xref: /dragonfly/sys/netinet/if_ether.c (revision 65cc0652)
1 /*
2  * Copyright (c) 2004, 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey M. Hsu.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of The DragonFly Project nor the names of its
16  *    contributors may be used to endorse or promote products derived
17  *    from this software without specific, prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1988, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
62  * $FreeBSD: src/sys/netinet/if_ether.c,v 1.64.2.23 2003/04/11 07:23:15 fjoe Exp $
63  */
64 
65 /*
66  * Ethernet address resolution protocol.
67  * TODO:
68  *	add "inuse/lock" bit (or ref. count) along with valid bit
69  */
70 
71 #include "opt_inet.h"
72 #include "opt_carp.h"
73 
74 #include <sys/param.h>
75 #include <sys/kernel.h>
76 #include <sys/queue.h>
77 #include <sys/sysctl.h>
78 #include <sys/systm.h>
79 #include <sys/mbuf.h>
80 #include <sys/malloc.h>
81 #include <sys/socket.h>
82 #include <sys/syslog.h>
83 #include <sys/lock.h>
84 
85 #include <net/if.h>
86 #include <net/if_dl.h>
87 #include <net/if_types.h>
88 #include <net/route.h>
89 #include <net/netisr.h>
90 #include <net/if_llc.h>
91 
92 #include <netinet/in.h>
93 #include <netinet/in_var.h>
94 #include <netinet/if_ether.h>
95 
96 #include <sys/thread2.h>
97 #include <sys/msgport2.h>
98 #include <net/netmsg2.h>
99 #include <net/netisr2.h>
100 #include <sys/mplock2.h>
101 
102 #ifdef CARP
103 #include <netinet/ip_carp.h>
104 #endif
105 
106 #define SIN(s) ((struct sockaddr_in *)s)
107 #define SDL(s) ((struct sockaddr_dl *)s)
108 
109 MALLOC_DEFINE(M_ARP, "arp", "ARP");
110 
111 SYSCTL_DECL(_net_link_ether);
112 SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
113 
114 /* timer values */
115 static int arpt_prune = (5*60*1); /* walk list every 5 minutes */
116 static int arpt_keep = (20*60); /* once resolved, good for 20 more minutes */
117 static int arpt_down = 20;	/* once declared down, don't send for 20 sec */
118 
119 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, prune_intvl, CTLFLAG_RW,
120 	   &arpt_prune, 0, "");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_RW,
122 	   &arpt_keep, 0, "");
123 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, host_down_time, CTLFLAG_RW,
124 	   &arpt_down, 0, "");
125 
126 #define	rt_expire	rt_rmx.rmx_expire
127 
128 struct llinfo_arp {
129 	LIST_ENTRY(llinfo_arp) la_le;
130 	struct	rtentry *la_rt;
131 	struct	mbuf *la_hold;	/* last packet until resolved/timeout */
132 	u_short	la_preempt;	/* countdown for pre-expiry arps */
133 	u_short	la_asked;	/* #times we QUERIED following expiration */
134 };
135 
136 static int	arp_maxtries = 5;
137 static int	useloopback = 1; /* use loopback interface for local traffic */
138 static int	arp_proxyall = 0;
139 static int	arp_refresh = 60; /* refresh arp cache ~60 (not impl yet) */
140 static int	arp_restricted_match = 0;
141 static int	arp_ignore_probes = 1;
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 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, ignore_probes, CTLFLAG_RW,
154 	   &arp_ignore_probes, 0, "Ignore ARP probes");
155 
156 static void	arp_rtrequest(int, struct rtentry *);
157 static void	arprequest(struct ifnet *, const struct in_addr *,
158 			   const struct in_addr *, const u_char *);
159 static void	arprequest_async(struct ifnet *, const struct in_addr *,
160 				 const struct in_addr *, const u_char *);
161 static void	arpintr(netmsg_t msg);
162 static void	arptfree(struct llinfo_arp *);
163 static void	arptimer(void *);
164 static struct llinfo_arp *
165 		arplookup(in_addr_t, boolean_t, boolean_t, boolean_t);
166 #ifdef INET
167 static void	in_arpinput(struct mbuf *);
168 static void	in_arpreply(struct mbuf *m, in_addr_t, in_addr_t);
169 static void	arp_update_msghandler(netmsg_t);
170 static void	arp_reply_msghandler(netmsg_t);
171 #endif
172 
173 struct arp_pcpu_data {
174 	LIST_HEAD(, llinfo_arp) llinfo_list;
175 	struct callout		timer_ch;
176 	struct netmsg_base	timer_nmsg;
177 };
178 
179 static struct arp_pcpu_data	*arp_data[MAXCPU];
180 
181 /*
182  * Timeout routine.  Age arp_tab entries periodically.
183  */
184 static void
185 arptimer_dispatch(netmsg_t nmsg)
186 {
187 	struct arp_pcpu_data *ad = nmsg->lmsg.u.ms_resultp;
188 	struct llinfo_arp *la, *nla;
189 
190 	ASSERT_NETISR_NCPUS(mycpuid);
191 
192 	/* Reply ASAP */
193 	crit_enter();
194 	netisr_replymsg(&nmsg->base, 0);
195 	crit_exit();
196 
197 	LIST_FOREACH_MUTABLE(la, &ad->llinfo_list, la_le, nla) {
198 		if (la->la_rt->rt_expire && la->la_rt->rt_expire <= time_uptime)
199 			arptfree(la);
200 	}
201 	callout_reset(&ad->timer_ch, arpt_prune * hz, arptimer, &ad->timer_nmsg);
202 }
203 
204 static void
205 arptimer(void *xnm)
206 {
207 	struct netmsg_base *nm = xnm;
208 
209 	KKASSERT(mycpuid < netisr_ncpus);
210 
211 	crit_enter();
212 	if (nm->lmsg.ms_flags & MSGF_DONE)
213 		netisr_sendmsg_oncpu(nm);
214 	crit_exit();
215 }
216 
217 /*
218  * Parallel to llc_rtrequest.
