xref: /dragonfly/sys/netinet/if_ether.c (revision 62dc643e)
1 /*
2  * Copyright (c) 2004, 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey M. Hsu.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of The DragonFly Project nor the names of its
16  *    contributors may be used to endorse or promote products derived
17  *    from this software without specific, prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * 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	8.1 (Berkeley) 6/10/93
62  * $FreeBSD: src/sys/netinet/if_ether.c,v 1.64.2.23 2003/04/11 07:23:15 fjoe Exp $
63  */
64 
65 /*
66  * Ethernet address resolution protocol.
67  * TODO:
68  *	add "inuse/lock" bit (or ref. count) along with valid bit
69  */
70 
71 #include "opt_inet.h"
72 #include "opt_carp.h"
73 
74 #include <sys/param.h>
75 #include <sys/kernel.h>
76 #include <sys/queue.h>
77 #include <sys/sysctl.h>
78 #include <sys/systm.h>
79 #include <sys/mbuf.h>
80 #include <sys/malloc.h>
81 #include <sys/socket.h>
82 #include <sys/syslog.h>
83 #include <sys/lock.h>
84 
85 #include <net/if.h>
86 #include <net/if_dl.h>
87 #include <net/if_types.h>
88 #include <net/route.h>
89 #include <net/netisr.h>
90 #include <net/if_llc.h>
91 
92 #include <netinet/in.h>
93 #include <netinet/in_var.h>
94 #include <netinet/if_ether.h>
95 
96 #include <sys/thread2.h>
97 #include <sys/msgport2.h>
98 #include <net/netmsg2.h>
99 #include <net/netisr2.h>
100 #include <sys/mplock2.h>
101 
102 #ifdef CARP
103 #include <netinet/ip_carp.h>
104 #endif
105 
106 #define SIN(s) ((struct sockaddr_in *)s)
107 #define SDL(s) ((struct sockaddr_dl *)s)
108 
109 SYSCTL_DECL(_net_link_ether);
110 SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
111 
112 /* timer values */
113 static int arpt_prune = (5*60*1); /* walk list every 5 minutes */
114 static int arpt_keep = (20*60); /* once resolved, good for 20 more minutes */
115 static int arpt_down = 20;	/* once declared down, don't send for 20 sec */
116 
117 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, prune_intvl, CTLFLAG_RW,
118 	   &arpt_prune, 0, "");
119 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_RW,
120 	   &arpt_keep, 0, "");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, host_down_time, CTLFLAG_RW,
122 	   &arpt_down, 0, "");
123 
124 #define	rt_expire	rt_rmx.rmx_expire
125 
126 struct llinfo_arp {
127 	LIST_ENTRY(llinfo_arp) la_le;
128 	struct	rtentry *la_rt;
129 	struct	mbuf *la_hold;	/* last packet until resolved/timeout */
130 	u_short	la_preempt;	/* countdown for pre-expiry arps */
131 	u_short	la_asked;	/* #times we QUERIED following expiration */
132 };
133 
134 static int	arp_maxtries = 5;
135 static int	useloopback = 1; /* use loopback interface for local traffic */
136 static int	arp_proxyall = 0;
137 static int	arp_refresh = 60; /* refresh arp cache ~60 (not impl yet) */
138 static int	arp_restricted_match = 0;
139 static int	arp_ignore_probes = 1;
140 
141 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_RW,
142 	   &arp_maxtries, 0, "ARP resolution attempts before returning error");
143 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, useloopback, CTLFLAG_RW,
144 	   &useloopback, 0, "Use the loopback interface for local traffic");
145 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_RW,
146 	   &arp_proxyall, 0, "Enable proxy ARP for all suitable requests");
147 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, restricted_match, CTLFLAG_RW,
148 	   &arp_restricted_match, 0, "Only match against the sender");
149 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, refresh, CTLFLAG_RW,
150 	   &arp_refresh, 0, "Preemptively refresh the ARP");
151 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, ignore_probes, CTLFLAG_RW,
152 	   &arp_ignore_probes, 0, "Ignore ARP probes");
153 
154 static void	arp_rtrequest(int, struct rtentry *);
155 static void	arprequest(struct ifnet *, const struct in_addr *,
156 			   const struct in_addr *, const u_char *);
157 static void	arprequest_async(struct ifnet *, const struct in_addr *,
158 				 const struct in_addr *, const u_char *);
159 static void	arpintr(netmsg_t msg);
160 static void	arptfree(struct llinfo_arp *);
161 static void	arptimer(void *);
162 static struct llinfo_arp *
163 		arplookup(in_addr_t, boolean_t, boolean_t, boolean_t);
164 #ifdef INET
165 static void	in_arpinput(struct mbuf *);
166 static void	in_arpreply(struct mbuf *m, in_addr_t, in_addr_t);
167 static void	arp_update_msghandler(netmsg_t);
168 static void	arp_reply_msghandler(netmsg_t);
169 #endif
170 
171 struct arp_pcpu_data {
172 	struct callout		timer_ch;
173 	struct netmsg_base	timer_nmsg;
174 	LIST_HEAD(, llinfo_arp) llinfo_list;
175 } __cachealign;
176 
177 static struct arp_pcpu_data	arp_data[MAXCPU];
178 
179 /*
180  * Timeout routine.  Age arp_tab entries periodically.
