xref: /openbsd-src/sys/netinet/if_ether.c (revision 5c389b79544373bccfce668b646e62e7ba9802a3)
1 /*	$OpenBSD: if_ether.c,v 1.265 2023/05/12 12:40:49 bluhm 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 =
109     LIST_HEAD_INITIALIZER(arp_list);	/* [mN] list of llinfo_arp structures */
110 struct	pool arp_pool;		/* [I] pool for llinfo_arp structures */
111 int	arp_maxtries = 5;	/* [I] arp requests before set to rejected */
112 unsigned int	la_hold_total;	/* [a] packets currently in the arp queue */
113 
114 #ifdef NFSCLIENT
115 /* revarp state */
116 struct in_addr revarp_myip, revarp_srvip;
117 int revarp_finished;
118 unsigned int revarp_ifidx;
119 #endif /* NFSCLIENT */
120 
121 /*
122  * Timeout routine.  Age arp_tab entries periodically.
123  */
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 	int refresh = 0, reject = 0;
343 
344 	if (m->m_flags & M_BCAST) {	/* broadcast */
345 		memcpy(desten, etherbroadcastaddr, sizeof(etherbroadcastaddr));
346 		return (0);
347 	}
348 	if (m->m_flags & M_MCAST) {	/* multicast */
349 		ETHER_MAP_IP_MULTICAST(&satosin(dst)->sin_addr, desten);
350 		return (0);
351 	}
352 
353 	uptime = getuptime();
354 	rt = rt_getll(rt0);
355 
356 	if (ISSET(rt->rt_flags, RTF_REJECT) &&
357 	    (rt->rt_expire == 0 || rt->rt_expire > uptime)) {
358 		m_freem(m);
359 		return (rt == rt0 ? EHOSTDOWN : EHOSTUNREACH);
360 	}
361 
362 	if (!ISSET(rt->rt_flags, RTF_LLINFO)) {
363 		log(LOG_DEBUG, "%s: %s: route contains no arp information\n",
364 		    __func__, inet_ntop(AF_INET, &satosin(rt_key(rt))->sin_addr,
365 		    addr, sizeof(addr)));
366 		goto bad;
367 	}
368 
369 	sdl = satosdl(rt->rt_gateway);
370 	if (sdl->sdl_alen > 0 && sdl->sdl_alen != ETHER_ADDR_LEN) {
371 		log(LOG_DEBUG, "%s: %s: incorrect arp information\n", __func__,
372 		    inet_ntop(AF_INET, &satosin(dst)->sin_addr,
373 			addr, sizeof(addr)));
374 		goto bad;
375 	}
376 
377 
378 	/*
379 	 * Check the address family and length is valid, the address
380 	 * is resolved; otherwise, try to resolve.
381 	 */
382 	if ((rt->rt_expire == 0 || rt->rt_expire > uptime) &&
383 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
384 		memcpy(desten, LLADDR(sdl), sdl->sdl_alen);
385 
386 		/* refresh ARP entry when timeout gets close */
387 		if (rt->rt_expire != 0 &&
388 		    rt->rt_expire - arpt_keep / 8 < uptime) {
389 
390 			mtx_enter(&arp_mtx);
391 			la = (struct llinfo_arp *)rt->rt_llinfo;
392 			if (la != NULL) {
393 				if (la->la_refreshed + 30 < uptime) {
394 					la->la_refreshed = uptime;
395 					refresh = 1;
396 				}
397 			}
398 			mtx_leave(&arp_mtx);
399 		}
400 		if (refresh) {
401 			arprequest(ifp,
402 			    &satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr,
403 			    &satosin(dst)->sin_addr.s_addr,
404 			    ac->ac_enaddr);
405 		}
406 		return (0);
407 	}
408 
409 	if (ifp->if_flags & (IFF_NOARP|IFF_STATICARP))
410 		goto bad;
411 
412 	mtx_enter(&arp_mtx);
413 	la = (struct llinfo_arp *)rt->rt_llinfo;
414 	if (la == NULL) {
415 		mtx_leave(&arp_mtx);
416 		goto bad;
417 	}
418 
419 	/*
420 	 * There is an arptab entry, but no ethernet address
421 	 * response yet. Insert mbuf in hold queue if below limit.
