xref: /dflybsd-src/sys/netinet/if_ether.c (revision d638c6eedc81671c3ceddd06ef20463940cb6a43)
1 /*
2  * Copyright (c) 2004, 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey M. Hsu.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of The DragonFly Project nor the names of its
16  *    contributors may be used to endorse or promote products derived
17  *    from this software without specific, prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
27  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
28  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
29  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1988, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
62  * $FreeBSD: src/sys/netinet/if_ether.c,v 1.64.2.23 2003/04/11 07:23:15 fjoe Exp $
63  */
64 
65 /*
66  * Ethernet address resolution protocol.
67  * TODO:
68  *	add "inuse/lock" bit (or ref. count) along with valid bit
69  */
70 
71 #include "opt_inet.h"
72 #include "opt_carp.h"
73 
74 #include <sys/param.h>
75 #include <sys/kernel.h>
76 #include <sys/queue.h>
77 #include <sys/sysctl.h>
78 #include <sys/systm.h>
79 #include <sys/mbuf.h>
80 #include <sys/malloc.h>
81 #include <sys/socket.h>
82 #include <sys/syslog.h>
83 #include <sys/lock.h>
84 
85 #include <net/if.h>
86 #include <net/if_dl.h>
87 #include <net/if_types.h>
88 #include <net/route.h>
89 #include <net/netisr.h>
90 #include <net/if_llc.h>
91 
92 #include <netinet/in.h>
93 #include <netinet/in_var.h>
94 #include <netinet/if_ether.h>
95 
96 #include <sys/thread2.h>
97 #include <sys/msgport2.h>
98 #include <net/netmsg2.h>
99 #include <net/netisr2.h>
100 #include <sys/mplock2.h>
101 
102 #ifdef CARP
103 #include <netinet/ip_carp.h>
104 #endif
105 
106 #define SIN(s) ((struct sockaddr_in *)s)
107 #define SDL(s) ((struct sockaddr_dl *)s)
108 
109 SYSCTL_DECL(_net_link_ether);
110 SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
111 
112 /* timer values */
113 static int arpt_prune = (5*60*1); /* walk list every 5 minutes */
114 static int arpt_keep = (20*60); /* once resolved, good for 20 more minutes */
115 static int arpt_down = 20;	/* once declared down, don't send for 20 sec */
116 
117 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, prune_intvl, CTLFLAG_RW,
118 	   &arpt_prune, 0, "");
119 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_RW,
120 	   &arpt_keep, 0, "");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, host_down_time, CTLFLAG_RW,
122 	   &arpt_down, 0, "");
123 
124 #define	rt_expire	rt_rmx.rmx_expire
125 
126 struct llinfo_arp {
127 	LIST_ENTRY(llinfo_arp) la_le;
128 	struct	rtentry *la_rt;
129 	struct	mbuf *la_hold;	/* last packet until resolved/timeout */
130 	u_short	la_preempt;	/* countdown for pre-expiry arps */
131 	u_short	la_asked;	/* #times we QUERIED following expiration */
132 };
133 
134 static	LIST_HEAD(, llinfo_arp) llinfo_arp_list[MAXCPU];
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 
142 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_RW,
143 	   &arp_maxtries, 0, "ARP resolution attempts before returning error");
144 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, useloopback, CTLFLAG_RW,
145 	   &useloopback, 0, "Use the loopback interface for local traffic");
146 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_RW,
147 	   &arp_proxyall, 0, "Enable proxy ARP for all suitable requests");
148 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, restricted_match, CTLFLAG_RW,
149 	   &arp_restricted_match, 0, "Only match against the sender");
150 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, refresh, CTLFLAG_RW,
151 	   &arp_refresh, 0, "Preemptively refresh the ARP");
152 
153 static void	arp_rtrequest(int, struct rtentry *);
154 static void	arprequest(struct ifnet *, const struct in_addr *,
155 			   const struct in_addr *, const u_char *);
156 static void	arprequest_async(struct ifnet *, const struct in_addr *,
157 				 const struct in_addr *, const u_char *);
158 static void	arpintr(netmsg_t msg);
159 static void	arptfree(struct llinfo_arp *);
160 static void	arptimer(void *);
161 static struct llinfo_arp *
162 		arplookup(in_addr_t, boolean_t, boolean_t, boolean_t);
163 #ifdef INET
164 static void	in_arpinput(struct mbuf *);
165 static void	in_arpreply(struct mbuf *m, in_addr_t, in_addr_t);
166 static void	arp_update_msghandler(netmsg_t);
167 static void	arp_reply_msghandler(netmsg_t);
168 #endif
169 
170 struct arptimer_ctx {
171 	struct callout		timer_ch;
172 	struct netmsg_base	timer_nmsg;
173 	int			timer_inited;
174 } __cachealign;
175 
176 static struct arptimer_ctx	arptimer_context[MAXCPU];
177 
178 /*
179  * Timeout routine.  Age arp_tab entries periodically.
180  */
181 static void
182 arptimer_dispatch(netmsg_t nmsg)
183 {
184 	struct llinfo_arp *la, *nla;
185 	int cpuid = mycpuid;
186 
187 	/* Reply ASAP */
188 	crit_enter();
189 	lwkt_replymsg(&nmsg->lmsg, 0);
190 	crit_exit();
191 
192 	LIST_FOREACH_MUTABLE(la, &llinfo_arp_list[cpuid], la_le, nla) {
193 		if (la->la_rt->rt_expire && la->la_rt->rt_expire <= time_uptime)
194 			arptfree(la);
195 	}
196 	callout_reset(&arptimer_context[cpuid].timer_ch, arpt_prune * hz,
197 	    arptimer, NULL);
198 }
199 
200 static void
201 arptimer(void *arg __unused)
202 {
203 	int cpuid = mycpuid;
204 	struct lwkt_msg *lmsg = &arptimer_context[cpuid].timer_nmsg.lmsg;
205 
206 	crit_enter();
207 	if (lmsg->ms_flags & MSGF_DONE)
208 		lwkt_sendmsg_oncpu(netisr_cpuport(cpuid), lmsg);
209 	crit_exit();
210 }
211 
212 /*
213  * Parallel to llc_rtrequest.