181  */
182 static void
183 arptimer_dispatch(netmsg_t nmsg)
184 {
185 	struct llinfo_arp *la, *nla;
186 	struct arp_pcpu_data *ad = &arp_data[mycpuid];
187 
188 	/* Reply ASAP */
189 	crit_enter();
190 	lwkt_replymsg(&nmsg->lmsg, 0);
191 	crit_exit();
192 
193 	LIST_FOREACH_MUTABLE(la, &ad->llinfo_list, la_le, nla) {
194 		if (la->la_rt->rt_expire && la->la_rt->rt_expire <= time_uptime)
195 			arptfree(la);
196 	}
197 	callout_reset(&ad->timer_ch, arpt_prune * hz, arptimer, NULL);
198 }
199 
200 static void
201 arptimer(void *arg __unused)
202 {
203 	int cpuid = mycpuid;
204 	struct lwkt_msg *lmsg = &arp_data[cpuid].timer_nmsg.lmsg;
205 
206 	crit_enter();
207 	if (lmsg->ms_flags & MSGF_DONE)
208 		lwkt_sendmsg_oncpu(netisr_cpuport(cpuid), lmsg);
209 	crit_exit();
210 }
211 
212 /*
213  * Parallel to llc_rtrequest.
214  *
215  * Called after a route is successfully added to the tree to fix-up the
216  * route and initiate arp operations if required.
217  */
218 static void
219 arp_rtrequest(int req, struct rtentry *rt)
220 {
221 	struct sockaddr *gate = rt->rt_gateway;
222 	struct llinfo_arp *la = rt->rt_llinfo;
223 
224 	struct sockaddr_dl null_sdl = { sizeof null_sdl, AF_LINK };
225 
226 	if (rt->rt_flags & RTF_GATEWAY)
227 		return;
228 
229 	switch (req) {
230 	case RTM_ADD:
231 		/*
232 		 * XXX: If this is a manually added route to interface
233 		 * such as older version of routed or gated might provide,
234 		 * restore cloning bit.
235 		 */
236 		if (!(rt->rt_flags & RTF_HOST) &&
237 		    SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
238 			rt->rt_flags |= RTF_CLONING;
239 		if (rt->rt_flags & RTF_CLONING) {
240 			/*
241 			 * Case 1: This route should come from a route to iface.
242 			 */
243 			rt_setgate(rt, rt_key(rt),
244 				   (struct sockaddr *)&null_sdl,
245 				   RTL_DONTREPORT);
246 			gate = rt->rt_gateway;
247 			SDL(gate)->sdl_type = rt->rt_ifp->if_type;
248 			SDL(gate)->sdl_index = rt->rt_ifp->if_index;
249 			rt->rt_expire = time_uptime;
250 			break;
251 		}
252 		/*
253 		 * Announce a new entry if requested, and only announce it
254 		 * once on cpu0.
255 		 */
256 		if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
257 			arprequest_async(rt->rt_ifp,
258 			    &SIN(rt_key(rt))->sin_addr,
259 			    &SIN(rt_key(rt))->sin_addr,
260 			    LLADDR(SDL(gate)));
261 		}
262 		/*FALLTHROUGH*/
263 	case RTM_RESOLVE:
264 		if (gate->sa_family != AF_LINK ||
265 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
266 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
267 			break;
268 		}
269 		SDL(gate)->sdl_type = rt->rt_ifp->if_type;
270 		SDL(gate)->sdl_index = rt->rt_ifp->if_index;
271 		if (la != NULL)
272 			break; /* This happens on a route change */
273 		/*
274 		 * Case 2:  This route may come from cloning, or a manual route
275 		 * add with a LL address.
276 		 */
277 		R_Malloc(la, struct llinfo_arp *, sizeof *la);
278 		rt->rt_llinfo = la;
279 		if (la == NULL) {
280 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
281 			break;
282 		}
283 		bzero(la, sizeof *la);
284 		la->la_rt = rt;
285 		rt->rt_flags |= RTF_LLINFO;
286 		LIST_INSERT_HEAD(&arp_data[mycpuid].llinfo_list, la, la_le);
287 
288 #ifdef INET
289 		/*
290 		 * This keeps the multicast addresses from showing up
291 		 * in `arp -a' listings as unresolved.  It's not actually
292 		 * functional.  Then the same for broadcast.
293 		 */
294 		if (IN_MULTICAST(ntohl(SIN(rt_key(rt))->sin_addr.s_addr))) {
295 			ETHER_MAP_IP_MULTICAST(&SIN(rt_key(rt))->sin_addr,
296 					       LLADDR(SDL(gate)));
297 			SDL(gate)->sdl_alen = 6;
298 			rt->rt_expire = 0;
299 		}
300 		if (in_broadcast(SIN(rt_key(rt))->sin_addr, rt->rt_ifp)) {
301 			memcpy(LLADDR(SDL(gate)), rt->rt_ifp->if_broadcastaddr,
302 			       rt->rt_ifp->if_addrlen);
303 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
304 			rt->rt_expire = 0;
305 		}
306 #endif
307 
308 		/*
309 		 * This fixes up the routing interface for local addresses.
310 		 * The route is adjusted to point at lo0 and the expiration
311 		 * timer is disabled.
312 		 *
313 		 * NOTE: This prevents locally targetted traffic from going
314 		 *	 out the hardware interface, which is inefficient
315 		 *	 and might not work if the hardware cannot listen
316 		 *	 to its own transmitted packets.   Setting
317 		 *	 net.link.ether.inet.useloopback to 0 will force
318 		 *	 packets for local addresses out the hardware (and
319 		 *	 it is expected to receive its own packet).
320 		 *
321 		 * XXX We should just be able to test RTF_LOCAL here instead
322 		 *     of having to compare IPs.
