xref: /freebsd/sys/netinet/if_ether.c (revision 5e53a4f90f82c4345f277dd87cc9292f26e04a29)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
32  */
33 
34 /*
35  * Ethernet address resolution protocol.
36  * TODO:
37  *	add "inuse/lock" bit (or ref. count) along with valid bit
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_inet.h"
44 
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/mbuf.h>
52 #include <sys/malloc.h>
53 #include <sys/proc.h>
54 #include <sys/rmlock.h>
55 #include <sys/socket.h>
56 #include <sys/syslog.h>
57 
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/netisr.h>
63 #include <net/ethernet.h>
64 #include <net/route.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_fib.h>
69 #include <netinet/in_var.h>
70 #include <net/if_llatbl.h>
71 #include <netinet/if_ether.h>
72 #ifdef INET
73 #include <netinet/ip_carp.h>
74 #endif
75 
76 #include <security/mac/mac_framework.h>
77 
78 #define SIN(s) ((const struct sockaddr_in *)(s))
79 
80 static struct timeval arp_lastlog;
81 static int arp_curpps;
82 static int arp_maxpps = 1;
83 
84 /* Simple ARP state machine */
85 enum arp_llinfo_state {
86 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
87 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
88 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
89 	ARP_LLINFO_DELETED,	/* LLE is deleted */
90 };
91 
92 SYSCTL_DECL(_net_link_ether);
93 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
94 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
95 
96 /* timer values */
97 static VNET_DEFINE(int, arpt_keep) = (20*60);	/* once resolved, good for 20
98 						 * minutes */
99 static VNET_DEFINE(int, arp_maxtries) = 5;
100 static VNET_DEFINE(int, arp_proxyall) = 0;
101 static VNET_DEFINE(int, arpt_down) = 20;	/* keep incomplete entries for
102 						 * 20 seconds */
103 static VNET_DEFINE(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
104 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
105 VNET_PCPUSTAT_SYSINIT(arpstat);
106 
107 #ifdef VIMAGE
108 VNET_PCPUSTAT_SYSUNINIT(arpstat);
109 #endif /* VIMAGE */
110 
111 static VNET_DEFINE(int, arp_maxhold) = 1;
112 
113 #define	V_arpt_keep		VNET(arpt_keep)
114 #define	V_arpt_down		VNET(arpt_down)
115 #define	V_arpt_rexmit		VNET(arpt_rexmit)
116 #define	V_arp_maxtries		VNET(arp_maxtries)
117 #define	V_arp_proxyall		VNET(arp_proxyall)
118 #define	V_arp_maxhold		VNET(arp_maxhold)
119 
120 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
121 	&VNET_NAME(arpt_keep), 0,
122 	"ARP entry lifetime in seconds");
123 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
124 	&VNET_NAME(arp_maxtries), 0,
125 	"ARP resolution attempts before returning error");
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
127 	&VNET_NAME(arp_proxyall), 0,
128 	"Enable proxy ARP for all suitable requests");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
130 	&VNET_NAME(arpt_down), 0,
131 	"Incomplete ARP entry lifetime in seconds");
132 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
133     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
134 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
135 	&VNET_NAME(arp_maxhold), 0,
136 	"Number of packets to hold per ARP entry");
137 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
138 	CTLFLAG_RW, &arp_maxpps, 0,
139 	"Maximum number of remotely triggered ARP messages that can be "
140 	"logged per second");
141 
142 /*
143  * Due to the exponential backoff algorithm used for the interval between GARP
144  * retransmissions, the maximum number of retransmissions is limited for
145  * sanity. This limit corresponds to a maximum interval between retransmissions
146  * of 2^16 seconds ~= 18 hours.
147  *
148  * Making this limit more dynamic is more complicated than worthwhile,
149  * especially since sending out GARPs spaced days apart would be of little
150  * use. A maximum dynamic limit would look something like:
151  *
152  * const int max = fls(INT_MAX / hz) - 1;
153  */
154 #define MAX_GARP_RETRANSMITS 16
155 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
156 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
157 
158 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
159     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
160     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
161     "Number of times to retransmit GARP packets;"
162     " 0 to disable, maximum of 16");
163 
164 #define	ARP_LOG(pri, ...)	do {					\
165 	if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
166 		log((pri), "arp: " __VA_ARGS__);			\
167 } while (0)
168 
169 
170 static void	arpintr(struct mbuf *);
171 static void	arptimer(void *);
172 #ifdef INET
173 static void	in_arpinput(struct mbuf *);
174 #endif
175 
176 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
177     struct ifnet *ifp, int bridged, struct llentry *la);
178 static void arp_mark_lle_reachable(struct llentry *la);
179 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
180 
181 static eventhandler_tag iflladdr_tag;
182 
183 static const struct netisr_handler arp_nh = {
184 	.nh_name = "arp",
185 	.nh_handler = arpintr,
186 	.nh_proto = NETISR_ARP,
187 	.nh_policy = NETISR_POLICY_SOURCE,
188 };
189 
190 /*
191  * Timeout routine.  Age arp_tab entries periodically.
192  */
193 static void
194 arptimer(void *arg)
195 {
196 	struct llentry *lle = (struct llentry *)arg;
197 	struct ifnet *ifp;
198 	int r_skip_req;
199 
200 	if (lle->la_flags & LLE_STATIC) {
201 		return;
202 	}
203 	LLE_WLOCK(lle);
204 	if (callout_pending(&lle->lle_timer)) {
205 		/*
206 		 * Here we are a bit odd here in the treatment of
207 		 * active/pending. If the pending bit is set, it got
208 		 * rescheduled before I ran. The active
209 		 * bit we ignore, since if it was stopped
210 		 * in ll_tablefree() and was currently running
211 		 * it would have return 0 so the code would
212 		 * not have deleted it since the callout could
213 		 * not be stopped so we want to go through
214 		 * with the delete here now. If the callout
215 		 * was restarted, the pending bit will be back on and
216 		 * we just want to bail since the callout_reset would
217 		 * return 1 and our reference would have been removed
218 		 * by arpresolve() below.
