid. */
2326 	if (ifp->if_link_state == link_state) {
2327 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2328 		return;
2329 	}
2330 
2331 #ifdef DEBUG
2332 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2333 		link_state == LINK_STATE_UP ? "UP" :
2334 		link_state == LINK_STATE_DOWN ? "DOWN" :
2335 		"UNKNOWN",
2336 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2337 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2338 		"UNKNOWN");
2339 #endif
2340 
2341 	/*
2342 	 * When going from UNKNOWN to UP, we need to mark existing
2343 	 * addresses as tentative and restart DAD as we may have
2344 	 * erroneously not found a duplicate.
2345 	 *
2346 	 * This needs to happen before rt_ifmsg to avoid a race where
2347 	 * listeners would have an address and expect it to work right
2348 	 * away.
2349 	 */
2350 	notify = (link_state == LINK_STATE_UP &&
2351 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2352 	ifp->if_link_state = link_state;
2353 	/* The following routines may sleep so release the spin mutex */
2354 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2355 
2356 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2357 	if (notify) {
2358 		DOMAIN_FOREACH(dp) {
2359 			if (dp->dom_if_link_state_change != NULL)
2360 				dp->dom_if_link_state_change(ifp,
2361 				    LINK_STATE_DOWN);
2362 		}
2363 	}
2364 
2365 	/* Notify that the link state has changed. */
2366 	rt_ifmsg(ifp);
2367 
2368 #if NCARP > 0
2369 	if (ifp->if_carp)
2370 		carp_carpdev_state(ifp);
2371 #endif
2372 
2373 	DOMAIN_FOREACH(dp) {
2374 		if (dp->dom_if_link_state_change != NULL)
2375 			dp->dom_if_link_state_change(ifp, link_state);
2376 	}
2377 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2378 	splx(s);
2379 }
2380 
2381 /*
2382  * Process the interface link state change queue.
2383  */
2384 static void
2385 if_link_state_change_si(void *arg)
2386 {
2387 	struct ifnet *ifp = arg;
2388 	int s;
2389 	uint8_t state;
2390 	bool schedule;
2391 
2392 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2393 	s = splnet();
2394 
2395 	/* Pop a link state change from the queue and process it. */
2396 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2397 	LQ_POP(ifp->if_link_queue, state);
2398 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2399 
2400 	if_link_state_change_softint(ifp, state);
2401 
2402 	/* If there is a link state change to come, schedule it. */
2403 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2404 	schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET);
2405 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2406 	if (schedule)
2407 		softint_schedule(ifp->if_link_si);
2408 
2409 	splx(s);
2410 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2411 }
2412 
2413 /*
2414  * Default action when installing a local route on a point-to-point
2415  * interface.
2416  */
2417 void
2418 p2p_rtrequest(int req, struct rtentry *rt,
2419     __unused const struct rt_addrinfo *info)
2420 {
2421 	struct ifnet *ifp = rt->rt_ifp;
2422 	struct ifaddr *ifa, *lo0ifa;
2423 	int s = pserialize_read_enter();
2424 
2425 	switch (req) {
2426 	case RTM_ADD:
2427 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2428 			break;
2429 
2430 		rt->rt_ifp = lo0ifp;
2431 
2432 		IFADDR_READER_FOREACH(ifa, ifp) {
2433 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2434 				break;
2435 		}
2436 		if (ifa == NULL)
2437 			break;
2438 
2439 		/*
2440 		 * Ensure lo0 has an address of the same family.
2441 		 */
2442 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2443 			if (lo0ifa->ifa_addr->sa_family ==
2444 			    ifa->ifa_addr->sa_family)
2445 				break;
2446 		}
2447 		if (lo0ifa == NULL)
2448 			break;
2449 
2450 		/*
2451 		 * Make sure to set rt->rt_ifa to the interface
2452 		 * address we are using, otherwise we will have trouble
2453 		 * with source address selection.
2454 		 */
2455 		if (ifa != rt->rt_ifa)
2456 			rt_replace_ifa(rt, ifa);
2457 		break;
2458 	case RTM_DELETE:
2459 	default:
2460 		break;
2461 	}
2462 	pserialize_read_exit(s);
2463 }
2464 
2465 static void
2466 _if_down(struct ifnet *ifp)
2467 {
2468 	struct ifaddr *ifa;
2469 	struct domain *dp;
2470 	int s, bound;
2471 	struct psref psref;
2472 
2473 	ifp->if_flags &= ~IFF_UP;
2474 	nanotime(&ifp->if_lastchange);
2475 
2476 	bound = curlwp_bind();
2477 	s = pserialize_read_enter();
2478 	IFADDR_READER_FOREACH(ifa, ifp) {
2479 		ifa_acquire(ifa, &psref);
2480 		pserialize_read_exit(s);
2481 
2482 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2483 
2484 		s = pserialize_read_enter();
2485 		ifa_release(ifa, &psref);
2486 	}
2487 	pserialize_read_exit(s);
2488 	curlwp_bindx(bound);
2489 
2490 	IFQ_PURGE(&ifp->if_snd);
2491 #if NCARP > 0
2492 	if (ifp->if_carp)
2493 		carp_carpdev_state(ifp);
2494 #endif
2495 	rt_ifmsg(ifp);
2496 	DOMAIN_FOREACH(dp) {
2497 		if (dp->dom_if_down)
2498 			dp->dom_if_down(ifp);
2499 	}
2500 }
2501 
2502 static void
2503 if_down_deactivated(struct ifnet *ifp)
2504 {
2505 
2506 	KASSERT(if_is_deactivated(ifp));
2507 	_if_down(ifp);
2508 }
2509 
2510 void
2511 if_down_locked(struct ifnet *ifp)
2512 {
2513 
2514 	KASSERT(IFNET_LOCKED(ifp));
2515 	_if_down(ifp);
2516 }
2517 
2518 /*
2519  * Mark an interface down and notify protocols of
2520  * the transition.
