xref: /openbsd-src/sys/netinet/if_ether.c (revision 0b7734b3d77bb9b21afec6f4621cae6c805dbd45)
1 /*	$OpenBSD: if_ether.c,v 1.220 2016/07/14 14:01:40 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 arptfree(struct rtentry *);
82 void arptimer(void *);
83 struct rtentry *arplookup(struct in_addr *, int, int, unsigned int);
84 void in_arpinput(struct ifnet *, struct mbuf *);
85 void in_revarpinput(struct ifnet *, struct mbuf *);
86 int arpcache(struct ifnet *, struct ether_arp *, struct rtentry *);
87 void arpreply(struct ifnet *, struct mbuf *, struct in_addr *, uint8_t *);
88 
89 struct niqueue arpinq = NIQUEUE_INITIALIZER(50, NETISR_ARP);
90 
91 LIST_HEAD(, llinfo_arp) arp_list;
92 struct	pool arp_pool;		/* pool for llinfo_arp structures */
93 int	arp_maxtries = 5;
94 int	arpinit_done;
95 int	la_hold_total;
96 
97 #ifdef NFSCLIENT
98 /* revarp state */
99 struct in_addr revarp_myip, revarp_srvip;
100 int revarp_finished;
101 unsigned int revarp_ifidx;
102 #endif /* NFSCLIENT */
103 
104 /*
105  * Timeout routine.  Age arp_tab entries periodically.
106  */
107 /* ARGSUSED */
108 void
109 arptimer(void *arg)
110 {
111 	struct timeout *to = (struct timeout *)arg;
112 	int s;
113 	struct llinfo_arp *la, *nla;
114 
115 	s = splsoftnet();
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 	splx(s);
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, 0, 0, "arp",
138 		    NULL);
139 
140 		timeout_set(&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 	}
219 }
220 
221 /*
222  * Broadcast an ARP request. Caller specifies:
223  *	- arp header source ip address
224  *	- arp header target ip address
225  *	- arp header source ethernet address
226  */
227 void
228 arprequest(struct ifnet *ifp, u_int32_t *sip, u_int32_t *tip, u_int8_t *enaddr)
229 {
230 	struct mbuf *m;
231 	struct ether_header *eh;
232 	struct ether_arp *ea;
233 	struct sockaddr sa;
234 
235 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
236 		return;
237 	m->m_len = sizeof(*ea);
238 	m->m_pkthdr.len = sizeof(*ea);
239 	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
240 	m->m_pkthdr.pf.prio = ifp->if_llprio;
241 	MH_ALIGN(m, sizeof(*ea));
242 	ea = mtod(m, struct ether_arp *);
243 	eh = (struct ether_header *)sa.sa_data;
244 	memset(ea, 0, sizeof(*ea));
245 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
246 	eh->ether_type = htons(ETHERTYPE_ARP);	/* if_output will not swap */
247 	ea->arp_hrd = htons(ARPHRD_ETHER);
248 	ea->arp_pro = htons(ETHERTYPE_IP);
249 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
250 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
251 	ea->arp_op = htons(ARPOP_REQUEST);
252 	memcpy(eh->ether_shost, enaddr, sizeof(eh->ether_shost));
253 	memcpy(ea->arp_sha, enaddr, sizeof(ea->arp_sha));
254 	memcpy(ea->arp_spa, sip, sizeof(ea->arp_spa));
255 	memcpy(ea->arp_tpa, tip, sizeof(ea->arp_tpa));
256 	sa.sa_family = pseudo_AF_HDRCMPLT;
257 	sa.sa_len = sizeof(sa);
258 	m->m_flags |= M_BCAST;
259 	ifp->if_output(ifp, m, &sa, NULL);
260 }
261 
262 void
263 arpreply(struct ifnet *ifp, struct mbuf *m, struct in_addr *sip, uint8_t *eaddr)
264 {
265 	struct ether_header *eh;
266 	struct ether_arp *ea;
267 	struct sockaddr sa;
268 
269 	ea = mtod(m, struct ether_arp *);
270 	ea->arp_op = htons(ARPOP_REPLY);
271 	ea->arp_pro = htons(ETHERTYPE_IP); /* let's be sure! */
272 
273 	/* We're replying to a request. */
274 	memcpy(ea->arp_tha, ea->arp_sha, sizeof(ea->arp_sha));
275 	memcpy(ea->arp_tpa, ea->arp_spa, sizeof(ea->arp_spa));
276 
277 	memcpy(ea->arp_sha, eaddr, sizeof(ea->arp_sha));
278 	memcpy(ea->arp_spa, sip, sizeof(ea->arp_spa));
279 
280 	eh = (struct ether_header *)sa.sa_data;
281 	memcpy(eh->ether_dhost, ea->arp_tha, sizeof(eh->ether_dhost));
282 	memcpy(eh->ether_shost, eaddr, sizeof(eh->ether_shost));
283 	eh->ether_type = htons(ETHERTYPE_ARP);
284 	sa.sa_family = pseudo_AF_HDRCMPLT;
285 	sa.sa_len = sizeof(sa);
286 	ifp->if_output(ifp, m, &sa, NULL);
287 }
288 
289 /*
290  * Resolve an IP address into an ethernet address.  If success,
291  * desten is filled in.  If there is no entry in arptab,
292  * set one up and broadcast a request for the IP address.