323 		 */
324 		if (SIN(rt_key(rt))->sin_addr.s_addr ==
325 		    (IA_SIN(rt->rt_ifa))->sin_addr.s_addr) {
326 			rt->rt_expire = 0;
327 			bcopy(IF_LLADDR(rt->rt_ifp), LLADDR(SDL(gate)),
328 			      SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen);
329 			if (useloopback)
330 				rt->rt_ifp = loif;
331 		}
332 		break;
333 
334 	case RTM_DELETE:
335 		if (la == NULL)
336 			break;
337 		LIST_REMOVE(la, la_le);
338 		rt->rt_llinfo = NULL;
339 		rt->rt_flags &= ~RTF_LLINFO;
340 		if (la->la_hold != NULL)
341 			m_freem(la->la_hold);
342 		Free(la);
343 		break;
344 	}
345 }
346 
347 static struct mbuf *
348 arpreq_alloc(struct ifnet *ifp, const struct in_addr *sip,
349 	     const struct in_addr *tip, const u_char *enaddr)
350 {
351 	struct mbuf *m;
352 	struct arphdr *ah;
353 	u_short ar_hrd;
354 
355 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
356 		return NULL;
357 	m->m_pkthdr.rcvif = NULL;
358 
359 	switch (ifp->if_type) {
360 	case IFT_ETHER:
361 		/*
362 		 * This may not be correct for types not explicitly
363 		 * listed, but this is our best guess
364 		 */
365 	default:
366 		ar_hrd = htons(ARPHRD_ETHER);
367 
368 		m->m_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
369 		m->m_pkthdr.len = m->m_len;
370 		MH_ALIGN(m, m->m_len);
371 
372 		ah = mtod(m, struct arphdr *);
373 		break;
374 	}
375 
376 	ah->ar_hrd = ar_hrd;
377 	ah->ar_pro = htons(ETHERTYPE_IP);
378 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
379 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
380 	ah->ar_op = htons(ARPOP_REQUEST);
381 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
382 	memset(ar_tha(ah), 0, ah->ar_hln);
383 	memcpy(ar_spa(ah), sip, ah->ar_pln);
384 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
385 
386 	return m;
387 }
388 
389 static void
390 arpreq_send(struct ifnet *ifp, struct mbuf *m)
391 {
392 	struct sockaddr sa;
393 	struct ether_header *eh;
394 
395 	switch (ifp->if_type) {
396 	case IFT_ETHER:
397 		/*
398 		 * This may not be correct for types not explicitly
399 		 * listed, but this is our best guess
400 		 */
401 	default:
402 		eh = (struct ether_header *)sa.sa_data;
403 		/* if_output() will not swap */
404 		eh->ether_type = htons(ETHERTYPE_ARP);
405 		memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ifp->if_addrlen);
406 		break;
407 	}
408 
409 	sa.sa_family = AF_UNSPEC;
410 	sa.sa_len = sizeof(sa);
411 	ifp->if_output(ifp, m, &sa, NULL);
412 }
413 
414 static void
415 arpreq_send_handler(netmsg_t msg)
416 {
417 	struct mbuf *m = msg->packet.nm_packet;
418 	struct ifnet *ifp = msg->lmsg.u.ms_resultp;
419 
420 	arpreq_send(ifp, m);
421 	/* nmsg was embedded in the mbuf, do not reply! */
422 }
423 
424 /*
425  * Broadcast an ARP request. Caller specifies:
426  *	- arp header source ip address
427  *	- arp header target ip address
428  *	- arp header source ethernet address
429  *
430  * NOTE: Caller MUST NOT hold ifp's serializer
431  */
432 static void
433 arprequest(struct ifnet *ifp, const struct in_addr *sip,
434 	   const struct in_addr *tip, const u_char *enaddr)
435 {
436 	struct mbuf *m;
437 
438 	if (enaddr == NULL) {
439 		if (ifp->if_bridge) {
440 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
441 		} else {
442 			enaddr = IF_LLADDR(ifp);
443 		}
444 	}
445 
446 	m = arpreq_alloc(ifp, sip, tip, enaddr);
447 	if (m == NULL)
448 		return;
449 	arpreq_send(ifp, m);
450 }
451 
452 /*
453  * Same as arprequest(), except:
454  * - Caller is allowed to hold ifp's serializer
455  * - Network output is done in protocol thead
456  */
457 static void
458 arprequest_async(struct ifnet *ifp, const struct in_addr *sip,
459 		 const struct in_addr *tip, const u_char *enaddr)
460 {
461 	struct mbuf *m;
462 	struct netmsg_packet *pmsg;
463 
464 	if (enaddr == NULL) {
465 		if (ifp->if_bridge) {
466 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
467 		} else {
468 			enaddr = IF_LLADDR(ifp);
469 		}
470 	}
471 	m = arpreq_alloc(ifp, sip, tip, enaddr);
472 	if (m == NULL)
473 		return;
474 
475 	pmsg = &m->m_hdr.mh_netmsg;
476 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
477 		    0, arpreq_send_handler);
478 	pmsg->nm_packet = m;
479 	pmsg->base.lmsg.u.ms_resultp = ifp;
480 
481 	lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), &pmsg->base.lmsg);
482 }
483 
484 /*
485  * Resolve an IP address into an ethernet address.  If success,
486  * desten is filled in.  If there is no entry in arptab,
487  * set one up and broadcast a request for the IP address.
488  * Hold onto this mbuf and resend it once the address
489  * is finally resolved.  A return value of 1 indicates
490  * that desten has been filled in and the packet should be sent
491  * normally; a 0 return indicates that the packet has been
492  * taken over here, either now or for later transmission.
