xref: /netbsd-src/sys/net/if.c (revision 16dce51364ebe8aeafbae46bc5aa167b8115bc45)
1 /*	$NetBSD: if.c,v 1.420 2018/04/12 18:44:59 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 1.420 2018/04/12 18:44:59 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 *);
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_sysctl_setup(NULL);
283 
284 #if (defined(INET) || defined(INET6))
285 	encapinit();
286 #endif
287 
288 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
289 	    if_listener_cb, NULL);
290 
291 	/* interfaces are available, inform socket code */
292 	ifioctl = doifioctl;
293 }
294 
295 /*
296  * XXX Initialization before configure().
297  * XXX hack to get pfil_add_hook working in autoconf.
298  */
299 void
300 ifinit1(void)
301 {
302 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
303 
304 	TAILQ_INIT(&ifnet_list);
305 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
306 	ifnet_psz = pserialize_create();
307 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
308 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
309 	PSLIST_INIT(&ifnet_pslist);
310 
311 	if_indexlim = 8;
312 
313 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
314 	KASSERT(if_pfil != NULL);
315 
316 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
317 	etherinit();
318 #endif
319 }
320 
321 ifnet_t *
322 if_alloc(u_char type)
323 {
324 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
325 }
326 
327 void
328 if_free(ifnet_t *ifp)
329 {
330 	kmem_free(ifp, sizeof(ifnet_t));
331 }
332 
333 void
334 if_initname(struct ifnet *ifp, const char *name, int unit)
335 {
336 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
337 	    "%s%d", name, unit);
338 }
339 
340 /*
341  * Null routines used while an interface is going away.  These routines
342  * just return an error.
343  */
344 
345 int
346 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
347     const struct sockaddr *so, const struct rtentry *rt)
348 {
349 
350 	return ENXIO;
351 }
352 
353 void
354 if_nullinput(struct ifnet *ifp, struct mbuf *m)
355 {
356 
357 	/* Nothing. */
358 }
359 
360 void
361 if_nullstart(struct ifnet *ifp)
362 {
363 
364 	/* Nothing. */
365 }
366 
367 int
368 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
369 {
370 
371 	m_freem(m);
372 	return ENXIO;
373 }
374 
375 int
376 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
377 {
378 
379 	return ENXIO;
380 }
381 
382 int
383 if_nullinit(struct ifnet *ifp)
384 {
385 
386 	return ENXIO;
387 }
388 
389 void
390 if_nullstop(struct ifnet *ifp, int disable)
391 {
392 
393 	/* Nothing. */
394 }
395 
396 void
397 if_nullslowtimo(struct ifnet *ifp)
398 {
399 
400 	/* Nothing. */
401 }
402 
403 void
404 if_nulldrain(struct ifnet *ifp)
405 {
406 
407 	/* Nothing. */
408 }
409 
410 void
411 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
412 {
413 	struct ifaddr *ifa;
414 	struct sockaddr_dl *sdl;
415 
416 	ifp->if_addrlen = addrlen;
417 	if_alloc_sadl(ifp);
418 	ifa = ifp->if_dl;
419 	sdl = satosdl(ifa->ifa_addr);
420 
421 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
422 	if (factory) {
423 		ifp->if_hwdl = ifp->if_dl;
424 		ifaref(ifp->if_hwdl);
425 	}
426 	/* TBD routing socket */
427 }
428 
429 struct ifaddr *
430 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
431 {
432 	unsigned socksize, ifasize;
433 	int addrlen, namelen;
434 	struct sockaddr_dl *mask, *sdl;
435 	struct ifaddr *ifa;
436 
437 	namelen = strlen(ifp->if_xname);
438 	addrlen = ifp->if_addrlen;
439 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
440 	ifasize = sizeof(*ifa) + 2 * socksize;
441 	ifa = malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
442 
443 	sdl = (struct sockaddr_dl *)(ifa + 1);
444 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
445 
446 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
447 	    ifp->if_xname, namelen, NULL, addrlen);
448 	mask->sdl_family = AF_LINK;
449 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
450 	memset(&mask->sdl_data[0], 0xff, namelen);
451 	ifa->ifa_rtrequest = link_rtrequest;
452 	ifa->ifa_addr = (struct sockaddr *)sdl;
453 	ifa->ifa_netmask = (struct sockaddr *)mask;
454 	ifa_psref_init(ifa);
455 
456 	*sdlp = sdl;
457 
458 	return ifa;
459 }
460 
461 static void
462 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
463 {
464 	const struct sockaddr_dl *sdl;
465 
466 	ifp->if_dl = ifa;
467 	ifaref(ifa);
468 	sdl = satosdl(ifa->ifa_addr);
469 	ifp->if_sadl = sdl;
470 }
471 
472 /*
473  * Allocate the link level name for the specified interface.  This
474  * is an attachment helper.  It must be called after ifp->if_addrlen
475  * is initialized, which may not be the case when if_attach() is
476  * called.
477  */
478 void
479 if_alloc_sadl(struct ifnet *ifp)
480 {
481 	struct ifaddr *ifa;
482 	const struct sockaddr_dl *sdl;
483 
484 	/*
485 	 * If the interface already has a link name, release it
486 	 * now.  This is useful for interfaces that can change
487 	 * link types, and thus switch link names often.
488 	 */
489 	if (ifp->if_sadl != NULL)
490 		if_free_sadl(ifp);
491 
492 	ifa = if_dl_create(ifp, &sdl);
493 
494 	ifa_insert(ifp, ifa);
495 	if_sadl_setrefs(ifp, ifa);
496 }
497 
498 static void
499 if_deactivate_sadl(struct ifnet *ifp)
500 {
501 	struct ifaddr *ifa;
502 
503 	KASSERT(ifp->if_dl != NULL);
504 
505 	ifa = ifp->if_dl;
506 
507 	ifp->if_sadl = NULL;
508 
509 	ifp->if_dl = NULL;
510 	ifafree(ifa);
511 }
512 
513 static void
514 if_replace_sadl(struct ifnet *ifp, struct ifaddr *ifa)
515 {
516 	struct ifaddr *old;
517 
518 	KASSERT(ifp->if_dl != NULL);
519 
520 	old = ifp->if_dl;
521 
522 	ifaref(ifa);
523 	/* XXX Update if_dl and if_sadl atomically */
524 	ifp->if_dl = ifa;
525 	ifp->if_sadl = satosdl(ifa->ifa_addr);
526 
527 	ifafree(old);
528 }
529 
530 void
531 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
532     const struct sockaddr_dl *sdl)
533 {
534 	int s, ss;
535 	struct ifaddr *ifa;
536 	int bound = curlwp_bind();
537 
538 	KASSERT(ifa_held(ifa0));
539 
540 	s = splsoftnet();
541 
542 	if_replace_sadl(ifp, ifa0);
543 
544 	ss = pserialize_read_enter();
545 	IFADDR_READER_FOREACH(ifa, ifp) {
546 		struct psref psref;
547 		ifa_acquire(ifa, &psref);
548 		pserialize_read_exit(ss);
549 
550 		rtinit(ifa, RTM_LLINFO_UPD, 0);
551 
552 		ss = pserialize_read_enter();
553 		ifa_release(ifa, &psref);
554 	}
555 	pserialize_read_exit(ss);
556 
557 	splx(s);
558 	curlwp_bindx(bound);
559 }
560 
561 /*
562  * Free the link level name for the specified interface.  This is
563  * a detach helper.  This is called from if_detach().
564  */
565 static void
566 if_free_sadl(struct ifnet *ifp)
567 {
568 	struct ifaddr *ifa;
569 	int s;
570 
571 	ifa = ifp->if_dl;
572 	if (ifa == NULL) {
573 		KASSERT(ifp->if_sadl == NULL);
574 		return;
575 	}
576 
577 	KASSERT(ifp->if_sadl != NULL);
578 
579 	s = splsoftnet();
580 	rtinit(ifa, RTM_DELETE, 0);
581 	ifa_remove(ifp, ifa);
582 	if_deactivate_sadl(ifp);
583 	if (ifp->if_hwdl == ifa) {
584 		ifafree(ifa);
585 		ifp->if_hwdl = NULL;
586 	}
587 	splx(s);
588 }
589 
590 static void
591 if_getindex(ifnet_t *ifp)
592 {
593 	bool hitlimit = false;
594 
595 	ifp->if_index_gen = index_gen++;
596 
597 	ifp->if_index = if_index;
598 	if (ifindex2ifnet == NULL) {
599 		if_index++;
600 		goto skip;
601 	}
602 	while (if_byindex(ifp->if_index)) {
603 		/*
604 		 * If we hit USHRT_MAX, we skip back to 0 since
605 		 * there are a number of places where the value
606 		 * of if_index or if_index itself is compared
607 		 * to or stored in an unsigned short.  By
608 		 * jumping back, we won't botch those assignments
609 		 * or comparisons.
610 		 */
611 		if (++if_index == 0) {
612 			if_index = 1;
613 		} else if (if_index == USHRT_MAX) {
614 			/*
615 			 * However, if we have to jump back to
616 			 * zero *twice* without finding an empty
617 			 * slot in ifindex2ifnet[], then there
618 			 * there are too many (>65535) interfaces.
619 			 */
620 			if (hitlimit) {
621 				panic("too many interfaces");
622 			}
623 			hitlimit = true;
624 			if_index = 1;
625 		}
626 		ifp->if_index = if_index;
627 	}
628 skip:
629 	/*
630 	 * ifindex2ifnet is indexed by if_index. Since if_index will
631 	 * grow dynamically, it should grow too.
632 	 */
633 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
634 		size_t m, n, oldlim;
635 		void *q;
636 
637 		oldlim = if_indexlim;
638 		while (ifp->if_index >= if_indexlim)
639 			if_indexlim <<= 1;
640 
641 		/* grow ifindex2ifnet */
642 		m = oldlim * sizeof(struct ifnet *);
643 		n = if_indexlim * sizeof(struct ifnet *);
644 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
645 		if (ifindex2ifnet != NULL) {
646 			memcpy(q, ifindex2ifnet, m);
647 			free(ifindex2ifnet, M_IFADDR);
648 		}
649 		ifindex2ifnet = (struct ifnet **)q;
650 	}
651 	ifindex2ifnet[ifp->if_index] = ifp;
652 }
653 
654 /*
655  * Initialize an interface and assign an index for it.
656  *
657  * It must be called prior to a device specific attach routine
658  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
659  * and be followed by if_register:
660  *
661  *     if_initialize(ifp);
662  *     ether_ifattach(ifp, enaddr);
663  *     if_register(ifp);
664  */
665 int
666 if_initialize(ifnet_t *ifp)
667 {
668 	int rv = 0;
669 
670 	KASSERT(if_indexlim > 0);
671 	TAILQ_INIT(&ifp->if_addrlist);
672 
673 	/*
674 	 * Link level name is allocated later by a separate call to
675 	 * if_alloc_sadl().
676 	 */
677 
678 	if (ifp->if_snd.ifq_maxlen == 0)
679 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
680 
681 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
682 
683 	ifp->if_link_state = LINK_STATE_UNKNOWN;
684 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
685 
686 	ifp->if_capenable = 0;
687 	ifp->if_csum_flags_tx = 0;
688 	ifp->if_csum_flags_rx = 0;
689 
690 #ifdef ALTQ
691 	ifp->if_snd.altq_type = 0;
692 	ifp->if_snd.altq_disc = NULL;
693 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
694 	ifp->if_snd.altq_tbr  = NULL;
695 	ifp->if_snd.altq_ifp  = ifp;
696 #endif
697 
698 	IFQ_LOCK_INIT(&ifp->if_snd);
699 
700 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
701 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_ATTACH, ifp);
702 
703 	IF_AFDATA_LOCK_INIT(ifp);
704 
705 	if (if_is_link_state_changeable(ifp)) {
706 		u_int flags = SOFTINT_NET;
707 		flags |= ISSET(ifp->if_extflags, IFEF_MPSAFE) ?
708 		    SOFTINT_MPSAFE : 0;
709 		ifp->if_link_si = softint_establish(flags,
710 		    if_link_state_change_si, ifp);
711 		if (ifp->if_link_si == NULL) {
712 			rv = ENOMEM;
713 			goto fail;
714 		}
715 	}
716 
717 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
718 	PSLIST_INIT(&ifp->if_addr_pslist);
719 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
720 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
721 	LIST_INIT(&ifp->if_multiaddrs);
722 
723 	IFNET_GLOBAL_LOCK();
724 	if_getindex(ifp);
725 	IFNET_GLOBAL_UNLOCK();
726 
727 	return 0;
728 
729 fail:
730 	IF_AFDATA_LOCK_DESTROY(ifp);
731 
732 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
733 	(void)pfil_head_destroy(ifp->if_pfil);
734 
735 	IFQ_LOCK_DESTROY(&ifp->if_snd);
736 
737 	return rv;
738 }
739 
740 /*
741  * Register an interface to the list of "active" interfaces.
742  */
743 void
744 if_register(ifnet_t *ifp)
745 {
746 	/*
747 	 * If the driver has not supplied its own if_ioctl, then
748 	 * supply the default.