293  * Hold onto this mbuf and resend it once the address
294  * is finally resolved.  A return value of 0 indicates
295  * that desten has been filled in and the packet should be sent
296  * normally; A return value of EAGAIN indicates that the packet
297  * has been taken over here, either now or for later transmission.
298  * Any other return value indicates an error.
299  */
300 int
301 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
302     struct sockaddr *dst, u_char *desten)
303 {
304 	struct arpcom *ac = (struct arpcom *)ifp;
305 	struct llinfo_arp *la = NULL;
306 	struct sockaddr_dl *sdl;
307 	struct rtentry *rt = NULL;
308 	char addr[INET_ADDRSTRLEN];
309 	int error;
310 
311 	if (m->m_flags & M_BCAST) {	/* broadcast */
312 		memcpy(desten, etherbroadcastaddr, sizeof(etherbroadcastaddr));
313 		return (0);
314 	}
315 	if (m->m_flags & M_MCAST) {	/* multicast */
316 		ETHER_MAP_IP_MULTICAST(&satosin(dst)->sin_addr, desten);
317 		return (0);
318 	}
319 
320 	error = rt_checkgate(rt0, &rt);
321 	if (error) {
322 		m_freem(m);
323 		return (error);
324 	}
325 
326 	if (!ISSET(rt->rt_flags, RTF_LLINFO)) {
327 		log(LOG_DEBUG, "%s: %s: route contains no arp information\n",
328 		    __func__, inet_ntop(AF_INET, &satosin(rt_key(rt))->sin_addr,
329 		    addr, sizeof(addr)));
330 		m_freem(m);
331 		return (EINVAL);
332 	}
333 
334 	sdl = satosdl(rt->rt_gateway);
335 	if (sdl->sdl_alen > 0 && sdl->sdl_alen != ETHER_ADDR_LEN) {
336 		log(LOG_DEBUG, "%s: %s: incorrect arp information\n", __func__,
337 		    inet_ntop(AF_INET, &satosin(dst)->sin_addr,
338 			addr, sizeof(addr)));
339 		goto bad;
340 	}
341 
342 	/*
343 	 * Check the address family and length is valid, the address
344 	 * is resolved; otherwise, try to resolve.
345 	 */
346 	if ((rt->rt_expire == 0 || rt->rt_expire > time_uptime) &&
347 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
348 		memcpy(desten, LLADDR(sdl), sdl->sdl_alen);
349 		return (0);
350 	}
351 
352 	if (ifp->if_flags & IFF_NOARP)
353 		goto bad;
354 
355 	/*
356 	 * There is an arptab entry, but no ethernet address
357 	 * response yet. Insert mbuf in hold queue if below limit
358 	 * if above the limit free the queue without queuing the new packet.
