xref: /netbsd-src/sys/netinet/if_arp.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
1 /*	$NetBSD: if_arp.c,v 1.282 2019/04/29 16:12:30 roy Exp $	*/
2 
3 /*
4  * Copyright (c) 1998, 2000, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Public Access Networks Corporation ("Panix").  It was developed under
9  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1988, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)if_ether.c	8.2 (Berkeley) 9/26/94
62  */
63 
64 /*
65  * Ethernet address resolution protocol.
66  * TODO:
67  *	add "inuse/lock" bit (or ref. count) along with valid bit
68  */
69 
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1.282 2019/04/29 16:12:30 roy Exp $");
72 
73 #ifdef _KERNEL_OPT
74 #include "opt_ddb.h"
75 #include "opt_inet.h"
76 #include "opt_net_mpsafe.h"
77 #endif
78 
79 #ifdef INET
80 
81 #include "arp.h"
82 #include "bridge.h"
83 
84 #include <sys/param.h>
85 #include <sys/systm.h>
86 #include <sys/callout.h>
87 #include <sys/kmem.h>
88 #include <sys/mbuf.h>
89 #include <sys/socket.h>
90 #include <sys/time.h>
91 #include <sys/timetc.h>
92 #include <sys/kernel.h>
93 #include <sys/errno.h>
94 #include <sys/ioctl.h>
95 #include <sys/syslog.h>
96 #include <sys/proc.h>
97 #include <sys/protosw.h>
98 #include <sys/domain.h>
99 #include <sys/sysctl.h>
100 #include <sys/socketvar.h>
101 #include <sys/percpu.h>
102 #include <sys/cprng.h>
103 #include <sys/kmem.h>
104 
105 #include <net/ethertypes.h>
106 #include <net/if.h>
107 #include <net/if_dl.h>
108 #include <net/if_token.h>
109 #include <net/if_types.h>
110 #include <net/if_ether.h>
111 #include <net/if_llatbl.h>
112 #include <net/route.h>
113 #include <net/net_stats.h>
114 
115 #include <netinet/in.h>
116 #include <netinet/in_systm.h>
117 #include <netinet/in_var.h>
118 #include <netinet/ip.h>
119 #include <netinet/if_inarp.h>
120 
121 #include "arcnet.h"
122 #if NARCNET > 0
123 #include <net/if_arc.h>
124 #endif
125 #include "fddi.h"
126 #if NFDDI > 0
127 #include <net/if_fddi.h>
128 #endif
129 #include "token.h"
130 #include "carp.h"
131 #if NCARP > 0
132 #include <netinet/ip_carp.h>
133 #endif
134 
135 /*
136  * ARP trailer negotiation.  Trailer protocol is not IP specific,
137  * but ARP request/response use IP addresses.
138  */
139 #define ETHERTYPE_IPTRAILERS ETHERTYPE_TRAIL
140 
141 /* timer values */
142 static int arpt_keep = (20*60);	/* once resolved, good for 20 more minutes */
143 static int arpt_down = 20;		/* once declared down, don't send for 20 secs */
144 static int arp_maxhold = 1;	/* number of packets to hold per ARP entry */
145 #define	rt_expire rt_rmx.rmx_expire
146 #define	rt_pksent rt_rmx.rmx_pksent
147 
148 int ip_dad_count = PROBE_NUM;
149 #ifdef ARP_DEBUG
150 int arp_debug = 1;
151 #else
152 int arp_debug = 0;
153 #endif
154 
155 static void arp_init(void);
156 
157 static void arprequest(struct ifnet *,
158     const struct in_addr *, const struct in_addr *,
159     const uint8_t *);
160 static void arpannounce1(struct ifaddr *);
161 static struct sockaddr *arp_setgate(struct rtentry *, struct sockaddr *,
162     const struct sockaddr *);
163 static void arptimer(void *);
164 static void arp_settimer(struct llentry *, int);
165 static struct llentry *arplookup(struct ifnet *,
166     const struct in_addr *, const struct sockaddr *, int);
167 static struct llentry *arpcreate(struct ifnet *,
168     const struct in_addr *, const struct sockaddr *, int);
169 static void in_arpinput(struct mbuf *);
170 static void in_revarpinput(struct mbuf *);
171 static void revarprequest(struct ifnet *);
172 
173 static void arp_drainstub(void);
174 
175 struct dadq;
176 static void arp_dad_timer(struct dadq *);
177 static void arp_dad_start(struct ifaddr *);
178 static void arp_dad_stop(struct ifaddr *);
179 static void arp_dad_duplicated(struct ifaddr *, const struct sockaddr_dl *);
180 
181 static void arp_init_llentry(struct ifnet *, struct llentry *);
182 #if NTOKEN > 0
183 static void arp_free_llentry_tokenring(struct llentry *);
184 #endif
185 
186 struct ifqueue arpintrq = {
187 	.ifq_head = NULL,
188 	.ifq_tail = NULL,
189 	.ifq_len = 0,
190 	.ifq_maxlen = 50,
191 	.ifq_drops = 0,
192 };
193 static int arp_maxtries = 5;
194 static int useloopback = 1;	/* use loopback interface for local traffic */
195 
196 static percpu_t *arpstat_percpu;
197 
198 #define	ARP_STAT_GETREF()	_NET_STAT_GETREF(arpstat_percpu)
199 #define	ARP_STAT_PUTREF()	_NET_STAT_PUTREF(arpstat_percpu)
200 
201 #define	ARP_STATINC(x)		_NET_STATINC(arpstat_percpu, x)
202 #define	ARP_STATADD(x, v)	_NET_STATADD(arpstat_percpu, x, v)
203 
204 /* revarp state */
205 static struct in_addr myip, srv_ip;
206 static int myip_initialized = 0;
207 static int revarp_in_progress = 0;
208 static struct ifnet *myip_ifp = NULL;
209 
210 static int arp_drainwanted;
211 
212 static int log_movements = 1;
213 static int log_permanent_modify = 1;
214 static int log_wrong_iface = 1;
215 static int log_unknown_network = 1;
216 
217 DOMAIN_DEFINE(arpdomain);	/* forward declare and add to link set */
218 
219 static void
220 arp_fasttimo(void)
221 {
222 	if (arp_drainwanted) {
223 		arp_drain();
224 		arp_drainwanted = 0;
225 	}
226 }
227 
228 static const struct protosw arpsw[] = {
229 	{
230 		.pr_type = 0,
231 		.pr_domain = &arpdomain,
232 		.pr_protocol = 0,
233 		.pr_flags = 0,
234 		.pr_input = 0,
235 		.pr_ctlinput = 0,
236 		.pr_ctloutput = 0,
237 		.pr_usrreqs = 0,
238 		.pr_init = arp_init,
239 		.pr_fasttimo = arp_fasttimo,
240 		.pr_slowtimo = 0,
241 		.pr_drain = arp_drainstub,
242 	}
243 };
244 
245 struct domain arpdomain = {
246 	.dom_family = PF_ARP,
247 	.dom_name = "arp",
248 	.dom_protosw = arpsw,
249 	.dom_protoswNPROTOSW = &arpsw[__arraycount(arpsw)],
250 };
251 
252 static void sysctl_net_inet_arp_setup(struct sysctllog **);
253 
254 void
255 arp_init(void)
256 {
257 
258 	sysctl_net_inet_arp_setup(NULL);
259 	arpstat_percpu = percpu_alloc(sizeof(uint64_t) * ARP_NSTATS);
260 	IFQ_LOCK_INIT(&arpintrq);
261 }
262 
263 static void
264 arp_drainstub(void)
265 {
266 	arp_drainwanted = 1;
267 }
268 
269 /*
270  * ARP protocol drain routine.  Called when memory is in short supply.
271  * Called at splvm();  don't acquire softnet_lock as can be called from
272  * hardware interrupt handlers.
273  */
274 void
275 arp_drain(void)
276 {
277 
278 	lltable_drain(AF_INET);
279 }
280 
281 static void
282 arptimer(void *arg)
283 {
284 	struct llentry *lle = arg;
285 	struct ifnet *ifp;
286 
287 	KASSERT((lle->la_flags & LLE_STATIC) == 0);
288 
289 	LLE_WLOCK(lle);
290 
291 	/*
292 	 * This shortcut is required to avoid trying to touch ifp that may be
293 	 * being destroyed.
294 	 */
295 	if ((lle->la_flags & LLE_LINKED) == 0) {
296 		LLE_FREE_LOCKED(lle);
297 		return;
298 	}
299 
300 	ifp = lle->lle_tbl->llt_ifp;
301 
302 	/* XXX: LOR avoidance. We still have ref on lle. */
303 	LLE_WUNLOCK(lle);
304 
305 	IF_AFDATA_LOCK(ifp);
306 	LLE_WLOCK(lle);
307 
308 	/* Guard against race with other llentry_free(). */
309 	if (lle->la_flags & LLE_LINKED) {
310 		size_t pkts_dropped;
311 
312 		LLE_REMREF(lle);
313 		pkts_dropped = llentry_free(lle);
314 		ARP_STATADD(ARP_STAT_DFRDROPPED, pkts_dropped);
315 		ARP_STATADD(ARP_STAT_DFRTOTAL, pkts_dropped);
316 	} else {
317 		LLE_FREE_LOCKED(lle);
318 	}
319 
320 	IF_AFDATA_UNLOCK(ifp);
321 }
322 
323 static void
324 arp_settimer(struct llentry *la, int sec)
325 {
326 
327 	LLE_WLOCK_ASSERT(la);
328 	KASSERT((la->la_flags & LLE_STATIC) == 0);
329 
330 	/*
331 	 * We have to take care of a reference leak which occurs if
332 	 * callout_reset overwrites a pending callout schedule.  Unfortunately
333 	 * we don't have a mean to know the overwrite, so we need to know it
334 	 * using callout_stop.  We need to call callout_pending first to exclude
335 	 * the case that the callout has never been scheduled.