493  */
494 int
495 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
496 	   struct sockaddr *dst, u_char *desten)
497 {
498 	struct rtentry *rt = NULL;
499 	struct llinfo_arp *la = NULL;
500 	struct sockaddr_dl *sdl;
501 
502 	if (m->m_flags & M_BCAST) {	/* broadcast */
503 		memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen);
504 		return (1);
505 	}
506 	if (m->m_flags & M_MCAST) {/* multicast */
507 		ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
508 		return (1);
509 	}
510 	if (rt0 != NULL) {
511 		if (rt_llroute(dst, rt0, &rt) != 0) {
512 			m_freem(m);
513 			return 0;
514 		}
515 		la = rt->rt_llinfo;
516 	}
517 	if (la == NULL) {
518 		la = arplookup(SIN(dst)->sin_addr.s_addr,
519 			       TRUE, RTL_REPORTMSG, FALSE);
520 		if (la != NULL)
521 			rt = la->la_rt;
522 	}
523 	if (la == NULL || rt == NULL) {
524 		char addr[INET_ADDRSTRLEN];
525 
526 		log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s%s%s\n",
527 		    kinet_ntoa(SIN(dst)->sin_addr, addr), la ? "la" : " ",
528 		    rt ? "rt" : "");
529 		m_freem(m);
530 		return (0);
531 	}
532 	sdl = SDL(rt->rt_gateway);
533 	/*
534 	 * Check the address family and length is valid, the address
535 	 * is resolved; otherwise, try to resolve.
536 	 */
537 	if ((rt->rt_expire == 0 || rt->rt_expire > time_uptime) &&
538 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
539 		/*
540 		 * If entry has an expiry time and it is approaching,
541 		 * see if we need to send an ARP request within this
542 		 * arpt_down interval.
543 		 */
544 		if ((rt->rt_expire != 0) &&
545 		    (time_uptime + la->la_preempt > rt->rt_expire)) {
546 			arprequest(ifp,
547 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
548 				   &SIN(dst)->sin_addr,
549 				   NULL);
550 			la->la_preempt--;
551 		}
552 
553 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
554 		return 1;
555 	}
556 	/*
557 	 * If ARP is disabled or static on this interface, stop.
558 	 * XXX
559 	 * Probably should not allocate empty llinfo struct if we are
560 	 * not going to be sending out an arp request.
561 	 */
562 	if (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) {
563 		m_freem(m);
564 		return (0);
565 	}
566 	/*
567 	 * There is an arptab entry, but no ethernet address
568 	 * response yet.  Replace the held mbuf with this
569 	 * latest one.
570 	 */
571 	if (la->la_hold != NULL)
572 		m_freem(la->la_hold);
573 	la->la_hold = m;
574 	if (rt->rt_expire || ((rt->rt_flags & RTF_STATIC) && !sdl->sdl_alen)) {
575 		rt->rt_flags &= ~RTF_REJECT;
576 		if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
577 			rt->rt_expire = time_uptime;
578 			if (la->la_asked++ < arp_maxtries) {
579 				arprequest(ifp,
580 					   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
581 					   &SIN(dst)->sin_addr,
582 					   NULL);
583 			} else {
584 				rt->rt_flags |= RTF_REJECT;
585 				rt->rt_expire += arpt_down;
586 				la->la_asked = 0;
587 				la->la_preempt = arp_maxtries;
588 			}
589 		}
590 	}
591 	return (0);
592 }
593 
594 /*
595  * Common length and type checks are done here,
596  * then the protocol-specific routine is called.
597  */
598 static void
599 arpintr(netmsg_t msg)
600 {
601 	struct mbuf *m = msg->packet.nm_packet;
602 	struct arphdr *ar;
603 	u_short ar_hrd;
604 	char hexstr[6];
605 
606 	if (m->m_len < sizeof(struct arphdr) &&
607 	    (m = m_pullup(m, sizeof(struct arphdr))) == NULL) {
608 		log(LOG_ERR, "arp: runt packet -- m_pullup failed\n");
609 		return;
610 	}
611 	ar = mtod(m, struct arphdr *);
612 
613 	ar_hrd = ntohs(ar->ar_hrd);
614 	if (ar_hrd != ARPHRD_ETHER && ar_hrd != ARPHRD_IEEE802) {
615 		hexncpy((unsigned char *)&ar->ar_hrd, 2, hexstr, 5, NULL);
616 		log(LOG_ERR, "arp: unknown hardware address format (0x%s)\n",
617 		    hexstr);
618 		m_freem(m);
619 		return;
620 	}
621 
622 	if (m->m_pkthdr.len < arphdr_len(ar)) {
623 		if ((m = m_pullup(m, arphdr_len(ar))) == NULL) {
624 			log(LOG_ERR, "arp: runt packet\n");
625 			return;
626 		}
627 		ar = mtod(m, struct arphdr *);
628 	}
629 
630 	switch (ntohs(ar->ar_pro)) {
631 #ifdef INET
632 	case ETHERTYPE_IP:
633 		in_arpinput(m);
634 		return;
635 #endif
636 	}
637 	m_freem(m);
638 	/* msg was embedded in the mbuf, do not reply! */
639 }
640 
641 #ifdef INET
642 /*
643  * ARP for Internet protocols on 10 Mb/s Ethernet.
644  * Algorithm is that given in RFC 826.
645  * In addition, a sanity check is performed on the sender
646  * protocol address, to catch impersonators.
647  * We no longer handle negotiations for use of trailer protocol:
648  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
649  * along with IP replies if we wanted trailers sent to us,
650  * and also sent them in response to IP replies.
651  * This allowed either end to announce the desire to receive
652  * trailer packets.
653  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
654  * but formerly didn't normally send requests.
655  */
656 
657 static int	log_arp_wrong_iface = 1;
658 static int	log_arp_movements = 1;
659 static int	log_arp_permanent_modify = 1;
660 static int	log_arp_creation_failure = 1;
661 
662 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
663 	   &log_arp_wrong_iface, 0,
664 	   "Log arp packets arriving on the wrong interface");
665 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
666 	   &log_arp_movements, 0,
667 	   "Log arp replies from MACs different than the one in the cache");
668 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
669 	   &log_arp_permanent_modify, 0,
670 	   "Log arp replies from MACs different than the one "
671 	   "in the permanent arp entry");
672 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_creation_failure, CTLFLAG_RW,
673 	   &log_arp_creation_failure, 0, "Log arp creation failure");
674 
675 /*
676  * Returns non-zero if the routine updated anything.
