xref: /openbsd-src/sys/netinet/if_ether.c (revision ae3cb403620ab940fbaabb3055fac045a63d56b7)
1 /*	$OpenBSD: if_ether.c,v 1.233 2018/01/16 10:33:55 mpi Exp $	*/
2 /*	$NetBSD: if_ether.c,v 1.31 1996/05/11 12:59:58 mycroft Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
33  */
34 
35 /*
36  * Ethernet address resolution protocol.
37  * TODO:
38  *	add "inuse/lock" bit (or ref. count) along with valid bit
39  */
40 
41 #include "carp.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/mbuf.h>
46 #include <sys/socket.h>
47 #include <sys/timeout.h>
48 #include <sys/kernel.h>
49 #include <sys/syslog.h>
50 #include <sys/queue.h>
51 #include <sys/pool.h>
52 
53 #include <net/if.h>
54 #include <net/if_var.h>
55 #include <net/if_dl.h>
56 #include <net/route.h>
57 #include <net/if_types.h>
58 #include <net/netisr.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_var.h>
62 #include <netinet/if_ether.h>
63 #if NCARP > 0
64 #include <netinet/ip_carp.h>
65 #endif
66 
67 struct llinfo_arp {
68 	LIST_ENTRY(llinfo_arp)	 la_list;
69 	struct rtentry		*la_rt;		/* backpointer to rtentry */
70 	long			 la_asked;	/* last time we QUERIED */
71 	struct mbuf_list	 la_ml;		/* packet hold queue */
72 };
73 #define LA_HOLD_QUEUE 10
74 #define LA_HOLD_TOTAL 100
75 
76 /* timer values */
77 int	arpt_prune = (5 * 60);	/* walk list every 5 minutes */
78 int	arpt_keep = (20 * 60);	/* once resolved, cache for 20 minutes */
79 int	arpt_down = 20;		/* once declared down, don't send for 20 secs */
80 
81 void arpinvalidate(struct rtentry *);
82 void arptfree(struct rtentry *);
83 void arptimer(void *);
84 struct rtentry *arplookup(struct in_addr *, int, int, unsigned int);
85 void in_arpinput(struct ifnet *, struct mbuf *);
86 void in_revarpinput(struct ifnet *, struct mbuf *);
87 int arpcache(struct ifnet *, struct ether_arp *, struct rtentry *);
88 void arpreply(struct ifnet *, struct mbuf *, struct in_addr *, uint8_t *);
89 
90 struct niqueue arpinq = NIQUEUE_INITIALIZER(50, NETISR_ARP);
91 
92 LIST_HEAD(, llinfo_arp) arp_list;
93 struct	pool arp_pool;		/* pool for llinfo_arp structures */
94 int	arp_maxtries = 5;
95 int	arpinit_done;
96 int	la_hold_total;
97 
98 #ifdef NFSCLIENT
99 /* revarp state */
100 struct in_addr revarp_myip, revarp_srvip;
101 int revarp_finished;
102 unsigned int revarp_ifidx;
103 #endif /* NFSCLIENT */
104 
105 /*
106  * Timeout routine.  Age arp_tab entries periodically.
107  */
108 /* ARGSUSED */
109 void
110 arptimer(void *arg)
111 {
112 	struct timeout *to = (struct timeout *)arg;
113 	struct llinfo_arp *la, *nla;
114 
115 	NET_LOCK();
116 	timeout_add_sec(to, arpt_prune);
117 	LIST_FOREACH_SAFE(la, &arp_list, la_list, nla) {
118 		struct rtentry *rt = la->la_rt;
119 
120 		if (rt->rt_expire && rt->rt_expire <= time_uptime)
121 			arptfree(rt); /* timer has expired; clear */
122 	}
123 	NET_UNLOCK();
124 }
125 
126 void
127 arp_rtrequest(struct ifnet *ifp, int req, struct rtentry *rt)
128 {
129 	struct sockaddr *gate = rt->rt_gateway;
130 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
131 	struct ifaddr *ifa;
132 
133 	if (!arpinit_done) {
134 		static struct timeout arptimer_to;
135 
136 		arpinit_done = 1;
137 		pool_init(&arp_pool, sizeof(struct llinfo_arp), 0,
138 		    IPL_SOFTNET, 0, "arp", NULL);
139 
140 		timeout_set_proc(&arptimer_to, arptimer, &arptimer_to);
141 		timeout_add_sec(&arptimer_to, 1);
142 	}
143 
144 	if (ISSET(rt->rt_flags, RTF_GATEWAY|RTF_BROADCAST|RTF_MULTICAST))
145 		return;
146 
147 	switch (req) {
148 
149 	case RTM_ADD:
150 		if (rt->rt_flags & RTF_CLONING ||
151 		    ((rt->rt_flags & (RTF_LLINFO | RTF_LOCAL)) && !la)) {
152 			/*
153 			 * Give this route an expiration time, even though
154 			 * it's a "permanent" route, so that routes cloned
155 			 * from it do not need their expiration time set.
