xref: /dragonfly/sys/netinet/if_ether.c (revision 3170ffd7)
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  * 4. 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 	struct	lwkt_port *la_msgport; /* last packet's msgport */
131 	u_short	la_preempt;	/* countdown for pre-expiry arps */
132 	u_short	la_asked;	/* #times we QUERIED following expiration */
133 };
134 
135 static	LIST_HEAD(, llinfo_arp) llinfo_arp_list[MAXCPU];
136 
137 static int	arp_maxtries = 5;
138 static int	useloopback = 1; /* use loopback interface for local traffic */
139 static int	arp_proxyall = 0;
140 static int	arp_refresh = 60; /* refresh arp cache ~60 (not impl yet) */
141 static int	arp_restricted_match = 0;
142 
143 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_RW,
144 	   &arp_maxtries, 0, "ARP resolution attempts before returning error");
145 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, useloopback, CTLFLAG_RW,
146 	   &useloopback, 0, "Use the loopback interface for local traffic");
147 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_RW,
148 	   &arp_proxyall, 0, "Enable proxy ARP for all suitable requests");
149 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, restricted_match, CTLFLAG_RW,
150 	   &arp_restricted_match, 0, "Only match against the sender");
151 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, refresh, CTLFLAG_RW,
152 	   &arp_refresh, 0, "Preemptively refresh the ARP");
153 
154 static void	arp_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
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 static struct callout	arptimer_ch[MAXCPU];
172 
173 /*
174  * Timeout routine.  Age arp_tab entries periodically.
175  */
176 /* ARGSUSED */
177 static void
178 arptimer(void *ignored_arg)
179 {
180 	struct llinfo_arp *la, *nla;
181 
182 	crit_enter();
183 	LIST_FOREACH_MUTABLE(la, &llinfo_arp_list[mycpuid], la_le, nla) {
184 		if (la->la_rt->rt_expire && la->la_rt->rt_expire <= time_second)
185 			arptfree(la);
186 	}
187 	callout_reset(&arptimer_ch[mycpuid], arpt_prune * hz, arptimer, NULL);
188 	crit_exit();
189 }
190 
191 /*
192  * Parallel to llc_rtrequest.
193  *
194  * Called after a route is successfully added to the tree to fix-up the
195  * route and initiate arp operations if required.
196  */
197 static void
198 arp_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
199 {
200 	struct sockaddr *gate = rt->rt_gateway;
201 	struct llinfo_arp *la = rt->rt_llinfo;
202 
203 	struct sockaddr_dl null_sdl = { sizeof null_sdl, AF_LINK };
204 	static boolean_t arpinit_done[MAXCPU];
205 
206 	if (!arpinit_done[mycpuid]) {
207 		arpinit_done[mycpuid] = TRUE;
208 		callout_init(&arptimer_ch[mycpuid]);
209 		callout_reset(&arptimer_ch[mycpuid], hz, arptimer, NULL);
210 	}
211 	if (rt->rt_flags & RTF_GATEWAY)
212 		return;
213 
214 	switch (req) {
215 	case RTM_ADD:
216 		/*
217 		 * XXX: If this is a manually added route to interface
218 		 * such as older version of routed or gated might provide,
219 		 * restore cloning bit.
220 		 */
221 		if (!(rt->rt_flags & RTF_HOST) &&
222 		    SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
223 			rt->rt_flags |= RTF_CLONING;
224 		if (rt->rt_flags & RTF_CLONING) {
225 			/*
226 			 * Case 1: This route should come from a route to iface.
227 			 */
228 			rt_setgate(rt, rt_key(rt),
229 				   (struct sockaddr *)&null_sdl,
230 				   RTL_DONTREPORT);
231 			gate = rt->rt_gateway;
232 			SDL(gate)->sdl_type = rt->rt_ifp->if_type;
233 			SDL(gate)->sdl_index = rt->rt_ifp->if_index;
234 			rt->rt_expire = time_second;
235 			break;
236 		}
237 		/* Announce a new entry if requested. */
238 		if (rt->rt_flags & RTF_ANNOUNCE) {
239 			arprequest_async(rt->rt_ifp,
240 			    &SIN(rt_key(rt))->sin_addr,
241 			    &SIN(rt_key(rt))->sin_addr,
242 			    LLADDR(SDL(gate)));
243 		}
244 		/*FALLTHROUGH*/
245 	case RTM_RESOLVE:
246 		if (gate->sa_family != AF_LINK ||
247 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
248 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
249 			break;
250 		}
251 		SDL(gate)->sdl_type = rt->rt_ifp->if_type;
252 		SDL(gate)->sdl_index = rt->rt_ifp->if_index;
253 		if (la != NULL)
254 			break; /* This happens on a route change */
255 		/*
256 		 * Case 2:  This route may come from cloning, or a manual route
257 		 * add with a LL address.
258 		 */
259 		R_Malloc(la, struct llinfo_arp *, sizeof *la);
260 		rt->rt_llinfo = la;
261 		if (la == NULL) {
262 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
263 			break;
264 		}
265 		bzero(la, sizeof *la);
266 		la->la_rt = rt;
267 		rt->rt_flags |= RTF_LLINFO;
268 		LIST_INSERT_HEAD(&llinfo_arp_list[mycpuid], la, la_le);
269 
270 #ifdef INET
271 		/*
272 		 * This keeps the multicast addresses from showing up
273 		 * in `arp -a' listings as unresolved.  It's not actually
274 		 * functional.  Then the same for broadcast.
