xref: /dragonfly/sys/netinet/if_ether.c (revision e7d467f4)
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 static void	in_arpreply(struct mbuf *m, in_addr_t, in_addr_t);
170 static void	arp_update_msghandler(netmsg_t);
171 static void	arp_reply_msghandler(netmsg_t);
172 #endif
173 
174 static struct callout	arptimer_ch[MAXCPU];
175 
176 /*
177  * Timeout routine.  Age arp_tab entries periodically.
178  */
179 /* ARGSUSED */
180 static void
181 arptimer(void *ignored_arg)
182 {
183 	struct llinfo_arp *la, *nla;
184 
185 	crit_enter();
186 	LIST_FOREACH_MUTABLE(la, &llinfo_arp_list[mycpuid], la_le, nla) {
187 		if (la->la_rt->rt_expire && la->la_rt->rt_expire <= time_second)
188 			arptfree(la);
189 	}
190 	callout_reset(&arptimer_ch[mycpuid], arpt_prune * hz, arptimer, NULL);
191 	crit_exit();
192 }
193 
194 /*
195  * Parallel to llc_rtrequest.
196  *
197  * Called after a route is successfully added to the tree to fix-up the
198  * route and initiate arp operations if required.
199  */
200 static void
201 arp_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
202 {
203 	struct sockaddr *gate = rt->rt_gateway;
204 	struct llinfo_arp *la = rt->rt_llinfo;
205 
206 	struct sockaddr_dl null_sdl = { sizeof null_sdl, AF_LINK };
207 	static boolean_t arpinit_done[MAXCPU];
208 
209 	if (!arpinit_done[mycpuid]) {
210 		arpinit_done[mycpuid] = TRUE;
211 		callout_init(&arptimer_ch[mycpuid]);
212 		callout_reset(&arptimer_ch[mycpuid], hz, arptimer, NULL);
213 	}
214 	if (rt->rt_flags & RTF_GATEWAY)
215 		return;
216 
217 	switch (req) {
218 	case RTM_ADD:
219 		/*
220 		 * XXX: If this is a manually added route to interface
221 		 * such as older version of routed or gated might provide,
222 		 * restore cloning bit.
223 		 */
224 		if (!(rt->rt_flags & RTF_HOST) &&
225 		    SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
226 			rt->rt_flags |= RTF_CLONING;
227 		if (rt->rt_flags & RTF_CLONING) {
228 			/*
229 			 * Case 1: This route should come from a route to iface.
230 			 */
231 			rt_setgate(rt, rt_key(rt),
232 				   (struct sockaddr *)&null_sdl,
233 				   RTL_DONTREPORT);
234 			gate = rt->rt_gateway;
235 			SDL(gate)->sdl_type = rt->rt_ifp->if_type;
236 			SDL(gate)->sdl_index = rt->rt_ifp->if_index;
237 			rt->rt_expire = time_second;
238 			break;
239 		}
240 		/* Announce a new entry if requested. */
241 		if (rt->rt_flags & RTF_ANNOUNCE) {
242 			arprequest_async(rt->rt_ifp,
243 			    &SIN(rt_key(rt))->sin_addr,
244 			    &SIN(rt_key(rt))->sin_addr,
245 			    LLADDR(SDL(gate)));
246 		}
247 		/*FALLTHROUGH*/
248 	case RTM_RESOLVE:
249 		if (gate->sa_family != AF_LINK ||
250 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
251 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
252 			break;
253 		}
254 		SDL(gate)->sdl_type = rt->rt_ifp->if_type;
255 		SDL(gate)->sdl_index = rt->rt_ifp->if_index;
256 		if (la != NULL)
257 			break; /* This happens on a route change */
258 		/*
259 		 * Case 2:  This route may come from cloning, or a manual route
260 		 * add with a LL address.
261 		 */
262 		R_Malloc(la, struct llinfo_arp *, sizeof *la);
263 		rt->rt_llinfo = la;
264 		if (la == NULL) {
265 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
266 			break;
267 		}
268 		bzero(la, sizeof *la);
269 		la->la_rt = rt;
270 		rt->rt_flags |= RTF_LLINFO;
271 		LIST_INSERT_HEAD(&llinfo_arp_list[mycpuid], la, la_le);
272 
273 #ifdef INET
274 		/*
275 		 * This keeps the multicast addresses from showing up
276 		 * in `arp -a' listings as unresolved.  It's not actually
277 		 * functional.  Then the same for broadcast.
278 		 */
279 		if (IN_MULTICAST(ntohl(SIN(rt_key(rt))->sin_addr.s_addr))) {
280 			ETHER_MAP_IP_MULTICAST(&SIN(rt_key(rt))->sin_addr,
281 					       LLADDR(SDL(gate)));
282 			SDL(gate)->sdl_alen = 6;
283 			rt->rt_expire = 0;
284 		}
285 		if (in_broadcast(SIN(rt_key(rt))->sin_addr, rt->rt_ifp)) {
286 			memcpy(LLADDR(SDL(gate)), rt->rt_ifp->if_broadcastaddr,
287 			       rt->rt_ifp->if_addrlen);
288 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
289 			rt->rt_expire = 0;
290 		}
291 #endif
292 
293 		/*
294 		 * This fixes up the routing interface for local addresses.
