xref: /netbsd-src/sys/net/if.c (revision 2e2322c9c07009df921d11b1268f8506affbb8ba)
1 /*	$NetBSD: if.c,v 1.367 2016/12/13 02:05:48 ozaki-r Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2001, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by William Studenmund and Jason R. Thorpe.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. Neither the name of the project nor the names of its contributors
45  *    may be used to endorse or promote products derived from this software
46  *    without specific prior written permission.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 /*
62  * Copyright (c) 1980, 1986, 1993
63  *	The Regents of the University of California.  All rights reserved.
64  *
65  * Redistribution and use in source and binary forms, with or without
66  * modification, are permitted provided that the following conditions
67  * are met:
68  * 1. Redistributions of source code must retain the above copyright
69  *    notice, this list of conditions and the following disclaimer.
70  * 2. Redistributions in binary form must reproduce the above copyright
71  *    notice, this list of conditions and the following disclaimer in the
72  *    documentation and/or other materials provided with the distribution.
73  * 3. Neither the name of the University nor the names of its contributors
74  *    may be used to endorse or promote products derived from this software
75  *    without specific prior written permission.
76  *
77  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
78  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
79  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
80  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
81  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
82  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
83  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
84  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
85  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
86  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
87  * SUCH DAMAGE.
88  *
89  *	@(#)if.c	8.5 (Berkeley) 1/9/95
90  */
91 
92 #include <sys/cdefs.h>
93 __KERNEL_RCSID(0, "$NetBSD: if.c,v 1.367 2016/12/13 02:05:48 ozaki-r Exp $");
94 
95 #if defined(_KERNEL_OPT)
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 
99 #include "opt_atalk.h"
100 #include "opt_natm.h"
101 #include "opt_wlan.h"
102 #include "opt_net_mpsafe.h"
103 #endif
104 
105 #include <sys/param.h>
106 #include <sys/mbuf.h>
107 #include <sys/systm.h>
108 #include <sys/callout.h>
109 #include <sys/proc.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/domain.h>
113 #include <sys/protosw.h>
114 #include <sys/kernel.h>
115 #include <sys/ioctl.h>
116 #include <sys/sysctl.h>
117 #include <sys/syslog.h>
118 #include <sys/kauth.h>
119 #include <sys/kmem.h>
120 #include <sys/xcall.h>
121 #include <sys/cpu.h>
122 #include <sys/intr.h>
123 
124 #include <net/if.h>
125 #include <net/if_dl.h>
126 #include <net/if_ether.h>
127 #include <net/if_media.h>
128 #include <net80211/ieee80211.h>
129 #include <net80211/ieee80211_ioctl.h>
130 #include <net/if_types.h>
131 #include <net/route.h>
132 #include <net/netisr.h>
133 #include <sys/module.h>
134 #ifdef NETATALK
135 #include <netatalk/at_extern.h>
136 #include <netatalk/at.h>
137 #endif
138 #include <net/pfil.h>
139 #include <netinet/in.h>
140 #include <netinet/in_var.h>
141 #ifndef IPSEC
142 #include <netinet/ip_encap.h>
143 #endif
144 
145 #ifdef INET6
146 #include <netinet6/in6_var.h>
147 #include <netinet6/nd6.h>
148 #endif
149 
150 #include "ether.h"
151 #include "fddi.h"
152 #include "token.h"
153 
154 #include "carp.h"
155 #if NCARP > 0
156 #include <netinet/ip_carp.h>
157 #endif
158 
159 #include <compat/sys/sockio.h>
160 #include <compat/sys/socket.h>
161 
162 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
163 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
164 
165 /*
166  * Global list of interfaces.
167  */
168 /* DEPRECATED. Remove it once kvm(3) users disappeared */
169 struct ifnet_head		ifnet_list;
170 
171 struct pslist_head		ifnet_pslist;
172 static ifnet_t **		ifindex2ifnet = NULL;
173 static u_int			if_index = 1;
174 static size_t			if_indexlim = 0;
175 static uint64_t			index_gen;
176 /* Mutex to protect the above objects. */
177 kmutex_t			ifnet_mtx __cacheline_aligned;
178 struct psref_class		*ifnet_psref_class __read_mostly;
179 static pserialize_t		ifnet_psz;
180 
181 static kmutex_t			if_clone_mtx;
182 
183 struct ifnet *lo0ifp;
184 int	ifqmaxlen = IFQ_MAXLEN;
185 
186 struct psref_class		*ifa_psref_class __read_mostly;
187 
188 static int	if_delroute_matcher(struct rtentry *, void *);
189 
190 static struct if_clone *if_clone_lookup(const char *, int *);
191 
192 static LIST_HEAD(, if_clone) if_cloners = LIST_HEAD_INITIALIZER(if_cloners);
193 static int if_cloners_count;
194 
195 /* Packet filtering hook for interfaces. */
196 pfil_head_t *	if_pfil;
197 
198 static kauth_listener_t if_listener;
199 
200 static int doifioctl(struct socket *, u_long, void *, struct lwp *);
201 static void if_detach_queues(struct ifnet *, struct ifqueue *);
202 static void sysctl_sndq_setup(struct sysctllog **, const char *,
203     struct ifaltq *);
204 static void if_slowtimo(void *);
205 static void if_free_sadl(struct ifnet *);
206 static void if_attachdomain1(struct ifnet *);
207 static int ifconf(u_long, void *);
208 static int if_transmit(struct ifnet *, struct mbuf *);
209 static int if_clone_create(const char *);
210 static int if_clone_destroy(const char *);
211 static void if_link_state_change_si(void *);
212 
213 struct if_percpuq {
214 	struct ifnet	*ipq_ifp;
215 	void		*ipq_si;
216 	struct percpu	*ipq_ifqs;	/* struct ifqueue */
217 };
218 
219 static struct mbuf *if_percpuq_dequeue(struct if_percpuq *);
220 
221 static void if_percpuq_drops(void *, void *, struct cpu_info *);
222 static int sysctl_percpuq_drops_handler(SYSCTLFN_PROTO);
223 static void sysctl_percpuq_setup(struct sysctllog **, const char *,
224     struct if_percpuq *);
225 
226 struct if_deferred_start {
227 	struct ifnet	*ids_ifp;
228 	void		(*ids_if_start)(struct ifnet *);
229 	void		*ids_si;
230 };
231 
232 static void if_deferred_start_softint(void *);
233 static void if_deferred_start_common(struct ifnet *);
234 static void if_deferred_start_destroy(struct ifnet *);
235 
236 #if defined(INET) || defined(INET6)
237 static void sysctl_net_pktq_setup(struct sysctllog **, int);
238 #endif
239 
240 /*
241  * Pointer to stub or real compat_cvtcmd() depending on presence of
242  * the compat module
243  */
244 u_long stub_compat_cvtcmd(u_long);
245 u_long (*vec_compat_cvtcmd)(u_long) = stub_compat_cvtcmd;
246 
247 /* Similarly, pointer to compat_ifioctl() if it is present */
248 
249 int (*vec_compat_ifioctl)(struct socket *, u_long, u_long, void *,
250 	struct lwp *) = NULL;
251 
252 /* The stub version of compat_cvtcmd() */
253 u_long stub_compat_cvtcmd(u_long cmd)
254 {
255 
256 	return cmd;
257 }
258 
259 static int
260 if_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
261     void *arg0, void *arg1, void *arg2, void *arg3)
262 {
263 	int result;
264 	enum kauth_network_req req;
265 
266 	result = KAUTH_RESULT_DEFER;
267 	req = (enum kauth_network_req)arg1;
268 
269 	if (action != KAUTH_NETWORK_INTERFACE)
270 		return result;
271 
272 	if ((req == KAUTH_REQ_NETWORK_INTERFACE_GET) ||
273 	    (req == KAUTH_REQ_NETWORK_INTERFACE_SET))
274 		result = KAUTH_RESULT_ALLOW;
275 
276 	return result;
277 }
278 
279 /*
280  * Network interface utility routines.
281  *
282  * Routines with ifa_ifwith* names take sockaddr *'s as
283  * parameters.
284  */
285 void
286 ifinit(void)
287 {
288 #if defined(INET)
289 	sysctl_net_pktq_setup(NULL, PF_INET);
290 #endif
291 #ifdef INET6
292 	if (in6_present)
293 		sysctl_net_pktq_setup(NULL, PF_INET6);
294 #endif
295 
296 #if (defined(INET) || defined(INET6)) && !defined(IPSEC)
297 	encapinit();
298 #endif
299 
300 	if_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
301 	    if_listener_cb, NULL);
302 
303 	/* interfaces are available, inform socket code */
304 	ifioctl = doifioctl;
305 }
306 
307 /*
308  * XXX Initialization before configure().
309  * XXX hack to get pfil_add_hook working in autoconf.
310  */
311 void
312 ifinit1(void)
313 {
314 	mutex_init(&if_clone_mtx, MUTEX_DEFAULT, IPL_NONE);
315 
316 	TAILQ_INIT(&ifnet_list);
317 	mutex_init(&ifnet_mtx, MUTEX_DEFAULT, IPL_NONE);
318 	ifnet_psz = pserialize_create();
319 	ifnet_psref_class = psref_class_create("ifnet", IPL_SOFTNET);
320 	ifa_psref_class = psref_class_create("ifa", IPL_SOFTNET);
321 	PSLIST_INIT(&ifnet_pslist);
322 
323 	if_indexlim = 8;
324 
325 	if_pfil = pfil_head_create(PFIL_TYPE_IFNET, NULL);
326 	KASSERT(if_pfil != NULL);
327 
328 #if NETHER > 0 || NFDDI > 0 || defined(NETATALK) || NTOKEN > 0 || defined(WLAN)
329 	etherinit();
330 #endif
331 }
332 
333 ifnet_t *
334 if_alloc(u_char type)
335 {
336 	return kmem_zalloc(sizeof(ifnet_t), KM_SLEEP);
337 }
338 
339 void
340 if_free(ifnet_t *ifp)
341 {
342 	kmem_free(ifp, sizeof(ifnet_t));
343 }
344 
345 void
346 if_initname(struct ifnet *ifp, const char *name, int unit)
347 {
348 	(void)snprintf(ifp->if_xname, sizeof(ifp->if_xname),
349 	    "%s%d", name, unit);
350 }
351 
352 /*
353  * Null routines used while an interface is going away.  These routines
354  * just return an error.
355  */
356 
357 int
358 if_nulloutput(struct ifnet *ifp, struct mbuf *m,
359     const struct sockaddr *so, const struct rtentry *rt)
360 {
361 
362 	return ENXIO;
363 }
364 
365 void
366 if_nullinput(struct ifnet *ifp, struct mbuf *m)
367 {
368 
369 	/* Nothing. */
370 }
371 
372 void
373 if_nullstart(struct ifnet *ifp)
374 {
375 
376 	/* Nothing. */
377 }
378 
379 int
380 if_nulltransmit(struct ifnet *ifp, struct mbuf *m)
381 {
382 
383 	return ENXIO;
384 }
385 
386 int
387 if_nullioctl(struct ifnet *ifp, u_long cmd, void *data)
388 {
389 
390 	return ENXIO;
391 }
392 
393 int
394 if_nullinit(struct ifnet *ifp)
395 {
396 
397 	return ENXIO;
398 }
399 
400 void
401 if_nullstop(struct ifnet *ifp, int disable)
402 {
403 
404 	/* Nothing. */
405 }
406 
407 void
408 if_nullslowtimo(struct ifnet *ifp)
409 {
410 
411 	/* Nothing. */
412 }
413 
414 void
415 if_nulldrain(struct ifnet *ifp)
416 {
417 
418 	/* Nothing. */
419 }
420 
421 void
422 if_set_sadl(struct ifnet *ifp, const void *lla, u_char addrlen, bool factory)
423 {
424 	struct ifaddr *ifa;
425 	struct sockaddr_dl *sdl;
426 
427 	ifp->if_addrlen = addrlen;
428 	if_alloc_sadl(ifp);
429 	ifa = ifp->if_dl;
430 	sdl = satosdl(ifa->ifa_addr);
431 
432 	(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lla, ifp->if_addrlen);
433 	if (factory) {
434 		ifp->if_hwdl = ifp->if_dl;
435 		ifaref(ifp->if_hwdl);
436 	}
437 	/* TBD routing socket */
438 }
439 
440 struct ifaddr *
441 if_dl_create(const struct ifnet *ifp, const struct sockaddr_dl **sdlp)
442 {
443 	unsigned socksize, ifasize;
444 	int addrlen, namelen;
445 	struct sockaddr_dl *mask, *sdl;
446 	struct ifaddr *ifa;
447 
448 	namelen = strlen(ifp->if_xname);
449 	addrlen = ifp->if_addrlen;
450 	socksize = roundup(sockaddr_dl_measure(namelen, addrlen), sizeof(long));
451 	ifasize = sizeof(*ifa) + 2 * socksize;
452 	ifa = (struct ifaddr *)malloc(ifasize, M_IFADDR, M_WAITOK|M_ZERO);
453 
454 	sdl = (struct sockaddr_dl *)(ifa + 1);
455 	mask = (struct sockaddr_dl *)(socksize + (char *)sdl);
456 
457 	sockaddr_dl_init(sdl, socksize, ifp->if_index, ifp->if_type,
458 	    ifp->if_xname, namelen, NULL, addrlen);
459 	mask->sdl_len = sockaddr_dl_measure(namelen, 0);
460 	memset(&mask->sdl_data[0], 0xff, namelen);
461 	ifa->ifa_rtrequest = link_rtrequest;
462 	ifa->ifa_addr = (struct sockaddr *)sdl;
463 	ifa->ifa_netmask = (struct sockaddr *)mask;
464 	ifa_psref_init(ifa);
465 
466 	*sdlp = sdl;
467 
468 	return ifa;
469 }
470 
471 static void
472 if_sadl_setrefs(struct ifnet *ifp, struct ifaddr *ifa)
473 {
474 	const struct sockaddr_dl *sdl;
475 
476 	ifp->if_dl = ifa;
477 	ifaref(ifa);
478 	sdl = satosdl(ifa->ifa_addr);
479 	ifp->if_sadl = sdl;
480 }
481 
482 /*
483  * Allocate the link level name for the specified interface.  This
484  * is an attachment helper.  It must be called after ifp->if_addrlen
485  * is initialized, which may not be the case when if_attach() is
486  * called.
487  */
488 void
489 if_alloc_sadl(struct ifnet *ifp)
490 {
491 	struct ifaddr *ifa;
492 	const struct sockaddr_dl *sdl;
493 
494 	/*
495 	 * If the interface already has a link name, release it
496 	 * now.  This is useful for interfaces that can change
497 	 * link types, and thus switch link names often.
498 	 */
499 	if (ifp->if_sadl != NULL)
500 		if_free_sadl(ifp);
501 
502 	ifa = if_dl_create(ifp, &sdl);
503 
504 	ifa_insert(ifp, ifa);
505 	if_sadl_setrefs(ifp, ifa);
506 }
507 
508 static void
509 if_deactivate_sadl(struct ifnet *ifp)
510 {
511 	struct ifaddr *ifa;
512 
513 	KASSERT(ifp->if_dl != NULL);
514 
515 	ifa = ifp->if_dl;
516 
517 	ifp->if_sadl = NULL;
518 
519 	ifp->if_dl = NULL;
520 	ifafree(ifa);
521 }
522 
523 void
524 if_activate_sadl(struct ifnet *ifp, struct ifaddr *ifa0,
525     const struct sockaddr_dl *sdl)
526 {
527 	int s, ss;
528 	struct ifaddr *ifa;
529 	int bound = curlwp_bind();
530 
531 	s = splnet();
532 
533 	if_deactivate_sadl(ifp);
534 
535 	if_sadl_setrefs(ifp, ifa0);
536 
537 	ss = pserialize_read_enter();
538 	IFADDR_READER_FOREACH(ifa, ifp) {
539 		struct psref psref;
540 		ifa_acquire(ifa, &psref);
541 		pserialize_read_exit(ss);
542 
543 		rtinit(ifa, RTM_LLINFO_UPD, 0);
544 
545 		ss = pserialize_read_enter();
546 		ifa_release(ifa, &psref);
547 	}
548 	pserialize_read_exit(ss);
549 
550 	splx(s);
551 	curlwp_bindx(bound);
552 }
553 
554 /*
555  * Free the link level name for the specified interface.  This is
556  * a detach helper.  This is called from if_detach().