336 	 */
337 	if (callout_pending(&la->la_timer)) {
338 		bool expired = callout_stop(&la->la_timer);
339 		if (!expired)
340 			/* A pending callout schedule is canceled. */
341 			LLE_REMREF(la);
342 	}
343 	LLE_ADDREF(la);
344 	callout_reset(&la->la_timer, hz * sec, arptimer, la);
345 }
346 
347 /*
348  * We set the gateway for RTF_CLONING routes to a "prototype"
349  * link-layer sockaddr whose interface type (if_type) and interface
350  * index (if_index) fields are prepared.
351  */
352 static struct sockaddr *
353 arp_setgate(struct rtentry *rt, struct sockaddr *gate,
354     const struct sockaddr *netmask)
355 {
356 	const struct ifnet *ifp = rt->rt_ifp;
357 	uint8_t namelen = strlen(ifp->if_xname);
358 	uint8_t addrlen = ifp->if_addrlen;
359 
360 	/*
361 	 * XXX: If this is a manually added route to interface
362 	 * such as older version of routed or gated might provide,
363 	 * restore cloning bit.
364 	 */
365 	if ((rt->rt_flags & RTF_HOST) == 0 && netmask != NULL &&
366 	    satocsin(netmask)->sin_addr.s_addr != 0xffffffff)
367 		rt->rt_flags |= RTF_CONNECTED;
368 
369 	if ((rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL))) {
370 		union {
371 			struct sockaddr sa;
372 			struct sockaddr_storage ss;
373 			struct sockaddr_dl sdl;
374 		} u;
375 		/*
376 		 * Case 1: This route should come from a route to iface.
377 		 */
378 		sockaddr_dl_init(&u.sdl, sizeof(u.ss),
379 		    ifp->if_index, ifp->if_type, NULL, namelen, NULL, addrlen);
380 		rt_setgate(rt, &u.sa);
381 		gate = rt->rt_gateway;
382 	}
383 	return gate;
384 }
385 
386 static void
387 arp_init_llentry(struct ifnet *ifp, struct llentry *lle)
388 {
389 
390 	switch (ifp->if_type) {
391 #if NTOKEN > 0
392 	case IFT_ISO88025:
393 		lle->la_opaque = kmem_intr_alloc(sizeof(struct token_rif),
394 		    KM_NOSLEEP);
395 		lle->lle_ll_free = arp_free_llentry_tokenring;
396 		break;
397 #endif
398 	}
399 }
400 
401 #if NTOKEN > 0
402 static void
403 arp_free_llentry_tokenring(struct llentry *lle)
404 {
405 
406 	kmem_intr_free(lle->la_opaque, sizeof(struct token_rif));
407 }
408 #endif
409 
410 /*
411  * Parallel to llc_rtrequest.
412  */
413 void
414 arp_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
415 {
416 	struct sockaddr *gate = rt->rt_gateway;
417 	struct in_ifaddr *ia;
418 	struct ifaddr *ifa;
419 	struct ifnet *ifp = rt->rt_ifp;
420 	int bound;
421 	int s;
422 
423 	if (req == RTM_LLINFO_UPD) {
424 		if ((ifa = info->rti_ifa) != NULL)
425 			arpannounce1(ifa);
426 		return;
427 	}
428 
429 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
430 		if (req != RTM_ADD)
431 			return;
432 
433 		/*
434 		 * linklayers with particular link MTU limitation.
435 		 */
436 		switch(ifp->if_type) {
437 #if NFDDI > 0
438 		case IFT_FDDI:
439 			if (ifp->if_mtu > FDDIIPMTU)
440 				rt->rt_rmx.rmx_mtu = FDDIIPMTU;
441 			break;
442 #endif
443 #if NARCNET > 0
444 		case IFT_ARCNET:
445 		    {
446 			int arcipifmtu;
447 
448 			if (ifp->if_flags & IFF_LINK0)
449 				arcipifmtu = arc_ipmtu;
450 			else
451 				arcipifmtu = ARCMTU;
452 			if (ifp->if_mtu > arcipifmtu)
453 				rt->rt_rmx.rmx_mtu = arcipifmtu;
454 			break;
455 		    }
456 #endif
457 		}
458 		return;
459 	}
460 
461 	switch (req) {
462 	case RTM_SETGATE:
463 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
464 		break;
465 	case RTM_ADD:
466 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
467 		if (gate == NULL) {
468 			log(LOG_ERR, "%s: arp_setgate failed\n", __func__);
469 			break;
470 		}
471 		if ((rt->rt_flags & RTF_CONNECTED) ||
472 		    (rt->rt_flags & RTF_LOCAL)) {
473 			/*
474 			 * Give this route an expiration time, even though
475 			 * it's a "permanent" route, so that routes cloned
476 			 * from it do not need their expiration time set.
477 			 */
478 			KASSERT(time_uptime != 0);
479 			rt->rt_expire = time_uptime;
480 			/*
481 			 * linklayers with particular link MTU limitation.
482 			 */
483 			switch (ifp->if_type) {
484 #if NFDDI > 0
485 			case IFT_FDDI:
486 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
487 				    (rt->rt_rmx.rmx_mtu > FDDIIPMTU ||
488 				     (rt->rt_rmx.rmx_mtu == 0 &&
489 				      ifp->if_mtu > FDDIIPMTU)))
490 					rt->rt_rmx.rmx_mtu = FDDIIPMTU;
491 				break;
492 #endif
493 #if NARCNET > 0
494 			case IFT_ARCNET:
495 			    {
496 				int arcipifmtu;
497 				if (ifp->if_flags & IFF_LINK0)
498 					arcipifmtu = arc_ipmtu;
499 				else
500 					arcipifmtu = ARCMTU;
501 
502 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
503 				    (rt->rt_rmx.rmx_mtu > arcipifmtu ||
504 				     (rt->rt_rmx.rmx_mtu == 0 &&
505 				      ifp->if_mtu > arcipifmtu)))
506 					rt->rt_rmx.rmx_mtu = arcipifmtu;
507 				break;
508 			    }
509 #endif
510 			}
511 			if (rt->rt_flags & RTF_CONNECTED)
512 				break;
513 		}
514 
515 		bound = curlwp_bind();
516 		/* Announce a new entry if requested. */
517 		if (rt->rt_flags & RTF_ANNOUNCE) {
518 			struct psref psref;
519 			ia = in_get_ia_on_iface_psref(
520 			    satocsin(rt_getkey(rt))->sin_addr, ifp, &psref);
521 			if (ia != NULL) {
522 				arpannounce(ifp, &ia->ia_ifa,
523 				    CLLADDR(satocsdl(gate)));
524 				ia4_release(ia, &psref);
525 			}
526 		}
527 
528 		if (gate->sa_family != AF_LINK ||
529 		    gate->sa_len < sockaddr_dl_measure(0, ifp->if_addrlen)) {
530 			log(LOG_DEBUG, "%s: bad gateway value\n", __func__);
531 			goto out;
532 		}
533 
534 		satosdl(gate)->sdl_type = ifp->if_type;
535 		satosdl(gate)->sdl_index = ifp->if_index;
536 
537 		/*
538 		 * If the route is for a broadcast address mark it as such.
539 		 * This way we can avoid an expensive call to in_broadcast()
540 		 * in ip_output() most of the time (because the route passed
541 		 * to ip_output() is almost always a host route).
542 		 */
543 		if (rt->rt_flags & RTF_HOST &&
544 		    !(rt->rt_flags & RTF_BROADCAST) &&
545 		    in_broadcast(satocsin(rt_getkey(rt))->sin_addr, rt->rt_ifp))
546 			rt->rt_flags |= RTF_BROADCAST;
547 		/* There is little point in resolving the broadcast address */
548 		if (rt->rt_flags & RTF_BROADCAST)
549 			goto out;
550 
551 		/*
552 		 * When called from rt_ifa_addlocal, we cannot depend on that
553 		 * the address (rt_getkey(rt)) exits in the address list of the
554 		 * interface. So check RTF_LOCAL instead.
555 		 */
556 		if (rt->rt_flags & RTF_LOCAL) {
557 			rt->rt_expire = 0;
558 			if (useloopback) {
559 				rt->rt_ifp = lo0ifp;
560 				rt->rt_rmx.rmx_mtu = 0;
561 			}
562 			goto out;
563 		}
564 
565 		s = pserialize_read_enter();
566 		ia = in_get_ia_on_iface(satocsin(rt_getkey(rt))->sin_addr, ifp);
567 		if (ia == NULL) {
568 			pserialize_read_exit(s);
569 			goto out;
570 		}
571 
572 		rt->rt_expire = 0;
573 		if (useloopback) {
574 			rt->rt_ifp = lo0ifp;
575 			rt->rt_rmx.rmx_mtu = 0;
576 		}
577 		rt->rt_flags |= RTF_LOCAL;
578 
579 		if (ISSET(info->rti_flags, RTF_DONTCHANGEIFA)) {
580 			pserialize_read_exit(s);
581 			goto out;
582 		}
583 		/*
584 		 * make sure to set rt->rt_ifa to the interface
585 		 * address we are using, otherwise we will have trouble
586 		 * with source address selection.
