xref: /freebsd/sys/netinet/if_ether.c (revision 344d411c676dc92e648be86de369c4d1dd070209)
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/eventhandler.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/rmlock.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/netisr.h>
64 #include <net/ethernet.h>
65 #include <net/route.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_fib.h>
70 #include <netinet/in_var.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #ifdef INET
74 #include <netinet/ip_carp.h>
75 #endif
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define SIN(s) ((const struct sockaddr_in *)(s))
80 
81 static struct timeval arp_lastlog;
82 static int arp_curpps;
83 static int arp_maxpps = 1;
84 
85 /* Simple ARP state machine */
86 enum arp_llinfo_state {
87 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
88 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
89 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
90 	ARP_LLINFO_DELETED,	/* LLE is deleted */
91 };
92 
93 SYSCTL_DECL(_net_link_ether);
94 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
95     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
96     "");
97 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
98     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
99     "");
100 
101 /* timer values */
102 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60);	/* once resolved, good for 20
103 						 * minutes */
104 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
105 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
106 VNET_DEFINE_STATIC(int, arpt_down) = 20;	/* keep incomplete entries for
107 						 * 20 seconds */
108 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
109 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
110 VNET_PCPUSTAT_SYSINIT(arpstat);
111 
112 #ifdef VIMAGE
113 VNET_PCPUSTAT_SYSUNINIT(arpstat);
114 #endif /* VIMAGE */
115 
116 VNET_DEFINE_STATIC(int, arp_maxhold) = 1;
117 
118 #define	V_arpt_keep		VNET(arpt_keep)
119 #define	V_arpt_down		VNET(arpt_down)
120 #define	V_arpt_rexmit		VNET(arpt_rexmit)
121 #define	V_arp_maxtries		VNET(arp_maxtries)
122 #define	V_arp_proxyall		VNET(arp_proxyall)
123 #define	V_arp_maxhold		VNET(arp_maxhold)
124 
125 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
126 	&VNET_NAME(arpt_keep), 0,
127 	"ARP entry lifetime in seconds");
128 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
129 	&VNET_NAME(arp_maxtries), 0,
130 	"ARP resolution attempts before returning error");
131 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
132 	&VNET_NAME(arp_proxyall), 0,
133 	"Enable proxy ARP for all suitable requests");
134 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
135 	&VNET_NAME(arpt_down), 0,
136 	"Incomplete ARP entry lifetime in seconds");
137 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
138     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
139 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
140 	&VNET_NAME(arp_maxhold), 0,
141 	"Number of packets to hold per ARP entry");
142 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
143 	CTLFLAG_RW, &arp_maxpps, 0,
144 	"Maximum number of remotely triggered ARP messages that can be "
145 	"logged per second");
146 
147 /*
148  * Due to the exponential backoff algorithm used for the interval between GARP
149  * retransmissions, the maximum number of retransmissions is limited for
150  * sanity. This limit corresponds to a maximum interval between retransmissions
151  * of 2^16 seconds ~= 18 hours.
152  *
153  * Making this limit more dynamic is more complicated than worthwhile,
154  * especially since sending out GARPs spaced days apart would be of little
155  * use. A maximum dynamic limit would look something like:
156  *
157  * const int max = fls(INT_MAX / hz) - 1;
158  */
159 #define MAX_GARP_RETRANSMITS 16
160 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
161 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
162 
163 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
164     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
165     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
166     "Number of times to retransmit GARP packets;"
167     " 0 to disable, maximum of 16");
168 
169 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO;	/* Min. log(9) level. */
170 #define	V_arp_log_level		VNET(arp_log_level)
171 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
172 	&VNET_NAME(arp_log_level), 0,
173 	"Minimum log(9) level for recording rate limited arp log messages. "
174 	"The higher will be log more (emerg=0, info=6 (default), debug=7).");
175 #define	ARP_LOG(pri, ...)	do {					\
176 	if ((pri) <= V_arp_log_level &&					\
177 	    ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
178 		log((pri), "arp: " __VA_ARGS__);			\
179 } while (0)
180 
181 
182 static void	arpintr(struct mbuf *);
183 static void	arptimer(void *);
184 #ifdef INET
185 static void	in_arpinput(struct mbuf *);
186 #endif
187 
188 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
189     struct ifnet *ifp, int bridged, struct llentry *la);
190 static void arp_mark_lle_reachable(struct llentry *la);
191 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
192 
193 static eventhandler_tag iflladdr_tag;
194 
195 static const struct netisr_handler arp_nh = {
196 	.nh_name = "arp",
197 	.nh_handler = arpintr,
198 	.nh_proto = NETISR_ARP,
199 	.nh_policy = NETISR_POLICY_SOURCE,
200 };
201 
202 /*
203  * Timeout routine.  Age arp_tab entries periodically.