749 	 */
750 	if (ifp->if_ioctl == NULL)
751 		ifp->if_ioctl = ifioctl_common;
752 
753 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
754 
755 	if (!STAILQ_EMPTY(&domains))
756 		if_attachdomain1(ifp);
757 
758 	/* Announce the interface. */
759 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
760 
761 	if (ifp->if_slowtimo != NULL) {
762 		ifp->if_slowtimo_ch =
763 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
764 		callout_init(ifp->if_slowtimo_ch, 0);
765 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
766 		if_slowtimo(ifp);
767 	}
768 
769 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
770 		ifp->if_transmit = if_transmit;
771 
772 	IFNET_GLOBAL_LOCK();
773 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
774 	IFNET_WRITER_INSERT_TAIL(ifp);
775 	IFNET_GLOBAL_UNLOCK();
776 }
777 
778 /*
779  * The if_percpuq framework
780  *
781  * It allows network device drivers to execute the network stack
782  * in softint (so called softint-based if_input). It utilizes
783  * softint and percpu ifqueue. It doesn't distribute any packets
784  * between CPUs, unlike pktqueue(9).
785  *
786  * Currently we support two options for device drivers to apply the framework:
787  * - Use it implicitly with less changes
788  *   - If you use if_attach in driver's _attach function and if_input in
789  *     driver's Rx interrupt handler, a packet is queued and a softint handles
790  *     the packet implicitly
791  * - Use it explicitly in each driver (recommended)
792  *   - You can use if_percpuq_* directly in your driver
793  *   - In this case, you need to allocate struct if_percpuq in driver's softc
794  *   - See wm(4) as a reference implementation
795  */
796 
797 static void
798 if_percpuq_softint(void *arg)
799 {
800 	struct if_percpuq *ipq = arg;
801 	struct ifnet *ifp = ipq->ipq_ifp;
802 	struct mbuf *m;
803 
804 	while ((m = if_percpuq_dequeue(ipq)) != NULL) {
805 		ifp->if_ipackets++;
806 		bpf_mtap(ifp, m);
807 
808 		ifp->_if_input(ifp, m);
809 	}
810 }
811 
812 static void
813 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
814 {
815 	struct ifqueue *const ifq = p;
816 
817 	memset(ifq, 0, sizeof(*ifq));
818 	ifq->ifq_maxlen = IFQ_MAXLEN;
819 }
820 
821 struct if_percpuq *
822 if_percpuq_create(struct ifnet *ifp)
823 {
824 	struct if_percpuq *ipq;
825 
826 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
827 	ipq->ipq_ifp = ifp;
828 	ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
829 	    if_percpuq_softint, ipq);
830 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
831 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
832 
833 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
834 
835 	return ipq;
836 }
837 
838 static struct mbuf *
839 if_percpuq_dequeue(struct if_percpuq *ipq)
840 {
841 	struct mbuf *m;
842 	struct ifqueue *ifq;
843 	int s;
844 
845 	s = splnet();
846 	ifq = percpu_getref(ipq->ipq_ifqs);
847 	IF_DEQUEUE(ifq, m);
848 	percpu_putref(ipq->ipq_ifqs);
849 	splx(s);
850 
851 	return m;
852 }
853 
854 static void
855 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
856 {
857 	struct ifqueue *const ifq = p;
858 
859 	IF_PURGE(ifq);
860 }
861 
862 void
863 if_percpuq_destroy(struct if_percpuq *ipq)
864 {
865 
866 	/* if_detach may already destroy it */
867 	if (ipq == NULL)
868 		return;
869 
870 	softint_disestablish(ipq->ipq_si);
871 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
872 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
873 	kmem_free(ipq, sizeof(*ipq));
874 }
875 
876 void
877 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
878 {
879 	struct ifqueue *ifq;
880 	int s;
881 
882 	KASSERT(ipq != NULL);
883 
884 	s = splnet();
885 	ifq = percpu_getref(ipq->ipq_ifqs);
886 	if (IF_QFULL(ifq)) {
887 		IF_DROP(ifq);
888 		percpu_putref(ipq->ipq_ifqs);
889 		m_freem(m);
890 		goto out;
891 	}
892 	IF_ENQUEUE(ifq, m);
893 	percpu_putref(ipq->ipq_ifqs);
894 
895 	softint_schedule(ipq->ipq_si);
896 out:
897 	splx(s);
898 }
899 
900 static void
901 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
902 {
903 	struct ifqueue *const ifq = p;
904 	int *sum = arg;
905 
906 	*sum += ifq->ifq_drops;
907 }
908 
909 static int
910 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
911 {
912 	struct sysctlnode node;
913 	struct if_percpuq *ipq;
914 	int sum = 0;
915 	int error;
916 
917 	node = *rnode;
918 	ipq = node.sysctl_data;
919 
920 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
921 
922 	node.sysctl_data = &sum;
923 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
924 	if (error != 0 || newp == NULL)
925 		return error;
926 
927 	return 0;
928 }
929 
930 static void
931 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
932     struct if_percpuq *ipq)
933 {
934 	const struct sysctlnode *cnode, *rnode;
935 
936 	if (sysctl_createv(clog, 0, NULL, &rnode,
937 		       CTLFLAG_PERMANENT,
938 		       CTLTYPE_NODE, "interfaces",
939 		       SYSCTL_DESCR("Per-interface controls"),
940 		       NULL, 0, NULL, 0,
941 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
942 		goto bad;
943 
944 	if (sysctl_createv(clog, 0, &rnode, &rnode,
945 		       CTLFLAG_PERMANENT,
946 		       CTLTYPE_NODE, ifname,
947 		       SYSCTL_DESCR("Interface controls"),
948 		       NULL, 0, NULL, 0,
949 		       CTL_CREATE, CTL_EOL) != 0)
950 		goto bad;
951 
952 	if (sysctl_createv(clog, 0, &rnode, &rnode,
953 		       CTLFLAG_PERMANENT,
954 		       CTLTYPE_NODE, "rcvq",
955 		       SYSCTL_DESCR("Interface input queue controls"),
956 		       NULL, 0, NULL, 0,
957 		       CTL_CREATE, CTL_EOL) != 0)
958 		goto bad;
959 
960 #ifdef NOTYET
961 	/* XXX Should show each per-CPU queue length? */
962 	if (sysctl_createv(clog, 0, &rnode, &rnode,
963 		       CTLFLAG_PERMANENT,
964 		       CTLTYPE_INT, "len",
965 		       SYSCTL_DESCR("Current input queue length"),
966 		       sysctl_percpuq_len, 0, NULL, 0,
967 		       CTL_CREATE, CTL_EOL) != 0)
968 		goto bad;
969 
970 	if (sysctl_createv(clog, 0, &rnode, &cnode,
971 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
972 		       CTLTYPE_INT, "maxlen",
973 		       SYSCTL_DESCR("Maximum allowed input queue length"),
974 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
975 		       CTL_CREATE, CTL_EOL) != 0)
976 		goto bad;
977 #endif
978 
979 	if (sysctl_createv(clog, 0, &rnode, &cnode,
980 		       CTLFLAG_PERMANENT,
981 		       CTLTYPE_INT, "drops",
982 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
983 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
984 		       CTL_CREATE, CTL_EOL) != 0)
985 		goto bad;
986 
987 	return;
988 bad:
989 	printf("%s: could not attach sysctl nodes\n", ifname);
990 	return;
991 }
992 
993 /*
994  * The deferred if_start framework
995  *
996  * The common APIs to defer if_start to softint when if_start is requested
997  * from a device driver running in hardware interrupt context.
998  */
999 /*
1000  * Call ifp->if_start (or equivalent) in a dedicated softint for
1001  * deferred if_start.
1002  */
1003 static void
1004 if_deferred_start_softint(void *arg)
1005 {
1006 	struct if_deferred_start *ids = arg;
1007 	struct ifnet *ifp = ids->ids_ifp;
1008 
1009 	ids->ids_if_start(ifp);
1010 }
1011 
1012 /*
1013  * The default callback function for deferred if_start.
1014  */
1015 static void
1016 if_deferred_start_common(struct ifnet *ifp)
1017 {
1018 	int s;
1019 
1020 	s = splnet();
1021 	if_start_lock(ifp);
1022 	splx(s);
1023 }
1024 
1025 static inline bool
1026 if_snd_is_used(struct ifnet *ifp)
1027 {
1028 
1029 	return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit ||
1030 	    ALTQ_IS_ENABLED(&ifp->if_snd);
1031 }
1032 
1033 /*
1034  * Schedule deferred if_start.
1035  */
1036 void
1037 if_schedule_deferred_start(struct ifnet *ifp)
1038 {
1039 
1040 	KASSERT(ifp->if_deferred_start != NULL);
1041 
1042 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1043 		return;
1044 
1045 	softint_schedule(ifp->if_deferred_start->ids_si);
1046 }
1047 
1048 /*
1049  * Create an instance of deferred if_start. A driver should call the function
1050  * only if the driver needs deferred if_start. Drivers can setup their own
1051  * deferred if_start function via 2nd argument.
1052  */
1053 void
1054 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1055 {
1056 	struct if_deferred_start *ids;
1057 
1058 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1059 	ids->ids_ifp = ifp;
1060 	ids->ids_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
1061 	    if_deferred_start_softint, ids);
1062 	if (func != NULL)
1063 		ids->ids_if_start = func;
1064 	else
1065 		ids->ids_if_start = if_deferred_start_common;
1066 
1067 	ifp->if_deferred_start = ids;
1068 }
1069 
1070 static void
1071 if_deferred_start_destroy(struct ifnet *ifp)
1072 {
1073 
1074 	if (ifp->if_deferred_start == NULL)
1075 		return;
1076 
1077 	softint_disestablish(ifp->if_deferred_start->ids_si);
1078 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1079 	ifp->if_deferred_start = NULL;
1080 }
1081 
1082 /*
1083  * The common interface input routine that is called by device drivers,
1084  * which should be used only when the driver's rx handler already runs
1085  * in softint.
1086  */
1087 void
1088 if_input(struct ifnet *ifp, struct mbuf *m)
1089 {
1090 
1091 	KASSERT(ifp->if_percpuq == NULL);
1092 	KASSERT(!cpu_intr_p());
1093 
1094 	ifp->if_ipackets++;
1095 	bpf_mtap(ifp, m);
1096 
1097 	ifp->_if_input(ifp, m);
1098 }
1099 
1100 /*
1101  * DEPRECATED. Use if_initialize and if_register instead.
1102  * See the above comment of if_initialize.
1103  *
1104  * Note that it implicitly enables if_percpuq to make drivers easy to
1105  * migrate softint-based if_input without much changes. If you don't
1106  * want to enable it, use if_initialize instead.
1107  */
1108 int
1109 if_attach(ifnet_t *ifp)
1110 {
1111 	int rv;
1112 
1113 	rv = if_initialize(ifp);
1114 	if (rv != 0)
1115 		return rv;
1116 
1117 	ifp->if_percpuq = if_percpuq_create(ifp);
1118 	if_register(ifp);
1119 
1120 	return 0;
1121 }
1122 
1123 void
1124 if_attachdomain(void)
1125 {
1126 	struct ifnet *ifp;
1127 	int s;
1128 	int bound = curlwp_bind();
1129 
1130 	s = pserialize_read_enter();
1131 	IFNET_READER_FOREACH(ifp) {
1132 		struct psref psref;
1133 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1134 		pserialize_read_exit(s);
1135 		if_attachdomain1(ifp);
1136 		s = pserialize_read_enter();
1137 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1138 	}
1139 	pserialize_read_exit(s);
1140 	curlwp_bindx(bound);
1141 }
1142 
1143 static void
1144 if_attachdomain1(struct ifnet *ifp)
1145 {
1146 	struct domain *dp;
1147 	int s;
1148 
1149 	s = splsoftnet();
1150 
1151 	/* address family dependent data region */
1152 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1153 	DOMAIN_FOREACH(dp) {
1154 		if (dp->dom_ifattach != NULL)
1155 			ifp->if_afdata[dp->dom_family] =
1156 			    (*dp->dom_ifattach)(ifp);
1157 	}
1158 
1159 	splx(s);
1160 }
1161 
1162 /*
1163  * Deactivate an interface.  This points all of the procedure
1164  * handles at error stubs.  May be called from interrupt context.
1165  */
1166 void
1167 if_deactivate(struct ifnet *ifp)
1168 {
1169 	int s;
1170 
1171 	s = splsoftnet();
1172 
1173 	ifp->if_output	 = if_nulloutput;
1174 	ifp->_if_input	 = if_nullinput;
1175 	ifp->if_start	 = if_nullstart;
1176 	ifp->if_transmit = if_nulltransmit;
1177 	ifp->if_ioctl	 = if_nullioctl;
1178 	ifp->if_init	 = if_nullinit;
1179 	ifp->if_stop	 = if_nullstop;
1180 	ifp->if_slowtimo = if_nullslowtimo;
1181 	ifp->if_drain	 = if_nulldrain;
1182 
1183 	/* No more packets may be enqueued. */
1184 	ifp->if_snd.ifq_maxlen = 0;
1185 
1186 	splx(s);
1187 }
1188 
1189 bool
1190 if_is_deactivated(const struct ifnet *ifp)
1191 {
1192 
1193 	return ifp->if_output == if_nulloutput;
1194 }
1195 
1196 void
1197 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1198 {
1199 	struct ifaddr *ifa, *nifa;
1200 	int s;
1201 
1202 	s = pserialize_read_enter();
1203 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1204 		nifa = IFADDR_READER_NEXT(ifa);
1205 		if (ifa->ifa_addr->sa_family != family)
1206 			continue;
1207 		pserialize_read_exit(s);
1208 
1209 		(*purgeaddr)(ifa);
1210 
1211 		s = pserialize_read_enter();
1212 	}
1213 	pserialize_read_exit(s);
1214 }
1215 
1216 #ifdef IFAREF_DEBUG
1217 static struct ifaddr **ifa_list;
1218 static int ifa_list_size;
1219 
1220 /* Depends on only one if_attach runs at once */
1221 static void
1222 if_build_ifa_list(struct ifnet *ifp)
1223 {
1224 	struct ifaddr *ifa;
1225 	int i;
1226 
1227 	KASSERT(ifa_list == NULL);
1228 	KASSERT(ifa_list_size == 0);
1229 
1230 	IFADDR_READER_FOREACH(ifa, ifp)
1231 		ifa_list_size++;
1232 
1233 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1234 	i = 0;
1235 	IFADDR_READER_FOREACH(ifa, ifp) {
1236 		ifa_list[i++] = ifa;
1237 		ifaref(ifa);
1238 	}
1239 }
1240 
1241 static void
1242 if_check_and_free_ifa_list(struct ifnet *ifp)
1243 {
1244 	int i;
1245 	struct ifaddr *ifa;
1246 
1247 	if (ifa_list == NULL)
1248 		return;
1249 
1250 	for (i = 0; i < ifa_list_size; i++) {
1251 		char buf[64];
1252 
1253 		ifa = ifa_list[i];
1254 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1255 		if (ifa->ifa_refcnt > 1) {
1256 			log(LOG_WARNING,
1257 			    "ifa(%s) still referenced (refcnt=%d)\n",
1258 			    buf, ifa->ifa_refcnt - 1);
1259 		} else
1260 			log(LOG_DEBUG,
1261 			    "ifa(%s) not referenced (refcnt=%d)\n",
1262 			    buf, ifa->ifa_refcnt - 1);
1263 		ifafree(ifa);
1264 	}
1265 
1266 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1267 	ifa_list = NULL;
1268 	ifa_list_size = 0;
1269 }
1270 #endif
1271 
1272 /*
1273  * Detach an interface from the list of "active" interfaces,
1274  * freeing any resources as we go along.