295 		 * The route is adjusted to point at lo0 and the expiration
296 		 * timer is disabled.
297 		 *
298 		 * NOTE: This prevents locally targetted traffic from going
299 		 *	 out the hardware interface, which is inefficient
300 		 *	 and might not work if the hardware cannot listen
301 		 *	 to its own transmitted packets.   Setting
302 		 *	 net.link.ether.inet.useloopback to 0 will force
303 		 *	 packets for local addresses out the hardware (and
304 		 *	 it is expected to receive its own packet).
305 		 *
306 		 * XXX We should just be able to test RTF_LOCAL here instead
307 		 *     of having to compare IPs.
308 		 */
309 		if (SIN(rt_key(rt))->sin_addr.s_addr ==
310 		    (IA_SIN(rt->rt_ifa))->sin_addr.s_addr) {
311 			rt->rt_expire = 0;
312 			bcopy(IF_LLADDR(rt->rt_ifp), LLADDR(SDL(gate)),
313 			      SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen);
314 			if (useloopback)
315 				rt->rt_ifp = loif;
316 		}
317 		break;
318 
319 	case RTM_DELETE:
320 		if (la == NULL)
321 			break;
322 		LIST_REMOVE(la, la_le);
323 		rt->rt_llinfo = NULL;
324 		rt->rt_flags &= ~RTF_LLINFO;
325 		if (la->la_hold != NULL)
326 			m_freem(la->la_hold);
327 		Free(la);
328 		break;
329 	}
330 }
331 
332 static struct mbuf *
333 arpreq_alloc(struct ifnet *ifp, const struct in_addr *sip,
334 	     const struct in_addr *tip, const u_char *enaddr)
335 {
336 	struct mbuf *m;
337 	struct arphdr *ah;
338 	u_short ar_hrd;
339 
340 	if ((m = m_gethdr(MB_DONTWAIT, MT_DATA)) == NULL)
341 		return NULL;
342 	m->m_pkthdr.rcvif = NULL;
343 
344 	switch (ifp->if_type) {
345 	case IFT_ETHER:
346 		/*
347 		 * This may not be correct for types not explicitly
348 		 * listed, but this is our best guess
349 		 */
350 	default:
351 		ar_hrd = htons(ARPHRD_ETHER);
352 
353 		m->m_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
354 		m->m_pkthdr.len = m->m_len;
355 		MH_ALIGN(m, m->m_len);
356 
357 		ah = mtod(m, struct arphdr *);
358 		break;
359 	}
360 
361 	ah->ar_hrd = ar_hrd;
362 	ah->ar_pro = htons(ETHERTYPE_IP);
363 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
364 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
365 	ah->ar_op = htons(ARPOP_REQUEST);
366 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
367 	memset(ar_tha(ah), 0, ah->ar_hln);
368 	memcpy(ar_spa(ah), sip, ah->ar_pln);
369 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
370 
371 	return m;
372 }
373 
374 static void
375 arpreq_send(struct ifnet *ifp, struct mbuf *m)
376 {
377 	struct sockaddr sa;
378 	struct ether_header *eh;
379 
380 	switch (ifp->if_type) {
381 	case IFT_ETHER:
382 		/*
383 		 * This may not be correct for types not explicitly
384 		 * listed, but this is our best guess
385 		 */
386 	default:
387 		eh = (struct ether_header *)sa.sa_data;
388 		/* if_output() will not swap */
389 		eh->ether_type = htons(ETHERTYPE_ARP);
390 		memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ifp->if_addrlen);
391 		break;
392 	}
393 
394 	sa.sa_family = AF_UNSPEC;
395 	sa.sa_len = sizeof(sa);
396 	ifp->if_output(ifp, m, &sa, NULL);
397 }
398 
399 static void
400 arpreq_send_handler(netmsg_t msg)
401 {
402 	struct mbuf *m = msg->packet.nm_packet;
403 	struct ifnet *ifp = msg->lmsg.u.ms_resultp;
404 
405 	arpreq_send(ifp, m);
406 	/* nmsg was embedded in the mbuf, do not reply! */
407 }
408 
409 /*
410  * Broadcast an ARP request. Caller specifies:
411  *	- arp header source ip address
412  *	- arp header target ip address
413  *	- arp header source ethernet address
414  *
415  * NOTE: Caller MUST NOT hold ifp's serializer
416  */
417 static void
418 arprequest(struct ifnet *ifp, const struct in_addr *sip,
419 	   const struct in_addr *tip, const u_char *enaddr)
420 {
421 	struct mbuf *m;
422 
423 	if (enaddr == NULL) {
424 		if (ifp->if_bridge) {
425 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
426 		} else {
427 			enaddr = IF_LLADDR(ifp);
428 		}
429 	}
430 
431 	m = arpreq_alloc(ifp, sip, tip, enaddr);
432 	if (m == NULL)
433 		return;
434 	arpreq_send(ifp, m);
435 }
436 
437 /*
438  * Same as arprequest(), except:
439  * - Caller is allowed to hold ifp's serializer
440  * - Network output is done in protocol thead
441  */
442 static void
443 arprequest_async(struct ifnet *ifp, const struct in_addr *sip,
444 		 const struct in_addr *tip, const u_char *enaddr)
445 {
446 	struct mbuf *m;
447 	struct netmsg_packet *pmsg;
448 
449 	if (enaddr == NULL) {
450 		if (ifp->if_bridge) {
451 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
452 		} else {
453 			enaddr = IF_LLADDR(ifp);
454 		}
455 	}
456 	m = arpreq_alloc(ifp, sip, tip, enaddr);
457 	if (m == NULL)
458 		return;
459 
460 	pmsg = &m->m_hdr.mh_netmsg;
461 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
462 		    0, arpreq_send_handler);
463 	pmsg->nm_packet = m;
464 	pmsg->base.lmsg.u.ms_resultp = ifp;
465 
466 	lwkt_sendmsg(netisr_portfn(mycpuid), &pmsg->base.lmsg);
467 }
468 
469 /*
470  * Resolve an IP address into an ethernet address.  If success,
471  * desten is filled in.  If there is no entry in arptab,
472  * set one up and broadcast a request for the IP address.