359 	 */
360 	la = (struct llinfo_arp *)rt->rt_llinfo;
361 	KASSERT(la != NULL);
362 	if (la_hold_total < LA_HOLD_TOTAL && la_hold_total < nmbclust / 64) {
363 		struct mbuf *mh;
364 
365 		if (ml_len(&la->la_ml) >= LA_HOLD_QUEUE) {
366 			mh = ml_dequeue(&la->la_ml);
367 			la_hold_total--;
368 			m_freem(mh);
369 		}
370 		ml_enqueue(&la->la_ml, m);
371 		la_hold_total++;
372 	} else {
373 		la_hold_total -= ml_purge(&la->la_ml);
374 		m_freem(m);
375 	}
376 
377 	/*
378 	 * Re-send the ARP request when appropriate.
379 	 */
380 #ifdef	DIAGNOSTIC
381 	if (rt->rt_expire == 0) {
382 		/* This should never happen. (Should it? -gwr) */
383 		printf("%s: unresolved and rt_expire == 0\n", __func__);
384 		/* Set expiration time to now (expired). */
385 		rt->rt_expire = time_uptime;
386 	}
387 #endif
388 	if (rt->rt_expire) {
389 		rt->rt_flags &= ~RTF_REJECT;
390 		if (la->la_asked == 0 || rt->rt_expire != time_uptime) {
391 			rt->rt_expire = time_uptime;
392 			if (la->la_asked++ < arp_maxtries)
393 				arprequest(ifp,
394 				    &satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr,
395 				    &satosin(dst)->sin_addr.s_addr,
396 				    ac->ac_enaddr);
397 			else {
398 				rt->rt_flags |= RTF_REJECT;
399 				rt->rt_expire += arpt_down;
400 				la->la_asked = 0;
401 				la_hold_total -= ml_purge(&la->la_ml);
402 			}
403 		}
404 	}
405 
406 	return (EAGAIN);
407 
408 bad:
409 	m_freem(m);
410 	return (EINVAL);
411 }
412 
413 /*
414  * Common length and type checks are done here,
415  * then the protocol-specific routine is called.
416  */
417 void
418 arpinput(struct ifnet *ifp, struct mbuf *m)
419 {
420 	struct arphdr *ar;
421 	int len;
422 
423 #ifdef DIAGNOSTIC
424 	if ((m->m_flags & M_PKTHDR) == 0)
425 		panic("arpintr");
426 #endif
427 
428 	len = sizeof(struct arphdr);
429 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
430 		return;
431 
432 	ar = mtod(m, struct arphdr *);
433 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER ||
434 	    ntohs(ar->ar_pro) != ETHERTYPE_IP) {
435 		m_freem(m);
436 		return;
437 	}
438 
439 	len += 2 * (ar->ar_hln + ar->ar_pln);
440 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
441 		return;
442 
443 	niq_enqueue(&arpinq, m);
444 }
445 
446 void
447 arpintr(void)
448 {
449 	struct mbuf_list ml;
450 	struct mbuf *m;
451 	struct ifnet *ifp;
452 
453 	niq_delist(&arpinq, &ml);
454 
455 	while ((m = ml_dequeue(&ml)) != NULL) {
456 		ifp = if_get(m->m_pkthdr.ph_ifidx);
457 
458 		if (ifp != NULL)
459 			in_arpinput(ifp, m);
460 		else
461 			m_freem(m);
462 
463 		if_put(ifp);
464 	}
465 }
466 
467 /*
468  * ARP for Internet protocols on Ethernet, RFC 826.
469  * In addition, a sanity check is performed on the sender
470  * protocol address, to catch impersonators.