2521  * NOTE: must be called at splsoftnet or equivalent.
2522  */
2523 void
2524 if_down(struct ifnet *ifp)
2525 {
2526 
2527 	IFNET_LOCK(ifp);
2528 	if_down_locked(ifp);
2529 	IFNET_UNLOCK(ifp);
2530 }
2531 
2532 /*
2533  * Must be called with holding if_ioctl_lock.
2534  */
2535 static void
2536 if_up_locked(struct ifnet *ifp)
2537 {
2538 #ifdef notyet
2539 	struct ifaddr *ifa;
2540 #endif
2541 	struct domain *dp;
2542 
2543 	KASSERT(IFNET_LOCKED(ifp));
2544 
2545 	KASSERT(!if_is_deactivated(ifp));
2546 	ifp->if_flags |= IFF_UP;
2547 	nanotime(&ifp->if_lastchange);
2548 #ifdef notyet
2549 	/* this has no effect on IP, and will kill all ISO connections XXX */
2550 	IFADDR_READER_FOREACH(ifa, ifp)
2551 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2552 #endif
2553 #if NCARP > 0
2554 	if (ifp->if_carp)
2555 		carp_carpdev_state(ifp);
2556 #endif
2557 	rt_ifmsg(ifp);
2558 	DOMAIN_FOREACH(dp) {
2559 		if (dp->dom_if_up)
2560 			dp->dom_if_up(ifp);
2561 	}
2562 }
2563 
2564 /*
2565  * Handle interface slowtimo timer routine.  Called
2566  * from softclock, we decrement timer (if set) and
2567  * call the appropriate interface routine on expiration.
2568  */
2569 static void
2570 if_slowtimo(void *arg)
2571 {
2572 	void (*slowtimo)(struct ifnet *);
2573 	struct ifnet *ifp = arg;
2574 	int s;
2575 
2576 	slowtimo = ifp->if_slowtimo;
2577 	if (__predict_false(slowtimo == NULL))
2578 		return;
2579 
2580 	s = splnet();
2581 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2582 		(*slowtimo)(ifp);
2583 
2584 	splx(s);
2585 
2586 	if (__predict_true(ifp->if_slowtimo != NULL))
2587 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2588 }
2589 
2590 /*
2591  * Mark an interface up and notify protocols of
2592  * the transition.
2593  * NOTE: must be called at splsoftnet or equivalent.
2594  */
2595 void
2596 if_up(struct ifnet *ifp)
2597 {
2598 
2599 	IFNET_LOCK(ifp);
2600 	if_up_locked(ifp);
2601 	IFNET_UNLOCK(ifp);
2602 }
2603 
2604 /*
2605  * Set/clear promiscuous mode on interface ifp based on the truth value
2606  * of pswitch.  The calls are reference counted so that only the first
2607  * "on" request actually has an effect, as does the final "off" request.
2608  * Results are undefined if the "off" and "on" requests are not matched.
2609  */
2610 int
2611 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2612 {
2613 	int pcount, ret = 0;
2614 	short nflags;
2615 
2616 	KASSERT(IFNET_LOCKED(ifp));
2617 
2618 	pcount = ifp->if_pcount;
2619 	if (pswitch) {
2620 		/*
2621 		 * Allow the device to be "placed" into promiscuous
2622 		 * mode even if it is not configured up.  It will
2623 		 * consult IFF_PROMISC when it is brought up.
2624 		 */
2625 		if (ifp->if_pcount++ != 0)
2626 			goto out;
2627 		nflags = ifp->if_flags | IFF_PROMISC;
2628 	} else {
2629 		if (--ifp->if_pcount > 0)
2630 			goto out;
2631 		nflags = ifp->if_flags & ~IFF_PROMISC;
2632 	}
2633 	ret = if_flags_set(ifp, nflags);
2634 	/* Restore interface state if not successful. */
2635 	if (ret != 0) {
2636 		ifp->if_pcount = pcount;
2637 	}
2638 out:
2639 	return ret;
2640 }
2641 
2642 int
2643 ifpromisc(struct ifnet *ifp, int pswitch)
2644 {
2645 	int e;
2646 
2647 	IFNET_LOCK(ifp);
2648 	e = ifpromisc_locked(ifp, pswitch);
2649 	IFNET_UNLOCK(ifp);
2650 
2651 	return e;
2652 }
2653 
2654 /*
2655  * Map interface name to
2656  * interface structure pointer.
2657  */
2658 struct ifnet *
2659 ifunit(const char *name)
2660 {
2661 	struct ifnet *ifp;
2662 	const char *cp = name;
2663 	u_int unit = 0;
2664 	u_int i;
2665 	int s;
2666 
2667 	/*
2668 	 * If the entire name is a number, treat it as an ifindex.
2669 	 */
2670 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2671 		unit = unit * 10 + (*cp - '0');
2672 	}
2673 
2674 	/*
2675 	 * If the number took all of the name, then it's a valid ifindex.
2676 	 */
2677 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2678 		return if_byindex(unit);
2679 
2680 	ifp = NULL;
2681 	s = pserialize_read_enter();
2682 	IFNET_READER_FOREACH(ifp) {
2683 		if (if_is_deactivated(ifp))
2684 			continue;
2685 	 	if (strcmp(ifp->if_xname, name) == 0)
2686 			goto out;
2687 	}
2688 out:
2689 	pserialize_read_exit(s);
2690 	return ifp;
2691 }
2692 
2693 /*
2694  * Get a reference of an ifnet object by an interface name.
2695  * The returned reference is protected by psref(9). The caller
2696  * must release a returned reference by if_put after use.
2697  */
2698 struct ifnet *
2699 if_get(const char *name, struct psref *psref)
2700 {
2701 	struct ifnet *ifp;
2702 	const char *cp = name;
2703 	u_int unit = 0;
2704 	u_int i;
2705 	int s;
2706 
2707 	/*
2708 	 * If the entire name is a number, treat it as an ifindex.