557  */
558 static void
559 if_free_sadl(struct ifnet *ifp)
560 {
561 	struct ifaddr *ifa;
562 	int s;
563 
564 	ifa = ifp->if_dl;
565 	if (ifa == NULL) {
566 		KASSERT(ifp->if_sadl == NULL);
567 		return;
568 	}
569 
570 	KASSERT(ifp->if_sadl != NULL);
571 
572 	s = splnet();
573 	rtinit(ifa, RTM_DELETE, 0);
574 	ifa_remove(ifp, ifa);
575 	if_deactivate_sadl(ifp);
576 	if (ifp->if_hwdl == ifa) {
577 		ifafree(ifa);
578 		ifp->if_hwdl = NULL;
579 	}
580 	splx(s);
581 }
582 
583 static void
584 if_getindex(ifnet_t *ifp)
585 {
586 	bool hitlimit = false;
587 
588 	ifp->if_index_gen = index_gen++;
589 
590 	ifp->if_index = if_index;
591 	if (ifindex2ifnet == NULL) {
592 		if_index++;
593 		goto skip;
594 	}
595 	while (if_byindex(ifp->if_index)) {
596 		/*
597 		 * If we hit USHRT_MAX, we skip back to 0 since
598 		 * there are a number of places where the value
599 		 * of if_index or if_index itself is compared
600 		 * to or stored in an unsigned short.  By
601 		 * jumping back, we won't botch those assignments
602 		 * or comparisons.
603 		 */
604 		if (++if_index == 0) {
605 			if_index = 1;
606 		} else if (if_index == USHRT_MAX) {
607 			/*
608 			 * However, if we have to jump back to
609 			 * zero *twice* without finding an empty
610 			 * slot in ifindex2ifnet[], then there
611 			 * there are too many (>65535) interfaces.
612 			 */
613 			if (hitlimit) {
614 				panic("too many interfaces");
615 			}
616 			hitlimit = true;
617 			if_index = 1;
618 		}
619 		ifp->if_index = if_index;
620 	}
621 skip:
622 	/*
623 	 * ifindex2ifnet is indexed by if_index. Since if_index will
624 	 * grow dynamically, it should grow too.
625 	 */
626 	if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
627 		size_t m, n, oldlim;
628 		void *q;
629 
630 		oldlim = if_indexlim;
631 		while (ifp->if_index >= if_indexlim)
632 			if_indexlim <<= 1;
633 
634 		/* grow ifindex2ifnet */
635 		m = oldlim * sizeof(struct ifnet *);
636 		n = if_indexlim * sizeof(struct ifnet *);
637 		q = malloc(n, M_IFADDR, M_WAITOK|M_ZERO);
638 		if (ifindex2ifnet != NULL) {
639 			memcpy(q, ifindex2ifnet, m);
640 			free(ifindex2ifnet, M_IFADDR);
641 		}
642 		ifindex2ifnet = (struct ifnet **)q;
643 	}
644 	ifindex2ifnet[ifp->if_index] = ifp;
645 }
646 
647 /*
648  * Initialize an interface and assign an index for it.
649  *
650  * It must be called prior to a device specific attach routine
651  * (e.g., ether_ifattach and ieee80211_ifattach) or if_alloc_sadl,
652  * and be followed by if_register:
653  *
654  *     if_initialize(ifp);
655  *     ether_ifattach(ifp, enaddr);
656  *     if_register(ifp);
657  */
658 void
659 if_initialize(ifnet_t *ifp)
660 {
661 	KASSERT(if_indexlim > 0);
662 	TAILQ_INIT(&ifp->if_addrlist);
663 
664 	/*
665 	 * Link level name is allocated later by a separate call to
666 	 * if_alloc_sadl().
667 	 */
668 
669 	if (ifp->if_snd.ifq_maxlen == 0)
670 		ifp->if_snd.ifq_maxlen = ifqmaxlen;
671 
672 	ifp->if_broadcastaddr = 0; /* reliably crash if used uninitialized */
673 
674 	ifp->if_link_state = LINK_STATE_UNKNOWN;
675 	ifp->if_link_queue = -1; /* all bits set, see link_state_change() */
676 
677 	ifp->if_capenable = 0;
678 	ifp->if_csum_flags_tx = 0;
679 	ifp->if_csum_flags_rx = 0;
680 
681 #ifdef ALTQ
682 	ifp->if_snd.altq_type = 0;
683 	ifp->if_snd.altq_disc = NULL;
684 	ifp->if_snd.altq_flags &= ALTQF_CANTCHANGE;
685 	ifp->if_snd.altq_tbr  = NULL;
686 	ifp->if_snd.altq_ifp  = ifp;
687 #endif
688 
689 	IFQ_LOCK_INIT(&ifp->if_snd);
690 
691 	ifp->if_pfil = pfil_head_create(PFIL_TYPE_IFNET, ifp);
692 	(void)pfil_run_hooks(if_pfil,
693 	    (struct mbuf **)PFIL_IFNET_ATTACH, ifp, PFIL_IFNET);
694 
695 	IF_AFDATA_LOCK_INIT(ifp);
696 
697 	if (if_is_link_state_changeable(ifp)) {
698 		ifp->if_link_si = softint_establish(SOFTINT_NET,
699 		    if_link_state_change_si, ifp);
700 		if (ifp->if_link_si == NULL)
701 			panic("%s: softint_establish() failed", __func__);
702 	}
703 
704 	PSLIST_ENTRY_INIT(ifp, if_pslist_entry);
705 	PSLIST_INIT(&ifp->if_addr_pslist);
706 	psref_target_init(&ifp->if_psref, ifnet_psref_class);
707 	ifp->if_ioctl_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
708 
709 	IFNET_LOCK();
710 	if_getindex(ifp);
711 	IFNET_UNLOCK();
712 }
713 
714 /*
715  * Register an interface to the list of "active" interfaces.
716  */
717 void
718 if_register(ifnet_t *ifp)
719 {
720 	/*
721 	 * If the driver has not supplied its own if_ioctl, then
722 	 * supply the default.
723 	 */
724 	if (ifp->if_ioctl == NULL)
725 		ifp->if_ioctl = ifioctl_common;
726 
727 	sysctl_sndq_setup(&ifp->if_sysctl_log, ifp->if_xname, &ifp->if_snd);
728 
729 	if (!STAILQ_EMPTY(&domains))
730 		if_attachdomain1(ifp);
731 
732 	/* Announce the interface. */
733 	rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
734 
735 	if (ifp->if_slowtimo != NULL) {
736 		ifp->if_slowtimo_ch =
737 		    kmem_zalloc(sizeof(*ifp->if_slowtimo_ch), KM_SLEEP);
738 		callout_init(ifp->if_slowtimo_ch, 0);
739 		callout_setfunc(ifp->if_slowtimo_ch, if_slowtimo, ifp);
740 		if_slowtimo(ifp);
741 	}
742 
743 	if (ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit)
744 		ifp->if_transmit = if_transmit;
745 
746 	IFNET_LOCK();
747 	TAILQ_INSERT_TAIL(&ifnet_list, ifp, if_list);
748 	IFNET_WRITER_INSERT_TAIL(ifp);
749 	IFNET_UNLOCK();
750 }
751 
752 /*
753  * The if_percpuq framework
754  *
755  * It allows network device drivers to execute the network stack
756  * in softint (so called softint-based if_input). It utilizes
757  * softint and percpu ifqueue. It doesn't distribute any packets
758  * between CPUs, unlike pktqueue(9).
759  *
760  * Currently we support two options for device drivers to apply the framework:
761  * - Use it implicitly with less changes
762  *   - If you use if_attach in driver's _attach function and if_input in
763  *     driver's Rx interrupt handler, a packet is queued and a softint handles
764  *     the packet implicitly
765  * - Use it explicitly in each driver (recommended)
766  *   - You can use if_percpuq_* directly in your driver
767  *   - In this case, you need to allocate struct if_percpuq in driver's softc
768  *   - See wm(4) as a reference implementation
769  */
770 
771 static void
772 if_percpuq_softint(void *arg)
773 {
774 	struct if_percpuq *ipq = arg;
775 	struct ifnet *ifp = ipq->ipq_ifp;
776 	struct mbuf *m;
777 
778 	while ((m = if_percpuq_dequeue(ipq)) != NULL)
779 		ifp->_if_input(ifp, m);
780 }
781 
782 static void
783 if_percpuq_init_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
784 {
785 	struct ifqueue *const ifq = p;
786 
787 	memset(ifq, 0, sizeof(*ifq));
788 	ifq->ifq_maxlen = IFQ_MAXLEN;
789 }
790 
791 struct if_percpuq *
792 if_percpuq_create(struct ifnet *ifp)
793 {
794 	struct if_percpuq *ipq;
795 
796 	ipq = kmem_zalloc(sizeof(*ipq), KM_SLEEP);
797 	if (ipq == NULL)
798 		panic("kmem_zalloc failed");
799 
800 	ipq->ipq_ifp = ifp;
801 	ipq->ipq_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
802 	    if_percpuq_softint, ipq);
803 	ipq->ipq_ifqs = percpu_alloc(sizeof(struct ifqueue));
804 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_init_ifq, NULL);
805 
806 	sysctl_percpuq_setup(&ifp->if_sysctl_log, ifp->if_xname, ipq);
807 
808 	return ipq;
809 }
810 
811 static struct mbuf *
812 if_percpuq_dequeue(struct if_percpuq *ipq)
813 {
814 	struct mbuf *m;
815 	struct ifqueue *ifq;
816 	int s;
817 
818 	s = splnet();
819 	ifq = percpu_getref(ipq->ipq_ifqs);
820 	IF_DEQUEUE(ifq, m);
821 	percpu_putref(ipq->ipq_ifqs);
822 	splx(s);
823 
824 	return m;
825 }
826 
827 static void
828 if_percpuq_purge_ifq(void *p, void *arg __unused, struct cpu_info *ci __unused)
829 {
830 	struct ifqueue *const ifq = p;
831 
832 	IF_PURGE(ifq);
833 }
834 
835 void
836 if_percpuq_destroy(struct if_percpuq *ipq)
837 {
838 
839 	/* if_detach may already destroy it */
840 	if (ipq == NULL)
841 		return;
842 
843 	softint_disestablish(ipq->ipq_si);
844 	percpu_foreach(ipq->ipq_ifqs, &if_percpuq_purge_ifq, NULL);
845 	percpu_free(ipq->ipq_ifqs, sizeof(struct ifqueue));
846 	kmem_free(ipq, sizeof(*ipq));
847 }
848 
849 void
850 if_percpuq_enqueue(struct if_percpuq *ipq, struct mbuf *m)
851 {
852 	struct ifqueue *ifq;
853 	int s;
854 
855 	KASSERT(ipq != NULL);
856 
857 	s = splnet();
858 	ifq = percpu_getref(ipq->ipq_ifqs);
859 	if (IF_QFULL(ifq)) {
860 		IF_DROP(ifq);
861 		percpu_putref(ipq->ipq_ifqs);
862 		m_freem(m);
863 		goto out;
864 	}
865 	IF_ENQUEUE(ifq, m);
866 	percpu_putref(ipq->ipq_ifqs);
867 
868 	softint_schedule(ipq->ipq_si);
869 out:
870 	splx(s);
871 }
872 
873 static void
874 if_percpuq_drops(void *p, void *arg, struct cpu_info *ci __unused)
875 {
876 	struct ifqueue *const ifq = p;
877 	int *sum = arg;
878 
879 	*sum += ifq->ifq_drops;
880 }
881 
882 static int
883 sysctl_percpuq_drops_handler(SYSCTLFN_ARGS)
884 {
885 	struct sysctlnode node;
886 	struct if_percpuq *ipq;
887 	int sum = 0;
888 	int error;
889 
890 	node = *rnode;
891 	ipq = node.sysctl_data;
892 
893 	percpu_foreach(ipq->ipq_ifqs, if_percpuq_drops, &sum);
894 
895 	node.sysctl_data = &sum;
896 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
897 	if (error != 0 || newp == NULL)
898 		return error;
899 
900 	return 0;
901 }
902 
903 static void
904 sysctl_percpuq_setup(struct sysctllog **clog, const char* ifname,
905     struct if_percpuq *ipq)
906 {
907 	const struct sysctlnode *cnode, *rnode;
908 
909 	if (sysctl_createv(clog, 0, NULL, &rnode,
910 		       CTLFLAG_PERMANENT,
911 		       CTLTYPE_NODE, "interfaces",
912 		       SYSCTL_DESCR("Per-interface controls"),
913 		       NULL, 0, NULL, 0,
914 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
915 		goto bad;
916 
917 	if (sysctl_createv(clog, 0, &rnode, &rnode,
918 		       CTLFLAG_PERMANENT,
919 		       CTLTYPE_NODE, ifname,
920 		       SYSCTL_DESCR("Interface controls"),
921 		       NULL, 0, NULL, 0,
922 		       CTL_CREATE, CTL_EOL) != 0)
923 		goto bad;
924 
925 	if (sysctl_createv(clog, 0, &rnode, &rnode,
926 		       CTLFLAG_PERMANENT,
927 		       CTLTYPE_NODE, "rcvq",
928 		       SYSCTL_DESCR("Interface input queue controls"),
929 		       NULL, 0, NULL, 0,
930 		       CTL_CREATE, CTL_EOL) != 0)
931 		goto bad;
932 
933 #ifdef NOTYET
934 	/* XXX Should show each per-CPU queue length? */
935 	if (sysctl_createv(clog, 0, &rnode, &rnode,
936 		       CTLFLAG_PERMANENT,
937 		       CTLTYPE_INT, "len",
938 		       SYSCTL_DESCR("Current input queue length"),
939 		       sysctl_percpuq_len, 0, NULL, 0,
940 		       CTL_CREATE, CTL_EOL) != 0)
941 		goto bad;
942 
943 	if (sysctl_createv(clog, 0, &rnode, &cnode,
944 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
945 		       CTLTYPE_INT, "maxlen",
946 		       SYSCTL_DESCR("Maximum allowed input queue length"),
947 		       sysctl_percpuq_maxlen_handler, 0, (void *)ipq, 0,
948 		       CTL_CREATE, CTL_EOL) != 0)
949 		goto bad;
950 #endif
951 
952 	if (sysctl_createv(clog, 0, &rnode, &cnode,
953 		       CTLFLAG_PERMANENT,
954 		       CTLTYPE_INT, "drops",
955 		       SYSCTL_DESCR("Total packets dropped due to full input queue"),
956 		       sysctl_percpuq_drops_handler, 0, (void *)ipq, 0,
957 		       CTL_CREATE, CTL_EOL) != 0)
958 		goto bad;
959 
960 	return;
961 bad:
962 	printf("%s: could not attach sysctl nodes\n", ifname);
963 	return;
964 }
965 
966 /*
967  * The deferred if_start framework
968  *
969  * The common APIs to defer if_start to softint when if_start is requested
970  * from a device driver running in hardware interrupt context.
971  */
972 /*
973  * Call ifp->if_start (or equivalent) in a dedicated softint for
974  * deferred if_start.
