xref: /freebsd/sys/netinet/if_ether.c (revision 28f4385e45a2681c14bd04b83fe1796eaefe8265)
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 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60);	/* once resolved, good for 20
98 						 * minutes */
99 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
100 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
101 VNET_DEFINE_STATIC(int, arpt_down) = 20;	/* keep incomplete entries for
102 						 * 20 seconds */
103 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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 epoch_tracker et;
363 		struct ifaddr *ifa;
364 
365 		NET_EPOCH_ENTER(et);
366 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
367 			if (ifa->ifa_addr->sa_family != AF_INET)
368 				continue;
369 
370 			if (ifa->ifa_carp) {
371 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
372 					continue;
373 				sip = &IA_SIN(ifa)->sin_addr;
374 			} else {
375 				carpaddr = NULL;
376 				sip = &IA_SIN(ifa)->sin_addr;
377 			}
378 
379 			if (0 == ((sip->s_addr ^ tip->s_addr) &
380 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
381 				break;  /* found it. */
382 		}
383 		NET_EPOCH_EXIT(et);
384 		if (sip == NULL) {
385 			printf("%s: cannot find matching address\n", __func__);
386 			return;
387 		}
388 	}
389 	if (enaddr == NULL)
390 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
391 
392 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
393 		return;
394 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
395 		2 * ifp->if_addrlen;
396 	m->m_pkthdr.len = m->m_len;
397 	M_ALIGN(m, m->m_len);
398 	ah = mtod(m, struct arphdr *);
399 	bzero((caddr_t)ah, m->m_len);
400 #ifdef MAC
401 	mac_netinet_arp_send(ifp, m);
402 #endif
403 	ah->ar_pro = htons(ETHERTYPE_IP);
404 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
405 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
406 	ah->ar_op = htons(ARPOP_REQUEST);
407 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
408 	bcopy(sip, ar_spa(ah), ah->ar_pln);
409 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
410 	sa.sa_family = AF_ARP;
411 	sa.sa_len = 2;
412 
413 	/* Calculate link header for sending frame */
414 	bzero(&ro, sizeof(ro));
415 	linkhdrsize = sizeof(linkhdr);
416 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
417 	if (error != 0 && error != EAFNOSUPPORT) {
418 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
419 		    if_name(ifp), error);
420 		return;
421 	}
422 
423 	ro.ro_prepend = linkhdr;
424 	ro.ro_plen = linkhdrsize;
425 	ro.ro_flags = 0;
426 
427 	m->m_flags |= M_BCAST;
428 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
429 	(*ifp->if_output)(ifp, m, &sa, &ro);
430 	ARPSTAT_INC(txrequests);
431 }
432 
433 
434 /*
435  * Resolve an IP address into an ethernet address - heavy version.
436  * Used internally by arpresolve().
437  * We have already checked that we can't use an existing lle without
438  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
439  *
440  * On success, desten and pflags are filled in and the function returns 0;
441  * If the packet must be held pending resolution, we return EWOULDBLOCK
442  * On other errors, we return the corresponding error code.
443  * Note that m_freem() handles NULL.
444  */
445 static int
446 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
447 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
448 	struct llentry **plle)
449 {
450 	struct llentry *la = NULL, *la_tmp;
451 	struct mbuf *curr = NULL;
452 	struct mbuf *next = NULL;
453 	int error, renew;
454 	char *lladdr;
455 	int ll_len;
456 
457 	if (pflags != NULL)
458 		*pflags = 0;
459 	if (plle != NULL)
460 		*plle = NULL;
461 
462 	if ((flags & LLE_CREATE) == 0) {
463 		struct epoch_tracker et;
464 
465 		NET_EPOCH_ENTER(et);
466 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
467 		NET_EPOCH_EXIT(et);
468 	}
469 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
470 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
471 		if (la == NULL) {
472 			char addrbuf[INET_ADDRSTRLEN];
473 
474 			log(LOG_DEBUG,
475 			    "arpresolve: can't allocate llinfo for %s on %s\n",
476 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
477 			    if_name(ifp));
478 			m_freem(m);
479 			return (EINVAL);
480 		}
481 
482 		IF_AFDATA_WLOCK(ifp);
483 		LLE_WLOCK(la);
484 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
485 		/* Prefer ANY existing lle over newly-created one */
486 		if (la_tmp == NULL)
487 			lltable_link_entry(LLTABLE(ifp), la);
488 		IF_AFDATA_WUNLOCK(ifp);
489 		if (la_tmp != NULL) {
490 			lltable_free_entry(LLTABLE(ifp), la);
491 			la = la_tmp;
492 		}
493 	}
494 	if (la == NULL) {
495 		m_freem(m);
496 		return (EINVAL);
497 	}
498 
499 	if ((la->la_flags & LLE_VALID) &&
500 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
501 		if (flags & LLE_ADDRONLY) {
502 			lladdr = la->ll_addr;
503 			ll_len = ifp->if_addrlen;
504 		} else {
505 			lladdr = la->r_linkdata;
506 			ll_len = la->r_hdrlen;
507 		}
508 		bcopy(lladdr, desten, ll_len);
509 
510 		/* Notify LLE code that the entry was used by datapath */
511 		llentry_mark_used(la);
512 		if (pflags != NULL)
513 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
514 		if (plle) {
515 			LLE_ADDREF(la);
516 			*plle = la;
517 		}
518 		LLE_WUNLOCK(la);
519 		return (0);
520 	}
521 
522 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
523 	/*
524 	 * There is an arptab entry, but no ethernet address
525 	 * response yet.  Add the mbuf to the list, dropping
526 	 * the oldest packet if we have exceeded the system
527 	 * setting.