677  */
678 static int
679 arp_update_oncpu(struct mbuf *m, in_addr_t saddr, boolean_t create,
680 		 boolean_t generate_report, boolean_t dologging)
681 {
682 	struct arphdr *ah = mtod(m, struct arphdr *);
683 	struct ifnet *ifp = m->m_pkthdr.rcvif;
684 	struct llinfo_arp *la;
685 	struct sockaddr_dl *sdl;
686 	struct rtentry *rt;
687 	char hexstr[2][64];
688 	char sbuf[INET_ADDRSTRLEN];
689 	int changed = create;
690 
691 	KASSERT(curthread->td_type == TD_TYPE_NETISR,
692 	    ("arp update not in netisr"));
693 
694 	la = arplookup(saddr, create, generate_report, FALSE);
695 	if (la && (rt = la->la_rt) && (sdl = SDL(rt->rt_gateway))) {
696 		struct in_addr isaddr = { saddr };
697 
698 		/*
699 		 * Normally arps coming in on the wrong interface are ignored,
700 		 * but if we are bridging and the two interfaces belong to
701 		 * the same bridge, or one is a member of the bridge which
702 		 * is the other, then it isn't an error.
703 		 */
704 		if (rt->rt_ifp != ifp) {
705 			/*
706 			 * (1) ifp and rt_ifp both members of same bridge
707 			 * (2) rt_ifp member of bridge ifp
708 			 * (3) ifp member of bridge rt_ifp
709 			 *
710 			 * Always replace rt_ifp with the bridge ifc.
711 			 */
712 			struct ifnet *nifp;
713 
714 			if (ifp->if_bridge &&
715 			    rt->rt_ifp->if_bridge == ifp->if_bridge) {
716 				nifp = ether_bridge_interface(ifp);
717 			} else if (rt->rt_ifp->if_bridge &&
718 				   ether_bridge_interface(rt->rt_ifp) == ifp) {
719 				nifp = ifp;
720 			} else if (ifp->if_bridge &&
721 				   ether_bridge_interface(ifp) == rt->rt_ifp) {
722 				nifp = rt->rt_ifp;
723 			} else {
724 				nifp = NULL;
725 			}
726 
727 			if ((log_arp_wrong_iface == 1 && nifp == NULL) ||
728 			    log_arp_wrong_iface == 2) {
729 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
730 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
731 				log(LOG_ERR,
732 				    "arp: %s is on %s "
733 				    "but got reply from %s on %s\n",
734 				    kinet_ntoa(isaddr, sbuf),
735 				    rt->rt_ifp->if_xname, hexstr[0],
736 				    ifp->if_xname);
737 			}
738 			if (nifp == NULL)
739 				return 0;
740 
741 			/*
742 			 * nifp is our man!  Replace rt_ifp and adjust
743 			 * the sdl.
744 			 */
745 			ifp = rt->rt_ifp = nifp;
746 			if (sdl->sdl_type != ifp->if_type) {
747 				sdl->sdl_type = ifp->if_type;
748 				changed = 1;
749 			}
750 			if (sdl->sdl_index != ifp->if_index) {
751 				sdl->sdl_index = ifp->if_index;
752 				changed = 1;
753 			}
754 		}
755 		if (sdl->sdl_alen &&
756 		    bcmp(ar_sha(ah), LLADDR(sdl), sdl->sdl_alen)) {
757 			changed = 1;
758 			if (rt->rt_expire != 0) {
759 				if (dologging && log_arp_movements) {
760 					hexncpy((u_char *)LLADDR(sdl), ifp->if_addrlen,
761 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
762 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
763 					    hexstr[1], HEX_NCPYLEN(ifp->if_addrlen), ":");
764 			    		log(LOG_INFO,
765 					    "arp: %s moved from %s to %s on %s\n",
766 					    kinet_ntoa(isaddr, sbuf), hexstr[0], hexstr[1],
767 					    ifp->if_xname);
768 				}
769 			} else {
770 				if (dologging && log_arp_permanent_modify) {
771 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
772 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
773 					log(LOG_ERR,
774 					"arp: %s attempts to modify "
775 					"permanent entry for %s on %s\n",
776 					hexstr[0], kinet_ntoa(isaddr, sbuf), ifp->if_xname);
777 				}
778 				return changed;
779 			}
780 		}
781 		/*
782 		 * sanity check for the address length.
783 		 * XXX this does not work for protocols with variable address
784 		 * length. -is
785 		 */
786 		if (dologging && sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
787 			hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
788 			    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
789 			log(LOG_WARNING,
790 			    "arp from %s: new addr len %d, was %d",
791 			    hexstr[0], ah->ar_hln, sdl->sdl_alen);
792 		}
793 		if (ifp->if_addrlen != ah->ar_hln) {
794 			if (dologging) {
795 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
796 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
797 				log(LOG_WARNING,
798 				"arp from %s: addr len: new %d, i/f %d "
799 				"(ignored)", hexstr[0],
800 				ah->ar_hln, ifp->if_addrlen);
801 			}
802 			return changed;
803 		}
804 		memcpy(LLADDR(sdl), ar_sha(ah), sdl->sdl_alen = ah->ar_hln);
805 		if (rt->rt_expire != 0) {
806 			if (rt->rt_expire != time_uptime + arpt_keep &&
807 			    rt->rt_expire != time_uptime + arpt_keep - 1) {
808 				rt->rt_expire = time_uptime + arpt_keep;
809 				changed = 1;
810 			}
811 		}
812 		if (rt->rt_flags & RTF_REJECT) {
813 			rt->rt_flags &= ~RTF_REJECT;
814 			changed = 1;
815 		}
816 		if (la->la_asked != 0) {
817 			la->la_asked = 0;
818 			changed = 1;
819 		}
820 		if (la->la_preempt != arp_maxtries) {
821 			la->la_preempt = arp_maxtries;
822 			changed = 1;
823 		}
824 
825 		/*
826 		 * This particular cpu might have been holding an mbuf
827 		 * pending ARP resolution.  If so, transmit the mbuf now.