471  */
472 void
473 in_arpinput(struct ifnet *ifp, struct mbuf *m)
474 {
475 	struct ether_arp *ea;
476 	struct rtentry *rt = NULL;
477 	struct sockaddr_in sin;
478 	struct in_addr isaddr, itaddr;
479 	char addr[INET_ADDRSTRLEN];
480 	int op, target = 0;
481 	unsigned int rdomain;
482 
483 	rdomain = rtable_l2(m->m_pkthdr.ph_rtableid);
484 
485 	ea = mtod(m, struct ether_arp *);
486 	op = ntohs(ea->arp_op);
487 	if ((op != ARPOP_REQUEST) && (op != ARPOP_REPLY))
488 		goto out;
489 
490 	memcpy(&itaddr, ea->arp_tpa, sizeof(itaddr));
491 	memcpy(&isaddr, ea->arp_spa, sizeof(isaddr));
492 	memset(&sin, 0, sizeof(sin));
493 	sin.sin_len = sizeof(sin);
494 	sin.sin_family = AF_INET;
495 
496 	if (ETHER_IS_MULTICAST(&ea->arp_sha[0]) &&
497 	    !memcmp(ea->arp_sha, etherbroadcastaddr, sizeof(ea->arp_sha))) {
498 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
499 		log(LOG_ERR, "arp: ether address is broadcast for IP address "
500 		    "%s!\n", addr);
501 		goto out;
502 	}
503 
504 	if (!memcmp(ea->arp_sha, LLADDR(ifp->if_sadl), sizeof(ea->arp_sha)))
505 		goto out;	/* it's from me, ignore it. */
506 
507 	/* Check target against our interface addresses. */
508 	sin.sin_addr = itaddr;
509 	rt = rtalloc(sintosa(&sin), 0, rdomain);
510 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
511 	    rt->rt_ifidx == ifp->if_index)
512 		target = 1;
513 	rtfree(rt);
514 	rt = NULL;
515 
516 #if NCARP > 0
517 	if (target && op == ARPOP_REQUEST && ifp->if_type == IFT_CARP &&
518 	    !carp_iamatch(ifp))
519 		goto out;
520 #endif
521 
522 	/* Do we have an ARP cache for the sender? Create if we are target. */
523 	rt = arplookup(&isaddr, target, 0, rdomain);
524 
525 	/* Check sender against our interface addresses. */
526 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
527 	    rt->rt_ifidx == ifp->if_index && isaddr.s_addr != INADDR_ANY) {
528 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
529 		log(LOG_ERR, "duplicate IP address %s sent from ethernet "
530 		    "address %s\n", addr, ether_sprintf(ea->arp_sha));
531 		itaddr = isaddr;
532 	} else if (rt != NULL) {
533 		int error;
534 
535 		KERNEL_LOCK();
536 		error = arpcache(ifp, ea, rt);
537 		KERNEL_UNLOCK();
538 		if (error)
539 			goto out;
540 	}
541 
542 	if (op == ARPOP_REQUEST) {
543 		uint8_t *eaddr;
544 
545 		if (target) {
546 			/* We already have all info for the reply */
547 			eaddr = LLADDR(ifp->if_sadl);
548 		} else {
549 			rtfree(rt);
550 			rt = arplookup(&itaddr, 0, SIN_PROXY, rdomain);
551 			/*
552 			 * Protect from possible duplicates, only owner
553 			 * should respond
554 			 */
555 			if ((rt == NULL) || (rt->rt_ifidx != ifp->if_index))
556 				goto out;
557 			eaddr = LLADDR(satosdl(rt->rt_gateway));
558 		}
559 		arpreply(ifp, m, &itaddr, eaddr);
560 		rtfree(rt);
561 		return;
562 	}
563 
564 out:
565 	rtfree(rt);
566 	m_freem(m);
567 }
568 
569 int
570 arpcache(struct ifnet *ifp, struct ether_arp *ea, struct rtentry *rt)
571 {
572 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
573 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
574 	struct in_addr *spa = (struct in_addr *)ea->arp_spa;
575 	char addr[INET_ADDRSTRLEN];
576 	struct ifnet *rifp;
577 	unsigned int len;
578 	int changed = 0;
579 
580 	KERNEL_ASSERT_LOCKED();
581 	KASSERT(sdl != NULL);
582 
583 	/*
584 	 * This can happen if the entry has been deleted by another CPU
585 	 * after we found it.