473  * Hold onto this mbuf and resend it once the address
474  * is finally resolved.  A return value of 1 indicates
475  * that desten has been filled in and the packet should be sent
476  * normally; a 0 return indicates that the packet has been
477  * taken over here, either now or for later transmission.
478  */
479 int
480 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
481 	   struct sockaddr *dst, u_char *desten)
482 {
483 	struct rtentry *rt = NULL;
484 	struct llinfo_arp *la = NULL;
485 	struct sockaddr_dl *sdl;
486 
487 	if (m->m_flags & M_BCAST) {	/* broadcast */
488 		memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen);
489 		return (1);
490 	}
491 	if (m->m_flags & M_MCAST) {/* multicast */
492 		ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
493 		return (1);
494 	}
495 	if (rt0 != NULL) {
496 		if (rt_llroute(dst, rt0, &rt) != 0) {
497 			m_freem(m);
498 			return 0;
499 		}
500 		la = rt->rt_llinfo;
501 	}
502 	if (la == NULL) {
503 		la = arplookup(SIN(dst)->sin_addr.s_addr,
504 			       TRUE, RTL_REPORTMSG, FALSE);
505 		if (la != NULL)
506 			rt = la->la_rt;
507 	}
508 	if (la == NULL || rt == NULL) {
509 		log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s%s%s\n",
510 		    inet_ntoa(SIN(dst)->sin_addr), la ? "la" : " ",
511 		    rt ? "rt" : "");
512 		m_freem(m);
513 		return (0);
514 	}
515 	sdl = SDL(rt->rt_gateway);
516 	/*
517 	 * Check the address family and length is valid, the address
518 	 * is resolved; otherwise, try to resolve.
519 	 */
520 	if ((rt->rt_expire == 0 || rt->rt_expire > time_second) &&
521 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
522 		/*
523 		 * If entry has an expiry time and it is approaching,
524 		 * see if we need to send an ARP request within this
525 		 * arpt_down interval.
526 		 */
527 		if ((rt->rt_expire != 0) &&
528 		    (time_second + la->la_preempt > rt->rt_expire)) {
529 			arprequest(ifp,
530 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
531 				   &SIN(dst)->sin_addr,
532 				   NULL);
533 			la->la_preempt--;
534 		}
535 
536 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
537 		return 1;
538 	}
539 	/*
540 	 * If ARP is disabled or static on this interface, stop.
541 	 * XXX
542 	 * Probably should not allocate empty llinfo struct if we are
543 	 * not going to be sending out an arp request.
544 	 */
545 	if (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) {
546 		m_freem(m);
547 		return (0);
548 	}
549 	/*
550 	 * There is an arptab entry, but no ethernet address
551 	 * response yet.  Replace the held mbuf with this
552 	 * latest one.
553 	 */
554 	if (la->la_hold != NULL)
555 		m_freem(la->la_hold);
556 	la->la_hold = m;
557 	la->la_msgport = netisr_curport();
558 	if (rt->rt_expire || ((rt->rt_flags & RTF_STATIC) && !sdl->sdl_alen)) {
559 		rt->rt_flags &= ~RTF_REJECT;
560 		if (la->la_asked == 0 || rt->rt_expire != time_second) {
561 			rt->rt_expire = time_second;
562 			if (la->la_asked++ < arp_maxtries) {
563 				arprequest(ifp,
564 					   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
565 					   &SIN(dst)->sin_addr,
566 					   NULL);
567 			} else {
568 				rt->rt_flags |= RTF_REJECT;
569 				rt->rt_expire += arpt_down;
570 				la->la_asked = 0;
571 				la->la_preempt = arp_maxtries;
572 			}
573 		}
574 	}
575 	return (0);
576 }
577 
578 /*
579  * Common length and type checks are done here,
580  * then the protocol-specific routine is called.