528 	 */
529 	if (m != NULL) {
530 		if (la->la_numheld >= V_arp_maxhold) {
531 			if (la->la_hold != NULL) {
532 				next = la->la_hold->m_nextpkt;
533 				m_freem(la->la_hold);
534 				la->la_hold = next;
535 				la->la_numheld--;
536 				ARPSTAT_INC(dropped);
537 			}
538 		}
539 		if (la->la_hold != NULL) {
540 			curr = la->la_hold;
541 			while (curr->m_nextpkt != NULL)
542 				curr = curr->m_nextpkt;
543 			curr->m_nextpkt = m;
544 		} else
545 			la->la_hold = m;
546 		la->la_numheld++;
547 	}
548 	/*
549 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
550 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
551 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
552 	 * ARP request, but not faster than one request per second.
553 	 */
554 	if (la->la_asked < V_arp_maxtries)
555 		error = EWOULDBLOCK;	/* First request. */
556 	else
557 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
558 
559 	if (renew) {
560 		int canceled;
561 
562 		LLE_ADDREF(la);
563 		la->la_expire = time_uptime;
564 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
565 		    arptimer, la);
566 		if (canceled)
567 			LLE_REMREF(la);
568 		la->la_asked++;
569 		LLE_WUNLOCK(la);
570 		arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
571 		return (error);
572 	}
573 
574 	LLE_WUNLOCK(la);
575 	return (error);
576 }
577 
578 /*
579  * Lookups link header based on an IP address.
580  * On input:
581  *    ifp is the interface we use
582  *    is_gw != 0 if @dst represents gateway to some destination
583  *    m is the mbuf. May be NULL if we don't have a packet.
584  *    dst is the next hop,
585  *    desten is the storage to put LL header.
586  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
587  *
588  * On success, full/partial link header and flags are filled in and
589  * the function returns 0.
590  * If the packet must be held pending resolution, we return EWOULDBLOCK
591  * On other errors, we return the corresponding error code.
592  * Note that m_freem() handles NULL.
593  */
594 int
595 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
596 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
597 	struct llentry **plle)
598 {
599 	struct epoch_tracker et;
600 	struct llentry *la = NULL;
601 
602 	if (pflags != NULL)
603 		*pflags = 0;
604 	if (plle != NULL)
605 		*plle = NULL;
606 
607 	if (m != NULL) {
608 		if (m->m_flags & M_BCAST) {
609 			/* broadcast */
610 			(void)memcpy(desten,
611 			    ifp->if_broadcastaddr, ifp->if_addrlen);
612 			return (0);
613 		}
614 		if (m->m_flags & M_MCAST) {
615 			/* multicast */
616 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
617 			return (0);
618 		}
619 	}
620 
621 	NET_EPOCH_ENTER(et);
622 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
623 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
624 		/* Entry found, let's copy lle info */
625 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
626 		if (pflags != NULL)
627 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
628 		/* Notify the LLE handling code that the entry was used. */
629 		llentry_mark_used(la);
630 		if (plle) {
631 			LLE_ADDREF(la);
632 			*plle = la;
633 			LLE_WUNLOCK(la);
634 		}
635 		NET_EPOCH_EXIT(et);
636 		return (0);
637 	}
638 	if (plle && la)
639 		LLE_WUNLOCK(la);
640 	NET_EPOCH_EXIT(et);
641 
642 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
643 	    desten, pflags, plle));
644 }
645 
646 /*
647  * Common length and type checks are done here,
648  * then the protocol-specific routine is called.