2709 	 */
2710 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2711 		unit = unit * 10 + (*cp - '0');
2712 	}
2713 
2714 	/*
2715 	 * If the number took all of the name, then it's a valid ifindex.
2716 	 */
2717 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2718 		return if_get_byindex(unit, psref);
2719 
2720 	ifp = NULL;
2721 	s = pserialize_read_enter();
2722 	IFNET_READER_FOREACH(ifp) {
2723 		if (if_is_deactivated(ifp))
2724 			continue;
2725 		if (strcmp(ifp->if_xname, name) == 0) {
2726 			psref_acquire(psref, &ifp->if_psref,
2727 			    ifnet_psref_class);
2728 			goto out;
2729 		}
2730 	}
2731 out:
2732 	pserialize_read_exit(s);
2733 	return ifp;
2734 }
2735 
2736 /*
2737  * Release a reference of an ifnet object given by if_get, if_get_byindex
2738  * or if_get_bylla.
2739  */
2740 void
2741 if_put(const struct ifnet *ifp, struct psref *psref)
2742 {
2743 
2744 	if (ifp == NULL)
2745 		return;
2746 
2747 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2748 }
2749 
2750 /*
2751  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2752  * should be used.
2753  */
2754 ifnet_t *
2755 _if_byindex(u_int idx)
2756 {
2757 
2758 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2759 }
2760 
2761 /*
2762  * Return ifp having idx. Return NULL if not found or the found ifp is
2763  * already deactivated.
2764  */
2765 ifnet_t *
2766 if_byindex(u_int idx)
2767 {
2768 	ifnet_t *ifp;
2769 
2770 	ifp = _if_byindex(idx);
2771 	if (ifp != NULL && if_is_deactivated(ifp))
2772 		ifp = NULL;
2773 	return ifp;
2774 }
2775 
2776 /*
2777  * Get a reference of an ifnet object by an interface index.
2778  * The returned reference is protected by psref(9). The caller
2779  * must release a returned reference by if_put after use.
2780  */
2781 ifnet_t *
2782 if_get_byindex(u_int idx, struct psref *psref)
2783 {
2784 	ifnet_t *ifp;
2785 	int s;
2786 
2787 	s = pserialize_read_enter();
2788 	ifp = if_byindex(idx);
2789 	if (__predict_true(ifp != NULL))
2790 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2791 	pserialize_read_exit(s);
2792 
2793 	return ifp;
2794 }
2795 
2796 ifnet_t *
2797 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2798 {
2799 	ifnet_t *ifp;
2800 	int s;
2801 
2802 	s = pserialize_read_enter();
2803 	IFNET_READER_FOREACH(ifp) {
2804 		if (if_is_deactivated(ifp))
2805 			continue;
2806 		if (ifp->if_addrlen != lla_len)
2807 			continue;
2808 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2809 			psref_acquire(psref, &ifp->if_psref,
2810 			    ifnet_psref_class);
2811 			break;
2812 		}
2813 	}
2814 	pserialize_read_exit(s);
2815 
2816 	return ifp;
2817 }
2818 
2819 /*
2820  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2821  * for example using pserialize or the ifp is already held or some other
2822  * object is held which guarantes the ifp to not be freed indirectly.
2823  */
2824 void
2825 if_acquire(struct ifnet *ifp, struct psref *psref)
2826 {
2827 
2828 	KASSERT(ifp->if_index != 0);
2829 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2830 }
2831 
2832 bool
2833 if_held(struct ifnet *ifp)
2834 {
2835 
2836 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2837 }
2838 
2839 /*
2840  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4.
2841  * Check the tunnel nesting count.
2842  * Return > 0, if tunnel nesting count is more than limit.
2843  * Return 0, if tunnel nesting count is equal or less than limit.
2844  */
2845 int
2846 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
2847 {
2848 	struct m_tag *mtag;
2849 	int *count;
2850 
2851 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL);
2852 	if (mtag != NULL) {
2853 		count = (int *)(mtag + 1);
2854 		if (++(*count) > limit) {
2855 			log(LOG_NOTICE,
2856 			    "%s: recursively called too many times(%d)\n",
2857 			    ifp->if_xname, *count);
2858 			return EIO;
2859 		}
2860 	} else {
2861 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
2862 		    M_NOWAIT);
2863 		if (mtag != NULL) {
2864 			m_tag_prepend(m, mtag);
2865 			count = (int *)(mtag + 1);
2866 			*count = 0;
2867 		} else {
2868 			log(LOG_DEBUG,
2869 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
2870 			    ifp->if_xname);
2871 		}
2872 	}
2873 
2874 	return 0;
2875 }
2876 
2877 /* common */
2878 int
2879 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2880 {
2881 	int s;
2882 	struct ifreq *ifr;
2883 	struct ifcapreq *ifcr;
2884 	struct ifdatareq *ifdr;
2885 
2886 	switch (cmd) {
2887 	case SIOCSIFCAP:
2888 		ifcr = data;
2889 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2890 			return EINVAL;
2891 
2892 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2893 			return 0;
2894 
2895 		ifp->if_capenable = ifcr->ifcr_capenable;
2896 
2897 		/* Pre-compute the checksum flags mask. */
2898 		ifp->if_csum_flags_tx = 0;
2899 		ifp->if_csum_flags_rx = 0;
2900 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2901 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2902 		}
2903 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2904 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2905 		}
2906 
2907 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2908 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2909 		}
2910 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2911 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2912 		}
2913 
2914 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2915 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2916 		}
2917 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2918 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2919 		}
2920 
2921 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2922 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2923 		}
2924 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2925 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2926 		}
2927 
2928 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2929 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2930 		}
2931 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2932 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2933 		}
2934 		if (ifp->if_flags & IFF_UP)
2935 			return ENETRESET;
2936 		return 0;
2937 	case SIOCSIFFLAGS:
2938 		ifr = data;
2939 		/*
2940 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
2941 		 * and if_down aren't MP-safe yet, so we must hold the lock.
