xref: /openbsd-src/sys/netinet/if_ether.c (revision 3374c67d44f9b75b98444cbf63020f777792342e)
1 /*	$OpenBSD: if_ether.c,v 1.252 2022/12/07 14:38:29 claudio 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 #include <netinet/ip_var.h>
64 #if NCARP > 0
65 #include <netinet/ip_carp.h>
66 #endif
67 
68 /*
69  *  Locks used to protect struct members in this file:
70  *	a	atomic operations
71  *	I	immutable after creation
72  *	K	kernel lock
73  *	m	arp mutex, needed when net lock is shared
74  *	N	net lock
75  */
76 
77 struct llinfo_arp {
78 	LIST_ENTRY(llinfo_arp)	 la_list;	/* [mN] global arp_list */
79 	struct rtentry		*la_rt;		/* [I] backpointer to rtentry */
80 	struct mbuf_queue	 la_mq;		/* packet hold queue */
81 	time_t			 la_refreshed;	/* when was refresh sent */
82 	int			 la_asked;	/* number of queries sent */
83 };
84 #define LA_HOLD_QUEUE 10
85 #define LA_HOLD_TOTAL 100
86 
87 /* timer values */
88 int 	arpt_prune = (5 * 60);	/* [I] walk list every 5 minutes */
89 int 	arpt_keep = (20 * 60);	/* [a] once resolved, cache for 20 minutes */
90 int 	arpt_down = 20;	/* [a] once declared down, don't send for 20 secs */
91 
92 struct mbuf *arppullup(struct mbuf *m);
93 void arpinvalidate(struct rtentry *);
94 void arptfree(struct rtentry *);
95 void arptimer(void *);
96 struct rtentry *arplookup(struct in_addr *, int, int, unsigned int);
97 void in_arpinput(struct ifnet *, struct mbuf *);
98 void in_revarpinput(struct ifnet *, struct mbuf *);
99 int arpcache(struct ifnet *, struct ether_arp *, struct rtentry *);
100 void arpreply(struct ifnet *, struct mbuf *, struct in_addr *, uint8_t *,
101     unsigned int);
102 
103 struct niqueue arpinq = NIQUEUE_INITIALIZER(50, NETISR_ARP);
104 
105 /* llinfo_arp live time, rt_llinfo and RTF_LLINFO are protected by arp_mtx */
106 struct mutex arp_mtx = MUTEX_INITIALIZER(IPL_SOFTNET);
107 
108 LIST_HEAD(, llinfo_arp) arp_list; /* [mN] list of all llinfo_arp structures */
109 struct	pool arp_pool;		/* [I] pool for llinfo_arp structures */
110 int	arp_maxtries = 5;	/* [I] arp requests before set to rejected */
111 int	la_hold_total;		/* [a] packets currently in the arp queue */
112 
113 #ifdef NFSCLIENT
114 /* revarp state */
115 struct in_addr revarp_myip, revarp_srvip;
116 int revarp_finished;
117 unsigned int revarp_ifidx;
118 #endif /* NFSCLIENT */
119 
120 /*
121  * Timeout routine.  Age arp_tab entries periodically.
122  */
123 /* ARGSUSED */
124 void
125 arptimer(void *arg)
126 {
127 	struct timeout *to = arg;
128 	struct llinfo_arp *la, *nla;
129 	time_t uptime;
130 
131 	NET_LOCK();
132 	uptime = getuptime();
133 	timeout_add_sec(to, arpt_prune);
134 	/* Net lock is exclusive, no arp mutex needed for arp_list here. */
135 	LIST_FOREACH_SAFE(la, &arp_list, la_list, nla) {
136 		struct rtentry *rt = la->la_rt;
137 
138 		if (rt->rt_expire && rt->rt_expire < uptime)
139 			arptfree(rt); /* timer has expired; clear */
140 	}
141 	NET_UNLOCK();
142 }
143 
144 void
145 arpinit(void)
146 {
147 	static struct timeout arptimer_to;
148 
149 	pool_init(&arp_pool, sizeof(struct llinfo_arp), 0,
150 	    IPL_SOFTNET, 0, "arp", NULL);
151 
152 	timeout_set_proc(&arptimer_to, arptimer, &arptimer_to);
153 	timeout_add_sec(&arptimer_to, arpt_prune);
154 }
155 
156 void
157 arp_rtrequest(struct ifnet *ifp, int req, struct rtentry *rt)
158 {
159 	struct sockaddr *gate = rt->rt_gateway;
160 	struct llinfo_arp *la;
161 	time_t uptime;
162 
163 	NET_ASSERT_LOCKED();
164 
165 	if (ISSET(rt->rt_flags,
166 	    RTF_GATEWAY|RTF_BROADCAST|RTF_MULTICAST|RTF_MPLS))
167 		return;
168 
169 	uptime = getuptime();
170 	switch (req) {
171 	case RTM_ADD:
172 		if (rt->rt_flags & RTF_CLONING) {
173 			rt->rt_expire = 0;
174 			break;
175 		}
176 		if ((rt->rt_flags & RTF_LOCAL) && rt->rt_llinfo == NULL)
177 			rt->rt_expire = 0;
178 		/*
179 		 * Announce a new entry if requested or warn the user
180 		 * if another station has this IP address.