156 			 */
157 			rt->rt_expire = time_uptime;
158 			if ((rt->rt_flags & RTF_CLONING) != 0)
159 				break;
160 		}
161 		/*
162 		 * Announce a new entry if requested or warn the user
163 		 * if another station has this IP address.
164 		 */
165 		if (rt->rt_flags & (RTF_ANNOUNCE|RTF_LOCAL))
166 			arprequest(ifp,
167 			    &satosin(rt_key(rt))->sin_addr.s_addr,
168 			    &satosin(rt_key(rt))->sin_addr.s_addr,
169 			    (u_char *)LLADDR(satosdl(gate)));
170 		/*FALLTHROUGH*/
171 	case RTM_RESOLVE:
172 		if (gate->sa_family != AF_LINK ||
173 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
174 			log(LOG_DEBUG, "%s: bad gateway value: %s\n", __func__,
175 			    ifp->if_xname);
176 			break;
177 		}
178 		satosdl(gate)->sdl_type = ifp->if_type;
179 		satosdl(gate)->sdl_index = ifp->if_index;
180 		if (la != 0)
181 			break; /* This happens on a route change */
182 		/*
183 		 * Case 2:  This route may come from cloning, or a manual route
184 		 * add with a LL address.
185 		 */
186 		la = pool_get(&arp_pool, PR_NOWAIT | PR_ZERO);
187 		rt->rt_llinfo = (caddr_t)la;
188 		if (la == NULL) {
189 			log(LOG_DEBUG, "%s: pool get failed\n", __func__);
190 			break;
191 		}
192 
193 		ml_init(&la->la_ml);
194 		la->la_rt = rt;
195 		rt->rt_flags |= RTF_LLINFO;
196 		LIST_INSERT_HEAD(&arp_list, la, la_list);
197 
198 		TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
199 			if ((ifa->ifa_addr->sa_family == AF_INET) &&
200 			    ifatoia(ifa)->ia_addr.sin_addr.s_addr ==
201 			    satosin(rt_key(rt))->sin_addr.s_addr)
202 				break;
203 		}
204 		if (ifa) {
205 			KASSERT(ifa == rt->rt_ifa);
206 			rt->rt_expire = 0;
207 		}
208 		break;
209 
210 	case RTM_DELETE:
211 		if (la == NULL)
212 			break;
213 		LIST_REMOVE(la, la_list);
214 		rt->rt_llinfo = NULL;
215 		rt->rt_flags &= ~RTF_LLINFO;
216 		la_hold_total -= ml_purge(&la->la_ml);
217 		pool_put(&arp_pool, la);
218 		break;
219 
220 	case RTM_INVALIDATE:
221 		if (!ISSET(rt->rt_flags, RTF_LOCAL))
222 			arpinvalidate(rt);
223 		break;
224 	}
225 }
226 
227 /*
228  * Broadcast an ARP request. Caller specifies:
229  *	- arp header source ip address
230  *	- arp header target ip address
231  *	- arp header source ethernet address
232  */
233 void
234 arprequest(struct ifnet *ifp, u_int32_t *sip, u_int32_t *tip, u_int8_t *enaddr)
235 {
236 	struct mbuf *m;
237 	struct ether_header *eh;
238 	struct ether_arp *ea;
239 	struct sockaddr sa;
240 
241 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
242 		return;
243 	m->m_len = sizeof(*ea);
244 	m->m_pkthdr.len = sizeof(*ea);
245 	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
246 	m->m_pkthdr.pf.prio = ifp->if_llprio;
247 	MH_ALIGN(m, sizeof(*ea));
248 	ea = mtod(m, struct ether_arp *);
249 	eh = (struct ether_header *)sa.sa_data;
250 	memset(ea, 0, sizeof(*ea));
251 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
252 	eh->ether_type = htons(ETHERTYPE_ARP);	/* if_output will not swap */
253 	ea->arp_hrd = htons(ARPHRD_ETHER);
254 	ea->arp_pro = htons(ETHERTYPE_IP);
255 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
256 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
257 	ea->arp_op = htons(ARPOP_REQUEST);
258 	memcpy(eh->ether_shost, enaddr, sizeof(eh->ether_shost));