219 		 */
220 		LLE_WUNLOCK(lle);
221  		return;
222  	}
223 	ifp = lle->lle_tbl->llt_ifp;
224 	CURVNET_SET(ifp->if_vnet);
225 
226 	switch (lle->ln_state) {
227 	case ARP_LLINFO_REACHABLE:
228 
229 		/*
230 		 * Expiration time is approaching.
231 		 * Let's try to refresh entry if it is still
232 		 * in use.
233 		 *
234 		 * Set r_skip_req to get feedback from
235 		 * fast path. Change state and re-schedule
236 		 * ourselves.
237 		 */
238 		LLE_REQ_LOCK(lle);
239 		lle->r_skip_req = 1;
240 		LLE_REQ_UNLOCK(lle);
241 		lle->ln_state = ARP_LLINFO_VERIFY;
242 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
243 		LLE_WUNLOCK(lle);
244 		CURVNET_RESTORE();
245 		return;
246 	case ARP_LLINFO_VERIFY:
247 		LLE_REQ_LOCK(lle);
248 		r_skip_req = lle->r_skip_req;
249 		LLE_REQ_UNLOCK(lle);
250 
251 		if (r_skip_req == 0 && lle->la_preempt > 0) {
252 			/* Entry was used, issue refresh request */
253 			struct in_addr dst;
254 			dst = lle->r_l3addr.addr4;
255 			lle->la_preempt--;
256 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
257 			LLE_WUNLOCK(lle);
258 			arprequest(ifp, NULL, &dst, NULL);
259 			CURVNET_RESTORE();
260 			return;
261 		}
262 		/* Nothing happened. Reschedule if not too late */
263 		if (lle->la_expire > time_uptime) {
264 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
265 			LLE_WUNLOCK(lle);
266 			CURVNET_RESTORE();
267 			return;
268 		}
269 		break;
270 	case ARP_LLINFO_INCOMPLETE:
271 	case ARP_LLINFO_DELETED:
272 		break;
273 	}
274 
275 	if ((lle->la_flags & LLE_DELETED) == 0) {
276 		int evt;
277 
278 		if (lle->la_flags & LLE_VALID)
279 			evt = LLENTRY_EXPIRED;
280 		else
281 			evt = LLENTRY_TIMEDOUT;
282 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
283 	}
284 
285 	callout_stop(&lle->lle_timer);
286 
287 	/* XXX: LOR avoidance. We still have ref on lle. */
288 	LLE_WUNLOCK(lle);
289 	IF_AFDATA_LOCK(ifp);
290 	LLE_WLOCK(lle);
291 
292 	/* Guard against race with other llentry_free(). */
293 	if (lle->la_flags & LLE_LINKED) {
294 		LLE_REMREF(lle);
295 		lltable_unlink_entry(lle->lle_tbl, lle);
296 	}
297 	IF_AFDATA_UNLOCK(ifp);
298 
299 	size_t pkts_dropped = llentry_free(lle);
300 
301 	ARPSTAT_ADD(dropped, pkts_dropped);
302 	ARPSTAT_INC(timeouts);
303 
304 	CURVNET_RESTORE();
305 }
306 
307 /*
308  * Stores link-layer header for @ifp in format suitable for if_output()
309  * into buffer @buf. Resulting header length is stored in @bufsize.
310  *
311  * Returns 0 on success.
312  */
313 static int
314 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
315     size_t *bufsize)
316 {
317 	struct if_encap_req ereq;
318 	int error;
319 
320 	bzero(buf, *bufsize);
321 	bzero(&ereq, sizeof(ereq));
322 	ereq.buf = buf;
323 	ereq.bufsize = *bufsize;
324 	ereq.rtype = IFENCAP_LL;
325 	ereq.family = AF_ARP;
326 	ereq.lladdr = ar_tha(ah);
327 	ereq.hdata = (u_char *)ah;
328 	if (bcast)
329 		ereq.flags = IFENCAP_FLAG_BROADCAST;
330 	error = ifp->if_requestencap(ifp, &ereq);
331 	if (error == 0)
332 		*bufsize = ereq.bufsize;
333 
334 	return (error);
335 }
336 
337 
338 /*
339  * Broadcast an ARP request. Caller specifies:
340  *	- arp header source ip address
341  *	- arp header target ip address
342  *	- arp header source ethernet address
343  */
344 void
345 arprequest(struct ifnet *ifp, const struct in_addr *sip,
346     const struct in_addr *tip, u_char *enaddr)
347 {
348 	struct mbuf *m;
349 	struct arphdr *ah;
350 	struct sockaddr sa;
351 	u_char *carpaddr = NULL;
352 	uint8_t linkhdr[LLE_MAX_LINKHDR];
353 	size_t linkhdrsize;
354 	struct route ro;
355 	int error;
356 
357 	if (sip == NULL) {
358 		/*
359 		 * The caller did not supply a source address, try to find
360 		 * a compatible one among those assigned to this interface.