219  *
220  * Called after a route is successfully added to the tree to fix-up the
221  * route and initiate arp operations if required.
222  */
223 static void
224 arp_rtrequest(int req, struct rtentry *rt)
225 {
226 	struct sockaddr *gate = rt->rt_gateway;
227 	struct llinfo_arp *la = rt->rt_llinfo;
228 
229 	struct sockaddr_dl null_sdl = { sizeof null_sdl, AF_LINK };
230 
231 	if (rt->rt_flags & RTF_GATEWAY)
232 		return;
233 
234 	switch (req) {
235 	case RTM_ADD:
236 		/*
237 		 * XXX: If this is a manually added route to interface
238 		 * such as older version of routed or gated might provide,
239 		 * restore cloning bit.
240 		 */
241 		if (!(rt->rt_flags & RTF_HOST) &&
242 		    SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
243 			rt->rt_flags |= RTF_CLONING;
244 		if (rt->rt_flags & RTF_CLONING) {
245 			/*
246 			 * Case 1: This route should come from a route to iface.
247 			 */
248 			rt_setgate(rt, rt_key(rt),
249 				   (struct sockaddr *)&null_sdl,
250 				   RTL_DONTREPORT);
251 			gate = rt->rt_gateway;
252 			SDL(gate)->sdl_type = rt->rt_ifp->if_type;
253 			SDL(gate)->sdl_index = rt->rt_ifp->if_index;
254 			rt->rt_expire = time_uptime;
255 			break;
256 		}
257 		/*
258 		 * Announce a new entry if requested, and only announce it
259 		 * once on cpu0.
260 		 */
261 		if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
262 			arprequest(rt->rt_ifp,
263 			    &SIN(rt_key(rt))->sin_addr,
264 			    &SIN(rt_key(rt))->sin_addr,
265 			    LLADDR(SDL(gate)));
266 		}
267 		/*FALLTHROUGH*/
268 	case RTM_RESOLVE:
269 		if (gate->sa_family != AF_LINK ||
270 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
271 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
272 			break;
273 		}
274 		SDL(gate)->sdl_type = rt->rt_ifp->if_type;
275 		SDL(gate)->sdl_index = rt->rt_ifp->if_index;
276 		if (la != NULL)
277 			break; /* This happens on a route change */
278 		/*
279 		 * Case 2:  This route may come from cloning, or a manual route
280 		 * add with a LL address.
281 		 */
282 		R_Malloc(la, struct llinfo_arp *, sizeof *la);
283 		rt->rt_llinfo = la;
284 		if (la == NULL) {
285 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
286 			break;
287 		}
288 		bzero(la, sizeof *la);
289 		la->la_rt = rt;
290 		rt->rt_flags |= RTF_LLINFO;
291 		LIST_INSERT_HEAD(&arp_data[mycpuid]->llinfo_list, la, la_le);
292 
293 #ifdef INET
294 		/*
295 		 * This keeps the multicast addresses from showing up
296 		 * in `arp -a' listings as unresolved.  It's not actually
297 		 * functional.  Then the same for broadcast.
298 		 */
299 		if (IN_MULTICAST(ntohl(SIN(rt_key(rt))->sin_addr.s_addr))) {
300 			ETHER_MAP_IP_MULTICAST(&SIN(rt_key(rt))->sin_addr,
301 					       LLADDR(SDL(gate)));
302 			SDL(gate)->sdl_alen = 6;
303 			rt->rt_expire = 0;
304 		}
305 		if (in_broadcast(SIN(rt_key(rt))->sin_addr, rt->rt_ifp)) {
306 			memcpy(LLADDR(SDL(gate)), rt->rt_ifp->if_broadcastaddr,
307 			       rt->rt_ifp->if_addrlen);
308 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
309 			rt->rt_expire = 0;
310 		}
311 #endif
312 
313 		/*
314 		 * This fixes up the routing interface for local addresses.
315 		 * The route is adjusted to point at lo0 and the expiration
316 		 * timer is disabled.
317 		 *
318 		 * NOTE: This prevents locally targetted traffic from going
319 		 *	 out the hardware interface, which is inefficient
320 		 *	 and might not work if the hardware cannot listen
321 		 *	 to its own transmitted packets.   Setting
322 		 *	 net.link.ether.inet.useloopback to 0 will force
323 		 *	 packets for local addresses out the hardware (and
324 		 *	 it is expected to receive its own packet).
325 		 *
326 		 * XXX We should just be able to test RTF_LOCAL here instead
327 		 *     of having to compare IPs.