422 	 * If above the limit free the queue without queuing the new packet.
423 	 */
424 	if (atomic_inc_int_nv(&la_hold_total) <= LA_HOLD_TOTAL) {
425 		if (mq_push(&la->la_mq, m) != 0)
426 			atomic_dec_int(&la_hold_total);
427 	} else {
428 		atomic_sub_int(&la_hold_total, mq_purge(&la->la_mq) + 1);
429 		m_freem(m);
430 	}
431 
432 	/*
433 	 * Re-send the ARP request when appropriate.
434 	 */
435 #ifdef	DIAGNOSTIC
436 	if (rt->rt_expire == 0) {
437 		/* This should never happen. (Should it? -gwr) */
438 		printf("%s: unresolved and rt_expire == 0\n", __func__);
439 		/* Set expiration time to now (expired). */
440 		rt->rt_expire = uptime;
441 	}
442 #endif
443 	if (rt->rt_expire) {
444 		reject = ~RTF_REJECT;
445 		if (la->la_asked == 0 || rt->rt_expire != uptime) {
446 			rt->rt_expire = uptime;
447 			if (la->la_asked++ < arp_maxtries)
448 				refresh = 1;
449 			else {
450 				reject = RTF_REJECT;
451 				rt->rt_expire += arpt_down;
452 				la->la_asked = 0;
453 				la->la_refreshed = 0;
454 				atomic_sub_int(&la_hold_total,
455 				    mq_purge(&la->la_mq));
456 			}
457 		}
458 	}
459 	mtx_leave(&arp_mtx);
460 
461 	if (reject == RTF_REJECT && !ISSET(rt->rt_flags, RTF_REJECT)) {
462 		KERNEL_LOCK();
463 		SET(rt->rt_flags, RTF_REJECT);
464 		KERNEL_UNLOCK();
465 	}
466 	if (reject == ~RTF_REJECT && ISSET(rt->rt_flags, RTF_REJECT)) {
467 		KERNEL_LOCK();
468 		CLR(rt->rt_flags, RTF_REJECT);
469 		KERNEL_UNLOCK();
470 	}
471 	if (refresh)
472 		arprequest(ifp, &satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr,
473 		    &satosin(dst)->sin_addr.s_addr, ac->ac_enaddr);
474 	return (EAGAIN);
475 
476 bad:
477 	m_freem(m);
478 	return (EINVAL);
479 }
480 
481 struct mbuf *
482 arppullup(struct mbuf *m)
483 {
484 	struct arphdr *ar;
485 	int len;
486 
487 #ifdef DIAGNOSTIC
488 	if ((m->m_flags & M_PKTHDR) == 0)
489 		panic("arp without packet header");
490 #endif
491 
492 	len = sizeof(struct arphdr);
493 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
494 		return NULL;
495 
496 	ar = mtod(m, struct arphdr *);
497 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER ||
498 	    ntohs(ar->ar_pro) != ETHERTYPE_IP ||
499 	    ar->ar_hln != ETHER_ADDR_LEN ||
500 	    ar->ar_pln != sizeof(struct in_addr)) {
501 		m_freem(m);
502 		return NULL;
503 	}
504 
505 	len += 2 * (ar->ar_hln + ar->ar_pln);
506 	if (m->m_len < len && (m = m_pullup(m, len)) == NULL)
507 		return NULL;
508 
509 	return m;
510 }
511 
512 /*
513  * Common length and type checks are done here,
514  * then the protocol-specific routine is called.
515  */
516 void
517 arpinput(struct ifnet *ifp, struct mbuf *m)
518 {
519 	if ((m = arppullup(m)) == NULL)
520 		return;
521 	niq_enqueue(&arpinq, m);
522 }
523 
524 void
525 arpintr(void)
526 {
527 	struct mbuf_list ml;
528 	struct mbuf *m;
529 	struct ifnet *ifp;
530 
531 	niq_delist(&arpinq, &ml);
532 
533 	while ((m = ml_dequeue(&ml)) != NULL) {
534 		ifp = if_get(m->m_pkthdr.ph_ifidx);
535 
536 		if (ifp != NULL)
537 			in_arpinput(ifp, m);
538 		else
539 			m_freem(m);
540 
541 		if_put(ifp);
542 	}
543 }
544 
545 /*
546  * ARP for Internet protocols on Ethernet, RFC 826.