204  */
205 static void
206 arptimer(void *arg)
207 {
208 	struct llentry *lle = (struct llentry *)arg;
209 	struct ifnet *ifp;
210 	int r_skip_req;
211 
212 	if (lle->la_flags & LLE_STATIC) {
213 		return;
214 	}
215 	LLE_WLOCK(lle);
216 	if (callout_pending(&lle->lle_timer)) {
217 		/*
218 		 * Here we are a bit odd here in the treatment of
219 		 * active/pending. If the pending bit is set, it got
220 		 * rescheduled before I ran. The active
221 		 * bit we ignore, since if it was stopped
222 		 * in ll_tablefree() and was currently running
223 		 * it would have return 0 so the code would
224 		 * not have deleted it since the callout could
225 		 * not be stopped so we want to go through
226 		 * with the delete here now. If the callout
227 		 * was restarted, the pending bit will be back on and
228 		 * we just want to bail since the callout_reset would
229 		 * return 1 and our reference would have been removed
230 		 * by arpresolve() below.
231 		 */
232 		LLE_WUNLOCK(lle);
233  		return;
234  	}
235 	ifp = lle->lle_tbl->llt_ifp;
236 	CURVNET_SET(ifp->if_vnet);
237 
238 	switch (lle->ln_state) {
239 	case ARP_LLINFO_REACHABLE:
240 
241 		/*
242 		 * Expiration time is approaching.
243 		 * Let's try to refresh entry if it is still
244 		 * in use.
245 		 *
246 		 * Set r_skip_req to get feedback from
247 		 * fast path. Change state and re-schedule
248 		 * ourselves.
249 		 */
250 		LLE_REQ_LOCK(lle);
251 		lle->r_skip_req = 1;
252 		LLE_REQ_UNLOCK(lle);
253 		lle->ln_state = ARP_LLINFO_VERIFY;
254 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
255 		LLE_WUNLOCK(lle);
256 		CURVNET_RESTORE();
257 		return;
258 	case ARP_LLINFO_VERIFY:
259 		LLE_REQ_LOCK(lle);
260 		r_skip_req = lle->r_skip_req;
261 		LLE_REQ_UNLOCK(lle);
262 
263 		if (r_skip_req == 0 && lle->la_preempt > 0) {
264 			/* Entry was used, issue refresh request */
265 			struct epoch_tracker et;
266 			struct in_addr dst;
267 
268 			dst = lle->r_l3addr.addr4;
269 			lle->la_preempt--;
270 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
271 			LLE_WUNLOCK(lle);
272 			NET_EPOCH_ENTER(et);
273 			arprequest(ifp, NULL, &dst, NULL);
274 			NET_EPOCH_EXIT(et);
275 			CURVNET_RESTORE();
276 			return;
277 		}
278 		/* Nothing happened. Reschedule if not too late */
279 		if (lle->la_expire > time_uptime) {
280 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
281 			LLE_WUNLOCK(lle);
282 			CURVNET_RESTORE();
283 			return;
284 		}
285 		break;
286 	case ARP_LLINFO_INCOMPLETE:
287 	case ARP_LLINFO_DELETED:
288 		break;
289 	}
290 
291 	if ((lle->la_flags & LLE_DELETED) == 0) {
292 		int evt;
293 
294 		if (lle->la_flags & LLE_VALID)
295 			evt = LLENTRY_EXPIRED;
296 		else
297 			evt = LLENTRY_TIMEDOUT;
298 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
299 	}
300 
301 	callout_stop(&lle->lle_timer);
302 
303 	/* XXX: LOR avoidance. We still have ref on lle. */
304 	LLE_WUNLOCK(lle);
305 	IF_AFDATA_LOCK(ifp);
306 	LLE_WLOCK(lle);
307 
308 	/* Guard against race with other llentry_free(). */
309 	if (lle->la_flags & LLE_LINKED) {
310 		LLE_REMREF(lle);
311 		lltable_unlink_entry(lle->lle_tbl, lle);
312 	}
313 	IF_AFDATA_UNLOCK(ifp);
314 
315 	size_t pkts_dropped = llentry_free(lle);
316 
317 	ARPSTAT_ADD(dropped, pkts_dropped);
318 	ARPSTAT_INC(timeouts);
319 
320 	CURVNET_RESTORE();
321 }
322 
323 /*
324  * Stores link-layer header for @ifp in format suitable for if_output()
325  * into buffer @buf. Resulting header length is stored in @bufsize.
326  *
327  * Returns 0 on success.