587 		 */
588 		ifa = &ia->ia_ifa;
589 		if (ifa != rt->rt_ifa)
590 			/* Assume it doesn't sleep */
591 			rt_replace_ifa(rt, ifa);
592 		pserialize_read_exit(s);
593 	out:
594 		curlwp_bindx(bound);
595 		break;
596 	}
597 }
598 
599 /*
600  * Broadcast an ARP request. Caller specifies:
601  *	- arp header source ip address
602  *	- arp header target ip address
603  *	- arp header source ethernet address
604  */
605 static void
606 arprequest(struct ifnet *ifp,
607     const struct in_addr *sip, const struct in_addr *tip,
608     const uint8_t *enaddr)
609 {
610 	struct mbuf *m;
611 	struct arphdr *ah;
612 	struct sockaddr sa;
613 	uint64_t *arps;
614 
615 	KASSERT(sip != NULL);
616 	KASSERT(tip != NULL);
617 	KASSERT(enaddr != NULL);
618 
619 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
620 		return;
621 	MCLAIM(m, &arpdomain.dom_mowner);
622 	switch (ifp->if_type) {
623 	case IFT_IEEE1394:
624 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
625 		    ifp->if_addrlen;
626 		break;
627 	default:
628 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
629 		    2 * ifp->if_addrlen;
630 		break;
631 	}
632 	m->m_pkthdr.len = m->m_len;
633 	m_align(m, m->m_len);
634 	ah = mtod(m, struct arphdr *);
635 	memset(ah, 0, m->m_len);
636 	switch (ifp->if_type) {
637 	case IFT_IEEE1394:	/* RFC2734 */
638 		/* fill it now for ar_tpa computation */
639 		ah->ar_hrd = htons(ARPHRD_IEEE1394);
640 		break;
641 	default:
642 		/* ifp->if_output will fill ar_hrd */
643 		break;
644 	}
645 	ah->ar_pro = htons(ETHERTYPE_IP);
646 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
647 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
648 	ah->ar_op = htons(ARPOP_REQUEST);
649 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
650 	memcpy(ar_spa(ah), sip, ah->ar_pln);
651 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
652 	sa.sa_family = AF_ARP;
653 	sa.sa_len = 2;
654 	m->m_flags |= M_BCAST;
655 	arps = ARP_STAT_GETREF();
656 	arps[ARP_STAT_SNDTOTAL]++;
657 	arps[ARP_STAT_SENDREQUEST]++;
658 	ARP_STAT_PUTREF();
659 	if_output_lock(ifp, ifp, m, &sa, NULL);
660 }
661 
662 void
663 arpannounce(struct ifnet *ifp, struct ifaddr *ifa, const uint8_t *enaddr)
664 {
665 	struct in_ifaddr *ia = ifatoia(ifa);
666 	struct in_addr *ip = &IA_SIN(ifa)->sin_addr;
667 
668 	if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED)) {
669 		ARPLOG(LOG_DEBUG, "%s not ready\n", ARPLOGADDR(ip));
670 		return;
671 	}
672 	arprequest(ifp, ip, ip, enaddr);
673 }
674 
675 static void
676 arpannounce1(struct ifaddr *ifa)
677 {
678 
679 	arpannounce(ifa->ifa_ifp, ifa, CLLADDR(ifa->ifa_ifp->if_sadl));
680 }
681 
682 /*
683  * Resolve an IP address into an ethernet address.  If success, desten is
684  * filled in. If there is no entry in arptab, set one up and broadcast a
685  * request for the IP address. Hold onto this mbuf and resend it once the
686  * address is finally resolved.
687  *
688  * A return value of 0 indicates that desten has been filled in and the packet
689  * should be sent normally; a return value of EWOULDBLOCK indicates that the
690  * packet has been held pending resolution. Any other value indicates an
691  * error.
692  */
693 int
694 arpresolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
695     const struct sockaddr *dst, void *desten, size_t destlen)
696 {
697 	struct llentry *la;
698 	const char *create_lookup;
699 	bool renew;
700 	int error;
701 	struct ifnet *origifp = ifp;
702 
703 #if NCARP > 0
704 	if (rt != NULL && rt->rt_ifp->if_type == IFT_CARP)
705 		ifp = rt->rt_ifp;
706 #endif
707 
708 	KASSERT(m != NULL);
709 
710 	la = arplookup(ifp, NULL, dst, 0);
711 	if (la == NULL)
712 		goto notfound;
713 
714 	if ((la->la_flags & LLE_VALID) &&
715 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
716 		KASSERT(destlen >= ifp->if_addrlen);
717 		memcpy(desten, &la->ll_addr, ifp->if_addrlen);
718 		LLE_RUNLOCK(la);
719 		return 0;
720 	}
721 
722 notfound:
723 	if (ifp->if_flags & IFF_NOARP) {
724 		if (la != NULL)
725 			LLE_RUNLOCK(la);
726 		error = ENOTSUP;
727 		goto bad;
728 	}
729 
730 	if (la == NULL) {
731 		struct rtentry *_rt;
732 
733 		create_lookup = "create";
734 		_rt = rtalloc1(dst, 0);
735 		IF_AFDATA_WLOCK(ifp);
736 		la = lla_create(LLTABLE(ifp), LLE_EXCLUSIVE, dst, _rt);
737 		IF_AFDATA_WUNLOCK(ifp);
738 		if (_rt != NULL)
739 			rt_unref(_rt);
740 		if (la == NULL)
741 			ARP_STATINC(ARP_STAT_ALLOCFAIL);
742 		else {
743 			struct sockaddr_in sin;
744 
745 			arp_init_llentry(ifp, la);
746 			sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
747 			rt_clonedmsg(sintosa(&sin), ifp, rt);
748 		}
749 	} else if (LLE_TRY_UPGRADE(la) == 0) {
750 		create_lookup = "lookup";
751 		LLE_RUNLOCK(la);
752 		IF_AFDATA_RLOCK(ifp);
753 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
754 		IF_AFDATA_RUNLOCK(ifp);
755 	}
756 
757 	error = EINVAL;
758 	if (la == NULL) {
759 		log(LOG_DEBUG,
760 		    "%s: failed to %s llentry for %s on %s\n",
761 		    __func__, create_lookup, inet_ntoa(satocsin(dst)->sin_addr),
762 		    ifp->if_xname);
763 		goto bad;
764 	}
765 
766 	if ((la->la_flags & LLE_VALID) &&
767 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime))
768 	{
769 		KASSERT(destlen >= ifp->if_addrlen);
770 		memcpy(desten, &la->ll_addr, ifp->if_addrlen);
771 		renew = false;
772 		/*
773 		 * If entry has an expiry time and it is approaching,
774 		 * see if we need to send an ARP request within this
775 		 * arpt_down interval.
776 		 */
777 		if (!(la->la_flags & LLE_STATIC) &&
778 		    time_uptime + la->la_preempt > la->la_expire)
779 		{
780 			renew = true;
781 			la->la_preempt--;
782 		}
783 
784 		LLE_WUNLOCK(la);
785 
786 		if (renew) {
787 			const uint8_t *enaddr = CLLADDR(ifp->if_sadl);
788 			arprequest(origifp,
789 			    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
790 			    &satocsin(dst)->sin_addr, enaddr);
791 		}
792 
793 		return 0;
794 	}
795 
796 	if (la->la_flags & LLE_STATIC) {   /* should not happen! */
797 		LLE_RUNLOCK(la);
798 		log(LOG_DEBUG, "%s: ouch, empty static llinfo for %s\n",
799 		    __func__, inet_ntoa(satocsin(dst)->sin_addr));
800 		error = EINVAL;
801 		goto bad;
802 	}
803 
804 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
805 
806 	/*
807 	 * There is an arptab entry, but no ethernet address
808 	 * response yet.  Add the mbuf to the list, dropping
809 	 * the oldest packet if we have exceeded the system
810 	 * setting.
811 	 */
812 	LLE_WLOCK_ASSERT(la);
813 	if (la->la_numheld >= arp_maxhold) {
814 		if (la->la_hold != NULL) {
815 			struct mbuf *next = la->la_hold->m_nextpkt;
816 			m_freem(la->la_hold);
817 			la->la_hold = next;
818 			la->la_numheld--;
819 			ARP_STATINC(ARP_STAT_DFRDROPPED);
820 			ARP_STATINC(ARP_STAT_DFRTOTAL);
821 		}
822 	}
823 	if (la->la_hold != NULL) {
824 		struct mbuf *curr = la->la_hold;
825 		while (curr->m_nextpkt != NULL)
826 			curr = curr->m_nextpkt;
827 		curr->m_nextpkt = m;
828 	} else
829 		la->la_hold = m;
830 	la->la_numheld++;
831 	if (!renew)
832 		LLE_DOWNGRADE(la);
833 
834 	/*
835 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
836 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
837 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
838 	 * ARP request, but not faster than one request per second.
839 	 */
840 	if (la->la_asked < arp_maxtries)
841 		error = EWOULDBLOCK;	/* First request. */
842 	else
843 		error = (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
844 		    EHOSTUNREACH : EHOSTDOWN;
845 
846 	if (renew) {
847 		const uint8_t *enaddr = CLLADDR(ifp->if_sadl);
848 		la->la_expire = time_uptime;
849 		arp_settimer(la, arpt_down);
850 		la->la_asked++;
851 		LLE_WUNLOCK(la);
852 
853 		if (rt != NULL) {
854 			arprequest(origifp,
855 			    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
856 			    &satocsin(dst)->sin_addr, enaddr);
857 		} else {
858 			struct sockaddr_in sin;
859 			struct rtentry *_rt;
860 
861 			sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
862 
863 			/* XXX */
864 			_rt = rtalloc1((struct sockaddr *)&sin, 0);
865 			if (_rt == NULL)
866 				goto bad;
867 			arprequest(origifp,
868 			    &satocsin(_rt->rt_ifa->ifa_addr)->sin_addr,
869 			    &satocsin(dst)->sin_addr, enaddr);
870 			rt_unref(_rt);
871 		}
872 		return error;
873 	}
874 
875 	LLE_RUNLOCK(la);
876 	return error;
877 
878 bad:
879 	m_freem(m);
880 	return error;
881 }
882 
883 /*
884  * Common length and type checks are done here,
885  * then the protocol-specific routine is called.