259 	memcpy(ea->arp_sha, enaddr, sizeof(ea->arp_sha));
260 	memcpy(ea->arp_spa, sip, sizeof(ea->arp_spa));
261 	memcpy(ea->arp_tpa, tip, sizeof(ea->arp_tpa));
262 	sa.sa_family = pseudo_AF_HDRCMPLT;
263 	sa.sa_len = sizeof(sa);
264 	m->m_flags |= M_BCAST;
265 	ifp->if_output(ifp, m, &sa, NULL);
266 }
267 
268 void
269 arpreply(struct ifnet *ifp, struct mbuf *m, struct in_addr *sip, uint8_t *eaddr)
270 {
271 	struct ether_header *eh;
272 	struct ether_arp *ea;
273 	struct sockaddr sa;
274 
275 	ea = mtod(m, struct ether_arp *);
276 	ea->arp_op = htons(ARPOP_REPLY);
277 	ea->arp_pro = htons(ETHERTYPE_IP); /* let's be sure! */
278 
279 	/* We're replying to a request. */
280 	memcpy(ea->arp_tha, ea->arp_sha, sizeof(ea->arp_sha));
281 	memcpy(ea->arp_tpa, ea->arp_spa, sizeof(ea->arp_spa));
282 
283 	memcpy(ea->arp_sha, eaddr, sizeof(ea->arp_sha));
284 	memcpy(ea->arp_spa, sip, sizeof(ea->arp_spa));
285 
286 	eh = (struct ether_header *)sa.sa_data;
287 	memcpy(eh->ether_dhost, ea->arp_tha, sizeof(eh->ether_dhost));
288 	memcpy(eh->ether_shost, eaddr, sizeof(eh->ether_shost));
289 	eh->ether_type = htons(ETHERTYPE_ARP);
290 	sa.sa_family = pseudo_AF_HDRCMPLT;
291 	sa.sa_len = sizeof(sa);
292 	ifp->if_output(ifp, m, &sa, NULL);
293 }
294 
295 /*
296  * Resolve an IP address into an ethernet address.  If success,
297  * desten is filled in.  If there is no entry in arptab,
298  * set one up and broadcast a request for the IP address.
299  * Hold onto this mbuf and resend it once the address
300  * is finally resolved.  A return value of 0 indicates
301  * that desten has been filled in and the packet should be sent
302  * normally; A return value of EAGAIN indicates that the packet
303  * has been taken over here, either now or for later transmission.
304  * Any other return value indicates an error.
305  */
306 int
307 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
308     struct sockaddr *dst, u_char *desten)
309 {
310 	struct arpcom *ac = (struct arpcom *)ifp;
311 	struct llinfo_arp *la = NULL;
312 	struct sockaddr_dl *sdl;
313 	struct rtentry *rt = NULL;
314 	char addr[INET_ADDRSTRLEN];
315 
316 	if (m->m_flags & M_BCAST) {	/* broadcast */
317 		memcpy(desten, etherbroadcastaddr, sizeof(etherbroadcastaddr));
318 		return (0);
319 	}
320 	if (m->m_flags & M_MCAST) {	/* multicast */
321 		ETHER_MAP_IP_MULTICAST(&satosin(dst)->sin_addr, desten);
322 		return (0);
323 	}
324 
325 	rt = rt_getll(rt0);
326 
327 	if (ISSET(rt->rt_flags, RTF_REJECT) &&
328 	    (rt->rt_expire == 0 || time_uptime < rt->rt_expire)) {
329 		m_freem(m);
330 		return (rt == rt0 ? EHOSTDOWN : EHOSTUNREACH);
331 	}
332 
333 	if (!ISSET(rt->rt_flags, RTF_LLINFO)) {
334 		log(LOG_DEBUG, "%s: %s: route contains no arp information\n",
335 		    __func__, inet_ntop(AF_INET, &satosin(rt_key(rt))->sin_addr,
336 		    addr, sizeof(addr)));
337 		m_freem(m);
338 		return (EINVAL);
339 	}
340 
341 	sdl = satosdl(rt->rt_gateway);
342 	if (sdl->sdl_alen > 0 && sdl->sdl_alen != ETHER_ADDR_LEN) {
343 		log(LOG_DEBUG, "%s: %s: incorrect arp information\n", __func__,
344 		    inet_ntop(AF_INET, &satosin(dst)->sin_addr,
345 			addr, sizeof(addr)));
346 		goto bad;
347 	}
348 
349 	/*
350 	 * Check the address family and length is valid, the address
351 	 * is resolved; otherwise, try to resolve.