214  *
215  * Called after a route is successfully added to the tree to fix-up the
216  * route and initiate arp operations if required.
217  */
218 static void
219 arp_rtrequest(int req, struct rtentry *rt)
220 {
221 	struct sockaddr *gate = rt->rt_gateway;
222 	struct llinfo_arp *la = rt->rt_llinfo;
223 
224 	struct sockaddr_dl null_sdl = { sizeof null_sdl, AF_LINK };
225 
226 	if (__predict_false(!arptimer_context[mycpuid].timer_inited)) {
227 		struct arptimer_ctx *ctx = &arptimer_context[mycpuid];
228 
229 		ctx->timer_inited = TRUE;
230 		netmsg_init(&ctx->timer_nmsg, NULL, &netisr_adone_rport,
231 		    MSGF_PRIORITY, arptimer_dispatch);
232 		callout_init_mp(&ctx->timer_ch);
233 		callout_reset(&ctx->timer_ch, hz, arptimer, NULL);
234 	}
235 	if (rt->rt_flags & RTF_GATEWAY)
236 		return;
237 
238 	switch (req) {
239 	case RTM_ADD:
240 		/*
241 		 * XXX: If this is a manually added route to interface
242 		 * such as older version of routed or gated might provide,
243 		 * restore cloning bit.
244 		 */
245 		if (!(rt->rt_flags & RTF_HOST) &&
246 		    SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
247 			rt->rt_flags |= RTF_CLONING;
248 		if (rt->rt_flags & RTF_CLONING) {
249 			/*
250 			 * Case 1: This route should come from a route to iface.
251 			 */
252 			rt_setgate(rt, rt_key(rt),
253 				   (struct sockaddr *)&null_sdl,
254 				   RTL_DONTREPORT);
255 			gate = rt->rt_gateway;
256 			SDL(gate)->sdl_type = rt->rt_ifp->if_type;
257 			SDL(gate)->sdl_index = rt->rt_ifp->if_index;
258 			rt->rt_expire = time_uptime;
259 			break;
260 		}
261 		/*
262 		 * Announce a new entry if requested, and only announce it
263 		 * once on cpu0.
264 		 */
265 		if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
266 			arprequest_async(rt->rt_ifp,
267 			    &SIN(rt_key(rt))->sin_addr,
268 			    &SIN(rt_key(rt))->sin_addr,
269 			    LLADDR(SDL(gate)));
270 		}
271 		/*FALLTHROUGH*/
272 	case RTM_RESOLVE:
273 		if (gate->sa_family != AF_LINK ||
274 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
275 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
276 			break;
277 		}
278 		SDL(gate)->sdl_type = rt->rt_ifp->if_type;
279 		SDL(gate)->sdl_index = rt->rt_ifp->if_index;
280 		if (la != NULL)
281 			break; /* This happens on a route change */
282 		/*
283 		 * Case 2:  This route may come from cloning, or a manual route
284 		 * add with a LL address.
285 		 */
286 		R_Malloc(la, struct llinfo_arp *, sizeof *la);
287 		rt->rt_llinfo = la;
288 		if (la == NULL) {
289 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
290 			break;
291 		}
292 		bzero(la, sizeof *la);
293 		la->la_rt = rt;
294 		rt->rt_flags |= RTF_LLINFO;
295 		LIST_INSERT_HEAD(&llinfo_arp_list[mycpuid], la, la_le);
296 
297 #ifdef INET
298 		/*
299 		 * This keeps the multicast addresses from showing up
300 		 * in `arp -a' listings as unresolved.  It's not actually
301 		 * functional.  Then the same for broadcast.
302 		 */
303 		if (IN_MULTICAST(ntohl(SIN(rt_key(rt))->sin_addr.s_addr))) {
304 			ETHER_MAP_IP_MULTICAST(&SIN(rt_key(rt))->sin_addr,
305 					       LLADDR(SDL(gate)));
306 			SDL(gate)->sdl_alen = 6;
307 			rt->rt_expire = 0;
308 		}
309 		if (in_broadcast(SIN(rt_key(rt))->sin_addr, rt->rt_ifp)) {
310 			memcpy(LLADDR(SDL(gate)), rt->rt_ifp->if_broadcastaddr,
311 			       rt->rt_ifp->if_addrlen);
312 			SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen;
313 			rt->rt_expire = 0;
314 		}
315 #endif
316 
317 		/*
318 		 * This fixes up the routing interface for local addresses.
319 		 * The route is adjusted to point at lo0 and the expiration
320 		 * timer is disabled.
321 		 *
322 		 * NOTE: This prevents locally targetted traffic from going
323 		 *	 out the hardware interface, which is inefficient
324 		 *	 and might not work if the hardware cannot listen
325 		 *	 to its own transmitted packets.   Setting
326 		 *	 net.link.ether.inet.useloopback to 0 will force
327 		 *	 packets for local addresses out the hardware (and
328 		 *	 it is expected to receive its own packet).
329 		 *
330 		 * XXX We should just be able to test RTF_LOCAL here instead
331 		 *     of having to compare IPs.
