xref: /dflybsd-src/sys/netinet/if_ether.c (revision d3c9c58ee7215956b01a6764f73d9b1f21812b6e)
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. All advertising materials mentioning features or use of this software
46  *    must display the following acknowledgement:
47  *	This product includes software developed by the University of
48  *	California, Berkeley and its contributors.
49  * 4. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
66  * $FreeBSD: src/sys/netinet/if_ether.c,v 1.64.2.23 2003/04/11 07:23:15 fjoe Exp $
67  */
68 
69 /*
70  * Ethernet address resolution protocol.
71  * TODO:
72  *	add "inuse/lock" bit (or ref. count) along with valid bit
73  */
74 
75 #include "opt_inet.h"
76 #include "opt_carp.h"
77 
78 #include <sys/param.h>
79 #include <sys/kernel.h>
80 #include <sys/queue.h>
81 #include <sys/sysctl.h>
82 #include <sys/systm.h>
83 #include <sys/mbuf.h>
84 #include <sys/malloc.h>
85 #include <sys/socket.h>
86 #include <sys/syslog.h>
87 #include <sys/lock.h>
88 
89 #include <net/if.h>
90 #include <net/if_dl.h>
91 #include <net/if_types.h>
92 #include <net/route.h>
93 #include <net/netisr.h>
94 #include <net/if_llc.h>
95 
96 #include <netinet/in.h>
97 #include <netinet/in_var.h>
98 #include <netinet/if_ether.h>
99 
100 #include <sys/thread2.h>
101 #include <sys/msgport2.h>
102 #include <net/netmsg2.h>
103 #include <sys/mplock2.h>
104 
105 #ifdef CARP
106 #include <netinet/ip_carp.h>
107 #endif
108 
109 #define SIN(s) ((struct sockaddr_in *)s)
110 #define SDL(s) ((struct sockaddr_dl *)s)
111 
112 SYSCTL_DECL(_net_link_ether);
113 SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
114 
115 /* timer values */
116 static int arpt_prune = (5*60*1); /* walk list every 5 minutes */
117 static int arpt_keep = (20*60); /* once resolved, good for 20 more minutes */
118 static int arpt_down = 20;	/* once declared down, don't send for 20 sec */
119 
120 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, prune_intvl, CTLFLAG_RW,
121 	   &arpt_prune, 0, "");
122 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_RW,
123 	   &arpt_keep, 0, "");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, host_down_time, CTLFLAG_RW,
125 	   &arpt_down, 0, "");
126 
127 #define	rt_expire	rt_rmx.rmx_expire
128 
129 struct llinfo_arp {
130 	LIST_ENTRY(llinfo_arp) la_le;
131 	struct	rtentry *la_rt;
132 	struct	mbuf *la_hold;	/* last packet until resolved/timeout */
133 	struct	lwkt_port *la_msgport; /* last packet's msgport */
134 	u_short	la_preempt;	/* countdown for pre-expiry arps */
135 	u_short	la_asked;	/* #times we QUERIED following expiration */
136 };
137 
138 static	LIST_HEAD(, llinfo_arp) llinfo_arp_list[MAXCPU];
139 
140 static int	arp_maxtries = 5;
141 static int	useloopback = 1; /* use loopback interface for local traffic */
142 static int	arp_proxyall = 0;
143 static int	arp_refresh = 60; /* refresh arp cache ~60 (not impl yet) */
144 static int	arp_restricted_match = 0;
145 
146 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_RW,
147 	   &arp_maxtries, 0, "ARP resolution attempts before returning error");
148 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, useloopback, CTLFLAG_RW,
149 	   &useloopback, 0, "Use the loopback interface for local traffic");
150 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_RW,
151 	   &arp_proxyall, 0, "Enable proxy ARP for all suitable requests");
152 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, restricted_match, CTLFLAG_RW,
153 	   &arp_restricted_match, 0, "Only match against the sender");
154 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, refresh, CTLFLAG_RW,
155 	   &arp_refresh, 0, "Preemptively refresh the ARP");
156 
157 static void	arp_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
158 static void	arprequest(struct ifnet *, const struct in_addr *,
159 			   const struct in_addr *, const u_char *);
160 static void	arprequest_async(struct ifnet *, const struct in_addr *,
161 				 const struct in_addr *, const u_char *);
162 static void	arpintr(netmsg_t msg);
163 static void	arptfree(struct llinfo_arp *);
164 static void	arptimer(void *);
165 static struct llinfo_arp *
166 		arplookup(in_addr_t, boolean_t, boolean_t, boolean_t);
167 #ifdef INET
168 static void	in_arpinput(struct mbuf *);
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_portfn(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;
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 = cur_netport();
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 struct netmsg_arp_update {
833 	struct netmsg_base base;
834 	struct mbuf	*m;
835 	in_addr_t	saddr;
836 	boolean_t	create;
837 };
838 
839 static void arp_update_msghandler(netmsg_t msg);
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 ether_header *eh;
850 	struct rtentry *rt;
851 	struct ifaddr_container *ifac;
852 	struct in_ifaddr_container *iac;
853 	struct in_ifaddr *ia = NULL;
854 	struct sockaddr sa;
855 	struct in_addr isaddr, itaddr, myaddr;
856 	struct netmsg_arp_update msg;
857 	uint8_t *enaddr = NULL;
858 	int op;
859 	int req_len;
860 	char hexstr[64];
861 
862 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
863 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
864 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
865 		return;
866 	}
867 
868 	ah = mtod(m, struct arphdr *);
869 	op = ntohs(ah->ar_op);
870 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
871 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
872 
873 	/*
874 	 * Check both target and sender IP addresses:
875 	 *
876 	 * If we receive the packet on the interface owning the address,
877 	 * then accept the address.