1275  *
1276  * NOTE: This routine must be called with a valid thread context,
1277  * as it may block.
1278  */
1279 void
1280 if_detach(struct ifnet *ifp)
1281 {
1282 	struct socket so;
1283 	struct ifaddr *ifa;
1284 #ifdef IFAREF_DEBUG
1285 	struct ifaddr *last_ifa = NULL;
1286 #endif
1287 	struct domain *dp;
1288 	const struct protosw *pr;
1289 	int s, i, family, purged;
1290 	uint64_t xc;
1291 
1292 #ifdef IFAREF_DEBUG
1293 	if_build_ifa_list(ifp);
1294 #endif
1295 	/*
1296 	 * XXX It's kind of lame that we have to have the
1297 	 * XXX socket structure...
1298 	 */
1299 	memset(&so, 0, sizeof(so));
1300 
1301 	s = splnet();
1302 
1303 	sysctl_teardown(&ifp->if_sysctl_log);
1304 	IFNET_LOCK(ifp);
1305 	if_deactivate(ifp);
1306 	IFNET_UNLOCK(ifp);
1307 
1308 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1309 		ifp->if_slowtimo = NULL;
1310 		callout_halt(ifp->if_slowtimo_ch, NULL);
1311 		callout_destroy(ifp->if_slowtimo_ch);
1312 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1313 	}
1314 	if_deferred_start_destroy(ifp);
1315 
1316 	/*
1317 	 * Do an if_down() to give protocols a chance to do something.
1318 	 */
1319 	if_down_deactivated(ifp);
1320 
1321 #ifdef ALTQ
1322 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1323 		altq_disable(&ifp->if_snd);
1324 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1325 		altq_detach(&ifp->if_snd);
1326 #endif
1327 
1328 #if NCARP > 0
1329 	/* Remove the interface from any carp group it is a part of.  */
1330 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1331 		carp_ifdetach(ifp);
1332 #endif
1333 
1334 	/*
1335 	 * Rip all the addresses off the interface.  This should make
1336 	 * all of the routes go away.
1337 	 *
1338 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1339 	 * from the list, including our "cursor", ifa.  For safety,
1340 	 * and to honor the TAILQ abstraction, I just restart the
1341 	 * loop after each removal.  Note that the loop will exit
1342 	 * when all of the remaining ifaddrs belong to the AF_LINK
1343 	 * family.  I am counting on the historical fact that at
1344 	 * least one pr_usrreq in each address domain removes at
1345 	 * least one ifaddr.
1346 	 */
1347 again:
1348 	/*
1349 	 * At this point, no other one tries to remove ifa in the list,
1350 	 * so we don't need to take a lock or psref.  Avoid using
1351 	 * IFADDR_READER_FOREACH to pass over an inspection of contract
1352 	 * violations of pserialize.
1353 	 */
1354 	IFADDR_WRITER_FOREACH(ifa, ifp) {
1355 		family = ifa->ifa_addr->sa_family;
1356 #ifdef IFAREF_DEBUG
1357 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1358 		    ifa, family, ifa->ifa_refcnt);
1359 		if (last_ifa != NULL && ifa == last_ifa)
1360 			panic("if_detach: loop detected");
1361 		last_ifa = ifa;
1362 #endif
1363 		if (family == AF_LINK)
1364 			continue;
1365 		dp = pffinddomain(family);
1366 		KASSERTMSG(dp != NULL, "no domain for AF %d", family);
1367 		/*
1368 		 * XXX These PURGEIF calls are redundant with the
1369 		 * purge-all-families calls below, but are left in for
1370 		 * now both to make a smaller change, and to avoid
1371 		 * unplanned interactions with clearing of
1372 		 * ifp->if_addrlist.
1373 		 */
1374 		purged = 0;
1375 		for (pr = dp->dom_protosw;
1376 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1377 			so.so_proto = pr;
1378 			if (pr->pr_usrreqs) {
1379 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1380 				purged = 1;
1381 			}
1382 		}
1383 		if (purged == 0) {
1384 			/*
1385 			 * XXX What's really the best thing to do
1386 			 * XXX here?  --thorpej@NetBSD.org
1387 			 */
1388 			printf("if_detach: WARNING: AF %d not purged\n",
1389 			    family);
1390 			ifa_remove(ifp, ifa);
1391 		}
1392 		goto again;
1393 	}
1394 
1395 	if_free_sadl(ifp);
1396 
1397 	/* Delete stray routes from the routing table. */
1398 	for (i = 0; i <= AF_MAX; i++)
1399 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1400 
1401 	DOMAIN_FOREACH(dp) {
1402 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1403 		{
1404 			void *p = ifp->if_afdata[dp->dom_family];
1405 			if (p) {
1406 				ifp->if_afdata[dp->dom_family] = NULL;
1407 				(*dp->dom_ifdetach)(ifp, p);
1408 			}
1409 		}
1410 
1411 		/*
1412 		 * One would expect multicast memberships (INET and
1413 		 * INET6) on UDP sockets to be purged by the PURGEIF
1414 		 * calls above, but if all addresses were removed from
1415 		 * the interface prior to destruction, the calls will
1416 		 * not be made (e.g. ppp, for which pppd(8) generally
1417 		 * removes addresses before destroying the interface).
1418 		 * Because there is no invariant that multicast
1419 		 * memberships only exist for interfaces with IPv4
1420 		 * addresses, we must call PURGEIF regardless of
1421 		 * addresses.  (Protocols which might store ifnet
1422 		 * pointers are marked with PR_PURGEIF.)
1423 		 */
1424 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1425 			so.so_proto = pr;
1426 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1427 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1428 		}
1429 	}
1430 
1431 	/* Wait for all readers to drain before freeing.  */
1432 	IFNET_GLOBAL_LOCK();
1433 	ifindex2ifnet[ifp->if_index] = NULL;
1434 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1435 	IFNET_WRITER_REMOVE(ifp);
1436 	pserialize_perform(ifnet_psz);
1437 	IFNET_GLOBAL_UNLOCK();
1438 
1439 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1440 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1441 
1442 	pfil_run_ifhooks(if_pfil, PFIL_IFNET_DETACH, ifp);
1443 	(void)pfil_head_destroy(ifp->if_pfil);
1444 
1445 	/* Announce that the interface is gone. */
1446 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1447 
1448 	IF_AFDATA_LOCK_DESTROY(ifp);
1449 
1450 	if (if_is_link_state_changeable(ifp)) {
1451 		softint_disestablish(ifp->if_link_si);
1452 		ifp->if_link_si = NULL;
1453 	}
1454 
1455 	/*
1456 	 * remove packets that came from ifp, from software interrupt queues.
1457 	 */
1458 	DOMAIN_FOREACH(dp) {
1459 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1460 			struct ifqueue *iq = dp->dom_ifqueues[i];
1461 			if (iq == NULL)
1462 				break;
1463 			dp->dom_ifqueues[i] = NULL;
1464 			if_detach_queues(ifp, iq);
1465 		}
1466 	}
1467 
1468 	/*
1469 	 * IP queues have to be processed separately: net-queue barrier
1470 	 * ensures that the packets are dequeued while a cross-call will
1471 	 * ensure that the interrupts have completed. FIXME: not quite..
1472 	 */
1473 #ifdef INET
1474 	pktq_barrier(ip_pktq);
1475 #endif
1476 #ifdef INET6
1477 	if (in6_present)
1478 		pktq_barrier(ip6_pktq);
1479 #endif
1480 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1481 	xc_wait(xc);
1482 
1483 	if (ifp->if_percpuq != NULL) {
1484 		if_percpuq_destroy(ifp->if_percpuq);
1485 		ifp->if_percpuq = NULL;
1486 	}
1487 
1488 	mutex_obj_free(ifp->if_ioctl_lock);
1489 	ifp->if_ioctl_lock = NULL;
1490 	mutex_obj_free(ifp->if_snd.ifq_lock);
1491 
1492 	splx(s);
1493 
1494 #ifdef IFAREF_DEBUG
1495 	if_check_and_free_ifa_list(ifp);
1496 #endif
1497 }
1498 
1499 static void
1500 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1501 {
1502 	struct mbuf *m, *prev, *next;
1503 
1504 	prev = NULL;
1505 	for (m = q->ifq_head; m != NULL; m = next) {
1506 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1507 
1508 		next = m->m_nextpkt;
1509 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1510 			prev = m;
1511 			continue;
1512 		}
1513 
1514 		if (prev != NULL)
1515 			prev->m_nextpkt = m->m_nextpkt;
1516 		else
1517 			q->ifq_head = m->m_nextpkt;
1518 		if (q->ifq_tail == m)
1519 			q->ifq_tail = prev;
1520 		q->ifq_len--;
1521 
1522 		m->m_nextpkt = NULL;
1523 		m_freem(m);
1524 		IF_DROP(q);
1525 	}
1526 }
1527 
1528 /*
1529  * Callback for a radix tree walk to delete all references to an
1530  * ifnet.
1531  */
1532 static int
1533 if_delroute_matcher(struct rtentry *rt, void *v)
1534 {
1535 	struct ifnet *ifp = (struct ifnet *)v;
1536 
1537 	if (rt->rt_ifp == ifp)
1538 		return 1;
1539 	else
1540 		return 0;
1541 }
1542 
1543 /*
1544  * Create a clone network interface.
1545  */
1546 static int
1547 if_clone_create(const char *name)
1548 {
1549 	struct if_clone *ifc;
1550 	int unit;
1551 	struct ifnet *ifp;
1552 	struct psref psref;
1553 
1554 	KASSERT(mutex_owned(&if_clone_mtx));
1555 
1556 	ifc = if_clone_lookup(name, &unit);
1557 	if (ifc == NULL)
1558 		return EINVAL;
1559 
1560 	ifp = if_get(name, &psref);
1561 	if (ifp != NULL) {
1562 		if_put(ifp, &psref);
1563 		return EEXIST;
1564 	}
1565 
1566 	return (*ifc->ifc_create)(ifc, unit);
1567 }
1568 
1569 /*
1570  * Destroy a clone network interface.
1571  */
1572 static int
1573 if_clone_destroy(const char *name)
1574 {
1575 	struct if_clone *ifc;
1576 	struct ifnet *ifp;
1577 	struct psref psref;
1578 
1579 	KASSERT(mutex_owned(&if_clone_mtx));
1580 
1581 	ifc = if_clone_lookup(name, NULL);
1582 	if (ifc == NULL)
1583 		return EINVAL;
1584 
1585 	if (ifc->ifc_destroy == NULL)
1586 		return EOPNOTSUPP;
1587 
1588 	ifp = if_get(name, &psref);
1589 	if (ifp == NULL)
1590 		return ENXIO;
1591 
1592 	/* We have to disable ioctls here */
1593 	IFNET_LOCK(ifp);
1594 	ifp->if_ioctl = if_nullioctl;
1595 	IFNET_UNLOCK(ifp);
1596 
1597 	/*
1598 	 * We cannot call ifc_destroy with holding ifp.
1599 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1600 	 */
1601 	if_put(ifp, &psref);
1602 
1603 	return (*ifc->ifc_destroy)(ifp);
1604 }
1605 
1606 static bool
1607 if_is_unit(const char *name)
1608 {
1609 
1610 	while(*name != '\0') {
1611 		if (*name < '0' || *name > '9')
1612 			return false;
1613 		name++;
1614 	}
1615 
1616 	return true;
1617 }
1618 
1619 /*
1620  * Look up a network interface cloner.