181 		 */
182 		if (rt->rt_flags & (RTF_ANNOUNCE|RTF_LOCAL))
183 			arprequest(ifp,
184 			    &satosin(rt_key(rt))->sin_addr.s_addr,
185 			    &satosin(rt_key(rt))->sin_addr.s_addr,
186 			    (u_char *)LLADDR(satosdl(gate)));
187 		/*FALLTHROUGH*/
188 	case RTM_RESOLVE:
189 		if (gate->sa_family != AF_LINK ||
190 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
191 			log(LOG_DEBUG, "%s: bad gateway value: %s\n", __func__,
192 			    ifp->if_xname);
193 			break;
194 		}
195 		satosdl(gate)->sdl_type = ifp->if_type;
196 		satosdl(gate)->sdl_index = ifp->if_index;
197 		/*
198 		 * Case 2:  This route may come from cloning, or a manual route
199 		 * add with a LL address.
200 		 */
201 		la = pool_get(&arp_pool, PR_NOWAIT | PR_ZERO);
202 		if (la == NULL) {
203 			log(LOG_DEBUG, "%s: pool get failed\n", __func__);
204 			break;
205 		}
206 
207 		mtx_enter(&arp_mtx);
208 		if (rt->rt_llinfo != NULL) {
209 			/* we lost the race, another thread has entered it */
210 			mtx_leave(&arp_mtx);
211 			pool_put(&arp_pool, la);
212 			break;
213 		}
214 		mq_init(&la->la_mq, LA_HOLD_QUEUE, IPL_SOFTNET);
215 		rt->rt_llinfo = (caddr_t)la;
216 		la->la_rt = rt;
217 		rt->rt_flags |= RTF_LLINFO;
218 		LIST_INSERT_HEAD(&arp_list, la, la_list);
219 		if ((rt->rt_flags & RTF_LOCAL) == 0)
220 			rt->rt_expire = uptime;
221 		mtx_leave(&arp_mtx);
222 
223 		break;
224 
225 	case RTM_DELETE:
226 		mtx_enter(&arp_mtx);
227 		la = (struct llinfo_arp *)rt->rt_llinfo;
228 		if (la == NULL) {
229 			/* we lost the race, another thread has removed it */
230 			mtx_leave(&arp_mtx);
231 			break;
232 		}
233 		LIST_REMOVE(la, la_list);
234 		rt->rt_llinfo = NULL;
235 		rt->rt_flags &= ~RTF_LLINFO;
236 		atomic_sub_int(&la_hold_total, mq_purge(&la->la_mq));
237 		mtx_leave(&arp_mtx);
238 
239 		pool_put(&arp_pool, la);
240 		break;
241 
242 	case RTM_INVALIDATE:
243 		if (!ISSET(rt->rt_flags, RTF_LOCAL))
244 			arpinvalidate(rt);
245 		break;
246 	}
247 }
248 
249 /*
250  * Broadcast an ARP request. Caller specifies:
251  *	- arp header source ip address
252  *	- arp header target ip address
253  *	- arp header source ethernet address
254  */
255 void
256 arprequest(struct ifnet *ifp, u_int32_t *sip, u_int32_t *tip, u_int8_t *enaddr)
257 {
258 	struct mbuf *m;
259 	struct ether_header *eh;
260 	struct ether_arp *ea;
261 	struct sockaddr sa;
262 
263 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
264 		return;
265 	m->m_len = sizeof(*ea);
266 	m->m_pkthdr.len = sizeof(*ea);
267 	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
268 	m->m_pkthdr.pf.prio = ifp->if_llprio;
269 	m_align(m, sizeof(*ea));
270 	ea = mtod(m, struct ether_arp *);
271 	eh = (struct ether_header *)sa.sa_data;
272 	memset(ea, 0, sizeof(*ea));
273 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
274 	eh->ether_type = htons(ETHERTYPE_ARP);	/* if_output will not swap */
275 	ea->arp_hrd = htons(ARPHRD_ETHER);
276 	ea->arp_pro = htons(ETHERTYPE_IP);
277 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
278 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
279 	ea->arp_op = htons(ARPOP_REQUEST);
280 	memcpy(eh->ether_shost, enaddr, sizeof(eh->ether_shost));
281 	memcpy(ea->arp_sha, enaddr, sizeof(ea->arp_sha));
282 	memcpy(ea->arp_spa, sip, sizeof(ea->arp_spa));