878 	 *
879 	 * For a bridge, we accept the address if the receive interface and
880 	 * the interface owning the address are on the same bridge, and
881 	 * use the bridge MAC as the is-at response.  The bridge will be
882 	 * responsible for handling the packet.
883 	 *
884 	 * (0) Check target IP against CARP IPs
885 	 */
886 #ifdef CARP
887 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
888 		int is_match = 0, is_parent = 0;
889 
890 		ia = iac->ia;
891 
892 		/* Skip all ia's which don't match */
893 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
894 			continue;
895 
896 		if (ia->ia_ifp->if_type != IFT_CARP)
897 			continue;
898 
899 		if (carp_parent(ia->ia_ifp) == ifp)
900 			is_parent = 1;
901 		if (is_parent || ia->ia_ifp == ifp)
902 			is_match = carp_iamatch(ia);
903 
904 		if (is_match) {
905 			if (is_parent) {
906 				/*
907 				 * The parent interface will also receive
908 				 * the ethernet broadcast packets, e.g. ARP
909 				 * REQUEST, so if we could find a CARP
910 				 * interface of the parent that could match
911 				 * the target IP address, we then drop the
912 				 * packets, which is delieverd to us through
913 				 * the parent interface.
914 				 */
915 				m_freem(m);
916 				return;
917 			}
918 			goto match;
919 		}
920 	}
921 #endif	/* CARP */
922 
923 	/*
924 	 * (1) Check target IP against our local IPs
925 	 */
926 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
927 		ia = iac->ia;
928 
929 		/* Skip all ia's which don't match */
930 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
931 			continue;
932 
933 #ifdef CARP
934 		/* CARP interfaces are checked in (0) */
935 		if (ia->ia_ifp->if_type == IFT_CARP)
936 			continue;
937 #endif
938 
939 		if (ifp->if_bridge && ia->ia_ifp &&
940 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
941 			ifp = ether_bridge_interface(ifp);
942 			goto match;
943 		}
944 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
945 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
946 			goto match;
947 		}
948 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
949 		    ia->ia_ifp) {
950 			goto match;
951 		}
952 		if (ia->ia_ifp == ifp) {
953 			goto match;
954 		}
955 	}
956 
957 	/*
958 	 * (2) Check sender IP against our local IPs
959 	 */
960 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
961 		ia = iac->ia;
962 
963 		/* Skip all ia's which don't match */
964 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
965 			continue;
966 
967 		if (ifp->if_bridge && ia->ia_ifp &&
968 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
969 			ifp = ether_bridge_interface(ifp);
970 			goto match;
971 		}
972 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
973 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
974 			goto match;
975 		}
976 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
977 		    ia->ia_ifp) {
978 			goto match;
979 		}
980 
981 		if (ia->ia_ifp == ifp)
982 			goto match;
983 	}
984 
985 	/*
986 	 * No match, use the first inet address on the receive interface
987 	 * as a dummy address for the rest of the function.
988 	 */
989 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
990 		struct ifaddr *ifa = ifac->ifa;
991 
992 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
993 			ia = ifatoia(ifa);
994 			goto match;
995 		}
996 	}
997 
998 	/*
999 	 * If we got here, we didn't find any suitable interface,
1000 	 * so drop the packet.