975  */
976 static void
977 if_deferred_start_softint(void *arg)
978 {
979 	struct if_deferred_start *ids = arg;
980 	struct ifnet *ifp = ids->ids_ifp;
981 
982 	ids->ids_if_start(ifp);
983 }
984 
985 /*
986  * The default callback function for deferred if_start.
987  */
988 static void
989 if_deferred_start_common(struct ifnet *ifp)
990 {
991 
992 	if_start_lock(ifp);
993 }
994 
995 static inline bool
996 if_snd_is_used(struct ifnet *ifp)
997 {
998 
999 	return ifp->if_transmit == NULL || ifp->if_transmit == if_nulltransmit ||
1000 	    ALTQ_IS_ENABLED(&ifp->if_snd);
1001 }
1002 
1003 /*
1004  * Schedule deferred if_start.
1005  */
1006 void
1007 if_schedule_deferred_start(struct ifnet *ifp)
1008 {
1009 
1010 	KASSERT(ifp->if_deferred_start != NULL);
1011 
1012 	if (if_snd_is_used(ifp) && IFQ_IS_EMPTY(&ifp->if_snd))
1013 		return;
1014 
1015 	softint_schedule(ifp->if_deferred_start->ids_si);
1016 }
1017 
1018 /*
1019  * Create an instance of deferred if_start. A driver should call the function
1020  * only if the driver needs deferred if_start. Drivers can setup their own
1021  * deferred if_start function via 2nd argument.
1022  */
1023 void
1024 if_deferred_start_init(struct ifnet *ifp, void (*func)(struct ifnet *))
1025 {
1026 	struct if_deferred_start *ids;
1027 
1028 	ids = kmem_zalloc(sizeof(*ids), KM_SLEEP);
1029 	if (ids == NULL)
1030 		panic("kmem_zalloc failed");
1031 
1032 	ids->ids_ifp = ifp;
1033 	ids->ids_si = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
1034 	    if_deferred_start_softint, ids);
1035 	if (func != NULL)
1036 		ids->ids_if_start = func;
1037 	else
1038 		ids->ids_if_start = if_deferred_start_common;
1039 
1040 	ifp->if_deferred_start = ids;
1041 }
1042 
1043 static void
1044 if_deferred_start_destroy(struct ifnet *ifp)
1045 {
1046 
1047 	if (ifp->if_deferred_start == NULL)
1048 		return;
1049 
1050 	softint_disestablish(ifp->if_deferred_start->ids_si);
1051 	kmem_free(ifp->if_deferred_start, sizeof(*ifp->if_deferred_start));
1052 	ifp->if_deferred_start = NULL;
1053 }
1054 
1055 /*
1056  * The common interface input routine that is called by device drivers,
1057  * which should be used only when the driver's rx handler already runs
1058  * in softint.
1059  */
1060 void
1061 if_input(struct ifnet *ifp, struct mbuf *m)
1062 {
1063 
1064 	KASSERT(ifp->if_percpuq == NULL);
1065 	KASSERT(!cpu_intr_p());
1066 
1067 	ifp->_if_input(ifp, m);
1068 }
1069 
1070 /*
1071  * DEPRECATED. Use if_initialize and if_register instead.
1072  * See the above comment of if_initialize.
1073  *
1074  * Note that it implicitly enables if_percpuq to make drivers easy to
1075  * migrate softint-based if_input without much changes. If you don't
1076  * want to enable it, use if_initialize instead.
1077  */
1078 void
1079 if_attach(ifnet_t *ifp)
1080 {
1081 
1082 	if_initialize(ifp);
1083 	ifp->if_percpuq = if_percpuq_create(ifp);
1084 	if_register(ifp);
1085 }
1086 
1087 void
1088 if_attachdomain(void)
1089 {
1090 	struct ifnet *ifp;
1091 	int s;
1092 	int bound = curlwp_bind();
1093 
1094 	s = pserialize_read_enter();
1095 	IFNET_READER_FOREACH(ifp) {
1096 		struct psref psref;
1097 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
1098 		pserialize_read_exit(s);
1099 		if_attachdomain1(ifp);
1100 		s = pserialize_read_enter();
1101 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
1102 	}
1103 	pserialize_read_exit(s);
1104 	curlwp_bindx(bound);
1105 }
1106 
1107 static void
1108 if_attachdomain1(struct ifnet *ifp)
1109 {
1110 	struct domain *dp;
1111 	int s;
1112 
1113 	s = splnet();
1114 
1115 	/* address family dependent data region */
1116 	memset(ifp->if_afdata, 0, sizeof(ifp->if_afdata));
1117 	DOMAIN_FOREACH(dp) {
1118 		if (dp->dom_ifattach != NULL)
1119 			ifp->if_afdata[dp->dom_family] =
1120 			    (*dp->dom_ifattach)(ifp);
1121 	}
1122 
1123 	splx(s);
1124 }
1125 
1126 /*
1127  * Deactivate an interface.  This points all of the procedure
1128  * handles at error stubs.  May be called from interrupt context.
1129  */
1130 void
1131 if_deactivate(struct ifnet *ifp)
1132 {
1133 	int s;
1134 
1135 	s = splnet();
1136 
1137 	ifp->if_output	 = if_nulloutput;
1138 	ifp->_if_input	 = if_nullinput;
1139 	ifp->if_start	 = if_nullstart;
1140 	ifp->if_transmit = if_nulltransmit;
1141 	ifp->if_ioctl	 = if_nullioctl;
1142 	ifp->if_init	 = if_nullinit;
1143 	ifp->if_stop	 = if_nullstop;
1144 	ifp->if_slowtimo = if_nullslowtimo;
1145 	ifp->if_drain	 = if_nulldrain;
1146 
1147 	/* No more packets may be enqueued. */
1148 	ifp->if_snd.ifq_maxlen = 0;
1149 
1150 	splx(s);
1151 }
1152 
1153 bool
1154 if_is_deactivated(const struct ifnet *ifp)
1155 {
1156 
1157 	return ifp->if_output == if_nulloutput;
1158 }
1159 
1160 void
1161 if_purgeaddrs(struct ifnet *ifp, int family, void (*purgeaddr)(struct ifaddr *))
1162 {
1163 	struct ifaddr *ifa, *nifa;
1164 	int s;
1165 
1166 	s = pserialize_read_enter();
1167 	for (ifa = IFADDR_READER_FIRST(ifp); ifa; ifa = nifa) {
1168 		nifa = IFADDR_READER_NEXT(ifa);
1169 		if (ifa->ifa_addr->sa_family != family)
1170 			continue;
1171 		pserialize_read_exit(s);
1172 
1173 		(*purgeaddr)(ifa);
1174 
1175 		s = pserialize_read_enter();
1176 	}
1177 	pserialize_read_exit(s);
1178 }
1179 
1180 #ifdef IFAREF_DEBUG
1181 static struct ifaddr **ifa_list;
1182 static int ifa_list_size;
1183 
1184 /* Depends on only one if_attach runs at once */
1185 static void
1186 if_build_ifa_list(struct ifnet *ifp)
1187 {
1188 	struct ifaddr *ifa;
1189 	int i;
1190 
1191 	KASSERT(ifa_list == NULL);
1192 	KASSERT(ifa_list_size == 0);
1193 
1194 	IFADDR_READER_FOREACH(ifa, ifp)
1195 		ifa_list_size++;
1196 
1197 	ifa_list = kmem_alloc(sizeof(*ifa) * ifa_list_size, KM_SLEEP);
1198 	if (ifa_list == NULL)
1199 		return;
1200 
1201 	i = 0;
1202 	IFADDR_READER_FOREACH(ifa, ifp) {
1203 		ifa_list[i++] = ifa;
1204 		ifaref(ifa);
1205 	}
1206 }
1207 
1208 static void
1209 if_check_and_free_ifa_list(struct ifnet *ifp)
1210 {
1211 	int i;
1212 	struct ifaddr *ifa;
1213 
1214 	if (ifa_list == NULL)
1215 		return;
1216 
1217 	for (i = 0; i < ifa_list_size; i++) {
1218 		char buf[64];
1219 
1220 		ifa = ifa_list[i];
1221 		sockaddr_format(ifa->ifa_addr, buf, sizeof(buf));
1222 		if (ifa->ifa_refcnt > 1) {
1223 			log(LOG_WARNING,
1224 			    "ifa(%s) still referenced (refcnt=%d)\n",
1225 			    buf, ifa->ifa_refcnt - 1);
1226 		} else
1227 			log(LOG_DEBUG,
1228 			    "ifa(%s) not referenced (refcnt=%d)\n",
1229 			    buf, ifa->ifa_refcnt - 1);
1230 		ifafree(ifa);
1231 	}
1232 
1233 	kmem_free(ifa_list, sizeof(*ifa) * ifa_list_size);
1234 	ifa_list = NULL;
1235 	ifa_list_size = 0;
1236 }
1237 #endif
1238 
1239 /*
1240  * Detach an interface from the list of "active" interfaces,
1241  * freeing any resources as we go along.
1242  *
1243  * NOTE: This routine must be called with a valid thread context,
1244  * as it may block.
1245  */
1246 void
1247 if_detach(struct ifnet *ifp)
1248 {
1249 	struct socket so;
1250 	struct ifaddr *ifa;
1251 #ifdef IFAREF_DEBUG
1252 	struct ifaddr *last_ifa = NULL;
1253 #endif
1254 	struct domain *dp;
1255 	const struct protosw *pr;
1256 	int s, i, family, purged;
1257 	uint64_t xc;
1258 
1259 #ifdef IFAREF_DEBUG
1260 	if_build_ifa_list(ifp);
1261 #endif
1262 	/*
1263 	 * XXX It's kind of lame that we have to have the
1264 	 * XXX socket structure...
1265 	 */
1266 	memset(&so, 0, sizeof(so));
1267 
1268 	s = splnet();
1269 
1270 	sysctl_teardown(&ifp->if_sysctl_log);
1271 	mutex_enter(ifp->if_ioctl_lock);
1272 	if_deactivate(ifp);
1273 	mutex_exit(ifp->if_ioctl_lock);
1274 
1275 	IFNET_LOCK();
1276 	ifindex2ifnet[ifp->if_index] = NULL;
1277 	TAILQ_REMOVE(&ifnet_list, ifp, if_list);
1278 	IFNET_WRITER_REMOVE(ifp);
1279 	pserialize_perform(ifnet_psz);
1280 	IFNET_UNLOCK();
1281 
1282 	/* Wait for all readers to drain before freeing.  */
1283 	psref_target_destroy(&ifp->if_psref, ifnet_psref_class);
1284 	PSLIST_ENTRY_DESTROY(ifp, if_pslist_entry);
1285 
1286 	mutex_obj_free(ifp->if_ioctl_lock);
1287 	ifp->if_ioctl_lock = NULL;
1288 
1289 	if (ifp->if_slowtimo != NULL && ifp->if_slowtimo_ch != NULL) {
1290 		ifp->if_slowtimo = NULL;
1291 		callout_halt(ifp->if_slowtimo_ch, NULL);
1292 		callout_destroy(ifp->if_slowtimo_ch);
1293 		kmem_free(ifp->if_slowtimo_ch, sizeof(*ifp->if_slowtimo_ch));
1294 	}
1295 	if_deferred_start_destroy(ifp);
1296 
1297 	/*
1298 	 * Do an if_down() to give protocols a chance to do something.
1299 	 */
1300 	if_down(ifp);
1301 
1302 #ifdef ALTQ
1303 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
1304 		altq_disable(&ifp->if_snd);
1305 	if (ALTQ_IS_ATTACHED(&ifp->if_snd))
1306 		altq_detach(&ifp->if_snd);
1307 #endif
1308 
1309 	mutex_obj_free(ifp->if_snd.ifq_lock);
1310 
1311 #if NCARP > 0
1312 	/* Remove the interface from any carp group it is a part of.  */
1313 	if (ifp->if_carp != NULL && ifp->if_type != IFT_CARP)
1314 		carp_ifdetach(ifp);
1315 #endif
1316 
1317 	/*
1318 	 * Rip all the addresses off the interface.  This should make
1319 	 * all of the routes go away.
1320 	 *
1321 	 * pr_usrreq calls can remove an arbitrary number of ifaddrs
1322 	 * from the list, including our "cursor", ifa.  For safety,
1323 	 * and to honor the TAILQ abstraction, I just restart the
1324 	 * loop after each removal.  Note that the loop will exit
1325 	 * when all of the remaining ifaddrs belong to the AF_LINK
1326 	 * family.  I am counting on the historical fact that at
1327 	 * least one pr_usrreq in each address domain removes at
1328 	 * least one ifaddr.
1329 	 */
1330 again:
1331 	/*
1332 	 * At this point, no other one tries to remove ifa in the list,
1333 	 * so we don't need to take a lock or psref.
1334 	 */
1335 	IFADDR_READER_FOREACH(ifa, ifp) {
1336 		family = ifa->ifa_addr->sa_family;
1337 #ifdef IFAREF_DEBUG
1338 		printf("if_detach: ifaddr %p, family %d, refcnt %d\n",
1339 		    ifa, family, ifa->ifa_refcnt);
1340 		if (last_ifa != NULL && ifa == last_ifa)
1341 			panic("if_detach: loop detected");
1342 		last_ifa = ifa;
1343 #endif
1344 		if (family == AF_LINK)
1345 			continue;
1346 		dp = pffinddomain(family);
1347 #ifdef DIAGNOSTIC
1348 		if (dp == NULL)
1349 			panic("if_detach: no domain for AF %d",
1350 			    family);
1351 #endif
1352 		/*
1353 		 * XXX These PURGEIF calls are redundant with the
1354 		 * purge-all-families calls below, but are left in for
1355 		 * now both to make a smaller change, and to avoid
1356 		 * unplanned interactions with clearing of
1357 		 * ifp->if_addrlist.
1358 		 */
1359 		purged = 0;
1360 		for (pr = dp->dom_protosw;
1361 		     pr < dp->dom_protoswNPROTOSW; pr++) {
1362 			so.so_proto = pr;
1363 			if (pr->pr_usrreqs) {
1364 				(void) (*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1365 				purged = 1;
1366 			}
1367 		}
1368 		if (purged == 0) {
1369 			/*
1370 			 * XXX What's really the best thing to do
1371 			 * XXX here?  --thorpej@NetBSD.org
1372 			 */
1373 			printf("if_detach: WARNING: AF %d not purged\n",
1374 			    family);
1375 			ifa_remove(ifp, ifa);
1376 		}
1377 		goto again;
1378 	}
1379 
1380 	if_free_sadl(ifp);
1381 
1382 	/* Delete stray routes from the routing table. */
1383 	for (i = 0; i <= AF_MAX; i++)
1384 		rt_delete_matched_entries(i, if_delroute_matcher, ifp);
1385 
1386 	DOMAIN_FOREACH(dp) {
1387 		if (dp->dom_ifdetach != NULL && ifp->if_afdata[dp->dom_family])
1388 		{
1389 			void *p = ifp->if_afdata[dp->dom_family];
1390 			if (p) {
1391 				ifp->if_afdata[dp->dom_family] = NULL;
1392 				(*dp->dom_ifdetach)(ifp, p);
1393 			}
1394 		}
1395 
1396 		/*
1397 		 * One would expect multicast memberships (INET and
1398 		 * INET6) on UDP sockets to be purged by the PURGEIF
1399 		 * calls above, but if all addresses were removed from
1400 		 * the interface prior to destruction, the calls will
1401 		 * not be made (e.g. ppp, for which pppd(8) generally
1402 		 * removes addresses before destroying the interface).
1403 		 * Because there is no invariant that multicast
1404 		 * memberships only exist for interfaces with IPv4
1405 		 * addresses, we must call PURGEIF regardless of
1406 		 * addresses.  (Protocols which might store ifnet
1407 		 * pointers are marked with PR_PURGEIF.)