886  */
887 void
888 arpintr(void)
889 {
890 	struct mbuf *m;
891 	struct arphdr *ar;
892 	int s;
893 	int arplen;
894 
895 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
896 	for (;;) {
897 		struct ifnet *rcvif;
898 
899 		IFQ_LOCK(&arpintrq);
900 		IF_DEQUEUE(&arpintrq, m);
901 		IFQ_UNLOCK(&arpintrq);
902 		if (m == NULL)
903 			goto out;
904 		if ((m->m_flags & M_PKTHDR) == 0)
905 			panic("arpintr");
906 
907 		MCLAIM(m, &arpdomain.dom_mowner);
908 		ARP_STATINC(ARP_STAT_RCVTOTAL);
909 
910 		arplen = sizeof(struct arphdr);
911 		if (m->m_len < arplen && (m = m_pullup(m, arplen)) == NULL)
912 			goto badlen;
913 		ar = mtod(m, struct arphdr *);
914 
915 		rcvif = m_get_rcvif(m, &s);
916 		if (__predict_false(rcvif == NULL)) {
917 			ARP_STATINC(ARP_STAT_RCVNOINT);
918 			goto free;
919 		}
920 
921 		/*
922 		 * We don't want non-IEEE1394 ARP packets on IEEE1394
923 		 * interfaces, and vice versa. Our life depends on that.
924 		 */
925 		switch (rcvif->if_type) {
926 		case IFT_IEEE1394:
927 			if (ntohs(ar->ar_hrd) != ARPHRD_IEEE1394) {
928 				m_put_rcvif(rcvif, &s);
929 				ARP_STATINC(ARP_STAT_RCVBADPROTO);
930 				goto free;
931 			}
932 
933 			arplen = sizeof(struct arphdr) +
934 			    ar->ar_hln + 2 * ar->ar_pln;
935 			break;
936 		default:
937 			if (ntohs(ar->ar_hrd) == ARPHRD_IEEE1394) {
938 				m_put_rcvif(rcvif, &s);
939 				ARP_STATINC(ARP_STAT_RCVBADPROTO);
940 				goto free;
941 			}
942 
943 			arplen = sizeof(struct arphdr) +
944 			    2 * ar->ar_hln + 2 * ar->ar_pln;
945 			break;
946 		}
947 
948 		m_put_rcvif(rcvif, &s);
949 
950 		if (m->m_len < arplen && (m = m_pullup(m, arplen)) == NULL)
951 			goto badlen;
952 		ar = mtod(m, struct arphdr *);
953 
954 		switch (ntohs(ar->ar_pro)) {
955 		case ETHERTYPE_IP:
956 		case ETHERTYPE_IPTRAILERS:
957 			in_arpinput(m);
958 			continue;
959 		default:
960 			ARP_STATINC(ARP_STAT_RCVBADPROTO);
961 			goto free;
962 		}
963 
964 badlen:
965 		ARP_STATINC(ARP_STAT_RCVBADLEN);
966 free:
967 		m_freem(m);
968 	}
969 
970 out:
971 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
972 	return; /* XXX gcc */
973 }
974 
975 /*
976  * ARP for Internet protocols on 10 Mb/s Ethernet. Algorithm is that given in
977  * RFC 826. In addition, a sanity check is performed on the sender protocol
978  * address, to catch impersonators.
979  *
980  * We no longer handle negotiations for use of trailer protocol: formerly, ARP
981  * replied for protocol type ETHERTYPE_TRAIL sent along with IP replies if we
982  * wanted trailers sent to us, and also sent them in response to IP replies.
983  * This allowed either end to announce the desire to receive trailer packets.
984  *
985  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either, but
986  * formerly didn't normally send requests.
987  */
988 static void
989 in_arpinput(struct mbuf *m)
990 {
991 	struct arphdr *ah;
992 	struct ifnet *ifp, *rcvif = NULL;
993 	struct llentry *la = NULL;
994 	struct in_ifaddr *ia = NULL;
995 #if NBRIDGE > 0
996 	struct in_ifaddr *bridge_ia = NULL;
997 #endif
998 #if NCARP > 0
999 	uint32_t count = 0, index = 0;
1000 #endif
1001 	struct sockaddr sa;
1002 	struct in_addr isaddr, itaddr, myaddr;
1003 	int op;
1004 	void *tha;
1005 	uint64_t *arps;
1006 	struct psref psref, psref_ia;
1007 	int s;
1008 	char ipbuf[INET_ADDRSTRLEN];
1009 	bool do_dad;
1010 
1011 	if (__predict_false(m_makewritable(&m, 0, m->m_pkthdr.len, M_DONTWAIT)))
1012 		goto out;
1013 	ah = mtod(m, struct arphdr *);
1014 	op = ntohs(ah->ar_op);
1015 
1016 	if (ah->ar_pln != sizeof(struct in_addr))
1017 		goto out;
1018 
1019 	ifp = if_get_bylla(ar_sha(ah), ah->ar_hln, &psref);
1020 	if (ifp) {
1021 		/* it's from me, ignore it. */
1022 		if_put(ifp, &psref);
1023 		ARP_STATINC(ARP_STAT_RCVLOCALSHA);
1024 		goto out;
1025 	}
1026 
1027 	rcvif = ifp = m_get_rcvif_psref(m, &psref);
1028 	if (__predict_false(rcvif == NULL))
1029 		goto out;
1030 	if (rcvif->if_flags & IFF_NOARP)
1031 		goto out;
1032 
1033 	memcpy(&isaddr, ar_spa(ah), sizeof(isaddr));
1034 	memcpy(&itaddr, ar_tpa(ah), sizeof(itaddr));
1035 
1036 	if (m->m_flags & (M_BCAST|M_MCAST))
1037 		ARP_STATINC(ARP_STAT_RCVMCAST);
1038 
1039 	/*
1040 	 * Search for a matching interface address
1041 	 * or any address on the interface to use
1042 	 * as a dummy address in the rest of this function.
1043 	 *
1044 	 * If the target IP address is zero then try and find
1045 	 * the sender address for DAD.
1046 	 */
1047 	myaddr = in_nullhost(itaddr) ? isaddr : itaddr;
1048 	s = pserialize_read_enter();
1049 	IN_ADDRHASH_READER_FOREACH(ia, myaddr.s_addr) {
1050 		if (!in_hosteq(ia->ia_addr.sin_addr, myaddr))
1051 			continue;
1052 #if NCARP > 0
1053 		if (ia->ia_ifp->if_type == IFT_CARP &&
1054 		    ((ia->ia_ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
1055 		    (IFF_UP|IFF_RUNNING))) {
1056 			index++;
1057 			/* XXX: ar_hln? */
1058 			if (ia->ia_ifp == rcvif && (ah->ar_hln >= 6) &&
1059 			    carp_iamatch(ia, ar_sha(ah),
1060 			    &count, index)) {
1061 				break;
1062 			}
1063 		} else
1064 #endif
1065 		if (ia->ia_ifp == rcvif)
1066 			break;
1067 #if NBRIDGE > 0
1068 		/*
1069 		 * If the interface we received the packet on
1070 		 * is part of a bridge, check to see if we need
1071 		 * to "bridge" the packet to ourselves at this
1072 		 * layer.  Note we still prefer a perfect match,
1073 		 * but allow this weaker match if necessary.
1074 		 */
1075 		if (rcvif->if_bridge != NULL &&
1076 		    rcvif->if_bridge == ia->ia_ifp->if_bridge)
1077 			bridge_ia = ia;
1078 #endif
1079 	}
1080 
1081 #if NBRIDGE > 0
1082 	if (ia == NULL && bridge_ia != NULL) {
1083 		ia = bridge_ia;
1084 		m_put_rcvif_psref(rcvif, &psref);
1085 		rcvif = NULL;
1086 		/* FIXME */
1087 		ifp = bridge_ia->ia_ifp;
1088 	}
1089 #endif
1090 	if (ia != NULL)
1091 		ia4_acquire(ia, &psref_ia);
1092 	pserialize_read_exit(s);
1093 
1094 	if (ah->ar_hln != ifp->if_addrlen) {
1095 		ARP_STATINC(ARP_STAT_RCVBADLEN);
1096 		log(LOG_WARNING,
1097 		    "arp from %s: addr len: new %d, i/f %d (ignored)\n",
1098 		    IN_PRINT(ipbuf, &isaddr), ah->ar_hln, ifp->if_addrlen);
1099 		goto out;
1100 	}
1101 
1102 	/* Only do DaD if we have a matching address. */
1103 	do_dad = (ia != NULL);
1104 
1105 	if (ia == NULL) {
1106 		ia = in_get_ia_on_iface_psref(isaddr, rcvif, &psref_ia);
1107 		if (ia == NULL) {
1108 			ia = in_get_ia_from_ifp_psref(ifp, &psref_ia);
1109 			if (ia == NULL) {
1110 				ARP_STATINC(ARP_STAT_RCVNOINT);
1111 				goto out;
1112 			}
1113 		}
1114 	}
1115 
1116 	myaddr = ia->ia_addr.sin_addr;
1117 
1118 	/* XXX checks for bridge case? */
1119 	if (!memcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1120 		ARP_STATINC(ARP_STAT_RCVBCASTSHA);
1121 		log(LOG_ERR,
1122 		    "%s: arp: link address is broadcast for IP address %s!\n",
1123 		    ifp->if_xname, IN_PRINT(ipbuf, &isaddr));
1124 		goto out;
1125 	}
1126 
1127 	/*
1128 	 * If the source IP address is zero, this is an RFC 5227 ARP probe
1129 	 */
1130 	if (in_nullhost(isaddr))
1131 		ARP_STATINC(ARP_STAT_RCVZEROSPA);
1132 	else if (in_hosteq(isaddr, myaddr))
1133 		ARP_STATINC(ARP_STAT_RCVLOCALSPA);
1134 
1135 	if (in_nullhost(itaddr))
1136 		ARP_STATINC(ARP_STAT_RCVZEROTPA);
1137 
1138 	/*
1139 	 * DAD check, RFC 5227.