283 	memcpy(ea->arp_tpa, tip, sizeof(ea->arp_tpa));
284 	sa.sa_family = pseudo_AF_HDRCMPLT;
285 	sa.sa_len = sizeof(sa);
286 	m->m_flags |= M_BCAST;
287 	ifp->if_output(ifp, m, &sa, NULL);
288 }
289 
290 void
291 arpreply(struct ifnet *ifp, struct mbuf *m, struct in_addr *sip, uint8_t *eaddr,
292     unsigned int rdomain)
293 {
294 	struct ether_header *eh;
295 	struct ether_arp *ea;
296 	struct sockaddr sa;
297 
298 	m_resethdr(m);
299 	m->m_pkthdr.ph_rtableid = rdomain;
300 
301 	ea = mtod(m, struct ether_arp *);
302 	ea->arp_op = htons(ARPOP_REPLY);
303 	ea->arp_pro = htons(ETHERTYPE_IP); /* let's be sure! */
304 
305 	/* We're replying to a request. */
306 	memcpy(ea->arp_tha, ea->arp_sha, sizeof(ea->arp_sha));
307 	memcpy(ea->arp_tpa, ea->arp_spa, sizeof(ea->arp_spa));
308 
309 	memcpy(ea->arp_sha, eaddr, sizeof(ea->arp_sha));
310 	memcpy(ea->arp_spa, sip, sizeof(ea->arp_spa));
311 
312 	eh = (struct ether_header *)sa.sa_data;
313 	memcpy(eh->ether_dhost, ea->arp_tha, sizeof(eh->ether_dhost));
314 	memcpy(eh->ether_shost, eaddr, sizeof(eh->ether_shost));
315 	eh->ether_type = htons(ETHERTYPE_ARP);
316 	sa.sa_family = pseudo_AF_HDRCMPLT;
317 	sa.sa_len = sizeof(sa);
318 	ifp->if_output(ifp, m, &sa, NULL);
319 }
320 
321 /*
322  * Resolve an IP address into an ethernet address.  If success,
323  * desten is filled in.  If there is no entry in arptab,
324  * set one up and broadcast a request for the IP address.
325  * Hold onto this mbuf and resend it once the address
326  * is finally resolved.  A return value of 0 indicates
327  * that desten has been filled in and the packet should be sent
328  * normally; A return value of EAGAIN indicates that the packet
329  * has been taken over here, either now or for later transmission.
330  * Any other return value indicates an error.
331  */
332 int
333 arpresolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
334     struct sockaddr *dst, u_char *desten)
335 {
336 	struct arpcom *ac = (struct arpcom *)ifp;
337 	struct llinfo_arp *la;
338 	struct sockaddr_dl *sdl;
339 	struct rtentry *rt = NULL;
340 	char addr[INET_ADDRSTRLEN];
341 	time_t uptime;
342 
343 	if (m->m_flags & M_BCAST) {	/* broadcast */
344 		memcpy(desten, etherbroadcastaddr, sizeof(etherbroadcastaddr));
345 		return (0);
346 	}
347 	if (m->m_flags & M_MCAST) {	/* multicast */
348 		ETHER_MAP_IP_MULTICAST(&satosin(dst)->sin_addr, desten);
349 		return (0);
350 	}
351 
352 	uptime = getuptime();
353 	rt = rt_getll(rt0);
354 
355 	if (ISSET(rt->rt_flags, RTF_REJECT) &&
356 	    (rt->rt_expire == 0 || rt->rt_expire > uptime)) {
357 		m_freem(m);
358 		return (rt == rt0 ? EHOSTDOWN : EHOSTUNREACH);
359 	}
360 
361 	if (!ISSET(rt->rt_flags, RTF_LLINFO)) {
362 		log(LOG_DEBUG, "%s: %s: route contains no arp information\n",
363 		    __func__, inet_ntop(AF_INET, &satosin(rt_key(rt))->sin_addr,
364 		    addr, sizeof(addr)));
365 		goto bad;
366 	}
367 
368 	sdl = satosdl(rt->rt_gateway);
369 	if (sdl->sdl_alen > 0 && sdl->sdl_alen != ETHER_ADDR_LEN) {
370 		log(LOG_DEBUG, "%s: %s: incorrect arp information\n", __func__,
371 		    inet_ntop(AF_INET, &satosin(dst)->sin_addr,
372 			addr, sizeof(addr)));
373 		goto bad;
374 	}
375 
376 
377 	/*
378 	 * Check the address family and length is valid, the address
379 	 * is resolved; otherwise, try to resolve.