361 		 */
362 		struct ifaddr *ifa;
363 
364 		IF_ADDR_RLOCK(ifp);
365 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
366 			if (ifa->ifa_addr->sa_family != AF_INET)
367 				continue;
368 
369 			if (ifa->ifa_carp) {
370 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
371 					continue;
372 				sip = &IA_SIN(ifa)->sin_addr;
373 			} else {
374 				carpaddr = NULL;
375 				sip = &IA_SIN(ifa)->sin_addr;
376 			}
377 
378 			if (0 == ((sip->s_addr ^ tip->s_addr) &
379 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
380 				break;  /* found it. */
381 		}
382 		IF_ADDR_RUNLOCK(ifp);
383 		if (sip == NULL) {
384 			printf("%s: cannot find matching address\n", __func__);
385 			return;
386 		}
387 	}
388 	if (enaddr == NULL)
389 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
390 
391 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
392 		return;
393 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
394 		2 * ifp->if_addrlen;
395 	m->m_pkthdr.len = m->m_len;
396 	M_ALIGN(m, m->m_len);
397 	ah = mtod(m, struct arphdr *);
398 	bzero((caddr_t)ah, m->m_len);
399 #ifdef MAC
400 	mac_netinet_arp_send(ifp, m);
401 #endif
402 	ah->ar_pro = htons(ETHERTYPE_IP);
403 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
404 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
405 	ah->ar_op = htons(ARPOP_REQUEST);
406 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
407 	bcopy(sip, ar_spa(ah), ah->ar_pln);
408 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
409 	sa.sa_family = AF_ARP;
410 	sa.sa_len = 2;
411 
412 	/* Calculate link header for sending frame */
413 	bzero(&ro, sizeof(ro));
414 	linkhdrsize = sizeof(linkhdr);
415 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
416 	if (error != 0 && error != EAFNOSUPPORT) {
417 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
418 		    if_name(ifp), error);
419 		return;
420 	}
421 
422 	ro.ro_prepend = linkhdr;
423 	ro.ro_plen = linkhdrsize;
424 	ro.ro_flags = 0;
425 
426 	m->m_flags |= M_BCAST;
427 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
428 	(*ifp->if_output)(ifp, m, &sa, &ro);
429 	ARPSTAT_INC(txrequests);
430 }
431 
432 
433 /*
434  * Resolve an IP address into an ethernet address - heavy version.
435  * Used internally by arpresolve().
436  * We have already checked than  we can't use existing lle without
437  * modification so we have to acquire LLE_EXCLUSIVE lle lock.
438  *
439  * On success, desten and flags are filled in and the function returns 0;
440  * If the packet must be held pending resolution, we return EWOULDBLOCK
441  * On other errors, we return the corresponding error code.
442  * Note that m_freem() handles NULL.
443  */
444 static int
445 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
446 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
447 	struct llentry **plle)
448 {
449 	struct llentry *la = NULL, *la_tmp;
450 	struct mbuf *curr = NULL;
451 	struct mbuf *next = NULL;
452 	int error, renew;
453 	char *lladdr;
454 	int ll_len;
455 
456 	if (pflags != NULL)
457 		*pflags = 0;
458 	if (plle != NULL)
459 		*plle = NULL;
460 
461 	if ((flags & LLE_CREATE) == 0) {
462 		IF_AFDATA_RLOCK(ifp);
463 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
464 		IF_AFDATA_RUNLOCK(ifp);
465 	}
466 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
467 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
468 		if (la == NULL) {
469 			char addrbuf[INET_ADDRSTRLEN];
470 
471 			log(LOG_DEBUG,
472 			    "arpresolve: can't allocate llinfo for %s on %s\n",
473 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
474 			    if_name(ifp));
475 			m_freem(m);
476 			return (EINVAL);
477 		}
478 
479 		IF_AFDATA_WLOCK(ifp);
480 		LLE_WLOCK(la);
481 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
482 		/* Prefer ANY existing lle over newly-created one */
483 		if (la_tmp == NULL)
484 			lltable_link_entry(LLTABLE(ifp), la);
485 		IF_AFDATA_WUNLOCK(ifp);
486 		if (la_tmp != NULL) {
487 			lltable_free_entry(LLTABLE(ifp), la);
488 			la = la_tmp;
489 		}
490 	}
491 	if (la == NULL) {
492 		m_freem(m);
493 		return (EINVAL);
494 	}
495 
496 	if ((la->la_flags & LLE_VALID) &&
497 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
498 		if (flags & LLE_ADDRONLY) {
499 			lladdr = la->ll_addr;
500 			ll_len = ifp->if_addrlen;
501 		} else {
502 			lladdr = la->r_linkdata;
503 			ll_len = la->r_hdrlen;
504 		}
505 		bcopy(lladdr, desten, ll_len);
506 
507 		/* Check if we have feedback request from arptimer() */
508 		if (la->r_skip_req != 0) {
509 			LLE_REQ_LOCK(la);
510 			la->r_skip_req = 0; /* Notify that entry was used */
511 			LLE_REQ_UNLOCK(la);
512 		}
513 		if (pflags != NULL)
514 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
515 		if (plle) {
516 			LLE_ADDREF(la);
517 			*plle = la;
518 		}
519 		LLE_WUNLOCK(la);
520 		return (0);
521 	}
522 
523 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
524 	/*
525 	 * There is an arptab entry, but no ethernet address
526 	 * response yet.  Add the mbuf to the list, dropping
527 	 * the oldest packet if we have exceeded the system
528 	 * setting.
529 	 */
530 	if (m != NULL) {
531 		if (la->la_numheld >= V_arp_maxhold) {
532 			if (la->la_hold != NULL) {
533 				next = la->la_hold->m_nextpkt;
534 				m_freem(la->la_hold);
535 				la->la_hold = next;
536 				la->la_numheld--;
537 				ARPSTAT_INC(dropped);
538 			}
539 		}
540 		if (la->la_hold != NULL) {
541 			curr = la->la_hold;
542 			while (curr->m_nextpkt != NULL)
543 				curr = curr->m_nextpkt;
544 			curr->m_nextpkt = m;
545 		} else
546 			la->la_hold = m;
547 		la->la_numheld++;
548 	}
549 	/*
550 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
551 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
552 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
553 	 * ARP request, but not faster than one request per second.
554 	 */
555 	if (la->la_asked < V_arp_maxtries)
556 		error = EWOULDBLOCK;	/* First request. */
557 	else
558 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
559 
560 	if (renew) {
561 		int canceled;
562 
563 		LLE_ADDREF(la);
564 		la->la_expire = time_uptime;
565 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
566 		    arptimer, la);
567 		if (canceled)
568 			LLE_REMREF(la);
569 		la->la_asked++;
570 		LLE_WUNLOCK(la);
571 		arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
572 		return (error);
573 	}
574 
575 	LLE_WUNLOCK(la);
576 	return (error);
577 }
578 
579 /*
580  * Resolve an IP address into an ethernet address.