2942 		 */
2943 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
2944 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2945 			s = splsoftnet();
2946 			if_down_locked(ifp);
2947 			splx(s);
2948 		}
2949 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2950 			s = splsoftnet();
2951 			if_up_locked(ifp);
2952 			splx(s);
2953 		}
2954 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
2955 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2956 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2957 		break;
2958 	case SIOCGIFFLAGS:
2959 		ifr = data;
2960 		ifr->ifr_flags = ifp->if_flags;
2961 		break;
2962 
2963 	case SIOCGIFMETRIC:
2964 		ifr = data;
2965 		ifr->ifr_metric = ifp->if_metric;
2966 		break;
2967 
2968 	case SIOCGIFMTU:
2969 		ifr = data;
2970 		ifr->ifr_mtu = ifp->if_mtu;
2971 		break;
2972 
2973 	case SIOCGIFDLT:
2974 		ifr = data;
2975 		ifr->ifr_dlt = ifp->if_dlt;
2976 		break;
2977 
2978 	case SIOCGIFCAP:
2979 		ifcr = data;
2980 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2981 		ifcr->ifcr_capenable = ifp->if_capenable;
2982 		break;
2983 
2984 	case SIOCSIFMETRIC:
2985 		ifr = data;
2986 		ifp->if_metric = ifr->ifr_metric;
2987 		break;
2988 
2989 	case SIOCGIFDATA:
2990 		ifdr = data;
2991 		ifdr->ifdr_data = ifp->if_data;
2992 		break;
2993 
2994 	case SIOCGIFINDEX:
2995 		ifr = data;
2996 		ifr->ifr_index = ifp->if_index;
2997 		break;
2998 
2999 	case SIOCZIFDATA:
3000 		ifdr = data;
3001 		ifdr->ifdr_data = ifp->if_data;
3002 		/*
3003 		 * Assumes that the volatile counters that can be
3004 		 * zero'ed are at the end of if_data.
3005 		 */
3006 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
3007 		    offsetof(struct if_data, ifi_ipackets));
3008 		/*
3009 		 * The memset() clears to the bottm of if_data. In the area,
3010 		 * if_lastchange is included. Please be careful if new entry
3011 		 * will be added into if_data or rewite this.
3012 		 *
3013 		 * And also, update if_lastchnage.
3014 		 */
3015 		getnanotime(&ifp->if_lastchange);
3016 		break;
3017 	case SIOCSIFMTU:
3018 		ifr = data;
3019 		if (ifp->if_mtu == ifr->ifr_mtu)
3020 			break;
3021 		ifp->if_mtu = ifr->ifr_mtu;
3022 		/*
3023 		 * If the link MTU changed, do network layer specific procedure.
3024 		 */
3025 #ifdef INET6
3026 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3027 		if (in6_present)
3028 			nd6_setmtu(ifp);
3029 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3030 #endif
3031 		return ENETRESET;
3032 	default:
3033 		return ENOTTY;
3034 	}
3035 	return 0;
3036 }
3037 
3038 int
3039 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3040 {
3041 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3042 	struct ifaddr *ifa;
3043 	const struct sockaddr *any, *sa;
3044 	union {
3045 		struct sockaddr sa;
3046 		struct sockaddr_storage ss;
3047 	} u, v;
3048 	int s, error = 0;
3049 
3050 	switch (cmd) {
3051 	case SIOCSIFADDRPREF:
3052 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
3053 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
3054 		    NULL) != 0)
3055 			return EPERM;
3056 	case SIOCGIFADDRPREF:
3057 		break;
3058 	default:
3059 		return EOPNOTSUPP;
3060 	}
3061 
3062 	/* sanity checks */
3063 	if (data == NULL || ifp == NULL) {
3064 		panic("invalid argument to %s", __func__);
3065 		/*NOTREACHED*/
3066 	}
3067 
3068 	/* address must be specified on ADD and DELETE */
3069 	sa = sstocsa(&ifap->ifap_addr);
3070 	if (sa->sa_family != sofamily(so))
3071 		return EINVAL;
3072 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3073 		return EINVAL;
3074 
3075 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3076 
3077 	s = pserialize_read_enter();
3078 	IFADDR_READER_FOREACH(ifa, ifp) {
3079 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3080 			continue;
3081 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3082 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3083 			break;
3084 	}
3085 	if (ifa == NULL) {
3086 		error = EADDRNOTAVAIL;
3087 		goto out;
3088 	}
3089 
3090 	switch (cmd) {
3091 	case SIOCSIFADDRPREF:
3092 		ifa->ifa_preference = ifap->ifap_preference;
3093 		goto out;
3094 	case SIOCGIFADDRPREF:
3095 		/* fill in the if_laddrreq structure */
3096 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3097 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3098 		ifap->ifap_preference = ifa->ifa_preference;
3099 		goto out;
3100 	default:
3101 		error = EOPNOTSUPP;
3102 	}
3103 out:
3104 	pserialize_read_exit(s);
3105 	return error;
3106 }
3107 
3108 /*
3109  * Interface ioctls.