547  * In addition, a sanity check is performed on the sender
548  * protocol address, to catch impersonators.
549  */
550 void
551 in_arpinput(struct ifnet *ifp, struct mbuf *m)
552 {
553 	struct ether_arp *ea;
554 	struct rtentry *rt = NULL;
555 	struct sockaddr_in sin;
556 	struct in_addr isaddr, itaddr;
557 	char addr[INET_ADDRSTRLEN];
558 	int op, target = 0;
559 	unsigned int rdomain;
560 
561 	rdomain = rtable_l2(m->m_pkthdr.ph_rtableid);
562 
563 	ea = mtod(m, struct ether_arp *);
564 	op = ntohs(ea->arp_op);
565 	if ((op != ARPOP_REQUEST) && (op != ARPOP_REPLY))
566 		goto out;
567 
568 	memcpy(&itaddr, ea->arp_tpa, sizeof(itaddr));
569 	memcpy(&isaddr, ea->arp_spa, sizeof(isaddr));
570 	memset(&sin, 0, sizeof(sin));
571 	sin.sin_len = sizeof(sin);
572 	sin.sin_family = AF_INET;
573 
574 	if (ETHER_IS_MULTICAST(ea->arp_sha) &&
575 	    ETHER_IS_BROADCAST(ea->arp_sha)) {
576 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
577 		log(LOG_ERR, "arp: ether address is broadcast for IP address "
578 		    "%s!\n", addr);
579 		goto out;
580 	}
581 
582 	if (!memcmp(ea->arp_sha, LLADDR(ifp->if_sadl), sizeof(ea->arp_sha)))
583 		goto out;	/* it's from me, ignore it. */
584 
585 	/* Check target against our interface addresses. */
586 	sin.sin_addr = itaddr;
587 	rt = rtalloc(sintosa(&sin), 0, rdomain);
588 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
589 	    rt->rt_ifidx == ifp->if_index)
590 		target = 1;
591 	rtfree(rt);
592 	rt = NULL;
593 
594 #if NCARP > 0
595 	if (target && op == ARPOP_REQUEST && ifp->if_type == IFT_CARP &&
596 	    !carp_iamatch(ifp))
597 		goto out;
598 #endif
599 
600 	/* Do we have an ARP cache for the sender? Create if we are target. */
601 	rt = arplookup(&isaddr, target, 0, rdomain);
602 
603 	/* Check sender against our interface addresses. */
604 	if (rtisvalid(rt) && ISSET(rt->rt_flags, RTF_LOCAL) &&
605 	    rt->rt_ifidx == ifp->if_index && isaddr.s_addr != INADDR_ANY) {
606 		inet_ntop(AF_INET, &isaddr, addr, sizeof(addr));
607 		log(LOG_ERR, "duplicate IP address %s sent from ethernet "
608 		    "address %s\n", addr, ether_sprintf(ea->arp_sha));
609 		itaddr = isaddr;
610 	} else if (rt != NULL) {
611 		if (arpcache(ifp, ea, rt))
612 			goto out;
613 	}
614 
615 	if (op == ARPOP_REQUEST) {
616 		uint8_t *eaddr;
617 
618 		if (target) {
619 			/* We already have all info for the reply */
620 			eaddr = LLADDR(ifp->if_sadl);
621 		} else {
622 			rtfree(rt);
623 			rt = arplookup(&itaddr, 0, SIN_PROXY, rdomain);
624 			/*
625 			 * Protect from possible duplicates, only owner
626 			 * should respond
627 			 */
628 			if ((rt == NULL) || (rt->rt_ifidx != ifp->if_index))
629 				goto out;
630 			eaddr = LLADDR(satosdl(rt->rt_gateway));
631 		}
632 		arpreply(ifp, m, &itaddr, eaddr, rdomain);
633 		rtfree(rt);
634 		return;
635 	}
636 
637 out:
638 	rtfree(rt);
639 	m_freem(m);
640 }
641 
642 int
643 arpcache(struct ifnet *ifp, struct ether_arp *ea, struct rtentry *rt)
644 {
645 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
646 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
647 	struct in_addr *spa = (struct in_addr *)ea->arp_spa;
648 	char addr[INET_ADDRSTRLEN];
649 	struct ifnet *rifp;
650 	time_t uptime;
651 	int changed = 0;
652 
653 	NET_ASSERT_LOCKED_EXCLUSIVE();
654 	KASSERT(sdl != NULL);
655 
656 	/*
657 	 * This can happen if the entry has been deleted by another CPU
658 	 * after we found it.