328  */
329 static int
330 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
331     size_t *bufsize)
332 {
333 	struct if_encap_req ereq;
334 	int error;
335 
336 	bzero(buf, *bufsize);
337 	bzero(&ereq, sizeof(ereq));
338 	ereq.buf = buf;
339 	ereq.bufsize = *bufsize;
340 	ereq.rtype = IFENCAP_LL;
341 	ereq.family = AF_ARP;
342 	ereq.lladdr = ar_tha(ah);
343 	ereq.hdata = (u_char *)ah;
344 	if (bcast)
345 		ereq.flags = IFENCAP_FLAG_BROADCAST;
346 	error = ifp->if_requestencap(ifp, &ereq);
347 	if (error == 0)
348 		*bufsize = ereq.bufsize;
349 
350 	return (error);
351 }
352 
353 
354 /*
355  * Broadcast an ARP request. Caller specifies:
356  *	- arp header source ip address
357  *	- arp header target ip address
358  *	- arp header source ethernet address
359  */
360 static int
361 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
362     const struct in_addr *tip, u_char *enaddr)
363 {
364 	struct mbuf *m;
365 	struct arphdr *ah;
366 	struct sockaddr sa;
367 	u_char *carpaddr = NULL;
368 	uint8_t linkhdr[LLE_MAX_LINKHDR];
369 	size_t linkhdrsize;
370 	struct route ro;
371 	int error;
372 
373 	NET_EPOCH_ASSERT();
374 
375 	if (sip == NULL) {
376 		/*
377 		 * The caller did not supply a source address, try to find
378 		 * a compatible one among those assigned to this interface.
379 		 */
380 		struct ifaddr *ifa;
381 
382 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
383 			if (ifa->ifa_addr->sa_family != AF_INET)
384 				continue;
385 
386 			if (ifa->ifa_carp) {
387 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
388 					continue;
389 				sip = &IA_SIN(ifa)->sin_addr;
390 			} else {
391 				carpaddr = NULL;
392 				sip = &IA_SIN(ifa)->sin_addr;
393 			}
394 
395 			if (0 == ((sip->s_addr ^ tip->s_addr) &
396 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
397 				break;  /* found it. */
398 		}
399 		if (sip == NULL) {
400 			printf("%s: cannot find matching address\n", __func__);
401 			return (EADDRNOTAVAIL);
402 		}
403 	}
404 	if (enaddr == NULL)
405 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
406 
407 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
408 		return (ENOMEM);
409 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
410 		2 * ifp->if_addrlen;
411 	m->m_pkthdr.len = m->m_len;
412 	M_ALIGN(m, m->m_len);
413 	ah = mtod(m, struct arphdr *);
414 	bzero((caddr_t)ah, m->m_len);
415 #ifdef MAC
416 	mac_netinet_arp_send(ifp, m);
417 #endif
418 	ah->ar_pro = htons(ETHERTYPE_IP);
419 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
420 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
421 	ah->ar_op = htons(ARPOP_REQUEST);
422 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
423 	bcopy(sip, ar_spa(ah), ah->ar_pln);
424 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
425 	sa.sa_family = AF_ARP;
426 	sa.sa_len = 2;
427 
428 	/* Calculate link header for sending frame */
429 	bzero(&ro, sizeof(ro));
430 	linkhdrsize = sizeof(linkhdr);
431 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
432 	if (error != 0 && error != EAFNOSUPPORT) {
433 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
434 		    if_name(ifp), error);
435 		return (error);
436 	}
437 
438 	ro.ro_prepend = linkhdr;
439 	ro.ro_plen = linkhdrsize;
440 	ro.ro_flags = 0;
441 
442 	m->m_flags |= M_BCAST;
443 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
444 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
445 	ARPSTAT_INC(txrequests);
446 	if (error) {
447 		ARPSTAT_INC(txerrors);
448 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
449 		    if_name(ifp), error);
450 	}
451 	return (error);
452 }
453 
454 void
455 arprequest(struct ifnet *ifp, const struct in_addr *sip,
456     const struct in_addr *tip, u_char *enaddr)
457 {
458 
459 	(void) arprequest_internal(ifp, sip, tip, enaddr);
460 }
461 
462 /*
463  * Resolve an IP address into an ethernet address - heavy version.
464  * Used internally by arpresolve().
465  * We have already checked that we can't use an existing lle without
466  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
467  *
468  * On success, desten and pflags are filled in and the function returns 0;
469  * If the packet must be held pending resolution, we return EWOULDBLOCK
470  * On other errors, we return the corresponding error code.
471  * Note that m_freem() handles NULL.