1140 	 * Collision on sender address is always a duplicate.
1141 	 * Collision on target address is only a duplicate IF
1142 	 * the sender address is the null host (ie a DAD probe) AND
1143 	 * the message was broadcast - if it's unicast then it's
1144 	 * a valid Unicast Poll from RFC 1122.
1145 	 */
1146 	if (do_dad &&
1147 	    (in_hosteq(isaddr, myaddr) ||
1148 	    (in_nullhost(isaddr) && in_hosteq(itaddr, myaddr) &&
1149 	     m->m_flags & M_BCAST)))
1150 	{
1151 		struct sockaddr_dl sdl, *sdlp;
1152 
1153 		sdlp = sockaddr_dl_init(&sdl, sizeof(sdl),
1154 		    ifp->if_index, ifp->if_type,
1155 		    NULL, 0, ar_sha(ah), ah->ar_hln);
1156 		arp_dad_duplicated((struct ifaddr *)ia, sdlp);
1157 		goto out;
1158 	}
1159 
1160 	/*
1161 	 * If the target IP address is zero, ignore the packet.
1162 	 * This prevents the code below from trying to answer
1163 	 * when we are using IP address zero (booting).
1164 	 */
1165 	if (in_nullhost(itaddr))
1166 		goto out;
1167 
1168 	if (in_nullhost(isaddr))
1169 		goto reply;
1170 
1171 	if (in_hosteq(itaddr, myaddr))
1172 		la = arpcreate(ifp, &isaddr, NULL, 1);
1173 	else
1174 		la = arplookup(ifp, &isaddr, NULL, 1);
1175 	if (la == NULL)
1176 		goto reply;
1177 
1178 	if ((la->la_flags & LLE_VALID) &&
1179 	    memcmp(ar_sha(ah), &la->ll_addr, ifp->if_addrlen))
1180 	{
1181 		char llabuf[LLA_ADDRSTRLEN], *llastr;
1182 
1183 		llastr = lla_snprintf(llabuf, sizeof(llabuf),
1184 		    ar_sha(ah), ah->ar_hln);
1185 
1186 		if (la->la_flags & LLE_STATIC) {
1187 			ARP_STATINC(ARP_STAT_RCVOVERPERM);
1188 			if (!log_permanent_modify)
1189 				goto out;
1190 			log(LOG_INFO,
1191 			    "%s tried to overwrite permanent arp info"
1192 			    " for %s\n", llastr, IN_PRINT(ipbuf, &isaddr));
1193 			goto out;
1194 		} else if (la->lle_tbl->llt_ifp != ifp) {
1195 			/* XXX should not happen? */
1196 			ARP_STATINC(ARP_STAT_RCVOVERINT);
1197 			if (!log_wrong_iface)
1198 				goto out;
1199 			log(LOG_INFO,
1200 			    "%s on %s tried to overwrite "
1201 			    "arp info for %s on %s\n",
1202 			    llastr,
1203 			    ifp->if_xname, IN_PRINT(ipbuf, &isaddr),
1204 			    la->lle_tbl->llt_ifp->if_xname);
1205 				goto out;
1206 		} else {
1207 			ARP_STATINC(ARP_STAT_RCVOVER);
1208 			if (log_movements)
1209 				log(LOG_INFO, "arp info overwritten "
1210 				    "for %s by %s\n",
1211 				    IN_PRINT(ipbuf, &isaddr), llastr);
1212 		}
1213 	}
1214 
1215 	KASSERT(ifp->if_sadl->sdl_alen == ifp->if_addrlen);
1216 
1217 #if NTOKEN > 0
1218 	/*
1219 	 * XXX uses m_data and assumes the complete answer including
1220 	 * XXX token-ring headers is in the same buf
1221 	 */
1222 	if (ifp->if_type == IFT_ISO88025) {
1223 		struct token_header *trh;
1224 
1225 		trh = (struct token_header *)M_TRHSTART(m);
1226 		if (trh->token_shost[0] & TOKEN_RI_PRESENT) {
1227 			struct token_rif *rif;
1228 			size_t riflen;
1229 
1230 			rif = TOKEN_RIF(trh);
1231 			riflen = (ntohs(rif->tr_rcf) &
1232 			    TOKEN_RCF_LEN_MASK) >> 8;
1233 
1234 			if (riflen > 2 &&
1235 			    riflen < sizeof(struct token_rif) &&
1236 			    (riflen & 1) == 0) {
1237 				rif->tr_rcf ^= htons(TOKEN_RCF_DIRECTION);
1238 				rif->tr_rcf &= htons(~TOKEN_RCF_BROADCAST_MASK);
1239 				memcpy(TOKEN_RIF_LLE(la), rif, riflen);
1240 			}
1241 		}
1242 	}
1243 #endif
1244 
1245 	KASSERT(sizeof(la->ll_addr) >= ifp->if_addrlen);
1246 	memcpy(&la->ll_addr, ar_sha(ah), ifp->if_addrlen);
1247 	la->la_flags |= LLE_VALID;
1248 	if ((la->la_flags & LLE_STATIC) == 0) {
1249 		la->la_expire = time_uptime + arpt_keep;
1250 		arp_settimer(la, arpt_keep);
1251 	}
1252 	la->la_asked = 0;
1253 	/* rt->rt_flags &= ~RTF_REJECT; */
1254 
1255 	if (la->la_hold != NULL) {
1256 		int n = la->la_numheld;
1257 		struct mbuf *m_hold, *m_hold_next;
1258 		struct sockaddr_in sin;
1259 
1260 		sockaddr_in_init(&sin, &la->r_l3addr.addr4, 0);
1261 
1262 		m_hold = la->la_hold;
1263 		la->la_hold = NULL;
1264 		la->la_numheld = 0;
1265 		/*
1266 		 * We have to unlock here because if_output would call
1267 		 * arpresolve
1268 		 */
1269 		LLE_WUNLOCK(la);
1270 		ARP_STATADD(ARP_STAT_DFRSENT, n);
1271 		ARP_STATADD(ARP_STAT_DFRTOTAL, n);
1272 		for (; m_hold != NULL; m_hold = m_hold_next) {
1273 			m_hold_next = m_hold->m_nextpkt;
1274 			m_hold->m_nextpkt = NULL;
1275 			if_output_lock(ifp, ifp, m_hold, sintosa(&sin), NULL);
1276 		}
1277 	} else
1278 		LLE_WUNLOCK(la);
1279 	la = NULL;
1280 
1281 reply:
1282 	if (la != NULL) {
1283 		LLE_WUNLOCK(la);
1284 		la = NULL;
1285 	}
1286 	if (op != ARPOP_REQUEST) {
1287 		if (op == ARPOP_REPLY)
1288 			ARP_STATINC(ARP_STAT_RCVREPLY);
1289 		goto out;
1290 	}
1291 	ARP_STATINC(ARP_STAT_RCVREQUEST);
1292 	if (in_hosteq(itaddr, myaddr)) {
1293 		/* If our address is unusable, don't reply */
1294 		if (ia->ia4_flags & (IN_IFF_NOTREADY | IN_IFF_DETACHED))
1295 			goto out;
1296 		/* I am the target */
1297 		tha = ar_tha(ah);
1298 		if (tha)
1299 			memcpy(tha, ar_sha(ah), ah->ar_hln);
1300 		memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
1301 	} else {
1302 		/* Proxy ARP */
1303 		struct llentry *lle = NULL;
1304 		struct sockaddr_in sin;
1305 
1306 #if NCARP > 0
1307 		if (ifp->if_type == IFT_CARP) {
1308 			struct ifnet *_rcvif = m_get_rcvif(m, &s);
1309 			int iftype = 0;
1310 			if (__predict_true(_rcvif != NULL))
1311 				iftype = _rcvif->if_type;
1312 			m_put_rcvif(_rcvif, &s);
1313 			if (iftype != IFT_CARP)
1314 				goto out;
1315 		}
1316 #endif
1317 
1318 		tha = ar_tha(ah);
1319 
1320 		sockaddr_in_init(&sin, &itaddr, 0);
1321 
1322 		IF_AFDATA_RLOCK(ifp);
1323 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1324 		IF_AFDATA_RUNLOCK(ifp);
1325 
1326 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1327 			if (tha)
1328 				memcpy(tha, ar_sha(ah), ah->ar_hln);
1329 			memcpy(ar_sha(ah), &lle->ll_addr, ah->ar_hln);
1330 			LLE_RUNLOCK(lle);
1331 		} else {
1332 			if (lle != NULL)
1333 				LLE_RUNLOCK(lle);
1334 			goto out;
1335 		}
1336 	}
1337 	ia4_release(ia, &psref_ia);
1338 
1339 	/*
1340 	 * XXX XXX: Here we're recycling the mbuf. But the mbuf could have
1341 	 * other mbufs in its chain, and just overwriting m->m_pkthdr.len
1342 	 * would be wrong in this case (the length becomes smaller than the
1343 	 * real chain size).