332 		 */
333 		if (SIN(rt_key(rt))->sin_addr.s_addr ==
334 		    (IA_SIN(rt->rt_ifa))->sin_addr.s_addr) {
335 			rt->rt_expire = 0;
336 			bcopy(IF_LLADDR(rt->rt_ifp), LLADDR(SDL(gate)),
337 			      SDL(gate)->sdl_alen = rt->rt_ifp->if_addrlen);
338 			if (useloopback)
339 				rt->rt_ifp = loif;
340 		}
341 		break;
342 
343 	case RTM_DELETE:
344 		if (la == NULL)
345 			break;
346 		LIST_REMOVE(la, la_le);
347 		rt->rt_llinfo = NULL;
348 		rt->rt_flags &= ~RTF_LLINFO;
349 		if (la->la_hold != NULL)
350 			m_freem(la->la_hold);
351 		Free(la);
352 		break;
353 	}
354 }
355 
356 static struct mbuf *
357 arpreq_alloc(struct ifnet *ifp, const struct in_addr *sip,
358 	     const struct in_addr *tip, const u_char *enaddr)
359 {
360 	struct mbuf *m;
361 	struct arphdr *ah;
362 	u_short ar_hrd;
363 
364 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
365 		return NULL;
366 	m->m_pkthdr.rcvif = NULL;
367 
368 	switch (ifp->if_type) {
369 	case IFT_ETHER:
370 		/*
371 		 * This may not be correct for types not explicitly
372 		 * listed, but this is our best guess
373 		 */
374 	default:
375 		ar_hrd = htons(ARPHRD_ETHER);
376 
377 		m->m_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
378 		m->m_pkthdr.len = m->m_len;
379 		MH_ALIGN(m, m->m_len);
380 
381 		ah = mtod(m, struct arphdr *);
382 		break;
383 	}
384 
385 	ah->ar_hrd = ar_hrd;
386 	ah->ar_pro = htons(ETHERTYPE_IP);
387 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
388 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
389 	ah->ar_op = htons(ARPOP_REQUEST);
390 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
391 	memset(ar_tha(ah), 0, ah->ar_hln);
392 	memcpy(ar_spa(ah), sip, ah->ar_pln);
393 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
394 
395 	return m;
396 }
397 
398 static void
399 arpreq_send(struct ifnet *ifp, struct mbuf *m)
400 {
401 	struct sockaddr sa;
402 	struct ether_header *eh;
403 
404 	switch (ifp->if_type) {
405 	case IFT_ETHER:
406 		/*
407 		 * This may not be correct for types not explicitly
408 		 * listed, but this is our best guess
409 		 */
410 	default:
411 		eh = (struct ether_header *)sa.sa_data;
412 		/* if_output() will not swap */
413 		eh->ether_type = htons(ETHERTYPE_ARP);
414 		memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ifp->if_addrlen);
415 		break;
416 	}
417 
418 	sa.sa_family = AF_UNSPEC;
419 	sa.sa_len = sizeof(sa);
420 	ifp->if_output(ifp, m, &sa, NULL);
421 }
422 
423 static void
424 arpreq_send_handler(netmsg_t msg)
425 {
426 	struct mbuf *m = msg->packet.nm_packet;
427 	struct ifnet *ifp = msg->lmsg.u.ms_resultp;
428 
429 	arpreq_send(ifp, m);
430 	/* nmsg was embedded in the mbuf, do not reply! */
431 }
432 
433 /*
434  * Broadcast an ARP request. Caller specifies:
435  *	- arp header source ip address
436  *	- arp header target ip address
437  *	- arp header source ethernet address
438  *
439  * NOTE: Caller MUST NOT hold ifp's serializer
440  */
441 static void
442 arprequest(struct ifnet *ifp, const struct in_addr *sip,
443 	   const struct in_addr *tip, const u_char *enaddr)
444 {
445 	struct mbuf *m;
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 
473 	if (enaddr == NULL) {
474 		if (ifp->if_bridge) {
475 			enaddr = IF_LLADDR(ether_bridge_interface(ifp));
476 		} else {
477 			enaddr = IF_LLADDR(ifp);
478 		}
479 	}
480 	m = arpreq_alloc(ifp, sip, tip, enaddr);
481 	if (m == NULL)
482 		return;
483 
484 	pmsg = &m->m_hdr.mh_netmsg;
485 	netmsg_init(&pmsg->base, NULL, &netisr_apanic_rport,
486 		    0, arpreq_send_handler);
487 	pmsg->nm_packet = m;
488 	pmsg->base.lmsg.u.ms_resultp = ifp;
489 
490 	lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), &pmsg->base.lmsg);
491 }
492 
493 /*
494  * Resolve an IP address into an ethernet address.  If success,
495  * desten is filled in.  If there is no entry in arptab,
496  * set one up and broadcast a request for the IP address.
497  * Hold onto this mbuf and resend it once the address
498  * is finally resolved.  A return value of 1 indicates
499  * that desten has been filled in and the packet should be sent
500  * normally; a 0 return indicates that the packet has been
501  * taken over here, either now or for later transmission.
502  */
503 int
504 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
505 	   struct sockaddr *dst, u_char *desten)
506 {
507 	struct rtentry *rt = NULL;
508 	struct llinfo_arp *la = NULL;
509 	struct sockaddr_dl *sdl;
510 
511 	if (m->m_flags & M_BCAST) {	/* broadcast */
512 		memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen);
513 		return (1);
514 	}
515 	if (m->m_flags & M_MCAST) {/* multicast */
516 		ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
517 		return (1);
518 	}
519 	if (rt0 != NULL) {
520 		if (rt_llroute(dst, rt0, &rt) != 0) {
521 			m_freem(m);
522 			return 0;
523 		}
524 		la = rt->rt_llinfo;
525 	}
526 	if (la == NULL) {
527 		la = arplookup(SIN(dst)->sin_addr.s_addr,
528 			       TRUE, RTL_REPORTMSG, FALSE);
529 		if (la != NULL)
530 			rt = la->la_rt;
531 	}
532 	if (la == NULL || rt == NULL) {
533 		char addr[INET_ADDRSTRLEN];
534 
535 		log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s%s%s\n",
536 		    kinet_ntoa(SIN(dst)->sin_addr, addr), la ? "la" : " ",
537 		    rt ? "rt" : "");
538 		m_freem(m);
539 		return (0);
540 	}
541 	sdl = SDL(rt->rt_gateway);
542 	/*
543 	 * Check the address family and length is valid, the address
544 	 * is resolved; otherwise, try to resolve.
545 	 */
546 	if ((rt->rt_expire == 0 || rt->rt_expire > time_uptime) &&
547 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
548 		/*
549 		 * If entry has an expiry time and it is approaching,
550 		 * see if we need to send an ARP request within this
551 		 * arpt_down interval.