649  */
650 static void
651 arpintr(struct mbuf *m)
652 {
653 	struct arphdr *ar;
654 	struct ifnet *ifp;
655 	char *layer;
656 	int hlen;
657 
658 	ifp = m->m_pkthdr.rcvif;
659 
660 	if (m->m_len < sizeof(struct arphdr) &&
661 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
662 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
663 		    if_name(ifp));
664 		return;
665 	}
666 	ar = mtod(m, struct arphdr *);
667 
668 	/* Check if length is sufficient */
669 	if (m->m_len <  arphdr_len(ar)) {
670 		m = m_pullup(m, arphdr_len(ar));
671 		if (m == NULL) {
672 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
673 			    if_name(ifp));
674 			return;
675 		}
676 		ar = mtod(m, struct arphdr *);
677 	}
678 
679 	hlen = 0;
680 	layer = "";
681 	switch (ntohs(ar->ar_hrd)) {
682 	case ARPHRD_ETHER:
683 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
684 		layer = "ethernet";
685 		break;
686 	case ARPHRD_INFINIBAND:
687 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
688 		layer = "infiniband";
689 		break;
690 	case ARPHRD_IEEE1394:
691 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
692 		layer = "firewire";
693 
694 		/*
695 		 * Restrict too long hardware addresses.
696 		 * Currently we are capable of handling 20-byte
697 		 * addresses ( sizeof(lle->ll_addr) )
698 		 */
699 		if (ar->ar_hln >= 20)
700 			hlen = 16;
701 		break;
702 	default:
703 		ARP_LOG(LOG_NOTICE,
704 		    "packet with unknown hardware format 0x%02d received on "
705 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
706 		m_freem(m);
707 		return;
708 	}
709 
710 	if (hlen != 0 && hlen != ar->ar_hln) {
711 		ARP_LOG(LOG_NOTICE,
712 		    "packet with invalid %s address length %d received on %s\n",
713 		    layer, ar->ar_hln, if_name(ifp));
714 		m_freem(m);
715 		return;
716 	}
717 
718 	ARPSTAT_INC(received);
719 	switch (ntohs(ar->ar_pro)) {
720 #ifdef INET
721 	case ETHERTYPE_IP:
722 		in_arpinput(m);
723 		return;
724 #endif
725 	}
726 	m_freem(m);
727 }
728 
729 #ifdef INET
730 /*
731  * ARP for Internet protocols on 10 Mb/s Ethernet.
732  * Algorithm is that given in RFC 826.
733  * In addition, a sanity check is performed on the sender
734  * protocol address, to catch impersonators.
735  * We no longer handle negotiations for use of trailer protocol:
736  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
737  * along with IP replies if we wanted trailers sent to us,
738  * and also sent them in response to IP replies.
739  * This allowed either end to announce the desire to receive
740  * trailer packets.
741  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
742  * but formerly didn't normally send requests.
743  */
744 static int log_arp_wrong_iface = 1;
745 static int log_arp_movements = 1;
746 static int log_arp_permanent_modify = 1;
747 static int allow_multicast = 0;
748 
749 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
750 	&log_arp_wrong_iface, 0,
751 	"log arp packets arriving on the wrong interface");
752 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
753 	&log_arp_movements, 0,
754 	"log arp replies from MACs different than the one in the cache");
755 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
756 	&log_arp_permanent_modify, 0,
757 	"log arp replies from MACs different than the one in the permanent arp entry");
758 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
759 	&allow_multicast, 0, "accept multicast addresses");
760 
761 static void
762 in_arpinput(struct mbuf *m)
763 {
764 	struct rm_priotracker in_ifa_tracker;
765 	struct arphdr *ah;
766 	struct ifnet *ifp = m->m_pkthdr.rcvif;
767 	struct llentry *la = NULL, *la_tmp;
768 	struct ifaddr *ifa;
769 	struct in_ifaddr *ia;
770 	struct sockaddr sa;
771 	struct in_addr isaddr, itaddr, myaddr;
772 	u_int8_t *enaddr = NULL;
773 	int op;
774 	int bridged = 0, is_bridge = 0;
775 	int carped;
776 	struct sockaddr_in sin;
777 	struct sockaddr *dst;
778 	struct nhop4_basic nh4;
779 	uint8_t linkhdr[LLE_MAX_LINKHDR];
780 	struct route ro;
781 	size_t linkhdrsize;
782 	int lladdr_off;
783 	int error;
784 	char addrbuf[INET_ADDRSTRLEN];
785 	struct epoch_tracker et;
786 
787 	sin.sin_len = sizeof(struct sockaddr_in);
788 	sin.sin_family = AF_INET;
789 	sin.sin_addr.s_addr = 0;
790 
791 	if (ifp->if_bridge)
792 		bridged = 1;
793 	if (ifp->if_type == IFT_BRIDGE)
794 		is_bridge = 1;
795 
796 	/*
797 	 * We already have checked that mbuf contains enough contiguous data
798 	 * to hold entire arp message according to the arp header.