581  */
582 static void
583 arpintr(netmsg_t msg)
584 {
585 	struct mbuf *m = msg->packet.nm_packet;
586 	struct arphdr *ar;
587 	u_short ar_hrd;
588 	char hexstr[6];
589 
590 	if (m->m_len < sizeof(struct arphdr) &&
591 	    (m = m_pullup(m, sizeof(struct arphdr))) == NULL) {
592 		log(LOG_ERR, "arp: runt packet -- m_pullup failed\n");
593 		return;
594 	}
595 	ar = mtod(m, struct arphdr *);
596 
597 	ar_hrd = ntohs(ar->ar_hrd);
598 	if (ar_hrd != ARPHRD_ETHER && ar_hrd != ARPHRD_IEEE802) {
599 		hexncpy((unsigned char *)&ar->ar_hrd, 2, hexstr, 5, NULL);
600 		log(LOG_ERR, "arp: unknown hardware address format (0x%s)\n",
601 		    hexstr);
602 		m_freem(m);
603 		return;
604 	}
605 
606 	if (m->m_pkthdr.len < arphdr_len(ar)) {
607 		if ((m = m_pullup(m, arphdr_len(ar))) == NULL) {
608 			log(LOG_ERR, "arp: runt packet\n");
609 			return;
610 		}
611 		ar = mtod(m, struct arphdr *);
612 	}
613 
614 	switch (ntohs(ar->ar_pro)) {
615 #ifdef INET
616 	case ETHERTYPE_IP:
617 		in_arpinput(m);
618 		return;
619 #endif
620 	}
621 	m_freem(m);
622 	/* msg was embedded in the mbuf, do not reply! */
623 }
624 
625 #ifdef INET
626 /*
627  * ARP for Internet protocols on 10 Mb/s Ethernet.
628  * Algorithm is that given in RFC 826.
629  * In addition, a sanity check is performed on the sender
630  * protocol address, to catch impersonators.
631  * We no longer handle negotiations for use of trailer protocol:
632  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
633  * along with IP replies if we wanted trailers sent to us,
634  * and also sent them in response to IP replies.
635  * This allowed either end to announce the desire to receive
636  * trailer packets.
637  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
638  * but formerly didn't normally send requests.
639  */
640 
641 static int	log_arp_wrong_iface = 1;
642 static int	log_arp_movements = 1;
643 static int	log_arp_permanent_modify = 1;
644 
645 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
646 	   &log_arp_wrong_iface, 0,
647 	   "Log arp packets arriving on the wrong interface");
648 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
649 	   &log_arp_movements, 0,
650 	   "Log arp replies from MACs different than the one in the cache");
651 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
652 	   &log_arp_permanent_modify, 0,
653 	   "Log arp replies from MACs different than the one "
654 	   "in the permanent arp entry");
655 
656 
657 static void
658 arp_hold_output(netmsg_t msg)
659 {
660 	struct mbuf *m = msg->packet.nm_packet;
661 	struct rtentry *rt;
662 	struct ifnet *ifp;
663 
664 	rt = msg->lmsg.u.ms_resultp;
665 	ifp = m->m_pkthdr.rcvif;
666 	m->m_pkthdr.rcvif = NULL;
667 
668 	ifp->if_output(ifp, m, rt_key(rt), rt);
669 
670 	/* Drop the reference count bumped by the sender */
671 	RTFREE(rt);
672 
673 	/* nmsg was embedded in the mbuf, do not reply! */
674 }
675 
676 static void
677 arp_update_oncpu(struct mbuf *m, in_addr_t saddr, boolean_t create,
678 		 boolean_t generate_report, boolean_t dologging)
679 {
680 	struct arphdr *ah = mtod(m, struct arphdr *);
681 	struct ifnet *ifp = m->m_pkthdr.rcvif;
682 	struct llinfo_arp *la;
683 	struct sockaddr_dl *sdl;
684 	struct rtentry *rt;
685 	char hexstr[2][64];
686 
687 	la = arplookup(saddr, create, generate_report, FALSE);
688 	if (la && (rt = la->la_rt) && (sdl = SDL(rt->rt_gateway))) {
689 		struct in_addr isaddr = { saddr };
690 
691 		/*
692 		 * Normally arps coming in on the wrong interface are ignored,
693 		 * but if we are bridging and the two interfaces belong to
694 		 * the same bridge, or one is a member of the bridge which
695 		 * is the other, then it isn't an error.
696 		 */
697 		if (rt->rt_ifp != ifp) {
698 			/*
699 			 * (1) ifp and rt_ifp both members of same bridge
700 			 * (2) rt_ifp member of bridge ifp
701 			 * (3) ifp member of bridge rt_ifp
702 			 *
703 			 * Always replace rt_ifp with the bridge ifc.