3110  */
3111 static int
3112 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3113 {
3114 	struct ifnet *ifp;
3115 	struct ifreq *ifr;
3116 	int error = 0;
3117 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
3118 	u_long ocmd = cmd;
3119 #endif
3120 	short oif_flags;
3121 #ifdef COMPAT_OIFREQ
3122 	struct ifreq ifrb;
3123 	struct oifreq *oifr = NULL;
3124 #endif
3125 	int r;
3126 	struct psref psref;
3127 	int bound;
3128 
3129 	switch (cmd) {
3130 	case SIOCGIFCONF:
3131 		return ifconf(cmd, data);
3132 	case SIOCINITIFADDR:
3133 		return EPERM;
3134 	default:
3135 		error = (*vec_compat_ifconf)(l, cmd, data);
3136 		if (error != ENOSYS)
3137 			return error;
3138 		error = (*vec_compat_ifdatareq)(l, cmd, data);
3139 		if (error != ENOSYS)
3140 			return error;
3141 		break;
3142 	}
3143 
3144 	ifr = data;
3145 #ifdef COMPAT_OIFREQ
3146 	if (vec_compat_cvtcmd) {
3147 		cmd = (*vec_compat_cvtcmd)(cmd);
3148 		if (cmd != ocmd) {
3149 			oifr = data;
3150 			data = ifr = &ifrb;
3151 			ifreqo2n(oifr, ifr);
3152 		}
3153 	}
3154 #endif
3155 
3156 	switch (cmd) {
3157 	case SIOCIFCREATE:
3158 	case SIOCIFDESTROY:
3159 		bound = curlwp_bind();
3160 		if (l != NULL) {
3161 			ifp = if_get(ifr->ifr_name, &psref);
3162 			error = kauth_authorize_network(l->l_cred,
3163 			    KAUTH_NETWORK_INTERFACE,
3164 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3165 			    (void *)cmd, NULL);
3166 			if (ifp != NULL)
3167 				if_put(ifp, &psref);
3168 			if (error != 0) {
3169 				curlwp_bindx(bound);
3170 				return error;
3171 			}
3172 		}
3173 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3174 		mutex_enter(&if_clone_mtx);
3175 		r = (cmd == SIOCIFCREATE) ?
3176 			if_clone_create(ifr->ifr_name) :
3177 			if_clone_destroy(ifr->ifr_name);
3178 		mutex_exit(&if_clone_mtx);
3179 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3180 		curlwp_bindx(bound);
3181 		return r;
3182 
3183 	case SIOCIFGCLONERS:
3184 		{
3185 			struct if_clonereq *req = (struct if_clonereq *)data;
3186 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3187 			    &req->ifcr_total);
3188 		}
3189 	}
3190 
3191 	bound = curlwp_bind();
3192 	ifp = if_get(ifr->ifr_name, &psref);
3193 	if (ifp == NULL) {
3194 		curlwp_bindx(bound);
3195 		return ENXIO;
3196 	}
3197 
3198 	switch (cmd) {
3199 	case SIOCALIFADDR:
3200 	case SIOCDLIFADDR:
3201 	case SIOCSIFADDRPREF:
3202 	case SIOCSIFFLAGS:
3203 	case SIOCSIFCAP:
3204 	case SIOCSIFMETRIC:
3205 	case SIOCZIFDATA:
3206 	case SIOCSIFMTU:
3207 	case SIOCSIFPHYADDR:
3208 	case SIOCDIFPHYADDR:
3209 #ifdef INET6
3210 	case SIOCSIFPHYADDR_IN6:
3211 #endif
3212 	case SIOCSLIFPHYADDR:
3213 	case SIOCADDMULTI:
3214 	case SIOCDELMULTI:
3215 	case SIOCSIFMEDIA:
3216 	case SIOCSDRVSPEC:
3217 	case SIOCG80211:
3218 	case SIOCS80211:
3219 	case SIOCS80211NWID:
3220 	case SIOCS80211NWKEY:
3221 	case SIOCS80211POWER:
3222 	case SIOCS80211BSSID:
3223 	case SIOCS80211CHANNEL:
3224 	case SIOCSLINKSTR:
3225 		if (l != NULL) {
3226 			error = kauth_authorize_network(l->l_cred,
3227 			    KAUTH_NETWORK_INTERFACE,
3228 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3229 			    (void *)cmd, NULL);
3230 			if (error != 0)
3231 				goto out;
3232 		}
3233 	}
3234 
3235 	oif_flags = ifp->if_flags;
3236 
3237 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3238 	IFNET_LOCK(ifp);
3239 
3240 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3241 	if (error != ENOTTY)
3242 		;
3243 	else if (so->so_proto == NULL)
3244 		error = EOPNOTSUPP;
3245 	else {
3246 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3247 #ifdef COMPAT_OSOCK
3248 		if (vec_compat_ifioctl != NULL)
3249 			error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l);
3250 		else
3251 #endif
3252 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3253 			    cmd, data, ifp);
3254 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3255 	}
3256 
3257 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3258 		if ((ifp->if_flags & IFF_UP) != 0) {
3259 			int s = splsoftnet();
3260 			if_up_locked(ifp);
3261 			splx(s);
3262 		}
3263 	}
3264 #ifdef COMPAT_OIFREQ
3265 	if (cmd != ocmd)
3266 		ifreqn2o(oifr, ifr);
3267 #endif
3268 
3269 	IFNET_UNLOCK(ifp);
3270 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3271 out:
3272 	if_put(ifp, &psref);
3273 	curlwp_bindx(bound);
3274 	return error;
3275 }
3276 
3277 /*
3278  * Return interface configuration
3279  * of system.  List may be used
3280  * in later ioctl's (above) to get
3281  * other information.
3282  *
3283  * Each record is a struct ifreq.  Before the addition of
3284  * sockaddr_storage, the API rule was that sockaddr flavors that did
3285  * not fit would extend beyond the struct ifreq, with the next struct
3286  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3287  * union in struct ifreq includes struct sockaddr_storage, every kind
3288  * of sockaddr must fit.  Thus, there are no longer any overlength
3289  * records.
3290  *
3291  * Records are added to the user buffer if they fit, and ifc_len is
3292  * adjusted to the length that was written.  Thus, the user is only
3293  * assured of getting the complete list if ifc_len on return is at
3294  * least sizeof(struct ifreq) less than it was on entry.
3295  *
3296  * If the user buffer pointer is NULL, this routine copies no data and
3297  * returns the amount of space that would be needed.