799 	 */
800 	ah = mtod(m, struct arphdr *);
801 
802 	/*
803 	 * ARP is only for IPv4 so we can reject packets with
804 	 * a protocol length not equal to an IPv4 address.
805 	 */
806 	if (ah->ar_pln != sizeof(struct in_addr)) {
807 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
808 		    sizeof(struct in_addr));
809 		goto drop;
810 	}
811 
812 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
813 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
814 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
815 		goto drop;
816 	}
817 
818 	op = ntohs(ah->ar_op);
819 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
820 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
821 
822 	if (op == ARPOP_REPLY)
823 		ARPSTAT_INC(rxreplies);
824 
825 	/*
826 	 * For a bridge, we want to check the address irrespective
827 	 * of the receive interface. (This will change slightly
828 	 * when we have clusters of interfaces).
829 	 */
830 	IN_IFADDR_RLOCK(&in_ifa_tracker);
831 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
832 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
833 		    ia->ia_ifp == ifp) &&
834 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
835 		    (ia->ia_ifa.ifa_carp == NULL ||
836 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
837 			ifa_ref(&ia->ia_ifa);
838 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
839 			goto match;
840 		}
841 	}
842 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
843 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
844 		    ia->ia_ifp == ifp) &&
845 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
846 			ifa_ref(&ia->ia_ifa);
847 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
848 			goto match;
849 		}
850 
851 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
852   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
853   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
854   addr == ia->ia_addr.sin_addr.s_addr)
855 	/*
856 	 * Check the case when bridge shares its MAC address with
857 	 * some of its children, so packets are claimed by bridge
858 	 * itself (bridge_input() does it first), but they are really
859 	 * meant to be destined to the bridge member.
860 	 */
861 	if (is_bridge) {
862 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
863 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
864 				ifa_ref(&ia->ia_ifa);
865 				ifp = ia->ia_ifp;
866 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
867 				goto match;
868 			}
869 		}
870 	}
871 #undef BDG_MEMBER_MATCHES_ARP
872 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
873 
874 	/*
875 	 * No match, use the first inet address on the receive interface
876 	 * as a dummy address for the rest of the function.
877 	 */
878 	NET_EPOCH_ENTER(et);
879 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
880 		if (ifa->ifa_addr->sa_family == AF_INET &&
881 		    (ifa->ifa_carp == NULL ||
882 		    (*carp_iamatch_p)(ifa, &enaddr))) {
883 			ia = ifatoia(ifa);
884 			ifa_ref(ifa);
885 			NET_EPOCH_EXIT(et);
886 			goto match;
887 		}
888 	NET_EPOCH_EXIT(et);
889 
890 	/*
891 	 * If bridging, fall back to using any inet address.
892 	 */
893 	IN_IFADDR_RLOCK(&in_ifa_tracker);
894 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
895 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
896 		goto drop;
897 	}
898 	ifa_ref(&ia->ia_ifa);
899 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
900 match:
901 	if (!enaddr)
902 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
903 	carped = (ia->ia_ifa.ifa_carp != NULL);
904 	myaddr = ia->ia_addr.sin_addr;
905 	ifa_free(&ia->ia_ifa);
906 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
907 		goto drop;	/* it's from me, ignore it. */
908 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
909 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
910 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
911 		goto drop;
912 	}
913 
914 	if (ifp->if_addrlen != ah->ar_hln) {
915 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
916 		    "i/f %d (ignored)\n", ifp->if_addrlen,
917 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
918 		    ifp->if_addrlen);
919 		goto drop;
920 	}
921 
922 	/*
923 	 * Warn if another host is using the same IP address, but only if the
924 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
925 	 * case we suppress the warning to avoid false positive complaints of
926 	 * potential misconfiguration.