659 	 */
660 	if (la == NULL)
661 		return (0);
662 
663 	uptime = getuptime();
664 	if (sdl->sdl_alen > 0) {
665 		if (memcmp(ea->arp_sha, LLADDR(sdl), sdl->sdl_alen)) {
666 			if (ISSET(rt->rt_flags, RTF_PERMANENT_ARP|RTF_LOCAL)) {
667 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
668 				log(LOG_WARNING, "arp: attempt to overwrite "
669 				   "permanent entry for %s by %s on %s\n", addr,
670 				   ether_sprintf(ea->arp_sha), ifp->if_xname);
671 				return (-1);
672 			} else if (rt->rt_ifidx != ifp->if_index) {
673 #if NCARP > 0
674 				if (ifp->if_type != IFT_CARP)
675 #endif
676 				{
677 					rifp = if_get(rt->rt_ifidx);
678 					if (rifp == NULL)
679 						return (-1);
680 					inet_ntop(AF_INET, spa, addr,
681 					    sizeof(addr));
682 					log(LOG_WARNING, "arp: attempt to "
683 					    "overwrite entry for %s on %s by "
684 					    "%s on %s\n", addr, rifp->if_xname,
685 					    ether_sprintf(ea->arp_sha),
686 					    ifp->if_xname);
687 					if_put(rifp);
688 				}
689 				return (-1);
690 			} else {
691 				inet_ntop(AF_INET, spa, addr, sizeof(addr));
692 				log(LOG_INFO, "arp info overwritten for %s by "
693 				    "%s on %s\n", addr,
694 				    ether_sprintf(ea->arp_sha), ifp->if_xname);
695 				rt->rt_expire = 1;/* no longer static */
696 			}
697 			changed = 1;
698 		}
699 	} else if (!if_isconnected(ifp, rt->rt_ifidx)) {
700 		rifp = if_get(rt->rt_ifidx);
701 		if (rifp == NULL)
702 			return (-1);
703 		inet_ntop(AF_INET, spa, addr, sizeof(addr));
704 		log(LOG_WARNING, "arp: attempt to add entry for %s on %s by %s"
705 		    " on %s\n", addr, rifp->if_xname,
706 		    ether_sprintf(ea->arp_sha), ifp->if_xname);
707 		if_put(rifp);
708 		return (-1);
709 	}
710 	sdl->sdl_alen = sizeof(ea->arp_sha);
711 	memcpy(LLADDR(sdl), ea->arp_sha, sizeof(ea->arp_sha));
712 	if (rt->rt_expire)
713 		rt->rt_expire = uptime + arpt_keep;
714 	rt->rt_flags &= ~RTF_REJECT;
715 
716 	/* Notify userland that an ARP resolution has been done. */
717 	if (la->la_asked || changed) {
718 		rtm_send(rt, RTM_RESOLVE, 0, ifp->if_rdomain);
719 	}
720 
721 	la->la_asked = 0;
722 	la->la_refreshed = 0;
723 	if_output_mq(ifp, &la->la_mq, &la_hold_total, rt_key(rt), rt);
724 
725 	return (0);
726 }
727 
728 void
729 arpinvalidate(struct rtentry *rt)
730 {
731 	struct llinfo_arp *la;
732 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
733 
734 	mtx_enter(&arp_mtx);
735 	la = (struct llinfo_arp *)rt->rt_llinfo;
736 	if (la == NULL) {
737 		mtx_leave(&arp_mtx);
738 		return;
739 	}
740 	atomic_sub_int(&la_hold_total, mq_purge(&la->la_mq));
741 	sdl->sdl_alen = 0;
742 	la->la_asked = 0;
743 	mtx_leave(&arp_mtx);
744 }
745 
746 /*
747  * Free an arp entry.