352 	 */
353 	if ((rt->rt_expire == 0 || rt->rt_expire > time_uptime) &&
354 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
355 		memcpy(desten, LLADDR(sdl), sdl->sdl_alen);
356 		return (0);
357 	}
358 
359 	if (ifp->if_flags & (IFF_NOARP|IFF_STATICARP))
360 		goto bad;
361 
362 	/*
363 	 * There is an arptab entry, but no ethernet address
364 	 * response yet. Insert mbuf in hold queue if below limit
365 	 * if above the limit free the queue without queuing the new packet.
366 	 */
367 	la = (struct llinfo_arp *)rt->rt_llinfo;
368 	KASSERT(la != NULL);
369 	if (la_hold_total < LA_HOLD_TOTAL && la_hold_total < nmbclust / 64) {
370 		struct mbuf *mh;
371 
372 		if (ml_len(&la->la_ml) >= LA_HOLD_QUEUE) {
373 			mh = ml_dequeue(&la->la_ml);
374 			la_hold_total--;
375 			m_freem(mh);
376 		}
377 		ml_enqueue(&la->la_ml, m);
378 		la_hold_total++;
379 	} else {
380 		la_hold_total -= ml_purge(&la->la_ml);
381 		m_freem(m);
382 	}
383 
384 	/*
385 	 * Re-send the ARP request when appropriate.
386 	 */
387 #ifdef	DIAGNOSTIC
388 	if (rt->rt_expire == 0) {
389 		/* This should never happen. (Should it? -gwr) */
390 		printf("%s: unresolved and rt_expire == 0\n", __func__);
391 		/* Set expiration time to now (expired). */
392 		rt->rt_expire = time_uptime;
393 	}
394 #endif
395 	if (rt->rt_expire) {
396 		rt->rt_flags &= ~RTF_REJECT;
397 		if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
398 			rt->rt_expire = time_uptime;
399 			if (la->la_asked++ < arp_maxtries)
400 				arprequest(ifp,
401 				    &satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr,
402 				    &satosin(dst)->sin_addr.s_addr,
403 				    ac->ac_enaddr);
404 			else {
405 				rt->rt_flags |= RTF_REJECT;
406 				rt->rt_expire += arpt_down;
407 				la->la_asked = 0;
408 				la_hold_total -= ml_purge(&la->la_ml);
409 			}
410 		}
411 	}
412 
413 	return (EAGAIN);
414 
415 bad:
416 	m_freem(m);
417 	return (EINVAL);
418 }
419 
420 /*
421  * Common length and type checks are done here,
422  * then the protocol-specific routine is called.
423  */
424 void
425 arpinput(struct ifnet *ifp, struct mbuf *m)
426 {
427 	struct arphdr *ar;
428 	int len;
429 
430 #ifdef DIAGNOSTIC
431 	if ((m->m_flags & M_PKTHDR) == 0)
432 		panic("arpintr");
433 #endif
434 
435 	len = sizeof(struct arphdr);
436 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
437 		return;
438 
439 	ar = mtod(m, struct arphdr *);
440 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER ||
441 	    ntohs(ar->ar_pro) != ETHERTYPE_IP) {
442 		m_freem(m);
443 		return;
444 	}
445 
446 	len += 2 * (ar->ar_hln + ar->ar_pln);
447 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
448 		return;
449 
450 	niq_enqueue(&arpinq, m);
451 }
452 
453 void
454 arpintr(void)
455 {
456 	struct mbuf_list ml;
457 	struct mbuf *m;
458 	struct ifnet *ifp;
459 
460 	niq_delist(&arpinq, &ml);
461 
462 	while ((m = ml_dequeue(&ml)) != NULL) {
463 		ifp = if_get(m->m_pkthdr.ph_ifidx);
464 
465 		if (ifp != NULL)
466 			in_arpinput(ifp, m);
467 		else
468 			m_freem(m);
469 
470 		if_put(ifp);
471 	}
472 }
473 
474 /*
475  * ARP for Internet protocols on Ethernet, RFC 826.