275 		 */
276 		if (IN_MULTICAST(ntohl(SIN(rt_key(rt))->sin_addr.s_addr))) {
277 			ETHER_MAP_IP_MULTICAST(&SIN(rt_key(rt))->sin_addr,
278 					       LLADDR(SDL(gate)));
279 			SDL(gate)->sdl_alen = 6;
280 			rt->rt_expire = 0;
281 		}
282 		if (in_broadcast(SIN(rt_key(rt))->sin_addr, rt->rt_ifp)) {
283 			memcpy(LLADDR(SDL(gate)), rt->rt_ifp->if_broadcastaddr,
284 			       rt->rt_ifp->if_addrlen);
285 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
286 			rt->rt_expire = 0;
287 		}
288 #endif
289 
290 		/*
291 		 * This fixes up the routing interface for local addresses.
292 		 * The route is adjusted to point at lo0 and the expiration
293 		 * timer is disabled.
294 		 *
295 		 * NOTE: This prevents locally targetted traffic from going
296 		 *	 out the hardware interface, which is inefficient
297 		 *	 and might not work if the hardware cannot listen
298 		 *	 to its own transmitted packets.   Setting
299 		 *	 net.link.ether.inet.useloopback to 0 will force
300 		 *	 packets for local addresses out the hardware (and
301 		 *	 it is expected to receive its own packet).
302 		 *
303 		 * XXX We should just be able to test RTF_LOCAL here instead
304 		 *     of having to compare IPs.
305 		 */
306 		if (SIN(rt_key(rt))->sin_addr.s_addr ==
307 		    (IA_SIN(rt->rt_ifa))->sin_addr.s_addr) {
308 			rt->rt_expire = 0;
309 			bcopy(IF_LLADDR(rt->rt_ifp), LLADDR(SDL(gate)),
310 			      SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen);
311 			if (useloopback)
312 				rt->rt_ifp = loif;
313 		}
314 		break;
315 
316 	case RTM_DELETE:
317 		if (la == NULL)
318 			break;
319 		LIST_REMOVE(la, la_le);
320 		rt->rt_llinfo = NULL;
321 		rt->rt_flags &= ~RTF_LLINFO;
322 		if (la->la_hold != NULL)
323 			m_freem(la->la_hold);
324 		Free(la);
325 		break;
326 	}
327 }
328 
329 static struct mbuf *
330 arpreq_alloc(struct ifnet *ifp, const struct in_addr *sip,
331 	     const struct in_addr *tip, const u_char *enaddr)
332 {
333 	struct mbuf *m;
334 	struct arphdr *ah;
335 	u_short ar_hrd;
336 
337 	if ((m = m_gethdr(MB_DONTWAIT, MT_DATA)) == NULL)
338 		return NULL;
339 	m->m_pkthdr.rcvif = NULL;
340 
341 	switch (ifp->if_type) {
342 	case IFT_ETHER:
343 		/*
344 		 * This may not be correct for types not explicitly
345 		 * listed, but this is our best guess
346 		 */
347 	default:
348 		ar_hrd = htons(ARPHRD_ETHER);
349 
350 		m->m_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
351 		m->m_pkthdr.len = m->m_len;
352 		MH_ALIGN(m, m->m_len);
353 
354 		ah = mtod(m, struct arphdr *);
355 		break;
356 	}
357 
358 	ah->ar_hrd = ar_hrd;
359 	ah->ar_pro = htons(ETHERTYPE_IP);
360 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
361 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
362 	ah->ar_op = htons(ARPOP_REQUEST);
363 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
364 	memset(ar_tha(ah), 0, ah->ar_hln);
365 	memcpy(ar_spa(ah), sip, ah->ar_pln);
366 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
367 
368 	return m;
369 }
370 
371 static void
372 arpreq_send(struct ifnet *ifp, struct mbuf *m)
373 {
374 	struct sockaddr sa;
375 	struct ether_header *eh;
376 
377 	switch (ifp->if_type) {
378 	case IFT_ETHER:
379 		/*
380 		 * This may not be correct for types not explicitly
381 		 * listed, but this is our best guess
382 		 */
383 	default:
384 		eh = (struct ether_header *)sa.sa_data;
385 		/* if_output() will not swap */
386 		eh->ether_type = htons(ETHERTYPE_ARP);
387 		memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ifp->if_addrlen);
388 		break;
389 	}
390 
391 	sa.sa_family = AF_UNSPEC;
392 	sa.sa_len = sizeof(sa);
393 	ifp->if_output(ifp, m, &sa, NULL);
394 }
395 
396 static void
397 arpreq_send_handler(netmsg_t msg)
398 {
399 	struct mbuf *m = msg->packet.nm_packet;
400 	struct ifnet *ifp = msg->lmsg.u.ms_resultp;
401 
402 	arpreq_send(ifp, m);
403 	/* nmsg was embedded in the mbuf, do not reply! */
404 }
405 
406 /*
407  * Broadcast an ARP request. Caller specifies:
408  *	- arp header source ip address
409  *	- arp header target ip address
410  *	- arp header source ethernet address
411  *
412  * NOTE: Caller MUST NOT hold ifp's serializer
413  */