328 		 */
329 		if (SIN(rt_key(rt))->sin_addr.s_addr ==
330 		    (IA_SIN(rt->rt_ifa))->sin_addr.s_addr) {
331 			rt->rt_expire = 0;
332 			bcopy(IF_LLADDR(rt->rt_ifp), LLADDR(SDL(gate)),
333 			      SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen);
334 			if (useloopback)
335 				rt->rt_ifp = loif;
336 		}
337 		break;
338 
339 	case RTM_DELETE:
340 		if (la == NULL)
341 			break;
342 		LIST_REMOVE(la, la_le);
343 		rt->rt_llinfo = NULL;
344 		rt->rt_flags &= ~RTF_LLINFO;
345 		if (la->la_hold != NULL)
346 			m_freem(la->la_hold);
347 		Free(la);
348 		break;
349 	}
350 }
351 
352 static struct mbuf *
353 arpreq_alloc(struct ifnet *ifp, const struct in_addr *sip,
354 	     const struct in_addr *tip, const u_char *enaddr)
355 {
356 	struct mbuf *m;
357 	struct arphdr *ah;
358 	u_short ar_hrd;
359 
360 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
361 		return NULL;
362 	m->m_pkthdr.rcvif = NULL;
363 
364 	switch (ifp->if_type) {
365 	case IFT_ETHER:
366 		/*
367 		 * This may not be correct for types not explicitly
368 		 * listed, but this is our best guess
369 		 */
370 	default:
371 		ar_hrd = htons(ARPHRD_ETHER);
372 
373 		m->m_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
374 		m->m_pkthdr.len = m->m_len;
375 		MH_ALIGN(m, m->m_len);
376 
377 		ah = mtod(m, struct arphdr *);
378 		break;
379 	}
380 
381 	ah->ar_hrd = ar_hrd;
382 	ah->ar_pro = htons(ETHERTYPE_IP);
383 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
384 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
385 	ah->ar_op = htons(ARPOP_REQUEST);
386 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
387 	memset(ar_tha(ah), 0, ah->ar_hln);
388 	memcpy(ar_spa(ah), sip, ah->ar_pln);
389 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
390 
391 	return m;
392 }
393 
394 static void
395 arpreq_send(struct ifnet *ifp, struct mbuf *m)
396 {
397 	struct sockaddr sa;
398 	struct ether_header *eh;
399 
400 	ASSERT_NETISR_NCPUS(mycpuid);
401 
402 	switch (ifp->if_type) {
403 	case IFT_ETHER:
404 		/*
405 		 * This may not be correct for types not explicitly
406 		 * listed, but this is our best guess
407 		 */
408 	default:
409 		eh = (struct ether_header *)sa.sa_data;
410 		/* if_output() will not swap */
411 		eh->ether_type = htons(ETHERTYPE_ARP);
412 		memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ifp->if_addrlen);
413 		break;
414 	}
415 
416 	sa.sa_family = AF_UNSPEC;
417 	sa.sa_len = sizeof(sa);
418 	ifp->if_output(ifp, m, &sa, NULL);
419 }
420 
421 static void
422 arpreq_send_handler(netmsg_t msg)
423 {
424 	struct mbuf *m = msg->packet.nm_packet;
425 	struct ifnet *ifp = msg->lmsg.u.ms_resultp;
426 
427 	arpreq_send(ifp, m);
428 	/* nmsg was embedded in the mbuf, do not reply! */
429 }
430 
431 /*
432  * Broadcast an ARP request. Caller specifies:
433  *	- arp header source ip address
434  *	- arp header target ip address
435  *	- arp header source ethernet address
436  *
437  * NOTE: Caller MUST NOT hold ifp's serializer
438  */
439 static void
440 arprequest(struct ifnet *ifp, const struct in_addr *sip,
441 	   const struct in_addr *tip, const u_char *enaddr)
442 {
443 	struct mbuf *m;
444 
445 	ASSERT_NETISR_NCPUS(mycpuid);
446 
447 	if (enaddr == NULL) {
448 		if (ifp->if_bridge) {
449 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
450 		} else {
451 			enaddr = IF_LLADDR(ifp);
452 		}
453 	}
454 
455 	m = arpreq_alloc(ifp, sip, tip, enaddr);
456 	if (m == NULL)
457 		return;
458 	arpreq_send(ifp, m);
459 }
460 
461 /*
462  * Same as arprequest(), except:
463  * - Caller is allowed to hold ifp's serializer
464  * - Network output is done in protocol thead
465  */
466 static void
467 arprequest_async(struct ifnet *ifp, const struct in_addr *sip,
468 		 const struct in_addr *tip, const u_char *enaddr)
469 {
470 	struct mbuf *m;
471 	struct netmsg_packet *pmsg;
472 	int cpu;
473 
474 	if (enaddr == NULL) {
475 		if (ifp->if_bridge) {
476 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
477 		} else {
478 			enaddr = IF_LLADDR(ifp);
479 		}
480 	}
481 	m = arpreq_alloc(ifp, sip, tip, enaddr);
482 	if (m == NULL)
483 		return;
484 
485 	pmsg = &m->m_hdr.mh_netmsg;
486 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
487 		    0, arpreq_send_handler);
488 	pmsg->nm_packet = m;
489 	pmsg->base.lmsg.u.ms_resultp = ifp;
490 
491 	if (mycpuid < netisr_ncpus)
492 		cpu = mycpuid;
493 	else
494 		cpu = 0;
495 	lwkt_sendmsg(netisr_cpuport(cpu), &pmsg->base.lmsg);
496 }
497 
498 /*
499  * Resolve an IP address into an ethernet address.  If success,
500  * desten is filled in.  If there is no entry in arptab,
501  * set one up and broadcast a request for the IP address.
502  * Hold onto this mbuf and resend it once the address
503  * is finally resolved.  A return value of 1 indicates
504  * that desten has been filled in and the packet should be sent
505  * normally; a 0 return indicates that the packet has been
506  * taken over here, either now or for later transmission.
507  */
508 int
509 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
510 	   struct sockaddr *dst, u_char *desten)
511 {
512 	struct rtentry *rt = NULL;
513 	struct llinfo_arp *la = NULL;
514 	struct sockaddr_dl *sdl;
515 
516 	if (m->m_flags & M_BCAST) {	/* broadcast */
517 		memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen);
518 		return (1);
519 	}
520 	if (m->m_flags & M_MCAST) {/* multicast */
521 		ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
522 		return (1);
523 	}
524 	if (rt0 != NULL) {
525 		if (rt_llroute(dst, rt0, &rt) != 0) {
526 			m_freem(m);
527 			return 0;
528 		}
529 		la = rt->rt_llinfo;
530 	}
531 	if (la == NULL) {
532 		la = arplookup(SIN(dst)->sin_addr.s_addr,
533 			       TRUE, RTL_REPORTMSG, FALSE);
534 		if (la != NULL)
535 			rt = la->la_rt;
536 	}
537 	if (la == NULL || rt == NULL) {
538 		char addr[INET_ADDRSTRLEN];
539 
540 		log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s%s%s\n",
541 		    kinet_ntoa(SIN(dst)->sin_addr, addr), la ? "la" : " ",
542 		    rt ? "rt" : "");
543 		m_freem(m);
544 		return (0);
545 	}
546 	sdl = SDL(rt->rt_gateway);
547 	/*
548 	 * Check the address family and length is valid, the address
549 	 * is resolved; otherwise, try to resolve.