1344 	 *
1345 	 * This can theoretically cause bugs in the lower layers (drivers,
1346 	 * and L2encap), in some corner cases.
1347 	 */
1348 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1349 	memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1350 	ah->ar_op = htons(ARPOP_REPLY);
1351 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1352 	switch (ifp->if_type) {
1353 	case IFT_IEEE1394:
1354 		/* ieee1394 arp reply is broadcast */
1355 		m->m_flags &= ~M_MCAST;
1356 		m->m_flags |= M_BCAST;
1357 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + ah->ar_hln;
1358 		break;
1359 	default:
1360 		m->m_flags &= ~(M_BCAST|M_MCAST); /* never reply by broadcast */
1361 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1362 		break;
1363 	}
1364 	m->m_pkthdr.len = m->m_len;
1365 	sa.sa_family = AF_ARP;
1366 	sa.sa_len = 2;
1367 	arps = ARP_STAT_GETREF();
1368 	arps[ARP_STAT_SNDTOTAL]++;
1369 	arps[ARP_STAT_SNDREPLY]++;
1370 	ARP_STAT_PUTREF();
1371 	if_output_lock(ifp, ifp, m, &sa, NULL);
1372 	if (rcvif != NULL)
1373 		m_put_rcvif_psref(rcvif, &psref);
1374 	return;
1375 
1376 out:
1377 	if (la != NULL)
1378 		LLE_WUNLOCK(la);
1379 	if (ia != NULL)
1380 		ia4_release(ia, &psref_ia);
1381 	if (rcvif != NULL)
1382 		m_put_rcvif_psref(rcvif, &psref);
1383 	m_freem(m);
1384 }
1385 
1386 /*
1387  * Lookup or a new address in arptab.
1388  */
1389 static struct llentry *
1390 arplookup(struct ifnet *ifp, const struct in_addr *addr,
1391     const struct sockaddr *sa, int wlock)
1392 {
1393 	struct sockaddr_in sin;
1394 	struct llentry *la;
1395 	int flags = wlock ? LLE_EXCLUSIVE : 0;
1396 
1397 	if (sa == NULL) {
1398 		KASSERT(addr != NULL);
1399 		sockaddr_in_init(&sin, addr, 0);
1400 		sa = sintocsa(&sin);
1401 	}
1402 
1403 	IF_AFDATA_RLOCK(ifp);
1404 	la = lla_lookup(LLTABLE(ifp), flags, sa);
1405 	IF_AFDATA_RUNLOCK(ifp);
1406 
1407 	return la;
1408 }
1409 
1410 static struct llentry *
1411 arpcreate(struct ifnet *ifp, const struct in_addr *addr,
1412     const struct sockaddr *sa, int wlock)
1413 {
1414 	struct sockaddr_in sin;
1415 	struct llentry *la;
1416 	int flags = wlock ? LLE_EXCLUSIVE : 0;
1417 
1418 	if (sa == NULL) {
1419 		KASSERT(addr != NULL);
1420 		sockaddr_in_init(&sin, addr, 0);
1421 		sa = sintocsa(&sin);
1422 	}
1423 
1424 	la = arplookup(ifp, addr, sa, wlock);
1425 
1426 	if (la == NULL) {
1427 		struct rtentry *rt;
1428 
1429 		rt = rtalloc1(sa, 0);
1430 		IF_AFDATA_WLOCK(ifp);
1431 		la = lla_create(LLTABLE(ifp), flags, sa, rt);
1432 		IF_AFDATA_WUNLOCK(ifp);
1433 		if (rt != NULL)
1434 			rt_unref(rt);
1435 
1436 		if (la != NULL)
1437 			arp_init_llentry(ifp, la);
1438 	}
1439 
1440 	return la;
1441 }
1442 
1443 int
1444 arpioctl(u_long cmd, void *data)
1445 {
1446 
1447 	return EOPNOTSUPP;
1448 }
1449 
1450 void
1451 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1452 {
1453 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
1454 
1455 	ifa->ifa_rtrequest = arp_rtrequest;
1456 	ifa->ifa_flags |= RTF_CONNECTED;
1457 
1458 	/* ARP will handle DAD for this address. */
1459 	if (in_nullhost(IA_SIN(ifa)->sin_addr)) {
1460 		if (ia->ia_dad_stop != NULL)	/* safety */
1461 			ia->ia_dad_stop(ifa);
1462 		ia->ia_dad_start = NULL;
1463 		ia->ia_dad_stop = NULL;
1464 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
1465 	} else {
1466 		ia->ia_dad_start = arp_dad_start;
1467 		ia->ia_dad_stop = arp_dad_stop;
1468 		if (ia->ia4_flags & IN_IFF_TRYTENTATIVE && ip_dad_enabled())
1469 			ia->ia4_flags |= IN_IFF_TENTATIVE;
1470 		else
1471 			arpannounce1(ifa);
1472 	}
1473 }
1474 
1475 TAILQ_HEAD(dadq_head, dadq);
1476 struct dadq {
1477 	TAILQ_ENTRY(dadq) dad_list;
1478 	struct ifaddr *dad_ifa;
1479 	int dad_count;		/* max ARP to send */
1480 	int dad_arp_tcount;	/* # of trials to send ARP */
1481 	int dad_arp_ocount;	/* ARP sent so far */
1482 	int dad_arp_announce;	/* max ARP announcements */
1483 	int dad_arp_acount;	/* # of announcements */
1484 	struct callout dad_timer_ch;
1485 };
1486 
1487 static struct dadq_head dadq;
1488 static int dad_init = 0;
1489 static int dad_maxtry = 15;     /* max # of *tries* to transmit DAD packet */
1490 static kmutex_t arp_dad_lock;
1491 
1492 static struct dadq *
1493 arp_dad_find(struct ifaddr *ifa)
1494 {
1495 	struct dadq *dp;
1496 
1497 	KASSERT(mutex_owned(&arp_dad_lock));
1498 
1499 	TAILQ_FOREACH(dp, &dadq, dad_list) {
1500 		if (dp->dad_ifa == ifa)
1501 			return dp;
1502 	}
1503 	return NULL;
1504 }
1505 
1506 static void
1507 arp_dad_starttimer(struct dadq *dp, int ticks)
1508 {
1509 
1510 	callout_reset(&dp->dad_timer_ch, ticks,
1511 	    (void (*)(void *))arp_dad_timer, dp);
1512 }
1513 
1514 static void
1515 arp_dad_stoptimer(struct dadq *dp)
1516 {
1517 
1518 	KASSERT(mutex_owned(&arp_dad_lock));
1519 
1520 	TAILQ_REMOVE(&dadq, dp, dad_list);
1521 	/* Tell the timer that dp is being destroyed. */
1522 	dp->dad_ifa = NULL;
1523 	callout_halt(&dp->dad_timer_ch, &arp_dad_lock);
1524 }
1525 
1526 static void
1527 arp_dad_destroytimer(struct dadq *dp)
1528 {
1529 
1530 	callout_destroy(&dp->dad_timer_ch);
1531 	KASSERT(dp->dad_ifa == NULL);
1532 	kmem_intr_free(dp, sizeof(*dp));
1533 }
1534 
1535 static void
1536 arp_dad_output(struct dadq *dp, struct ifaddr *ifa)
1537 {
1538 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
1539 	struct ifnet *ifp = ifa->ifa_ifp;
1540 	struct in_addr sip;
1541 
1542 	dp->dad_arp_tcount++;
1543 	if ((ifp->if_flags & IFF_UP) == 0)
1544 		return;
1545 	if ((ifp->if_flags & IFF_RUNNING) == 0)
1546 		return;
1547 
1548 	dp->dad_arp_tcount = 0;
1549 	dp->dad_arp_ocount++;
1550 
1551 	memset(&sip, 0, sizeof(sip));
1552 	arprequest(ifa->ifa_ifp, &sip, &ia->ia_addr.sin_addr,
1553 	    CLLADDR(ifa->ifa_ifp->if_sadl));
1554 }
1555 
1556 /*
1557  * Start Duplicate Address Detection (DAD) for specified interface address.
1558  */
1559 static void
1560 arp_dad_start(struct ifaddr *ifa)
1561 {
1562 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
1563 	struct dadq *dp;
1564 	char ipbuf[INET_ADDRSTRLEN];
1565 
1566 	if (!dad_init) {
1567 		TAILQ_INIT(&dadq);
1568 		mutex_init(&arp_dad_lock, MUTEX_DEFAULT, IPL_NONE);
1569 		dad_init++;
1570 	}
1571 
1572 	/*
1573 	 * If we don't need DAD, don't do it.
1574 	 * - DAD is disabled
1575 	 */
1576 	if (!(ia->ia4_flags & IN_IFF_TENTATIVE)) {
1577 		log(LOG_DEBUG,
1578 		    "%s: called with non-tentative address %s(%s)\n", __func__,
1579 		    IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
1580 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1581 		return;
1582 	}
1583 	if (!ip_dad_enabled()) {
1584 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
1585 		rt_addrmsg(RTM_NEWADDR, ifa);
1586 		arpannounce1(ifa);
1587 		return;
1588 	}
1589 	KASSERT(ifa->ifa_ifp != NULL);
1590 	if (!(ifa->ifa_ifp->if_flags & IFF_UP))
1591 		return;
1592 
1593 	dp = kmem_intr_alloc(sizeof(*dp), KM_NOSLEEP);
1594 
1595 	mutex_enter(&arp_dad_lock);
1596 	if (arp_dad_find(ifa) != NULL) {
1597 		mutex_exit(&arp_dad_lock);
1598 		/* DAD already in progress */
1599 		if (dp != NULL)
1600 			kmem_intr_free(dp, sizeof(*dp));
1601 		return;
1602 	}
1603 
1604 	if (dp == NULL) {
1605 		mutex_exit(&arp_dad_lock);
1606 		log(LOG_ERR, "%s: memory allocation failed for %s(%s)\n",
1607 		    __func__, IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
1608 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1609 		return;
1610 	}
1611 
1612 	/*
1613 	 * Send ARP packet for DAD, ip_dad_count times.