1001 	 */
1002 	m_freem(m);
1003 	return;
1004 
1005 match:
1006 	if (!enaddr)
1007 		enaddr = (uint8_t *)IF_LLADDR(ifp);
1008 	myaddr = ia->ia_addr.sin_addr;
1009 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
1010 		m_freem(m);	/* it's from me, ignore it. */
1011 		return;
1012 	}
1013 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1014 		log(LOG_ERR,
1015 		    "arp: link address is broadcast for IP address %s!\n",
1016 		    inet_ntoa(isaddr));
1017 		m_freem(m);
1018 		return;
1019 	}
1020 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1021 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1022 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1023 		log(LOG_ERR,
1024 		   "arp: %s is using my IP address %s!\n",
1025 		    hexstr, inet_ntoa(isaddr));
1026 		itaddr = myaddr;
1027 		goto reply;
1028 	}
1029 	if (ifp->if_flags & IFF_STATICARP)
1030 		goto reply;
1031 
1032 	/*
1033 	 * When arp_restricted_match is true and the ARP response is not
1034 	 * specifically targetted to me, ignore it.  Otherwise the entry
1035 	 * timeout may be updated for an old MAC.
1036 	 */
1037 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1038 		m_freem(m);
1039 		return;
1040 	}
1041 
1042 	netmsg_init(&msg.base, NULL, &curthread->td_msgport,
1043 		    0, arp_update_msghandler);
1044 	msg.m = m;
1045 	msg.saddr = isaddr.s_addr;
1046 	msg.create = (itaddr.s_addr == myaddr.s_addr);
1047 	lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0);
1048 reply:
1049 	if (op != ARPOP_REQUEST) {
1050 		m_freem(m);
1051 		return;
1052 	}
1053 	if (itaddr.s_addr == myaddr.s_addr) {
1054 		/* I am the target */
1055 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1056 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1057 	} else {
1058 		struct llinfo_arp *la;
1059 
1060 		la = arplookup(itaddr.s_addr, FALSE, RTL_DONTREPORT, SIN_PROXY);
1061 		if (la == NULL) {
1062 			struct sockaddr_in sin;
1063 
1064 			if (!arp_proxyall) {
1065 				m_freem(m);
1066 				return;
1067 			}
1068 
1069 			bzero(&sin, sizeof sin);
1070 			sin.sin_family = AF_INET;
1071 			sin.sin_len = sizeof sin;
1072 			sin.sin_addr = itaddr;
1073 
1074 			rt = rtpurelookup((struct sockaddr *)&sin);
1075 			if (rt == NULL) {
1076 				m_freem(m);
1077 				return;
1078 			}
1079 			--rt->rt_refcnt;
1080 			/*
1081 			 * Don't send proxies for nodes on the same interface
1082 			 * as this one came out of, or we'll get into a fight
1083 			 * over who claims what Ether address.
1084 			 */
1085 			if (rt->rt_ifp == ifp) {
1086 				m_freem(m);
1087 				return;
1088 			}
1089 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1090 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1091 #ifdef DEBUG_PROXY
1092 			kprintf("arp: proxying for %s\n", inet_ntoa(itaddr));
1093 #endif
1094 		} else {
1095 			struct sockaddr_dl *sdl;
1096 
1097 			rt = la->la_rt;
1098 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1099 			sdl = SDL(rt->rt_gateway);
1100 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1101 		}
1102 	}
1103 
1104 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1105 	memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1106 	ah->ar_op = htons(ARPOP_REPLY);
1107 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1108 	switch (ifp->if_type) {
1109 	case IFT_ETHER:
1110 		/*
1111 		 * May not be correct for types not explictly
1112 		 * listed, but it is our best guess.
1113 		 */
1114 	default:
1115 		eh = (struct ether_header *)sa.sa_data;
1116 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1117 		eh->ether_type = htons(ETHERTYPE_ARP);
1118 		break;
1119 	}
1120 	sa.sa_family = AF_UNSPEC;
1121 	sa.sa_len = sizeof sa;
1122 	ifp->if_output(ifp, m, &sa, NULL);
1123 }
1124 
1125 static void
1126 arp_update_msghandler(netmsg_t msg)
1127 {
1128 	struct netmsg_arp_update *rmsg = (struct netmsg_arp_update *)msg;
1129 	int nextcpu;
1130 
1131 	/*
1132 	 * This message handler will be called on all of the CPUs,
1133 	 * however, we only need to generate rtmsg on CPU0.