828 		 */
829 		if (la->la_hold != NULL) {
830 			struct mbuf *m = la->la_hold;
831 
832 			la->la_hold = NULL;
833 			m_adj(m, sizeof(struct ether_header));
834 			ifp->if_output(ifp, m, rt_key(rt), rt);
835 			changed = 1;
836 		}
837 	}
838 	return changed;
839 }
840 
841 /*
842  * Called from arpintr() - this routine is run from a single cpu.
843  */
844 static void
845 in_arpinput(struct mbuf *m)
846 {
847 	struct arphdr *ah;
848 	struct ifnet *ifp = m->m_pkthdr.rcvif;
849 	struct ifaddr_container *ifac;
850 	struct in_ifaddr_container *iac;
851 	struct in_ifaddr *ia = NULL;
852 	struct in_addr isaddr, itaddr, myaddr;
853 	uint8_t *enaddr = NULL;
854 	int req_len;
855 	int changed;
856 	char hexstr[64], sbuf[INET_ADDRSTRLEN];
857 
858 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
859 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
860 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
861 		return;
862 	}
863 
864 	ah = mtod(m, struct arphdr *);
865 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
866 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
867 
868 	/*
869 	 * Check both target and sender IP addresses:
870 	 *
871 	 * If we receive the packet on the interface owning the address,
872 	 * then accept the address.
873 	 *
874 	 * For a bridge, we accept the address if the receive interface and
875 	 * the interface owning the address are on the same bridge, and
876 	 * use the bridge MAC as the is-at response.  The bridge will be
877 	 * responsible for handling the packet.
878 	 *
879 	 * (0) Check target IP against CARP IPs
880 	 */
881 #ifdef CARP
882 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
883 		int is_match = 0, is_parent = 0;
884 
885 		ia = iac->ia;
886 
887 		/* Skip all ia's which don't match */
888 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
889 			continue;
890 
891 		if (ia->ia_ifp->if_type != IFT_CARP)
892 			continue;
893 
894 		if (carp_parent(ia->ia_ifp) == ifp)
895 			is_parent = 1;
896 		if (is_parent || ia->ia_ifp == ifp)
897 			is_match = carp_iamatch(ia);
898 
899 		if (is_match) {
900 			if (is_parent) {
901 				/*
902 				 * The parent interface will also receive
903 				 * the ethernet broadcast packets, e.g. ARP
904 				 * REQUEST, so if we could find a CARP
905 				 * interface of the parent that could match
906 				 * the target IP address, we then drop the
907 				 * packets, which is delieverd to us through
908 				 * the parent interface.
909 				 */
910 				m_freem(m);
911 				return;
912 			}
913 			goto match;
914 		}
915 	}
916 #endif	/* CARP */
917 
918 	/*
919 	 * (1) Check target IP against our local IPs
920 	 */
921 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
922 		ia = iac->ia;
923 
924 		/* Skip all ia's which don't match */
925 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
926 			continue;
927 
928 #ifdef CARP
929 		/* CARP interfaces are checked in (0) */
930 		if (ia->ia_ifp->if_type == IFT_CARP)
931 			continue;
932 #endif
933 
934 		if (ifp->if_bridge && ia->ia_ifp &&
935 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
936 			ifp = ether_bridge_interface(ifp);
937 			goto match;
938 		}
939 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
940 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
941 			goto match;
942 		}
943 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
944 		    ia->ia_ifp) {
945 			goto match;
946 		}
947 		if (ia->ia_ifp == ifp) {
948 			goto match;
949 		}
950 	}
951 
952 	/*
953 	 * (2) Check sender IP against our local IPs
954 	 */
955 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
956 		ia = iac->ia;
957 
958 		/* Skip all ia's which don't match */
959 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
960 			continue;
961 
962 		if (ifp->if_bridge && ia->ia_ifp &&
963 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
964 			ifp = ether_bridge_interface(ifp);
965 			goto match;
966 		}
967 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
968 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
969 			goto match;
970 		}
971 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
972 		    ia->ia_ifp) {
973 			goto match;
974 		}
975 
976 		if (ia->ia_ifp == ifp)
977 			goto match;
978 	}
979 
980 	/*
981 	 * No match, use the first inet address on the receive interface
982 	 * as a dummy address for the rest of the function.
983 	 */
984 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
985 		struct ifaddr *ifa = ifac->ifa;
986 
987 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
988 			ia = ifatoia(ifa);
989 			goto match;
990 		}
991 	}
992 
993 	/*
994 	 * If we got here, we didn't find any suitable interface,
995 	 * so drop the packet.
996 	 */
997 	m_freem(m);
998 	return;
999 
1000 match:
1001 	if (!enaddr)
1002 		enaddr = (uint8_t *)IF_LLADDR(ifp);
1003 	myaddr = ia->ia_addr.sin_addr;
1004 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
1005 		m_freem(m);	/* it's from me, ignore it. */
1006 		return;
1007 	}
1008 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1009 		log(LOG_ERR,
1010 		    "arp: link address is broadcast for IP address %s!\n",
1011 		    kinet_ntoa(isaddr, sbuf));
1012 		m_freem(m);
1013 		return;
1014 	}
1015 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1016 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1017 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1018 		log(LOG_ERR,
1019 		   "arp: %s is using my IP address %s!\n",
1020 		    hexstr, kinet_ntoa(isaddr, sbuf));
1021 		itaddr = myaddr;
1022 		goto reply;
1023 	}
1024 	if (ifp->if_flags & IFF_STATICARP)
1025 		goto reply;
1026 
1027 	/*
1028 	 * When arp_restricted_match is true and the ARP response is not
1029 	 * specifically targetted to me, ignore it.  Otherwise the entry
1030 	 * timeout may be updated for an old MAC.