581  */
582 int
583 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
584     char *desten, uint32_t *pflags, struct llentry **plle)
585 {
586 	int error;
587 
588 	flags |= LLE_ADDRONLY;
589 	error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
590 	return (error);
591 }
592 
593 
594 /*
595  * Lookups link header based on an IP address.
596  * On input:
597  *    ifp is the interface we use
598  *    is_gw != 0 if @dst represents gateway to some destination
599  *    m is the mbuf. May be NULL if we don't have a packet.
600  *    dst is the next hop,
601  *    desten is the storage to put LL header.
602  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
603  *
604  * On success, full/partial link header and flags are filled in and
605  * the function returns 0.
606  * If the packet must be held pending resolution, we return EWOULDBLOCK
607  * On other errors, we return the corresponding error code.
608  * Note that m_freem() handles NULL.
609  */
610 int
611 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
612 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
613 	struct llentry **plle)
614 {
615 	struct llentry *la = NULL;
616 
617 	if (pflags != NULL)
618 		*pflags = 0;
619 	if (plle != NULL)
620 		*plle = NULL;
621 
622 	if (m != NULL) {
623 		if (m->m_flags & M_BCAST) {
624 			/* broadcast */
625 			(void)memcpy(desten,
626 			    ifp->if_broadcastaddr, ifp->if_addrlen);
627 			return (0);
628 		}
629 		if (m->m_flags & M_MCAST) {
630 			/* multicast */
631 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
632 			return (0);
633 		}
634 	}
635 
636 	IF_AFDATA_RLOCK(ifp);
637 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
638 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
639 		/* Entry found, let's copy lle info */
640 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
641 		if (pflags != NULL)
642 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
643 		/* Check if we have feedback request from arptimer() */
644 		if (la->r_skip_req != 0) {
645 			LLE_REQ_LOCK(la);
646 			la->r_skip_req = 0; /* Notify that entry was used */
647 			LLE_REQ_UNLOCK(la);
648 		}
649 		if (plle) {
650 			LLE_ADDREF(la);
651 			*plle = la;
652 			LLE_WUNLOCK(la);
653 		}
654 		IF_AFDATA_RUNLOCK(ifp);
655 		return (0);
656 	}
657 	if (plle && la)
658 		LLE_WUNLOCK(la);
659 	IF_AFDATA_RUNLOCK(ifp);
660 
661 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
662 	    desten, pflags, plle));
663 }
664 
665 /*
666  * Common length and type checks are done here,
667  * then the protocol-specific routine is called.
668  */
669 static void
670 arpintr(struct mbuf *m)
671 {
672 	struct arphdr *ar;
673 	struct ifnet *ifp;
674 	char *layer;
675 	int hlen;
676 
677 	ifp = m->m_pkthdr.rcvif;
678 
679 	if (m->m_len < sizeof(struct arphdr) &&
680 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
681 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
682 		    if_name(ifp));
683 		return;
684 	}
685 	ar = mtod(m, struct arphdr *);
686 
687 	/* Check if length is sufficient */
688 	if (m->m_len <  arphdr_len(ar)) {
689 		m = m_pullup(m, arphdr_len(ar));
690 		if (m == NULL) {
691 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
692 			    if_name(ifp));
693 			return;
694 		}
695 		ar = mtod(m, struct arphdr *);
696 	}
697 
698 	hlen = 0;
699 	layer = "";
700 	switch (ntohs(ar->ar_hrd)) {
701 	case ARPHRD_ETHER:
702 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
703 		layer = "ethernet";
704 		break;
705 	case ARPHRD_IEEE802:
706 		hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
707 		layer = "fddi";
708 		break;
709 	case ARPHRD_ARCNET:
710 		hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
711 		layer = "arcnet";
712 		break;
713 	case ARPHRD_INFINIBAND:
714 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
715 		layer = "infiniband";
716 		break;
717 	case ARPHRD_IEEE1394:
718 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
719 		layer = "firewire";
720 
721 		/*
722 		 * Restrict too long hardware addresses.
723 		 * Currently we are capable of handling 20-byte
724 		 * addresses ( sizeof(lle->ll_addr) )
725 		 */
726 		if (ar->ar_hln >= 20)
727 			hlen = 16;
728 		break;
729 	default:
730 		ARP_LOG(LOG_NOTICE,
731 		    "packet with unknown hardware format 0x%02d received on "
732 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
733 		m_freem(m);
734 		return;
735 	}
736 
737 	if (hlen != 0 && hlen != ar->ar_hln) {
738 		ARP_LOG(LOG_NOTICE,
739 		    "packet with invalid %s address length %d received on %s\n",
740 		    layer, ar->ar_hln, if_name(ifp));
741 		m_freem(m);
742 		return;
743 	}
744 
745 	ARPSTAT_INC(received);
746 	switch (ntohs(ar->ar_pro)) {
747 #ifdef INET
748 	case ETHERTYPE_IP:
749 		in_arpinput(m);
750 		return;
751 #endif
752 	}
753 	m_freem(m);
754 }
755 
756 #ifdef INET
757 /*
758  * ARP for Internet protocols on 10 Mb/s Ethernet.
759  * Algorithm is that given in RFC 826.
760  * In addition, a sanity check is performed on the sender
761  * protocol address, to catch impersonators.
762  * We no longer handle negotiations for use of trailer protocol:
763  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
764  * along with IP replies if we wanted trailers sent to us,
765  * and also sent them in response to IP replies.
766  * This allowed either end to announce the desire to receive
767  * trailer packets.
768  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
769  * but formerly didn't normally send requests.