472  */
473 static int
474 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
475 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
476 	struct llentry **plle)
477 {
478 	struct llentry *la = NULL, *la_tmp;
479 	struct mbuf *curr = NULL;
480 	struct mbuf *next = NULL;
481 	int error, renew;
482 	char *lladdr;
483 	int ll_len;
484 
485 	NET_EPOCH_ASSERT();
486 
487 	if (pflags != NULL)
488 		*pflags = 0;
489 	if (plle != NULL)
490 		*plle = NULL;
491 
492 	if ((flags & LLE_CREATE) == 0)
493 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
494 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
495 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
496 		if (la == NULL) {
497 			char addrbuf[INET_ADDRSTRLEN];
498 
499 			log(LOG_DEBUG,
500 			    "arpresolve: can't allocate llinfo for %s on %s\n",
501 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
502 			    if_name(ifp));
503 			m_freem(m);
504 			return (EINVAL);
505 		}
506 
507 		IF_AFDATA_WLOCK(ifp);
508 		LLE_WLOCK(la);
509 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
510 		/* Prefer ANY existing lle over newly-created one */
511 		if (la_tmp == NULL)
512 			lltable_link_entry(LLTABLE(ifp), la);
513 		IF_AFDATA_WUNLOCK(ifp);
514 		if (la_tmp != NULL) {
515 			lltable_free_entry(LLTABLE(ifp), la);
516 			la = la_tmp;
517 		}
518 	}
519 	if (la == NULL) {
520 		m_freem(m);
521 		return (EINVAL);
522 	}
523 
524 	if ((la->la_flags & LLE_VALID) &&
525 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
526 		if (flags & LLE_ADDRONLY) {
527 			lladdr = la->ll_addr;
528 			ll_len = ifp->if_addrlen;
529 		} else {
530 			lladdr = la->r_linkdata;
531 			ll_len = la->r_hdrlen;
532 		}
533 		bcopy(lladdr, desten, ll_len);
534 
535 		/* Notify LLE code that the entry was used by datapath */
536 		llentry_mark_used(la);
537 		if (pflags != NULL)
538 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
539 		if (plle) {
540 			LLE_ADDREF(la);
541 			*plle = la;
542 		}
543 		LLE_WUNLOCK(la);
544 		return (0);
545 	}
546 
547 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
548 	/*
549 	 * There is an arptab entry, but no ethernet address
550 	 * response yet.  Add the mbuf to the list, dropping
551 	 * the oldest packet if we have exceeded the system
552 	 * setting.
553 	 */
554 	if (m != NULL) {
555 		if (la->la_numheld >= V_arp_maxhold) {
556 			if (la->la_hold != NULL) {
557 				next = la->la_hold->m_nextpkt;
558 				m_freem(la->la_hold);
559 				la->la_hold = next;
560 				la->la_numheld--;
561 				ARPSTAT_INC(dropped);
562 			}
563 		}
564 		if (la->la_hold != NULL) {
565 			curr = la->la_hold;
566 			while (curr->m_nextpkt != NULL)
567 				curr = curr->m_nextpkt;
568 			curr->m_nextpkt = m;
569 		} else
570 			la->la_hold = m;
571 		la->la_numheld++;
572 	}
573 	/*
574 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
575 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
576 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
577 	 * ARP request, but not faster than one request per second.
578 	 */
579 	if (la->la_asked < V_arp_maxtries)
580 		error = EWOULDBLOCK;	/* First request. */
581 	else
582 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
583 
584 	if (renew) {
585 		int canceled, e;
586 
587 		LLE_ADDREF(la);
588 		la->la_expire = time_uptime;
589 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
590 		    arptimer, la);
591 		if (canceled)
592 			LLE_REMREF(la);
593 		la->la_asked++;
594 		LLE_WUNLOCK(la);
595 		e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
596 		/*
597 		 * Only overwrite 'error' in case of error; in case of success
598 		 * the proper return value was already set above.
599 		 */
600 		if (e != 0)
601 			return (e);
602 		return (error);
603 	}
604 
605 	LLE_WUNLOCK(la);
606 	return (error);
607 }
608 
609 /*
610  * Lookups link header based on an IP address.
611  * On input:
612  *    ifp is the interface we use
613  *    is_gw != 0 if @dst represents gateway to some destination
614  *    m is the mbuf. May be NULL if we don't have a packet.
615  *    dst is the next hop,
616  *    desten is the storage to put LL header.
617  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
618  *
619  * On success, full/partial link header and flags are filled in and
620  * the function returns 0.
621  * If the packet must be held pending resolution, we return EWOULDBLOCK
622  * On other errors, we return the corresponding error code.
623  * Note that m_freem() handles NULL.