414 static void
415 arprequest(struct ifnet *ifp, const struct in_addr *sip,
416 	   const struct in_addr *tip, const u_char *enaddr)
417 {
418 	struct mbuf *m;
419 
420 	if (enaddr == NULL) {
421 		if (ifp->if_bridge) {
422 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
423 		} else {
424 			enaddr = IF_LLADDR(ifp);
425 		}
426 	}
427 
428 	m = arpreq_alloc(ifp, sip, tip, enaddr);
429 	if (m == NULL)
430 		return;
431 	arpreq_send(ifp, m);
432 }
433 
434 /*
435  * Same as arprequest(), except:
436  * - Caller is allowed to hold ifp's serializer
437  * - Network output is done in protocol thead
438  */
439 static void
440 arprequest_async(struct ifnet *ifp, const struct in_addr *sip,
441 		 const struct in_addr *tip, const u_char *enaddr)
442 {
443 	struct mbuf *m;
444 	struct netmsg_packet *pmsg;
445 
446 	if (enaddr == NULL) {
447 		if (ifp->if_bridge) {
448 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
449 		} else {
450 			enaddr = IF_LLADDR(ifp);
451 		}
452 	}
453 	m = arpreq_alloc(ifp, sip, tip, enaddr);
454 	if (m == NULL)
455 		return;
456 
457 	pmsg = &m->m_hdr.mh_netmsg;
458 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
459 		    0, arpreq_send_handler);
460 	pmsg->nm_packet = m;
461 	pmsg->base.lmsg.u.ms_resultp = ifp;
462 
463 	lwkt_sendmsg(netisr_cpuport(mycpuid), &pmsg->base.lmsg);
464 }
465 
466 /*
467  * Resolve an IP address into an ethernet address.  If success,
468  * desten is filled in.  If there is no entry in arptab,
469  * set one up and broadcast a request for the IP address.
470  * Hold onto this mbuf and resend it once the address
471  * is finally resolved.  A return value of 1 indicates
472  * that desten has been filled in and the packet should be sent
473  * normally; a 0 return indicates that the packet has been
474  * taken over here, either now or for later transmission.
475  */
476 int
477 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
478 	   struct sockaddr *dst, u_char *desten)
479 {
480 	struct rtentry *rt = NULL;
481 	struct llinfo_arp *la = NULL;
482 	struct sockaddr_dl *sdl;
483 
484 	if (m->m_flags & M_BCAST) {	/* broadcast */
485 		memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen);
486 		return (1);
487 	}
488 	if (m->m_flags & M_MCAST) {/* multicast */
489 		ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
490 		return (1);
491 	}
492 	if (rt0 != NULL) {
493 		if (rt_llroute(dst, rt0, &rt) != 0) {
494 			m_freem(m);
495 			return 0;
496 		}
497 		la = rt->rt_llinfo;
498 	}
499 	if (la == NULL) {
500 		la = arplookup(SIN(dst)->sin_addr.s_addr,
501 			       TRUE, RTL_REPORTMSG, FALSE);
502 		if (la != NULL)
503 			rt = la->la_rt;
504 	}
505 	if (la == NULL || rt == NULL) {
506 		log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s%s%s\n",
507 		    inet_ntoa(SIN(dst)->sin_addr), la ? "la" : " ",
508 		    rt ? "rt" : "");
509 		m_freem(m);
510 		return (0);
511 	}
512 	sdl = SDL(rt->rt_gateway);
513 	/*
514 	 * Check the address family and length is valid, the address
515 	 * is resolved; otherwise, try to resolve.
516 	 */
517 	if ((rt->rt_expire == 0 || rt->rt_expire > time_second) &&
518 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
519 		/*
520 		 * If entry has an expiry time and it is approaching,
521 		 * see if we need to send an ARP request within this
522 		 * arpt_down interval.
523 		 */
524 		if ((rt->rt_expire != 0) &&
525 		    (time_second + la->la_preempt > rt->rt_expire)) {
526 			arprequest(ifp,
527 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
528 				   &SIN(dst)->sin_addr,
529 				   NULL);
530 			la->la_preempt--;
531 		}
532 
533 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
534 		return 1;
535 	}
536 	/*
537 	 * If ARP is disabled or static on this interface, stop.
538 	 * XXX
539 	 * Probably should not allocate empty llinfo struct if we are
540 	 * not going to be sending out an arp request.
541 	 */
542 	if (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) {
543 		m_freem(m);
544 		return (0);
545 	}
546 	/*
547 	 * There is an arptab entry, but no ethernet address
548 	 * response yet.  Replace the held mbuf with this
549 	 * latest one.