1408 		 */
1409 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) {
1410 			so.so_proto = pr;
1411 			if (pr->pr_usrreqs && pr->pr_flags & PR_PURGEIF)
1412 				(void)(*pr->pr_usrreqs->pr_purgeif)(&so, ifp);
1413 		}
1414 	}
1415 
1416 	(void)pfil_run_hooks(if_pfil,
1417 	    (struct mbuf **)PFIL_IFNET_DETACH, ifp, PFIL_IFNET);
1418 	(void)pfil_head_destroy(ifp->if_pfil);
1419 
1420 	/* Announce that the interface is gone. */
1421 	rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
1422 
1423 	IF_AFDATA_LOCK_DESTROY(ifp);
1424 
1425 	if (if_is_link_state_changeable(ifp)) {
1426 		softint_disestablish(ifp->if_link_si);
1427 		ifp->if_link_si = NULL;
1428 	}
1429 
1430 	/*
1431 	 * remove packets that came from ifp, from software interrupt queues.
1432 	 */
1433 	DOMAIN_FOREACH(dp) {
1434 		for (i = 0; i < __arraycount(dp->dom_ifqueues); i++) {
1435 			struct ifqueue *iq = dp->dom_ifqueues[i];
1436 			if (iq == NULL)
1437 				break;
1438 			dp->dom_ifqueues[i] = NULL;
1439 			if_detach_queues(ifp, iq);
1440 		}
1441 	}
1442 
1443 	/*
1444 	 * IP queues have to be processed separately: net-queue barrier
1445 	 * ensures that the packets are dequeued while a cross-call will
1446 	 * ensure that the interrupts have completed. FIXME: not quite..
1447 	 */
1448 #ifdef INET
1449 	pktq_barrier(ip_pktq);
1450 #endif
1451 #ifdef INET6
1452 	if (in6_present)
1453 		pktq_barrier(ip6_pktq);
1454 #endif
1455 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
1456 	xc_wait(xc);
1457 
1458 	if (ifp->if_percpuq != NULL) {
1459 		if_percpuq_destroy(ifp->if_percpuq);
1460 		ifp->if_percpuq = NULL;
1461 	}
1462 
1463 	splx(s);
1464 
1465 #ifdef IFAREF_DEBUG
1466 	if_check_and_free_ifa_list(ifp);
1467 #endif
1468 }
1469 
1470 static void
1471 if_detach_queues(struct ifnet *ifp, struct ifqueue *q)
1472 {
1473 	struct mbuf *m, *prev, *next;
1474 
1475 	prev = NULL;
1476 	for (m = q->ifq_head; m != NULL; m = next) {
1477 		KASSERT((m->m_flags & M_PKTHDR) != 0);
1478 
1479 		next = m->m_nextpkt;
1480 		if (m->m_pkthdr.rcvif_index != ifp->if_index) {
1481 			prev = m;
1482 			continue;
1483 		}
1484 
1485 		if (prev != NULL)
1486 			prev->m_nextpkt = m->m_nextpkt;
1487 		else
1488 			q->ifq_head = m->m_nextpkt;
1489 		if (q->ifq_tail == m)
1490 			q->ifq_tail = prev;
1491 		q->ifq_len--;
1492 
1493 		m->m_nextpkt = NULL;
1494 		m_freem(m);
1495 		IF_DROP(q);
1496 	}
1497 }
1498 
1499 /*
1500  * Callback for a radix tree walk to delete all references to an
1501  * ifnet.
1502  */
1503 static int
1504 if_delroute_matcher(struct rtentry *rt, void *v)
1505 {
1506 	struct ifnet *ifp = (struct ifnet *)v;
1507 
1508 	if (rt->rt_ifp == ifp)
1509 		return 1;
1510 	else
1511 		return 0;
1512 }
1513 
1514 /*
1515  * Create a clone network interface.
1516  */
1517 static int
1518 if_clone_create(const char *name)
1519 {
1520 	struct if_clone *ifc;
1521 	int unit;
1522 	struct ifnet *ifp;
1523 	struct psref psref;
1524 
1525 	ifc = if_clone_lookup(name, &unit);
1526 	if (ifc == NULL)
1527 		return EINVAL;
1528 
1529 	ifp = if_get(name, &psref);
1530 	if (ifp != NULL) {
1531 		if_put(ifp, &psref);
1532 		return EEXIST;
1533 	}
1534 
1535 	return (*ifc->ifc_create)(ifc, unit);
1536 }
1537 
1538 /*
1539  * Destroy a clone network interface.
1540  */
1541 static int
1542 if_clone_destroy(const char *name)
1543 {
1544 	struct if_clone *ifc;
1545 	struct ifnet *ifp;
1546 	struct psref psref;
1547 
1548 	ifc = if_clone_lookup(name, NULL);
1549 	if (ifc == NULL)
1550 		return EINVAL;
1551 
1552 	if (ifc->ifc_destroy == NULL)
1553 		return EOPNOTSUPP;
1554 
1555 	ifp = if_get(name, &psref);
1556 	if (ifp == NULL)
1557 		return ENXIO;
1558 
1559 	/* We have to disable ioctls here */
1560 	mutex_enter(ifp->if_ioctl_lock);
1561 	ifp->if_ioctl = if_nullioctl;
1562 	mutex_exit(ifp->if_ioctl_lock);
1563 
1564 	/*
1565 	 * We cannot call ifc_destroy with holding ifp.
1566 	 * Releasing ifp here is safe thanks to if_clone_mtx.
1567 	 */
1568 	if_put(ifp, &psref);
1569 
1570 	return (*ifc->ifc_destroy)(ifp);
1571 }
1572 
1573 /*
1574  * Look up a network interface cloner.
1575  */
1576 static struct if_clone *
1577 if_clone_lookup(const char *name, int *unitp)
1578 {
1579 	struct if_clone *ifc;
1580 	const char *cp;
1581 	char *dp, ifname[IFNAMSIZ + 3];
1582 	int unit;
1583 
1584 	strcpy(ifname, "if_");
1585 	/* separate interface name from unit */
1586 	for (dp = ifname + 3, cp = name; cp - name < IFNAMSIZ &&
1587 	    *cp && (*cp < '0' || *cp > '9');)
1588 		*dp++ = *cp++;
1589 
1590 	if (cp == name || cp - name == IFNAMSIZ || !*cp)
1591 		return NULL;	/* No name or unit number */
1592 	*dp++ = '\0';
1593 
1594 again:
1595 	LIST_FOREACH(ifc, &if_cloners, ifc_list) {
1596 		if (strcmp(ifname + 3, ifc->ifc_name) == 0)
1597 			break;
1598 	}
1599 
1600 	if (ifc == NULL) {
1601 		if (*ifname == '\0' ||
1602 		    module_autoload(ifname, MODULE_CLASS_DRIVER))
1603 			return NULL;
1604 		*ifname = '\0';
1605 		goto again;
1606 	}
1607 
1608 	unit = 0;
1609 	while (cp - name < IFNAMSIZ && *cp) {
1610 		if (*cp < '0' || *cp > '9' || unit >= INT_MAX / 10) {
1611 			/* Bogus unit number. */
1612 			return NULL;
1613 		}
1614 		unit = (unit * 10) + (*cp++ - '0');
1615 	}
1616 
1617 	if (unitp != NULL)
1618 		*unitp = unit;
1619 	return ifc;
1620 }
1621 
1622 /*
1623  * Register a network interface cloner.
1624  */
1625 void
1626 if_clone_attach(struct if_clone *ifc)
1627 {
1628 
1629 	LIST_INSERT_HEAD(&if_cloners, ifc, ifc_list);
1630 	if_cloners_count++;
1631 }
1632 
1633 /*
1634  * Unregister a network interface cloner.
1635  */
1636 void
1637 if_clone_detach(struct if_clone *ifc)
1638 {
1639 
1640 	LIST_REMOVE(ifc, ifc_list);
1641 	if_cloners_count--;
1642 }
1643 
1644 /*
1645  * Provide list of interface cloners to userspace.
1646  */
1647 int
1648 if_clone_list(int buf_count, char *buffer, int *total)
1649 {
1650 	char outbuf[IFNAMSIZ], *dst;
1651 	struct if_clone *ifc;
1652 	int count, error = 0;
1653 
1654 	*total = if_cloners_count;
1655 	if ((dst = buffer) == NULL) {
1656 		/* Just asking how many there are. */
1657 		return 0;
1658 	}
1659 
1660 	if (buf_count < 0)
1661 		return EINVAL;
1662 
1663 	count = (if_cloners_count < buf_count) ?
1664 	    if_cloners_count : buf_count;
1665 
1666 	for (ifc = LIST_FIRST(&if_cloners); ifc != NULL && count != 0;
1667 	     ifc = LIST_NEXT(ifc, ifc_list), count--, dst += IFNAMSIZ) {
1668 		(void)strncpy(outbuf, ifc->ifc_name, sizeof(outbuf));
1669 		if (outbuf[sizeof(outbuf) - 1] != '\0')
1670 			return ENAMETOOLONG;
1671 		error = copyout(outbuf, dst, sizeof(outbuf));
1672 		if (error != 0)
1673 			break;
1674 	}
1675 
1676 	return error;
1677 }
1678 
1679 void
1680 ifa_psref_init(struct ifaddr *ifa)
1681 {
1682 
1683 	psref_target_init(&ifa->ifa_psref, ifa_psref_class);
1684 }
1685 
1686 void
1687 ifaref(struct ifaddr *ifa)
1688 {
1689 	KASSERT(!ISSET(ifa->ifa_flags, IFA_DESTROYING));
1690 	ifa->ifa_refcnt++;
1691 }
1692 
1693 void
1694 ifafree(struct ifaddr *ifa)
1695 {
1696 	KASSERT(ifa != NULL);
1697 	KASSERT(ifa->ifa_refcnt > 0);
1698 
1699 	if (--ifa->ifa_refcnt == 0) {
1700 		free(ifa, M_IFADDR);
1701 	}
1702 }
1703 
1704 bool
1705 ifa_is_destroying(struct ifaddr *ifa)
1706 {
1707 
1708 	return ISSET(ifa->ifa_flags, IFA_DESTROYING);
1709 }
1710 
1711 void
1712 ifa_insert(struct ifnet *ifp, struct ifaddr *ifa)
1713 {
1714 
1715 	ifa->ifa_ifp = ifp;
1716 
1717 	IFNET_LOCK();
1718 	TAILQ_INSERT_TAIL(&ifp->if_addrlist, ifa, ifa_list);
1719 	IFADDR_ENTRY_INIT(ifa);
1720 	IFADDR_WRITER_INSERT_TAIL(ifp, ifa);
1721 	IFNET_UNLOCK();
1722 
1723 	ifaref(ifa);
1724 }
1725 
1726 void
1727 ifa_remove(struct ifnet *ifp, struct ifaddr *ifa)
1728 {
1729 
1730 	KASSERT(ifa->ifa_ifp == ifp);
1731 
1732 	IFNET_LOCK();
1733 	TAILQ_REMOVE(&ifp->if_addrlist, ifa, ifa_list);
1734 	IFADDR_WRITER_REMOVE(ifa);
1735 #ifdef NET_MPSAFE
1736 	pserialize_perform(ifnet_psz);
1737 #endif
1738 	IFNET_UNLOCK();
1739 
1740 #ifdef NET_MPSAFE
1741 	psref_target_destroy(&ifa->ifa_psref, ifa_psref_class);
1742 #endif
1743 	IFADDR_ENTRY_DESTROY(ifa);
1744 	ifafree(ifa);
1745 }
1746 
1747 void
1748 ifa_acquire(struct ifaddr *ifa, struct psref *psref)
1749 {
1750 
1751 	psref_acquire(psref, &ifa->ifa_psref, ifa_psref_class);
1752 }
1753 
1754 void
1755 ifa_release(struct ifaddr *ifa, struct psref *psref)
1756 {
1757 
1758 	if (ifa == NULL)
1759 		return;
1760 
1761 	psref_release(psref, &ifa->ifa_psref, ifa_psref_class);
1762 }
1763 
1764 bool
1765 ifa_held(struct ifaddr *ifa)
1766 {
1767 
1768 	return psref_held(&ifa->ifa_psref, ifa_psref_class);
1769 }
1770 
1771 static inline int
1772 equal(const struct sockaddr *sa1, const struct sockaddr *sa2)
1773 {
1774 	return sockaddr_cmp(sa1, sa2) == 0;
1775 }
1776 
1777 /*
1778  * Locate an interface based on a complete address.
1779  */
1780 /*ARGSUSED*/
1781 struct ifaddr *
1782 ifa_ifwithaddr(const struct sockaddr *addr)
1783 {
1784 	struct ifnet *ifp;
1785 	struct ifaddr *ifa;
1786 
1787 	IFNET_READER_FOREACH(ifp) {
1788 		if (if_is_deactivated(ifp))
1789 			continue;
1790 		IFADDR_READER_FOREACH(ifa, ifp) {
1791 			if (ifa->ifa_addr->sa_family != addr->sa_family)
1792 				continue;
1793 			if (equal(addr, ifa->ifa_addr))
1794 				return ifa;
1795 			if ((ifp->if_flags & IFF_BROADCAST) &&
1796 			    ifa->ifa_broadaddr &&
1797 			    /* IP6 doesn't have broadcast */
1798 			    ifa->ifa_broadaddr->sa_len != 0 &&
1799 			    equal(ifa->ifa_broadaddr, addr))
1800 				return ifa;
1801 		}
1802 	}
1803 	return NULL;
1804 }
1805 
1806 struct ifaddr *
1807 ifa_ifwithaddr_psref(const struct sockaddr *addr, struct psref *psref)
1808 {
1809 	struct ifaddr *ifa;
1810 	int s = pserialize_read_enter();
1811 
1812 	ifa = ifa_ifwithaddr(addr);
1813 	if (ifa != NULL)
1814 		ifa_acquire(ifa, psref);
1815 	pserialize_read_exit(s);
1816 
1817 	return ifa;
1818 }
1819 
1820 /*
1821  * Locate the point to point interface with a given destination address.
1822  */
1823 /*ARGSUSED*/
1824 struct ifaddr *
1825 ifa_ifwithdstaddr(const struct sockaddr *addr)
1826 {
1827 	struct ifnet *ifp;
1828 	struct ifaddr *ifa;
1829 
1830 	IFNET_READER_FOREACH(ifp) {
1831 		if (if_is_deactivated(ifp))
1832 			continue;
1833 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1834 			continue;
1835 		IFADDR_READER_FOREACH(ifa, ifp) {
1836 			if (ifa->ifa_addr->sa_family != addr->sa_family ||
1837 			    ifa->ifa_dstaddr == NULL)
1838 				continue;
1839 			if (equal(addr, ifa->ifa_dstaddr))
1840 				return ifa;
1841 		}
1842 	}
1843 
1844 	return NULL;
1845 }
1846 
1847 struct ifaddr *
1848 ifa_ifwithdstaddr_psref(const struct sockaddr *addr, struct psref *psref)
1849 {
1850 	struct ifaddr *ifa;
1851 	int s;
1852 
1853 	s = pserialize_read_enter();
1854 	ifa = ifa_ifwithdstaddr(addr);
1855 	if (ifa != NULL)
1856 		ifa_acquire(ifa, psref);
1857 	pserialize_read_exit(s);
1858 
1859 	return ifa;
1860 }
1861 
1862 /*
1863  * Find an interface on a specific network.  If many, choice
1864  * is most specific found.