586 	 */
587 	if (la == NULL)
588 		return (0);
589 
590 	if (sdl->sdl_alen > 0) {
591 		if (memcmp(ea->arp_sha, LLADDR(sdl), sdl->sdl_alen)) {
592 			if (ISSET(rt->rt_flags, RTF_PERMANENT_ARP|RTF_LOCAL)) {
593 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
594 				log(LOG_WARNING, "arp: attempt to overwrite "
595 				   "permanent entry for %s by %s on %s\n", addr,
596 				   ether_sprintf(ea->arp_sha), ifp->if_xname);
597 				return (-1);
598 			} else if (rt->rt_ifidx != ifp->if_index) {
599 #if NCARP > 0
600 				if (ifp->if_type != IFT_CARP)
601 #endif
602 				{
603 					rifp = if_get(rt->rt_ifidx);
604 					if (rifp == NULL)
605 						return (-1);
606 					inet_ntop(AF_INET, spa, addr,
607 					    sizeof(addr));
608 					log(LOG_WARNING, "arp: attempt to "
609 					    "overwrite entry for %s on %s by "
610 					    "%s on %s\n", addr, rifp->if_xname,
611 					    ether_sprintf(ea->arp_sha),
612 					    ifp->if_xname);
613 					if_put(rifp);
614 				}
615 				return (-1);
616 			} else {
617 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
618 				log(LOG_INFO, "arp info overwritten for %s by "
619 				    "%s on %s\n", addr,
620 				    ether_sprintf(ea->arp_sha), ifp->if_xname);
621 				rt->rt_expire = 1;/* no longer static */
622 			}
623 			changed = 1;
624 		}
625 	} else if (!if_isconnected(ifp, rt->rt_ifidx)) {
626 		rifp = if_get(rt->rt_ifidx);
627 		if (rifp == NULL)
628 			return (-1);
629 		inet_ntop(AF_INET, spa, addr, sizeof(addr));
630 		log(LOG_WARNING, "arp: attempt to add entry for %s on %s by %s"
631 		    " on %s\n", addr, rifp->if_xname,
632 		    ether_sprintf(ea->arp_sha), ifp->if_xname);
633 		if_put(rifp);
634 		return (-1);
635 	}
636 	sdl->sdl_alen = sizeof(ea->arp_sha);
637 	memcpy(LLADDR(sdl), ea->arp_sha, sizeof(ea->arp_sha));
638 	if (rt->rt_expire)
639 		rt->rt_expire = time_uptime + arpt_keep;
640 	rt->rt_flags &= ~RTF_REJECT;
641 
642 	/* Notify userland that an ARP resolution has been done. */
643 	if (la->la_asked || changed) {
644 		KERNEL_LOCK();
645 		rt_sendmsg(rt, RTM_RESOLVE, ifp->if_rdomain);
646 		KERNEL_UNLOCK();
647 	}
648 
649 	la->la_asked = 0;
650 	while ((len = ml_len(&la->la_ml)) != 0) {
651 		struct mbuf *mh;
652 
653 		mh = ml_dequeue(&la->la_ml);
654 		la_hold_total--;
655 
656 		ifp->if_output(ifp, mh, rt_key(rt), rt);
657 
658 		if (ml_len(&la->la_ml) == len) {
659 			/* mbuf is back in queue. Discard. */
660 			while ((mh = ml_dequeue(&la->la_ml)) != NULL) {
661 				la_hold_total--;
662 				m_freem(mh);
663 			}
664 			break;
665 		}
666 	}
667 
668 	return (0);
669 }
670 /*
671  * Free an arp entry.
672  */
673 void
674 arptfree(struct rtentry *rt)
675 {
676 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
677 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
678 	struct ifnet *ifp;
679 
680 	ifp = if_get(rt->rt_ifidx);
681 	if ((sdl != NULL) && (sdl->sdl_family == AF_LINK)) {
682 		sdl->sdl_alen = 0;
683 		la->la_asked = 0;
684 	}
685 
686 	if (!ISSET(rt->rt_flags, RTF_STATIC))
687 		rtdeletemsg(rt, ifp, ifp->if_rdomain);
688 	if_put(ifp);
689 }
690 
691 /*
692  * Lookup or enter a new address in arptab.