476  * In addition, a sanity check is performed on the sender
477  * protocol address, to catch impersonators.
478  */
479 void
480 in_arpinput(struct ifnet *ifp, struct mbuf *m)
481 {
482 	struct ether_arp *ea;
483 	struct rtentry *rt = NULL;
484 	struct sockaddr_in sin;
485 	struct in_addr isaddr, itaddr;
486 	char addr[INET_ADDRSTRLEN];
487 	int op, target = 0;
488 	unsigned int rdomain;
489 
490 	rdomain = rtable_l2(m->m_pkthdr.ph_rtableid);
491 
492 	ea = mtod(m, struct ether_arp *);
493 	op = ntohs(ea->arp_op);
494 	if ((op != ARPOP_REQUEST) && (op != ARPOP_REPLY))
495 		goto out;
496 
497 	memcpy(&itaddr, ea->arp_tpa, sizeof(itaddr));
498 	memcpy(&isaddr, ea->arp_spa, sizeof(isaddr));
499 	memset(&sin, 0, sizeof(sin));
500 	sin.sin_len = sizeof(sin);
501 	sin.sin_family = AF_INET;
502 
503 	if (ETHER_IS_MULTICAST(&ea->arp_sha[0]) &&
504 	    !memcmp(ea->arp_sha, etherbroadcastaddr, sizeof(ea->arp_sha))) {
505 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
506 		log(LOG_ERR, "arp: ether address is broadcast for IP address "
507 		    "%s!\n", addr);
508 		goto out;
509 	}
510 
511 	if (!memcmp(ea->arp_sha, LLADDR(ifp->if_sadl), sizeof(ea->arp_sha)))
512 		goto out;	/* it's from me, ignore it. */
513 
514 	/* Check target against our interface addresses. */
515 	sin.sin_addr = itaddr;
516 	rt = rtalloc(sintosa(&sin), 0, rdomain);
517 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
518 	    rt->rt_ifidx == ifp->if_index)
519 		target = 1;
520 	rtfree(rt);
521 	rt = NULL;
522 
523 #if NCARP > 0
524 	if (target && op == ARPOP_REQUEST && ifp->if_type == IFT_CARP &&
525 	    !carp_iamatch(ifp))
526 		goto out;
527 #endif
528 
529 	/* Do we have an ARP cache for the sender? Create if we are target. */
530 	rt = arplookup(&isaddr, target, 0, rdomain);
531 
532 	/* Check sender against our interface addresses. */
533 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
534 	    rt->rt_ifidx == ifp->if_index && isaddr.s_addr != INADDR_ANY) {
535 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
536 		log(LOG_ERR, "duplicate IP address %s sent from ethernet "
537 		    "address %s\n", addr, ether_sprintf(ea->arp_sha));
538 		itaddr = isaddr;
539 	} else if (rt != NULL) {
540 		int error;
541 
542 		KERNEL_LOCK();
543 		error = arpcache(ifp, ea, rt);
544 		KERNEL_UNLOCK();
545 		if (error)
546 			goto out;
547 	}
548 
549 	if (op == ARPOP_REQUEST) {
550 		uint8_t *eaddr;
551 
552 		if (target) {
553 			/* We already have all info for the reply */
554 			eaddr = LLADDR(ifp->if_sadl);
555 		} else {
556 			rtfree(rt);
557 			rt = arplookup(&itaddr, 0, SIN_PROXY, rdomain);
558 			/*
559 			 * Protect from possible duplicates, only owner
560 			 * should respond
561 			 */
562 			if ((rt == NULL) || (rt->rt_ifidx != ifp->if_index))
563 				goto out;
564 			eaddr = LLADDR(satosdl(rt->rt_gateway));
565 		}
566 		arpreply(ifp, m, &itaddr, eaddr);
567 		rtfree(rt);
568 		return;
569 	}
570 
571 out:
572 	rtfree(rt);
573 	m_freem(m);
574 }
575 
576 int
577 arpcache(struct ifnet *ifp, struct ether_arp *ea, struct rtentry *rt)
578 {
579 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
580 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
581 	struct in_addr *spa = (struct in_addr *)ea->arp_spa;
582 	char addr[INET_ADDRSTRLEN];
583 	struct ifnet *rifp;
584 	unsigned int len;
585 	int changed = 0;
586 
587 	KERNEL_ASSERT_LOCKED();
588 	KASSERT(sdl != NULL);
589 
590 	/*
591 	 * This can happen if the entry has been deleted by another CPU
592 	 * after we found it.