1865  */
1866 struct ifaddr *
1867 ifa_ifwithnet(const struct sockaddr *addr)
1868 {
1869 	struct ifnet *ifp;
1870 	struct ifaddr *ifa, *ifa_maybe = NULL;
1871 	const struct sockaddr_dl *sdl;
1872 	u_int af = addr->sa_family;
1873 	const char *addr_data = addr->sa_data, *cplim;
1874 
1875 	if (af == AF_LINK) {
1876 		sdl = satocsdl(addr);
1877 		if (sdl->sdl_index && sdl->sdl_index < if_indexlim &&
1878 		    ifindex2ifnet[sdl->sdl_index] &&
1879 		    !if_is_deactivated(ifindex2ifnet[sdl->sdl_index])) {
1880 			return ifindex2ifnet[sdl->sdl_index]->if_dl;
1881 		}
1882 	}
1883 #ifdef NETATALK
1884 	if (af == AF_APPLETALK) {
1885 		const struct sockaddr_at *sat, *sat2;
1886 		sat = (const struct sockaddr_at *)addr;
1887 		IFNET_READER_FOREACH(ifp) {
1888 			if (if_is_deactivated(ifp))
1889 				continue;
1890 			ifa = at_ifawithnet((const struct sockaddr_at *)addr, ifp);
1891 			if (ifa == NULL)
1892 				continue;
1893 			sat2 = (struct sockaddr_at *)ifa->ifa_addr;
1894 			if (sat2->sat_addr.s_net == sat->sat_addr.s_net)
1895 				return ifa; /* exact match */
1896 			if (ifa_maybe == NULL) {
1897 				/* else keep the if with the right range */
1898 				ifa_maybe = ifa;
1899 			}
1900 		}
1901 		return ifa_maybe;
1902 	}
1903 #endif
1904 	IFNET_READER_FOREACH(ifp) {
1905 		if (if_is_deactivated(ifp))
1906 			continue;
1907 		IFADDR_READER_FOREACH(ifa, ifp) {
1908 			const char *cp, *cp2, *cp3;
1909 
1910 			if (ifa->ifa_addr->sa_family != af ||
1911 			    ifa->ifa_netmask == NULL)
1912  next:				continue;
1913 			cp = addr_data;
1914 			cp2 = ifa->ifa_addr->sa_data;
1915 			cp3 = ifa->ifa_netmask->sa_data;
1916 			cplim = (const char *)ifa->ifa_netmask +
1917 			    ifa->ifa_netmask->sa_len;
1918 			while (cp3 < cplim) {
1919 				if ((*cp++ ^ *cp2++) & *cp3++) {
1920 					/* want to continue for() loop */
1921 					goto next;
1922 				}
1923 			}
1924 			if (ifa_maybe == NULL ||
1925 			    rt_refines(ifa->ifa_netmask,
1926 			               ifa_maybe->ifa_netmask))
1927 				ifa_maybe = ifa;
1928 		}
1929 	}
1930 	return ifa_maybe;
1931 }
1932 
1933 struct ifaddr *
1934 ifa_ifwithnet_psref(const struct sockaddr *addr, struct psref *psref)
1935 {
1936 	struct ifaddr *ifa;
1937 	int s;
1938 
1939 	s = pserialize_read_enter();
1940 	ifa = ifa_ifwithnet(addr);
1941 	if (ifa != NULL)
1942 		ifa_acquire(ifa, psref);
1943 	pserialize_read_exit(s);
1944 
1945 	return ifa;
1946 }
1947 
1948 /*
1949  * Find the interface of the addresss.
1950  */
1951 struct ifaddr *
1952 ifa_ifwithladdr(const struct sockaddr *addr)
1953 {
1954 	struct ifaddr *ia;
1955 
1956 	if ((ia = ifa_ifwithaddr(addr)) || (ia = ifa_ifwithdstaddr(addr)) ||
1957 	    (ia = ifa_ifwithnet(addr)))
1958 		return ia;
1959 	return NULL;
1960 }
1961 
1962 struct ifaddr *
1963 ifa_ifwithladdr_psref(const struct sockaddr *addr, struct psref *psref)
1964 {
1965 	struct ifaddr *ifa;
1966 	int s;
1967 
1968 	s = pserialize_read_enter();
1969 	ifa = ifa_ifwithladdr(addr);
1970 	if (ifa != NULL)
1971 		ifa_acquire(ifa, psref);
1972 	pserialize_read_exit(s);
1973 
1974 	return ifa;
1975 }
1976 
1977 /*
1978  * Find an interface using a specific address family
1979  */
1980 struct ifaddr *
1981 ifa_ifwithaf(int af)
1982 {
1983 	struct ifnet *ifp;
1984 	struct ifaddr *ifa = NULL;
1985 	int s;
1986 
1987 	s = pserialize_read_enter();
1988 	IFNET_READER_FOREACH(ifp) {
1989 		if (if_is_deactivated(ifp))
1990 			continue;
1991 		IFADDR_READER_FOREACH(ifa, ifp) {
1992 			if (ifa->ifa_addr->sa_family == af)
1993 				goto out;
1994 		}
1995 	}
1996 out:
1997 	pserialize_read_exit(s);
1998 	return ifa;
1999 }
2000 
2001 /*
2002  * Find an interface address specific to an interface best matching
2003  * a given address.
2004  */
2005 struct ifaddr *
2006 ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp)
2007 {
2008 	struct ifaddr *ifa;
2009 	const char *cp, *cp2, *cp3;
2010 	const char *cplim;
2011 	struct ifaddr *ifa_maybe = 0;
2012 	u_int af = addr->sa_family;
2013 
2014 	if (if_is_deactivated(ifp))
2015 		return NULL;
2016 
2017 	if (af >= AF_MAX)
2018 		return NULL;
2019 
2020 	IFADDR_READER_FOREACH(ifa, ifp) {
2021 		if (ifa->ifa_addr->sa_family != af)
2022 			continue;
2023 		ifa_maybe = ifa;
2024 		if (ifa->ifa_netmask == NULL) {
2025 			if (equal(addr, ifa->ifa_addr) ||
2026 			    (ifa->ifa_dstaddr &&
2027 			     equal(addr, ifa->ifa_dstaddr)))
2028 				return ifa;
2029 			continue;
2030 		}
2031 		cp = addr->sa_data;
2032 		cp2 = ifa->ifa_addr->sa_data;
2033 		cp3 = ifa->ifa_netmask->sa_data;
2034 		cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2035 		for (; cp3 < cplim; cp3++) {
2036 			if ((*cp++ ^ *cp2++) & *cp3)
2037 				break;
2038 		}
2039 		if (cp3 == cplim)
2040 			return ifa;
2041 	}
2042 	return ifa_maybe;
2043 }
2044 
2045 struct ifaddr *
2046 ifaof_ifpforaddr_psref(const struct sockaddr *addr, struct ifnet *ifp,
2047     struct psref *psref)
2048 {
2049 	struct ifaddr *ifa;
2050 	int s;
2051 
2052 	s = pserialize_read_enter();
2053 	ifa = ifaof_ifpforaddr(addr, ifp);
2054 	if (ifa != NULL)
2055 		ifa_acquire(ifa, psref);
2056 	pserialize_read_exit(s);
2057 
2058 	return ifa;
2059 }
2060 
2061 /*
2062  * Default action when installing a route with a Link Level gateway.
2063  * Lookup an appropriate real ifa to point to.
2064  * This should be moved to /sys/net/link.c eventually.
2065  */
2066 void
2067 link_rtrequest(int cmd, struct rtentry *rt, const struct rt_addrinfo *info)
2068 {
2069 	struct ifaddr *ifa;
2070 	const struct sockaddr *dst;
2071 	struct ifnet *ifp;
2072 	struct psref psref;
2073 
2074 	if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
2075 	    (ifp = ifa->ifa_ifp) == NULL || (dst = rt_getkey(rt)) == NULL)
2076 		return;
2077 	if ((ifa = ifaof_ifpforaddr_psref(dst, ifp, &psref)) != NULL) {
2078 		rt_replace_ifa(rt, ifa);
2079 		if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2080 			ifa->ifa_rtrequest(cmd, rt, info);
2081 		ifa_release(ifa, &psref);
2082 	}
2083 }
2084 
2085 /*
2086  * bitmask macros to manage a densely packed link_state change queue.
2087  * Because we need to store LINK_STATE_UNKNOWN(0), LINK_STATE_DOWN(1) and
2088  * LINK_STATE_UP(2) we need 2 bits for each state change.
2089  * As a state change to store is 0, treat all bits set as an unset item.
2090  */
2091 #define LQ_ITEM_BITS		2
2092 #define LQ_ITEM_MASK		((1 << LQ_ITEM_BITS) - 1)
2093 #define LQ_MASK(i)		(LQ_ITEM_MASK << (i) * LQ_ITEM_BITS)
2094 #define LINK_STATE_UNSET	LQ_ITEM_MASK
2095 #define LQ_ITEM(q, i)		(((q) & LQ_MASK((i))) >> (i) * LQ_ITEM_BITS)
2096 #define LQ_STORE(q, i, v)						      \
2097 	do {								      \
2098 		(q) &= ~LQ_MASK((i));					      \
2099 		(q) |= (v) << (i) * LQ_ITEM_BITS;			      \
2100 	} while (0 /* CONSTCOND */)
2101 #define LQ_MAX(q)		((sizeof((q)) * NBBY) / LQ_ITEM_BITS)
2102 #define LQ_POP(q, v)							      \
2103 	do {								      \
2104 		(v) = LQ_ITEM((q), 0);					      \
2105 		(q) >>= LQ_ITEM_BITS;					      \
2106 		(q) |= LINK_STATE_UNSET << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;  \
2107 	} while (0 /* CONSTCOND */)
2108 #define LQ_PUSH(q, v)							      \
2109 	do {								      \
2110 		(q) >>= LQ_ITEM_BITS;					      \
2111 		(q) |= (v) << (LQ_MAX((q)) - 1) * LQ_ITEM_BITS;		      \
2112 	} while (0 /* CONSTCOND */)
2113 #define LQ_FIND_UNSET(q, i)						      \
2114 	for ((i) = 0; i < LQ_MAX((q)); (i)++) {				      \
2115 		if (LQ_ITEM((q), (i)) == LINK_STATE_UNSET)		      \
2116 			break;						      \
2117 	}
2118 /*
2119  * Handle a change in the interface link state and
2120  * queue notifications.
2121  */
2122 void
2123 if_link_state_change(struct ifnet *ifp, int link_state)
2124 {
2125 	int s, idx;
2126 
2127 	KASSERTMSG(if_is_link_state_changeable(ifp),
2128 	    "%s: IFEF_NO_LINK_STATE_CHANGE must not be set, but if_extflags=0x%x",
2129 	    ifp->if_xname, ifp->if_extflags);
2130 
2131 	/* Ensure change is to a valid state */
2132 	switch (link_state) {
2133 	case LINK_STATE_UNKNOWN:	/* FALLTHROUGH */
2134 	case LINK_STATE_DOWN:		/* FALLTHROUGH */
2135 	case LINK_STATE_UP:
2136 		break;
2137 	default:
2138 #ifdef DEBUG
2139 		printf("%s: invalid link state %d\n",
2140 		    ifp->if_xname, link_state);
2141 #endif
2142 		return;
2143 	}
2144 
2145 	s = splnet();
2146 
2147 	/* Find the last unset event in the queue. */
2148 	LQ_FIND_UNSET(ifp->if_link_queue, idx);
2149 
2150 	/*
2151 	 * Ensure link_state doesn't match the last event in the queue.
2152 	 * ifp->if_link_state is not checked and set here because
2153 	 * that would present an inconsistent picture to the system.
2154 	 */
2155 	if (idx != 0 &&
2156 	    LQ_ITEM(ifp->if_link_queue, idx - 1) == (uint8_t)link_state)
2157 		goto out;
2158 
2159 	/* Handle queue overflow. */
2160 	if (idx == LQ_MAX(ifp->if_link_queue)) {
2161 		uint8_t lost;
2162 
2163 		/*
2164 		 * The DOWN state must be protected from being pushed off
2165 		 * the queue to ensure that userland will always be
2166 		 * in a sane state.
2167 		 * Because DOWN is protected, there is no need to protect
2168 		 * UNKNOWN.
2169 		 * It should be invalid to change from any other state to
2170 		 * UNKNOWN anyway ...
2171 		 */
2172 		lost = LQ_ITEM(ifp->if_link_queue, 0);
2173 		LQ_PUSH(ifp->if_link_queue, (uint8_t)link_state);
2174 		if (lost == LINK_STATE_DOWN) {
2175 			lost = LQ_ITEM(ifp->if_link_queue, 0);
2176 			LQ_STORE(ifp->if_link_queue, 0, LINK_STATE_DOWN);
2177 		}
2178 		printf("%s: lost link state change %s\n",
2179 		    ifp->if_xname,
2180 		    lost == LINK_STATE_UP ? "UP" :
2181 		    lost == LINK_STATE_DOWN ? "DOWN" :
2182 		    "UNKNOWN");
2183 	} else
2184 		LQ_STORE(ifp->if_link_queue, idx, (uint8_t)link_state);
2185 
2186 	softint_schedule(ifp->if_link_si);
2187 
2188 out:
2189 	splx(s);
2190 }
2191 
2192 /*
2193  * Handle interface link state change notifications.
2194  * Must be called at splnet().
2195  */
2196 static void
2197 if_link_state_change0(struct ifnet *ifp, int link_state)
2198 {
2199 	struct domain *dp;
2200 
2201 	/* Ensure the change is still valid. */
2202 	if (ifp->if_link_state == link_state)
2203 		return;
2204 
2205 #ifdef DEBUG
2206 	log(LOG_DEBUG, "%s: link state %s (was %s)\n", ifp->if_xname,
2207 		link_state == LINK_STATE_UP ? "UP" :
2208 		link_state == LINK_STATE_DOWN ? "DOWN" :
2209 		"UNKNOWN",
2210 		ifp->if_link_state == LINK_STATE_UP ? "UP" :
2211 		ifp->if_link_state == LINK_STATE_DOWN ? "DOWN" :
2212 		"UNKNOWN");
2213 #endif
2214 
2215 	/*
2216 	 * When going from UNKNOWN to UP, we need to mark existing
2217 	 * addresses as tentative and restart DAD as we may have
2218 	 * erroneously not found a duplicate.
2219 	 *
2220 	 * This needs to happen before rt_ifmsg to avoid a race where
2221 	 * listeners would have an address and expect it to work right
2222 	 * away.
2223 	 */
2224 	if (link_state == LINK_STATE_UP &&
2225 	    ifp->if_link_state == LINK_STATE_UNKNOWN)
2226 	{
2227 		DOMAIN_FOREACH(dp) {
2228 			if (dp->dom_if_link_state_change != NULL)
2229 				dp->dom_if_link_state_change(ifp,
2230 				    LINK_STATE_DOWN);
2231 		}
2232 	}
2233 
2234 	ifp->if_link_state = link_state;
2235 
2236 	/* Notify that the link state has changed. */
2237 	rt_ifmsg(ifp);
2238 
2239 #if NCARP > 0
2240 	if (ifp->if_carp)
2241 		carp_carpdev_state(ifp);
2242 #endif
2243 
2244 	DOMAIN_FOREACH(dp) {
2245 		if (dp->dom_if_link_state_change != NULL)
2246 			dp->dom_if_link_state_change(ifp, link_state);
2247 	}
2248 }
2249 
2250 /*
2251  * Process the interface link state change queue.
2252  */
2253 static void
2254 if_link_state_change_si(void *arg)
2255 {
2256 	struct ifnet *ifp = arg;
2257 	int s;
2258 	uint8_t state;
2259 
2260 	s = splnet();
2261 
2262 	/* Pop a link state change from the queue and process it. */
2263 	LQ_POP(ifp->if_link_queue, state);
2264 	if_link_state_change0(ifp, state);
2265 
2266 	/* If there is a link state change to come, schedule it. */
2267 	if (LQ_ITEM(ifp->if_link_queue, 0) != LINK_STATE_UNSET)
2268 		softint_schedule(ifp->if_link_si);
2269 
2270 	splx(s);
2271 }
2272 
2273 /*
2274  * Default action when installing a local route on a point-to-point
2275  * interface.