704 			 */
705 			struct ifnet *nifp;
706 
707 			if (ifp->if_bridge &&
708 			    rt->rt_ifp->if_bridge == ifp->if_bridge) {
709 				nifp = ether_bridge_interface(ifp);
710 			} else if (rt->rt_ifp->if_bridge &&
711 				   ether_bridge_interface(rt->rt_ifp) == ifp) {
712 				nifp = ifp;
713 			} else if (ifp->if_bridge &&
714 				   ether_bridge_interface(ifp) == rt->rt_ifp) {
715 				nifp = rt->rt_ifp;
716 			} else {
717 				nifp = NULL;
718 			}
719 
720 			if ((log_arp_wrong_iface == 1 && nifp == NULL) ||
721 			    log_arp_wrong_iface == 2) {
722 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
723 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
724 				log(LOG_ERR,
725 				    "arp: %s is on %s "
726 				    "but got reply from %s on %s\n",
727 				    inet_ntoa(isaddr),
728 				    rt->rt_ifp->if_xname, hexstr[0],
729 				    ifp->if_xname);
730 			}
731 			if (nifp == NULL)
732 				return;
733 
734 			/*
735 			 * nifp is our man!  Replace rt_ifp and adjust
736 			 * the sdl.
737 			 */
738 			ifp = rt->rt_ifp = nifp;
739 			sdl->sdl_type = ifp->if_type;
740 			sdl->sdl_index = ifp->if_index;
741 		}
742 		if (sdl->sdl_alen &&
743 		    bcmp(ar_sha(ah), LLADDR(sdl), sdl->sdl_alen)) {
744 			if (rt->rt_expire != 0) {
745 				if (dologging && log_arp_movements) {
746 					hexncpy((u_char *)LLADDR(sdl), ifp->if_addrlen,
747 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
748 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
749 					    hexstr[1], HEX_NCPYLEN(ifp->if_addrlen), ":");
750 			    		log(LOG_INFO,
751 					    "arp: %s moved from %s to %s on %s\n",
752 					    inet_ntoa(isaddr), hexstr[0], hexstr[1],
753 					    ifp->if_xname);
754 				}
755 			} else {
756 				if (dologging && log_arp_permanent_modify) {
757 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
758 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
759 					log(LOG_ERR,
760 					"arp: %s attempts to modify "
761 					"permanent entry for %s on %s\n",
762 					hexstr[0], inet_ntoa(isaddr), ifp->if_xname);
763 				}
764 				return;
765 			}
766 		}
767 		/*
768 		 * sanity check for the address length.
769 		 * XXX this does not work for protocols with variable address
770 		 * length. -is
771 		 */
772 		if (dologging && sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
773 			hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
774 			    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
775 			log(LOG_WARNING,
776 			    "arp from %s: new addr len %d, was %d",
777 			    hexstr[0], ah->ar_hln, sdl->sdl_alen);
778 		}
779 		if (ifp->if_addrlen != ah->ar_hln) {
780 			if (dologging) {
781 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
782 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
783 				log(LOG_WARNING,
784 				"arp from %s: addr len: new %d, i/f %d "
785 				"(ignored)", hexstr[0],
786 				ah->ar_hln, ifp->if_addrlen);
787 			}
788 			return;
789 		}
790 		memcpy(LLADDR(sdl), ar_sha(ah), sdl->sdl_alen = ah->ar_hln);
791 		if (rt->rt_expire != 0) {
792 			rt->rt_expire = time_second + arpt_keep;
793 		}
794 		rt->rt_flags &= ~RTF_REJECT;
795 		la->la_asked = 0;
796 		la->la_preempt = arp_maxtries;
797 
798 		/*
799 		 * This particular cpu might have been holding an mbuf
800 		 * pending ARP resolution.  If so, transmit the mbuf now.
801 		 */
802 		if (la->la_hold != NULL) {
803 			struct mbuf *m = la->la_hold;
804 			struct lwkt_port *port = la->la_msgport;
805 			struct netmsg_packet *pmsg;
806 
807 			la->la_hold = NULL;
808 			la->la_msgport = NULL;
809 
810 			m_adj(m, sizeof(struct ether_header));
811 
812 			/*
813 			 * Make sure that this rtentry will not be freed
814 			 * before the packet is processed on the target
815 			 * msgport.  The reference count will be dropped
816 			 * in the handler associated with this packet.
817 			 */
818 			rt->rt_refcnt++;
819 
820 			pmsg = &m->m_hdr.mh_netmsg;
821 			netmsg_init(&pmsg->base, NULL,
822 				    &netisr_apanic_rport,
823 				    MSGF_PRIORITY, arp_hold_output);
824 			pmsg->nm_packet = m;
825 
826 			/* Record necessary information */
827 			m->m_pkthdr.rcvif = ifp;
828 			pmsg->base.lmsg.u.ms_resultp = rt;
829 
830 			lwkt_sendmsg(port, &pmsg->base.lmsg);
831 		}
832 	}
833 }
834 
835 /*
836  * Called from arpintr() - this routine is run from a single cpu.