380 	 */
381 	if ((rt->rt_expire == 0 || rt->rt_expire > uptime) &&
382 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
383 		int refresh = 0;
384 
385 		memcpy(desten, LLADDR(sdl), sdl->sdl_alen);
386 
387 		/* refresh ARP entry when timeout gets close */
388 		if (rt->rt_expire != 0 &&
389 		    rt->rt_expire - arpt_keep / 8 < uptime) {
390 
391 			mtx_enter(&arp_mtx);
392 			if (ISSET(rt->rt_flags, RTF_LLINFO)) {
393 				la = (struct llinfo_arp *)rt->rt_llinfo;
394 				KASSERT(la != NULL);
395 
396 				if (la->la_refreshed + 30 < uptime) {
397 					la->la_refreshed = uptime;
398 					refresh = 1;
399 				}
400 			}
401 			mtx_leave(&arp_mtx);
402 		}
403 		if (refresh) {
404 			arprequest(ifp,
405 			    &satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr,
406 			    &satosin(dst)->sin_addr.s_addr,
407 			    ac->ac_enaddr);
408 		}
409 		return (0);
410 	}
411 
412 	if (ifp->if_flags & (IFF_NOARP|IFF_STATICARP))
413 		goto bad;
414 
415 	KERNEL_LOCK();
416 	/*
417 	 * Re-check since we grab the kernel lock after the first check.
418 	 * rtrequest_delete() can be called with shared netlock.  From
419 	 * there arp_rtrequest() is reached which touches RTF_LLINFO
420 	 * and rt_llinfo.  As this is called with kernel lock we grab the
421 	 * kernel lock here and are safe.  XXXSMP
422 	 */
423 	if (!ISSET(rt->rt_flags, RTF_LLINFO)) {
424 		KERNEL_UNLOCK();
425 		goto bad;
426 	}
427 	la = (struct llinfo_arp *)rt->rt_llinfo;
428 	KASSERT(la != NULL);
429 
430 	/*
431 	 * There is an arptab entry, but no ethernet address
432 	 * response yet. Insert mbuf in hold queue if below limit.
433 	 * If above the limit free the queue without queuing the new packet.
434 	 */
435 	if (atomic_inc_int_nv(&la_hold_total) <= LA_HOLD_TOTAL) {
436 		if (mq_push(&la->la_mq, m) != 0)
437 			atomic_dec_int(&la_hold_total);
438 	} else {
439 		atomic_sub_int(&la_hold_total, mq_purge(&la->la_mq) + 1);
440 		m_freem(m);
441 	}
442 
443 	/*
444 	 * Re-send the ARP request when appropriate.
445 	 */
446 #ifdef	DIAGNOSTIC
447 	if (rt->rt_expire == 0) {
448 		/* This should never happen. (Should it? -gwr) */
449 		printf("%s: unresolved and rt_expire == 0\n", __func__);
450 		/* Set expiration time to now (expired). */
451 		rt->rt_expire = uptime;
452 	}
453 #endif
454 	if (rt->rt_expire) {
455 		rt->rt_flags &= ~RTF_REJECT;
456 		if (la->la_asked == 0 || rt->rt_expire != uptime) {
457 			rt->rt_expire = uptime;
458 			if (la->la_asked++ < arp_maxtries)
459 				arprequest(ifp,
460 				    &satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr,
461 				    &satosin(dst)->sin_addr.s_addr,
462 				    ac->ac_enaddr);
463 			else {
464 				rt->rt_flags |= RTF_REJECT;
465 				rt->rt_expire += arpt_down;
466 				la->la_asked = 0;
467 				la->la_refreshed = 0;
468 				atomic_sub_int(&la_hold_total,
469 				    mq_purge(&la->la_mq));
470 			}
471 		}
472 	}
473 
474 	KERNEL_UNLOCK();
475 	return (EAGAIN);
476 
477 bad:
478 	m_freem(m);
479 	return (EINVAL);
480 }
481 
482 struct mbuf *
483 arppullup(struct mbuf *m)
484 {
485 	struct arphdr *ar;
486 	int len;
487 
488 #ifdef DIAGNOSTIC
489 	if ((m->m_flags & M_PKTHDR) == 0)
490 		panic("arp without packet header");
491 #endif
492 
493 	len = sizeof(struct arphdr);
494 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
495 		return NULL;
496 
497 	ar = mtod(m, struct arphdr *);
498 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER ||
499 	    ntohs(ar->ar_pro) != ETHERTYPE_IP ||
500 	    ar->ar_hln != ETHER_ADDR_LEN ||
501 	    ar->ar_pln != sizeof(struct in_addr)) {
502 		m_freem(m);
503 		return NULL;
504 	}
505 
506 	len += 2 * (ar->ar_hln + ar->ar_pln);
507 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
508 		return NULL;
509 
510 	return m;
511 }
512 
513 /*
514  * Common length and type checks are done here,
515  * then the protocol-specific routine is called.