770  */
771 static int log_arp_wrong_iface = 1;
772 static int log_arp_movements = 1;
773 static int log_arp_permanent_modify = 1;
774 static int allow_multicast = 0;
775 
776 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
777 	&log_arp_wrong_iface, 0,
778 	"log arp packets arriving on the wrong interface");
779 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
780 	&log_arp_movements, 0,
781 	"log arp replies from MACs different than the one in the cache");
782 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
783 	&log_arp_permanent_modify, 0,
784 	"log arp replies from MACs different than the one in the permanent arp entry");
785 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
786 	&allow_multicast, 0, "accept multicast addresses");
787 
788 static void
789 in_arpinput(struct mbuf *m)
790 {
791 	struct rm_priotracker in_ifa_tracker;
792 	struct arphdr *ah;
793 	struct ifnet *ifp = m->m_pkthdr.rcvif;
794 	struct llentry *la = NULL, *la_tmp;
795 	struct ifaddr *ifa;
796 	struct in_ifaddr *ia;
797 	struct sockaddr sa;
798 	struct in_addr isaddr, itaddr, myaddr;
799 	u_int8_t *enaddr = NULL;
800 	int op;
801 	int bridged = 0, is_bridge = 0;
802 	int carped;
803 	struct sockaddr_in sin;
804 	struct sockaddr *dst;
805 	struct nhop4_basic nh4;
806 	uint8_t linkhdr[LLE_MAX_LINKHDR];
807 	struct route ro;
808 	size_t linkhdrsize;
809 	int lladdr_off;
810 	int error;
811 	char addrbuf[INET_ADDRSTRLEN];
812 
813 	sin.sin_len = sizeof(struct sockaddr_in);
814 	sin.sin_family = AF_INET;
815 	sin.sin_addr.s_addr = 0;
816 
817 	if (ifp->if_bridge)
818 		bridged = 1;
819 	if (ifp->if_type == IFT_BRIDGE)
820 		is_bridge = 1;
821 
822 	/*
823 	 * We already have checked that mbuf contains enough contiguous data
824 	 * to hold entire arp message according to the arp header.
825 	 */
826 	ah = mtod(m, struct arphdr *);
827 
828 	/*
829 	 * ARP is only for IPv4 so we can reject packets with
830 	 * a protocol length not equal to an IPv4 address.
831 	 */
832 	if (ah->ar_pln != sizeof(struct in_addr)) {
833 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
834 		    sizeof(struct in_addr));
835 		goto drop;
836 	}
837 
838 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
839 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
840 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
841 		goto drop;
842 	}
843 
844 	op = ntohs(ah->ar_op);
845 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
846 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
847 
848 	if (op == ARPOP_REPLY)
849 		ARPSTAT_INC(rxreplies);
850 
851 	/*
852 	 * For a bridge, we want to check the address irrespective
853 	 * of the receive interface. (This will change slightly
854 	 * when we have clusters of interfaces).
855 	 */
856 	IN_IFADDR_RLOCK(&in_ifa_tracker);
857 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
858 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
859 		    ia->ia_ifp == ifp) &&
860 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
861 		    (ia->ia_ifa.ifa_carp == NULL ||
862 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
863 			ifa_ref(&ia->ia_ifa);
864 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
865 			goto match;
866 		}
867 	}
868 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
869 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
870 		    ia->ia_ifp == ifp) &&
871 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
872 			ifa_ref(&ia->ia_ifa);
873 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
874 			goto match;
875 		}
876 
877 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
878   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
879   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
880   addr == ia->ia_addr.sin_addr.s_addr)
881 	/*
882 	 * Check the case when bridge shares its MAC address with
883 	 * some of its children, so packets are claimed by bridge
884 	 * itself (bridge_input() does it first), but they are really
885 	 * meant to be destined to the bridge member.
886 	 */
887 	if (is_bridge) {
888 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
889 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
890 				ifa_ref(&ia->ia_ifa);
891 				ifp = ia->ia_ifp;
892 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
893 				goto match;
894 			}
895 		}
896 	}
897 #undef BDG_MEMBER_MATCHES_ARP
898 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
899 
900 	/*
901 	 * No match, use the first inet address on the receive interface
902 	 * as a dummy address for the rest of the function.
903 	 */
904 	IF_ADDR_RLOCK(ifp);
905 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
906 		if (ifa->ifa_addr->sa_family == AF_INET &&
907 		    (ifa->ifa_carp == NULL ||
908 		    (*carp_iamatch_p)(ifa, &enaddr))) {
909 			ia = ifatoia(ifa);
910 			ifa_ref(ifa);
911 			IF_ADDR_RUNLOCK(ifp);
912 			goto match;
913 		}
914 	IF_ADDR_RUNLOCK(ifp);
915 
916 	/*
917 	 * If bridging, fall back to using any inet address.
918 	 */
919 	IN_IFADDR_RLOCK(&in_ifa_tracker);
920 	if (!bridged || (ia = TAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
921 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
922 		goto drop;
923 	}
924 	ifa_ref(&ia->ia_ifa);
925 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
926 match:
927 	if (!enaddr)
928 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
929 	carped = (ia->ia_ifa.ifa_carp != NULL);
930 	myaddr = ia->ia_addr.sin_addr;
931 	ifa_free(&ia->ia_ifa);
932 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
933 		goto drop;	/* it's from me, ignore it. */
934 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
935 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
936 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
937 		goto drop;
938 	}
939 
940 	if (ifp->if_addrlen != ah->ar_hln) {
941 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
942 		    "i/f %d (ignored)\n", ifp->if_addrlen,
943 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
944 		    ifp->if_addrlen);
945 		goto drop;
946 	}
947 
948 	/*
949 	 * Warn if another host is using the same IP address, but only if the
950 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
951 	 * case we suppress the warning to avoid false positive complaints of
952 	 * potential misconfiguration.