837  */
838 static void
839 in_arpinput(struct mbuf *m)
840 {
841 	struct arphdr *ah;
842 	struct ifnet *ifp = m->m_pkthdr.rcvif;
843 	struct ifaddr_container *ifac;
844 	struct in_ifaddr_container *iac;
845 	struct in_ifaddr *ia = NULL;
846 	struct in_addr isaddr, itaddr, myaddr;
847 	struct netmsg_inarp *msg;
848 	uint8_t *enaddr = NULL;
849 	int req_len;
850 	char hexstr[64];
851 
852 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
853 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
854 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
855 		return;
856 	}
857 
858 	ah = mtod(m, struct arphdr *);
859 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
860 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
861 
862 	/*
863 	 * Check both target and sender IP addresses:
864 	 *
865 	 * If we receive the packet on the interface owning the address,
866 	 * then accept the address.
867 	 *
868 	 * For a bridge, we accept the address if the receive interface and
869 	 * the interface owning the address are on the same bridge, and
870 	 * use the bridge MAC as the is-at response.  The bridge will be
871 	 * responsible for handling the packet.
872 	 *
873 	 * (0) Check target IP against CARP IPs
874 	 */
875 #ifdef CARP
876 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
877 		int is_match = 0, is_parent = 0;
878 
879 		ia = iac->ia;
880 
881 		/* Skip all ia's which don't match */
882 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
883 			continue;
884 
885 		if (ia->ia_ifp->if_type != IFT_CARP)
886 			continue;
887 
888 		if (carp_parent(ia->ia_ifp) == ifp)
889 			is_parent = 1;
890 		if (is_parent || ia->ia_ifp == ifp)
891 			is_match = carp_iamatch(ia);
892 
893 		if (is_match) {
894 			if (is_parent) {
895 				/*
896 				 * The parent interface will also receive
897 				 * the ethernet broadcast packets, e.g. ARP
898 				 * REQUEST, so if we could find a CARP
899 				 * interface of the parent that could match
900 				 * the target IP address, we then drop the
901 				 * packets, which is delieverd to us through
902 				 * the parent interface.
903 				 */
904 				m_freem(m);
905 				return;
906 			}
907 			goto match;
908 		}
909 	}
910 #endif	/* CARP */
911 
912 	/*
913 	 * (1) Check target IP against our local IPs
914 	 */
915 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
916 		ia = iac->ia;
917 
918 		/* Skip all ia's which don't match */
919 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
920 			continue;
921 
922 #ifdef CARP
923 		/* CARP interfaces are checked in (0) */
924 		if (ia->ia_ifp->if_type == IFT_CARP)
925 			continue;
926 #endif
927 
928 		if (ifp->if_bridge && ia->ia_ifp &&
929 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
930 			ifp = ether_bridge_interface(ifp);
931 			goto match;
932 		}
933 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
934 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
935 			goto match;
936 		}
937 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
938 		    ia->ia_ifp) {
939 			goto match;
940 		}
941 		if (ia->ia_ifp == ifp) {
942 			goto match;
943 		}
944 	}
945 
946 	/*
947 	 * (2) Check sender IP against our local IPs
948 	 */
949 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
950 		ia = iac->ia;
951 
952 		/* Skip all ia's which don't match */
953 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
954 			continue;
955 
956 		if (ifp->if_bridge && ia->ia_ifp &&
957 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
958 			ifp = ether_bridge_interface(ifp);
959 			goto match;
960 		}
961 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
962 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
963 			goto match;
964 		}
965 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
966 		    ia->ia_ifp) {
967 			goto match;
968 		}
969 
970 		if (ia->ia_ifp == ifp)
971 			goto match;
972 	}
973 
974 	/*
975 	 * No match, use the first inet address on the receive interface
976 	 * as a dummy address for the rest of the function.
977 	 */
978 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
979 		struct ifaddr *ifa = ifac->ifa;
980 
981 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
982 			ia = ifatoia(ifa);
983 			goto match;
984 		}
985 	}
986 
987 	/*
988 	 * If we got here, we didn't find any suitable interface,
989 	 * so drop the packet.
990 	 */
991 	m_freem(m);
992 	return;
993 
994 match:
995 	if (!enaddr)
996 		enaddr = (uint8_t *)IF_LLADDR(ifp);
997 	myaddr = ia->ia_addr.sin_addr;
998 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
999 		m_freem(m);	/* it's from me, ignore it. */
1000 		return;
1001 	}
1002 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1003 		log(LOG_ERR,
1004 		    "arp: link address is broadcast for IP address %s!\n",
1005 		    inet_ntoa(isaddr));
1006 		m_freem(m);
1007 		return;
1008 	}
1009 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1010 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1011 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1012 		log(LOG_ERR,
1013 		   "arp: %s is using my IP address %s!\n",
1014 		    hexstr, inet_ntoa(isaddr));
1015 		itaddr = myaddr;
1016 		goto reply;
1017 	}
1018 	if (ifp->if_flags & IFF_STATICARP)
1019 		goto reply;
1020 
1021 	/*
1022 	 * When arp_restricted_match is true and the ARP response is not
1023 	 * specifically targetted to me, ignore it.  Otherwise the entry
1024 	 * timeout may be updated for an old MAC.