516  */
517 void
518 arpinput(struct ifnet *ifp, struct mbuf *m)
519 {
520 	if ((m = arppullup(m)) == NULL)
521 		return;
522 	niq_enqueue(&arpinq, m);
523 }
524 
525 void
526 arpintr(void)
527 {
528 	struct mbuf_list ml;
529 	struct mbuf *m;
530 	struct ifnet *ifp;
531 
532 	niq_delist(&arpinq, &ml);
533 
534 	while ((m = ml_dequeue(&ml)) != NULL) {
535 		ifp = if_get(m->m_pkthdr.ph_ifidx);
536 
537 		if (ifp != NULL)
538 			in_arpinput(ifp, m);
539 		else
540 			m_freem(m);
541 
542 		if_put(ifp);
543 	}
544 }
545 
546 /*
547  * ARP for Internet protocols on Ethernet, RFC 826.
548  * In addition, a sanity check is performed on the sender
549  * protocol address, to catch impersonators.
550  */
551 void
552 in_arpinput(struct ifnet *ifp, struct mbuf *m)
553 {
554 	struct ether_arp *ea;
555 	struct rtentry *rt = NULL;
556 	struct sockaddr_in sin;
557 	struct in_addr isaddr, itaddr;
558 	char addr[INET_ADDRSTRLEN];
559 	int op, target = 0;
560 	unsigned int rdomain;
561 
562 	rdomain = rtable_l2(m->m_pkthdr.ph_rtableid);
563 
564 	ea = mtod(m, struct ether_arp *);
565 	op = ntohs(ea->arp_op);
566 	if ((op != ARPOP_REQUEST) && (op != ARPOP_REPLY))
567 		goto out;
568 
569 	memcpy(&itaddr, ea->arp_tpa, sizeof(itaddr));
570 	memcpy(&isaddr, ea->arp_spa, sizeof(isaddr));
571 	memset(&sin, 0, sizeof(sin));
572 	sin.sin_len = sizeof(sin);
573 	sin.sin_family = AF_INET;
574 
575 	if (ETHER_IS_MULTICAST(ea->arp_sha) &&
576 	    ETHER_IS_BROADCAST(ea->arp_sha)) {
577 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
578 		log(LOG_ERR, "arp: ether address is broadcast for IP address "
579 		    "%s!\n", addr);
580 		goto out;
581 	}
582 
583 	if (!memcmp(ea->arp_sha, LLADDR(ifp->if_sadl), sizeof(ea->arp_sha)))
584 		goto out;	/* it's from me, ignore it. */
585 
586 	/* Check target against our interface addresses. */
587 	sin.sin_addr = itaddr;
588 	rt = rtalloc(sintosa(&sin), 0, rdomain);
589 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
590 	    rt->rt_ifidx == ifp->if_index)
591 		target = 1;
592 	rtfree(rt);
593 	rt = NULL;
594 
595 #if NCARP > 0
596 	if (target && op == ARPOP_REQUEST && ifp->if_type == IFT_CARP &&
597 	    !carp_iamatch(ifp))
598 		goto out;
599 #endif
600 
601 	/* Do we have an ARP cache for the sender? Create if we are target. */
602 	rt = arplookup(&isaddr, target, 0, rdomain);
603 
604 	/* Check sender against our interface addresses. */
605 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
606 	    rt->rt_ifidx == ifp->if_index && isaddr.s_addr != INADDR_ANY) {
607 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
608 		log(LOG_ERR, "duplicate IP address %s sent from ethernet "
609 		    "address %s\n", addr, ether_sprintf(ea->arp_sha));
610 		itaddr = isaddr;
611 	} else if (rt != NULL) {
612 		int error;
613 
614 		KERNEL_LOCK();
615 		error = arpcache(ifp, ea, rt);
616 		KERNEL_UNLOCK();
617 		if (error)
618 			goto out;
619 	}
620 
621 	if (op == ARPOP_REQUEST) {
622 		uint8_t *eaddr;
623 
624 		if (target) {
625 			/* We already have all info for the reply */
626 			eaddr = LLADDR(ifp->if_sadl);
627 		} else {
628 			rtfree(rt);
629 			rt = arplookup(&itaddr, 0, SIN_PROXY, rdomain);
630 			/*
631 			 * Protect from possible duplicates, only owner
632 			 * should respond
633 			 */
634 			if ((rt == NULL) || (rt->rt_ifidx != ifp->if_index))
635 				goto out;
636 			eaddr = LLADDR(satosdl(rt->rt_gateway));
637 		}
638 		arpreply(ifp, m, &itaddr, eaddr, rdomain);
639 		rtfree(rt);
640 		return;
641 	}
642 
643 out:
644 	rtfree(rt);
645 	m_freem(m);
646 }
647 
648 int
649 arpcache(struct ifnet *ifp, struct ether_arp *ea, struct rtentry *rt)
650 {
651 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
652 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
653 	struct in_addr *spa = (struct in_addr *)ea->arp_spa;
654 	char addr[INET_ADDRSTRLEN];
655 	struct ifnet *rifp;
656 	struct mbuf_list ml;
657 	struct mbuf *m;
658 	time_t uptime;
659 	unsigned int len;
660 	int changed = 0;
661 
662 	KERNEL_ASSERT_LOCKED();
663 	KASSERT(sdl != NULL);
664 
665 	/*
666 	 * This can happen if the entry has been deleted by another CPU
667 	 * after we found it.