550 	 */
551 	if (la->la_hold != NULL)
552 		m_freem(la->la_hold);
553 	la->la_hold = m;
554 	la->la_msgport = netisr_curport();
555 	if (rt->rt_expire || ((rt->rt_flags & RTF_STATIC) && !sdl->sdl_alen)) {
556 		rt->rt_flags &= ~RTF_REJECT;
557 		if (la->la_asked == 0 || rt->rt_expire != time_second) {
558 			rt->rt_expire = time_second;
559 			if (la->la_asked++ < arp_maxtries) {
560 				arprequest(ifp,
561 					   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
562 					   &SIN(dst)->sin_addr,
563 					   NULL);
564 			} else {
565 				rt->rt_flags |= RTF_REJECT;
566 				rt->rt_expire += arpt_down;
567 				la->la_asked = 0;
568 				la->la_preempt = arp_maxtries;
569 			}
570 		}
571 	}
572 	return (0);
573 }
574 
575 /*
576  * Common length and type checks are done here,
577  * then the protocol-specific routine is called.
578  */
579 static void
580 arpintr(netmsg_t msg)
581 {
582 	struct mbuf *m = msg->packet.nm_packet;
583 	struct arphdr *ar;
584 	u_short ar_hrd;
585 	char hexstr[6];
586 
587 	if (m->m_len < sizeof(struct arphdr) &&
588 	    (m = m_pullup(m, sizeof(struct arphdr))) == NULL) {
589 		log(LOG_ERR, "arp: runt packet -- m_pullup failed\n");
590 		return;
591 	}
592 	ar = mtod(m, struct arphdr *);
593 
594 	ar_hrd = ntohs(ar->ar_hrd);
595 	if (ar_hrd != ARPHRD_ETHER && ar_hrd != ARPHRD_IEEE802) {
596 		hexncpy((unsigned char *)&ar->ar_hrd, 2, hexstr, 5, NULL);
597 		log(LOG_ERR, "arp: unknown hardware address format (0x%s)\n",
598 		    hexstr);
599 		m_freem(m);
600 		return;
601 	}
602 
603 	if (m->m_pkthdr.len < arphdr_len(ar)) {
604 		if ((m = m_pullup(m, arphdr_len(ar))) == NULL) {
605 			log(LOG_ERR, "arp: runt packet\n");
606 			return;
607 		}
608 		ar = mtod(m, struct arphdr *);
609 	}
610 
611 	switch (ntohs(ar->ar_pro)) {
612 #ifdef INET
613 	case ETHERTYPE_IP:
614 		in_arpinput(m);
615 		return;
616 #endif
617 	}
618 	m_freem(m);
619 	/* msg was embedded in the mbuf, do not reply! */
620 }
621 
622 #ifdef INET
623 /*
624  * ARP for Internet protocols on 10 Mb/s Ethernet.
625  * Algorithm is that given in RFC 826.
626  * In addition, a sanity check is performed on the sender
627  * protocol address, to catch impersonators.
628  * We no longer handle negotiations for use of trailer protocol:
629  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
630  * along with IP replies if we wanted trailers sent to us,
631  * and also sent them in response to IP replies.
632  * This allowed either end to announce the desire to receive
633  * trailer packets.
634  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
635  * but formerly didn't normally send requests.
636  */
637 
638 static int	log_arp_wrong_iface = 1;
639 static int	log_arp_movements = 1;
640 static int	log_arp_permanent_modify = 1;
641 
642 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
643 	   &log_arp_wrong_iface, 0,
644 	   "Log arp packets arriving on the wrong interface");
645 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
646 	   &log_arp_movements, 0,
647 	   "Log arp replies from MACs different than the one in the cache");
648 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
649 	   &log_arp_permanent_modify, 0,
650 	   "Log arp replies from MACs different than the one "
651 	   "in the permanent arp entry");
652 
653 
654 static void
655 arp_hold_output(netmsg_t msg)
656 {
657 	struct mbuf *m = msg->packet.nm_packet;
658 	struct rtentry *rt;
659 	struct ifnet *ifp;
660 
661 	rt = msg->lmsg.u.ms_resultp;
662 	ifp = m->m_pkthdr.rcvif;
663 	m->m_pkthdr.rcvif = NULL;
664 
665 	ifp->if_output(ifp, m, rt_key(rt), rt);
666 
667 	/* Drop the reference count bumped by the sender */
668 	RTFREE(rt);
669 
670 	/* nmsg was embedded in the mbuf, do not reply! */
671 }
672 
673 static void
674 arp_update_oncpu(struct mbuf *m, in_addr_t saddr, boolean_t create,
675 		 boolean_t generate_report, boolean_t dologging)
676 {
677 	struct arphdr *ah = mtod(m, struct arphdr *);
678 	struct ifnet *ifp = m->m_pkthdr.rcvif;
679 	struct llinfo_arp *la;
680 	struct sockaddr_dl *sdl;
681 	struct rtentry *rt;
682 	char hexstr[2][64];
683 
684 	la = arplookup(saddr, create, generate_report, FALSE);
685 	if (la && (rt = la->la_rt) && (sdl = SDL(rt->rt_gateway))) {
686 		struct in_addr isaddr = { saddr };
687 
688 		/*
689 		 * Normally arps coming in on the wrong interface are ignored,
690 		 * but if we are bridging and the two interfaces belong to
691 		 * the same bridge, or one is a member of the bridge which
692 		 * is the other, then it isn't an error.