2276  */
2277 void
2278 p2p_rtrequest(int req, struct rtentry *rt,
2279     __unused const struct rt_addrinfo *info)
2280 {
2281 	struct ifnet *ifp = rt->rt_ifp;
2282 	struct ifaddr *ifa, *lo0ifa;
2283 	int s = pserialize_read_enter();
2284 
2285 	switch (req) {
2286 	case RTM_ADD:
2287 		if ((rt->rt_flags & RTF_LOCAL) == 0)
2288 			break;
2289 
2290 		rt->rt_ifp = lo0ifp;
2291 
2292 		IFADDR_READER_FOREACH(ifa, ifp) {
2293 			if (equal(rt_getkey(rt), ifa->ifa_addr))
2294 				break;
2295 		}
2296 		if (ifa == NULL)
2297 			break;
2298 
2299 		/*
2300 		 * Ensure lo0 has an address of the same family.
2301 		 */
2302 		IFADDR_READER_FOREACH(lo0ifa, lo0ifp) {
2303 			if (lo0ifa->ifa_addr->sa_family ==
2304 			    ifa->ifa_addr->sa_family)
2305 				break;
2306 		}
2307 		if (lo0ifa == NULL)
2308 			break;
2309 
2310 		/*
2311 		 * Make sure to set rt->rt_ifa to the interface
2312 		 * address we are using, otherwise we will have trouble
2313 		 * with source address selection.
2314 		 */
2315 		if (ifa != rt->rt_ifa)
2316 			rt_replace_ifa(rt, ifa);
2317 		break;
2318 	case RTM_DELETE:
2319 	default:
2320 		break;
2321 	}
2322 	pserialize_read_exit(s);
2323 }
2324 
2325 /*
2326  * Mark an interface down and notify protocols of
2327  * the transition.
2328  * NOTE: must be called at splsoftnet or equivalent.
2329  */
2330 void
2331 if_down(struct ifnet *ifp)
2332 {
2333 	struct ifaddr *ifa;
2334 	struct domain *dp;
2335 	int s, bound;
2336 	struct psref psref;
2337 
2338 	ifp->if_flags &= ~IFF_UP;
2339 	nanotime(&ifp->if_lastchange);
2340 
2341 	bound = curlwp_bind();
2342 	s = pserialize_read_enter();
2343 	IFADDR_READER_FOREACH(ifa, ifp) {
2344 		ifa_acquire(ifa, &psref);
2345 		pserialize_read_exit(s);
2346 
2347 		pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2348 
2349 		s = pserialize_read_enter();
2350 		ifa_release(ifa, &psref);
2351 	}
2352 	pserialize_read_exit(s);
2353 	curlwp_bindx(bound);
2354 
2355 	IFQ_PURGE(&ifp->if_snd);
2356 #if NCARP > 0
2357 	if (ifp->if_carp)
2358 		carp_carpdev_state(ifp);
2359 #endif
2360 	rt_ifmsg(ifp);
2361 	DOMAIN_FOREACH(dp) {
2362 		if (dp->dom_if_down)
2363 			dp->dom_if_down(ifp);
2364 	}
2365 }
2366 
2367 /*
2368  * Mark an interface up and notify protocols of
2369  * the transition.
2370  * NOTE: must be called at splsoftnet or equivalent.
2371  */
2372 void
2373 if_up(struct ifnet *ifp)
2374 {
2375 #ifdef notyet
2376 	struct ifaddr *ifa;
2377 #endif
2378 	struct domain *dp;
2379 
2380 	ifp->if_flags |= IFF_UP;
2381 	nanotime(&ifp->if_lastchange);
2382 #ifdef notyet
2383 	/* this has no effect on IP, and will kill all ISO connections XXX */
2384 	IFADDR_READER_FOREACH(ifa, ifp)
2385 		pfctlinput(PRC_IFUP, ifa->ifa_addr);
2386 #endif
2387 #if NCARP > 0
2388 	if (ifp->if_carp)
2389 		carp_carpdev_state(ifp);
2390 #endif
2391 	rt_ifmsg(ifp);
2392 	DOMAIN_FOREACH(dp) {
2393 		if (dp->dom_if_up)
2394 			dp->dom_if_up(ifp);
2395 	}
2396 }
2397 
2398 /*
2399  * Handle interface slowtimo timer routine.  Called
2400  * from softclock, we decrement timer (if set) and
2401  * call the appropriate interface routine on expiration.
2402  */
2403 static void
2404 if_slowtimo(void *arg)
2405 {
2406 	void (*slowtimo)(struct ifnet *);
2407 	struct ifnet *ifp = arg;
2408 	int s;
2409 
2410 	slowtimo = ifp->if_slowtimo;
2411 	if (__predict_false(slowtimo == NULL))
2412 		return;
2413 
2414 	s = splnet();
2415 	if (ifp->if_timer != 0 && --ifp->if_timer == 0)
2416 		(*slowtimo)(ifp);
2417 
2418 	splx(s);
2419 
2420 	if (__predict_true(ifp->if_slowtimo != NULL))
2421 		callout_schedule(ifp->if_slowtimo_ch, hz / IFNET_SLOWHZ);
2422 }
2423 
2424 /*
2425  * Set/clear promiscuous mode on interface ifp based on the truth value
2426  * of pswitch.  The calls are reference counted so that only the first
2427  * "on" request actually has an effect, as does the final "off" request.
2428  * Results are undefined if the "off" and "on" requests are not matched.
2429  */
2430 int
2431 ifpromisc(struct ifnet *ifp, int pswitch)
2432 {
2433 	int pcount, ret;
2434 	short nflags;
2435 
2436 	pcount = ifp->if_pcount;
2437 	if (pswitch) {
2438 		/*
2439 		 * Allow the device to be "placed" into promiscuous
2440 		 * mode even if it is not configured up.  It will
2441 		 * consult IFF_PROMISC when it is brought up.
2442 		 */
2443 		if (ifp->if_pcount++ != 0)
2444 			return 0;
2445 		nflags = ifp->if_flags | IFF_PROMISC;
2446 	} else {
2447 		if (--ifp->if_pcount > 0)
2448 			return 0;
2449 		nflags = ifp->if_flags & ~IFF_PROMISC;
2450 	}
2451 	ret = if_flags_set(ifp, nflags);
2452 	/* Restore interface state if not successful. */
2453 	if (ret != 0) {
2454 		ifp->if_pcount = pcount;
2455 	}
2456 	return ret;
2457 }
2458 
2459 /*
2460  * Map interface name to
2461  * interface structure pointer.
2462  */
2463 struct ifnet *
2464 ifunit(const char *name)
2465 {
2466 	struct ifnet *ifp;
2467 	const char *cp = name;
2468 	u_int unit = 0;
2469 	u_int i;
2470 	int s;
2471 
2472 	/*
2473 	 * If the entire name is a number, treat it as an ifindex.
2474 	 */
2475 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2476 		unit = unit * 10 + (*cp - '0');
2477 	}
2478 
2479 	/*
2480 	 * If the number took all of the name, then it's a valid ifindex.
2481 	 */
2482 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2483 		if (unit >= if_indexlim)
2484 			return NULL;
2485 		ifp = ifindex2ifnet[unit];
2486 		if (ifp == NULL || if_is_deactivated(ifp))
2487 			return NULL;
2488 		return ifp;
2489 	}
2490 
2491 	ifp = NULL;
2492 	s = pserialize_read_enter();
2493 	IFNET_READER_FOREACH(ifp) {
2494 		if (if_is_deactivated(ifp))
2495 			continue;
2496 	 	if (strcmp(ifp->if_xname, name) == 0)
2497 			goto out;
2498 	}
2499 out:
2500 	pserialize_read_exit(s);
2501 	return ifp;
2502 }
2503 
2504 /*
2505  * Get a reference of an ifnet object by an interface name.
2506  * The returned reference is protected by psref(9). The caller
2507  * must release a returned reference by if_put after use.
2508  */
2509 struct ifnet *
2510 if_get(const char *name, struct psref *psref)
2511 {
2512 	struct ifnet *ifp;
2513 	const char *cp = name;
2514 	u_int unit = 0;
2515 	u_int i;
2516 	int s;
2517 
2518 	/*
2519 	 * If the entire name is a number, treat it as an ifindex.
2520 	 */
2521 	for (i = 0; i < IFNAMSIZ && *cp >= '0' && *cp <= '9'; i++, cp++) {
2522 		unit = unit * 10 + (*cp - '0');
2523 	}
2524 
2525 	/*
2526 	 * If the number took all of the name, then it's a valid ifindex.
2527 	 */
2528 	if (i == IFNAMSIZ || (cp != name && *cp == '\0')) {
2529 		if (unit >= if_indexlim)
2530 			return NULL;
2531 		ifp = ifindex2ifnet[unit];
2532 		if (ifp == NULL || if_is_deactivated(ifp))
2533 			return NULL;
2534 		return ifp;
2535 	}
2536 
2537 	ifp = NULL;
2538 	s = pserialize_read_enter();
2539 	IFNET_READER_FOREACH(ifp) {
2540 		if (if_is_deactivated(ifp))
2541 			continue;
2542 		if (strcmp(ifp->if_xname, name) == 0) {
2543 			psref_acquire(psref, &ifp->if_psref,
2544 			    ifnet_psref_class);
2545 			goto out;
2546 		}
2547 	}
2548 out:
2549 	pserialize_read_exit(s);
2550 	return ifp;
2551 }
2552 
2553 /*
2554  * Release a reference of an ifnet object given by if_get or
2555  * if_get_byindex.
2556  */
2557 void
2558 if_put(const struct ifnet *ifp, struct psref *psref)
2559 {
2560 
2561 	if (ifp == NULL)
2562 		return;
2563 
2564 	psref_release(psref, &ifp->if_psref, ifnet_psref_class);
2565 }
2566 
2567 ifnet_t *
2568 if_byindex(u_int idx)
2569 {
2570 	ifnet_t *ifp;
2571 
2572 	ifp = (idx < if_indexlim) ? ifindex2ifnet[idx] : NULL;
2573 	if (ifp != NULL && if_is_deactivated(ifp))
2574 		ifp = NULL;
2575 	return ifp;
2576 }
2577 
2578 /*
2579  * Get a reference of an ifnet object by an interface index.
2580  * The returned reference is protected by psref(9). The caller
2581  * must release a returned reference by if_put after use.
2582  */
2583 ifnet_t *
2584 if_get_byindex(u_int idx, struct psref *psref)
2585 {
2586 	ifnet_t *ifp;
2587 	int s;
2588 
2589 	s = pserialize_read_enter();
2590 	ifp = (__predict_true(idx < if_indexlim)) ? ifindex2ifnet[idx] : NULL;
2591 	if (ifp != NULL && if_is_deactivated(ifp))
2592 		ifp = NULL;
2593 	if (__predict_true(ifp != NULL))
2594 		psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2595 	pserialize_read_exit(s);
2596 
2597 	return ifp;
2598 }
2599 
2600 /*
2601  * XXX it's safe only if the passed ifp is guaranteed to not be freed,
2602  * for example the ifp is already held or some other object is held which
2603  * guarantes the ifp to not be freed indirectly.
2604  */
2605 void
2606 if_acquire_NOMPSAFE(struct ifnet *ifp, struct psref *psref)
2607 {
2608 
2609 	KASSERT(ifp->if_index != 0);
2610 	psref_acquire(psref, &ifp->if_psref, ifnet_psref_class);
2611 }
2612 
2613 bool
2614 if_held(struct ifnet *ifp)
2615 {
2616 
2617 	return psref_held(&ifp->if_psref, ifnet_psref_class);
2618 }
2619 
2620 
2621 /* common */
2622 int
2623 ifioctl_common(struct ifnet *ifp, u_long cmd, void *data)
2624 {
2625 	int s;
2626 	struct ifreq *ifr;
2627 	struct ifcapreq *ifcr;
2628 	struct ifdatareq *ifdr;
2629 
2630 	switch (cmd) {
2631 	case SIOCSIFCAP:
2632 		ifcr = data;
2633 		if ((ifcr->ifcr_capenable & ~ifp->if_capabilities) != 0)
2634 			return EINVAL;
2635 
2636 		if (ifcr->ifcr_capenable == ifp->if_capenable)
2637 			return 0;
2638 
2639 		ifp->if_capenable = ifcr->ifcr_capenable;
2640 
2641 		/* Pre-compute the checksum flags mask. */
2642 		ifp->if_csum_flags_tx = 0;
2643 		ifp->if_csum_flags_rx = 0;
2644 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Tx) {
2645 			ifp->if_csum_flags_tx |= M_CSUM_IPv4;
2646 		}
2647 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) {
2648 			ifp->if_csum_flags_rx |= M_CSUM_IPv4;
2649 		}
2650 
2651 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Tx) {
2652 			ifp->if_csum_flags_tx |= M_CSUM_TCPv4;
2653 		}
2654 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) {
2655 			ifp->if_csum_flags_rx |= M_CSUM_TCPv4;
2656 		}
2657 
2658 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Tx) {
2659 			ifp->if_csum_flags_tx |= M_CSUM_UDPv4;
2660 		}
2661 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) {
2662 			ifp->if_csum_flags_rx |= M_CSUM_UDPv4;
2663 		}
2664 
2665 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Tx) {
2666 			ifp->if_csum_flags_tx |= M_CSUM_TCPv6;
2667 		}
2668 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) {
2669 			ifp->if_csum_flags_rx |= M_CSUM_TCPv6;
2670 		}
2671 
2672 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Tx) {
2673 			ifp->if_csum_flags_tx |= M_CSUM_UDPv6;
2674 		}
2675 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) {
2676 			ifp->if_csum_flags_rx |= M_CSUM_UDPv6;
2677 		}
2678 		if (ifp->if_flags & IFF_UP)
2679 			return ENETRESET;
2680 		return 0;
2681 	case SIOCSIFFLAGS:
2682 		ifr = data;
2683 		if (ifp->if_flags & IFF_UP && (ifr->ifr_flags & IFF_UP) == 0) {
2684 			s = splnet();
2685 			if_down(ifp);
2686 			splx(s);
2687 		}
2688 		if (ifr->ifr_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) {
2689 			s = splnet();
2690 			if_up(ifp);
2691 			splx(s);
2692 		}
2693 		ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2694 			(ifr->ifr_flags &~ IFF_CANTCHANGE);
2695 		break;
2696 	case SIOCGIFFLAGS:
2697 		ifr = data;
2698 		ifr->ifr_flags = ifp->if_flags;
2699 		break;
2700 
2701 	case SIOCGIFMETRIC:
2702 		ifr = data;
2703 		ifr->ifr_metric = ifp->if_metric;
2704 		break;
2705 
2706 	case SIOCGIFMTU:
2707 		ifr = data;
2708 		ifr->ifr_mtu = ifp->if_mtu;
2709 		break;
2710 
2711 	case SIOCGIFDLT:
2712 		ifr = data;
2713 		ifr->ifr_dlt = ifp->if_dlt;
2714 		break;
2715 
2716 	case SIOCGIFCAP:
2717 		ifcr = data;
2718 		ifcr->ifcr_capabilities = ifp->if_capabilities;
2719 		ifcr->ifcr_capenable = ifp->if_capenable;
2720 		break;
2721 
2722 	case SIOCSIFMETRIC:
2723 		ifr = data;
2724 		ifp->if_metric = ifr->ifr_metric;
2725 		break;
2726 
2727 	case SIOCGIFDATA:
2728 		ifdr = data;
2729 		ifdr->ifdr_data = ifp->if_data;
2730 		break;
2731 
2732 	case SIOCGIFINDEX:
2733 		ifr = data;
2734 		ifr->ifr_index = ifp->if_index;
2735 		break;
2736 
2737 	case SIOCZIFDATA:
2738 		ifdr = data;
2739 		ifdr->ifdr_data = ifp->if_data;
2740 		/*
2741 		 * Assumes that the volatile counters that can be
2742 		 * zero'ed are at the end of if_data.