1025 	 */
1026 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1027 		m_freem(m);
1028 		return;
1029 	}
1030 
1031 	/*
1032 	 * Update all CPU's routing tables with this ARP packet
1033 	 */
1034 	msg = &m->m_hdr.mh_arpmsg;
1035 	netmsg_init(&msg->base, NULL, &netisr_apanic_rport,
1036 	    0, arp_update_msghandler);
1037 	msg->m = m;
1038 	msg->saddr = isaddr.s_addr;
1039 	msg->taddr = itaddr.s_addr;
1040 	msg->myaddr = myaddr.s_addr;
1041 	lwkt_sendmsg(rtable_portfn(0), &msg->base.lmsg);
1042 
1043 	/*
1044 	 * Just return here; after all CPUs's routing tables are
1045 	 * properly updated by this ARP packet, an ARP reply will
1046 	 * be generated if appropriate.
1047 	 */
1048 	return;
1049 reply:
1050 	in_arpreply(m, itaddr.s_addr, myaddr.s_addr);
1051 }
1052 
1053 static void
1054 arp_reply_msghandler(netmsg_t msg)
1055 {
1056 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1057 
1058 	in_arpreply(rmsg->m, rmsg->taddr, rmsg->myaddr);
1059 	/* Don't reply this netmsg; netmsg_inarp is embedded in mbuf */
1060 }
1061 
1062 static void
1063 arp_update_msghandler(netmsg_t msg)
1064 {
1065 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1066 	int nextcpu;
1067 
1068 	/*
1069 	 * This message handler will be called on all of the CPUs,
1070 	 * however, we only need to generate rtmsg on CPU0.
1071 	 */
1072 	arp_update_oncpu(rmsg->m, rmsg->saddr, rmsg->taddr == rmsg->myaddr,
1073 	    mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT, mycpuid == 0);
1074 
1075 	nextcpu = mycpuid + 1;
1076 	if (nextcpu < ncpus) {
1077 		lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg);
1078 	} else {
1079 		struct mbuf *m = rmsg->m;
1080 		in_addr_t saddr = rmsg->saddr;
1081 		in_addr_t taddr = rmsg->taddr;
1082 		in_addr_t myaddr = rmsg->myaddr;
1083 
1084 		/*
1085 		 * Dispatch this mbuf to netisr0 to perform ARP reply,
1086 		 * if appropriate.
1087 		 * NOTE: netmsg_inarp is embedded in this mbuf.
1088 		 */
1089 		netmsg_init(&rmsg->base, NULL, &netisr_apanic_rport,
1090 		    0, arp_reply_msghandler);
1091 		rmsg->m = m;
1092 		rmsg->saddr = saddr;
1093 		rmsg->taddr = taddr;
1094 		rmsg->myaddr = myaddr;
1095 		lwkt_sendmsg(netisr_portfn(0), &rmsg->base.lmsg);
1096 	}
1097 }
1098 
1099 static void
1100 in_arpreply(struct mbuf *m, in_addr_t taddr, in_addr_t myaddr)
1101 {
1102 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1103 	const uint8_t *enaddr;
1104 	struct arphdr *ah;
1105 	struct sockaddr sa;
1106 	struct ether_header *eh;
1107 
1108 	ah = mtod(m, struct arphdr *);
1109 	if (ntohs(ah->ar_op) != ARPOP_REQUEST) {
1110 		m_freem(m);
1111 		return;
1112 	}
1113 
1114 	enaddr = (const uint8_t *)IF_LLADDR(ifp);
1115 	if (taddr == myaddr) {
1116 		/* I am the target */
1117 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1118 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1119 	} else {
1120 		struct llinfo_arp *la;
1121 		struct rtentry *rt;
1122 
1123 		la = arplookup(taddr, FALSE, RTL_DONTREPORT, SIN_PROXY);
1124 		if (la == NULL) {
1125 			struct sockaddr_in sin;
1126 
1127 			if (!arp_proxyall) {
1128 				m_freem(m);
1129 				return;
1130 			}
1131 
1132 			bzero(&sin, sizeof sin);
1133 			sin.sin_family = AF_INET;
1134 			sin.sin_len = sizeof sin;
1135 			sin.sin_addr.s_addr = taddr;
1136 
1137 			rt = rtpurelookup((struct sockaddr *)&sin);
1138 			if (rt == NULL) {
1139 				m_freem(m);
1140 				return;
1141 			}
1142 			--rt->rt_refcnt;
1143 			/*
1144 			 * Don't send proxies for nodes on the same interface
1145 			 * as this one came out of, or we'll get into a fight
1146 			 * over who claims what Ether address.