624  */
625 int
626 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
627 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
628 	struct llentry **plle)
629 {
630 	struct llentry *la = NULL;
631 
632 	NET_EPOCH_ASSERT();
633 
634 	if (pflags != NULL)
635 		*pflags = 0;
636 	if (plle != NULL)
637 		*plle = NULL;
638 
639 	if (m != NULL) {
640 		if (m->m_flags & M_BCAST) {
641 			/* broadcast */
642 			(void)memcpy(desten,
643 			    ifp->if_broadcastaddr, ifp->if_addrlen);
644 			return (0);
645 		}
646 		if (m->m_flags & M_MCAST) {
647 			/* multicast */
648 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
649 			return (0);
650 		}
651 	}
652 
653 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
654 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
655 		/* Entry found, let's copy lle info */
656 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
657 		if (pflags != NULL)
658 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
659 		/* Notify the LLE handling code that the entry was used. */
660 		llentry_mark_used(la);
661 		if (plle) {
662 			LLE_ADDREF(la);
663 			*plle = la;
664 			LLE_WUNLOCK(la);
665 		}
666 		return (0);
667 	}
668 	if (plle && la)
669 		LLE_WUNLOCK(la);
670 
671 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
672 	    desten, pflags, plle));
673 }
674 
675 /*
676  * Common length and type checks are done here,
677  * then the protocol-specific routine is called.
678  */
679 static void
680 arpintr(struct mbuf *m)
681 {
682 	struct arphdr *ar;
683 	struct ifnet *ifp;
684 	char *layer;
685 	int hlen;
686 
687 	ifp = m->m_pkthdr.rcvif;
688 
689 	if (m->m_len < sizeof(struct arphdr) &&
690 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
691 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
692 		    if_name(ifp));
693 		return;
694 	}
695 	ar = mtod(m, struct arphdr *);
696 
697 	/* Check if length is sufficient */
698 	if (m->m_len <  arphdr_len(ar)) {
699 		m = m_pullup(m, arphdr_len(ar));
700 		if (m == NULL) {
701 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
702 			    if_name(ifp));
703 			return;
704 		}
705 		ar = mtod(m, struct arphdr *);
706 	}
707 
708 	hlen = 0;
709 	layer = "";
710 	switch (ntohs(ar->ar_hrd)) {
711 	case ARPHRD_ETHER:
712 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
713 		layer = "ethernet";
714 		break;
715 	case ARPHRD_INFINIBAND:
716 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
717 		layer = "infiniband";
718 		break;
719 	case ARPHRD_IEEE1394:
720 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
721 		layer = "firewire";
722 
723 		/*
724 		 * Restrict too long hardware addresses.
725 		 * Currently we are capable of handling 20-byte
726 		 * addresses ( sizeof(lle->ll_addr) )
727 		 */
728 		if (ar->ar_hln >= 20)
729 			hlen = 16;
730 		break;
731 	default:
732 		ARP_LOG(LOG_NOTICE,
733 		    "packet with unknown hardware format 0x%02d received on "
734 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
735 		m_freem(m);
736 		return;
737 	}
738 
739 	if (hlen != 0 && hlen != ar->ar_hln) {
740 		ARP_LOG(LOG_NOTICE,
741 		    "packet with invalid %s address length %d received on %s\n",
742 		    layer, ar->ar_hln, if_name(ifp));
743 		m_freem(m);
744 		return;
745 	}
746 
747 	ARPSTAT_INC(received);
748 	switch (ntohs(ar->ar_pro)) {
749 #ifdef INET
750 	case ETHERTYPE_IP:
751 		in_arpinput(m);
752 		return;
753 #endif
754 	}
755 	m_freem(m);
756 }
757 
758 #ifdef INET
759 /*
760  * ARP for Internet protocols on 10 Mb/s Ethernet.
761  * Algorithm is that given in RFC 826.
762  * In addition, a sanity check is performed on the sender
763  * protocol address, to catch impersonators.
764  * We no longer handle negotiations for use of trailer protocol:
765  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
766  * along with IP replies if we wanted trailers sent to us,
767  * and also sent them in response to IP replies.
768  * This allowed either end to announce the desire to receive
769  * trailer packets.
770  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
771  * but formerly didn't normally send requests.