927 	 */
928 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
929 	    myaddr.s_addr != 0) {
930 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
931 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
932 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
933 		itaddr = myaddr;
934 		ARPSTAT_INC(dupips);
935 		goto reply;
936 	}
937 	if (ifp->if_flags & IFF_STATICARP)
938 		goto reply;
939 
940 	bzero(&sin, sizeof(sin));
941 	sin.sin_len = sizeof(struct sockaddr_in);
942 	sin.sin_family = AF_INET;
943 	sin.sin_addr = isaddr;
944 	dst = (struct sockaddr *)&sin;
945 	NET_EPOCH_ENTER(et);
946 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
947 	NET_EPOCH_EXIT(et);
948 	if (la != NULL)
949 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
950 	else if (itaddr.s_addr == myaddr.s_addr) {
951 		/*
952 		 * Request/reply to our address, but no lle exists yet.
953 		 * Calculate full link prepend to use in lle.
954 		 */
955 		linkhdrsize = sizeof(linkhdr);
956 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
957 		    &linkhdrsize, &lladdr_off) != 0)
958 			goto reply;
959 
960 		/* Allocate new entry */
961 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
962 		if (la == NULL) {
963 
964 			/*
965 			 * lle creation may fail if source address belongs
966 			 * to non-directly connected subnet. However, we
967 			 * will try to answer the request instead of dropping
968 			 * frame.
969 			 */
970 			goto reply;
971 		}
972 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
973 		    lladdr_off);
974 
975 		IF_AFDATA_WLOCK(ifp);
976 		LLE_WLOCK(la);
977 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
978 
979 		/*
980 		 * Check if lle still does not exists.
981 		 * If it does, that means that we either
982 		 * 1) have configured it explicitly, via
983 		 * 1a) 'arp -s' static entry or
984 		 * 1b) interface address static record
985 		 * or
986 		 * 2) it was the result of sending first packet to-host
987 		 * or
988 		 * 3) it was another arp reply packet we handled in
989 		 * different thread.
990 		 *
991 		 * In all cases except 3) we definitely need to prefer
992 		 * existing lle. For the sake of simplicity, prefer any
993 		 * existing lle over newly-create one.
994 		 */
995 		if (la_tmp == NULL)
996 			lltable_link_entry(LLTABLE(ifp), la);
997 		IF_AFDATA_WUNLOCK(ifp);
998 
999 		if (la_tmp == NULL) {
1000 			arp_mark_lle_reachable(la);
1001 			LLE_WUNLOCK(la);
1002 		} else {
1003 			/* Free newly-create entry and handle packet */
1004 			lltable_free_entry(LLTABLE(ifp), la);
1005 			la = la_tmp;
1006 			la_tmp = NULL;
1007 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1008 			/* arp_check_update_lle() returns @la unlocked */
1009 		}
1010 		la = NULL;
1011 	}
1012 reply:
1013 	if (op != ARPOP_REQUEST)
1014 		goto drop;
1015 	ARPSTAT_INC(rxrequests);
1016 
1017 	if (itaddr.s_addr == myaddr.s_addr) {
1018 		/* Shortcut.. the receiving interface is the target. */
1019 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1020 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1021 	} else {
1022 		struct llentry *lle = NULL;
1023 
1024 		sin.sin_addr = itaddr;
1025 		NET_EPOCH_ENTER(et);
1026 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1027 		NET_EPOCH_EXIT(et);
1028 
1029 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1030 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1031 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1032 			LLE_RUNLOCK(lle);
1033 		} else {
1034 
1035 			if (lle != NULL)
1036 				LLE_RUNLOCK(lle);
1037 
1038 			if (!V_arp_proxyall)
1039 				goto drop;
1040 
1041 			/* XXX MRT use table 0 for arp reply  */
1042 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1043 				goto drop;
1044 
1045 			/*
1046 			 * Don't send proxies for nodes on the same interface
1047 			 * as this one came out of, or we'll get into a fight
1048 			 * over who claims what Ether address.
1049 			 */
1050 			if (nh4.nh_ifp == ifp)
1051 				goto drop;
1052 
1053 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1054 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1055 
1056 			/*
1057 			 * Also check that the node which sent the ARP packet
1058 			 * is on the interface we expect it to be on. This
1059 			 * avoids ARP chaos if an interface is connected to the
1060 			 * wrong network.