748  */
749 void
750 arptfree(struct rtentry *rt)
751 {
752 	struct ifnet *ifp;
753 
754 	KASSERT(!ISSET(rt->rt_flags, RTF_LOCAL));
755 	arpinvalidate(rt);
756 
757 	ifp = if_get(rt->rt_ifidx);
758 	KASSERT(ifp != NULL);
759 	if (!ISSET(rt->rt_flags, RTF_STATIC|RTF_CACHED))
760 		rtdeletemsg(rt, ifp, ifp->if_rdomain);
761 	if_put(ifp);
762 }
763 
764 /*
765  * Lookup or enter a new address in arptab.
766  */
767 struct rtentry *
768 arplookup(struct in_addr *inp, int create, int proxy, u_int tableid)
769 {
770 	struct rtentry *rt;
771 	struct sockaddr_inarp sin;
772 	int flags;
773 
774 	memset(&sin, 0, sizeof(sin));
775 	sin.sin_len = sizeof(sin);
776 	sin.sin_family = AF_INET;
777 	sin.sin_addr.s_addr = inp->s_addr;
778 	sin.sin_other = proxy ? SIN_PROXY : 0;
779 	flags = (create) ? RT_RESOLVE : 0;
780 
781 	rt = rtalloc((struct sockaddr *)&sin, flags, tableid);
782 	if (!rtisvalid(rt) || ISSET(rt->rt_flags, RTF_GATEWAY) ||
783 	    !ISSET(rt->rt_flags, RTF_LLINFO) ||
784 	    rt->rt_gateway->sa_family != AF_LINK) {
785 		rtfree(rt);
786 		return (NULL);
787 	}
788 
789 	if (proxy && !ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
790 		while ((rt = rtable_iterate(rt)) != NULL) {
791 			if (ISSET(rt->rt_flags, RTF_ANNOUNCE)) {
792 				break;
793 			}
794 		}
795 	}
796 
797 	return (rt);
798 }
799 
800 /*
801  * Check whether we do proxy ARP for this address and we point to ourselves.
802  */
803 int
804 arpproxy(struct in_addr in, unsigned int rtableid)
805 {
806 	struct sockaddr_dl *sdl;
807 	struct rtentry *rt;
808 	struct ifnet *ifp;
809 	int found = 0;
810 
811 	rt = arplookup(&in, 0, SIN_PROXY, rtableid);
812 	if (!rtisvalid(rt)) {
813 		rtfree(rt);
814 		return (0);
815 	}
816 
817 	/* Check that arp information are correct. */
818 	sdl = satosdl(rt->rt_gateway);
819 	if (sdl->sdl_alen != ETHER_ADDR_LEN) {
820 		rtfree(rt);
821 		return (0);
822 	}
823 
824 	ifp = if_get(rt->rt_ifidx);
825 	if (ifp == NULL) {
826 		rtfree(rt);
827 		return (0);
828 	}
829 
830 	if (!memcmp(LLADDR(sdl), LLADDR(ifp->if_sadl), sdl->sdl_alen))
831 		found = 1;
832 
833 	if_put(ifp);
834 	rtfree(rt);
835 	return (found);
836 }
837 
838 /*
839  * Called from Ethernet interrupt handlers
840  * when ether packet type ETHERTYPE_REVARP
841  * is received.  Common length and type checks are done here,
842  * then the protocol-specific routine is called.
843  */
844 void
845 revarpinput(struct ifnet *ifp, struct mbuf *m)
846 {
847 	if ((m = arppullup(m)) == NULL)
848 		return;
849 	in_revarpinput(ifp, m);
850 }
851 
852 /*
853  * RARP for Internet protocols on Ethernet.
854  * Algorithm is that given in RFC 903.
855  * We are only using for bootstrap purposes to get an ip address for one of
856  * our interfaces.  Thus we support no user-interface.
857  *
858  * Since the contents of the RARP reply are specific to the interface that
859  * sent the request, this code must ensure that they are properly associated.
860  *
861  * Note: also supports ARP via RARP packets, per the RFC.