772  */
773 static int log_arp_wrong_iface = 1;
774 static int log_arp_movements = 1;
775 static int log_arp_permanent_modify = 1;
776 static int allow_multicast = 0;
777 
778 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
779 	&log_arp_wrong_iface, 0,
780 	"log arp packets arriving on the wrong interface");
781 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
782 	&log_arp_movements, 0,
783 	"log arp replies from MACs different than the one in the cache");
784 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
785 	&log_arp_permanent_modify, 0,
786 	"log arp replies from MACs different than the one in the permanent arp entry");
787 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
788 	&allow_multicast, 0, "accept multicast addresses");
789 
790 static void
791 in_arpinput(struct mbuf *m)
792 {
793 	struct rm_priotracker in_ifa_tracker;
794 	struct arphdr *ah;
795 	struct ifnet *ifp = m->m_pkthdr.rcvif;
796 	struct llentry *la = NULL, *la_tmp;
797 	struct ifaddr *ifa;
798 	struct in_ifaddr *ia;
799 	struct sockaddr sa;
800 	struct in_addr isaddr, itaddr, myaddr;
801 	u_int8_t *enaddr = NULL;
802 	int op;
803 	int bridged = 0, is_bridge = 0;
804 	int carped;
805 	struct sockaddr_in sin;
806 	struct sockaddr *dst;
807 	struct nhop4_basic nh4;
808 	uint8_t linkhdr[LLE_MAX_LINKHDR];
809 	struct route ro;
810 	size_t linkhdrsize;
811 	int lladdr_off;
812 	int error;
813 	char addrbuf[INET_ADDRSTRLEN];
814 
815 	NET_EPOCH_ASSERT();
816 
817 	sin.sin_len = sizeof(struct sockaddr_in);
818 	sin.sin_family = AF_INET;
819 	sin.sin_addr.s_addr = 0;
820 
821 	if (ifp->if_bridge)
822 		bridged = 1;
823 	if (ifp->if_type == IFT_BRIDGE)
824 		is_bridge = 1;
825 
826 	/*
827 	 * We already have checked that mbuf contains enough contiguous data
828 	 * to hold entire arp message according to the arp header.
829 	 */
830 	ah = mtod(m, struct arphdr *);
831 
832 	/*
833 	 * ARP is only for IPv4 so we can reject packets with
834 	 * a protocol length not equal to an IPv4 address.
835 	 */
836 	if (ah->ar_pln != sizeof(struct in_addr)) {
837 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
838 		    sizeof(struct in_addr));
839 		goto drop;
840 	}
841 
842 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
843 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
844 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
845 		goto drop;
846 	}
847 
848 	op = ntohs(ah->ar_op);
849 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
850 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
851 
852 	if (op == ARPOP_REPLY)
853 		ARPSTAT_INC(rxreplies);
854 
855 	/*
856 	 * For a bridge, we want to check the address irrespective
857 	 * of the receive interface. (This will change slightly
858 	 * when we have clusters of interfaces).
859 	 */
860 	IN_IFADDR_RLOCK(&in_ifa_tracker);
861 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
862 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
863 		    ia->ia_ifp == ifp) &&
864 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
865 		    (ia->ia_ifa.ifa_carp == NULL ||
866 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
867 			ifa_ref(&ia->ia_ifa);
868 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
869 			goto match;
870 		}
871 	}
872 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
873 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
874 		    ia->ia_ifp == ifp) &&
875 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
876 			ifa_ref(&ia->ia_ifa);
877 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
878 			goto match;
879 		}
880 
881 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
882   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
883   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
884   addr == ia->ia_addr.sin_addr.s_addr)
885 	/*
886 	 * Check the case when bridge shares its MAC address with
887 	 * some of its children, so packets are claimed by bridge
888 	 * itself (bridge_input() does it first), but they are really
889 	 * meant to be destined to the bridge member.
890 	 */
891 	if (is_bridge) {
892 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
893 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
894 				ifa_ref(&ia->ia_ifa);
895 				ifp = ia->ia_ifp;
896 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
897 				goto match;
898 			}
899 		}
900 	}
901 #undef BDG_MEMBER_MATCHES_ARP
902 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
903 
904 	/*
905 	 * No match, use the first inet address on the receive interface
906 	 * as a dummy address for the rest of the function.
907 	 */
908 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
909 		if (ifa->ifa_addr->sa_family == AF_INET &&
910 		    (ifa->ifa_carp == NULL ||
911 		    (*carp_iamatch_p)(ifa, &enaddr))) {
912 			ia = ifatoia(ifa);
913 			ifa_ref(ifa);
914 			goto match;
915 		}
916 
917 	/*
918 	 * If bridging, fall back to using any inet address.
919 	 */
920 	IN_IFADDR_RLOCK(&in_ifa_tracker);
921 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
922 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
923 		goto drop;
924 	}
925 	ifa_ref(&ia->ia_ifa);
926 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
927 match:
928 	if (!enaddr)
929 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
930 	carped = (ia->ia_ifa.ifa_carp != NULL);
931 	myaddr = ia->ia_addr.sin_addr;
932 	ifa_free(&ia->ia_ifa);
933 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
934 		goto drop;	/* it's from me, ignore it. */
935 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
936 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
937 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
938 		goto drop;
939 	}
940 
941 	if (ifp->if_addrlen != ah->ar_hln) {
942 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
943 		    "i/f %d (ignored)\n", ifp->if_addrlen,
944 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
945 		    ifp->if_addrlen);
946 		goto drop;
947 	}
948 
949 	/*
950 	 * Warn if another host is using the same IP address, but only if the
951 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
952 	 * case we suppress the warning to avoid false positive complaints of
953 	 * potential misconfiguration.