693  */
694 struct rtentry *
695 arplookup(struct in_addr *inp, int create, int proxy, u_int tableid)
696 {
697 	struct rtentry *rt;
698 	struct sockaddr_inarp sin;
699 	int flags;
700 
701 	memset(&sin, 0, sizeof(sin));
702 	sin.sin_len = sizeof(sin);
703 	sin.sin_family = AF_INET;
704 	sin.sin_addr.s_addr = inp->s_addr;
705 	sin.sin_other = proxy ? SIN_PROXY : 0;
706 	flags = (create) ? RT_RESOLVE : 0;
707 
708 	rt = rtalloc((struct sockaddr *)&sin, flags, tableid);
709 	if (!rtisvalid(rt) || ISSET(rt->rt_flags, RTF_GATEWAY) ||
710 	    !ISSET(rt->rt_flags, RTF_LLINFO) ||
711 	    rt->rt_gateway->sa_family != AF_LINK) {
712 		rtfree(rt);
713 		return (NULL);
714 	}
715 
716 	if (proxy && !ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
717 		struct rtentry *mrt = NULL;
718 #if defined(ART) && !defined(SMALL_KERNEL)
719 		mrt = rt;
720 		KERNEL_LOCK();
721 		while ((mrt = rtable_mpath_next(mrt)) != NULL) {
722 			if (ISSET(mrt->rt_flags, RTF_ANNOUNCE)) {
723 				rtref(mrt);
724 				break;
725 			}
726 		}
727 		KERNEL_UNLOCK();
728 #endif /* ART && !SMALL_KERNEL */
729 		rtfree(rt);
730 		return (mrt);
731 	}
732 
733 	return (rt);
734 }
735 
736 /*
737  * Check whether we do proxy ARP for this address and we point to ourselves.
738  */
739 int
740 arpproxy(struct in_addr in, unsigned int rtableid)
741 {
742 	struct sockaddr_dl *sdl;
743 	struct rtentry *rt;
744 	struct ifnet *ifp;
745 	int found = 0;
746 
747 	rt = arplookup(&in, 0, SIN_PROXY, rtableid);
748 	if (!rtisvalid(rt)) {
749 		rtfree(rt);
750 		return (0);
751 	}
752 
753 	/* Check that arp information are correct. */
754 	sdl = satosdl(rt->rt_gateway);
755 	if (sdl->sdl_alen != ETHER_ADDR_LEN) {
756 		rtfree(rt);
757 		return (0);
758 	}
759 
760 	ifp = if_get(rt->rt_ifidx);
761 	if (ifp == NULL) {
762 		rtfree(rt);
763 		return (0);
764 	}
765 
766 	if (!memcmp(LLADDR(sdl), LLADDR(ifp->if_sadl), sdl->sdl_alen))
767 		found = 1;
768 
769 	if_put(ifp);
770 	rtfree(rt);
771 	return (found);
772 }
773 
774 /*
775  * Called from Ethernet interrupt handlers
776  * when ether packet type ETHERTYPE_REVARP
777  * is received.  Common length and type checks are done here,
778  * then the protocol-specific routine is called.
779  */
780 void
781 revarpinput(struct ifnet *ifp, struct mbuf *m)
782 {
783 	struct arphdr *ar;
784 
785 	if (m->m_len < sizeof(struct arphdr))
786 		goto out;
787 	ar = mtod(m, struct arphdr *);
788 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER)
789 		goto out;
790 	if (m->m_len < sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln))
791 		goto out;
792 	switch (ntohs(ar->ar_pro)) {
793 
794 	case ETHERTYPE_IP:
795 		in_revarpinput(ifp, m);
796 		return;
797 
798 	default:
799 		break;
800 	}
801 out:
802 	m_freem(m);
803 }
804 
805 /*
806  * RARP for Internet protocols on Ethernet.
807  * Algorithm is that given in RFC 903.
808  * We are only using for bootstrap purposes to get an ip address for one of
809  * our interfaces.  Thus we support no user-interface.
810  *
811  * Since the contents of the RARP reply are specific to the interface that
812  * sent the request, this code must ensure that they are properly associated.
813  *
814  * Note: also supports ARP via RARP packets, per the RFC.