953 	 */
954 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
955 	    myaddr.s_addr != 0) {
956 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
957 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
958 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
959 		itaddr = myaddr;
960 		ARPSTAT_INC(dupips);
961 		goto reply;
962 	}
963 	if (ifp->if_flags & IFF_STATICARP)
964 		goto reply;
965 
966 	bzero(&sin, sizeof(sin));
967 	sin.sin_len = sizeof(struct sockaddr_in);
968 	sin.sin_family = AF_INET;
969 	sin.sin_addr = isaddr;
970 	dst = (struct sockaddr *)&sin;
971 	IF_AFDATA_RLOCK(ifp);
972 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
973 	IF_AFDATA_RUNLOCK(ifp);
974 	if (la != NULL)
975 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
976 	else if (itaddr.s_addr == myaddr.s_addr) {
977 		/*
978 		 * Request/reply to our address, but no lle exists yet.
979 		 * Calculate full link prepend to use in lle.
980 		 */
981 		linkhdrsize = sizeof(linkhdr);
982 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
983 		    &linkhdrsize, &lladdr_off) != 0)
984 			goto reply;
985 
986 		/* Allocate new entry */
987 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
988 		if (la == NULL) {
989 
990 			/*
991 			 * lle creation may fail if source address belongs
992 			 * to non-directly connected subnet. However, we
993 			 * will try to answer the request instead of dropping
994 			 * frame.
995 			 */
996 			goto reply;
997 		}
998 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
999 		    lladdr_off);
1000 
1001 		IF_AFDATA_WLOCK(ifp);
1002 		LLE_WLOCK(la);
1003 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1004 
1005 		/*
1006 		 * Check if lle still does not exists.
1007 		 * If it does, that means that we either
1008 		 * 1) have configured it explicitly, via
1009 		 * 1a) 'arp -s' static entry or
1010 		 * 1b) interface address static record
1011 		 * or
1012 		 * 2) it was the result of sending first packet to-host
1013 		 * or
1014 		 * 3) it was another arp reply packet we handled in
1015 		 * different thread.
1016 		 *
1017 		 * In all cases except 3) we definitely need to prefer
1018 		 * existing lle. For the sake of simplicity, prefer any
1019 		 * existing lle over newly-create one.
1020 		 */
1021 		if (la_tmp == NULL)
1022 			lltable_link_entry(LLTABLE(ifp), la);
1023 		IF_AFDATA_WUNLOCK(ifp);
1024 
1025 		if (la_tmp == NULL) {
1026 			arp_mark_lle_reachable(la);
1027 			LLE_WUNLOCK(la);
1028 		} else {
1029 			/* Free newly-create entry and handle packet */
1030 			lltable_free_entry(LLTABLE(ifp), la);
1031 			la = la_tmp;
1032 			la_tmp = NULL;
1033 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1034 			/* arp_check_update_lle() returns @la unlocked */
1035 		}
1036 		la = NULL;
1037 	}
1038 reply:
1039 	if (op != ARPOP_REQUEST)
1040 		goto drop;
1041 	ARPSTAT_INC(rxrequests);
1042 
1043 	if (itaddr.s_addr == myaddr.s_addr) {
1044 		/* Shortcut.. the receiving interface is the target. */
1045 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1046 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1047 	} else {
1048 		struct llentry *lle = NULL;
1049 
1050 		sin.sin_addr = itaddr;
1051 		IF_AFDATA_RLOCK(ifp);
1052 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1053 		IF_AFDATA_RUNLOCK(ifp);
1054 
1055 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1056 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1057 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1058 			LLE_RUNLOCK(lle);
1059 		} else {
1060 
1061 			if (lle != NULL)
1062 				LLE_RUNLOCK(lle);
1063 
1064 			if (!V_arp_proxyall)
1065 				goto drop;
1066 
1067 			/* XXX MRT use table 0 for arp reply  */
1068 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1069 				goto drop;
1070 
1071 			/*
1072 			 * Don't send proxies for nodes on the same interface
1073 			 * as this one came out of, or we'll get into a fight
1074 			 * over who claims what Ether address.
1075 			 */
1076 			if (nh4.nh_ifp == ifp)
1077 				goto drop;
1078 
1079 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1080 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1081 
1082 			/*
1083 			 * Also check that the node which sent the ARP packet
1084 			 * is on the interface we expect it to be on. This
1085 			 * avoids ARP chaos if an interface is connected to the
1086 			 * wrong network.
1087 			 */
1088 
1089 			/* XXX MRT use table 0 for arp checks */
1090 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1091 				goto drop;
1092 			if (nh4.nh_ifp != ifp) {
1093 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1094 				    " from %s via %s\n",
1095 				    inet_ntoa_r(isaddr, addrbuf),
1096 				    ifp->if_xname);
1097 				goto drop;
1098 			}
1099 
1100 #ifdef DEBUG_PROXY
1101 			printf("arp: proxying for %s\n",
1102 			    inet_ntoa_r(itaddr, addrbuf));
1103 #endif
1104 		}
1105 	}
1106 
1107 	if (itaddr.s_addr == myaddr.s_addr &&
1108 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1109 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1110 #ifdef DEBUG_LINKLOCAL
1111 		printf("arp: sending reply for link-local addr %s\n",
1112 		    inet_ntoa_r(itaddr, addrbuf));
1113 #endif
1114 		m->m_flags |= M_BCAST;
1115 		m->m_flags &= ~M_MCAST;
1116 	} else {
1117 		/* default behaviour; never reply by broadcast. */
1118 		m->m_flags &= ~(M_BCAST|M_MCAST);
1119 	}
1120 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1121 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1122 	ah->ar_op = htons(ARPOP_REPLY);
1123 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1124 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1125 	m->m_pkthdr.len = m->m_len;
1126 	m->m_pkthdr.rcvif = NULL;
1127 	sa.sa_family = AF_ARP;
1128 	sa.sa_len = 2;
1129 
1130 	/* Calculate link header for sending frame */
1131 	bzero(&ro, sizeof(ro));
1132 	linkhdrsize = sizeof(linkhdr);
1133 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1134 
1135 	/*
1136 	 * arp_fillheader() may fail due to lack of support inside encap request
1137 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1138 	 * by if_output().