593 	 */
594 	if (la == NULL)
595 		return (0);
596 
597 	if (sdl->sdl_alen > 0) {
598 		if (memcmp(ea->arp_sha, LLADDR(sdl), sdl->sdl_alen)) {
599 			if (ISSET(rt->rt_flags, RTF_PERMANENT_ARP|RTF_LOCAL)) {
600 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
601 				log(LOG_WARNING, "arp: attempt to overwrite "
602 				   "permanent entry for %s by %s on %s\n", addr,
603 				   ether_sprintf(ea->arp_sha), ifp->if_xname);
604 				return (-1);
605 			} else if (rt->rt_ifidx != ifp->if_index) {
606 #if NCARP > 0
607 				if (ifp->if_type != IFT_CARP)
608 #endif
609 				{
610 					rifp = if_get(rt->rt_ifidx);
611 					if (rifp == NULL)
612 						return (-1);
613 					inet_ntop(AF_INET, spa, addr,
614 					    sizeof(addr));
615 					log(LOG_WARNING, "arp: attempt to "
616 					    "overwrite entry for %s on %s by "
617 					    "%s on %s\n", addr, rifp->if_xname,
618 					    ether_sprintf(ea->arp_sha),
619 					    ifp->if_xname);
620 					if_put(rifp);
621 				}
622 				return (-1);
623 			} else {
624 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
625 				log(LOG_INFO, "arp info overwritten for %s by "
626 				    "%s on %s\n", addr,
627 				    ether_sprintf(ea->arp_sha), ifp->if_xname);
628 				rt->rt_expire = 1;/* no longer static */
629 			}
630 			changed = 1;
631 		}
632 	} else if (!if_isconnected(ifp, rt->rt_ifidx)) {
633 		rifp = if_get(rt->rt_ifidx);
634 		if (rifp == NULL)
635 			return (-1);
636 		inet_ntop(AF_INET, spa, addr, sizeof(addr));
637 		log(LOG_WARNING, "arp: attempt to add entry for %s on %s by %s"
638 		    " on %s\n", addr, rifp->if_xname,
639 		    ether_sprintf(ea->arp_sha), ifp->if_xname);
640 		if_put(rifp);
641 		return (-1);
642 	}
643 	sdl->sdl_alen = sizeof(ea->arp_sha);
644 	memcpy(LLADDR(sdl), ea->arp_sha, sizeof(ea->arp_sha));
645 	if (rt->rt_expire)
646 		rt->rt_expire = time_uptime + arpt_keep;
647 	rt->rt_flags &= ~RTF_REJECT;
648 
649 	/* Notify userland that an ARP resolution has been done. */
650 	if (la->la_asked || changed) {
651 		KERNEL_LOCK();
652 		rtm_send(rt, RTM_RESOLVE, 0, ifp->if_rdomain);
653 		KERNEL_UNLOCK();
654 	}
655 
656 	la->la_asked = 0;
657 	while ((len = ml_len(&la->la_ml)) != 0) {
658 		struct mbuf *mh;
659 
660 		mh = ml_dequeue(&la->la_ml);
661 		la_hold_total--;
662 
663 		ifp->if_output(ifp, mh, rt_key(rt), rt);
664 
665 		if (ml_len(&la->la_ml) == len) {
666 			/* mbuf is back in queue. Discard. */
667 			while ((mh = ml_dequeue(&la->la_ml)) != NULL) {
668 				la_hold_total--;
669 				m_freem(mh);
670 			}
671 			break;
672 		}
673 	}
674 
675 	return (0);
676 }
677 
678 void
679 arpinvalidate(struct rtentry *rt)
680 {
681 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
682 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
683 
684 	la_hold_total -= ml_purge(&la->la_ml);
685 	sdl->sdl_alen = 0;
686 	la->la_asked = 0;
687 }
688 
689 /*
690  * Free an arp entry.