1031 	 */
1032 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1033 		m_freem(m);
1034 		return;
1035 	}
1036 
1037 	/*
1038 	 * Update all CPU's routing tables with this ARP packet.
1039 	 *
1040 	 * However, we only need to generate rtmsg on CPU0.
1041 	 */
1042 	ASSERT_IN_NETISR(0);
1043 	changed = arp_update_oncpu(m, isaddr.s_addr,
1044 				   itaddr.s_addr == myaddr.s_addr,
1045 				   RTL_REPORTMSG, TRUE);
1046 
1047 	if (ncpus > 1 && changed) {
1048 		struct netmsg_inarp *msg = &m->m_hdr.mh_arpmsg;
1049 
1050 		netmsg_init(&msg->base, NULL, &netisr_apanic_rport,
1051 			    0, arp_update_msghandler);
1052 		msg->m = m;
1053 		msg->saddr = isaddr.s_addr;
1054 		msg->taddr = itaddr.s_addr;
1055 		msg->myaddr = myaddr.s_addr;
1056 		lwkt_sendmsg(netisr_cpuport(1), &msg->base.lmsg);
1057 	} else {
1058 		goto reply;
1059 	}
1060 
1061 	/*
1062 	 * Just return here; after all CPUs's routing tables are
1063 	 * properly updated by this ARP packet, an ARP reply will
1064 	 * be generated if appropriate.
1065 	 */
1066 	return;
1067 reply:
1068 	in_arpreply(m, itaddr.s_addr, myaddr.s_addr);
1069 }
1070 
1071 static void
1072 arp_reply_msghandler(netmsg_t msg)
1073 {
1074 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1075 
1076 	in_arpreply(rmsg->m, rmsg->taddr, rmsg->myaddr);
1077 	/* Don't reply this netmsg; netmsg_inarp is embedded in mbuf */
1078 }
1079 
1080 static void
1081 arp_update_msghandler(netmsg_t msg)
1082 {
1083 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1084 	int nextcpu;
1085 
1086 	/*
1087 	 * This message handler will be called on all of the APs;
1088 	 * no need to generate rtmsg on them.
1089 	 */
1090 	KASSERT(mycpuid > 0, ("arp update msg on cpu%d", mycpuid));
1091 	arp_update_oncpu(rmsg->m, rmsg->saddr,
1092 			 rmsg->taddr == rmsg->myaddr,
1093 			 RTL_DONTREPORT, FALSE);
1094 
1095 	nextcpu = mycpuid + 1;
1096 	if (nextcpu < ncpus) {
1097 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
1098 	} else {
1099 		struct mbuf *m = rmsg->m;
1100 		in_addr_t saddr = rmsg->saddr;
1101 		in_addr_t taddr = rmsg->taddr;
1102 		in_addr_t myaddr = rmsg->myaddr;
1103 
1104 		/*
1105 		 * Dispatch this mbuf to netisr0 to perform ARP reply,
1106 		 * if appropriate.
1107 		 * NOTE: netmsg_inarp is embedded in this mbuf.
1108 		 */
1109 		netmsg_init(&rmsg->base, NULL, &netisr_apanic_rport,
1110 		    0, arp_reply_msghandler);
1111 		rmsg->m = m;
1112 		rmsg->saddr = saddr;
1113 		rmsg->taddr = taddr;
1114 		rmsg->myaddr = myaddr;
1115 		lwkt_sendmsg(netisr_cpuport(0), &rmsg->base.lmsg);
1116 	}
1117 }
1118 
1119 static void
1120 in_arpreply(struct mbuf *m, in_addr_t taddr, in_addr_t myaddr)
1121 {
1122 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1123 	const uint8_t *enaddr;
1124 	struct arphdr *ah;
1125 	struct sockaddr sa;
1126 	struct ether_header *eh;
1127 
1128 	ah = mtod(m, struct arphdr *);
1129 	if (ntohs(ah->ar_op) != ARPOP_REQUEST) {
1130 		m_freem(m);
1131 		return;
1132 	}
1133 
1134 	enaddr = (const uint8_t *)IF_LLADDR(ifp);
1135 	if (taddr == myaddr) {
1136 		/* I am the target */
1137 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1138 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1139 	} else {
1140 		struct llinfo_arp *la;
1141 		struct rtentry *rt;
1142 
1143 		la = arplookup(taddr, FALSE, RTL_DONTREPORT, SIN_PROXY);
1144 		if (la == NULL) {
1145 			struct sockaddr_in sin;
1146 #ifdef DEBUG_PROXY
1147 			char tbuf[INET_ADDRSTRLEN];
1148 #endif
1149 
1150 			if (!arp_proxyall) {
1151 				m_freem(m);
1152 				return;
1153 			}
1154 
1155 			bzero(&sin, sizeof sin);
1156 			sin.sin_family = AF_INET;
1157 			sin.sin_len = sizeof sin;
1158 			sin.sin_addr.s_addr = taddr;
1159 
1160 			rt = rtpurelookup((struct sockaddr *)&sin);
1161 			if (rt == NULL) {
1162 				m_freem(m);
1163 				return;
1164 			}
1165 			--rt->rt_refcnt;
1166 			/*
1167 			 * Don't send proxies for nodes on the same interface
1168 			 * as this one came out of, or we'll get into a fight
1169 			 * over who claims what Ether address.