1061 			 */
1062 
1063 			/* XXX MRT use table 0 for arp checks */
1064 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1065 				goto drop;
1066 			if (nh4.nh_ifp != ifp) {
1067 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1068 				    " from %s via %s\n",
1069 				    inet_ntoa_r(isaddr, addrbuf),
1070 				    ifp->if_xname);
1071 				goto drop;
1072 			}
1073 
1074 #ifdef DEBUG_PROXY
1075 			printf("arp: proxying for %s\n",
1076 			    inet_ntoa_r(itaddr, addrbuf));
1077 #endif
1078 		}
1079 	}
1080 
1081 	if (itaddr.s_addr == myaddr.s_addr &&
1082 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1083 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1084 #ifdef DEBUG_LINKLOCAL
1085 		printf("arp: sending reply for link-local addr %s\n",
1086 		    inet_ntoa_r(itaddr, addrbuf));
1087 #endif
1088 		m->m_flags |= M_BCAST;
1089 		m->m_flags &= ~M_MCAST;
1090 	} else {
1091 		/* default behaviour; never reply by broadcast. */
1092 		m->m_flags &= ~(M_BCAST|M_MCAST);
1093 	}
1094 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1095 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1096 	ah->ar_op = htons(ARPOP_REPLY);
1097 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1098 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1099 	m->m_pkthdr.len = m->m_len;
1100 	m->m_pkthdr.rcvif = NULL;
1101 	sa.sa_family = AF_ARP;
1102 	sa.sa_len = 2;
1103 
1104 	/* Calculate link header for sending frame */
1105 	bzero(&ro, sizeof(ro));
1106 	linkhdrsize = sizeof(linkhdr);
1107 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1108 
1109 	/*
1110 	 * arp_fillheader() may fail due to lack of support inside encap request
1111 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1112 	 * by if_output().
1113 	 */
1114 	if (error != 0 && error != EAFNOSUPPORT) {
1115 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1116 		    if_name(ifp), error);
1117 		return;
1118 	}
1119 
1120 	ro.ro_prepend = linkhdr;
1121 	ro.ro_plen = linkhdrsize;
1122 	ro.ro_flags = 0;
1123 
1124 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1125 	(*ifp->if_output)(ifp, m, &sa, &ro);
1126 	ARPSTAT_INC(txreplies);
1127 	return;
1128 
1129 drop:
1130 	m_freem(m);
1131 }
1132 #endif
1133 
1134 /*
1135  * Checks received arp data against existing @la.
1136  * Updates lle state/performs notification if necessary.
1137  */
1138 static void
1139 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1140     int bridged, struct llentry *la)
1141 {
1142 	struct sockaddr sa;
1143 	struct mbuf *m_hold, *m_hold_next;
1144 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1145 	size_t linkhdrsize;
1146 	int lladdr_off;
1147 	char addrbuf[INET_ADDRSTRLEN];
1148 
1149 	LLE_WLOCK_ASSERT(la);
1150 
1151 	/* the following is not an error when doing bridging */
1152 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1153 		if (log_arp_wrong_iface)
1154 			ARP_LOG(LOG_WARNING, "%s is on %s "
1155 			    "but got reply from %*D on %s\n",
1156 			    inet_ntoa_r(isaddr, addrbuf),
1157 			    la->lle_tbl->llt_ifp->if_xname,
1158 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1159 			    ifp->if_xname);
1160 		LLE_WUNLOCK(la);
1161 		return;
1162 	}
1163 	if ((la->la_flags & LLE_VALID) &&
1164 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1165 		if (la->la_flags & LLE_STATIC) {
1166 			LLE_WUNLOCK(la);
1167 			if (log_arp_permanent_modify)
1168 				ARP_LOG(LOG_ERR,
1169 				    "%*D attempts to modify "
1170 				    "permanent entry for %s on %s\n",
1171 				    ifp->if_addrlen,
1172 				    (u_char *)ar_sha(ah), ":",
1173 				    inet_ntoa_r(isaddr, addrbuf),
1174 				    ifp->if_xname);
1175 			return;
1176 		}
1177 		if (log_arp_movements) {
1178 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1179 			    "to %*D on %s\n",
1180 			    inet_ntoa_r(isaddr, addrbuf),
1181 			    ifp->if_addrlen,
1182 			    (u_char *)la->ll_addr, ":",
1183 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1184 			    ifp->if_xname);
1185 		}
1186 	}
1187 
1188 	/* Calculate full link prepend to use in lle */
1189 	linkhdrsize = sizeof(linkhdr);
1190 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1191 	    &linkhdrsize, &lladdr_off) != 0)
1192 		return;
1193 
1194 	/* Check if something has changed */
1195 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1196 	    (la->la_flags & LLE_VALID) == 0) {
1197 		/* Try to perform LLE update */
1198 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1199 		    lladdr_off) == 0)
1200 			return;
1201 
1202 		/* Clear fast path feedback request if set */
1203 		la->r_skip_req = 0;
1204 	}
1205 
1206 	arp_mark_lle_reachable(la);
1207 
1208 	/*
1209 	 * The packets are all freed within the call to the output
1210 	 * routine.