691  */
692 void
693 arptfree(struct rtentry *rt)
694 {
695 	struct ifnet *ifp;
696 
697 	KASSERT(!ISSET(rt->rt_flags, RTF_LOCAL));
698 	arpinvalidate(rt);
699 
700 	ifp = if_get(rt->rt_ifidx);
701 	KASSERT(ifp != NULL);
702 	if (!ISSET(rt->rt_flags, RTF_STATIC|RTF_CACHED))
703 		rtdeletemsg(rt, ifp, ifp->if_rdomain);
704 	if_put(ifp);
705 }
706 
707 /*
708  * Lookup or enter a new address in arptab.
709  */
710 struct rtentry *
711 arplookup(struct in_addr *inp, int create, int proxy, u_int tableid)
712 {
713 	struct rtentry *rt;
714 	struct sockaddr_inarp sin;
715 	int flags;
716 
717 	memset(&sin, 0, sizeof(sin));
718 	sin.sin_len = sizeof(sin);
719 	sin.sin_family = AF_INET;
720 	sin.sin_addr.s_addr = inp->s_addr;
721 	sin.sin_other = proxy ? SIN_PROXY : 0;
722 	flags = (create) ? RT_RESOLVE : 0;
723 
724 	rt = rtalloc((struct sockaddr *)&sin, flags, tableid);
725 	if (!rtisvalid(rt) || ISSET(rt->rt_flags, RTF_GATEWAY) ||
726 	    !ISSET(rt->rt_flags, RTF_LLINFO) ||
727 	    rt->rt_gateway->sa_family != AF_LINK) {
728 		rtfree(rt);
729 		return (NULL);
730 	}
731 
732 	if (proxy && !ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
733 		while ((rt = rtable_iterate(rt)) != NULL) {
734 			if (ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
735 				break;
736 			}
737 		}
738 	}
739 
740 	return (rt);
741 }
742 
743 /*
744  * Check whether we do proxy ARP for this address and we point to ourselves.
745  */
746 int
747 arpproxy(struct in_addr in, unsigned int rtableid)
748 {
749 	struct sockaddr_dl *sdl;
750 	struct rtentry *rt;
751 	struct ifnet *ifp;
752 	int found = 0;
753 
754 	rt = arplookup(&in, 0, SIN_PROXY, rtableid);
755 	if (!rtisvalid(rt)) {
756 		rtfree(rt);
757 		return (0);
758 	}
759 
760 	/* Check that arp information are correct. */
761 	sdl = satosdl(rt->rt_gateway);
762 	if (sdl->sdl_alen != ETHER_ADDR_LEN) {
763 		rtfree(rt);
764 		return (0);
765 	}
766 
767 	ifp = if_get(rt->rt_ifidx);
768 	if (ifp == NULL) {
769 		rtfree(rt);
770 		return (0);
771 	}
772 
773 	if (!memcmp(LLADDR(sdl), LLADDR(ifp->if_sadl), sdl->sdl_alen))
774 		found = 1;
775 
776 	if_put(ifp);
777 	rtfree(rt);
778 	return (found);
779 }
780 
781 /*
782  * Called from Ethernet interrupt handlers
783  * when ether packet type ETHERTYPE_REVARP
784  * is received.  Common length and type checks are done here,
785  * then the protocol-specific routine is called.
786  */
787 void
788 revarpinput(struct ifnet *ifp, struct mbuf *m)
789 {
790 	struct arphdr *ar;
791 
792 	if (m->m_len < sizeof(struct arphdr))
793 		goto out;
794 	ar = mtod(m, struct arphdr *);
795 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER)
796 		goto out;
797 	if (m->m_len < sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln))
798 		goto out;
799 	switch (ntohs(ar->ar_pro)) {
800 
801 	case ETHERTYPE_IP:
802 		in_revarpinput(ifp, m);
803 		return;
804 
805 	default:
806 		break;
807 	}
808 out:
809 	m_freem(m);
810 }
811 
812 /*
813  * RARP for Internet protocols on Ethernet.
814  * Algorithm is that given in RFC 903.
815  * We are only using for bootstrap purposes to get an ip address for one of
816  * our interfaces.  Thus we support no user-interface.
817  *
818  * Since the contents of the RARP reply are specific to the interface that
819  * sent the request, this code must ensure that they are properly associated.
820  *
821  * Note: also supports ARP via RARP packets, per the RFC.