1621  */
1622 static struct if_clone *
1623 if_clone_lookup(const char *name, int *unitp)
1624 {
1625 	struct if_clone *ifc;
1626 	const char *cp;
1627 	char *dp, ifname[IFNAMSIZ + 3];
1628 	int unit;
1629 
1630 	KASSERT(mutex_owned(&if_clone_mtx));
1631 
1632 	strcpy(ifname, "if_");
1633 	/* separate interface name from unit */
1634 	/* TODO: search unit number from backward */
1635 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1636 	    *cp && !if_is_unit(cp);)
1637 		*dp++ = *cp++;
1638 
1639 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1640 		return NULL;	/* No name or unit number */
1641 	*dp++ = '\0';
1642 
1643 again:
1644 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1645 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1646 			break;
1647 	}
1648 
1649 	if (ifc == NULL) {
1650 		int error;
1651 		if (*ifname == '\0')
1652 			return NULL;
1653 		mutex_exit(&if_clone_mtx);
1654 		error = module_autoload(ifname, MODULE_CLASS_DRIVER);
1655 		mutex_enter(&if_clone_mtx);
1656 		if (error)
1657 			return NULL;
1658 		*ifname = '\0';
1659 		goto again;
1660 	}
1661 
1662 	unit = 0;
1663 	while (cp - name < IFNAMSIZ && *cp) {
1664 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1665 			/* Bogus unit number. */
1666 			return NULL;
1667 		}
1668 		unit = (unit * 10) + (*cp++ - '0');
1669 	}
1670 
1671 	if (unitp != NULL)
1672 		*unitp = unit;
1673 	return ifc;
1674 }
1675 
1676 /*
1677  * Register a network interface cloner.
1678  */
1679 void
1680 if_clone_attach(struct if_clone *ifc)
1681 {
1682 
1683 	mutex_enter(&if_clone_mtx);
1684 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1685 	if_cloners_count++;
1686 	mutex_exit(&if_clone_mtx);
1687 }
1688 
1689 /*
1690  * Unregister a network interface cloner.
1691  */
1692 void
1693 if_clone_detach(struct if_clone *ifc)
1694 {
1695 
1696 	mutex_enter(&if_clone_mtx);
1697 	LIST_REMOVE(ifc, ifc_list);
1698 	if_cloners_count--;
1699 	mutex_exit(&if_clone_mtx);
1700 }
1701 
1702 /*
1703  * Provide list of interface cloners to userspace.
1704  */
1705 int
1706 if_clone_list(int buf_count, char *buffer, int *total)
1707 {
1708 	char outbuf[IFNAMSIZ], *dst;
1709 	struct if_clone *ifc;
1710 	int count, error = 0;
1711 
1712 	mutex_enter(&if_clone_mtx);
1713 	*total = if_cloners_count;
1714 	if ((dst = buffer) == NULL) {
1715 		/* Just asking how many there are. */
1716 		goto out;
1717 	}
1718 
1719 	if (buf_count < 0) {
1720 		error = EINVAL;
1721 		goto out;
1722 	}
1723 
1724 	count = (if_cloners_count < buf_count) ?
1725 	    if_cloners_count : buf_count;
1726 
1727 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1728 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1729 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1730 		if (outbuf[sizeof(outbuf) - 1] != '\0') {
1731 			error = ENAMETOOLONG;
1732 			goto out;
1733 		}
1734 		error = copyout(outbuf, dst, sizeof(outbuf));
1735 		if (error != 0)
1736 			break;
1737 	}
1738 
1739 out:
1740 	mutex_exit(&if_clone_mtx);
1741 	return error;
1742 }
1743 
1744 void
1745 ifa_psref_init(struct ifaddr *ifa)
1746 {
1747 
1748 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1749 }
1750 
1751 void
1752 ifaref(struct ifaddr *ifa)
1753 {
1754 	KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING));
1755 	ifa->ifa_refcnt++;
1756 }
1757 
1758 void
1759 ifafree(struct ifaddr *ifa)
1760 {
1761 	KASSERT(ifa != NULL);
1762 	KASSERT(ifa->ifa_refcnt > 0);
1763 
1764 	if (--ifa->ifa_refcnt == 0) {
1765 		free(ifa, M_IFADDR);
1766 	}
1767 }
1768 
1769 bool
1770 ifa_is_destroying(struct ifaddr *ifa)
1771 {
1772 
1773 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1774 }
1775 
1776 void
1777 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1778 {
1779 
1780 	ifa->ifa_ifp = ifp;
1781 
1782 	/*
1783 	 * Check MP-safety for IFEF_MPSAFE drivers.
1784 	 * Check !IFF_RUNNING for initialization routines that normally don't
1785 	 * take IFNET_LOCK but it's safe because there is no competitor.
1786 	 * XXX there are false positive cases because IFF_RUNNING can be off on
1787 	 * if_stop.
1788 	 */
1789 	KASSERT(!if_is_mpsafe(ifp) || !ISSET(ifp->if_flags, IFF_RUNNING) ||
1790 	    IFNET_LOCKED(ifp));
1791 
1792 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1793 	IFADDR_ENTRY_INIT(ifa);
1794 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1795 
1796 	ifaref(ifa);
1797 }
1798 
1799 void
1800 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1801 {
1802 
1803 	KASSERT(ifa->ifa_ifp == ifp);
1804 	/*
1805 	 * Check MP-safety for IFEF_MPSAFE drivers.
1806 	 * if_is_deactivated indicates ifa_remove is called form if_detach
1807 	 * where is safe even if IFNET_LOCK isn't held.
1808 	 */
1809 	KASSERT(!if_is_mpsafe(ifp) || if_is_deactivated(ifp) || IFNET_LOCKED(ifp));
1810 
1811 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1812 	IFADDR_WRITER_REMOVE(ifa);
1813 #ifdef NET_MPSAFE
1814 	IFNET_GLOBAL_LOCK();
1815 	pserialize_perform(ifnet_psz);
1816 	IFNET_GLOBAL_UNLOCK();
1817 #endif
1818 
1819 #ifdef NET_MPSAFE
1820 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1821 #endif
1822 	IFADDR_ENTRY_DESTROY(ifa);
1823 	ifafree(ifa);
1824 }
1825 
1826 void
1827 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1828 {
1829 
1830 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1831 }
1832 
1833 void
1834 ifa_release(struct ifaddr *ifa, struct psref *psref)
1835 {
1836 
1837 	if (ifa == NULL)
1838 		return;
1839 
1840 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1841 }
1842 
1843 bool
1844 ifa_held(struct ifaddr *ifa)
1845 {
1846 
1847 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1848 }
1849 
1850 static inline int
1851 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1852 {
1853 	return sockaddr_cmp(sa1, sa2) == 0;
1854 }
1855 
1856 /*
1857  * Locate an interface based on a complete address.
1858  */
1859 /*ARGSUSED*/
1860 struct ifaddr *
1861 ifa_ifwithaddr(const struct sockaddr *addr)
1862 {
1863 	struct ifnet *ifp;
1864 	struct ifaddr *ifa;
1865 
1866 	IFNET_READER_FOREACH(ifp) {
1867 		if (if_is_deactivated(ifp))
1868 			continue;
1869 		IFADDR_READER_FOREACH(ifa, ifp) {
1870 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1871 				continue;
1872 			if (equal(addr, ifa->ifa_addr))
1873 				return ifa;
1874 			if ((ifp->if_flags & IFF_BROADCAST) &&
1875 			    ifa->ifa_broadaddr &&
1876 			    /* IP6 doesn't have broadcast */
1877 			    ifa->ifa_broadaddr->sa_len != 0 &&
1878 			    equal(ifa->ifa_broadaddr, addr))
1879 				return ifa;
1880 		}
1881 	}
1882 	return NULL;
1883 }
1884 
1885 struct ifaddr *
1886 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1887 {
1888 	struct ifaddr *ifa;
1889 	int s = pserialize_read_enter();
1890 
1891 	ifa = ifa_ifwithaddr(addr);
1892 	if (ifa != NULL)
1893 		ifa_acquire(ifa, psref);
1894 	pserialize_read_exit(s);
1895 
1896 	return ifa;
1897 }
1898 
1899 /*
1900  * Locate the point to point interface with a given destination address.
1901  */
1902 /*ARGSUSED*/
1903 struct ifaddr *
1904 ifa_ifwithdstaddr(const struct sockaddr *addr)
1905 {
1906 	struct ifnet *ifp;
1907 	struct ifaddr *ifa;
1908 
1909 	IFNET_READER_FOREACH(ifp) {
1910 		if (if_is_deactivated(ifp))
1911 			continue;
1912 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1913 			continue;
1914 		IFADDR_READER_FOREACH(ifa, ifp) {
1915 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1916 			    ifa->ifa_dstaddr == NULL)
1917 				continue;
1918 			if (equal(addr, ifa->ifa_dstaddr))
1919 				return ifa;
1920 		}
1921 	}
1922 
1923 	return NULL;
1924 }
1925 
1926 struct ifaddr *
1927 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1928 {
1929 	struct ifaddr *ifa;
1930 	int s;
1931 
1932 	s = pserialize_read_enter();
1933 	ifa = ifa_ifwithdstaddr(addr);
1934 	if (ifa != NULL)
1935 		ifa_acquire(ifa, psref);
1936 	pserialize_read_exit(s);
1937 
1938 	return ifa;
1939 }
1940 
1941 /*
1942  * Find an interface on a specific network.  If many, choice
1943  * is most specific found.
1944  */
1945 struct ifaddr *
1946 ifa_ifwithnet(const struct sockaddr *addr)
1947 {
1948 	struct ifnet *ifp;
1949 	struct ifaddr *ifa, *ifa_maybe = NULL;
1950 	const struct sockaddr_dl *sdl;
1951 	u_int af = addr->sa_family;
1952 	const char *addr_data = addr->sa_data, *cplim;
1953 
1954 	if (af == AF_LINK) {
1955 		sdl = satocsdl(addr);
1956 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1957 		    ifindex2ifnet[sdl->sdl_index] &&
1958 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1959 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1960 		}
1961 	}
1962 #ifdef NETATALK
1963 	if (af == AF_APPLETALK) {
1964 		const struct sockaddr_at *sat, *sat2;
1965 		sat = (const struct sockaddr_at *)addr;
1966 		IFNET_READER_FOREACH(ifp) {
1967 			if (if_is_deactivated(ifp))
1968 				continue;
1969 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1970 			if (ifa == NULL)
1971 				continue;
1972 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1973 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1974 				return ifa; /* exact match */
1975 			if (ifa_maybe == NULL) {
1976 				/* else keep the if with the right range */
1977 				ifa_maybe = ifa;
1978 			}
1979 		}
1980 		return ifa_maybe;
1981 	}
1982 #endif
1983 	IFNET_READER_FOREACH(ifp) {
1984 		if (if_is_deactivated(ifp))
1985 			continue;
1986 		IFADDR_READER_FOREACH(ifa, ifp) {
1987 			const char *cp, *cp2, *cp3;
1988 
1989 			if (ifa->ifa_addr->sa_family != af ||
1990 			    ifa->ifa_netmask == NULL)
1991  next:				continue;
1992 			cp = addr_data;
1993 			cp2 = ifa->ifa_addr->sa_data;
1994 			cp3 = ifa->ifa_netmask->sa_data;
1995 			cplim = (const char *)ifa->ifa_netmask +
1996 			    ifa->ifa_netmask->sa_len;
1997 			while (cp3 < cplim) {
1998 				if ((*cp++ ^ *cp2++) & *cp3++) {
1999 					/* want to continue for() loop */
2000 					goto next;
2001 				}
2002 			}
2003 			if (ifa_maybe == NULL ||
2004 			    rt_refines(ifa->ifa_netmask,
2005 			               ifa_maybe->ifa_netmask))
2006 				ifa_maybe = ifa;
2007 		}
2008 	}
2009 	return ifa_maybe;
2010 }
2011 
2012 struct ifaddr *
2013 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
2014 {
2015 	struct ifaddr *ifa;
2016 	int s;
2017 
2018 	s = pserialize_read_enter();
2019 	ifa = ifa_ifwithnet(addr);
2020 	if (ifa != NULL)
2021 		ifa_acquire(ifa, psref);
2022 	pserialize_read_exit(s);
2023 
2024 	return ifa;
2025 }
2026 
2027 /*
2028  * Find the interface of the addresss.
2029  */
2030 struct ifaddr *
2031 ifa_ifwithladdr(const struct sockaddr *addr)
2032 {
2033 	struct ifaddr *ia;
2034 
2035 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
2036 	    (ia = ifa_ifwithnet(addr)))
2037 		return ia;
2038 	return NULL;
2039 }
2040 
2041 struct ifaddr *
2042 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
2043 {
2044 	struct ifaddr *ifa;
2045 	int s;
2046 
2047 	s = pserialize_read_enter();
2048 	ifa = ifa_ifwithladdr(addr);
2049 	if (ifa != NULL)
2050 		ifa_acquire(ifa, psref);
2051 	pserialize_read_exit(s);
2052 
2053 	return ifa;
2054 }
2055 
2056 /*
2057  * Find an interface using a specific address family
2058  */
2059 struct ifaddr *
2060 ifa_ifwithaf(int af)
2061 {
2062 	struct ifnet *ifp;
2063 	struct ifaddr *ifa = NULL;
2064 	int s;
2065 
2066 	s = pserialize_read_enter();
2067 	IFNET_READER_FOREACH(ifp) {
2068 		if (if_is_deactivated(ifp))
2069 			continue;
2070 		IFADDR_READER_FOREACH(ifa, ifp) {
2071 			if (ifa->ifa_addr->sa_family == af)
2072 				goto out;
2073 		}
2074 	}
2075 out:
2076 	pserialize_read_exit(s);
2077 	return ifa;
2078 }
2079 
2080 /*
2081  * Find an interface address specific to an interface best matching
2082  * a given address.