954 	 */
955 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
956 	    myaddr.s_addr != 0) {
957 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
958 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
959 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
960 		itaddr = myaddr;
961 		ARPSTAT_INC(dupips);
962 		goto reply;
963 	}
964 	if (ifp->if_flags & IFF_STATICARP)
965 		goto reply;
966 
967 	bzero(&sin, sizeof(sin));
968 	sin.sin_len = sizeof(struct sockaddr_in);
969 	sin.sin_family = AF_INET;
970 	sin.sin_addr = isaddr;
971 	dst = (struct sockaddr *)&sin;
972 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
973 	if (la != NULL)
974 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
975 	else if (itaddr.s_addr == myaddr.s_addr) {
976 		/*
977 		 * Request/reply to our address, but no lle exists yet.
978 		 * Calculate full link prepend to use in lle.
979 		 */
980 		linkhdrsize = sizeof(linkhdr);
981 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
982 		    &linkhdrsize, &lladdr_off) != 0)
983 			goto reply;
984 
985 		/* Allocate new entry */
986 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
987 		if (la == NULL) {
988 
989 			/*
990 			 * lle creation may fail if source address belongs
991 			 * to non-directly connected subnet. However, we
992 			 * will try to answer the request instead of dropping
993 			 * frame.
994 			 */
995 			goto reply;
996 		}
997 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
998 		    lladdr_off);
999 
1000 		IF_AFDATA_WLOCK(ifp);
1001 		LLE_WLOCK(la);
1002 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1003 
1004 		/*
1005 		 * Check if lle still does not exists.
1006 		 * If it does, that means that we either
1007 		 * 1) have configured it explicitly, via
1008 		 * 1a) 'arp -s' static entry or
1009 		 * 1b) interface address static record
1010 		 * or
1011 		 * 2) it was the result of sending first packet to-host
1012 		 * or
1013 		 * 3) it was another arp reply packet we handled in
1014 		 * different thread.
1015 		 *
1016 		 * In all cases except 3) we definitely need to prefer
1017 		 * existing lle. For the sake of simplicity, prefer any
1018 		 * existing lle over newly-create one.
1019 		 */
1020 		if (la_tmp == NULL)
1021 			lltable_link_entry(LLTABLE(ifp), la);
1022 		IF_AFDATA_WUNLOCK(ifp);
1023 
1024 		if (la_tmp == NULL) {
1025 			arp_mark_lle_reachable(la);
1026 			LLE_WUNLOCK(la);
1027 		} else {
1028 			/* Free newly-create entry and handle packet */
1029 			lltable_free_entry(LLTABLE(ifp), la);
1030 			la = la_tmp;
1031 			la_tmp = NULL;
1032 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1033 			/* arp_check_update_lle() returns @la unlocked */
1034 		}
1035 		la = NULL;
1036 	}
1037 reply:
1038 	if (op != ARPOP_REQUEST)
1039 		goto drop;
1040 	ARPSTAT_INC(rxrequests);
1041 
1042 	if (itaddr.s_addr == myaddr.s_addr) {
1043 		/* Shortcut.. the receiving interface is the target. */
1044 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1045 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1046 	} else {
1047 		/*
1048 		 * Destination address is not ours. Check if
1049 		 * proxyarp entry exists or proxyarp is turned on globally.
1050 		 */
1051 		struct llentry *lle;
1052 
1053 		sin.sin_addr = itaddr;
1054 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1055 
1056 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1057 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1058 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1059 			LLE_RUNLOCK(lle);
1060 		} else {
1061 
1062 			if (lle != NULL)
1063 				LLE_RUNLOCK(lle);
1064 
1065 			if (!V_arp_proxyall)
1066 				goto drop;
1067 
1068 			if (fib4_lookup_nh_basic(ifp->if_fib, itaddr, 0, 0,
1069 			    &nh4) != 0)
1070 				goto drop;
1071 
1072 			/*
1073 			 * Don't send proxies for nodes on the same interface
1074 			 * as this one came out of, or we'll get into a fight
1075 			 * over who claims what Ether address.
1076 			 */
1077 			if (nh4.nh_ifp == ifp)
1078 				goto drop;
1079 
1080 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1081 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1082 
1083 			/*
1084 			 * Also check that the node which sent the ARP packet
1085 			 * is on the interface we expect it to be on. This
1086 			 * avoids ARP chaos if an interface is connected to the
1087 			 * wrong network.