668 	 */
669 	if (la == NULL)
670 		return (0);
671 
672 	uptime = getuptime();
673 	if (sdl->sdl_alen > 0) {
674 		if (memcmp(ea->arp_sha, LLADDR(sdl), sdl->sdl_alen)) {
675 			if (ISSET(rt->rt_flags, RTF_PERMANENT_ARP|RTF_LOCAL)) {
676 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
677 				log(LOG_WARNING, "arp: attempt to overwrite "
678 				   "permanent entry for %s by %s on %s\n", addr,
679 				   ether_sprintf(ea->arp_sha), ifp->if_xname);
680 				return (-1);
681 			} else if (rt->rt_ifidx != ifp->if_index) {
682 #if NCARP > 0
683 				if (ifp->if_type != IFT_CARP)
684 #endif
685 				{
686 					rifp = if_get(rt->rt_ifidx);
687 					if (rifp == NULL)
688 						return (-1);
689 					inet_ntop(AF_INET, spa, addr,
690 					    sizeof(addr));
691 					log(LOG_WARNING, "arp: attempt to "
692 					    "overwrite entry for %s on %s by "
693 					    "%s on %s\n", addr, rifp->if_xname,
694 					    ether_sprintf(ea->arp_sha),
695 					    ifp->if_xname);
696 					if_put(rifp);
697 				}
698 				return (-1);
699 			} else {
700 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
701 				log(LOG_INFO, "arp info overwritten for %s by "
702 				    "%s on %s\n", addr,
703 				    ether_sprintf(ea->arp_sha), ifp->if_xname);
704 				rt->rt_expire = 1;/* no longer static */
705 			}
706 			changed = 1;
707 		}
708 	} else if (!if_isconnected(ifp, rt->rt_ifidx)) {
709 		rifp = if_get(rt->rt_ifidx);
710 		if (rifp == NULL)
711 			return (-1);
712 		inet_ntop(AF_INET, spa, addr, sizeof(addr));
713 		log(LOG_WARNING, "arp: attempt to add entry for %s on %s by %s"
714 		    " on %s\n", addr, rifp->if_xname,
715 		    ether_sprintf(ea->arp_sha), ifp->if_xname);
716 		if_put(rifp);
717 		return (-1);
718 	}
719 	sdl->sdl_alen = sizeof(ea->arp_sha);
720 	memcpy(LLADDR(sdl), ea->arp_sha, sizeof(ea->arp_sha));
721 	if (rt->rt_expire)
722 		rt->rt_expire = uptime + arpt_keep;
723 	rt->rt_flags &= ~RTF_REJECT;
724 
725 	/* Notify userland that an ARP resolution has been done. */
726 	if (la->la_asked || changed) {
727 		rtm_send(rt, RTM_RESOLVE, 0, ifp->if_rdomain);
728 	}
729 
730 	la->la_asked = 0;
731 	la->la_refreshed = 0;
732 	mq_delist(&la->la_mq, &ml);
733 	len = ml_len(&ml);
734 	while ((m = ml_dequeue(&ml)) != NULL)
735 		ifp->if_output(ifp, m, rt_key(rt), rt);
736 	/* XXXSMP we discard if other CPU enqueues */
737 	if (mq_len(&la->la_mq) > 0) {
738 		/* mbuf is back in queue. Discard. */
739 		atomic_sub_int(&la_hold_total, len + mq_purge(&la->la_mq));
740 	} else
741 		atomic_sub_int(&la_hold_total, len);
742 
743 	return (0);
744 }
745 
746 void
747 arpinvalidate(struct rtentry *rt)
748 {
749 	struct llinfo_arp *la;
750 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
751 
752 	mtx_enter(&arp_mtx);
753 	la = (struct llinfo_arp *)rt->rt_llinfo;
754 	if (la == NULL) {
755 		mtx_leave(&arp_mtx);
756 		return;
757 	}
758 	atomic_sub_int(&la_hold_total, mq_purge(&la->la_mq));
759 	sdl->sdl_alen = 0;
760 	la->la_asked = 0;
761 	mtx_leave(&arp_mtx);
762 }
763 
764 /*
765  * Free an arp entry.