552 		 */
553 		if ((rt->rt_expire != 0) &&
554 		    (time_uptime + la->la_preempt > rt->rt_expire)) {
555 			arprequest(ifp,
556 				   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
557 				   &SIN(dst)->sin_addr,
558 				   NULL);
559 			la->la_preempt--;
560 		}
561 
562 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
563 		return 1;
564 	}
565 	/*
566 	 * If ARP is disabled or static on this interface, stop.
567 	 * XXX
568 	 * Probably should not allocate empty llinfo struct if we are
569 	 * not going to be sending out an arp request.
570 	 */
571 	if (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) {
572 		m_freem(m);
573 		return (0);
574 	}
575 	/*
576 	 * There is an arptab entry, but no ethernet address
577 	 * response yet.  Replace the held mbuf with this
578 	 * latest one.
579 	 */
580 	if (la->la_hold != NULL)
581 		m_freem(la->la_hold);
582 	la->la_hold = m;
583 	if (rt->rt_expire || ((rt->rt_flags & RTF_STATIC) && !sdl->sdl_alen)) {
584 		rt->rt_flags &= ~RTF_REJECT;
585 		if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
586 			rt->rt_expire = time_uptime;
587 			if (la->la_asked++ < arp_maxtries) {
588 				arprequest(ifp,
589 					   &SIN(rt->rt_ifa->ifa_addr)->sin_addr,
590 					   &SIN(dst)->sin_addr,
591 					   NULL);
592 			} else {
593 				rt->rt_flags |= RTF_REJECT;
594 				rt->rt_expire += arpt_down;
595 				la->la_asked = 0;
596 				la->la_preempt = arp_maxtries;
597 			}
598 		}
599 	}
600 	return (0);
601 }
602 
603 /*
604  * Common length and type checks are done here,
605  * then the protocol-specific routine is called.
606  */
607 static void
608 arpintr(netmsg_t msg)
609 {
610 	struct mbuf *m = msg->packet.nm_packet;
611 	struct arphdr *ar;
612 	u_short ar_hrd;
613 	char hexstr[6];
614 
615 	if (m->m_len < sizeof(struct arphdr) &&
616 	    (m = m_pullup(m, sizeof(struct arphdr))) == NULL) {
617 		log(LOG_ERR, "arp: runt packet -- m_pullup failed\n");
618 		return;
619 	}
620 	ar = mtod(m, struct arphdr *);
621 
622 	ar_hrd = ntohs(ar->ar_hrd);
623 	if (ar_hrd != ARPHRD_ETHER && ar_hrd != ARPHRD_IEEE802) {
624 		hexncpy((unsigned char *)&ar->ar_hrd, 2, hexstr, 5, NULL);
625 		log(LOG_ERR, "arp: unknown hardware address format (0x%s)\n",
626 		    hexstr);
627 		m_freem(m);
628 		return;
629 	}
630 
631 	if (m->m_pkthdr.len < arphdr_len(ar)) {
632 		if ((m = m_pullup(m, arphdr_len(ar))) == NULL) {
633 			log(LOG_ERR, "arp: runt packet\n");
634 			return;
635 		}
636 		ar = mtod(m, struct arphdr *);
637 	}
638 
639 	switch (ntohs(ar->ar_pro)) {
640 #ifdef INET
641 	case ETHERTYPE_IP:
642 		in_arpinput(m);
643 		return;
644 #endif
645 	}
646 	m_freem(m);
647 	/* msg was embedded in the mbuf, do not reply! */
648 }
649 
650 #ifdef INET
651 /*
652  * ARP for Internet protocols on 10 Mb/s Ethernet.
653  * Algorithm is that given in RFC 826.
654  * In addition, a sanity check is performed on the sender
655  * protocol address, to catch impersonators.
656  * We no longer handle negotiations for use of trailer protocol:
657  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
658  * along with IP replies if we wanted trailers sent to us,
659  * and also sent them in response to IP replies.
660  * This allowed either end to announce the desire to receive
661  * trailer packets.
662  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
663  * but formerly didn't normally send requests.
664  */
665 
666 static int	log_arp_wrong_iface = 1;
667 static int	log_arp_movements = 1;
668 static int	log_arp_permanent_modify = 1;
669 static int	log_arp_creation_failure = 1;
670 
671 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
672 	   &log_arp_wrong_iface, 0,
673 	   "Log arp packets arriving on the wrong interface");
674 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
675 	   &log_arp_movements, 0,
676 	   "Log arp replies from MACs different than the one in the cache");
677 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
678 	   &log_arp_permanent_modify, 0,
679 	   "Log arp replies from MACs different than the one "
680 	   "in the permanent arp entry");
681 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_creation_failure, CTLFLAG_RW,
682 	   &log_arp_creation_failure, 0, "Log arp creation failure");
683 
684 /*
685  * Returns non-zero if the routine updated anything.
686  */
687 static int
688 arp_update_oncpu(struct mbuf *m, in_addr_t saddr, boolean_t create,
689 		 boolean_t generate_report, boolean_t dologging)
690 {
691 	struct arphdr *ah = mtod(m, struct arphdr *);
692 	struct ifnet *ifp = m->m_pkthdr.rcvif;
693 	struct llinfo_arp *la;
694 	struct sockaddr_dl *sdl;
695 	struct rtentry *rt;
696 	char hexstr[2][64];
697 	char sbuf[INET_ADDRSTRLEN];
698 	int changed = create;
699 
700 	KASSERT(curthread->td_type == TD_TYPE_NETISR,
701 	    ("arp update not in netisr"));
702 
703 	la = arplookup(saddr, create, generate_report, FALSE);
704 	if (la && (rt = la->la_rt) && (sdl = SDL(rt->rt_gateway))) {
705 		struct in_addr isaddr = { saddr };
706 
707 		/*
708 		 * Normally arps coming in on the wrong interface are ignored,
709 		 * but if we are bridging and the two interfaces belong to
710 		 * the same bridge, or one is a member of the bridge which
711 		 * is the other, then it isn't an error.