550 	 */
551 	if ((rt->rt_expire == 0 || rt->rt_expire > time_uptime) &&
552 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
553 		/*
554 		 * If entry has an expiry time and it is approaching,
555 		 * see if we need to send an ARP request within this
556 		 * arpt_down interval.
557 		 */
558 		if ((rt->rt_expire != 0) &&
559 		    (time_uptime + la->la_preempt > rt->rt_expire)) {
560 			arprequest(ifp,
561 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
562 				   &SIN(dst)->sin_addr,
563 				   NULL);
564 			la->la_preempt--;
565 		}
566 
567 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
568 		return 1;
569 	}
570 	/*
571 	 * If ARP is disabled or static on this interface, stop.
572 	 * XXX
573 	 * Probably should not allocate empty llinfo struct if we are
574 	 * not going to be sending out an arp request.
575 	 */
576 	if (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) {
577 		m_freem(m);
578 		return (0);
579 	}
580 	/*
581 	 * There is an arptab entry, but no ethernet address
582 	 * response yet.  Replace the held mbuf with this
583 	 * latest one.
584 	 */
585 	if (la->la_hold != NULL)
586 		m_freem(la->la_hold);
587 	la->la_hold = m;
588 	if (rt->rt_expire || ((rt->rt_flags & RTF_STATIC) && !sdl->sdl_alen)) {
589 		rt->rt_flags &= ~RTF_REJECT;
590 		if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
591 			rt->rt_expire = time_uptime;
592 			if (la->la_asked++ < arp_maxtries) {
593 				arprequest(ifp,
594 					   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
595 					   &SIN(dst)->sin_addr,
596 					   NULL);
597 			} else {
598 				rt->rt_flags |= RTF_REJECT;
599 				rt->rt_expire += arpt_down;
600 				la->la_asked = 0;
601 				la->la_preempt = arp_maxtries;
602 			}
603 		}
604 	}
605 	return (0);
606 }
607 
608 /*
609  * Common length and type checks are done here,
610  * then the protocol-specific routine is called.
611  */
612 static void
613 arpintr(netmsg_t msg)
614 {
615 	struct mbuf *m = msg->packet.nm_packet;
616 	struct arphdr *ar;
617 	u_short ar_hrd;
618 	char hexstr[6];
619 
620 	if (m->m_len < sizeof(struct arphdr) &&
621 	    (m = m_pullup(m, sizeof(struct arphdr))) == NULL) {
622 		log(LOG_ERR, "arp: runt packet -- m_pullup failed\n");
623 		return;
624 	}
625 	ar = mtod(m, struct arphdr *);
626 
627 	ar_hrd = ntohs(ar->ar_hrd);
628 	if (ar_hrd != ARPHRD_ETHER && ar_hrd != ARPHRD_IEEE802) {
629 		hexncpy((unsigned char *)&ar->ar_hrd, 2, hexstr, 5, NULL);
630 		log(LOG_ERR, "arp: unknown hardware address format (0x%s)\n",
631 		    hexstr);
632 		m_freem(m);
633 		return;
634 	}
635 
636 	if (m->m_pkthdr.len < arphdr_len(ar)) {
637 		if ((m = m_pullup(m, arphdr_len(ar))) == NULL) {
638 			log(LOG_ERR, "arp: runt packet\n");
639 			return;
640 		}
641 		ar = mtod(m, struct arphdr *);
642 	}
643 
644 	switch (ntohs(ar->ar_pro)) {
645 #ifdef INET
646 	case ETHERTYPE_IP:
647 		in_arpinput(m);
648 		return;
649 #endif
650 	}
651 	m_freem(m);
652 	/* msg was embedded in the mbuf, do not reply! */
653 }
654 
655 #ifdef INET
656 /*
657  * ARP for Internet protocols on 10 Mb/s Ethernet.
658  * Algorithm is that given in RFC 826.
659  * In addition, a sanity check is performed on the sender
660  * protocol address, to catch impersonators.
661  * We no longer handle negotiations for use of trailer protocol:
662  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
663  * along with IP replies if we wanted trailers sent to us,
664  * and also sent them in response to IP replies.
665  * This allowed either end to announce the desire to receive
666  * trailer packets.
667  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
668  * but formerly didn't normally send requests.
669  */
670 
671 static int	log_arp_wrong_iface = 1;
672 static int	log_arp_movements = 1;
673 static int	log_arp_permanent_modify = 1;
674 static int	log_arp_creation_failure = 1;
675 
676 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
677 	   &log_arp_wrong_iface, 0,
678 	   "Log arp packets arriving on the wrong interface");
679 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
680 	   &log_arp_movements, 0,
681 	   "Log arp replies from MACs different than the one in the cache");
682 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
683 	   &log_arp_permanent_modify, 0,
684 	   "Log arp replies from MACs different than the one "
685 	   "in the permanent arp entry");
686 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_creation_failure, CTLFLAG_RW,
687 	   &log_arp_creation_failure, 0, "Log arp creation failure");
688 
689 /*
690  * Returns non-zero if the routine updated anything.
691  */
692 static int
693 arp_update_oncpu(struct mbuf *m, in_addr_t saddr, boolean_t create,
694 		 boolean_t generate_report, boolean_t dologging)
695 {
696 	struct arphdr *ah = mtod(m, struct arphdr *);
697 	struct ifnet *ifp = m->m_pkthdr.rcvif;
698 	struct llinfo_arp *la;
699 	struct sockaddr_dl *sdl;
700 	struct rtentry *rt;
701 	char hexstr[2][64];
702 	char sbuf[INET_ADDRSTRLEN];
703 	int changed = create;
704 
705 	KASSERT(curthread->td_type == TD_TYPE_NETISR,
706 	    ("arp update not in netisr"));
707 
708 	la = arplookup(saddr, create, generate_report, FALSE);
709 	if (la && (rt = la->la_rt) && (sdl = SDL(rt->rt_gateway))) {
710 		struct in_addr isaddr = { saddr };
711 
712 		/*
713 		 * Normally arps coming in on the wrong interface are ignored,
714 		 * but if we are bridging and the two interfaces belong to
715 		 * the same bridge, or one is a member of the bridge which
716 		 * is the other, then it isn't an error.