693 		 */
694 		if (rt->rt_ifp != ifp) {
695 			/*
696 			 * (1) ifp and rt_ifp both members of same bridge
697 			 * (2) rt_ifp member of bridge ifp
698 			 * (3) ifp member of bridge rt_ifp
699 			 *
700 			 * Always replace rt_ifp with the bridge ifc.
701 			 */
702 			struct ifnet *nifp;
703 
704 			if (ifp->if_bridge &&
705 			    rt->rt_ifp->if_bridge == ifp->if_bridge) {
706 				nifp = ether_bridge_interface(ifp);
707 			} else if (rt->rt_ifp->if_bridge &&
708 				   ether_bridge_interface(rt->rt_ifp) == ifp) {
709 				nifp = ifp;
710 			} else if (ifp->if_bridge &&
711 				   ether_bridge_interface(ifp) == rt->rt_ifp) {
712 				nifp = rt->rt_ifp;
713 			} else {
714 				nifp = NULL;
715 			}
716 
717 			if ((log_arp_wrong_iface == 1 && nifp == NULL) ||
718 			    log_arp_wrong_iface == 2) {
719 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
720 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
721 				log(LOG_ERR,
722 				    "arp: %s is on %s "
723 				    "but got reply from %s on %s\n",
724 				    inet_ntoa(isaddr),
725 				    rt->rt_ifp->if_xname, hexstr[0],
726 				    ifp->if_xname);
727 			}
728 			if (nifp == NULL)
729 				return;
730 
731 			/*
732 			 * nifp is our man!  Replace rt_ifp and adjust
733 			 * the sdl.
734 			 */
735 			ifp = rt->rt_ifp = nifp;
736 			sdl->sdl_type = ifp->if_type;
737 			sdl->sdl_index = ifp->if_index;
738 		}
739 		if (sdl->sdl_alen &&
740 		    bcmp(ar_sha(ah), LLADDR(sdl), sdl->sdl_alen)) {
741 			if (rt->rt_expire != 0) {
742 				if (dologging && log_arp_movements) {
743 					hexncpy((u_char *)LLADDR(sdl), ifp->if_addrlen,
744 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
745 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
746 					    hexstr[1], HEX_NCPYLEN(ifp->if_addrlen), ":");
747 			    		log(LOG_INFO,
748 					    "arp: %s moved from %s to %s on %s\n",
749 					    inet_ntoa(isaddr), hexstr[0], hexstr[1],
750 					    ifp->if_xname);
751 				}
752 			} else {
753 				if (dologging && log_arp_permanent_modify) {
754 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
755 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
756 					log(LOG_ERR,
757 					"arp: %s attempts to modify "
758 					"permanent entry for %s on %s\n",
759 					hexstr[0], inet_ntoa(isaddr), ifp->if_xname);
760 				}
761 				return;
762 			}
763 		}
764 		/*
765 		 * sanity check for the address length.
766 		 * XXX this does not work for protocols with variable address
767 		 * length. -is
768 		 */
769 		if (dologging && sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
770 			hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
771 			    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
772 			log(LOG_WARNING,
773 			    "arp from %s: new addr len %d, was %d",
774 			    hexstr[0], ah->ar_hln, sdl->sdl_alen);
775 		}
776 		if (ifp->if_addrlen != ah->ar_hln) {
777 			if (dologging) {
778 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
779 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
780 				log(LOG_WARNING,
781 				"arp from %s: addr len: new %d, i/f %d "
782 				"(ignored)", hexstr[0],
783 				ah->ar_hln, ifp->if_addrlen);
784 			}
785 			return;
786 		}
787 		memcpy(LLADDR(sdl), ar_sha(ah), sdl->sdl_alen = ah->ar_hln);
788 		if (rt->rt_expire != 0) {
789 			rt->rt_expire = time_second + arpt_keep;
790 		}
791 		rt->rt_flags &= ~RTF_REJECT;
792 		la->la_asked = 0;
793 		la->la_preempt = arp_maxtries;
794 
795 		/*
796 		 * This particular cpu might have been holding an mbuf
797 		 * pending ARP resolution.  If so, transmit the mbuf now.
798 		 */
799 		if (la->la_hold != NULL) {
800 			struct mbuf *m = la->la_hold;
801 			struct lwkt_port *port = la->la_msgport;
802 			struct netmsg_packet *pmsg;
803 
804 			la->la_hold = NULL;
805 			la->la_msgport = NULL;
806 
807 			m_adj(m, sizeof(struct ether_header));
808 
809 			/*
810 			 * Make sure that this rtentry will not be freed
811 			 * before the packet is processed on the target
812 			 * msgport.  The reference count will be dropped
813 			 * in the handler associated with this packet.
814 			 */
815 			rt->rt_refcnt++;
816 
817 			pmsg = &m->m_hdr.mh_netmsg;
818 			netmsg_init(&pmsg->base, NULL,
819 				    &netisr_apanic_rport,
820 				    MSGF_PRIORITY, arp_hold_output);
821 			pmsg->nm_packet = m;
822 
823 			/* Record necessary information */
824 			m->m_pkthdr.rcvif = ifp;
825 			pmsg->base.lmsg.u.ms_resultp = rt;
826 
827 			lwkt_sendmsg(port, &pmsg->base.lmsg);
828 		}
829 	}
830 }
831 
832 /*
833  * Called from arpintr() - this routine is run from a single cpu.