2083  */
2084 struct ifaddr *
2085 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2086 {
2087 	struct ifaddr *ifa;
2088 	const char *cp, *cp2, *cp3;
2089 	const char *cplim;
2090 	struct ifaddr *ifa_maybe = 0;
2091 	u_int af = addr->sa_family;
2092 
2093 	if (if_is_deactivated(ifp))
2094 		return NULL;
2095 
2096 	if (af >= AF_MAX)
2097 		return NULL;
2098 
2099 	IFADDR_READER_FOREACH(ifa, ifp) {
2100 		if (ifa->ifa_addr->sa_family != af)
2101 			continue;
2102 		ifa_maybe = ifa;
2103 		if (ifa->ifa_netmask == NULL) {
2104 			if (equal(addr, ifa->ifa_addr) ||
2105 			    (ifa->ifa_dstaddr &&
2106 			     equal(addr, ifa->ifa_dstaddr)))
2107 				return ifa;
2108 			continue;
2109 		}
2110 		cp = addr->sa_data;
2111 		cp2 = ifa->ifa_addr->sa_data;
2112 		cp3 = ifa->ifa_netmask->sa_data;
2113 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2114 		for (; cp3 < cplim; cp3++) {
2115 			if ((*cp++ ^ *cp2++) & *cp3)
2116 				break;
2117 		}
2118 		if (cp3 == cplim)
2119 			return ifa;
2120 	}
2121 	return ifa_maybe;
2122 }
2123 
2124 struct ifaddr *
2125 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2126     struct psref *psref)
2127 {
2128 	struct ifaddr *ifa;
2129 	int s;
2130 
2131 	s = pserialize_read_enter();
2132 	ifa = ifaof_ifpforaddr(addr, ifp);
2133 	if (ifa != NULL)
2134 		ifa_acquire(ifa, psref);
2135 	pserialize_read_exit(s);
2136 
2137 	return ifa;
2138 }
2139 
2140 /*
2141  * Default action when installing a route with a Link Level gateway.
2142  * Lookup an appropriate real ifa to point to.
2143  * This should be moved to /sys/net/link.c eventually.
2144  */
2145 void
2146 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2147 {
2148 	struct ifaddr *ifa;
2149 	const struct sockaddr *dst;
2150 	struct ifnet *ifp;
2151 	struct psref psref;
2152 
2153 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2154 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
2155 		return;
2156 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2157 		rt_replace_ifa(rt, ifa);
2158 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2159 			ifa->ifa_rtrequest(cmd, rt, info);
2160 		ifa_release(ifa, &psref);
2161 	}
2162 }
2163 
2164 /*
2165  * bitmask macros to manage a densely packed link_state change queue.
2166  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2167  * LINK_STATE_UP(2) we need 2 bits for each state change.
2168  * As a state change to store is 0, treat all bits set as an unset item.
2169  */
2170 #define LQ_ITEM_BITS		2
2171 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2172 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2173 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2174 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2175 #define LQ_STORE(q, i, v)						      \
2176 	do {								      \
2177 		(q) &= ~LQ_MASK((i));					      \
2178 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2179 	} while (0 /* CONSTCOND */)
2180 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2181 #define LQ_POP(q, v)							      \
2182 	do {								      \
2183 		(v) = LQ_ITEM((q), 0);					      \
2184 		(q) >>= LQ_ITEM_BITS;					      \
2185 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2186 	} while (0 /* CONSTCOND */)
2187 #define LQ_PUSH(q, v)							      \
2188 	do {								      \
2189 		(q) >>= LQ_ITEM_BITS;					      \
2190 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2191 	} while (0 /* CONSTCOND */)
2192 #define LQ_FIND_UNSET(q, i)						      \
2193 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2194 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2195 			break;						      \
2196 	}
2197 
2198 /*
2199  * XXX reusing (ifp)->if_snd->ifq_lock rather than having another spin mutex
2200  * for each ifnet.  It doesn't matter because:
2201  * - if IFEF_MPSAFE is enabled, if_snd isn't used and lock contentions on
2202  *   ifq_lock don't happen
2203  * - if IFEF_MPSAFE is disabled, there is no lock contention on ifq_lock
2204  *   because if_snd, if_link_state_change and if_link_state_change_softint
2205  *   are all called with KERNEL_LOCK
2206  */
2207 #define IF_LINK_STATE_CHANGE_LOCK(ifp)		\
2208 	mutex_enter((ifp)->if_snd.ifq_lock)
2209 #define IF_LINK_STATE_CHANGE_UNLOCK(ifp)	\
2210 	mutex_exit((ifp)->if_snd.ifq_lock)
2211 
2212 /*
2213  * Handle a change in the interface link state and
2214  * queue notifications.
2215  */
2216 void
2217 if_link_state_change(struct ifnet *ifp, int link_state)
2218 {
2219 	int idx;
2220 
2221 	KASSERTMSG(if_is_link_state_changeable(ifp),
2222 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2223 	    ifp->if_xname, ifp->if_extflags);
2224 
2225 	/* Ensure change is to a valid state */
2226 	switch (link_state) {
2227 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2228 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2229 	case LINK_STATE_UP:
2230 		break;
2231 	default:
2232 #ifdef DEBUG
2233 		printf("%s: invalid link state %d\n",
2234 		    ifp->if_xname, link_state);
2235 #endif
2236 		return;
2237 	}
2238 
2239 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2240 
2241 	/* Find the last unset event in the queue. */
2242 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2243 
2244 	/*
2245 	 * Ensure link_state doesn't match the last event in the queue.
2246 	 * ifp->if_link_state is not checked and set here because
2247 	 * that would present an inconsistent picture to the system.
2248 	 */
2249 	if (idx != 0 &&
2250 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2251 		goto out;
2252 
2253 	/* Handle queue overflow. */
2254 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2255 		uint8_t lost;
2256 
2257 		/*
2258 		 * The DOWN state must be protected from being pushed off
2259 		 * the queue to ensure that userland will always be
2260 		 * in a sane state.
2261 		 * Because DOWN is protected, there is no need to protect
2262 		 * UNKNOWN.
2263 		 * It should be invalid to change from any other state to
2264 		 * UNKNOWN anyway ...
2265 		 */
2266 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2267 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2268 		if (lost == LINK_STATE_DOWN) {
2269 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2270 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2271 		}
2272 		printf("%s: lost link state change %s\n",
2273 		    ifp->if_xname,
2274 		    lost == LINK_STATE_UP ? "UP" :
2275 		    lost == LINK_STATE_DOWN ? "DOWN" :
2276 		    "UNKNOWN");
2277 	} else
2278 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2279 
2280 	softint_schedule(ifp->if_link_si);
2281 
2282 out:
2283 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2284 }
2285 
2286 /*
2287  * Handle interface link state change notifications.
2288  */
2289 void
2290 if_link_state_change_softint(struct ifnet *ifp, int link_state)
2291 {
2292 	struct domain *dp;
2293 	int s = splnet();
2294 	bool notify;
2295 
2296 	KASSERT(!cpu_intr_p());
2297 
2298 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2299 
2300 	/* Ensure the change is still valid. */
2301 	if (ifp->if_link_state == link_state) {
2302 		IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2303 		return;
2304 	}
2305 
2306 #ifdef DEBUG
2307 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2308 		link_state == LINK_STATE_UP ? "UP" :
2309 		link_state == LINK_STATE_DOWN ? "DOWN" :
2310 		"UNKNOWN",
2311 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2312 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2313 		"UNKNOWN");
2314 #endif
2315 
2316 	/*
2317 	 * When going from UNKNOWN to UP, we need to mark existing
2318 	 * addresses as tentative and restart DAD as we may have
2319 	 * erroneously not found a duplicate.
2320 	 *
2321 	 * This needs to happen before rt_ifmsg to avoid a race where
2322 	 * listeners would have an address and expect it to work right
2323 	 * away.
2324 	 */
2325 	notify = (link_state == LINK_STATE_UP &&
2326 	    ifp->if_link_state == LINK_STATE_UNKNOWN);
2327 	ifp->if_link_state = link_state;
2328 	/* The following routines may sleep so release the spin mutex */
2329 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2330 
2331 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
2332 	if (notify) {
2333 		DOMAIN_FOREACH(dp) {
2334 			if (dp->dom_if_link_state_change != NULL)
2335 				dp->dom_if_link_state_change(ifp,
2336 				    LINK_STATE_DOWN);
2337 		}
2338 	}
2339 
2340 	/* Notify that the link state has changed. */
2341 	rt_ifmsg(ifp);
2342 
2343 #if NCARP > 0
2344 	if (ifp->if_carp)
2345 		carp_carpdev_state(ifp);
2346 #endif
2347 
2348 	DOMAIN_FOREACH(dp) {
2349 		if (dp->dom_if_link_state_change != NULL)
2350 			dp->dom_if_link_state_change(ifp, link_state);
2351 	}
2352 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2353 	splx(s);
2354 }
2355 
2356 /*
2357  * Process the interface link state change queue.
2358  */
2359 static void
2360 if_link_state_change_si(void *arg)
2361 {
2362 	struct ifnet *ifp = arg;
2363 	int s;
2364 	uint8_t state;
2365 	bool schedule;
2366 
2367 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2368 	s = splnet();
2369 
2370 	/* Pop a link state change from the queue and process it. */
2371 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2372 	LQ_POP(ifp->if_link_queue, state);
2373 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2374 
2375 	if_link_state_change_softint(ifp, state);
2376 
2377 	/* If there is a link state change to come, schedule it. */
2378 	IF_LINK_STATE_CHANGE_LOCK(ifp);
2379 	schedule = (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET);
2380 	IF_LINK_STATE_CHANGE_UNLOCK(ifp);
2381 	if (schedule)
2382 		softint_schedule(ifp->if_link_si);
2383 
2384 	splx(s);
2385 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2386 }
2387 
2388 /*
2389  * Default action when installing a local route on a point-to-point
2390  * interface.
2391  */
2392 void
2393 p2p_rtrequest(int req, struct rtentry *rt,
2394     __unused const struct rt_addrinfo *info)
2395 {
2396 	struct ifnet *ifp = rt->rt_ifp;
2397 	struct ifaddr *ifa, *lo0ifa;
2398 	int s = pserialize_read_enter();
2399 
2400 	switch (req) {
2401 	case RTM_ADD:
2402 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2403 			break;
2404 
2405 		rt->rt_ifp = lo0ifp;
2406 
2407 		IFADDR_READER_FOREACH(ifa, ifp) {
2408 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2409 				break;
2410 		}
2411 		if (ifa == NULL)
2412 			break;
2413 
2414 		/*
2415 		 * Ensure lo0 has an address of the same family.
2416 		 */
2417 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2418 			if (lo0ifa->ifa_addr->sa_family ==
2419 			    ifa->ifa_addr->sa_family)
2420 				break;
2421 		}
2422 		if (lo0ifa == NULL)
2423 			break;
2424 
2425 		/*
2426 		 * Make sure to set rt->rt_ifa to the interface
2427 		 * address we are using, otherwise we will have trouble
2428 		 * with source address selection.
2429 		 */
2430 		if (ifa != rt->rt_ifa)
2431 			rt_replace_ifa(rt, ifa);
2432 		break;
2433 	case RTM_DELETE:
2434 	default:
2435 		break;
2436 	}
2437 	pserialize_read_exit(s);
2438 }
2439 
2440 static void
2441 _if_down(struct ifnet *ifp)
2442 {
2443 	struct ifaddr *ifa;
2444 	struct domain *dp;
2445 	int s, bound;
2446 	struct psref psref;
2447 
2448 	ifp->if_flags &= ~IFF_UP;
2449 	nanotime(&ifp->if_lastchange);
2450 
2451 	bound = curlwp_bind();
2452 	s = pserialize_read_enter();
2453 	IFADDR_READER_FOREACH(ifa, ifp) {
2454 		ifa_acquire(ifa, &psref);
2455 		pserialize_read_exit(s);
2456 
2457 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2458 
2459 		s = pserialize_read_enter();
2460 		ifa_release(ifa, &psref);
2461 	}
2462 	pserialize_read_exit(s);
2463 	curlwp_bindx(bound);
2464 
2465 	IFQ_PURGE(&ifp->if_snd);
2466 #if NCARP > 0
2467 	if (ifp->if_carp)
2468 		carp_carpdev_state(ifp);
2469 #endif
2470 	rt_ifmsg(ifp);
2471 	DOMAIN_FOREACH(dp) {
2472 		if (dp->dom_if_down)
2473 			dp->dom_if_down(ifp);
2474 	}
2475 }
2476 
2477 static void
2478 if_down_deactivated(struct ifnet *ifp)
2479 {
2480 
2481 	KASSERT(if_is_deactivated(ifp));
2482 	_if_down(ifp);
2483 }
2484 
2485 void
2486 if_down_locked(struct ifnet *ifp)
2487 {
2488 
2489 	KASSERT(IFNET_LOCKED(ifp));
2490 	_if_down(ifp);
2491 }
2492 
2493 /*
2494  * Mark an interface down and notify protocols of
2495  * the transition.
2496  * NOTE: must be called at splsoftnet or equivalent.
2497  */
2498 void
2499 if_down(struct ifnet *ifp)
2500 {
2501 
2502 	IFNET_LOCK(ifp);
2503 	if_down_locked(ifp);
2504 	IFNET_UNLOCK(ifp);
2505 }
2506 
2507 /*
2508  * Must be called with holding if_ioctl_lock.