3298  *
3299  * Invariants:
3300  * ifrp points to the next part of the user's buffer to be used.  If
3301  * ifrp != NULL, space holds the number of bytes remaining that we may
3302  * write at ifrp.  Otherwise, space holds the number of bytes that
3303  * would have been written had there been adequate space.
3304  */
3305 /*ARGSUSED*/
3306 static int
3307 ifconf(u_long cmd, void *data)
3308 {
3309 	struct ifconf *ifc = (struct ifconf *)data;
3310 	struct ifnet *ifp;
3311 	struct ifaddr *ifa;
3312 	struct ifreq ifr, *ifrp = NULL;
3313 	int space = 0, error = 0;
3314 	const int sz = (int)sizeof(struct ifreq);
3315 	const bool docopy = ifc->ifc_req != NULL;
3316 	int s;
3317 	int bound;
3318 	struct psref psref;
3319 
3320 	if (docopy) {
3321 		space = ifc->ifc_len;
3322 		ifrp = ifc->ifc_req;
3323 	}
3324 
3325 	bound = curlwp_bind();
3326 	s = pserialize_read_enter();
3327 	IFNET_READER_FOREACH(ifp) {
3328 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3329 		pserialize_read_exit(s);
3330 
3331 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3332 		    sizeof(ifr.ifr_name));
3333 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3334 			error = ENAMETOOLONG;
3335 			goto release_exit;
3336 		}
3337 		if (IFADDR_READER_EMPTY(ifp)) {
3338 			/* Interface with no addresses - send zero sockaddr. */
3339 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3340 			if (!docopy) {
3341 				space += sz;
3342 				goto next;
3343 			}
3344 			if (space >= sz) {
3345 				error = copyout(&ifr, ifrp, sz);
3346 				if (error != 0)
3347 					goto release_exit;
3348 				ifrp++;
3349 				space -= sz;
3350 			}
3351 		}
3352 
3353 		s = pserialize_read_enter();
3354 		IFADDR_READER_FOREACH(ifa, ifp) {
3355 			struct sockaddr *sa = ifa->ifa_addr;
3356 			/* all sockaddrs must fit in sockaddr_storage */
3357 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3358 
3359 			if (!docopy) {
3360 				space += sz;
3361 				continue;
3362 			}
3363 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3364 			pserialize_read_exit(s);
3365 
3366 			if (space >= sz) {
3367 				error = copyout(&ifr, ifrp, sz);
3368 				if (error != 0)
3369 					goto release_exit;
3370 				ifrp++; space -= sz;
3371 			}
3372 			s = pserialize_read_enter();
3373 		}
3374 		pserialize_read_exit(s);
3375 
3376         next:
3377 		s = pserialize_read_enter();
3378 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3379 	}
3380 	pserialize_read_exit(s);
3381 	curlwp_bindx(bound);
3382 
3383 	if (docopy) {
3384 		KASSERT(0 <= space && space <= ifc->ifc_len);
3385 		ifc->ifc_len -= space;
3386 	} else {
3387 		KASSERT(space >= 0);
3388 		ifc->ifc_len = space;
3389 	}
3390 	return (0);
3391 
3392 release_exit:
3393 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3394 	curlwp_bindx(bound);
3395 	return error;
3396 }
3397 
3398 int
3399 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3400 {
3401 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3402 #ifdef COMPAT_OIFREQ
3403 	struct ifreq ifrb;
3404 	struct oifreq *oifr = NULL;
3405 	u_long ocmd = cmd;
3406 
3407 	if (vec_compat_cvtcmd) {
3408 		    cmd = (*vec_compat_cvtcmd)(cmd);
3409 		    if (cmd != ocmd) {
3410 			    oifr = (struct oifreq *)(void *)ifr;
3411 			    ifr = &ifrb;
3412 			    ifreqo2n(oifr, ifr);
3413 			    len = sizeof(oifr->ifr_addr);
3414 		    }
3415 	}
3416 #endif
3417 	if (len < sa->sa_len)
3418 		return EFBIG;
3419 
3420 	memset(&ifr->ifr_addr, 0, len);
3421 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3422 
3423 #ifdef COMPAT_OIFREQ
3424 	if (cmd != ocmd)
3425 		ifreqn2o(oifr, ifr);
3426 #endif
3427 	return 0;
3428 }
3429 
3430 /*
3431  * wrapper function for the drivers which doesn't have if_transmit().
3432  */
3433 static int
3434 if_transmit(struct ifnet *ifp, struct mbuf *m)
3435 {
3436 	int s, error;
3437 	size_t pktlen = m->m_pkthdr.len;
3438 	bool mcast = (m->m_flags & M_MCAST) != 0;
3439 
3440 	s = splnet();
3441 
3442 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3443 	if (error != 0) {
3444 		/* mbuf is already freed */
3445 		goto out;
3446 	}
3447 
3448 	ifp->if_obytes += pktlen;
3449 	if (mcast)
3450 		ifp->if_omcasts++;
3451 
3452 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3453 		if_start_lock(ifp);
3454 out:
3455 	splx(s);
3456 
3457 	return error;
3458 }
3459 
3460 int
3461 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3462 {
3463 	int error;
3464 
3465 #ifdef ALTQ
3466 	KERNEL_LOCK(1, NULL);
3467 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3468 		error = if_transmit(ifp, m);
3469 		KERNEL_UNLOCK_ONE(NULL);
3470 	} else {
3471 		KERNEL_UNLOCK_ONE(NULL);
3472 		error = (*ifp->if_transmit)(ifp, m);
3473 		/* mbuf is alredy freed */
3474 	}
3475 #else /* !ALTQ */
3476 	error = (*ifp->if_transmit)(ifp, m);
3477 	/* mbuf is alredy freed */
3478 #endif /* !ALTQ */
3479 
3480 	return error;
3481 }
3482 
3483 /*
3484  * Queue message on interface, and start output if interface
3485  * not yet active.