712 		 */
713 		if (rt->rt_ifp != ifp) {
714 			/*
715 			 * (1) ifp and rt_ifp both members of same bridge
716 			 * (2) rt_ifp member of bridge ifp
717 			 * (3) ifp member of bridge rt_ifp
718 			 *
719 			 * Always replace rt_ifp with the bridge ifc.
720 			 */
721 			struct ifnet *nifp;
722 
723 			if (ifp->if_bridge &&
724 			    rt->rt_ifp->if_bridge == ifp->if_bridge) {
725 				nifp = ether_bridge_interface(ifp);
726 			} else if (rt->rt_ifp->if_bridge &&
727 				   ether_bridge_interface(rt->rt_ifp) == ifp) {
728 				nifp = ifp;
729 			} else if (ifp->if_bridge &&
730 				   ether_bridge_interface(ifp) == rt->rt_ifp) {
731 				nifp = rt->rt_ifp;
732 			} else {
733 				nifp = NULL;
734 			}
735 
736 			if ((log_arp_wrong_iface == 1 && nifp == NULL) ||
737 			    log_arp_wrong_iface == 2) {
738 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
739 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
740 				log(LOG_ERR,
741 				    "arp: %s is on %s "
742 				    "but got reply from %s on %s\n",
743 				    kinet_ntoa(isaddr, sbuf),
744 				    rt->rt_ifp->if_xname, hexstr[0],
745 				    ifp->if_xname);
746 			}
747 			if (nifp == NULL)
748 				return 0;
749 
750 			/*
751 			 * nifp is our man!  Replace rt_ifp and adjust
752 			 * the sdl.
753 			 */
754 			ifp = rt->rt_ifp = nifp;
755 			if (sdl->sdl_type != ifp->if_type) {
756 				sdl->sdl_type = ifp->if_type;
757 				changed = 1;
758 			}
759 			if (sdl->sdl_index != ifp->if_index) {
760 				sdl->sdl_index = ifp->if_index;
761 				changed = 1;
762 			}
763 		}
764 		if (sdl->sdl_alen &&
765 		    bcmp(ar_sha(ah), LLADDR(sdl), sdl->sdl_alen)) {
766 			changed = 1;
767 			if (rt->rt_expire != 0) {
768 				if (dologging && log_arp_movements) {
769 					hexncpy((u_char *)LLADDR(sdl), ifp->if_addrlen,
770 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
771 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
772 					    hexstr[1], HEX_NCPYLEN(ifp->if_addrlen), ":");
773 			    		log(LOG_INFO,
774 					    "arp: %s moved from %s to %s on %s\n",
775 					    kinet_ntoa(isaddr, sbuf), hexstr[0], hexstr[1],
776 					    ifp->if_xname);
777 				}
778 			} else {
779 				if (dologging && log_arp_permanent_modify) {
780 					hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
781 					    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
782 					log(LOG_ERR,
783 					"arp: %s attempts to modify "
784 					"permanent entry for %s on %s\n",
785 					hexstr[0], kinet_ntoa(isaddr, sbuf), ifp->if_xname);
786 				}
787 				return changed;
788 			}
789 		}
790 		/*
791 		 * sanity check for the address length.
792 		 * XXX this does not work for protocols with variable address
793 		 * length. -is
794 		 */
795 		if (dologging && sdl->sdl_alen && sdl->sdl_alen != ah->ar_hln) {
796 			hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
797 			    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
798 			log(LOG_WARNING,
799 			    "arp from %s: new addr len %d, was %d",
800 			    hexstr[0], ah->ar_hln, sdl->sdl_alen);
801 		}
802 		if (ifp->if_addrlen != ah->ar_hln) {
803 			if (dologging) {
804 				hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
805 				    hexstr[0], HEX_NCPYLEN(ifp->if_addrlen), ":");
806 				log(LOG_WARNING,
807 				"arp from %s: addr len: new %d, i/f %d "
808 				"(ignored)", hexstr[0],
809 				ah->ar_hln, ifp->if_addrlen);
810 			}
811 			return changed;
812 		}
813 		memcpy(LLADDR(sdl), ar_sha(ah), sdl->sdl_alen = ah->ar_hln);
814 		if (rt->rt_expire != 0) {
815 			if (rt->rt_expire != time_uptime + arpt_keep &&
816 			    rt->rt_expire != time_uptime + arpt_keep - 1) {
817 				rt->rt_expire = time_uptime + arpt_keep;
818 				changed = 1;
819 			}
820 		}
821 		if (rt->rt_flags & RTF_REJECT) {
822 			rt->rt_flags &= ~RTF_REJECT;
823 			changed = 1;
824 		}
825 		if (la->la_asked != 0) {
826 			la->la_asked = 0;
827 			changed = 1;
828 		}
829 		if (la->la_preempt != arp_maxtries) {
830 			la->la_preempt = arp_maxtries;
831 			changed = 1;
832 		}
833 
834 		/*
835 		 * This particular cpu might have been holding an mbuf
836 		 * pending ARP resolution.  If so, transmit the mbuf now.
837 		 */
838 		if (la->la_hold != NULL) {
839 			struct mbuf *m = la->la_hold;
840 
841 			la->la_hold = NULL;
842 			m_adj(m, sizeof(struct ether_header));
843 			ifp->if_output(ifp, m, rt_key(rt), rt);
844 			changed = 1;
845 		}
846 	}
847 	return changed;
848 }
849 
850 /*
851  * Called from arpintr() - this routine is run from a single cpu.