2509  */
2510 static void
2511 if_up_locked(struct ifnet *ifp)
2512 {
2513 #ifdef notyet
2514 	struct ifaddr *ifa;
2515 #endif
2516 	struct domain *dp;
2517 
2518 	KASSERT(IFNET_LOCKED(ifp));
2519 
2520 	KASSERT(!if_is_deactivated(ifp));
2521 	ifp->if_flags |= IFF_UP;
2522 	nanotime(&ifp->if_lastchange);
2523 #ifdef notyet
2524 	/* this has no effect on IP, and will kill all ISO connections XXX */
2525 	IFADDR_READER_FOREACH(ifa, ifp)
2526 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2527 #endif
2528 #if NCARP > 0
2529 	if (ifp->if_carp)
2530 		carp_carpdev_state(ifp);
2531 #endif
2532 	rt_ifmsg(ifp);
2533 	DOMAIN_FOREACH(dp) {
2534 		if (dp->dom_if_up)
2535 			dp->dom_if_up(ifp);
2536 	}
2537 }
2538 
2539 /*
2540  * Handle interface slowtimo timer routine.  Called
2541  * from softclock, we decrement timer (if set) and
2542  * call the appropriate interface routine on expiration.
2543  */
2544 static void
2545 if_slowtimo(void *arg)
2546 {
2547 	void (*slowtimo)(struct ifnet *);
2548 	struct ifnet *ifp = arg;
2549 	int s;
2550 
2551 	slowtimo = ifp->if_slowtimo;
2552 	if (__predict_false(slowtimo == NULL))
2553 		return;
2554 
2555 	s = splnet();
2556 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2557 		(*slowtimo)(ifp);
2558 
2559 	splx(s);
2560 
2561 	if (__predict_true(ifp->if_slowtimo != NULL))
2562 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2563 }
2564 
2565 /*
2566  * Mark an interface up and notify protocols of
2567  * the transition.
2568  * NOTE: must be called at splsoftnet or equivalent.
2569  */
2570 void
2571 if_up(struct ifnet *ifp)
2572 {
2573 
2574 	IFNET_LOCK(ifp);
2575 	if_up_locked(ifp);
2576 	IFNET_UNLOCK(ifp);
2577 }
2578 
2579 /*
2580  * Set/clear promiscuous mode on interface ifp based on the truth value
2581  * of pswitch.  The calls are reference counted so that only the first
2582  * "on" request actually has an effect, as does the final "off" request.
2583  * Results are undefined if the "off" and "on" requests are not matched.
2584  */
2585 int
2586 ifpromisc_locked(struct ifnet *ifp, int pswitch)
2587 {
2588 	int pcount, ret = 0;
2589 	short nflags;
2590 
2591 	KASSERT(IFNET_LOCKED(ifp));
2592 
2593 	pcount = ifp->if_pcount;
2594 	if (pswitch) {
2595 		/*
2596 		 * Allow the device to be "placed" into promiscuous
2597 		 * mode even if it is not configured up.  It will
2598 		 * consult IFF_PROMISC when it is brought up.
2599 		 */
2600 		if (ifp->if_pcount++ != 0)
2601 			goto out;
2602 		nflags = ifp->if_flags | IFF_PROMISC;
2603 	} else {
2604 		if (--ifp->if_pcount > 0)
2605 			goto out;
2606 		nflags = ifp->if_flags & ~IFF_PROMISC;
2607 	}
2608 	ret = if_flags_set(ifp, nflags);
2609 	/* Restore interface state if not successful. */
2610 	if (ret != 0) {
2611 		ifp->if_pcount = pcount;
2612 	}
2613 out:
2614 	return ret;
2615 }
2616 
2617 int
2618 ifpromisc(struct ifnet *ifp, int pswitch)
2619 {
2620 	int e;
2621 
2622 	IFNET_LOCK(ifp);
2623 	e = ifpromisc_locked(ifp, pswitch);
2624 	IFNET_UNLOCK(ifp);
2625 
2626 	return e;
2627 }
2628 
2629 /*
2630  * Map interface name to
2631  * interface structure pointer.
2632  */
2633 struct ifnet *
2634 ifunit(const char *name)
2635 {
2636 	struct ifnet *ifp;
2637 	const char *cp = name;
2638 	u_int unit = 0;
2639 	u_int i;
2640 	int s;
2641 
2642 	/*
2643 	 * If the entire name is a number, treat it as an ifindex.
2644 	 */
2645 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2646 		unit = unit * 10 + (*cp - '0');
2647 	}
2648 
2649 	/*
2650 	 * If the number took all of the name, then it's a valid ifindex.
2651 	 */
2652 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2653 		return if_byindex(unit);
2654 
2655 	ifp = NULL;
2656 	s = pserialize_read_enter();
2657 	IFNET_READER_FOREACH(ifp) {
2658 		if (if_is_deactivated(ifp))
2659 			continue;
2660 	 	if (strcmp(ifp->if_xname, name) == 0)
2661 			goto out;
2662 	}
2663 out:
2664 	pserialize_read_exit(s);
2665 	return ifp;
2666 }
2667 
2668 /*
2669  * Get a reference of an ifnet object by an interface name.
2670  * The returned reference is protected by psref(9). The caller
2671  * must release a returned reference by if_put after use.
2672  */
2673 struct ifnet *
2674 if_get(const char *name, struct psref *psref)
2675 {
2676 	struct ifnet *ifp;
2677 	const char *cp = name;
2678 	u_int unit = 0;
2679 	u_int i;
2680 	int s;
2681 
2682 	/*
2683 	 * If the entire name is a number, treat it as an ifindex.
2684 	 */
2685 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2686 		unit = unit * 10 + (*cp - '0');
2687 	}
2688 
2689 	/*
2690 	 * If the number took all of the name, then it's a valid ifindex.
2691 	 */
2692 	if (i == IFNAMSIZ || (cp != name && *cp == '\0'))
2693 		return if_get_byindex(unit, psref);
2694 
2695 	ifp = NULL;
2696 	s = pserialize_read_enter();
2697 	IFNET_READER_FOREACH(ifp) {
2698 		if (if_is_deactivated(ifp))
2699 			continue;
2700 		if (strcmp(ifp->if_xname, name) == 0) {
2701 			psref_acquire(psref, &ifp->if_psref,
2702 			    ifnet_psref_class);
2703 			goto out;
2704 		}
2705 	}
2706 out:
2707 	pserialize_read_exit(s);
2708 	return ifp;
2709 }
2710 
2711 /*
2712  * Release a reference of an ifnet object given by if_get, if_get_byindex
2713  * or if_get_bylla.
2714  */
2715 void
2716 if_put(const struct ifnet *ifp, struct psref *psref)
2717 {
2718 
2719 	if (ifp == NULL)
2720 		return;
2721 
2722 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2723 }
2724 
2725 /*
2726  * Return ifp having idx. Return NULL if not found.  Normally if_byindex
2727  * should be used.
2728  */
2729 ifnet_t *
2730 _if_byindex(u_int idx)
2731 {
2732 
2733 	return (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2734 }
2735 
2736 /*
2737  * Return ifp having idx. Return NULL if not found or the found ifp is
2738  * already deactivated.
2739  */
2740 ifnet_t *
2741 if_byindex(u_int idx)
2742 {
2743 	ifnet_t *ifp;
2744 
2745 	ifp = _if_byindex(idx);
2746 	if (ifp != NULL && if_is_deactivated(ifp))
2747 		ifp = NULL;
2748 	return ifp;
2749 }
2750 
2751 /*
2752  * Get a reference of an ifnet object by an interface index.
2753  * The returned reference is protected by psref(9). The caller
2754  * must release a returned reference by if_put after use.
2755  */
2756 ifnet_t *
2757 if_get_byindex(u_int idx, struct psref *psref)
2758 {
2759 	ifnet_t *ifp;
2760 	int s;
2761 
2762 	s = pserialize_read_enter();
2763 	ifp = if_byindex(idx);
2764 	if (__predict_true(ifp != NULL))
2765 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2766 	pserialize_read_exit(s);
2767 
2768 	return ifp;
2769 }
2770 
2771 ifnet_t *
2772 if_get_bylla(const void *lla, unsigned char lla_len, struct psref *psref)
2773 {
2774 	ifnet_t *ifp;
2775 	int s;
2776 
2777 	s = pserialize_read_enter();
2778 	IFNET_READER_FOREACH(ifp) {
2779 		if (if_is_deactivated(ifp))
2780 			continue;
2781 		if (ifp->if_addrlen != lla_len)
2782 			continue;
2783 		if (memcmp(lla, CLLADDR(ifp->if_sadl), lla_len) == 0) {
2784 			psref_acquire(psref, &ifp->if_psref,
2785 			    ifnet_psref_class);
2786 			break;
2787 		}
2788 	}
2789 	pserialize_read_exit(s);
2790 
2791 	return ifp;
2792 }
2793 
2794 /*
2795  * Note that it's safe only if the passed ifp is guaranteed to not be freed,
2796  * for example using pserialize or the ifp is already held or some other
2797  * object is held which guarantes the ifp to not be freed indirectly.
2798  */
2799 void
2800 if_acquire(struct ifnet *ifp, struct psref *psref)
2801 {
2802 
2803 	KASSERT(ifp->if_index != 0);
2804 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2805 }
2806 
2807 bool
2808 if_held(struct ifnet *ifp)
2809 {
2810 
2811 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2812 }
2813 
2814 /*
2815  * Some tunnel interfaces can nest, e.g. IPv4 over IPv4 gif(4) tunnel over IPv4.
2816  * Check the tunnel nesting count.
2817  * Return > 0, if tunnel nesting count is more than limit.
2818  * Return 0, if tunnel nesting count is equal or less than limit.
2819  */
2820 int
2821 if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, int limit)
2822 {
2823 	struct m_tag *mtag;
2824 	int *count;
2825 
2826 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL);
2827 	if (mtag != NULL) {
2828 		count = (int *)(mtag + 1);
2829 		if (++(*count) > limit) {
2830 			log(LOG_NOTICE,
2831 			    "%s: recursively called too many times(%d)\n",
2832 			    ifp->if_xname, *count);
2833 			return EIO;
2834 		}
2835 	} else {
2836 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
2837 		    M_NOWAIT);
2838 		if (mtag != NULL) {
2839 			m_tag_prepend(m, mtag);
2840 			count = (int *)(mtag + 1);
2841 			*count = 0;
2842 		} else {
2843 			log(LOG_DEBUG,
2844 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
2845 			    ifp->if_xname);
2846 		}
2847 	}
2848 
2849 	return 0;
2850 }
2851 
2852 /* common */
2853 int
2854 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2855 {
2856 	int s;
2857 	struct ifreq *ifr;
2858 	struct ifcapreq *ifcr;
2859 	struct ifdatareq *ifdr;
2860 
2861 	switch (cmd) {
2862 	case SIOCSIFCAP:
2863 		ifcr = data;
2864 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2865 			return EINVAL;
2866 
2867 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2868 			return 0;
2869 
2870 		ifp->if_capenable = ifcr->ifcr_capenable;
2871 
2872 		/* Pre-compute the checksum flags mask. */
2873 		ifp->if_csum_flags_tx = 0;
2874 		ifp->if_csum_flags_rx = 0;
2875 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2876 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2877 		}
2878 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2879 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2880 		}
2881 
2882 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2883 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2884 		}
2885 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2886 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2887 		}
2888 
2889 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2890 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2891 		}
2892 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2893 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2894 		}
2895 
2896 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2897 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2898 		}
2899 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2900 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2901 		}
2902 
2903 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2904 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2905 		}
2906 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2907 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2908 		}
2909 		if (ifp->if_flags & IFF_UP)
2910 			return ENETRESET;
2911 		return 0;
2912 	case SIOCSIFFLAGS:
2913 		ifr = data;
2914 		/*
2915 		 * If if_is_mpsafe(ifp), KERNEL_LOCK isn't held here, but if_up
2916 		 * and if_down aren't MP-safe yet, so we must hold the lock.
2917 		 */
2918 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
2919 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2920 			s = splsoftnet();
2921 			if_down_locked(ifp);
2922 			splx(s);
2923 		}
2924 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2925 			s = splsoftnet();
2926 			if_up_locked(ifp);
2927 			splx(s);
2928 		}
2929 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
2930 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2931 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2932 		break;
2933 	case SIOCGIFFLAGS:
2934 		ifr = data;
2935 		ifr->ifr_flags = ifp->if_flags;
2936 		break;
2937 
2938 	case SIOCGIFMETRIC:
2939 		ifr = data;
2940 		ifr->ifr_metric = ifp->if_metric;
2941 		break;
2942 
2943 	case SIOCGIFMTU:
2944 		ifr = data;
2945 		ifr->ifr_mtu = ifp->if_mtu;
2946 		break;
2947 
2948 	case SIOCGIFDLT:
2949 		ifr = data;
2950 		ifr->ifr_dlt = ifp->if_dlt;
2951 		break;
2952 
2953 	case SIOCGIFCAP:
2954 		ifcr = data;
2955 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2956 		ifcr->ifcr_capenable = ifp->if_capenable;
2957 		break;
2958 
2959 	case SIOCSIFMETRIC:
2960 		ifr = data;
2961 		ifp->if_metric = ifr->ifr_metric;
2962 		break;
2963 
2964 	case SIOCGIFDATA:
2965 		ifdr = data;
2966 		ifdr->ifdr_data = ifp->if_data;
2967 		break;
2968 
2969 	case SIOCGIFINDEX:
2970 		ifr = data;
2971 		ifr->ifr_index = ifp->if_index;
2972 		break;
2973 
2974 	case SIOCZIFDATA:
2975 		ifdr = data;
2976 		ifdr->ifdr_data = ifp->if_data;
2977 		/*
2978 		 * Assumes that the volatile counters that can be
2979 		 * zero'ed are at the end of if_data.