834  */
835 static void
836 in_arpinput(struct mbuf *m)
837 {
838 	struct arphdr *ah;
839 	struct ifnet *ifp = m->m_pkthdr.rcvif;
840 	struct ifaddr_container *ifac;
841 	struct in_ifaddr_container *iac;
842 	struct in_ifaddr *ia = NULL;
843 	struct in_addr isaddr, itaddr, myaddr;
844 	struct netmsg_inarp *msg;
845 	uint8_t *enaddr = NULL;
846 	int req_len;
847 	char hexstr[64];
848 
849 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
850 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
851 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
852 		return;
853 	}
854 
855 	ah = mtod(m, struct arphdr *);
856 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
857 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
858 
859 	/*
860 	 * Check both target and sender IP addresses:
861 	 *
862 	 * If we receive the packet on the interface owning the address,
863 	 * then accept the address.
864 	 *
865 	 * For a bridge, we accept the address if the receive interface and
866 	 * the interface owning the address are on the same bridge, and
867 	 * use the bridge MAC as the is-at response.  The bridge will be
868 	 * responsible for handling the packet.
869 	 *
870 	 * (0) Check target IP against CARP IPs
871 	 */
872 #ifdef CARP
873 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
874 		int is_match = 0, is_parent = 0;
875 
876 		ia = iac->ia;
877 
878 		/* Skip all ia's which don't match */
879 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
880 			continue;
881 
882 		if (ia->ia_ifp->if_type != IFT_CARP)
883 			continue;
884 
885 		if (carp_parent(ia->ia_ifp) == ifp)
886 			is_parent = 1;
887 		if (is_parent || ia->ia_ifp == ifp)
888 			is_match = carp_iamatch(ia);
889 
890 		if (is_match) {
891 			if (is_parent) {
892 				/*
893 				 * The parent interface will also receive
894 				 * the ethernet broadcast packets, e.g. ARP
895 				 * REQUEST, so if we could find a CARP
896 				 * interface of the parent that could match
897 				 * the target IP address, we then drop the
898 				 * packets, which is delieverd to us through
899 				 * the parent interface.
900 				 */
901 				m_freem(m);
902 				return;
903 			}
904 			goto match;
905 		}
906 	}
907 #endif	/* CARP */
908 
909 	/*
910 	 * (1) Check target IP against our local IPs
911 	 */
912 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
913 		ia = iac->ia;
914 
915 		/* Skip all ia's which don't match */
916 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
917 			continue;
918 
919 #ifdef CARP
920 		/* CARP interfaces are checked in (0) */
921 		if (ia->ia_ifp->if_type == IFT_CARP)
922 			continue;
923 #endif
924 
925 		if (ifp->if_bridge && ia->ia_ifp &&
926 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
927 			ifp = ether_bridge_interface(ifp);
928 			goto match;
929 		}
930 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
931 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
932 			goto match;
933 		}
934 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
935 		    ia->ia_ifp) {
936 			goto match;
937 		}
938 		if (ia->ia_ifp == ifp) {
939 			goto match;
940 		}
941 	}
942 
943 	/*
944 	 * (2) Check sender IP against our local IPs
945 	 */
946 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
947 		ia = iac->ia;
948 
949 		/* Skip all ia's which don't match */
950 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
951 			continue;
952 
953 		if (ifp->if_bridge && ia->ia_ifp &&
954 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
955 			ifp = ether_bridge_interface(ifp);
956 			goto match;
957 		}
958 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
959 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
960 			goto match;
961 		}
962 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
963 		    ia->ia_ifp) {
964 			goto match;
965 		}
966 
967 		if (ia->ia_ifp == ifp)
968 			goto match;
969 	}
970 
971 	/*
972 	 * No match, use the first inet address on the receive interface
973 	 * as a dummy address for the rest of the function.
974 	 */
975 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
976 		struct ifaddr *ifa = ifac->ifa;
977 
978 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
979 			ia = ifatoia(ifa);
980 			goto match;
981 		}
982 	}
983 
984 	/*
985 	 * If we got here, we didn't find any suitable interface,
986 	 * so drop the packet.
987 	 */
988 	m_freem(m);
989 	return;
990 
991 match:
992 	if (!enaddr)
993 		enaddr = (uint8_t *)IF_LLADDR(ifp);
994 	myaddr = ia->ia_addr.sin_addr;
995 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
996 		m_freem(m);	/* it's from me, ignore it. */
997 		return;
998 	}
999 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1000 		log(LOG_ERR,
1001 		    "arp: link address is broadcast for IP address %s!\n",
1002 		    inet_ntoa(isaddr));
1003 		m_freem(m);
1004 		return;
1005 	}
1006 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1007 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1008 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1009 		log(LOG_ERR,
1010 		   "arp: %s is using my IP address %s!\n",
1011 		    hexstr, inet_ntoa(isaddr));
1012 		itaddr = myaddr;
1013 		goto reply;
1014 	}
1015 	if (ifp->if_flags & IFF_STATICARP)
1016 		goto reply;
1017 
1018 	/*
1019 	 * When arp_restricted_match is true and the ARP response is not
1020 	 * specifically targetted to me, ignore it.  Otherwise the entry
1021 	 * timeout may be updated for an old MAC.