1614 	 * Note that we must delay the first transmission.
1615 	 */
1616 	callout_init(&dp->dad_timer_ch, CALLOUT_MPSAFE);
1617 	dp->dad_ifa = ifa;
1618 	ifaref(ifa);	/* just for safety */
1619 	dp->dad_count = ip_dad_count;
1620 	dp->dad_arp_announce = 0; /* Will be set when starting to announce */
1621 	dp->dad_arp_acount = dp->dad_arp_ocount = dp->dad_arp_tcount = 0;
1622 	TAILQ_INSERT_TAIL(&dadq, (struct dadq *)dp, dad_list);
1623 
1624 	ARPLOG(LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp),
1625 	    ARPLOGADDR(&ia->ia_addr.sin_addr));
1626 
1627 	arp_dad_starttimer(dp, cprng_fast32() % (PROBE_WAIT * hz));
1628 
1629 	mutex_exit(&arp_dad_lock);
1630 }
1631 
1632 /*
1633  * terminate DAD unconditionally.  used for address removals.
1634  */
1635 static void
1636 arp_dad_stop(struct ifaddr *ifa)
1637 {
1638 	struct dadq *dp;
1639 
1640 	if (!dad_init)
1641 		return;
1642 
1643 	mutex_enter(&arp_dad_lock);
1644 	dp = arp_dad_find(ifa);
1645 	if (dp == NULL) {
1646 		mutex_exit(&arp_dad_lock);
1647 		/* DAD wasn't started yet */
1648 		return;
1649 	}
1650 
1651 	arp_dad_stoptimer(dp);
1652 
1653 	mutex_exit(&arp_dad_lock);
1654 
1655 	arp_dad_destroytimer(dp);
1656 	ifafree(ifa);
1657 }
1658 
1659 static void
1660 arp_dad_timer(struct dadq *dp)
1661 {
1662 	struct ifaddr *ifa;
1663 	struct in_ifaddr *ia;
1664 	char ipbuf[INET_ADDRSTRLEN];
1665 	bool need_free = false;
1666 
1667 	KERNEL_LOCK_UNLESS_NET_MPSAFE();
1668 	mutex_enter(&arp_dad_lock);
1669 
1670 	ifa = dp->dad_ifa;
1671 	if (ifa == NULL) {
1672 		/* dp is being destroyed by someone.  Do nothing. */
1673 		goto done;
1674 	}
1675 
1676 	ia = (struct in_ifaddr *)ifa;
1677 	if (ia->ia4_flags & IN_IFF_DUPLICATED) {
1678 		log(LOG_ERR, "%s: called with duplicate address %s(%s)\n",
1679 		    __func__, IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
1680 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1681 		goto done;
1682 	}
1683 	if ((ia->ia4_flags & IN_IFF_TENTATIVE) == 0 && dp->dad_arp_acount == 0)
1684 	{
1685 		log(LOG_ERR, "%s: called with non-tentative address %s(%s)\n",
1686 		    __func__, IN_PRINT(ipbuf, &ia->ia_addr.sin_addr),
1687 		    ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???");
1688 		goto done;
1689 	}
1690 
1691 	/* timeouted with IFF_{RUNNING,UP} check */
1692 	if (dp->dad_arp_tcount > dad_maxtry) {
1693 		ARPLOG(LOG_INFO, "%s: could not run DAD, driver problem?\n",
1694 		    if_name(ifa->ifa_ifp));
1695 
1696 		arp_dad_stoptimer(dp);
1697 		need_free = true;
1698 		goto done;
1699 	}
1700 
1701 	/* Need more checks? */
1702 	if (dp->dad_arp_ocount < dp->dad_count) {
1703 		int adelay;
1704 
1705 		/*
1706 		 * We have more ARP to go.  Send ARP packet for DAD.
1707 		 */
1708 		arp_dad_output(dp, ifa);
1709 		if (dp->dad_arp_ocount < dp->dad_count)
1710 			adelay = (PROBE_MIN * hz) +
1711 			    (cprng_fast32() %
1712 			    ((PROBE_MAX * hz) - (PROBE_MIN * hz)));
1713 		else
1714 			adelay = ANNOUNCE_WAIT * hz;
1715 		arp_dad_starttimer(dp, adelay);
1716 		goto done;
1717 	} else if (dp->dad_arp_acount == 0) {
1718 		/*
1719 		 * We are done with DAD.
1720 		 * No duplicate address found.
1721 		 */
1722 		ia->ia4_flags &= ~IN_IFF_TENTATIVE;
1723 		rt_addrmsg(RTM_NEWADDR, ifa);
1724 		ARPLOG(LOG_DEBUG,
1725 		    "%s: DAD complete for %s - no duplicates found\n",
1726 		    if_name(ifa->ifa_ifp), ARPLOGADDR(&ia->ia_addr.sin_addr));
1727 		dp->dad_arp_announce = ANNOUNCE_NUM;
1728 		goto announce;
1729 	} else if (dp->dad_arp_acount < dp->dad_arp_announce) {
1730 announce:
1731 		/*
1732 		 * Announce the address.
1733 		 */
1734 		arpannounce1(ifa);
1735 		dp->dad_arp_acount++;
1736 		if (dp->dad_arp_acount < dp->dad_arp_announce) {
1737 			arp_dad_starttimer(dp, ANNOUNCE_INTERVAL * hz);
1738 			goto done;
1739 		}
1740 		ARPLOG(LOG_DEBUG,
1741 		    "%s: ARP announcement complete for %s\n",
1742 		    if_name(ifa->ifa_ifp), ARPLOGADDR(&ia->ia_addr.sin_addr));
1743 	}
1744 
1745 	arp_dad_stoptimer(dp);
1746 	need_free = true;
1747 done:
1748 	mutex_exit(&arp_dad_lock);
1749 
1750 	if (need_free) {
1751 		arp_dad_destroytimer(dp);
1752 		KASSERT(ifa != NULL);
1753 		ifafree(ifa);
1754 	}
1755 
1756 	KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
1757 }
1758 
1759 static void
1760 arp_dad_duplicated(struct ifaddr *ifa, const struct sockaddr_dl *from)
1761 {
1762 	struct in_ifaddr *ia = (struct in_ifaddr *)ifa;
1763 	struct ifnet *ifp = ifa->ifa_ifp;
1764 	char ipbuf[INET_ADDRSTRLEN], llabuf[LLA_ADDRSTRLEN];
1765 	const char *iastr, *llastr;
1766 
1767 	iastr = IN_PRINT(ipbuf, &ia->ia_addr.sin_addr);
1768 	if (__predict_false(from == NULL))
1769 		llastr = NULL;
1770 	else
1771 		llastr = lla_snprintf(llabuf, sizeof(llabuf),
1772 		    CLLADDR(from), from->sdl_alen);
1773 
1774 	if (ia->ia4_flags & (IN_IFF_TENTATIVE|IN_IFF_DUPLICATED)) {
1775 		log(LOG_ERR,
1776 		    "%s: DAD duplicate address %s from %s\n",
1777 		    if_name(ifp), iastr, llastr);
1778 	} else if (ia->ia_dad_defended == 0 ||
1779 		   ia->ia_dad_defended < time_uptime - DEFEND_INTERVAL) {
1780 		ia->ia_dad_defended = time_uptime;
1781 		arpannounce1(ifa);
1782 		log(LOG_ERR,
1783 		    "%s: DAD defended address %s from %s\n",
1784 		    if_name(ifp), iastr, llastr);
1785 		return;
1786 	} else {
1787 		/* If DAD is disabled, just report the duplicate. */
1788 		if (!ip_dad_enabled()) {
1789 			log(LOG_ERR,
1790 			    "%s: DAD ignoring duplicate address %s from %s\n",
1791 			    if_name(ifp), iastr, llastr);
1792 			return;
1793 		}
1794 		log(LOG_ERR,
1795 		    "%s: DAD defence failed for %s from %s\n",
1796 		    if_name(ifp), iastr, llastr);
1797 	}
1798 
1799 	arp_dad_stop(ifa);
1800 
1801 	ia->ia4_flags &= ~IN_IFF_TENTATIVE;
1802 	if ((ia->ia4_flags & IN_IFF_DUPLICATED) == 0) {
1803 		ia->ia4_flags |= IN_IFF_DUPLICATED;
1804 		/* Inform the routing socket of the duplicate address */
1805 		rt_addrmsg_src(RTM_NEWADDR, ifa, (const struct sockaddr *)from);
1806 	}
1807 }
1808 
1809 /*
1810  * Called from 10 Mb/s Ethernet interrupt handlers
1811  * when ether packet type ETHERTYPE_REVARP
1812  * is received.  Common length and type checks are done here,
1813  * then the protocol-specific routine is called.