1134 	 */
1135 	arp_update_oncpu(rmsg->m, rmsg->saddr, rmsg->create,
1136 			 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT,
1137 			 mycpuid == 0);
1138 
1139 	nextcpu = mycpuid + 1;
1140 	if (nextcpu < ncpus)
1141 		lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
1142 	else
1143 		lwkt_replymsg(&rmsg->base.lmsg, 0);
1144 }
1145 
1146 #endif	/* INET */
1147 
1148 /*
1149  * Free an arp entry.  If the arp entry is actively referenced or represents
1150  * a static entry we only clear it back to an unresolved state, otherwise
1151  * we destroy the entry entirely.
1152  *
1153  * Note that static entries are created when route add ... -interface is used
1154  * to create an interface route to a (direct) destination.
1155  */
1156 static void
1157 arptfree(struct llinfo_arp *la)
1158 {
1159 	struct rtentry *rt = la->la_rt;
1160 	struct sockaddr_dl *sdl;
1161 
1162 	if (rt == NULL)
1163 		panic("arptfree");
1164 	sdl = SDL(rt->rt_gateway);
1165 	if (sdl != NULL &&
1166 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1167 	     (rt->rt_flags & RTF_STATIC))) {
1168 		sdl->sdl_alen = 0;
1169 		la->la_preempt = la->la_asked = 0;
1170 		rt->rt_flags &= ~RTF_REJECT;
1171 		return;
1172 	}
1173 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1174 }
1175 
1176 /*
1177  * Lookup or enter a new address in arptab.
1178  */
1179 static struct llinfo_arp *
1180 arplookup(in_addr_t addr, boolean_t create, boolean_t generate_report,
1181 	  boolean_t proxy)
1182 {
1183 	struct rtentry *rt;
1184 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1185 	const char *why = NULL;
1186 
1187 	sin.sin_addr.s_addr = addr;
1188 	sin.sin_other = proxy ? SIN_PROXY : 0;
1189 	if (create) {
1190 		rt = _rtlookup((struct sockaddr *)&sin,
1191 			       generate_report, RTL_DOCLONE);
1192 	} else {
1193 		rt = rtpurelookup((struct sockaddr *)&sin);
1194 	}
1195 	if (rt == NULL)
1196 		return (NULL);
1197 	rt->rt_refcnt--;
1198 
1199 	if (rt->rt_flags & RTF_GATEWAY)
1200 		why = "host is not on local network";
1201 	else if (!(rt->rt_flags & RTF_LLINFO))
1202 		why = "could not allocate llinfo";
1203 	else if (rt->rt_gateway->sa_family != AF_LINK)
1204 		why = "gateway route is not ours";
1205 
1206 	if (why) {
1207 		if (create) {
1208 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1209 			    inet_ntoa(sin.sin_addr), why);
1210 		}
1211 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1212 			/* No references to this route.  Purge it. */
1213 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1214 				  rt_mask(rt), rt->rt_flags, NULL);
1215 		}
1216 		return (NULL);
1217 	}
1218 	return (rt->rt_llinfo);
1219 }
1220 
1221 void
1222 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1223 {
1224 	ifa->ifa_rtrequest = arp_rtrequest;
1225 	ifa->ifa_flags |= RTF_CLONING;
1226 }
1227 
1228 void
1229 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1230 {
1231 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1232 		arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1233 				 &IA_SIN(ifa)->sin_addr, NULL);
1234 	}
1235 }
1236 
1237 static void
1238 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1239     enum ifaddr_event event, struct ifaddr *ifa)
1240 {
1241 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1242 		return;
1243 	if (ifa->ifa_addr->sa_family != AF_INET)
1244 		return;
1245 	if (event == IFADDR_EVENT_DELETE)
1246 		return;
1247 
1248 	/*
1249 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1250 	 * - If we are the CARP interface's parent, don't send gratuitous
1251 	 *   ARP to avoid unnecessary confusion.
1252 	 */
1253 #ifdef CARP
1254 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1255 #endif
1256 	{
1257 		arp_gratuitous(ifp, ifa);
1258 	}
1259 }
1260 
1261 static void
1262 arp_init(void)
1263 {
1264 	int cpu;
1265 
1266 	for (cpu = 0; cpu < ncpus2; cpu++)
1267 		LIST_INIT(&llinfo_arp_list[cpu]);
1268 
1269 	netisr_register(NETISR_ARP, arpintr, NULL);
1270 
1271 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1272 	    EVENTHANDLER_PRI_LAST);
1273 }
1274 
1275 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1276