2980 		 */
2981 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
2982 		    offsetof(struct if_data, ifi_ipackets));
2983 		/*
2984 		 * The memset() clears to the bottm of if_data. In the area,
2985 		 * if_lastchange is included. Please be careful if new entry
2986 		 * will be added into if_data or rewite this.
2987 		 *
2988 		 * And also, update if_lastchnage.
2989 		 */
2990 		getnanotime(&ifp->if_lastchange);
2991 		break;
2992 	case SIOCSIFMTU:
2993 		ifr = data;
2994 		if (ifp->if_mtu == ifr->ifr_mtu)
2995 			break;
2996 		ifp->if_mtu = ifr->ifr_mtu;
2997 		/*
2998 		 * If the link MTU changed, do network layer specific procedure.
2999 		 */
3000 #ifdef INET6
3001 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3002 		if (in6_present)
3003 			nd6_setmtu(ifp);
3004 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3005 #endif
3006 		return ENETRESET;
3007 	default:
3008 		return ENOTTY;
3009 	}
3010 	return 0;
3011 }
3012 
3013 int
3014 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
3015 {
3016 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
3017 	struct ifaddr *ifa;
3018 	const struct sockaddr *any, *sa;
3019 	union {
3020 		struct sockaddr sa;
3021 		struct sockaddr_storage ss;
3022 	} u, v;
3023 	int s, error = 0;
3024 
3025 	switch (cmd) {
3026 	case SIOCSIFADDRPREF:
3027 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
3028 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
3029 		    NULL) != 0)
3030 			return EPERM;
3031 	case SIOCGIFADDRPREF:
3032 		break;
3033 	default:
3034 		return EOPNOTSUPP;
3035 	}
3036 
3037 	/* sanity checks */
3038 	if (data == NULL || ifp == NULL) {
3039 		panic("invalid argument to %s", __func__);
3040 		/*NOTREACHED*/
3041 	}
3042 
3043 	/* address must be specified on ADD and DELETE */
3044 	sa = sstocsa(&ifap->ifap_addr);
3045 	if (sa->sa_family != sofamily(so))
3046 		return EINVAL;
3047 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
3048 		return EINVAL;
3049 
3050 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
3051 
3052 	s = pserialize_read_enter();
3053 	IFADDR_READER_FOREACH(ifa, ifp) {
3054 		if (ifa->ifa_addr->sa_family != sa->sa_family)
3055 			continue;
3056 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
3057 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
3058 			break;
3059 	}
3060 	if (ifa == NULL) {
3061 		error = EADDRNOTAVAIL;
3062 		goto out;
3063 	}
3064 
3065 	switch (cmd) {
3066 	case SIOCSIFADDRPREF:
3067 		ifa->ifa_preference = ifap->ifap_preference;
3068 		goto out;
3069 	case SIOCGIFADDRPREF:
3070 		/* fill in the if_laddrreq structure */
3071 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
3072 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
3073 		ifap->ifap_preference = ifa->ifa_preference;
3074 		goto out;
3075 	default:
3076 		error = EOPNOTSUPP;
3077 	}
3078 out:
3079 	pserialize_read_exit(s);
3080 	return error;
3081 }
3082 
3083 /*
3084  * Interface ioctls.
3085  */
3086 static int
3087 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
3088 {
3089 	struct ifnet *ifp;
3090 	struct ifreq *ifr;
3091 	int error = 0;
3092 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
3093 	u_long ocmd = cmd;
3094 #endif
3095 	short oif_flags;
3096 #ifdef COMPAT_OIFREQ
3097 	struct ifreq ifrb;
3098 	struct oifreq *oifr = NULL;
3099 #endif
3100 	int r;
3101 	struct psref psref;
3102 	int bound;
3103 
3104 	switch (cmd) {
3105 	case SIOCGIFCONF:
3106 		return ifconf(cmd, data);
3107 	case SIOCINITIFADDR:
3108 		return EPERM;
3109 	default:
3110 		error = (*vec_compat_ifconf)(l, cmd, data);
3111 		if (error != ENOSYS)
3112 			return error;
3113 		error = (*vec_compat_ifdatareq)(l, cmd, data);
3114 		if (error != ENOSYS)
3115 			return error;
3116 		break;
3117 	}
3118 
3119 	ifr = data;
3120 #ifdef COMPAT_OIFREQ
3121 	if (vec_compat_cvtcmd) {
3122 		cmd = (*vec_compat_cvtcmd)(cmd);
3123 		if (cmd != ocmd) {
3124 			oifr = data;
3125 			data = ifr = &ifrb;
3126 			ifreqo2n(oifr, ifr);
3127 		}
3128 	}
3129 #endif
3130 
3131 	switch (cmd) {
3132 	case SIOCIFCREATE:
3133 	case SIOCIFDESTROY:
3134 		bound = curlwp_bind();
3135 		if (l != NULL) {
3136 			ifp = if_get(ifr->ifr_name, &psref);
3137 			error = kauth_authorize_network(l->l_cred,
3138 			    KAUTH_NETWORK_INTERFACE,
3139 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3140 			    (void *)cmd, NULL);
3141 			if (ifp != NULL)
3142 				if_put(ifp, &psref);
3143 			if (error != 0) {
3144 				curlwp_bindx(bound);
3145 				return error;
3146 			}
3147 		}
3148 		KERNEL_LOCK_UNLESS_NET_MPSAFE();
3149 		mutex_enter(&if_clone_mtx);
3150 		r = (cmd == SIOCIFCREATE) ?
3151 			if_clone_create(ifr->ifr_name) :
3152 			if_clone_destroy(ifr->ifr_name);
3153 		mutex_exit(&if_clone_mtx);
3154 		KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
3155 		curlwp_bindx(bound);
3156 		return r;
3157 
3158 	case SIOCIFGCLONERS:
3159 		{
3160 			struct if_clonereq *req = (struct if_clonereq *)data;
3161 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
3162 			    &req->ifcr_total);
3163 		}
3164 	}
3165 
3166 	bound = curlwp_bind();
3167 	ifp = if_get(ifr->ifr_name, &psref);
3168 	if (ifp == NULL) {
3169 		curlwp_bindx(bound);
3170 		return ENXIO;
3171 	}
3172 
3173 	switch (cmd) {
3174 	case SIOCALIFADDR:
3175 	case SIOCDLIFADDR:
3176 	case SIOCSIFADDRPREF:
3177 	case SIOCSIFFLAGS:
3178 	case SIOCSIFCAP:
3179 	case SIOCSIFMETRIC:
3180 	case SIOCZIFDATA:
3181 	case SIOCSIFMTU:
3182 	case SIOCSIFPHYADDR:
3183 	case SIOCDIFPHYADDR:
3184 #ifdef INET6
3185 	case SIOCSIFPHYADDR_IN6:
3186 #endif
3187 	case SIOCSLIFPHYADDR:
3188 	case SIOCADDMULTI:
3189 	case SIOCDELMULTI:
3190 	case SIOCSIFMEDIA:
3191 	case SIOCSDRVSPEC:
3192 	case SIOCG80211:
3193 	case SIOCS80211:
3194 	case SIOCS80211NWID:
3195 	case SIOCS80211NWKEY:
3196 	case SIOCS80211POWER:
3197 	case SIOCS80211BSSID:
3198 	case SIOCS80211CHANNEL:
3199 	case SIOCSLINKSTR:
3200 		if (l != NULL) {
3201 			error = kauth_authorize_network(l->l_cred,
3202 			    KAUTH_NETWORK_INTERFACE,
3203 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
3204 			    (void *)cmd, NULL);
3205 			if (error != 0)
3206 				goto out;
3207 		}
3208 	}
3209 
3210 	oif_flags = ifp->if_flags;
3211 
3212 	KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp);
3213 	IFNET_LOCK(ifp);
3214 
3215 	error = (*ifp->if_ioctl)(ifp, cmd, data);
3216 	if (error != ENOTTY)
3217 		;
3218 	else if (so->so_proto == NULL)
3219 		error = EOPNOTSUPP;
3220 	else {
3221 		KERNEL_LOCK_IF_IFP_MPSAFE(ifp);
3222 #ifdef COMPAT_OSOCK
3223 		if (vec_compat_ifioctl != NULL)
3224 			error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l);
3225 		else
3226 #endif
3227 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
3228 			    cmd, data, ifp);
3229 		KERNEL_UNLOCK_IF_IFP_MPSAFE(ifp);
3230 	}
3231 
3232 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
3233 		if ((ifp->if_flags & IFF_UP) != 0) {
3234 			int s = splsoftnet();
3235 			if_up_locked(ifp);
3236 			splx(s);
3237 		}
3238 	}
3239 #ifdef COMPAT_OIFREQ
3240 	if (cmd != ocmd)
3241 		ifreqn2o(oifr, ifr);
3242 #endif
3243 
3244 	IFNET_UNLOCK(ifp);
3245 	KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp);
3246 out:
3247 	if_put(ifp, &psref);
3248 	curlwp_bindx(bound);
3249 	return error;
3250 }
3251 
3252 /*
3253  * Return interface configuration
3254  * of system.  List may be used
3255  * in later ioctl's (above) to get
3256  * other information.
3257  *
3258  * Each record is a struct ifreq.  Before the addition of
3259  * sockaddr_storage, the API rule was that sockaddr flavors that did
3260  * not fit would extend beyond the struct ifreq, with the next struct
3261  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3262  * union in struct ifreq includes struct sockaddr_storage, every kind
3263  * of sockaddr must fit.  Thus, there are no longer any overlength
3264  * records.
3265  *
3266  * Records are added to the user buffer if they fit, and ifc_len is
3267  * adjusted to the length that was written.  Thus, the user is only
3268  * assured of getting the complete list if ifc_len on return is at
3269  * least sizeof(struct ifreq) less than it was on entry.
3270  *
3271  * If the user buffer pointer is NULL, this routine copies no data and
3272  * returns the amount of space that would be needed.
3273  *
3274  * Invariants:
3275  * ifrp points to the next part of the user's buffer to be used.  If
3276  * ifrp != NULL, space holds the number of bytes remaining that we may
3277  * write at ifrp.  Otherwise, space holds the number of bytes that
3278  * would have been written had there been adequate space.
3279  */
3280 /*ARGSUSED*/
3281 static int
3282 ifconf(u_long cmd, void *data)
3283 {
3284 	struct ifconf *ifc = (struct ifconf *)data;
3285 	struct ifnet *ifp;
3286 	struct ifaddr *ifa;
3287 	struct ifreq ifr, *ifrp = NULL;
3288 	int space = 0, error = 0;
3289 	const int sz = (int)sizeof(struct ifreq);
3290 	const bool docopy = ifc->ifc_req != NULL;
3291 	int s;
3292 	int bound;
3293 	struct psref psref;
3294 
3295 	if (docopy) {
3296 		space = ifc->ifc_len;
3297 		ifrp = ifc->ifc_req;
3298 	}
3299 
3300 	bound = curlwp_bind();
3301 	s = pserialize_read_enter();
3302 	IFNET_READER_FOREACH(ifp) {
3303 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3304 		pserialize_read_exit(s);
3305 
3306 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3307 		    sizeof(ifr.ifr_name));
3308 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3309 			error = ENAMETOOLONG;
3310 			goto release_exit;
3311 		}
3312 		if (IFADDR_READER_EMPTY(ifp)) {
3313 			/* Interface with no addresses - send zero sockaddr. */
3314 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3315 			if (!docopy) {
3316 				space += sz;
3317 				goto next;
3318 			}
3319 			if (space >= sz) {
3320 				error = copyout(&ifr, ifrp, sz);
3321 				if (error != 0)
3322 					goto release_exit;
3323 				ifrp++;
3324 				space -= sz;
3325 			}
3326 		}
3327 
3328 		s = pserialize_read_enter();
3329 		IFADDR_READER_FOREACH(ifa, ifp) {
3330 			struct sockaddr *sa = ifa->ifa_addr;
3331 			/* all sockaddrs must fit in sockaddr_storage */
3332 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3333 
3334 			if (!docopy) {
3335 				space += sz;
3336 				continue;
3337 			}
3338 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3339 			pserialize_read_exit(s);
3340 
3341 			if (space >= sz) {
3342 				error = copyout(&ifr, ifrp, sz);
3343 				if (error != 0)
3344 					goto release_exit;
3345 				ifrp++; space -= sz;
3346 			}
3347 			s = pserialize_read_enter();
3348 		}
3349 		pserialize_read_exit(s);
3350 
3351         next:
3352 		s = pserialize_read_enter();
3353 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3354 	}
3355 	pserialize_read_exit(s);
3356 	curlwp_bindx(bound);
3357 
3358 	if (docopy) {
3359 		KASSERT(0 <= space && space <= ifc->ifc_len);
3360 		ifc->ifc_len -= space;
3361 	} else {
3362 		KASSERT(space >= 0);
3363 		ifc->ifc_len = space;
3364 	}
3365 	return (0);
3366 
3367 release_exit:
3368 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3369 	curlwp_bindx(bound);
3370 	return error;
3371 }
3372 
3373 int
3374 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3375 {
3376 	uint8_t len = sizeof(ifr->ifr_ifru.ifru_space);
3377 #ifdef COMPAT_OIFREQ
3378 	struct ifreq ifrb;
3379 	struct oifreq *oifr = NULL;
3380 	u_long ocmd = cmd;
3381 
3382 	if (vec_compat_cvtcmd) {
3383 		    cmd = (*vec_compat_cvtcmd)(cmd);
3384 		    if (cmd != ocmd) {
3385 			    oifr = (struct oifreq *)(void *)ifr;
3386 			    ifr = &ifrb;
3387 			    ifreqo2n(oifr, ifr);
3388 			    len = sizeof(oifr->ifr_addr);
3389 		    }
3390 	}
3391 #endif
3392 	if (len < sa->sa_len)
3393 		return EFBIG;
3394 
3395 	memset(&ifr->ifr_addr, 0, len);
3396 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3397 
3398 #ifdef COMPAT_OIFREQ
3399 	if (cmd != ocmd)
3400 		ifreqn2o(oifr, ifr);
3401 #endif
3402 	return 0;
3403 }
3404 
3405 /*
3406  * wrapper function for the drivers which doesn't have if_transmit().