3486  */
3487 int
3488 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3489 {
3490 
3491 	return if_transmit_lock(ifp, m);
3492 }
3493 
3494 /*
3495  * Queue message on interface, possibly using a second fast queue
3496  */
3497 int
3498 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3499 {
3500 	int error = 0;
3501 
3502 	if (ifq != NULL
3503 #ifdef ALTQ
3504 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3505 #endif
3506 	    ) {
3507 		if (IF_QFULL(ifq)) {
3508 			IF_DROP(&ifp->if_snd);
3509 			m_freem(m);
3510 			if (error == 0)
3511 				error = ENOBUFS;
3512 		} else
3513 			IF_ENQUEUE(ifq, m);
3514 	} else
3515 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3516 	if (error != 0) {
3517 		++ifp->if_oerrors;
3518 		return error;
3519 	}
3520 	return 0;
3521 }
3522 
3523 int
3524 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3525 {
3526 	int rc;
3527 
3528 	KASSERT(IFNET_LOCKED(ifp));
3529 	if (ifp->if_initaddr != NULL)
3530 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3531 	else if (src ||
3532 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3533 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3534 
3535 	return rc;
3536 }
3537 
3538 int
3539 if_do_dad(struct ifnet *ifp)
3540 {
3541 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3542 		return 0;
3543 
3544 	switch (ifp->if_type) {
3545 	case IFT_FAITH:
3546 		/*
3547 		 * These interfaces do not have the IFF_LOOPBACK flag,
3548 		 * but loop packets back.  We do not have to do DAD on such
3549 		 * interfaces.  We should even omit it, because loop-backed
3550 		 * responses would confuse the DAD procedure.
3551 		 */
3552 		return 0;
3553 	default:
3554 		/*
3555 		 * Our DAD routine requires the interface up and running.
3556 		 * However, some interfaces can be up before the RUNNING
3557 		 * status.  Additionaly, users may try to assign addresses
3558 		 * before the interface becomes up (or running).
3559 		 * We simply skip DAD in such a case as a work around.
3560 		 * XXX: we should rather mark "tentative" on such addresses,
3561 		 * and do DAD after the interface becomes ready.
3562 		 */
3563 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
3564 		    (IFF_UP|IFF_RUNNING))
3565 			return 0;
3566 
3567 		return 1;
3568 	}
3569 }
3570 
3571 int
3572 if_flags_set(ifnet_t *ifp, const short flags)
3573 {
3574 	int rc;
3575 
3576 	KASSERT(IFNET_LOCKED(ifp));
3577 
3578 	if (ifp->if_setflags != NULL)
3579 		rc = (*ifp->if_setflags)(ifp, flags);
3580 	else {
3581 		short cantflags, chgdflags;
3582 		struct ifreq ifr;
3583 
3584 		chgdflags = ifp->if_flags ^ flags;
3585 		cantflags = chgdflags & IFF_CANTCHANGE;
3586 
3587 		if (cantflags != 0)
3588 			ifp->if_flags ^= cantflags;
3589 
3590                 /* Traditionally, we do not call if_ioctl after
3591                  * setting/clearing only IFF_PROMISC if the interface
3592                  * isn't IFF_UP.  Uphold that tradition.
3593 		 */
3594 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3595 			return 0;
3596 
3597 		memset(&ifr, 0, sizeof(ifr));
3598 
3599 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3600 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3601 
3602 		if (rc != 0 && cantflags != 0)
3603 			ifp->if_flags ^= cantflags;
3604 	}
3605 
3606 	return rc;
3607 }
3608 
3609 int
3610 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3611 {
3612 	int rc;
3613 	struct ifreq ifr;
3614 
3615 	/* There remain some paths that don't hold IFNET_LOCK yet */
3616 #ifdef NET_MPSAFE
3617 	/* CARP and MROUTING still don't deal with the lock yet */
3618 #if (!defined(NCARP) || (NCARP == 0)) && !defined(MROUTING)
3619 	KASSERT(IFNET_LOCKED(ifp));
3620 #endif
3621 #endif
3622 	if (ifp->if_mcastop != NULL)
3623 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3624 	else {
3625 		ifreq_setaddr(cmd, &ifr, sa);
3626 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3627 	}
3628 
3629 	return rc;
3630 }
3631 
3632 static void
3633 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3634     struct ifaltq *ifq)
3635 {
3636 	const struct sysctlnode *cnode, *rnode;
3637 
3638 	if (sysctl_createv(clog, 0, NULL, &rnode,
3639 		       CTLFLAG_PERMANENT,
3640 		       CTLTYPE_NODE, "interfaces",
3641 		       SYSCTL_DESCR("Per-interface controls"),
3642 		       NULL, 0, NULL, 0,
3643 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3644 		goto bad;
3645 
3646 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3647 		       CTLFLAG_PERMANENT,
3648 		       CTLTYPE_NODE, ifname,
3649 		       SYSCTL_DESCR("Interface controls"),
3650 		       NULL, 0, NULL, 0,
3651 		       CTL_CREATE, CTL_EOL) != 0)
3652 		goto bad;
3653 
3654 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3655 		       CTLFLAG_PERMANENT,
3656 		       CTLTYPE_NODE, "sndq",
3657 		       SYSCTL_DESCR("Interface output queue controls"),
3658 		       NULL, 0, NULL, 0,
3659 		       CTL_CREATE, CTL_EOL) != 0)
3660 		goto bad;
3661 
3662 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3663 		       CTLFLAG_PERMANENT,
3664 		       CTLTYPE_INT, "len",
3665 		       SYSCTL_DESCR("Current output queue length"),
3666 		       NULL, 0, &ifq->ifq_len, 0,
3667 		       CTL_CREATE, CTL_EOL) != 0)
3668 		goto bad;
3669 