2743 		 */
2744 		memset(&ifp->if_data.ifi_ipackets, 0, sizeof(ifp->if_data) -
2745 		    offsetof(struct if_data, ifi_ipackets));
2746 		/*
2747 		 * The memset() clears to the bottm of if_data. In the area,
2748 		 * if_lastchange is included. Please be careful if new entry
2749 		 * will be added into if_data or rewite this.
2750 		 *
2751 		 * And also, update if_lastchnage.
2752 		 */
2753 		getnanotime(&ifp->if_lastchange);
2754 		break;
2755 	case SIOCSIFMTU:
2756 		ifr = data;
2757 		if (ifp->if_mtu == ifr->ifr_mtu)
2758 			break;
2759 		ifp->if_mtu = ifr->ifr_mtu;
2760 		/*
2761 		 * If the link MTU changed, do network layer specific procedure.
2762 		 */
2763 #ifdef INET6
2764 		if (in6_present)
2765 			nd6_setmtu(ifp);
2766 #endif
2767 		return ENETRESET;
2768 	default:
2769 		return ENOTTY;
2770 	}
2771 	return 0;
2772 }
2773 
2774 int
2775 ifaddrpref_ioctl(struct socket *so, u_long cmd, void *data, struct ifnet *ifp)
2776 {
2777 	struct if_addrprefreq *ifap = (struct if_addrprefreq *)data;
2778 	struct ifaddr *ifa;
2779 	const struct sockaddr *any, *sa;
2780 	union {
2781 		struct sockaddr sa;
2782 		struct sockaddr_storage ss;
2783 	} u, v;
2784 	int s, error = 0;
2785 
2786 	switch (cmd) {
2787 	case SIOCSIFADDRPREF:
2788 		if (kauth_authorize_network(curlwp->l_cred, KAUTH_NETWORK_INTERFACE,
2789 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
2790 		    NULL) != 0)
2791 			return EPERM;
2792 	case SIOCGIFADDRPREF:
2793 		break;
2794 	default:
2795 		return EOPNOTSUPP;
2796 	}
2797 
2798 	/* sanity checks */
2799 	if (data == NULL || ifp == NULL) {
2800 		panic("invalid argument to %s", __func__);
2801 		/*NOTREACHED*/
2802 	}
2803 
2804 	/* address must be specified on ADD and DELETE */
2805 	sa = sstocsa(&ifap->ifap_addr);
2806 	if (sa->sa_family != sofamily(so))
2807 		return EINVAL;
2808 	if ((any = sockaddr_any(sa)) == NULL || sa->sa_len != any->sa_len)
2809 		return EINVAL;
2810 
2811 	sockaddr_externalize(&v.sa, sizeof(v.ss), sa);
2812 
2813 	s = pserialize_read_enter();
2814 	IFADDR_READER_FOREACH(ifa, ifp) {
2815 		if (ifa->ifa_addr->sa_family != sa->sa_family)
2816 			continue;
2817 		sockaddr_externalize(&u.sa, sizeof(u.ss), ifa->ifa_addr);
2818 		if (sockaddr_cmp(&u.sa, &v.sa) == 0)
2819 			break;
2820 	}
2821 	if (ifa == NULL) {
2822 		error = EADDRNOTAVAIL;
2823 		goto out;
2824 	}
2825 
2826 	switch (cmd) {
2827 	case SIOCSIFADDRPREF:
2828 		ifa->ifa_preference = ifap->ifap_preference;
2829 		goto out;
2830 	case SIOCGIFADDRPREF:
2831 		/* fill in the if_laddrreq structure */
2832 		(void)sockaddr_copy(sstosa(&ifap->ifap_addr),
2833 		    sizeof(ifap->ifap_addr), ifa->ifa_addr);
2834 		ifap->ifap_preference = ifa->ifa_preference;
2835 		goto out;
2836 	default:
2837 		error = EOPNOTSUPP;
2838 	}
2839 out:
2840 	pserialize_read_exit(s);
2841 	return error;
2842 }
2843 
2844 /*
2845  * Interface ioctls.
2846  */
2847 static int
2848 doifioctl(struct socket *so, u_long cmd, void *data, struct lwp *l)
2849 {
2850 	struct ifnet *ifp;
2851 	struct ifreq *ifr;
2852 	int error = 0;
2853 #if defined(COMPAT_OSOCK) || defined(COMPAT_OIFREQ)
2854 	u_long ocmd = cmd;
2855 #endif
2856 	short oif_flags;
2857 #ifdef COMPAT_OIFREQ
2858 	struct ifreq ifrb;
2859 	struct oifreq *oifr = NULL;
2860 #endif
2861 	int r;
2862 	struct psref psref;
2863 	int bound;
2864 
2865 	switch (cmd) {
2866 #ifdef COMPAT_OIFREQ
2867 	case OSIOCGIFCONF:
2868 	case OOSIOCGIFCONF:
2869 		return compat_ifconf(cmd, data);
2870 #endif
2871 #ifdef COMPAT_OIFDATA
2872 	case OSIOCGIFDATA:
2873 	case OSIOCZIFDATA:
2874 		return compat_ifdatareq(l, cmd, data);
2875 #endif
2876 	case SIOCGIFCONF:
2877 		return ifconf(cmd, data);
2878 	case SIOCINITIFADDR:
2879 		return EPERM;
2880 	}
2881 
2882 #ifdef COMPAT_OIFREQ
2883 	cmd = (*vec_compat_cvtcmd)(cmd);
2884 	if (cmd != ocmd) {
2885 		oifr = data;
2886 		data = ifr = &ifrb;
2887 		ifreqo2n(oifr, ifr);
2888 	} else
2889 #endif
2890 		ifr = data;
2891 
2892 	switch (cmd) {
2893 	case SIOCIFCREATE:
2894 	case SIOCIFDESTROY:
2895 		bound = curlwp_bind();
2896 		if (l != NULL) {
2897 			ifp = if_get(ifr->ifr_name, &psref);
2898 			error = kauth_authorize_network(l->l_cred,
2899 			    KAUTH_NETWORK_INTERFACE,
2900 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2901 			    (void *)cmd, NULL);
2902 			if (ifp != NULL)
2903 				if_put(ifp, &psref);
2904 			if (error != 0) {
2905 				curlwp_bindx(bound);
2906 				return error;
2907 			}
2908 		}
2909 		mutex_enter(&if_clone_mtx);
2910 		r = (cmd == SIOCIFCREATE) ?
2911 			if_clone_create(ifr->ifr_name) :
2912 			if_clone_destroy(ifr->ifr_name);
2913 		mutex_exit(&if_clone_mtx);
2914 		curlwp_bindx(bound);
2915 		return r;
2916 
2917 	case SIOCIFGCLONERS:
2918 		{
2919 			struct if_clonereq *req = (struct if_clonereq *)data;
2920 			return if_clone_list(req->ifcr_count, req->ifcr_buffer,
2921 			    &req->ifcr_total);
2922 		}
2923 	}
2924 
2925 	bound = curlwp_bind();
2926 	ifp = if_get(ifr->ifr_name, &psref);
2927 	if (ifp == NULL) {
2928 		curlwp_bindx(bound);
2929 		return ENXIO;
2930 	}
2931 
2932 	switch (cmd) {
2933 	case SIOCALIFADDR:
2934 	case SIOCDLIFADDR:
2935 	case SIOCSIFADDRPREF:
2936 	case SIOCSIFFLAGS:
2937 	case SIOCSIFCAP:
2938 	case SIOCSIFMETRIC:
2939 	case SIOCZIFDATA:
2940 	case SIOCSIFMTU:
2941 	case SIOCSIFPHYADDR:
2942 	case SIOCDIFPHYADDR:
2943 #ifdef INET6
2944 	case SIOCSIFPHYADDR_IN6:
2945 #endif
2946 	case SIOCSLIFPHYADDR:
2947 	case SIOCADDMULTI:
2948 	case SIOCDELMULTI:
2949 	case SIOCSIFMEDIA:
2950 	case SIOCSDRVSPEC:
2951 	case SIOCG80211:
2952 	case SIOCS80211:
2953 	case SIOCS80211NWID:
2954 	case SIOCS80211NWKEY:
2955 	case SIOCS80211POWER:
2956 	case SIOCS80211BSSID:
2957 	case SIOCS80211CHANNEL:
2958 	case SIOCSLINKSTR:
2959 		if (l != NULL) {
2960 			error = kauth_authorize_network(l->l_cred,
2961 			    KAUTH_NETWORK_INTERFACE,
2962 			    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp,
2963 			    (void *)cmd, NULL);
2964 			if (error != 0)
2965 				goto out;
2966 		}
2967 	}
2968 
2969 	oif_flags = ifp->if_flags;
2970 
2971 	mutex_enter(ifp->if_ioctl_lock);
2972 
2973 	error = (*ifp->if_ioctl)(ifp, cmd, data);
2974 	if (error != ENOTTY)
2975 		;
2976 	else if (so->so_proto == NULL)
2977 		error = EOPNOTSUPP;
2978 	else {
2979 #ifdef COMPAT_OSOCK
2980 		if (vec_compat_ifioctl != NULL)
2981 			error = (*vec_compat_ifioctl)(so, ocmd, cmd, data, l);
2982 		else
2983 #endif
2984 			error = (*so->so_proto->pr_usrreqs->pr_ioctl)(so,
2985 			    cmd, data, ifp);
2986 	}
2987 
2988 	if (((oif_flags ^ ifp->if_flags) & IFF_UP) != 0) {
2989 		if ((ifp->if_flags & IFF_UP) != 0) {
2990 			int s = splnet();
2991 			if_up(ifp);
2992 			splx(s);
2993 		}
2994 	}
2995 #ifdef COMPAT_OIFREQ
2996 	if (cmd != ocmd)
2997 		ifreqn2o(oifr, ifr);
2998 #endif
2999 
3000 	mutex_exit(ifp->if_ioctl_lock);
3001 out:
3002 	if_put(ifp, &psref);
3003 	curlwp_bindx(bound);
3004 	return error;
3005 }
3006 
3007 /*
3008  * Return interface configuration
3009  * of system.  List may be used
3010  * in later ioctl's (above) to get
3011  * other information.
3012  *
3013  * Each record is a struct ifreq.  Before the addition of
3014  * sockaddr_storage, the API rule was that sockaddr flavors that did
3015  * not fit would extend beyond the struct ifreq, with the next struct
3016  * ifreq starting sa_len beyond the struct sockaddr.  Because the
3017  * union in struct ifreq includes struct sockaddr_storage, every kind
3018  * of sockaddr must fit.  Thus, there are no longer any overlength
3019  * records.
3020  *
3021  * Records are added to the user buffer if they fit, and ifc_len is
3022  * adjusted to the length that was written.  Thus, the user is only
3023  * assured of getting the complete list if ifc_len on return is at
3024  * least sizeof(struct ifreq) less than it was on entry.
3025  *
3026  * If the user buffer pointer is NULL, this routine copies no data and
3027  * returns the amount of space that would be needed.
3028  *
3029  * Invariants:
3030  * ifrp points to the next part of the user's buffer to be used.  If
3031  * ifrp != NULL, space holds the number of bytes remaining that we may
3032  * write at ifrp.  Otherwise, space holds the number of bytes that
3033  * would have been written had there been adequate space.
3034  */
3035 /*ARGSUSED*/
3036 static int
3037 ifconf(u_long cmd, void *data)
3038 {
3039 	struct ifconf *ifc = (struct ifconf *)data;
3040 	struct ifnet *ifp;
3041 	struct ifaddr *ifa;
3042 	struct ifreq ifr, *ifrp = NULL;
3043 	int space = 0, error = 0;
3044 	const int sz = (int)sizeof(struct ifreq);
3045 	const bool docopy = ifc->ifc_req != NULL;
3046 	int s;
3047 	int bound;
3048 	struct psref psref;
3049 
3050 	if (docopy) {
3051 		space = ifc->ifc_len;
3052 		ifrp = ifc->ifc_req;
3053 	}
3054 
3055 	bound = curlwp_bind();
3056 	s = pserialize_read_enter();
3057 	IFNET_READER_FOREACH(ifp) {
3058 		psref_acquire(&psref, &ifp->if_psref, ifnet_psref_class);
3059 		pserialize_read_exit(s);
3060 
3061 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
3062 		    sizeof(ifr.ifr_name));
3063 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
3064 			error = ENAMETOOLONG;
3065 			goto release_exit;
3066 		}
3067 		if (IFADDR_READER_EMPTY(ifp)) {
3068 			/* Interface with no addresses - send zero sockaddr. */
3069 			memset(&ifr.ifr_addr, 0, sizeof(ifr.ifr_addr));
3070 			if (!docopy) {
3071 				space += sz;
3072 				continue;
3073 			}
3074 			if (space >= sz) {
3075 				error = copyout(&ifr, ifrp, sz);
3076 				if (error != 0)
3077 					goto release_exit;
3078 				ifrp++;
3079 				space -= sz;
3080 			}
3081 		}
3082 
3083 		IFADDR_READER_FOREACH(ifa, ifp) {
3084 			struct sockaddr *sa = ifa->ifa_addr;
3085 			/* all sockaddrs must fit in sockaddr_storage */
3086 			KASSERT(sa->sa_len <= sizeof(ifr.ifr_ifru));
3087 
3088 			if (!docopy) {
3089 				space += sz;
3090 				continue;
3091 			}
3092 			memcpy(&ifr.ifr_space, sa, sa->sa_len);
3093 			if (space >= sz) {
3094 				error = copyout(&ifr, ifrp, sz);
3095 				if (error != 0)
3096 					goto release_exit;
3097 				ifrp++; space -= sz;
3098 			}
3099 		}
3100 
3101 		s = pserialize_read_enter();
3102 		psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3103 	}
3104 	pserialize_read_exit(s);
3105 	curlwp_bindx(bound);
3106 
3107 	if (docopy) {
3108 		KASSERT(0 <= space && space <= ifc->ifc_len);
3109 		ifc->ifc_len -= space;
3110 	} else {
3111 		KASSERT(space >= 0);
3112 		ifc->ifc_len = space;
3113 	}
3114 	return (0);
3115 
3116 release_exit:
3117 	psref_release(&psref, &ifp->if_psref, ifnet_psref_class);
3118 	curlwp_bindx(bound);
3119 	return error;
3120 }
3121 
3122 int
3123 ifreq_setaddr(u_long cmd, struct ifreq *ifr, const struct sockaddr *sa)
3124 {
3125 	uint8_t len;
3126 #ifdef COMPAT_OIFREQ
3127 	struct ifreq ifrb;
3128 	struct oifreq *oifr = NULL;
3129 	u_long ocmd = cmd;
3130 	cmd = (*vec_compat_cvtcmd)(cmd);
3131 	if (cmd != ocmd) {
3132 		oifr = (struct oifreq *)(void *)ifr;
3133 		ifr = &ifrb;
3134 		ifreqo2n(oifr, ifr);
3135 		len = sizeof(oifr->ifr_addr);
3136 	} else
3137 #endif
3138 		len = sizeof(ifr->ifr_ifru.ifru_space);
3139 
3140 	if (len < sa->sa_len)
3141 		return EFBIG;
3142 
3143 	memset(&ifr->ifr_addr, 0, len);
3144 	sockaddr_copy(&ifr->ifr_addr, len, sa);
3145 
3146 #ifdef COMPAT_OIFREQ
3147 	if (cmd != ocmd)
3148 		ifreqn2o(oifr, ifr);
3149 #endif
3150 	return 0;
3151 }
3152 
3153 /*
3154  * wrapper function for the drivers which doesn't have if_transmit().