717 		 */
718 		if (rt->rt_ifp != ifp) {
719 			/*
720 			 * (1) ifp and rt_ifp both members of same bridge
721 			 * (2) rt_ifp member of bridge ifp
722 			 * (3) ifp member of bridge rt_ifp
723 			 *
724 			 * Always replace rt_ifp with the bridge ifc.
725 			 */
726 			struct ifnet *nifp;
727 
728 			if (ifp->if_bridge &&
729 			    rt->rt_ifp->if_bridge == ifp->if_bridge) {
730 				nifp = ether_bridge_interface(ifp);
731 			} else if (rt->rt_ifp->if_bridge &&
732 				   ether_bridge_interface(rt->rt_ifp) == ifp) {
733 				nifp = ifp;
734 			} else if (ifp->if_bridge &&
735 				   ether_bridge_interface(ifp) == rt->rt_ifp) {
736 				nifp = rt->rt_ifp;
737 			} else {
738 				nifp = NULL;
739 			}
740 
741 			if ((log_arp_wrong_iface == 1 && nifp == NULL) ||
742 			    log_arp_wrong_iface == 2) {
743 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
744 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
745 				log(LOG_ERR,
746 				    "arp: %s is on %s "
747 				    "but got reply from %s on %s\n",
748 				    kinet_ntoa(isaddr, sbuf),
749 				    rt->rt_ifp->if_xname, hexstr[0],
750 				    ifp->if_xname);
751 			}
752 			if (nifp == NULL)
753 				return 0;
754 
755 			/*
756 			 * nifp is our man!  Replace rt_ifp and adjust
757 			 * the sdl.
758 			 */
759 			ifp = rt->rt_ifp = nifp;
760 			if (sdl->sdl_type != ifp->if_type) {
761 				sdl->sdl_type = ifp->if_type;
762 				changed = 1;
763 			}
764 			if (sdl->sdl_index != ifp->if_index) {
765 				sdl->sdl_index = ifp->if_index;
766 				changed = 1;
767 			}
768 		}
769 		if (sdl->sdl_alen &&
770 		    bcmp(ar_sha(ah), LLADDR(sdl), sdl->sdl_alen)) {
771 			changed = 1;
772 			if (rt->rt_expire != 0) {
773 				if (dologging && log_arp_movements) {
774 					hexncpy((u_char *)LLADDR(sdl), ifp->if_addrlen,
775 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
776 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
777 					    hexstr[1], HEX_NCPYLEN(ifp->if_addrlen), ":");
778 			    		log(LOG_INFO,
779 					    "arp: %s moved from %s to %s on %s\n",
780 					    kinet_ntoa(isaddr, sbuf), hexstr[0], hexstr[1],
781 					    ifp->if_xname);
782 				}
783 			} else {
784 				if (dologging && log_arp_permanent_modify) {
785 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
786 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
787 					log(LOG_ERR,
788 					"arp: %s attempts to modify "
789 					"permanent entry for %s on %s\n",
790 					hexstr[0], kinet_ntoa(isaddr, sbuf), ifp->if_xname);
791 				}
792 				return changed;
793 			}
794 		}
795 		/*
796 		 * sanity check for the address length.
797 		 * XXX this does not work for protocols with variable address
798 		 * length. -is
799 		 */
800 		if (dologging && sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
801 			hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
802 			    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
803 			log(LOG_WARNING,
804 			    "arp from %s: new addr len %d, was %d",
805 			    hexstr[0], ah->ar_hln, sdl->sdl_alen);
806 		}
807 		if (ifp->if_addrlen != ah->ar_hln) {
808 			if (dologging) {
809 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
810 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
811 				log(LOG_WARNING,
812 				"arp from %s: addr len: new %d, i/f %d "
813 				"(ignored)", hexstr[0],
814 				ah->ar_hln, ifp->if_addrlen);
815 			}
816 			return changed;
817 		}
818 		memcpy(LLADDR(sdl), ar_sha(ah), sdl->sdl_alen = ah->ar_hln);
819 		if (rt->rt_expire != 0) {
820 			if (rt->rt_expire != time_uptime + arpt_keep &&
821 			    rt->rt_expire != time_uptime + arpt_keep - 1) {
822 				rt->rt_expire = time_uptime + arpt_keep;
823 				changed = 1;
824 			}
825 		}
826 		if (rt->rt_flags & RTF_REJECT) {
827 			rt->rt_flags &= ~RTF_REJECT;
828 			changed = 1;
829 		}
830 		if (la->la_asked != 0) {
831 			la->la_asked = 0;
832 			changed = 1;
833 		}
834 		if (la->la_preempt != arp_maxtries) {
835 			la->la_preempt = arp_maxtries;
836 			changed = 1;
837 		}
838 
839 		/*
840 		 * This particular cpu might have been holding an mbuf
841 		 * pending ARP resolution.  If so, transmit the mbuf now.
842 		 */
843 		if (la->la_hold != NULL) {
844 			struct mbuf *m = la->la_hold;
845 
846 			la->la_hold = NULL;
847 			m_adj(m, sizeof(struct ether_header));
848 			ifp->if_output(ifp, m, rt_key(rt), rt);
849 			changed = 1;
850 		}
851 	}
852 	return changed;
853 }
854 
855 /*
856  * Called from arpintr() - this routine is run from a single cpu.