1139 	 */
1140 	if (error != 0 && error != EAFNOSUPPORT) {
1141 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1142 		    if_name(ifp), error);
1143 		return;
1144 	}
1145 
1146 	ro.ro_prepend = linkhdr;
1147 	ro.ro_plen = linkhdrsize;
1148 	ro.ro_flags = 0;
1149 
1150 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1151 	(*ifp->if_output)(ifp, m, &sa, &ro);
1152 	ARPSTAT_INC(txreplies);
1153 	return;
1154 
1155 drop:
1156 	m_freem(m);
1157 }
1158 #endif
1159 
1160 /*
1161  * Checks received arp data against existing @la.
1162  * Updates lle state/performs notification if necessary.
1163  */
1164 static void
1165 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1166     int bridged, struct llentry *la)
1167 {
1168 	struct sockaddr sa;
1169 	struct mbuf *m_hold, *m_hold_next;
1170 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1171 	size_t linkhdrsize;
1172 	int lladdr_off;
1173 	char addrbuf[INET_ADDRSTRLEN];
1174 
1175 	LLE_WLOCK_ASSERT(la);
1176 
1177 	/* the following is not an error when doing bridging */
1178 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1179 		if (log_arp_wrong_iface)
1180 			ARP_LOG(LOG_WARNING, "%s is on %s "
1181 			    "but got reply from %*D on %s\n",
1182 			    inet_ntoa_r(isaddr, addrbuf),
1183 			    la->lle_tbl->llt_ifp->if_xname,
1184 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1185 			    ifp->if_xname);
1186 		LLE_WUNLOCK(la);
1187 		return;
1188 	}
1189 	if ((la->la_flags & LLE_VALID) &&
1190 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1191 		if (la->la_flags & LLE_STATIC) {
1192 			LLE_WUNLOCK(la);
1193 			if (log_arp_permanent_modify)
1194 				ARP_LOG(LOG_ERR,
1195 				    "%*D attempts to modify "
1196 				    "permanent entry for %s on %s\n",
1197 				    ifp->if_addrlen,
1198 				    (u_char *)ar_sha(ah), ":",
1199 				    inet_ntoa_r(isaddr, addrbuf),
1200 				    ifp->if_xname);
1201 			return;
1202 		}
1203 		if (log_arp_movements) {
1204 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1205 			    "to %*D on %s\n",
1206 			    inet_ntoa_r(isaddr, addrbuf),
1207 			    ifp->if_addrlen,
1208 			    (u_char *)la->ll_addr, ":",
1209 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1210 			    ifp->if_xname);
1211 		}
1212 	}
1213 
1214 	/* Calculate full link prepend to use in lle */
1215 	linkhdrsize = sizeof(linkhdr);
1216 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1217 	    &linkhdrsize, &lladdr_off) != 0)
1218 		return;
1219 
1220 	/* Check if something has changed */
1221 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1222 	    (la->la_flags & LLE_VALID) == 0) {
1223 		/* Try to perform LLE update */
1224 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1225 		    lladdr_off) == 0)
1226 			return;
1227 
1228 		/* Clear fast path feedback request if set */
1229 		la->r_skip_req = 0;
1230 	}
1231 
1232 	arp_mark_lle_reachable(la);
1233 
1234 	/*
1235 	 * The packets are all freed within the call to the output
1236 	 * routine.
1237 	 *
1238 	 * NB: The lock MUST be released before the call to the
1239 	 * output routine.
1240 	 */
1241 	if (la->la_hold != NULL) {
1242 		m_hold = la->la_hold;
1243 		la->la_hold = NULL;
1244 		la->la_numheld = 0;
1245 		lltable_fill_sa_entry(la, &sa);
1246 		LLE_WUNLOCK(la);
1247 		for (; m_hold != NULL; m_hold = m_hold_next) {
1248 			m_hold_next = m_hold->m_nextpkt;
1249 			m_hold->m_nextpkt = NULL;
1250 			/* Avoid confusing lower layers. */
1251 			m_clrprotoflags(m_hold);
1252 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1253 		}
1254 	} else
1255 		LLE_WUNLOCK(la);
1256 }
1257 
1258 static void
1259 arp_mark_lle_reachable(struct llentry *la)
1260 {
1261 	int canceled, wtime;
1262 
1263 	LLE_WLOCK_ASSERT(la);
1264 
1265 	la->ln_state = ARP_LLINFO_REACHABLE;
1266 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1267 
1268 	if (!(la->la_flags & LLE_STATIC)) {
1269 		LLE_ADDREF(la);
1270 		la->la_expire = time_uptime + V_arpt_keep;
1271 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1272 		if (wtime < 0)
1273 			wtime = V_arpt_keep;
1274 		canceled = callout_reset(&la->lle_timer,
1275 		    hz * wtime, arptimer, la);
1276 		if (canceled)
1277 			LLE_REMREF(la);
1278 	}
1279 	la->la_asked = 0;
1280 	la->la_preempt = V_arp_maxtries;
1281 }
1282 
1283 /*
1284  * Add permanent link-layer record for given interface address.
1285  */
1286 static __noinline void
1287 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1288 {
1289 	struct llentry *lle, *lle_tmp;
1290 
1291 	/*
1292 	 * Interface address LLE record is considered static
1293 	 * because kernel code relies on LLE_STATIC flag to check
1294 	 * if these entries can be rewriten by arp updates.