822  */
823 void
824 in_revarpinput(struct ifnet *ifp, struct mbuf *m)
825 {
826 	struct ether_arp *ar;
827 	int op;
828 
829 	ar = mtod(m, struct ether_arp *);
830 	op = ntohs(ar->arp_op);
831 	switch (op) {
832 	case ARPOP_REQUEST:
833 	case ARPOP_REPLY:	/* per RFC */
834 		niq_enqueue(&arpinq, m);
835 		return;
836 	case ARPOP_REVREPLY:
837 		break;
838 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
839 	default:
840 		goto out;
841 	}
842 #ifdef NFSCLIENT
843 	if (revarp_ifidx == 0)
844 		goto out;
845 	if (revarp_ifidx != m->m_pkthdr.ph_ifidx) /* !same interface */
846 		goto out;
847 	if (revarp_finished)
848 		goto wake;
849 	if (memcmp(ar->arp_tha, LLADDR(ifp->if_sadl), sizeof(ar->arp_tha)))
850 		goto out;
851 	memcpy(&revarp_srvip, ar->arp_spa, sizeof(revarp_srvip));
852 	memcpy(&revarp_myip, ar->arp_tpa, sizeof(revarp_myip));
853 	revarp_finished = 1;
854 wake:	/* Do wakeup every time in case it was missed. */
855 	wakeup((caddr_t)&revarp_myip);
856 #endif /* NFSCLIENT */
857 
858 out:
859 	m_freem(m);
860 }
861 
862 /*
863  * Send a RARP request for the ip address of the specified interface.
864  * The request should be RFC 903-compliant.
865  */
866 void
867 revarprequest(struct ifnet *ifp)
868 {
869 	struct sockaddr sa;
870 	struct mbuf *m;
871 	struct ether_header *eh;
872 	struct ether_arp *ea;
873 	struct arpcom *ac = (struct arpcom *)ifp;
874 
875 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
876 		return;
877 	m->m_len = sizeof(*ea);
878 	m->m_pkthdr.len = sizeof(*ea);
879 	m->m_pkthdr.pf.prio = ifp->if_llprio;
880 	MH_ALIGN(m, sizeof(*ea));
881 	ea = mtod(m, struct ether_arp *);
882 	eh = (struct ether_header *)sa.sa_data;
883 	memset(ea, 0, sizeof(*ea));
884 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
885 	eh->ether_type = htons(ETHERTYPE_REVARP);
886 	ea->arp_hrd = htons(ARPHRD_ETHER);
887 	ea->arp_pro = htons(ETHERTYPE_IP);
888 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
889 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
890 	ea->arp_op = htons(ARPOP_REVREQUEST);
891 	memcpy(eh->ether_shost, ac->ac_enaddr, sizeof(ea->arp_tha));
892 	memcpy(ea->arp_sha, ac->ac_enaddr, sizeof(ea->arp_sha));
893 	memcpy(ea->arp_tha, ac->ac_enaddr, sizeof(ea->arp_tha));
894 	sa.sa_family = pseudo_AF_HDRCMPLT;
895 	sa.sa_len = sizeof(sa);
896 	m->m_flags |= M_BCAST;
897 	ifp->if_output(ifp, m, &sa, NULL);
898 }
899 
900 #ifdef NFSCLIENT
901 /*
902  * RARP for the ip address of the specified interface, but also
903  * save the ip address of the server that sent the answer.
904  * Timeout if no response is received.
905  */
906 int
907 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
908     struct in_addr *clnt_in)
909 {
910 	int result, count = 20;
911 
912 	if (revarp_finished)
913 		return EIO;
914 
915 	revarp_ifidx = ifp->if_index;
916 	while (count--) {
917 		revarprequest(ifp);
918 		result = tsleep((caddr_t)&revarp_myip, PSOCK, "revarp", hz/2);
919 		if (result != EWOULDBLOCK)
920 			break;
921 	}
922 	revarp_ifidx = 0;
923 	if (!revarp_finished)
924 		return ENETUNREACH;
925 
926 	memcpy(serv_in, &revarp_srvip, sizeof(*serv_in));
927 	memcpy(clnt_in, &revarp_myip, sizeof(*clnt_in));
928 	return 0;
929 }
930 
931 /* For compatibility: only saves interface address. */
932 int
933 revarpwhoami(struct in_addr *in, struct ifnet *ifp)
934 {
935 	struct in_addr server;
936 	return (revarpwhoarewe(ifp, &server, in));
937 }
938 #endif /* NFSCLIENT */
939