3407  */
3408 static int
3409 if_transmit(struct ifnet *ifp, struct mbuf *m)
3410 {
3411 	int s, error;
3412 	size_t pktlen = m->m_pkthdr.len;
3413 	bool mcast = (m->m_flags & M_MCAST) != 0;
3414 
3415 	s = splnet();
3416 
3417 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3418 	if (error != 0) {
3419 		/* mbuf is already freed */
3420 		goto out;
3421 	}
3422 
3423 	ifp->if_obytes += pktlen;
3424 	if (mcast)
3425 		ifp->if_omcasts++;
3426 
3427 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3428 		if_start_lock(ifp);
3429 out:
3430 	splx(s);
3431 
3432 	return error;
3433 }
3434 
3435 int
3436 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3437 {
3438 	int error;
3439 
3440 #ifdef ALTQ
3441 	KERNEL_LOCK(1, NULL);
3442 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3443 		error = if_transmit(ifp, m);
3444 		KERNEL_UNLOCK_ONE(NULL);
3445 	} else {
3446 		KERNEL_UNLOCK_ONE(NULL);
3447 		error = (*ifp->if_transmit)(ifp, m);
3448 		/* mbuf is alredy freed */
3449 	}
3450 #else /* !ALTQ */
3451 	error = (*ifp->if_transmit)(ifp, m);
3452 	/* mbuf is alredy freed */
3453 #endif /* !ALTQ */
3454 
3455 	return error;
3456 }
3457 
3458 /*
3459  * Queue message on interface, and start output if interface
3460  * not yet active.
3461  */
3462 int
3463 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3464 {
3465 
3466 	return if_transmit_lock(ifp, m);
3467 }
3468 
3469 /*
3470  * Queue message on interface, possibly using a second fast queue
3471  */
3472 int
3473 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3474 {
3475 	int error = 0;
3476 
3477 	if (ifq != NULL
3478 #ifdef ALTQ
3479 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3480 #endif
3481 	    ) {
3482 		if (IF_QFULL(ifq)) {
3483 			IF_DROP(&ifp->if_snd);
3484 			m_freem(m);
3485 			if (error == 0)
3486 				error = ENOBUFS;
3487 		} else
3488 			IF_ENQUEUE(ifq, m);
3489 	} else
3490 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3491 	if (error != 0) {
3492 		++ifp->if_oerrors;
3493 		return error;
3494 	}
3495 	return 0;
3496 }
3497 
3498 int
3499 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3500 {
3501 	int rc;
3502 
3503 	KASSERT(IFNET_LOCKED(ifp));
3504 	if (ifp->if_initaddr != NULL)
3505 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3506 	else if (src ||
3507 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3508 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3509 
3510 	return rc;
3511 }
3512 
3513 int
3514 if_do_dad(struct ifnet *ifp)
3515 {
3516 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3517 		return 0;
3518 
3519 	switch (ifp->if_type) {
3520 	case IFT_FAITH:
3521 		/*
3522 		 * These interfaces do not have the IFF_LOOPBACK flag,
3523 		 * but loop packets back.  We do not have to do DAD on such
3524 		 * interfaces.  We should even omit it, because loop-backed
3525 		 * responses would confuse the DAD procedure.
3526 		 */
3527 		return 0;
3528 	default:
3529 		/*
3530 		 * Our DAD routine requires the interface up and running.
3531 		 * However, some interfaces can be up before the RUNNING
3532 		 * status.  Additionaly, users may try to assign addresses
3533 		 * before the interface becomes up (or running).
3534 		 * We simply skip DAD in such a case as a work around.
3535 		 * XXX: we should rather mark "tentative" on such addresses,
3536 		 * and do DAD after the interface becomes ready.
3537 		 */
3538 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
3539 		    (IFF_UP|IFF_RUNNING))
3540 			return 0;
3541 
3542 		return 1;
3543 	}
3544 }
3545 
3546 int
3547 if_flags_set(ifnet_t *ifp, const short flags)
3548 {
3549 	int rc;
3550 
3551 	KASSERT(IFNET_LOCKED(ifp));
3552 
3553 	if (ifp->if_setflags != NULL)
3554 		rc = (*ifp->if_setflags)(ifp, flags);
3555 	else {
3556 		short cantflags, chgdflags;
3557 		struct ifreq ifr;
3558 
3559 		chgdflags = ifp->if_flags ^ flags;
3560 		cantflags = chgdflags & IFF_CANTCHANGE;
3561 
3562 		if (cantflags != 0)
3563 			ifp->if_flags ^= cantflags;
3564 
3565                 /* Traditionally, we do not call if_ioctl after
3566                  * setting/clearing only IFF_PROMISC if the interface
3567                  * isn't IFF_UP.  Uphold that tradition.
3568 		 */
3569 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3570 			return 0;
3571 
3572 		memset(&ifr, 0, sizeof(ifr));
3573 
3574 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3575 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3576 
3577 		if (rc != 0 && cantflags != 0)
3578 			ifp->if_flags ^= cantflags;
3579 	}
3580 
3581 	return rc;
3582 }
3583 
3584 int
3585 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3586 {
3587 	int rc;
3588 	struct ifreq ifr;
3589 
3590 	/* CARP and MROUTING still don't deal with the lock yet */
3591 #if (!defined(NCARP) || (NCARP == 0)) && !defined(MROUTING)
3592 	KASSERT(IFNET_LOCKED(ifp));
3593 #endif
3594 	if (ifp->if_mcastop != NULL)
3595 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3596 	else {
3597 		ifreq_setaddr(cmd, &ifr, sa);
3598 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3599 	}
3600 
3601 	return rc;
3602 }
3603 
3604 static void
3605 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3606     struct ifaltq *ifq)
3607 {
3608 	const struct sysctlnode *cnode, *rnode;
3609 
3610 	if (sysctl_createv(clog, 0, NULL, &rnode,
3611 		       CTLFLAG_PERMANENT,
3612 		       CTLTYPE_NODE, "interfaces",
3613 		       SYSCTL_DESCR("Per-interface controls"),
3614 		       NULL, 0, NULL, 0,
3615 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3616 		goto bad;
3617 
3618 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3619 		       CTLFLAG_PERMANENT,
3620 		       CTLTYPE_NODE, ifname,
3621 		       SYSCTL_DESCR("Interface controls"),
3622 		       NULL, 0, NULL, 0,
3623 		       CTL_CREATE, CTL_EOL) != 0)
3624 		goto bad;
3625 
3626 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3627 		       CTLFLAG_PERMANENT,
3628 		       CTLTYPE_NODE, "sndq",
3629 		       SYSCTL_DESCR("Interface output queue controls"),
3630 		       NULL, 0, NULL, 0,
3631 		       CTL_CREATE, CTL_EOL) != 0)
3632 		goto bad;
3633 
3634 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3635 		       CTLFLAG_PERMANENT,
3636 		       CTLTYPE_INT, "len",
3637 		       SYSCTL_DESCR("Current output queue length"),
3638 		       NULL, 0, &ifq->ifq_len, 0,
3639 		       CTL_CREATE, CTL_EOL) != 0)
3640 		goto bad;
3641 
3642 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3643 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3644 		       CTLTYPE_INT, "maxlen",
3645 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3646 		       NULL, 0, &ifq->ifq_maxlen, 0,
3647 		       CTL_CREATE, CTL_EOL) != 0)
3648 		goto bad;
3649 
3650 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3651 		       CTLFLAG_PERMANENT,
3652 		       CTLTYPE_INT, "drops",
3653 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3654 		       NULL, 0, &ifq->ifq_drops, 0,
3655 		       CTL_CREATE, CTL_EOL) != 0)
3656 		goto bad;
3657 
3658 	return;
3659 bad:
3660 	printf("%s: could not attach sysctl nodes\n", ifname);
3661 	return;
3662 }
3663 
3664 #if defined(INET) || defined(INET6)
3665 
3666 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3667 	static int							\
3668 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3669 	{								\
3670 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3671 	}
3672 
3673 #if defined(INET)
3674 static int
3675 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3676 {
3677 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3678 }
3679 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3680 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3681 #endif
3682 
3683 #if defined(INET6)
3684 static int
3685 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3686 {
3687 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3688 }
3689 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3690 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3691 #endif
3692 
3693 static void
3694 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3695 {
3696 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3697 	const char *pfname = NULL, *ipname = NULL;
3698 	int ipn = 0, qid = 0;
3699 
3700 	switch (pf) {
3701 #if defined(INET)
3702 	case PF_INET:
3703 		len_func = sysctl_net_ip_pktq_items;
3704 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3705 		drops_func = sysctl_net_ip_pktq_drops;
3706 		pfname = "inet", ipn = IPPROTO_IP;
3707 		ipname = "ip", qid = IPCTL_IFQ;
3708 		break;
3709 #endif
3710 #if defined(INET6)
3711 	case PF_INET6:
3712 		len_func = sysctl_net_ip6_pktq_items;
3713 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3714 		drops_func = sysctl_net_ip6_pktq_drops;
3715 		pfname = "inet6", ipn = IPPROTO_IPV6;
3716 		ipname = "ip6", qid = IPV6CTL_IFQ;
3717 		break;
3718 #endif
3719 	default:
3720 		KASSERT(false);
3721 	}
3722 
3723 	sysctl_createv(clog, 0, NULL, NULL,
3724 		       CTLFLAG_PERMANENT,
3725 		       CTLTYPE_NODE, pfname, NULL,
3726 		       NULL, 0, NULL, 0,
3727 		       CTL_NET, pf, CTL_EOL);
3728 	sysctl_createv(clog, 0, NULL, NULL,
3729 		       CTLFLAG_PERMANENT,
3730 		       CTLTYPE_NODE, ipname, NULL,
3731 		       NULL, 0, NULL, 0,
3732 		       CTL_NET, pf, ipn, CTL_EOL);
3733 	sysctl_createv(clog, 0, NULL, NULL,
3734 		       CTLFLAG_PERMANENT,
3735 		       CTLTYPE_NODE, "ifq",
3736 		       SYSCTL_DESCR("Protocol input queue controls"),
3737 		       NULL, 0, NULL, 0,
3738 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3739 
3740 	sysctl_createv(clog, 0, NULL, NULL,
3741 		       CTLFLAG_PERMANENT,
3742 		       CTLTYPE_INT, "len",
3743 		       SYSCTL_DESCR("Current input queue length"),
3744 		       len_func, 0, NULL, 0,
3745 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3746 	sysctl_createv(clog, 0, NULL, NULL,
3747 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3748 		       CTLTYPE_INT, "maxlen",
3749 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3750 		       maxlen_func, 0, NULL, 0,
3751 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3752 	sysctl_createv(clog, 0, NULL, NULL,
3753 		       CTLFLAG_PERMANENT,
3754 		       CTLTYPE_INT, "drops",
3755 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3756 		       drops_func, 0, NULL, 0,
3757 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3758 }
3759 #endif /* INET || INET6 */
3760 
3761 static int
3762 if_sdl_sysctl(SYSCTLFN_ARGS)
3763 {
3764 	struct ifnet *ifp;
3765 	const struct sockaddr_dl *sdl;
3766 	struct psref psref;
3767 	int error = 0;
3768 	int bound;
3769 
3770 	if (namelen != 1)
3771 		return EINVAL;
3772 
3773 	bound = curlwp_bind();
3774 	ifp = if_get_byindex(name[0], &psref);
3775 	if (ifp == NULL) {
3776 		error = ENODEV;
3777 		goto out0;
3778 	}
3779 
3780 	sdl = ifp->if_sadl;
3781 	if (sdl == NULL) {
3782 		*oldlenp = 0;
3783 		goto out1;
3784 	}
3785 
3786 	if (oldp == NULL) {
3787 		*oldlenp = sdl->sdl_alen;
3788 		goto out1;
3789 	}
3790 
3791 	if (*oldlenp >= sdl->sdl_alen)
3792 		*oldlenp = sdl->sdl_alen;
3793 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3794 out1:
3795 	if_put(ifp, &psref);
3796 out0:
3797 	curlwp_bindx(bound);
3798 	return error;
3799 }
3800 
3801 static void
3802 if_sysctl_setup(struct sysctllog **clog)
3803 {
3804 	const struct sysctlnode *rnode = NULL;
3805 
3806 	sysctl_createv(clog, 0, NULL, &rnode,
3807 		       CTLFLAG_PERMANENT,
3808 		       CTLTYPE_NODE, "sdl",
3809 		       SYSCTL_DESCR("Get active link-layer address"),
3810 		       if_sdl_sysctl, 0, NULL, 0,
3811 		       CTL_NET, CTL_CREATE, CTL_EOL);
3812 
3813 #if defined(INET)
3814 	sysctl_net_pktq_setup(NULL, PF_INET);
3815 #endif
3816 #ifdef INET6
3817 	if (in6_present)
3818 		sysctl_net_pktq_setup(NULL, PF_INET6);
3819 #endif
3820 }
3821