1295 	 */
1296 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1297 	if (lle == NULL) {
1298 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1299 		    "entry for interface address\n");
1300 		return;
1301 	}
1302 
1303 	IF_AFDATA_WLOCK(ifp);
1304 	LLE_WLOCK(lle);
1305 	/* Unlink any entry if exists */
1306 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1307 	if (lle_tmp != NULL)
1308 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1309 
1310 	lltable_link_entry(LLTABLE(ifp), lle);
1311 	IF_AFDATA_WUNLOCK(ifp);
1312 
1313 	if (lle_tmp != NULL)
1314 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1315 
1316 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1317 	LLE_WUNLOCK(lle);
1318 	if (lle_tmp != NULL)
1319 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1320 }
1321 
1322 /*
1323  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1324  * of valid values.
1325  */
1326 static int
1327 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1328 {
1329 	int error;
1330 	int rexmit_count = *(int *)arg1;
1331 
1332 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1333 
1334 	/* Enforce limits on any new value that may have been set. */
1335 	if (!error && req->newptr) {
1336 		/* A new value was set. */
1337 		if (rexmit_count < 0) {
1338 			rexmit_count = 0;
1339 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1340 			rexmit_count = MAX_GARP_RETRANSMITS;
1341 		}
1342 		*(int *)arg1 = rexmit_count;
1343 	}
1344 
1345 	return (error);
1346 }
1347 
1348 /*
1349  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1350  * retransmit it again. A pending callout owns a reference to the ifa.
1351  */
1352 static void
1353 garp_rexmit(void *arg)
1354 {
1355 	struct in_ifaddr *ia = arg;
1356 
1357 	if (callout_pending(&ia->ia_garp_timer) ||
1358 	    !callout_active(&ia->ia_garp_timer)) {
1359 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1360 		ifa_free(&ia->ia_ifa);
1361 		return;
1362 	}
1363 
1364 	/*
1365 	 * Drop lock while the ARP request is generated.
1366 	 */
1367 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1368 
1369 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1370 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1371 
1372 	/*
1373 	 * Increment the count of retransmissions. If the count has reached the
1374 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1375 	 * the callout to retransmit another GARP packet.
1376 	 */
1377 	++ia->ia_garp_count;
1378 	if (ia->ia_garp_count >= garp_rexmit_count) {
1379 		ifa_free(&ia->ia_ifa);
1380 	} else {
1381 		int rescheduled;
1382 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1383 		rescheduled = callout_reset(&ia->ia_garp_timer,
1384 		    (1 << ia->ia_garp_count) * hz,
1385 		    garp_rexmit, ia);
1386 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1387 		if (rescheduled) {
1388 			ifa_free(&ia->ia_ifa);
1389 		}
1390 	}
1391 }
1392 
1393 /*
1394  * Start the GARP retransmit timer.
1395  *
1396  * A single GARP is always transmitted when an IPv4 address is added
1397  * to an interface and that is usually sufficient. However, in some
1398  * circumstances, such as when a shared address is passed between
1399  * cluster nodes, this single GARP may occasionally be dropped or
1400  * lost. This can lead to neighbors on the network link working with a
1401  * stale ARP cache and sending packets destined for that address to
1402  * the node that previously owned the address, which may not respond.
1403  *
1404  * To avoid this situation, GARP retransmits can be enabled by setting
1405  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1406  * than zero. The setting represents the maximum number of
1407  * retransmissions. The interval between retransmissions is calculated
1408  * using an exponential backoff algorithm, doubling each time, so the
1409  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1410  */
1411 static void
1412 garp_timer_start(struct ifaddr *ifa)
1413 {
1414 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1415 
1416 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1417 	ia->ia_garp_count = 0;
1418 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1419 	    garp_rexmit, ia) == 0) {
1420 		ifa_ref(ifa);
1421 	}
1422 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1423 }
1424 
1425 void
1426 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1427 {
1428 	const struct sockaddr_in *dst_in;
1429 	const struct sockaddr *dst;
1430 
1431 	if (ifa->ifa_carp != NULL)
1432 		return;
1433 
1434 	dst = ifa->ifa_addr;
1435 	dst_in = (const struct sockaddr_in *)dst;
1436 
1437 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1438 		return;
1439 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1440 	if (garp_rexmit_count > 0) {
1441 		garp_timer_start(ifa);
1442 	}
1443 
1444 	arp_add_ifa_lle(ifp, dst);
1445 }
1446 
1447 void
1448 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1449 {
1450 
1451 	if (ntohl(addr.s_addr) != INADDR_ANY)
1452 		arprequest(ifp, &addr, &addr, enaddr);
1453 }
1454 
1455 /*
1456  * Sends gratuitous ARPs for each ifaddr to notify other
1457  * nodes about the address change.
1458  */
1459 static __noinline void
1460 arp_handle_ifllchange(struct ifnet *ifp)
1461 {
1462 	struct ifaddr *ifa;
1463 
1464 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1465 		if (ifa->ifa_addr->sa_family == AF_INET)
1466 			arp_ifinit(ifp, ifa);
1467 	}
1468 }
1469 
1470 /*
1471  * A handler for interface link layer address change event.
1472  */
1473 static void
1474 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1475 {
1476 
1477 	lltable_update_ifaddr(LLTABLE(ifp));
1478 
1479 	if ((ifp->if_flags & IFF_UP) != 0)
1480 		arp_handle_ifllchange(ifp);
1481 }
1482 
1483 static void
1484 vnet_arp_init(void)
1485 {
1486 
1487 	if (IS_DEFAULT_VNET(curvnet)) {
1488 		netisr_register(&arp_nh);
1489 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1490 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1491 	}
1492 #ifdef VIMAGE
1493 	else
1494 		netisr_register_vnet(&arp_nh);
1495 #endif
1496 }
1497 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1498     vnet_arp_init, 0);
1499 
1500 #ifdef VIMAGE
1501 /*
1502  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1503  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1504  */
1505 static void
1506 vnet_arp_destroy(__unused void *arg)
1507 {
1508 
1509 	netisr_unregister_vnet(&arp_nh);
1510 }
1511 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1512     vnet_arp_destroy, NULL);
1513 #endif
1514