3155  */
3156 static int
3157 if_transmit(struct ifnet *ifp, struct mbuf *m)
3158 {
3159 	int s, error;
3160 
3161 	s = splnet();
3162 
3163 	IFQ_ENQUEUE(&ifp->if_snd, m, error);
3164 	if (error != 0) {
3165 		/* mbuf is already freed */
3166 		goto out;
3167 	}
3168 
3169 	ifp->if_obytes += m->m_pkthdr.len;;
3170 	if (m->m_flags & M_MCAST)
3171 		ifp->if_omcasts++;
3172 
3173 	if ((ifp->if_flags & IFF_OACTIVE) == 0)
3174 		if_start_lock(ifp);
3175 out:
3176 	splx(s);
3177 
3178 	return error;
3179 }
3180 
3181 int
3182 if_transmit_lock(struct ifnet *ifp, struct mbuf *m)
3183 {
3184 	int error;
3185 
3186 #ifdef ALTQ
3187 	KERNEL_LOCK(1, NULL);
3188 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3189 		error = if_transmit(ifp, m);
3190 		KERNEL_UNLOCK_ONE(NULL);
3191 	} else {
3192 		KERNEL_UNLOCK_ONE(NULL);
3193 		error = (*ifp->if_transmit)(ifp, m);
3194 	}
3195 #else /* !ALTQ */
3196 	error = (*ifp->if_transmit)(ifp, m);
3197 #endif /* !ALTQ */
3198 
3199 	return error;
3200 }
3201 
3202 /*
3203  * Queue message on interface, and start output if interface
3204  * not yet active.
3205  */
3206 int
3207 ifq_enqueue(struct ifnet *ifp, struct mbuf *m)
3208 {
3209 
3210 	return if_transmit_lock(ifp, m);
3211 }
3212 
3213 /*
3214  * Queue message on interface, possibly using a second fast queue
3215  */
3216 int
3217 ifq_enqueue2(struct ifnet *ifp, struct ifqueue *ifq, struct mbuf *m)
3218 {
3219 	int error = 0;
3220 
3221 	if (ifq != NULL
3222 #ifdef ALTQ
3223 	    && ALTQ_IS_ENABLED(&ifp->if_snd) == 0
3224 #endif
3225 	    ) {
3226 		if (IF_QFULL(ifq)) {
3227 			IF_DROP(&ifp->if_snd);
3228 			m_freem(m);
3229 			if (error == 0)
3230 				error = ENOBUFS;
3231 		} else
3232 			IF_ENQUEUE(ifq, m);
3233 	} else
3234 		IFQ_ENQUEUE(&ifp->if_snd, m, error);
3235 	if (error != 0) {
3236 		++ifp->if_oerrors;
3237 		return error;
3238 	}
3239 	return 0;
3240 }
3241 
3242 int
3243 if_addr_init(ifnet_t *ifp, struct ifaddr *ifa, const bool src)
3244 {
3245 	int rc;
3246 
3247 	if (ifp->if_initaddr != NULL)
3248 		rc = (*ifp->if_initaddr)(ifp, ifa, src);
3249 	else if (src ||
3250 		/* FIXME: may not hold if_ioctl_lock */
3251 	         (rc = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR, ifa)) == ENOTTY)
3252 		rc = (*ifp->if_ioctl)(ifp, SIOCINITIFADDR, ifa);
3253 
3254 	return rc;
3255 }
3256 
3257 int
3258 if_do_dad(struct ifnet *ifp)
3259 {
3260 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
3261 		return 0;
3262 
3263 	switch (ifp->if_type) {
3264 	case IFT_FAITH:
3265 		/*
3266 		 * These interfaces do not have the IFF_LOOPBACK flag,
3267 		 * but loop packets back.  We do not have to do DAD on such
3268 		 * interfaces.  We should even omit it, because loop-backed
3269 		 * responses would confuse the DAD procedure.
3270 		 */
3271 		return 0;
3272 	default:
3273 		/*
3274 		 * Our DAD routine requires the interface up and running.
3275 		 * However, some interfaces can be up before the RUNNING
3276 		 * status.  Additionaly, users may try to assign addresses
3277 		 * before the interface becomes up (or running).
3278 		 * We simply skip DAD in such a case as a work around.
3279 		 * XXX: we should rather mark "tentative" on such addresses,
3280 		 * and do DAD after the interface becomes ready.
3281 		 */
3282 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) !=
3283 		    (IFF_UP|IFF_RUNNING))
3284 			return 0;
3285 
3286 		return 1;
3287 	}
3288 }
3289 
3290 int
3291 if_flags_set(ifnet_t *ifp, const short flags)
3292 {
3293 	int rc;
3294 
3295 	if (ifp->if_setflags != NULL)
3296 		rc = (*ifp->if_setflags)(ifp, flags);
3297 	else {
3298 		short cantflags, chgdflags;
3299 		struct ifreq ifr;
3300 
3301 		chgdflags = ifp->if_flags ^ flags;
3302 		cantflags = chgdflags & IFF_CANTCHANGE;
3303 
3304 		if (cantflags != 0)
3305 			ifp->if_flags ^= cantflags;
3306 
3307                 /* Traditionally, we do not call if_ioctl after
3308                  * setting/clearing only IFF_PROMISC if the interface
3309                  * isn't IFF_UP.  Uphold that tradition.
3310 		 */
3311 		if (chgdflags == IFF_PROMISC && (ifp->if_flags & IFF_UP) == 0)
3312 			return 0;
3313 
3314 		memset(&ifr, 0, sizeof(ifr));
3315 
3316 		ifr.ifr_flags = flags & ~IFF_CANTCHANGE;
3317 		/* FIXME: may not hold if_ioctl_lock */
3318 		rc = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, &ifr);
3319 
3320 		if (rc != 0 && cantflags != 0)
3321 			ifp->if_flags ^= cantflags;
3322 	}
3323 
3324 	return rc;
3325 }
3326 
3327 int
3328 if_mcast_op(ifnet_t *ifp, const unsigned long cmd, const struct sockaddr *sa)
3329 {
3330 	int rc;
3331 	struct ifreq ifr;
3332 
3333 	if (ifp->if_mcastop != NULL)
3334 		rc = (*ifp->if_mcastop)(ifp, cmd, sa);
3335 	else {
3336 		ifreq_setaddr(cmd, &ifr, sa);
3337 		rc = (*ifp->if_ioctl)(ifp, cmd, &ifr);
3338 	}
3339 
3340 	return rc;
3341 }
3342 
3343 static void
3344 sysctl_sndq_setup(struct sysctllog **clog, const char *ifname,
3345     struct ifaltq *ifq)
3346 {
3347 	const struct sysctlnode *cnode, *rnode;
3348 
3349 	if (sysctl_createv(clog, 0, NULL, &rnode,
3350 		       CTLFLAG_PERMANENT,
3351 		       CTLTYPE_NODE, "interfaces",
3352 		       SYSCTL_DESCR("Per-interface controls"),
3353 		       NULL, 0, NULL, 0,
3354 		       CTL_NET, CTL_CREATE, CTL_EOL) != 0)
3355 		goto bad;
3356 
3357 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3358 		       CTLFLAG_PERMANENT,
3359 		       CTLTYPE_NODE, ifname,
3360 		       SYSCTL_DESCR("Interface controls"),
3361 		       NULL, 0, NULL, 0,
3362 		       CTL_CREATE, CTL_EOL) != 0)
3363 		goto bad;
3364 
3365 	if (sysctl_createv(clog, 0, &rnode, &rnode,
3366 		       CTLFLAG_PERMANENT,
3367 		       CTLTYPE_NODE, "sndq",
3368 		       SYSCTL_DESCR("Interface output queue controls"),
3369 		       NULL, 0, NULL, 0,
3370 		       CTL_CREATE, CTL_EOL) != 0)
3371 		goto bad;
3372 
3373 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3374 		       CTLFLAG_PERMANENT,
3375 		       CTLTYPE_INT, "len",
3376 		       SYSCTL_DESCR("Current output queue length"),
3377 		       NULL, 0, &ifq->ifq_len, 0,
3378 		       CTL_CREATE, CTL_EOL) != 0)
3379 		goto bad;
3380 
3381 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3382 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3383 		       CTLTYPE_INT, "maxlen",
3384 		       SYSCTL_DESCR("Maximum allowed output queue length"),
3385 		       NULL, 0, &ifq->ifq_maxlen, 0,
3386 		       CTL_CREATE, CTL_EOL) != 0)
3387 		goto bad;
3388 
3389 	if (sysctl_createv(clog, 0, &rnode, &cnode,
3390 		       CTLFLAG_PERMANENT,
3391 		       CTLTYPE_INT, "drops",
3392 		       SYSCTL_DESCR("Packets dropped due to full output queue"),
3393 		       NULL, 0, &ifq->ifq_drops, 0,
3394 		       CTL_CREATE, CTL_EOL) != 0)
3395 		goto bad;
3396 
3397 	return;
3398 bad:
3399 	printf("%s: could not attach sysctl nodes\n", ifname);
3400 	return;
3401 }
3402 
3403 #if defined(INET) || defined(INET6)
3404 
3405 #define	SYSCTL_NET_PKTQ(q, cn, c)					\
3406 	static int							\
3407 	sysctl_net_##q##_##cn(SYSCTLFN_ARGS)				\
3408 	{								\
3409 		return sysctl_pktq_count(SYSCTLFN_CALL(rnode), q, c);	\
3410 	}
3411 
3412 #if defined(INET)
3413 static int
3414 sysctl_net_ip_pktq_maxlen(SYSCTLFN_ARGS)
3415 {
3416 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip_pktq);
3417 }
3418 SYSCTL_NET_PKTQ(ip_pktq, items, PKTQ_NITEMS)
3419 SYSCTL_NET_PKTQ(ip_pktq, drops, PKTQ_DROPS)
3420 #endif
3421 
3422 #if defined(INET6)
3423 static int
3424 sysctl_net_ip6_pktq_maxlen(SYSCTLFN_ARGS)
3425 {
3426 	return sysctl_pktq_maxlen(SYSCTLFN_CALL(rnode), ip6_pktq);
3427 }
3428 SYSCTL_NET_PKTQ(ip6_pktq, items, PKTQ_NITEMS)
3429 SYSCTL_NET_PKTQ(ip6_pktq, drops, PKTQ_DROPS)
3430 #endif
3431 
3432 static void
3433 sysctl_net_pktq_setup(struct sysctllog **clog, int pf)
3434 {
3435 	sysctlfn len_func = NULL, maxlen_func = NULL, drops_func = NULL;
3436 	const char *pfname = NULL, *ipname = NULL;
3437 	int ipn = 0, qid = 0;
3438 
3439 	switch (pf) {
3440 #if defined(INET)
3441 	case PF_INET:
3442 		len_func = sysctl_net_ip_pktq_items;
3443 		maxlen_func = sysctl_net_ip_pktq_maxlen;
3444 		drops_func = sysctl_net_ip_pktq_drops;
3445 		pfname = "inet", ipn = IPPROTO_IP;
3446 		ipname = "ip", qid = IPCTL_IFQ;
3447 		break;
3448 #endif
3449 #if defined(INET6)
3450 	case PF_INET6:
3451 		len_func = sysctl_net_ip6_pktq_items;
3452 		maxlen_func = sysctl_net_ip6_pktq_maxlen;
3453 		drops_func = sysctl_net_ip6_pktq_drops;
3454 		pfname = "inet6", ipn = IPPROTO_IPV6;
3455 		ipname = "ip6", qid = IPV6CTL_IFQ;
3456 		break;
3457 #endif
3458 	default:
3459 		KASSERT(false);
3460 	}
3461 
3462 	sysctl_createv(clog, 0, NULL, NULL,
3463 		       CTLFLAG_PERMANENT,
3464 		       CTLTYPE_NODE, pfname, NULL,
3465 		       NULL, 0, NULL, 0,
3466 		       CTL_NET, pf, CTL_EOL);
3467 	sysctl_createv(clog, 0, NULL, NULL,
3468 		       CTLFLAG_PERMANENT,
3469 		       CTLTYPE_NODE, ipname, NULL,
3470 		       NULL, 0, NULL, 0,
3471 		       CTL_NET, pf, ipn, CTL_EOL);
3472 	sysctl_createv(clog, 0, NULL, NULL,
3473 		       CTLFLAG_PERMANENT,
3474 		       CTLTYPE_NODE, "ifq",
3475 		       SYSCTL_DESCR("Protocol input queue controls"),
3476 		       NULL, 0, NULL, 0,
3477 		       CTL_NET, pf, ipn, qid, CTL_EOL);
3478 
3479 	sysctl_createv(clog, 0, NULL, NULL,
3480 		       CTLFLAG_PERMANENT,
3481 		       CTLTYPE_INT, "len",
3482 		       SYSCTL_DESCR("Current input queue length"),
3483 		       len_func, 0, NULL, 0,
3484 		       CTL_NET, pf, ipn, qid, IFQCTL_LEN, CTL_EOL);
3485 	sysctl_createv(clog, 0, NULL, NULL,
3486 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
3487 		       CTLTYPE_INT, "maxlen",
3488 		       SYSCTL_DESCR("Maximum allowed input queue length"),
3489 		       maxlen_func, 0, NULL, 0,
3490 		       CTL_NET, pf, ipn, qid, IFQCTL_MAXLEN, CTL_EOL);
3491 	sysctl_createv(clog, 0, NULL, NULL,
3492 		       CTLFLAG_PERMANENT,
3493 		       CTLTYPE_INT, "drops",
3494 		       SYSCTL_DESCR("Packets dropped due to full input queue"),
3495 		       drops_func, 0, NULL, 0,
3496 		       CTL_NET, pf, ipn, qid, IFQCTL_DROPS, CTL_EOL);
3497 }
3498 #endif /* INET || INET6 */
3499 
3500 static int
3501 if_sdl_sysctl(SYSCTLFN_ARGS)
3502 {
3503 	struct ifnet *ifp;
3504 	const struct sockaddr_dl *sdl;
3505 	struct psref psref;
3506 	int error = 0;
3507 	int bound;
3508 
3509 	if (namelen != 1)
3510 		return EINVAL;
3511 
3512 	bound = curlwp_bind();
3513 	ifp = if_get_byindex(name[0], &psref);
3514 	if (ifp == NULL) {
3515 		error = ENODEV;
3516 		goto out0;
3517 	}
3518 
3519 	sdl = ifp->if_sadl;
3520 	if (sdl == NULL) {
3521 		*oldlenp = 0;
3522 		goto out1;
3523 	}
3524 
3525 	if (oldp == NULL) {
3526 		*oldlenp = sdl->sdl_alen;
3527 		goto out1;
3528 	}
3529 
3530 	if (*oldlenp >= sdl->sdl_alen)
3531 		*oldlenp = sdl->sdl_alen;
3532 	error = sysctl_copyout(l, &sdl->sdl_data[sdl->sdl_nlen], oldp, *oldlenp);
3533 out1:
3534 	if_put(ifp, &psref);
3535 out0:
3536 	curlwp_bindx(bound);
3537 	return error;
3538 }
3539 
3540 SYSCTL_SETUP(sysctl_net_sdl_setup, "sysctl net.sdl subtree setup")
3541 {
3542 	const struct sysctlnode *rnode = NULL;
3543 
3544 	sysctl_createv(clog, 0, NULL, &rnode,
3545 		       CTLFLAG_PERMANENT,
3546 		       CTLTYPE_NODE, "sdl",
3547 		       SYSCTL_DESCR("Get active link-layer address"),
3548 		       if_sdl_sysctl, 0, NULL, 0,
3549 		       CTL_NET, CTL_CREATE, CTL_EOL);
3550 }
3551