852  */
853 static void
854 in_arpinput(struct mbuf *m)
855 {
856 	struct arphdr *ah;
857 	struct ifnet *ifp = m->m_pkthdr.rcvif;
858 	struct ifaddr_container *ifac;
859 	struct in_ifaddr_container *iac;
860 	struct in_ifaddr *ia = NULL;
861 	struct in_addr isaddr, itaddr, myaddr;
862 	uint8_t *enaddr = NULL;
863 	int req_len;
864 	int changed;
865 	char hexstr[64], sbuf[INET_ADDRSTRLEN];
866 
867 	req_len = arphdr_len2(ifp->if_addrlen, sizeof(struct in_addr));
868 	if (m->m_len < req_len && (m = m_pullup(m, req_len)) == NULL) {
869 		log(LOG_ERR, "in_arp: runt packet -- m_pullup failed\n");
870 		return;
871 	}
872 
873 	ah = mtod(m, struct arphdr *);
874 	memcpy(&isaddr, ar_spa(ah), sizeof isaddr);
875 	memcpy(&itaddr, ar_tpa(ah), sizeof itaddr);
876 
877 	/*
878 	 * Check both target and sender IP addresses:
879 	 *
880 	 * If we receive the packet on the interface owning the address,
881 	 * then accept the address.
882 	 *
883 	 * For a bridge, we accept the address if the receive interface and
884 	 * the interface owning the address are on the same bridge, and
885 	 * use the bridge MAC as the is-at response.  The bridge will be
886 	 * responsible for handling the packet.
887 	 *
888 	 * (0) Check target IP against CARP IPs
889 	 */
890 #ifdef CARP
891 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
892 		int is_match = 0, is_parent = 0;
893 
894 		ia = iac->ia;
895 
896 		/* Skip all ia's which don't match */
897 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
898 			continue;
899 
900 		if (ia->ia_ifp->if_type != IFT_CARP)
901 			continue;
902 
903 		if (carp_parent(ia->ia_ifp) == ifp)
904 			is_parent = 1;
905 		if (is_parent || ia->ia_ifp == ifp)
906 			is_match = carp_iamatch(ia);
907 
908 		if (is_match) {
909 			if (is_parent) {
910 				/*
911 				 * The parent interface will also receive
912 				 * the ethernet broadcast packets, e.g. ARP
913 				 * REQUEST, so if we could find a CARP
914 				 * interface of the parent that could match
915 				 * the target IP address, we then drop the
916 				 * packets, which is delieverd to us through
917 				 * the parent interface.
918 				 */
919 				m_freem(m);
920 				return;
921 			}
922 			goto match;
923 		}
924 	}
925 #endif	/* CARP */
926 
927 	/*
928 	 * (1) Check target IP against our local IPs
929 	 */
930 	LIST_FOREACH(iac, INADDR_HASH(itaddr.s_addr), ia_hash) {
931 		ia = iac->ia;
932 
933 		/* Skip all ia's which don't match */
934 		if (itaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
935 			continue;
936 
937 #ifdef CARP
938 		/* CARP interfaces are checked in (0) */
939 		if (ia->ia_ifp->if_type == IFT_CARP)
940 			continue;
941 #endif
942 
943 		if (ifp->if_bridge && ia->ia_ifp &&
944 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
945 			ifp = ether_bridge_interface(ifp);
946 			goto match;
947 		}
948 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
949 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
950 			goto match;
951 		}
952 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
953 		    ia->ia_ifp) {
954 			goto match;
955 		}
956 		if (ia->ia_ifp == ifp) {
957 			goto match;
958 		}
959 	}
960 
961 	/*
962 	 * (2) Check sender IP against our local IPs
963 	 */
964 	LIST_FOREACH(iac, INADDR_HASH(isaddr.s_addr), ia_hash) {
965 		ia = iac->ia;
966 
967 		/* Skip all ia's which don't match */
968 		if (isaddr.s_addr != ia->ia_addr.sin_addr.s_addr)
969 			continue;
970 
971 		if (ifp->if_bridge && ia->ia_ifp &&
972 		    ifp->if_bridge == ia->ia_ifp->if_bridge) {
973 			ifp = ether_bridge_interface(ifp);
974 			goto match;
975 		}
976 		if (ia->ia_ifp && ia->ia_ifp->if_bridge &&
977 		    ether_bridge_interface(ia->ia_ifp) == ifp) {
978 			goto match;
979 		}
980 		if (ifp->if_bridge && ether_bridge_interface(ifp) ==
981 		    ia->ia_ifp) {
982 			goto match;
983 		}
984 
985 		if (ia->ia_ifp == ifp)
986 			goto match;
987 	}
988 
989 	/*
990 	 * No match, use the first inet address on the receive interface
991 	 * as a dummy address for the rest of the function.
992 	 */
993 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
994 		struct ifaddr *ifa = ifac->ifa;
995 
996 		if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
997 			ia = ifatoia(ifa);
998 			goto match;
999 		}
1000 	}
1001 
1002 	/*
1003 	 * If we got here, we didn't find any suitable interface,
1004 	 * so drop the packet.
1005 	 */
1006 	m_freem(m);
1007 	return;
1008 
1009 match:
1010 	if (!enaddr)
1011 		enaddr = (uint8_t *)IF_LLADDR(ifp);
1012 	myaddr = ia->ia_addr.sin_addr;
1013 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) {
1014 		m_freem(m);	/* it's from me, ignore it. */
1015 		return;
1016 	}
1017 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1018 		log(LOG_ERR,
1019 		    "arp: link address is broadcast for IP address %s!\n",
1020 		    kinet_ntoa(isaddr, sbuf));
1021 		m_freem(m);
1022 		return;
1023 	}
1024 	if (isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) {
1025 		hexncpy((u_char *)ar_sha(ah), ifp->if_addrlen,
1026 		    hexstr, HEX_NCPYLEN(ifp->if_addrlen), ":");
1027 		log(LOG_ERR,
1028 		   "arp: %s is using my IP address %s!\n",
1029 		    hexstr, kinet_ntoa(isaddr, sbuf));
1030 		itaddr = myaddr;
1031 		goto reply;
1032 	}
1033 	if (ifp->if_flags & IFF_STATICARP)
1034 		goto reply;
1035 
1036 	/*
1037 	 * When arp_restricted_match is true and the ARP response is not
1038 	 * specifically targetted to me, ignore it.  Otherwise the entry
1039 	 * timeout may be updated for an old MAC.