1147 			 */
1148 			if (rt->rt_ifp == ifp) {
1149 				m_freem(m);
1150 				return;
1151 			}
1152 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1153 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1154 #ifdef DEBUG_PROXY
1155 			kprintf("arp: proxying for %s\n", inet_ntoa(itaddr));
1156 #endif
1157 		} else {
1158 			struct sockaddr_dl *sdl;
1159 
1160 			rt = la->la_rt;
1161 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1162 			sdl = SDL(rt->rt_gateway);
1163 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1164 		}
1165 	}
1166 
1167 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1168 	memcpy(ar_spa(ah), &taddr, ah->ar_pln);
1169 	ah->ar_op = htons(ARPOP_REPLY);
1170 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1171 	switch (ifp->if_type) {
1172 	case IFT_ETHER:
1173 		/*
1174 		 * May not be correct for types not explictly
1175 		 * listed, but it is our best guess.
1176 		 */
1177 	default:
1178 		eh = (struct ether_header *)sa.sa_data;
1179 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1180 		eh->ether_type = htons(ETHERTYPE_ARP);
1181 		break;
1182 	}
1183 	sa.sa_family = AF_UNSPEC;
1184 	sa.sa_len = sizeof sa;
1185 	ifp->if_output(ifp, m, &sa, NULL);
1186 }
1187 
1188 #endif	/* INET */
1189 
1190 /*
1191  * Free an arp entry.  If the arp entry is actively referenced or represents
1192  * a static entry we only clear it back to an unresolved state, otherwise
1193  * we destroy the entry entirely.
1194  *
1195  * Note that static entries are created when route add ... -interface is used
1196  * to create an interface route to a (direct) destination.
1197  */
1198 static void
1199 arptfree(struct llinfo_arp *la)
1200 {
1201 	struct rtentry *rt = la->la_rt;
1202 	struct sockaddr_dl *sdl;
1203 
1204 	if (rt == NULL)
1205 		panic("arptfree");
1206 	sdl = SDL(rt->rt_gateway);
1207 	if (sdl != NULL &&
1208 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1209 	     (rt->rt_flags & RTF_STATIC))) {
1210 		sdl->sdl_alen = 0;
1211 		la->la_preempt = la->la_asked = 0;
1212 		rt->rt_flags &= ~RTF_REJECT;
1213 		return;
1214 	}
1215 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1216 }
1217 
1218 /*
1219  * Lookup or enter a new address in arptab.
1220  */
1221 static struct llinfo_arp *
1222 arplookup(in_addr_t addr, boolean_t create, boolean_t generate_report,
1223 	  boolean_t proxy)
1224 {
1225 	struct rtentry *rt;
1226 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1227 	const char *why = NULL;
1228 
1229 	sin.sin_addr.s_addr = addr;
1230 	sin.sin_other = proxy ? SIN_PROXY : 0;
1231 	if (create) {
1232 		rt = _rtlookup((struct sockaddr *)&sin,
1233 			       generate_report, RTL_DOCLONE);
1234 	} else {
1235 		rt = rtpurelookup((struct sockaddr *)&sin);
1236 	}
1237 	if (rt == NULL)
1238 		return (NULL);
1239 	rt->rt_refcnt--;
1240 
1241 	if (rt->rt_flags & RTF_GATEWAY)
1242 		why = "host is not on local network";
1243 	else if (!(rt->rt_flags & RTF_LLINFO))
1244 		why = "could not allocate llinfo";
1245 	else if (rt->rt_gateway->sa_family != AF_LINK)
1246 		why = "gateway route is not ours";
1247 
1248 	if (why) {
1249 		if (create) {
1250 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1251 			    inet_ntoa(sin.sin_addr), why);
1252 		}
1253 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1254 			/* No references to this route.  Purge it. */
1255 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1256 				  rt_mask(rt), rt->rt_flags, NULL);
1257 		}
1258 		return (NULL);
1259 	}
1260 	return (rt->rt_llinfo);
1261 }
1262 
1263 void
1264 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1265 {
1266 	ifa->ifa_rtrequest = arp_rtrequest;
1267 	ifa->ifa_flags |= RTF_CLONING;
1268 }
1269 
1270 void
1271 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1272 {
1273 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1274 		arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1275 				 &IA_SIN(ifa)->sin_addr, NULL);
1276 	}
1277 }
1278 
1279 static void
1280 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1281     enum ifaddr_event event, struct ifaddr *ifa)
1282 {
1283 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1284 		return;
1285 	if (ifa->ifa_addr->sa_family != AF_INET)
1286 		return;
1287 	if (event == IFADDR_EVENT_DELETE)
1288 		return;
1289 
1290 	/*
1291 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1292 	 * - If we are the CARP interface's parent, don't send gratuitous
1293 	 *   ARP to avoid unnecessary confusion.
1294 	 */
1295 #ifdef CARP
1296 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1297 #endif
1298 	{
1299 		arp_gratuitous(ifp, ifa);
1300 	}
1301 }
1302 
1303 static void
1304 arp_init(void)
1305 {
1306 	int cpu;
1307 
1308 	for (cpu = 0; cpu < ncpus2; cpu++)
1309 		LIST_INIT(&llinfo_arp_list[cpu]);
1310 
1311 	netisr_register(NETISR_ARP, arpintr, NULL);
1312 
1313 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1314 	    EVENTHANDLER_PRI_LAST);
1315 }
1316 
1317 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1318