857  */
858 static void
859 in_arpinput(struct mbuf *m)
860 {
861 	struct arphdr *ah;
862 	struct ifnet *ifp = m->m_pkthdr.rcvif;
863 	struct ifaddr_container *ifac;
864 	struct in_ifaddr_container *iac;
865 	struct in_ifaddr *ia = NULL;
866 	struct in_addr isaddr, itaddr, myaddr;
867 	uint8_t *enaddr = NULL;
868 	int req_len;
869 	int changed;
870 	char hexstr[64], sbuf[INET_ADDRSTRLEN];
871 
872 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
873 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
874 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
875 		return;
876 	}
877 
878 	ah = mtod(m, struct arphdr *);
879 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
880 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
881 
882 	/*
883 	 * Check both target and sender IP addresses:
884 	 *
885 	 * If we receive the packet on the interface owning the address,
886 	 * then accept the address.
887 	 *
888 	 * For a bridge, we accept the address if the receive interface and
889 	 * the interface owning the address are on the same bridge, and
890 	 * use the bridge MAC as the is-at response.  The bridge will be
891 	 * responsible for handling the packet.
892 	 *
893 	 * (0) Check target IP against CARP IPs
894 	 */
895 #ifdef CARP
896 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
897 		int is_match = 0, is_parent = 0;
898 
899 		ia = iac->ia;
900 
901 		/* Skip all ia's which don't match */
902 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
903 			continue;
904 
905 		if (ia->ia_ifp->if_type != IFT_CARP)
906 			continue;
907 
908 		if (carp_parent(ia->ia_ifp) == ifp)
909 			is_parent = 1;
910 		if (is_parent || ia->ia_ifp == ifp)
911 			is_match = carp_iamatch(ia);
912 
913 		if (is_match) {
914 			if (is_parent) {
915 				/*
916 				 * The parent interface will also receive
917 				 * the ethernet broadcast packets, e.g. ARP
918 				 * REQUEST, so if we could find a CARP
919 				 * interface of the parent that could match
920 				 * the target IP address, we then drop the
921 				 * packets, which is delieverd to us through
922 				 * the parent interface.
923 				 */
924 				m_freem(m);
925 				return;
926 			}
927 			goto match;
928 		}
929 	}
930 #endif	/* CARP */
931 
932 	/*
933 	 * (1) Check target IP against our local IPs
934 	 */
935 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
936 		ia = iac->ia;
937 
938 		/* Skip all ia's which don't match */
939 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
940 			continue;
941 
942 #ifdef CARP
943 		/* CARP interfaces are checked in (0) */
944 		if (ia->ia_ifp->if_type == IFT_CARP)
945 			continue;
946 #endif
947 
948 		if (ifp->if_bridge && ia->ia_ifp &&
949 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
950 			ifp = ether_bridge_interface(ifp);
951 			goto match;
952 		}
953 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
954 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
955 			goto match;
956 		}
957 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
958 		    ia->ia_ifp) {
959 			goto match;
960 		}
961 		if (ia->ia_ifp == ifp) {
962 			goto match;
963 		}
964 	}
965 
966 	/*
967 	 * (2) Check sender IP against our local IPs
968 	 */
969 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
970 		ia = iac->ia;
971 
972 		/* Skip all ia's which don't match */
973 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
974 			continue;
975 
976 		if (ifp->if_bridge && ia->ia_ifp &&
977 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
978 			ifp = ether_bridge_interface(ifp);
979 			goto match;
980 		}
981 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
982 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
983 			goto match;
984 		}
985 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
986 		    ia->ia_ifp) {
987 			goto match;
988 		}
989 
990 		if (ia->ia_ifp == ifp)
991 			goto match;
992 	}
993 
994 	/*
995 	 * No match, use the first inet address on the receive interface
996 	 * as a dummy address for the rest of the function.
997 	 */
998 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
999 		struct ifaddr *ifa = ifac->ifa;
1000 
1001 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
1002 			ia = ifatoia(ifa);
1003 			goto match;
1004 		}
1005 	}
1006 
1007 	/*
1008 	 * If we got here, we didn't find any suitable interface,
1009 	 * so drop the packet.
1010 	 */
1011 	m_freem(m);
1012 	return;
1013 
1014 match:
1015 	if (!enaddr)
1016 		enaddr = (uint8_t *)IF_LLADDR(ifp);
1017 	myaddr = ia->ia_addr.sin_addr;
1018 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
1019 		m_freem(m);	/* it's from me, ignore it. */
1020 		return;
1021 	}
1022 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1023 		log(LOG_ERR,
1024 		    "arp: link address is broadcast for IP address %s!\n",
1025 		    kinet_ntoa(isaddr, sbuf));
1026 		m_freem(m);
1027 		return;
1028 	}
1029 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1030 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1031 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1032 		log(LOG_ERR,
1033 		   "arp: %s is using my IP address %s!\n",
1034 		    hexstr, kinet_ntoa(isaddr, sbuf));
1035 		itaddr = myaddr;
1036 		goto reply;
1037 	}
1038 	if (ifp->if_flags & IFF_STATICARP)
1039 		goto reply;
1040 
1041 	/*
1042 	 * When arp_restricted_match is true and the ARP response is not
1043 	 * specifically targetted to me, ignore it.  Otherwise the entry
1044 	 * timeout may be updated for an old MAC.
1045 	 */
1046 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1047 		m_freem(m);
1048 		return;
1049 	}
1050 
1051 	/*
1052 	 * Update all CPU's routing tables with this ARP packet.
1053 	 *
1054 	 * However, we only need to generate rtmsg on CPU0.