1040 	 */
1041 	if (arp_restricted_match && itaddr.s_addr != myaddr.s_addr) {
1042 		m_freem(m);
1043 		return;
1044 	}
1045 
1046 	/*
1047 	 * Update all CPU's routing tables with this ARP packet.
1048 	 *
1049 	 * However, we only need to generate rtmsg on CPU0.
1050 	 */
1051 	ASSERT_IN_NETISR(0);
1052 	changed = arp_update_oncpu(m, isaddr.s_addr,
1053 				   itaddr.s_addr == myaddr.s_addr,
1054 				   RTL_REPORTMSG, TRUE);
1055 
1056 	if (ncpus > 1 && changed) {
1057 		struct netmsg_inarp *msg = &m->m_hdr.mh_arpmsg;
1058 
1059 		netmsg_init(&msg->base, NULL, &netisr_apanic_rport,
1060 			    0, arp_update_msghandler);
1061 		msg->m = m;
1062 		msg->saddr = isaddr.s_addr;
1063 		msg->taddr = itaddr.s_addr;
1064 		msg->myaddr = myaddr.s_addr;
1065 		lwkt_sendmsg(netisr_cpuport(1), &msg->base.lmsg);
1066 	} else {
1067 		goto reply;
1068 	}
1069 
1070 	/*
1071 	 * Just return here; after all CPUs's routing tables are
1072 	 * properly updated by this ARP packet, an ARP reply will
1073 	 * be generated if appropriate.
1074 	 */
1075 	return;
1076 reply:
1077 	in_arpreply(m, itaddr.s_addr, myaddr.s_addr);
1078 }
1079 
1080 static void
1081 arp_reply_msghandler(netmsg_t msg)
1082 {
1083 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1084 
1085 	in_arpreply(rmsg->m, rmsg->taddr, rmsg->myaddr);
1086 	/* Don't reply this netmsg; netmsg_inarp is embedded in mbuf */
1087 }
1088 
1089 static void
1090 arp_update_msghandler(netmsg_t msg)
1091 {
1092 	struct netmsg_inarp *rmsg = (struct netmsg_inarp *)msg;
1093 	int nextcpu;
1094 
1095 	/*
1096 	 * This message handler will be called on all of the APs;
1097 	 * no need to generate rtmsg on them.
1098 	 */
1099 	KASSERT(mycpuid > 0, ("arp update msg on cpu%d", mycpuid));
1100 	arp_update_oncpu(rmsg->m, rmsg->saddr,
1101 			 rmsg->taddr == rmsg->myaddr,
1102 			 RTL_DONTREPORT, FALSE);
1103 
1104 	nextcpu = mycpuid + 1;
1105 	if (nextcpu < ncpus) {
1106 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
1107 	} else {
1108 		struct mbuf *m = rmsg->m;
1109 		in_addr_t saddr = rmsg->saddr;
1110 		in_addr_t taddr = rmsg->taddr;
1111 		in_addr_t myaddr = rmsg->myaddr;
1112 
1113 		/*
1114 		 * Dispatch this mbuf to netisr0 to perform ARP reply,
1115 		 * if appropriate.
1116 		 * NOTE: netmsg_inarp is embedded in this mbuf.
1117 		 */
1118 		netmsg_init(&rmsg->base, NULL, &netisr_apanic_rport,
1119 		    0, arp_reply_msghandler);
1120 		rmsg->m = m;
1121 		rmsg->saddr = saddr;
1122 		rmsg->taddr = taddr;
1123 		rmsg->myaddr = myaddr;
1124 		lwkt_sendmsg(netisr_cpuport(0), &rmsg->base.lmsg);
1125 	}
1126 }
1127 
1128 static void
1129 in_arpreply(struct mbuf *m, in_addr_t taddr, in_addr_t myaddr)
1130 {
1131 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1132 	const uint8_t *enaddr;
1133 	struct arphdr *ah;
1134 	struct sockaddr sa;
1135 	struct ether_header *eh;
1136 
1137 	ah = mtod(m, struct arphdr *);
1138 	if (ntohs(ah->ar_op) != ARPOP_REQUEST) {
1139 		m_freem(m);
1140 		return;
1141 	}
1142 
1143 	enaddr = (const uint8_t *)IF_LLADDR(ifp);
1144 	if (taddr == myaddr) {
1145 		/* I am the target */
1146 		memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1147 		memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1148 	} else {
1149 		struct llinfo_arp *la;
1150 		struct rtentry *rt;
1151 
1152 		la = arplookup(taddr, FALSE, RTL_DONTREPORT, SIN_PROXY);
1153 		if (la == NULL) {
1154 			struct sockaddr_in sin;
1155 #ifdef DEBUG_PROXY
1156 			char tbuf[INET_ADDRSTRLEN];
1157 #endif
1158 
1159 			if (!arp_proxyall) {
1160 				m_freem(m);
1161 				return;
1162 			}
1163 
1164 			bzero(&sin, sizeof sin);
1165 			sin.sin_family = AF_INET;
1166 			sin.sin_len = sizeof sin;
1167 			sin.sin_addr.s_addr = taddr;
1168 
1169 			rt = rtpurelookup((struct sockaddr *)&sin);
1170 			if (rt == NULL) {
1171 				m_freem(m);
1172 				return;
1173 			}
1174 			--rt->rt_refcnt;
1175 			/*
1176 			 * Don't send proxies for nodes on the same interface
1177 			 * as this one came out of, or we'll get into a fight
1178 			 * over who claims what Ether address.
1179 			 */
1180 			if (rt->rt_ifp == ifp) {
1181 				m_freem(m);
1182 				return;
1183 			}
1184 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1185 			memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1186 #ifdef DEBUG_PROXY
1187 			kprintf("arp: proxying for %s\n",
1188 			    kinet_ntoa(itaddr, tbuf));
1189 #endif
1190 		} else {
1191 			struct sockaddr_dl *sdl;
1192 
1193 			rt = la->la_rt;
1194 			memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1195 			sdl = SDL(rt->rt_gateway);
1196 			memcpy(ar_sha(ah), LLADDR(sdl), ah->ar_hln);
1197 		}
1198 	}
1199 
1200 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1201 	memcpy(ar_spa(ah), &taddr, ah->ar_pln);
1202 	ah->ar_op = htons(ARPOP_REPLY);
1203 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1204 	switch (ifp->if_type) {
1205 	case IFT_ETHER:
1206 		/*
1207 		 * May not be correct for types not explictly
1208 		 * listed, but it is our best guess.