1814  */
1815 void
1816 revarpinput(struct mbuf *m)
1817 {
1818 	struct arphdr *ar;
1819 	int arplen;
1820 
1821 	arplen = sizeof(struct arphdr);
1822 	if (m->m_len < arplen && (m = m_pullup(m, arplen)) == NULL)
1823 		return;
1824 	ar = mtod(m, struct arphdr *);
1825 
1826 	if (ntohs(ar->ar_hrd) == ARPHRD_IEEE1394) {
1827 		goto out;
1828 	}
1829 
1830 	arplen = sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln);
1831 	if (m->m_len < arplen && (m = m_pullup(m, arplen)) == NULL)
1832 		return;
1833 	ar = mtod(m, struct arphdr *);
1834 
1835 	switch (ntohs(ar->ar_pro)) {
1836 	case ETHERTYPE_IP:
1837 	case ETHERTYPE_IPTRAILERS:
1838 		in_revarpinput(m);
1839 		return;
1840 
1841 	default:
1842 		break;
1843 	}
1844 
1845 out:
1846 	m_freem(m);
1847 }
1848 
1849 /*
1850  * RARP for Internet protocols on 10 Mb/s Ethernet.
1851  * Algorithm is that given in RFC 903.
1852  * We are only using for bootstrap purposes to get an ip address for one of
1853  * our interfaces.  Thus we support no user-interface.
1854  *
1855  * Since the contents of the RARP reply are specific to the interface that
1856  * sent the request, this code must ensure that they are properly associated.
1857  *
1858  * Note: also supports ARP via RARP packets, per the RFC.
1859  */
1860 void
1861 in_revarpinput(struct mbuf *m)
1862 {
1863 	struct arphdr *ah;
1864 	void *tha;
1865 	int op;
1866 	struct ifnet *rcvif;
1867 	int s;
1868 
1869 	ah = mtod(m, struct arphdr *);
1870 	op = ntohs(ah->ar_op);
1871 
1872 	rcvif = m_get_rcvif(m, &s);
1873 	if (__predict_false(rcvif == NULL))
1874 		goto out;
1875 	if (rcvif->if_flags & IFF_NOARP)
1876 		goto out;
1877 
1878 	switch (rcvif->if_type) {
1879 	case IFT_IEEE1394:
1880 		/* ARP without target hardware address is not supported */
1881 		goto out;
1882 	default:
1883 		break;
1884 	}
1885 
1886 	switch (op) {
1887 	case ARPOP_REQUEST:
1888 	case ARPOP_REPLY:	/* per RFC */
1889 		m_put_rcvif(rcvif, &s);
1890 		in_arpinput(m);
1891 		return;
1892 	case ARPOP_REVREPLY:
1893 		break;
1894 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
1895 	default:
1896 		goto out;
1897 	}
1898 	if (!revarp_in_progress)
1899 		goto out;
1900 	if (rcvif != myip_ifp) /* !same interface */
1901 		goto out;
1902 	if (myip_initialized)
1903 		goto wake;
1904 	tha = ar_tha(ah);
1905 	if (tha == NULL)
1906 		goto out;
1907 	if (ah->ar_pln != sizeof(struct in_addr))
1908 		goto out;
1909 	if (ah->ar_hln != rcvif->if_sadl->sdl_alen)
1910 		goto out;
1911 	if (memcmp(tha, CLLADDR(rcvif->if_sadl), rcvif->if_sadl->sdl_alen))
1912 		goto out;
1913 	memcpy(&srv_ip, ar_spa(ah), sizeof(srv_ip));
1914 	memcpy(&myip, ar_tpa(ah), sizeof(myip));
1915 	myip_initialized = 1;
1916 wake:	/* Do wakeup every time in case it was missed. */
1917 	wakeup((void *)&myip);
1918 
1919 out:
1920 	m_put_rcvif(rcvif, &s);
1921 	m_freem(m);
1922 }
1923 
1924 /*
1925  * Send a RARP request for the ip address of the specified interface.
1926  * The request should be RFC 903-compliant.
1927  */
1928 static void
1929 revarprequest(struct ifnet *ifp)
1930 {
1931 	struct sockaddr sa;
1932 	struct mbuf *m;
1933 	struct arphdr *ah;
1934 	void *tha;
1935 
1936 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
1937 		return;
1938 	MCLAIM(m, &arpdomain.dom_mowner);
1939 	m->m_len = sizeof(*ah) + 2*sizeof(struct in_addr) +
1940 	    2*ifp->if_addrlen;
1941 	m->m_pkthdr.len = m->m_len;
1942 	m_align(m, m->m_len);
1943 	ah = mtod(m, struct arphdr *);
1944 	memset(ah, 0, m->m_len);
1945 	ah->ar_pro = htons(ETHERTYPE_IP);
1946 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
1947 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
1948 	ah->ar_op = htons(ARPOP_REVREQUEST);
1949 
1950 	memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
1951 	tha = ar_tha(ah);
1952 	if (tha == NULL) {
1953 		m_free(m);
1954 		return;
1955 	}
1956 	memcpy(tha, CLLADDR(ifp->if_sadl), ah->ar_hln);
1957 
1958 	sa.sa_family = AF_ARP;
1959 	sa.sa_len = 2;
1960 	m->m_flags |= M_BCAST;
1961 
1962 	if_output_lock(ifp, ifp, m, &sa, NULL);
1963 }
1964 
1965 /*
1966  * RARP for the ip address of the specified interface, but also
1967  * save the ip address of the server that sent the answer.
1968  * Timeout if no response is received.
1969  */
1970 int
1971 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
1972     struct in_addr *clnt_in)
1973 {
1974 	int result, count = 20;
1975 
1976 	myip_initialized = 0;
1977 	myip_ifp = ifp;
1978 
1979 	revarp_in_progress = 1;
1980 	while (count--) {
1981 		revarprequest(ifp);
1982 		result = tsleep((void *)&myip, PSOCK, "revarp", hz/2);
1983 		if (result != EWOULDBLOCK)
1984 			break;
1985 	}
1986 	revarp_in_progress = 0;
1987 
1988 	if (!myip_initialized)
1989 		return ENETUNREACH;
1990 
1991 	memcpy(serv_in, &srv_ip, sizeof(*serv_in));
1992 	memcpy(clnt_in, &myip, sizeof(*clnt_in));
1993 	return 0;
1994 }
1995 
1996 void
1997 arp_stat_add(int type, uint64_t count)
1998 {
1999 	ARP_STATADD(type, count);
2000 }
2001 
2002 static int
2003 sysctl_net_inet_arp_stats(SYSCTLFN_ARGS)
2004 {
2005 
2006 	return NETSTAT_SYSCTL(arpstat_percpu, ARP_NSTATS);
2007 }
2008 
2009 static void
2010 sysctl_net_inet_arp_setup(struct sysctllog **clog)
2011 {
2012 	const struct sysctlnode *node;
2013 
2014 	sysctl_createv(clog, 0, NULL, NULL,
2015 			CTLFLAG_PERMANENT,
2016 			CTLTYPE_NODE, "inet", NULL,
2017 			NULL, 0, NULL, 0,
2018 			CTL_NET, PF_INET, CTL_EOL);
2019 	sysctl_createv(clog, 0, NULL, &node,
2020 			CTLFLAG_PERMANENT,
2021 			CTLTYPE_NODE, "arp",
2022 			SYSCTL_DESCR("Address Resolution Protocol"),
2023 			NULL, 0, NULL, 0,
2024 			CTL_NET, PF_INET, CTL_CREATE, CTL_EOL);
2025 
2026 	sysctl_createv(clog, 0, NULL, NULL,
2027 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2028 			CTLTYPE_INT, "keep",
2029 			SYSCTL_DESCR("Valid ARP entry lifetime in seconds"),
2030 			NULL, 0, &arpt_keep, 0,
2031 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2032 
2033 	sysctl_createv(clog, 0, NULL, NULL,
2034 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2035 			CTLTYPE_INT, "down",
2036 			SYSCTL_DESCR("Failed ARP entry lifetime in seconds"),
2037 			NULL, 0, &arpt_down, 0,
2038 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2039 
2040 	sysctl_createv(clog, 0, NULL, NULL,
2041 			CTLFLAG_PERMANENT,
2042 			CTLTYPE_STRUCT, "stats",
2043 			SYSCTL_DESCR("ARP statistics"),
2044 			sysctl_net_inet_arp_stats, 0, NULL, 0,
2045 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2046 
2047 	sysctl_createv(clog, 0, NULL, NULL,
2048 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2049 			CTLTYPE_INT, "log_movements",
2050 			SYSCTL_DESCR("log ARP replies from MACs different than"
2051 			    " the one in the cache"),
2052 			NULL, 0, &log_movements, 0,
2053 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2054 
2055 	sysctl_createv(clog, 0, NULL, NULL,
2056 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2057 			CTLTYPE_INT, "log_permanent_modify",
2058 			SYSCTL_DESCR("log ARP replies from MACs different than"
2059 			    " the one in the permanent arp entry"),
2060 			NULL, 0, &log_permanent_modify, 0,
2061 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2062 
2063 	sysctl_createv(clog, 0, NULL, NULL,
2064 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2065 			CTLTYPE_INT, "log_wrong_iface",
2066 			SYSCTL_DESCR("log ARP packets arriving on the wrong"
2067 			    " interface"),
2068 			NULL, 0, &log_wrong_iface, 0,
2069 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2070 
2071 	sysctl_createv(clog, 0, NULL, NULL,
2072 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2073 			CTLTYPE_INT, "log_unknown_network",
2074 			SYSCTL_DESCR("log ARP packets from non-local network"),
2075 			NULL, 0, &log_unknown_network, 0,
2076 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2077 
2078 	sysctl_createv(clog, 0, NULL, NULL,
2079 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2080 			CTLTYPE_INT, "debug",
2081 			SYSCTL_DESCR("Enable ARP DAD debug output"),
2082 			NULL, 0, &arp_debug, 0,
2083 			CTL_NET, PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
2084 }
2085 
2086 #endif /* INET */
2087