1055 	 */
1056 	ASSERT_NETISR0;
1057 	changed = arp_update_oncpu(m, isaddr.s_addr,
1058 				   itaddr.s_addr == myaddr.s_addr,
1059 				   RTL_REPORTMSG, TRUE);
1060 
1061 	if (netisr_ncpus > 1 && changed) {
1062 		struct netmsg_inarp *msg = &m->m_hdr.mh_arpmsg;
1063 
1064 		netmsg_init(&msg->base, NULL, &netisr_apanic_rport,
1065 			    0, arp_update_msghandler);
1066 		msg->m = m;
1067 		msg->saddr = isaddr.s_addr;
1068 		msg->taddr = itaddr.s_addr;
1069 		msg->myaddr = myaddr.s_addr;
1070 		lwkt_sendmsg(netisr_cpuport(1), &msg->base.lmsg);
1071 	} else {
1072 		goto reply;
1073 	}
1074 
1075 	/*
1076 	 * Just return here; after all CPUs's routing tables are
1077 	 * properly updated by this ARP packet, an ARP reply will
1078 	 * be generated if appropriate.
1079 	 */
1080 	return;
1081 reply:
1082 	in_arpreply(m, itaddr.s_addr, myaddr.s_addr);
1083 }
1084 
1085 static void
1086 arp_reply_msghandler(netmsg_t msg)
1087 {
1088 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1089 
1090 	in_arpreply(rmsg->m, rmsg->taddr, rmsg->myaddr);
1091 	/* Don't reply this netmsg; netmsg_inarp is embedded in mbuf */
1092 }
1093 
1094 static void
1095 arp_update_msghandler(netmsg_t msg)
1096 {
1097 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1098 	int nextcpu;
1099 
1100 	ASSERT_NETISR_NCPUS(mycpuid);
1101 
1102 	/*
1103 	 * This message handler will be called on all of the APs;
1104 	 * no need to generate rtmsg on them.
1105 	 */
1106 	KASSERT(mycpuid > 0, ("arp update msg on cpu%d", mycpuid));
1107 	arp_update_oncpu(rmsg->m, rmsg->saddr,
1108 			 rmsg->taddr == rmsg->myaddr,
1109 			 RTL_DONTREPORT, FALSE);
1110 
1111 	nextcpu = mycpuid + 1;
1112 	if (nextcpu < netisr_ncpus) {
1113 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
1114 	} else {
1115 		struct mbuf *m = rmsg->m;
1116 		in_addr_t saddr = rmsg->saddr;
1117 		in_addr_t taddr = rmsg->taddr;
1118 		in_addr_t myaddr = rmsg->myaddr;
1119 
1120 		/*
1121 		 * Dispatch this mbuf to netisr0 to perform ARP reply,
1122 		 * if appropriate.
1123 		 * NOTE: netmsg_inarp is embedded in this mbuf.
1124 		 */
1125 		netmsg_init(&rmsg->base, NULL, &netisr_apanic_rport,
1126 		    0, arp_reply_msghandler);
1127 		rmsg->m = m;
1128 		rmsg->saddr = saddr;
1129 		rmsg->taddr = taddr;
1130 		rmsg->myaddr = myaddr;
1131 		lwkt_sendmsg(netisr_cpuport(0), &rmsg->base.lmsg);
1132 	}
1133 }
1134 
1135 static void
1136 in_arpreply(struct mbuf *m, in_addr_t taddr, in_addr_t myaddr)
1137 {
1138 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1139 	const uint8_t *enaddr;
1140 	struct arphdr *ah;
1141 	struct sockaddr sa;
1142 	struct ether_header *eh;
1143 
1144 	ASSERT_NETISR0;
1145 
1146 	ah = mtod(m, struct arphdr *);
1147 	if (ntohs(ah->ar_op) != ARPOP_REQUEST) {
1148 		m_freem(m);
1149 		return;
1150 	}
1151 
1152 	enaddr = (const uint8_t *)IF_LLADDR(ifp);
1153 	if (taddr == myaddr) {
1154 		/* I am the target */
1155 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1156 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1157 	} else {
1158 		struct llinfo_arp *la;
1159 		struct rtentry *rt;
1160 
1161 		la = arplookup(taddr, FALSE, RTL_DONTREPORT, SIN_PROXY);
1162 		if (la == NULL) {
1163 			struct sockaddr_in sin;
1164 #ifdef DEBUG_PROXY
1165 			char tbuf[INET_ADDRSTRLEN];
1166 #endif
1167 
1168 			if (!arp_proxyall) {
1169 				m_freem(m);
1170 				return;
1171 			}
1172 
1173 			bzero(&sin, sizeof sin);
1174 			sin.sin_family = AF_INET;
1175 			sin.sin_len = sizeof sin;
1176 			sin.sin_addr.s_addr = taddr;
1177 
1178 			rt = rtpurelookup((struct sockaddr *)&sin);
1179 			if (rt == NULL) {
1180 				m_freem(m);
1181 				return;
1182 			}
1183 			--rt->rt_refcnt;
1184 			/*
1185 			 * Don't send proxies for nodes on the same interface
1186 			 * as this one came out of, or we'll get into a fight
1187 			 * over who claims what Ether address.
1188 			 */
1189 			if (rt->rt_ifp == ifp) {
1190 				m_freem(m);
1191 				return;
1192 			}
1193 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1194 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1195 #ifdef DEBUG_PROXY
1196 			kprintf("arp: proxying for %s\n",
1197 			    kinet_ntoa(itaddr, tbuf));
1198 #endif
1199 		} else {
1200 			struct sockaddr_dl *sdl;
1201 
1202 			rt = la->la_rt;
1203 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1204 			sdl = SDL(rt->rt_gateway);
1205 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1206 		}
1207 	}
1208 
1209 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1210 	memcpy(ar_spa(ah), &taddr, ah->ar_pln);
1211 	ah->ar_op = htons(ARPOP_REPLY);
1212 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1213 	switch (ifp->if_type) {
1214 	case IFT_ETHER:
1215 		/*
1216 		 * May not be correct for types not explictly
1217 		 * listed, but it is our best guess.