1209 		 */
1210 	default:
1211 		eh = (struct ether_header *)sa.sa_data;
1212 		memcpy(eh->ether_dhost, ar_tha(ah), sizeof eh->ether_dhost);
1213 		eh->ether_type = htons(ETHERTYPE_ARP);
1214 		break;
1215 	}
1216 	sa.sa_family = AF_UNSPEC;
1217 	sa.sa_len = sizeof sa;
1218 	ifp->if_output(ifp, m, &sa, NULL);
1219 }
1220 
1221 #endif	/* INET */
1222 
1223 /*
1224  * Free an arp entry.  If the arp entry is actively referenced or represents
1225  * a static entry we only clear it back to an unresolved state, otherwise
1226  * we destroy the entry entirely.
1227  *
1228  * Note that static entries are created when route add ... -interface is used
1229  * to create an interface route to a (direct) destination.
1230  */
1231 static void
1232 arptfree(struct llinfo_arp *la)
1233 {
1234 	struct rtentry *rt = la->la_rt;
1235 	struct sockaddr_dl *sdl;
1236 
1237 	if (rt == NULL)
1238 		panic("arptfree");
1239 	sdl = SDL(rt->rt_gateway);
1240 	if (sdl != NULL &&
1241 	    ((rt->rt_refcnt > 0 && sdl->sdl_family == AF_LINK) ||
1242 	     (rt->rt_flags & RTF_STATIC))) {
1243 		sdl->sdl_alen = 0;
1244 		la->la_preempt = la->la_asked = 0;
1245 		rt->rt_flags &= ~RTF_REJECT;
1246 		return;
1247 	}
1248 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1249 }
1250 
1251 /*
1252  * Lookup or enter a new address in arptab.
1253  */
1254 static struct llinfo_arp *
1255 arplookup(in_addr_t addr, boolean_t create, boolean_t generate_report,
1256 	  boolean_t proxy)
1257 {
1258 	struct rtentry *rt;
1259 	struct sockaddr_inarp sin = { sizeof sin, AF_INET };
1260 	const char *why = NULL;
1261 
1262 	sin.sin_addr.s_addr = addr;
1263 	sin.sin_other = proxy ? SIN_PROXY : 0;
1264 	if (create) {
1265 		rt = _rtlookup((struct sockaddr *)&sin,
1266 			       generate_report, RTL_DOCLONE);
1267 	} else {
1268 		rt = rtpurelookup((struct sockaddr *)&sin);
1269 	}
1270 	if (rt == NULL)
1271 		return (NULL);
1272 	rt->rt_refcnt--;
1273 
1274 	if (rt->rt_flags & RTF_GATEWAY)
1275 		why = "host is not on local network";
1276 	else if (!(rt->rt_flags & RTF_LLINFO))
1277 		why = "could not allocate llinfo";
1278 	else if (rt->rt_gateway->sa_family != AF_LINK)
1279 		why = "gateway route is not ours";
1280 
1281 	if (why) {
1282 		if (create && log_arp_creation_failure) {
1283 			char abuf[INET_ADDRSTRLEN];
1284 
1285 			log(LOG_DEBUG, "arplookup %s failed: %s\n",
1286 			    kinet_ntoa(sin.sin_addr, abuf), why);
1287 		}
1288 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_WASCLONED)) {
1289 			/* No references to this route.  Purge it. */
1290 			rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1291 				  rt_mask(rt), rt->rt_flags, NULL);
1292 		}
1293 		return (NULL);
1294 	}
1295 	return (rt->rt_llinfo);
1296 }
1297 
1298 void
1299 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1300 {
1301 	ifa->ifa_rtrequest = arp_rtrequest;
1302 	ifa->ifa_flags |= RTF_CLONING;
1303 }
1304 
1305 void
1306 arp_gratuitous(struct ifnet *ifp, struct ifaddr *ifa)
1307 {
1308 	if (IA_SIN(ifa)->sin_addr.s_addr != INADDR_ANY) {
1309 		arprequest_async(ifp, &IA_SIN(ifa)->sin_addr,
1310 				 &IA_SIN(ifa)->sin_addr, NULL);
1311 	}
1312 }
1313 
1314 static void
1315 arp_ifaddr(void *arg __unused, struct ifnet *ifp,
1316     enum ifaddr_event event, struct ifaddr *ifa)
1317 {
1318 	if (ifa->ifa_rtrequest != arp_rtrequest) /* XXX need a generic way */
1319 		return;
1320 	if (ifa->ifa_addr->sa_family != AF_INET)
1321 		return;
1322 	if (event == IFADDR_EVENT_DELETE)
1323 		return;
1324 
1325 	/*
1326 	 * - CARP interfaces will take care of gratuitous ARP themselves.
1327 	 * - If we are the CARP interface's parent, don't send gratuitous
1328 	 *   ARP to avoid unnecessary confusion.
1329 	 */
1330 #ifdef CARP
1331 	if (ifp->if_type != IFT_CARP && ifp->if_carp == NULL)
1332 #endif
1333 	{
1334 		arp_gratuitous(ifp, ifa);
1335 	}
1336 }
1337 
1338 static void
1339 arp_init(void)
1340 {
1341 	int cpu;
1342 
1343 	for (cpu = 0; cpu < ncpus2; cpu++)
1344 		LIST_INIT(&llinfo_arp_list[cpu]);
1345 
1346 	netisr_register(NETISR_ARP, arpintr, NULL);
1347 
1348 	EVENTHANDLER_REGISTER(ifaddr_event, arp_ifaddr, NULL,
1349 	    EVENTHANDLER_PRI_LAST);
1350 }
1351 
1352 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1353