766  */
767 void
768 arptfree(struct rtentry *rt)
769 {
770 	struct ifnet *ifp;
771 
772 	KASSERT(!ISSET(rt->rt_flags, RTF_LOCAL));
773 	arpinvalidate(rt);
774 
775 	ifp = if_get(rt->rt_ifidx);
776 	KASSERT(ifp != NULL);
777 	if (!ISSET(rt->rt_flags, RTF_STATIC|RTF_CACHED))
778 		rtdeletemsg(rt, ifp, ifp->if_rdomain);
779 	if_put(ifp);
780 }
781 
782 /*
783  * Lookup or enter a new address in arptab.
784  */
785 struct rtentry *
786 arplookup(struct in_addr *inp, int create, int proxy, u_int tableid)
787 {
788 	struct rtentry *rt;
789 	struct sockaddr_inarp sin;
790 	int flags;
791 
792 	memset(&sin, 0, sizeof(sin));
793 	sin.sin_len = sizeof(sin);
794 	sin.sin_family = AF_INET;
795 	sin.sin_addr.s_addr = inp->s_addr;
796 	sin.sin_other = proxy ? SIN_PROXY : 0;
797 	flags = (create) ? RT_RESOLVE : 0;
798 
799 	rt = rtalloc((struct sockaddr *)&sin, flags, tableid);
800 	if (!rtisvalid(rt) || ISSET(rt->rt_flags, RTF_GATEWAY) ||
801 	    !ISSET(rt->rt_flags, RTF_LLINFO) ||
802 	    rt->rt_gateway->sa_family != AF_LINK) {
803 		rtfree(rt);
804 		return (NULL);
805 	}
806 
807 	if (proxy && !ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
808 		while ((rt = rtable_iterate(rt)) != NULL) {
809 			if (ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
810 				break;
811 			}
812 		}
813 	}
814 
815 	return (rt);
816 }
817 
818 /*
819  * Check whether we do proxy ARP for this address and we point to ourselves.
820  */
821 int
822 arpproxy(struct in_addr in, unsigned int rtableid)
823 {
824 	struct sockaddr_dl *sdl;
825 	struct rtentry *rt;
826 	struct ifnet *ifp;
827 	int found = 0;
828 
829 	rt = arplookup(&in, 0, SIN_PROXY, rtableid);
830 	if (!rtisvalid(rt)) {
831 		rtfree(rt);
832 		return (0);
833 	}
834 
835 	/* Check that arp information are correct. */
836 	sdl = satosdl(rt->rt_gateway);
837 	if (sdl->sdl_alen != ETHER_ADDR_LEN) {
838 		rtfree(rt);
839 		return (0);
840 	}
841 
842 	ifp = if_get(rt->rt_ifidx);
843 	if (ifp == NULL) {
844 		rtfree(rt);
845 		return (0);
846 	}
847 
848 	if (!memcmp(LLADDR(sdl), LLADDR(ifp->if_sadl), sdl->sdl_alen))
849 		found = 1;
850 
851 	if_put(ifp);
852 	rtfree(rt);
853 	return (found);
854 }
855 
856 /*
857  * Called from Ethernet interrupt handlers
858  * when ether packet type ETHERTYPE_REVARP
859  * is received.  Common length and type checks are done here,
860  * then the protocol-specific routine is called.
861  */
862 void
863 revarpinput(struct ifnet *ifp, struct mbuf *m)
864 {
865 	if ((m = arppullup(m)) == NULL)
866 		return;
867 	in_revarpinput(ifp, m);
868 }
869 
870 /*
871  * RARP for Internet protocols on Ethernet.
872  * Algorithm is that given in RFC 903.
873  * We are only using for bootstrap purposes to get an ip address for one of
874  * our interfaces.  Thus we support no user-interface.
875  *
876  * Since the contents of the RARP reply are specific to the interface that
877  * sent the request, this code must ensure that they are properly associated.