1211 	 *
1212 	 * NB: The lock MUST be released before the call to the
1213 	 * output routine.
1214 	 */
1215 	if (la->la_hold != NULL) {
1216 		m_hold = la->la_hold;
1217 		la->la_hold = NULL;
1218 		la->la_numheld = 0;
1219 		lltable_fill_sa_entry(la, &sa);
1220 		LLE_WUNLOCK(la);
1221 		for (; m_hold != NULL; m_hold = m_hold_next) {
1222 			m_hold_next = m_hold->m_nextpkt;
1223 			m_hold->m_nextpkt = NULL;
1224 			/* Avoid confusing lower layers. */
1225 			m_clrprotoflags(m_hold);
1226 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1227 		}
1228 	} else
1229 		LLE_WUNLOCK(la);
1230 }
1231 
1232 static void
1233 arp_mark_lle_reachable(struct llentry *la)
1234 {
1235 	int canceled, wtime;
1236 
1237 	LLE_WLOCK_ASSERT(la);
1238 
1239 	la->ln_state = ARP_LLINFO_REACHABLE;
1240 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1241 
1242 	if (!(la->la_flags & LLE_STATIC)) {
1243 		LLE_ADDREF(la);
1244 		la->la_expire = time_uptime + V_arpt_keep;
1245 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1246 		if (wtime < 0)
1247 			wtime = V_arpt_keep;
1248 		canceled = callout_reset(&la->lle_timer,
1249 		    hz * wtime, arptimer, la);
1250 		if (canceled)
1251 			LLE_REMREF(la);
1252 	}
1253 	la->la_asked = 0;
1254 	la->la_preempt = V_arp_maxtries;
1255 }
1256 
1257 /*
1258  * Add permanent link-layer record for given interface address.
1259  */
1260 static __noinline void
1261 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1262 {
1263 	struct llentry *lle, *lle_tmp;
1264 
1265 	/*
1266 	 * Interface address LLE record is considered static
1267 	 * because kernel code relies on LLE_STATIC flag to check
1268 	 * if these entries can be rewriten by arp updates.
1269 	 */
1270 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1271 	if (lle == NULL) {
1272 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1273 		    "entry for interface address\n");
1274 		return;
1275 	}
1276 
1277 	IF_AFDATA_WLOCK(ifp);
1278 	LLE_WLOCK(lle);
1279 	/* Unlink any entry if exists */
1280 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1281 	if (lle_tmp != NULL)
1282 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1283 
1284 	lltable_link_entry(LLTABLE(ifp), lle);
1285 	IF_AFDATA_WUNLOCK(ifp);
1286 
1287 	if (lle_tmp != NULL)
1288 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1289 
1290 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1291 	LLE_WUNLOCK(lle);
1292 	if (lle_tmp != NULL)
1293 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1294 }
1295 
1296 /*
1297  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1298  * of valid values.
1299  */
1300 static int
1301 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1302 {
1303 	int error;
1304 	int rexmit_count = *(int *)arg1;
1305 
1306 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1307 
1308 	/* Enforce limits on any new value that may have been set. */
1309 	if (!error && req->newptr) {
1310 		/* A new value was set. */
1311 		if (rexmit_count < 0) {
1312 			rexmit_count = 0;
1313 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1314 			rexmit_count = MAX_GARP_RETRANSMITS;
1315 		}
1316 		*(int *)arg1 = rexmit_count;
1317 	}
1318 
1319 	return (error);
1320 }
1321 
1322 /*
1323  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1324  * retransmit it again. A pending callout owns a reference to the ifa.
1325  */
1326 static void
1327 garp_rexmit(void *arg)
1328 {
1329 	struct in_ifaddr *ia = arg;
1330 
1331 	if (callout_pending(&ia->ia_garp_timer) ||
1332 	    !callout_active(&ia->ia_garp_timer)) {
1333 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1334 		ifa_free(&ia->ia_ifa);
1335 		return;
1336 	}
1337 
1338 	/*
1339 	 * Drop lock while the ARP request is generated.