815  */
816 void
817 in_revarpinput(struct ifnet *ifp, struct mbuf *m)
818 {
819 	struct ether_arp *ar;
820 	int op;
821 
822 	ar = mtod(m, struct ether_arp *);
823 	op = ntohs(ar->arp_op);
824 	switch (op) {
825 	case ARPOP_REQUEST:
826 	case ARPOP_REPLY:	/* per RFC */
827 		niq_enqueue(&arpinq, m);
828 		return;
829 	case ARPOP_REVREPLY:
830 		break;
831 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
832 	default:
833 		goto out;
834 	}
835 #ifdef NFSCLIENT
836 	if (revarp_ifidx == 0)
837 		goto out;
838 	if (revarp_ifidx != m->m_pkthdr.ph_ifidx) /* !same interface */
839 		goto out;
840 	if (revarp_finished)
841 		goto wake;
842 	if (memcmp(ar->arp_tha, LLADDR(ifp->if_sadl), sizeof(ar->arp_tha)))
843 		goto out;
844 	memcpy(&revarp_srvip, ar->arp_spa, sizeof(revarp_srvip));
845 	memcpy(&revarp_myip, ar->arp_tpa, sizeof(revarp_myip));
846 	revarp_finished = 1;
847 wake:	/* Do wakeup every time in case it was missed. */
848 	wakeup((caddr_t)&revarp_myip);
849 #endif /* NFSCLIENT */
850 
851 out:
852 	m_freem(m);
853 }
854 
855 /*
856  * Send a RARP request for the ip address of the specified interface.
857  * The request should be RFC 903-compliant.
858  */
859 void
860 revarprequest(struct ifnet *ifp)
861 {
862 	struct sockaddr sa;
863 	struct mbuf *m;
864 	struct ether_header *eh;
865 	struct ether_arp *ea;
866 	struct arpcom *ac = (struct arpcom *)ifp;
867 
868 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
869 		return;
870 	m->m_len = sizeof(*ea);
871 	m->m_pkthdr.len = sizeof(*ea);
872 	m->m_pkthdr.pf.prio = ifp->if_llprio;
873 	MH_ALIGN(m, sizeof(*ea));
874 	ea = mtod(m, struct ether_arp *);
875 	eh = (struct ether_header *)sa.sa_data;
876 	memset(ea, 0, sizeof(*ea));
877 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
878 	eh->ether_type = htons(ETHERTYPE_REVARP);
879 	ea->arp_hrd = htons(ARPHRD_ETHER);
880 	ea->arp_pro = htons(ETHERTYPE_IP);
881 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
882 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
883 	ea->arp_op = htons(ARPOP_REVREQUEST);
884 	memcpy(eh->ether_shost, ac->ac_enaddr, sizeof(ea->arp_tha));
885 	memcpy(ea->arp_sha, ac->ac_enaddr, sizeof(ea->arp_sha));
886 	memcpy(ea->arp_tha, ac->ac_enaddr, sizeof(ea->arp_tha));
887 	sa.sa_family = pseudo_AF_HDRCMPLT;
888 	sa.sa_len = sizeof(sa);
889 	m->m_flags |= M_BCAST;
890 	ifp->if_output(ifp, m, &sa, NULL);
891 }
892 
893 #ifdef NFSCLIENT
894 /*
895  * RARP for the ip address of the specified interface, but also
896  * save the ip address of the server that sent the answer.
897  * Timeout if no response is received.
898  */
899 int
900 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
901     struct in_addr *clnt_in)
902 {
903 	int result, count = 20;
904 
905 	if (revarp_finished)
906 		return EIO;
907 
908 	revarp_ifidx = ifp->if_index;
909 	while (count--) {
910 		revarprequest(ifp);
911 		result = tsleep((caddr_t)&revarp_myip, PSOCK, "revarp", hz/2);
912 		if (result != EWOULDBLOCK)
913 			break;
914 	}
915 	revarp_ifidx = 0;
916 	if (!revarp_finished)
917 		return ENETUNREACH;
918 
919 	memcpy(serv_in, &revarp_srvip, sizeof(*serv_in));
920 	memcpy(clnt_in, &revarp_myip, sizeof(*clnt_in));
921 	return 0;
922 }
923 
924 /* For compatibility: only saves interface address. */
925 int
926 revarpwhoami(struct in_addr *in, struct ifnet *ifp)
927 {
928 	struct in_addr server;
929 	return (revarpwhoarewe(ifp, &server, in));
930 }
931 #endif /* NFSCLIENT */
932