3670 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3671 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3672 		       CTLTYPE_INT, "maxlen",
3673 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3674 		       NULL, 0, &ifq->ifq_maxlen, 0,
3675 		       CTL_CREATE, CTL_EOL) != 0)
3676 		goto bad;
3677 
3678 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3679 		       CTLFLAG_PERMANENT,
3680 		       CTLTYPE_INT, "drops",
3681 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3682 		       NULL, 0, &ifq->ifq_drops, 0,
3683 		       CTL_CREATE, CTL_EOL) != 0)
3684 		goto bad;
3685 
3686 	return;
3687 bad:
3688 	printf("%s: could not attach sysctl nodes\n", ifname);
3689 	return;
3690 }
3691 
3692 #if defined(INET) || defined(INET6)
3693 
3694 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3695 	static int							\
3696 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3697 	{								\
3698 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3699 	}
3700 
3701 #if defined(INET)
3702 static int
3703 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3704 {
3705 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3706 }
3707 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3708 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3709 #endif
3710 
3711 #if defined(INET6)
3712 static int
3713 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3714 {
3715 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3716 }
3717 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3718 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3719 #endif
3720 
3721 static void
3722 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3723 {
3724 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3725 	const char *pfname = NULL, *ipname = NULL;
3726 	int ipn = 0, qid = 0;
3727 
3728 	switch (pf) {
3729 #if defined(INET)
3730 	case PF_INET:
3731 		len_func = sysctl_net_ip_pktq_items;
3732 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3733 		drops_func = sysctl_net_ip_pktq_drops;
3734 		pfname = "inet", ipn = IPPROTO_IP;
3735 		ipname = "ip", qid = IPCTL_IFQ;
3736 		break;
3737 #endif
3738 #if defined(INET6)
3739 	case PF_INET6:
3740 		len_func = sysctl_net_ip6_pktq_items;
3741 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3742 		drops_func = sysctl_net_ip6_pktq_drops;
3743 		pfname = "inet6", ipn = IPPROTO_IPV6;
3744 		ipname = "ip6", qid = IPV6CTL_IFQ;
3745 		break;
3746 #endif
3747 	default:
3748 		KASSERT(false);
3749 	}
3750 
3751 	sysctl_createv(clog, 0, NULL, NULL,
3752 		       CTLFLAG_PERMANENT,
3753 		       CTLTYPE_NODE, pfname, NULL,
3754 		       NULL, 0, NULL, 0,
3755 		       CTL_NET, pf, CTL_EOL);
3756 	sysctl_createv(clog, 0, NULL, NULL,
3757 		       CTLFLAG_PERMANENT,
3758 		       CTLTYPE_NODE, ipname, NULL,
3759 		       NULL, 0, NULL, 0,
3760 		       CTL_NET, pf, ipn, CTL_EOL);
3761 	sysctl_createv(clog, 0, NULL, NULL,
3762 		       CTLFLAG_PERMANENT,
3763 		       CTLTYPE_NODE, "ifq",
3764 		       SYSCTL_DESCR("Protocol input queue controls"),
3765 		       NULL, 0, NULL, 0,
3766 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3767 
3768 	sysctl_createv(clog, 0, NULL, NULL,
3769 		       CTLFLAG_PERMANENT,
3770 		       CTLTYPE_INT, "len",
3771 		       SYSCTL_DESCR("Current input queue length"),
3772 		       len_func, 0, NULL, 0,
3773 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3774 	sysctl_createv(clog, 0, NULL, NULL,
3775 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3776 		       CTLTYPE_INT, "maxlen",
3777 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3778 		       maxlen_func, 0, NULL, 0,
3779 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3780 	sysctl_createv(clog, 0, NULL, NULL,
3781 		       CTLFLAG_PERMANENT,
3782 		       CTLTYPE_INT, "drops",
3783 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3784 		       drops_func, 0, NULL, 0,
3785 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3786 }
3787 #endif /* INET || INET6 */
3788 
3789 static int
3790 if_sdl_sysctl(SYSCTLFN_ARGS)
3791 {
3792 	struct ifnet *ifp;
3793 	const struct sockaddr_dl *sdl;
3794 	struct psref psref;
3795 	int error = 0;
3796 	int bound;
3797 
3798 	if (namelen != 1)
3799 		return EINVAL;
3800 
3801 	bound = curlwp_bind();
3802 	ifp = if_get_byindex(name[0], &psref);
3803 	if (ifp == NULL) {
3804 		error = ENODEV;
3805 		goto out0;
3806 	}
3807 
3808 	sdl = ifp->if_sadl;
3809 	if (sdl == NULL) {
3810 		*oldlenp = 0;
3811 		goto out1;
3812 	}
3813 
3814 	if (oldp == NULL) {
3815 		*oldlenp = sdl->sdl_alen;
3816 		goto out1;
3817 	}
3818 
3819 	if (*oldlenp >= sdl->sdl_alen)
3820 		*oldlenp = sdl->sdl_alen;
3821 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3822 out1:
3823 	if_put(ifp, &psref);
3824 out0:
3825 	curlwp_bindx(bound);
3826 	return error;
3827 }
3828 
3829 static void
3830 if_sysctl_setup(struct sysctllog **clog)
3831 {
3832 	const struct sysctlnode *rnode = NULL;
3833 
3834 	sysctl_createv(clog, 0, NULL, &rnode,
3835 		       CTLFLAG_PERMANENT,
3836 		       CTLTYPE_NODE, "sdl",
3837 		       SYSCTL_DESCR("Get active link-layer address"),
3838 		       if_sdl_sysctl, 0, NULL, 0,
3839 		       CTL_NET, CTL_CREATE, CTL_EOL);
3840 
3841 #if defined(INET)
3842 	sysctl_net_pktq_setup(NULL, PF_INET);
3843 #endif
3844 #ifdef INET6
3845 	if (in6_present)
3846 		sysctl_net_pktq_setup(NULL, PF_INET6);
3847 #endif
3848 }
3849