1218 		 */
1219 	default:
1220 		eh = (struct ether_header *)sa.sa_data;
1221 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1222 		eh->ether_type = htons(ETHERTYPE_ARP);
1223 		break;
1224 	}
1225 	sa.sa_family = AF_UNSPEC;
1226 	sa.sa_len = sizeof sa;
1227 	ifp->if_output(ifp, m, &sa, NULL);
1228 }
1229 
1230 #endif	/* INET */
1231 
1232 /*
1233  * Free an arp entry.  If the arp entry is actively referenced or represents
1234  * a static entry we only clear it back to an unresolved state, otherwise
1235  * we destroy the entry entirely.
1236  *
1237  * Note that static entries are created when route add ... -interface is used
1238  * to create an interface route to a (direct) destination.
1239  */
1240 static void
1241 arptfree(struct llinfo_arp *la)
1242 {
1243 	struct rtentry *rt = la->la_rt;
1244 	struct sockaddr_dl *sdl;
1245 
1246 	if (rt == NULL)
1247 		panic("arptfree");
1248 	sdl = SDL(rt->rt_gateway);
1249 	if (sdl != NULL &&
1250 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1251 	     (rt->rt_flags & RTF_STATIC))) {
1252 		sdl->sdl_alen = 0;
1253 		la->la_preempt = la->la_asked = 0;
1254 		rt->rt_flags &= ~RTF_REJECT;
1255 		return;
1256 	}
1257 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1258 }
1259 
1260 /*
1261  * Lookup or enter a new address in arptab.
1262  */
1263 static struct llinfo_arp *
1264 arplookup(in_addr_t addr, boolean_t create, boolean_t generate_report,
1265 	  boolean_t proxy)
1266 {
1267 	struct rtentry *rt;
1268 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1269 	const char *why = NULL;
1270 
1271 	/* Check ARP probes, e.g. from Cisco switches. */
1272 	if (addr == INADDR_ANY && arp_ignore_probes)
1273 		return (NULL);
1274 
1275 	sin.sin_addr.s_addr = addr;
1276 	sin.sin_other = proxy ? SIN_PROXY : 0;
1277 	if (create) {
1278 		rt = _rtlookup((struct sockaddr *)&sin,
1279 			       generate_report, RTL_DOCLONE);
1280 	} else {
1281 		rt = rtpurelookup((struct sockaddr *)&sin);
1282 	}
1283 	if (rt == NULL)
1284 		return (NULL);
1285 	rt->rt_refcnt--;
1286 
1287 	if (rt->rt_flags & RTF_GATEWAY)
1288 		why = "host is not on local network";
1289 	else if (!(rt->rt_flags & RTF_LLINFO))
1290 		why = "could not allocate llinfo";
1291 	else if (rt->rt_gateway->sa_family != AF_LINK)
1292 		why = "gateway route is not ours";
1293 
1294 	if (why) {
1295 		if (create && log_arp_creation_failure) {
1296 			char abuf[INET_ADDRSTRLEN];
1297 
1298 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1299 			    kinet_ntoa(sin.sin_addr, abuf), why);
1300 		}
1301 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1302 			/* No references to this route.  Purge it. */
1303 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1304 				  rt_mask(rt), rt->rt_flags, NULL);
1305 		}
1306 		return (NULL);
1307 	}
1308 	return (rt->rt_llinfo);
1309 }
1310 
1311 void
1312 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1313 {
1314 	ifa->ifa_rtrequest = arp_rtrequest;
1315 	ifa->ifa_flags |= RTF_CLONING;
1316 }
1317 
1318 void
1319 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1320 {
1321 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1322 		if (IN_NETISR_NCPUS(mycpuid)) {
1323 			arprequest(ifp, &IA_SIN(ifa)->sin_addr,
1324 			    &IA_SIN(ifa)->sin_addr, NULL);
1325 		} else {
1326 			arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1327 			    &IA_SIN(ifa)->sin_addr, NULL);
1328 		}
1329 	}
1330 }
1331 
1332 static void
1333 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1334     enum ifaddr_event event, struct ifaddr *ifa)
1335 {
1336 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1337 		return;
1338 	if (ifa->ifa_addr->sa_family != AF_INET)
1339 		return;
1340 	if (event == IFADDR_EVENT_DELETE)
1341 		return;
1342 
1343 	/*
1344 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1345 	 * - If we are the CARP interface's parent, don't send gratuitous
1346 	 *   ARP to avoid unnecessary confusion.
1347 	 */
1348 #ifdef CARP
1349 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1350 #endif
1351 	{
1352 		arp_gratuitous(ifp, ifa);
1353 	}
1354 }
1355 
1356 static void
1357 arp_init_dispatch(netmsg_t nm)
1358 {
1359 	struct arp_pcpu_data *ad;
1360 
1361 	ASSERT_NETISR_NCPUS(mycpuid);
1362 
1363 	ad = kmalloc(sizeof(*ad), M_ARP, M_WAITOK | M_ZERO);
1364 
1365 	LIST_INIT(&ad->llinfo_list);
1366 	callout_init_mp(&ad->timer_ch);
1367 	netmsg_init(&ad->timer_nmsg, NULL, &netisr_adone_rport,
1368 	    MSGF_PRIORITY, arptimer_dispatch);
1369 	ad->timer_nmsg.lmsg.u.ms_resultp = ad;
1370 
1371 	arp_data[mycpuid] = ad;
1372 
1373 	callout_reset(&ad->timer_ch, hz, arptimer, &ad->timer_nmsg);
1374 
1375 	netisr_forwardmsg(&nm->base, mycpuid + 1);
1376 }
1377 
1378 static void
1379 arp_init(void)
1380 {
1381 	struct netmsg_base nm;
1382 
1383 	netmsg_init(&nm, NULL, &curthread->td_msgport, 0, arp_init_dispatch);
1384 	netisr_domsg_global(&nm);
1385 
1386 	netisr_register(NETISR_ARP, arpintr, NULL);
1387 
1388 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1389 	    EVENTHANDLER_PRI_LAST);
1390 }
1391 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1392