1170 			 */
1171 			if (rt->rt_ifp == ifp) {
1172 				m_freem(m);
1173 				return;
1174 			}
1175 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1176 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1177 #ifdef DEBUG_PROXY
1178 			kprintf("arp: proxying for %s\n",
1179 			    kinet_ntoa(itaddr, tbuf));
1180 #endif
1181 		} else {
1182 			struct sockaddr_dl *sdl;
1183 
1184 			rt = la->la_rt;
1185 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1186 			sdl = SDL(rt->rt_gateway);
1187 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1188 		}
1189 	}
1190 
1191 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1192 	memcpy(ar_spa(ah), &taddr, ah->ar_pln);
1193 	ah->ar_op = htons(ARPOP_REPLY);
1194 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1195 	switch (ifp->if_type) {
1196 	case IFT_ETHER:
1197 		/*
1198 		 * May not be correct for types not explictly
1199 		 * listed, but it is our best guess.
1200 		 */
1201 	default:
1202 		eh = (struct ether_header *)sa.sa_data;
1203 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1204 		eh->ether_type = htons(ETHERTYPE_ARP);
1205 		break;
1206 	}
1207 	sa.sa_family = AF_UNSPEC;
1208 	sa.sa_len = sizeof sa;
1209 	ifp->if_output(ifp, m, &sa, NULL);
1210 }
1211 
1212 #endif	/* INET */
1213 
1214 /*
1215  * Free an arp entry.  If the arp entry is actively referenced or represents
1216  * a static entry we only clear it back to an unresolved state, otherwise
1217  * we destroy the entry entirely.
1218  *
1219  * Note that static entries are created when route add ... -interface is used
1220  * to create an interface route to a (direct) destination.
1221  */
1222 static void
1223 arptfree(struct llinfo_arp *la)
1224 {
1225 	struct rtentry *rt = la->la_rt;
1226 	struct sockaddr_dl *sdl;
1227 
1228 	if (rt == NULL)
1229 		panic("arptfree");
1230 	sdl = SDL(rt->rt_gateway);
1231 	if (sdl != NULL &&
1232 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1233 	     (rt->rt_flags & RTF_STATIC))) {
1234 		sdl->sdl_alen = 0;
1235 		la->la_preempt = la->la_asked = 0;
1236 		rt->rt_flags &= ~RTF_REJECT;
1237 		return;
1238 	}
1239 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1240 }
1241 
1242 /*
1243  * Lookup or enter a new address in arptab.
1244  */
1245 static struct llinfo_arp *
1246 arplookup(in_addr_t addr, boolean_t create, boolean_t generate_report,
1247 	  boolean_t proxy)
1248 {
1249 	struct rtentry *rt;
1250 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1251 	const char *why = NULL;
1252 
1253 	/* Check ARP probes, e.g. from Cisco switches. */
1254 	if (addr == INADDR_ANY && arp_ignore_probes)
1255 		return (NULL);
1256 
1257 	sin.sin_addr.s_addr = addr;
1258 	sin.sin_other = proxy ? SIN_PROXY : 0;
1259 	if (create) {
1260 		rt = _rtlookup((struct sockaddr *)&sin,
1261 			       generate_report, RTL_DOCLONE);
1262 	} else {
1263 		rt = rtpurelookup((struct sockaddr *)&sin);
1264 	}
1265 	if (rt == NULL)
1266 		return (NULL);
1267 	rt->rt_refcnt--;
1268 
1269 	if (rt->rt_flags & RTF_GATEWAY)
1270 		why = "host is not on local network";
1271 	else if (!(rt->rt_flags & RTF_LLINFO))
1272 		why = "could not allocate llinfo";
1273 	else if (rt->rt_gateway->sa_family != AF_LINK)
1274 		why = "gateway route is not ours";
1275 
1276 	if (why) {
1277 		if (create && log_arp_creation_failure) {
1278 			char abuf[INET_ADDRSTRLEN];
1279 
1280 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1281 			    kinet_ntoa(sin.sin_addr, abuf), why);
1282 		}
1283 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1284 			/* No references to this route.  Purge it. */
1285 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1286 				  rt_mask(rt), rt->rt_flags, NULL);
1287 		}
1288 		return (NULL);
1289 	}
1290 	return (rt->rt_llinfo);
1291 }
1292 
1293 void
1294 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1295 {
1296 	ifa->ifa_rtrequest = arp_rtrequest;
1297 	ifa->ifa_flags |= RTF_CLONING;
1298 }
1299 
1300 void
1301 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1302 {
1303 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1304 		arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1305 				 &IA_SIN(ifa)->sin_addr, NULL);
1306 	}
1307 }
1308 
1309 static void
1310 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1311     enum ifaddr_event event, struct ifaddr *ifa)
1312 {
1313 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1314 		return;
1315 	if (ifa->ifa_addr->sa_family != AF_INET)
1316 		return;
1317 	if (event == IFADDR_EVENT_DELETE)
1318 		return;
1319 
1320 	/*
1321 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1322 	 * - If we are the CARP interface's parent, don't send gratuitous
1323 	 *   ARP to avoid unnecessary confusion.
1324 	 */
1325 #ifdef CARP
1326 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1327 #endif
1328 	{
1329 		arp_gratuitous(ifp, ifa);
1330 	}
1331 }
1332 
1333 static void
1334 arp_init(void)
1335 {
1336 	int cpu;
1337 
1338 	for (cpu = 0; cpu < ncpus; cpu++) {
1339 		struct arp_pcpu_data *ad = &arp_data[cpu];
1340 
1341 		LIST_INIT(&ad->llinfo_list);
1342 		netmsg_init(&ad->timer_nmsg, NULL, &netisr_adone_rport,
1343 		    MSGF_PRIORITY, arptimer_dispatch);
1344 		callout_init_mp(&ad->timer_ch);
1345 
1346 		callout_reset_bycpu(&ad->timer_ch, hz, arptimer, NULL, cpu);
1347 	}
1348 
1349 	netisr_register(NETISR_ARP, arpintr, NULL);
1350 
1351 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1352 	    EVENTHANDLER_PRI_LAST);
1353 }
1354 
1355 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1356