1022 	 */
1023 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1024 		m_freem(m);
1025 		return;
1026 	}
1027 
1028 	/*
1029 	 * Update all CPU's routing tables with this ARP packet
1030 	 */
1031 	msg = &m->m_hdr.mh_arpmsg;
1032 	netmsg_init(&msg->base, NULL, &netisr_apanic_rport,
1033 	    0, arp_update_msghandler);
1034 	msg->m = m;
1035 	msg->saddr = isaddr.s_addr;
1036 	msg->taddr = itaddr.s_addr;
1037 	msg->myaddr = myaddr.s_addr;
1038 	lwkt_sendmsg(rtable_portfn(0), &msg->base.lmsg);
1039 
1040 	/*
1041 	 * Just return here; after all CPUs's routing tables are
1042 	 * properly updated by this ARP packet, an ARP reply will
1043 	 * be generated if appropriate.
1044 	 */
1045 	return;
1046 reply:
1047 	in_arpreply(m, itaddr.s_addr, myaddr.s_addr);
1048 }
1049 
1050 static void
1051 arp_reply_msghandler(netmsg_t msg)
1052 {
1053 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1054 
1055 	in_arpreply(rmsg->m, rmsg->taddr, rmsg->myaddr);
1056 	/* Don't reply this netmsg; netmsg_inarp is embedded in mbuf */
1057 }
1058 
1059 static void
1060 arp_update_msghandler(netmsg_t msg)
1061 {
1062 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1063 	int nextcpu;
1064 
1065 	/*
1066 	 * This message handler will be called on all of the CPUs,
1067 	 * however, we only need to generate rtmsg on CPU0.
1068 	 */
1069 	arp_update_oncpu(rmsg->m, rmsg->saddr, rmsg->taddr == rmsg->myaddr,
1070 	    mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT, mycpuid == 0);
1071 
1072 	nextcpu = mycpuid + 1;
1073 	if (nextcpu < ncpus) {
1074 		lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
1075 	} else {
1076 		struct mbuf *m = rmsg->m;
1077 		in_addr_t saddr = rmsg->saddr;
1078 		in_addr_t taddr = rmsg->taddr;
1079 		in_addr_t myaddr = rmsg->myaddr;
1080 
1081 		/*
1082 		 * Dispatch this mbuf to netisr0 to perform ARP reply,
1083 		 * if appropriate.
1084 		 * NOTE: netmsg_inarp is embedded in this mbuf.
1085 		 */
1086 		netmsg_init(&rmsg->base, NULL, &netisr_apanic_rport,
1087 		    0, arp_reply_msghandler);
1088 		rmsg->m = m;
1089 		rmsg->saddr = saddr;
1090 		rmsg->taddr = taddr;
1091 		rmsg->myaddr = myaddr;
1092 		lwkt_sendmsg(netisr_cpuport(0), &rmsg->base.lmsg);
1093 	}
1094 }
1095 
1096 static void
1097 in_arpreply(struct mbuf *m, in_addr_t taddr, in_addr_t myaddr)
1098 {
1099 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1100 	const uint8_t *enaddr;
1101 	struct arphdr *ah;
1102 	struct sockaddr sa;
1103 	struct ether_header *eh;
1104 
1105 	ah = mtod(m, struct arphdr *);
1106 	if (ntohs(ah->ar_op) != ARPOP_REQUEST) {
1107 		m_freem(m);
1108 		return;
1109 	}
1110 
1111 	enaddr = (const uint8_t *)IF_LLADDR(ifp);
1112 	if (taddr == myaddr) {
1113 		/* I am the target */
1114 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1115 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1116 	} else {
1117 		struct llinfo_arp *la;
1118 		struct rtentry *rt;
1119 
1120 		la = arplookup(taddr, FALSE, RTL_DONTREPORT, SIN_PROXY);
1121 		if (la == NULL) {
1122 			struct sockaddr_in sin;
1123 
1124 			if (!arp_proxyall) {
1125 				m_freem(m);
1126 				return;
1127 			}
1128 
1129 			bzero(&sin, sizeof sin);
1130 			sin.sin_family = AF_INET;
1131 			sin.sin_len = sizeof sin;
1132 			sin.sin_addr.s_addr = taddr;
1133 
1134 			rt = rtpurelookup((struct sockaddr *)&sin);
1135 			if (rt == NULL) {
1136 				m_freem(m);
1137 				return;
1138 			}
1139 			--rt->rt_refcnt;
1140 			/*
1141 			 * Don't send proxies for nodes on the same interface
1142 			 * as this one came out of, or we'll get into a fight
1143 			 * over who claims what Ether address.