1340 	 */
1341 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1342 
1343 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1344 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1345 
1346 	/*
1347 	 * Increment the count of retransmissions. If the count has reached the
1348 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1349 	 * the callout to retransmit another GARP packet.
1350 	 */
1351 	++ia->ia_garp_count;
1352 	if (ia->ia_garp_count >= garp_rexmit_count) {
1353 		ifa_free(&ia->ia_ifa);
1354 	} else {
1355 		int rescheduled;
1356 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1357 		rescheduled = callout_reset(&ia->ia_garp_timer,
1358 		    (1 << ia->ia_garp_count) * hz,
1359 		    garp_rexmit, ia);
1360 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1361 		if (rescheduled) {
1362 			ifa_free(&ia->ia_ifa);
1363 		}
1364 	}
1365 }
1366 
1367 /*
1368  * Start the GARP retransmit timer.
1369  *
1370  * A single GARP is always transmitted when an IPv4 address is added
1371  * to an interface and that is usually sufficient. However, in some
1372  * circumstances, such as when a shared address is passed between
1373  * cluster nodes, this single GARP may occasionally be dropped or
1374  * lost. This can lead to neighbors on the network link working with a
1375  * stale ARP cache and sending packets destined for that address to
1376  * the node that previously owned the address, which may not respond.
1377  *
1378  * To avoid this situation, GARP retransmits can be enabled by setting
1379  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1380  * than zero. The setting represents the maximum number of
1381  * retransmissions. The interval between retransmissions is calculated
1382  * using an exponential backoff algorithm, doubling each time, so the
1383  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1384  */
1385 static void
1386 garp_timer_start(struct ifaddr *ifa)
1387 {
1388 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1389 
1390 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1391 	ia->ia_garp_count = 0;
1392 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1393 	    garp_rexmit, ia) == 0) {
1394 		ifa_ref(ifa);
1395 	}
1396 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1397 }
1398 
1399 void
1400 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1401 {
1402 	const struct sockaddr_in *dst_in;
1403 	const struct sockaddr *dst;
1404 
1405 	if (ifa->ifa_carp != NULL)
1406 		return;
1407 
1408 	dst = ifa->ifa_addr;
1409 	dst_in = (const struct sockaddr_in *)dst;
1410 
1411 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1412 		return;
1413 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1414 	if (garp_rexmit_count > 0) {
1415 		garp_timer_start(ifa);
1416 	}
1417 
1418 	arp_add_ifa_lle(ifp, dst);
1419 }
1420 
1421 void
1422 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1423 {
1424 
1425 	if (ntohl(addr.s_addr) != INADDR_ANY)
1426 		arprequest(ifp, &addr, &addr, enaddr);
1427 }
1428 
1429 /*
1430  * Sends gratuitous ARPs for each ifaddr to notify other
1431  * nodes about the address change.
1432  */
1433 static __noinline void
1434 arp_handle_ifllchange(struct ifnet *ifp)
1435 {
1436 	struct ifaddr *ifa;
1437 
1438 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1439 		if (ifa->ifa_addr->sa_family == AF_INET)
1440 			arp_ifinit(ifp, ifa);
1441 	}
1442 }
1443 
1444 /*
1445  * A handler for interface link layer address change event.
1446  */
1447 static void
1448 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1449 {
1450 
1451 	lltable_update_ifaddr(LLTABLE(ifp));
1452 
1453 	if ((ifp->if_flags & IFF_UP) != 0)
1454 		arp_handle_ifllchange(ifp);
1455 }
1456 
1457 static void
1458 vnet_arp_init(void)
1459 {
1460 
1461 	if (IS_DEFAULT_VNET(curvnet)) {
1462 		netisr_register(&arp_nh);
1463 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1464 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1465 	}
1466 #ifdef VIMAGE
1467 	else
1468 		netisr_register_vnet(&arp_nh);
1469 #endif
1470 }
1471 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1472     vnet_arp_init, 0);
1473 
1474 #ifdef VIMAGE
1475 /*
1476  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1477  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1478  */
1479 static void
1480 vnet_arp_destroy(__unused void *arg)
1481 {
1482 
1483 	netisr_unregister_vnet(&arp_nh);
1484 }
1485 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1486     vnet_arp_destroy, NULL);
1487 #endif
1488