862  */
863 void
864 in_revarpinput(struct ifnet *ifp, struct mbuf *m)
865 {
866 	struct ether_arp *ar;
867 	int op;
868 
869 	ar = mtod(m, struct ether_arp *);
870 	op = ntohs(ar->arp_op);
871 	switch (op) {
872 	case ARPOP_REQUEST:
873 	case ARPOP_REPLY:	/* per RFC */
874 		niq_enqueue(&arpinq, m);
875 		return;
876 	case ARPOP_REVREPLY:
877 		break;
878 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
879 	default:
880 		goto out;
881 	}
882 #ifdef NFSCLIENT
883 	if (revarp_ifidx == 0)
884 		goto out;
885 	if (revarp_ifidx != m->m_pkthdr.ph_ifidx) /* !same interface */
886 		goto out;
887 	if (revarp_finished)
888 		goto wake;
889 	if (memcmp(ar->arp_tha, LLADDR(ifp->if_sadl), sizeof(ar->arp_tha)))
890 		goto out;
891 	memcpy(&revarp_srvip, ar->arp_spa, sizeof(revarp_srvip));
892 	memcpy(&revarp_myip, ar->arp_tpa, sizeof(revarp_myip));
893 	revarp_finished = 1;
894 wake:	/* Do wakeup every time in case it was missed. */
895 	wakeup((caddr_t)&revarp_myip);
896 #endif /* NFSCLIENT */
897 
898 out:
899 	m_freem(m);
900 }
901 
902 /*
903  * Send a RARP request for the ip address of the specified interface.
904  * The request should be RFC 903-compliant.
905  */
906 void
907 revarprequest(struct ifnet *ifp)
908 {
909 	struct sockaddr sa;
910 	struct mbuf *m;
911 	struct ether_header *eh;
912 	struct ether_arp *ea;
913 	struct arpcom *ac = (struct arpcom *)ifp;
914 
915 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
916 		return;
917 	m->m_len = sizeof(*ea);
918 	m->m_pkthdr.len = sizeof(*ea);
919 	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
920 	m->m_pkthdr.pf.prio = ifp->if_llprio;
921 	m_align(m, sizeof(*ea));
922 	ea = mtod(m, struct ether_arp *);
923 	eh = (struct ether_header *)sa.sa_data;
924 	memset(ea, 0, sizeof(*ea));
925 	memcpy(eh->ether_dhost, etherbroadcastaddr, sizeof(eh->ether_dhost));
926 	eh->ether_type = htons(ETHERTYPE_REVARP);
927 	ea->arp_hrd = htons(ARPHRD_ETHER);
928 	ea->arp_pro = htons(ETHERTYPE_IP);
929 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
930 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
931 	ea->arp_op = htons(ARPOP_REVREQUEST);
932 	memcpy(eh->ether_shost, ac->ac_enaddr, sizeof(ea->arp_tha));
933 	memcpy(ea->arp_sha, ac->ac_enaddr, sizeof(ea->arp_sha));
934 	memcpy(ea->arp_tha, ac->ac_enaddr, sizeof(ea->arp_tha));
935 	sa.sa_family = pseudo_AF_HDRCMPLT;
936 	sa.sa_len = sizeof(sa);
937 	m->m_flags |= M_BCAST;
938 	ifp->if_output(ifp, m, &sa, NULL);
939 }
940 
941 #ifdef NFSCLIENT
942 /*
943  * RARP for the ip address of the specified interface, but also
944  * save the ip address of the server that sent the answer.
945  * Timeout if no response is received.
946  */
947 int
948 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
949     struct in_addr *clnt_in)
950 {
951 	int result, count = 20;
952 
953 	if (revarp_finished)
954 		return EIO;
955 
956 	revarp_ifidx = ifp->if_index;
957 	while (count--) {
958 		revarprequest(ifp);
959 		result = tsleep_nsec(&revarp_myip, PSOCK, "revarp",
960 		    MSEC_TO_NSEC(500));
961 		if (result != EWOULDBLOCK)
962 			break;
963 	}
964 	revarp_ifidx = 0;
965 	if (!revarp_finished)
966 		return ENETUNREACH;
967 
968 	memcpy(serv_in, &revarp_srvip, sizeof(*serv_in));
969 	memcpy(clnt_in, &revarp_myip, sizeof(*clnt_in));
970 	return 0;
971 }
972 
973 /* For compatibility: only saves interface address. */
974 int
975 revarpwhoami(struct in_addr *in, struct ifnet *ifp)
976 {
977 	struct in_addr server;
978 	return (revarpwhoarewe(ifp, &server, in));
979 }
980 #endif /* NFSCLIENT */
981