878  *
879  * Note: also supports ARP via RARP packets, per the RFC.
880  */
881 void
882 in_revarpinput(struct ifnet *ifp, struct mbuf *m)
883 {
884 	struct ether_arp *ar;
885 	int op;
886 
887 	ar = mtod(m, struct ether_arp *);
888 	op = ntohs(ar->arp_op);
889 	switch (op) {
890 	case ARPOP_REQUEST:
891 	case ARPOP_REPLY:	/* per RFC */
892 		niq_enqueue(&arpinq, m);
893 		return;
894 	case ARPOP_REVREPLY:
895 		break;
896 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
897 	default:
898 		goto out;
899 	}
900 #ifdef NFSCLIENT
901 	if (revarp_ifidx == 0)
902 		goto out;
903 	if (revarp_ifidx != m->m_pkthdr.ph_ifidx) /* !same interface */
904 		goto out;
905 	if (revarp_finished)
906 		goto wake;
907 	if (memcmp(ar->arp_tha, LLADDR(ifp->if_sadl), sizeof(ar->arp_tha)))
908 		goto out;
909 	memcpy(&revarp_srvip, ar->arp_spa, sizeof(revarp_srvip));
910 	memcpy(&revarp_myip, ar->arp_tpa, sizeof(revarp_myip));
911 	revarp_finished = 1;
912 wake:	/* Do wakeup every time in case it was missed. */
913 	wakeup((caddr_t)&revarp_myip);
914 #endif /* NFSCLIENT */
915 
916 out:
917 	m_freem(m);
918 }
919 
920 /*
921  * Send a RARP request for the ip address of the specified interface.
922  * The request should be RFC 903-compliant.
923  */
924 void
925 revarprequest(struct ifnet *ifp)
926 {
927 	struct sockaddr sa;
928 	struct mbuf *m;
929 	struct ether_header *eh;
930 	struct ether_arp *ea;
931 	struct arpcom *ac = (struct arpcom *)ifp;
932 
933 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
934 		return;
935 	m->m_len = sizeof(*ea);
936 	m->m_pkthdr.len = sizeof(*ea);
937 	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
938 	m->m_pkthdr.pf.prio = ifp->if_llprio;
939 	m_align(m, sizeof(*ea));
940 	ea = mtod(m, struct ether_arp *);
941 	eh = (struct ether_header *)sa.sa_data;
942 	memset(ea, 0, sizeof(*ea));
943 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
944 	eh->ether_type = htons(ETHERTYPE_REVARP);
945 	ea->arp_hrd = htons(ARPHRD_ETHER);
946 	ea->arp_pro = htons(ETHERTYPE_IP);
947 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
948 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
949 	ea->arp_op = htons(ARPOP_REVREQUEST);
950 	memcpy(eh->ether_shost, ac->ac_enaddr, sizeof(ea->arp_tha));
951 	memcpy(ea->arp_sha, ac->ac_enaddr, sizeof(ea->arp_sha));
952 	memcpy(ea->arp_tha, ac->ac_enaddr, sizeof(ea->arp_tha));
953 	sa.sa_family = pseudo_AF_HDRCMPLT;
954 	sa.sa_len = sizeof(sa);
955 	m->m_flags |= M_BCAST;
956 	ifp->if_output(ifp, m, &sa, NULL);
957 }
958 
959 #ifdef NFSCLIENT
960 /*
961  * RARP for the ip address of the specified interface, but also
962  * save the ip address of the server that sent the answer.
963  * Timeout if no response is received.
964  */
965 int
966 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
967     struct in_addr *clnt_in)
968 {
969 	int result, count = 20;
970 
971 	if (revarp_finished)
972 		return EIO;
973 
974 	revarp_ifidx = ifp->if_index;
975 	while (count--) {
976 		revarprequest(ifp);
977 		result = tsleep_nsec(&revarp_myip, PSOCK, "revarp",
978 		    MSEC_TO_NSEC(500));
979 		if (result != EWOULDBLOCK)
980 			break;
981 	}
982 	revarp_ifidx = 0;
983 	if (!revarp_finished)
984 		return ENETUNREACH;
985 
986 	memcpy(serv_in, &revarp_srvip, sizeof(*serv_in));
987 	memcpy(clnt_in, &revarp_myip, sizeof(*clnt_in));
988 	return 0;
989 }
990 
991 /* For compatibility: only saves interface address. */
992 int
993 revarpwhoami(struct in_addr *in, struct ifnet *ifp)
994 {
995 	struct in_addr server;
996 	return (revarpwhoarewe(ifp, &server, in));
997 }
998 #endif /* NFSCLIENT */
999