1144 			 */
1145 			if (rt->rt_ifp == ifp) {
1146 				m_freem(m);
1147 				return;
1148 			}
1149 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1150 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1151 #ifdef DEBUG_PROXY
1152 			kprintf("arp: proxying for %s\n", inet_ntoa(itaddr));
1153 #endif
1154 		} else {
1155 			struct sockaddr_dl *sdl;
1156 
1157 			rt = la->la_rt;
1158 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1159 			sdl = SDL(rt->rt_gateway);
1160 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1161 		}
1162 	}
1163 
1164 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1165 	memcpy(ar_spa(ah), &taddr, ah->ar_pln);
1166 	ah->ar_op = htons(ARPOP_REPLY);
1167 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1168 	switch (ifp->if_type) {
1169 	case IFT_ETHER:
1170 		/*
1171 		 * May not be correct for types not explictly
1172 		 * listed, but it is our best guess.
1173 		 */
1174 	default:
1175 		eh = (struct ether_header *)sa.sa_data;
1176 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1177 		eh->ether_type = htons(ETHERTYPE_ARP);
1178 		break;
1179 	}
1180 	sa.sa_family = AF_UNSPEC;
1181 	sa.sa_len = sizeof sa;
1182 	ifp->if_output(ifp, m, &sa, NULL);
1183 }
1184 
1185 #endif	/* INET */
1186 
1187 /*
1188  * Free an arp entry.  If the arp entry is actively referenced or represents
1189  * a static entry we only clear it back to an unresolved state, otherwise
1190  * we destroy the entry entirely.
1191  *
1192  * Note that static entries are created when route add ... -interface is used
1193  * to create an interface route to a (direct) destination.
1194  */
1195 static void
1196 arptfree(struct llinfo_arp *la)
1197 {
1198 	struct rtentry *rt = la->la_rt;
1199 	struct sockaddr_dl *sdl;
1200 
1201 	if (rt == NULL)
1202 		panic("arptfree");
1203 	sdl = SDL(rt->rt_gateway);
1204 	if (sdl != NULL &&
1205 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1206 	     (rt->rt_flags & RTF_STATIC))) {
1207 		sdl->sdl_alen = 0;
1208 		la->la_preempt = la->la_asked = 0;
1209 		rt->rt_flags &= ~RTF_REJECT;
1210 		return;
1211 	}
1212 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1213 }
1214 
1215 /*
1216  * Lookup or enter a new address in arptab.
1217  */
1218 static struct llinfo_arp *
1219 arplookup(in_addr_t addr, boolean_t create, boolean_t generate_report,
1220 	  boolean_t proxy)
1221 {
1222 	struct rtentry *rt;
1223 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1224 	const char *why = NULL;
1225 
1226 	sin.sin_addr.s_addr = addr;
1227 	sin.sin_other = proxy ? SIN_PROXY : 0;
1228 	if (create) {
1229 		rt = _rtlookup((struct sockaddr *)&sin,
1230 			       generate_report, RTL_DOCLONE);
1231 	} else {
1232 		rt = rtpurelookup((struct sockaddr *)&sin);
1233 	}
1234 	if (rt == NULL)
1235 		return (NULL);
1236 	rt->rt_refcnt--;
1237 
1238 	if (rt->rt_flags & RTF_GATEWAY)
1239 		why = "host is not on local network";
1240 	else if (!(rt->rt_flags & RTF_LLINFO))
1241 		why = "could not allocate llinfo";
1242 	else if (rt->rt_gateway->sa_family != AF_LINK)
1243 		why = "gateway route is not ours";
1244 
1245 	if (why) {
1246 		if (create) {
1247 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1248 			    inet_ntoa(sin.sin_addr), why);
1249 		}
1250 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1251 			/* No references to this route.  Purge it. */
1252 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1253 				  rt_mask(rt), rt->rt_flags, NULL);
1254 		}
1255 		return (NULL);
1256 	}
1257 	return (rt->rt_llinfo);
1258 }
1259 
1260 void
1261 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1262 {
1263 	ifa->ifa_rtrequest = arp_rtrequest;
1264 	ifa->ifa_flags |= RTF_CLONING;
1265 }
1266 
1267 void
1268 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1269 {
1270 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1271 		arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1272 				 &IA_SIN(ifa)->sin_addr, NULL);
1273 	}
1274 }
1275 
1276 static void
1277 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1278     enum ifaddr_event event, struct ifaddr *ifa)
1279 {
1280 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1281 		return;
1282 	if (ifa->ifa_addr->sa_family != AF_INET)
1283 		return;
1284 	if (event == IFADDR_EVENT_DELETE)
1285 		return;
1286 
1287 	/*
1288 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1289 	 * - If we are the CARP interface's parent, don't send gratuitous
1290 	 *   ARP to avoid unnecessary confusion.
1291 	 */
1292 #ifdef CARP
1293 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1294 #endif
1295 	{
1296 		arp_gratuitous(ifp, ifa);
1297 	}
1298 }
1299 
1300 static void
1301 arp_init(void)
1302 {
1303 	int cpu;
1304 
1305 	for (cpu = 0; cpu < ncpus2; cpu++)
1306 		LIST_INIT(&llinfo_arp_list[cpu]);
1307 
1308 	netisr_register(NETISR_ARP, arpintr, NULL);
1309 
1310 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1311 	    EVENTHANDLER_PRI_LAST);
1312 }
1313 
1314 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1315