1088 			 */
1089 
1090 			if (fib4_lookup_nh_basic(ifp->if_fib, isaddr, 0, 0,
1091 			    &nh4) != 0)
1092 				goto drop;
1093 			if (nh4.nh_ifp != ifp) {
1094 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1095 				    " from %s via %s\n",
1096 				    inet_ntoa_r(isaddr, addrbuf),
1097 				    ifp->if_xname);
1098 				goto drop;
1099 			}
1100 
1101 #ifdef DEBUG_PROXY
1102 			printf("arp: proxying for %s\n",
1103 			    inet_ntoa_r(itaddr, addrbuf));
1104 #endif
1105 		}
1106 	}
1107 
1108 	if (itaddr.s_addr == myaddr.s_addr &&
1109 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1110 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1111 #ifdef DEBUG_LINKLOCAL
1112 		printf("arp: sending reply for link-local addr %s\n",
1113 		    inet_ntoa_r(itaddr, addrbuf));
1114 #endif
1115 		m->m_flags |= M_BCAST;
1116 		m->m_flags &= ~M_MCAST;
1117 	} else {
1118 		/* default behaviour; never reply by broadcast. */
1119 		m->m_flags &= ~(M_BCAST|M_MCAST);
1120 	}
1121 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1122 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1123 	ah->ar_op = htons(ARPOP_REPLY);
1124 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1125 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1126 	m->m_pkthdr.len = m->m_len;
1127 	m->m_pkthdr.rcvif = NULL;
1128 	sa.sa_family = AF_ARP;
1129 	sa.sa_len = 2;
1130 
1131 	/* Calculate link header for sending frame */
1132 	bzero(&ro, sizeof(ro));
1133 	linkhdrsize = sizeof(linkhdr);
1134 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1135 
1136 	/*
1137 	 * arp_fillheader() may fail due to lack of support inside encap request
1138 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1139 	 * by if_output().
1140 	 */
1141 	if (error != 0 && error != EAFNOSUPPORT) {
1142 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1143 		    if_name(ifp), error);
1144 		return;
1145 	}
1146 
1147 	ro.ro_prepend = linkhdr;
1148 	ro.ro_plen = linkhdrsize;
1149 	ro.ro_flags = 0;
1150 
1151 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1152 	(*ifp->if_output)(ifp, m, &sa, &ro);
1153 	ARPSTAT_INC(txreplies);
1154 	return;
1155 
1156 drop:
1157 	m_freem(m);
1158 }
1159 #endif
1160 
1161 /*
1162  * Checks received arp data against existing @la.
1163  * Updates lle state/performs notification if necessary.
1164  */
1165 static void
1166 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1167     int bridged, struct llentry *la)
1168 {
1169 	struct sockaddr sa;
1170 	struct mbuf *m_hold, *m_hold_next;
1171 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1172 	size_t linkhdrsize;
1173 	int lladdr_off;
1174 	char addrbuf[INET_ADDRSTRLEN];
1175 
1176 	LLE_WLOCK_ASSERT(la);
1177 
1178 	/* the following is not an error when doing bridging */
1179 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1180 		if (log_arp_wrong_iface)
1181 			ARP_LOG(LOG_WARNING, "%s is on %s "
1182 			    "but got reply from %*D on %s\n",
1183 			    inet_ntoa_r(isaddr, addrbuf),
1184 			    la->lle_tbl->llt_ifp->if_xname,
1185 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1186 			    ifp->if_xname);
1187 		LLE_WUNLOCK(la);
1188 		return;
1189 	}
1190 	if ((la->la_flags & LLE_VALID) &&
1191 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1192 		if (la->la_flags & LLE_STATIC) {
1193 			LLE_WUNLOCK(la);
1194 			if (log_arp_permanent_modify)
1195 				ARP_LOG(LOG_ERR,
1196 				    "%*D attempts to modify "
1197 				    "permanent entry for %s on %s\n",
1198 				    ifp->if_addrlen,
1199 				    (u_char *)ar_sha(ah), ":",
1200 				    inet_ntoa_r(isaddr, addrbuf),
1201 				    ifp->if_xname);
1202 			return;
1203 		}
1204 		if (log_arp_movements) {
1205 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1206 			    "to %*D on %s\n",
1207 			    inet_ntoa_r(isaddr, addrbuf),
1208 			    ifp->if_addrlen,
1209 			    (u_char *)la->ll_addr, ":",
1210 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1211 			    ifp->if_xname);
1212 		}
1213 	}
1214 
1215 	/* Calculate full link prepend to use in lle */
1216 	linkhdrsize = sizeof(linkhdr);
1217 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1218 	    &linkhdrsize, &lladdr_off) != 0)
1219 		return;
1220 
1221 	/* Check if something has changed */
1222 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1223 	    (la->la_flags & LLE_VALID) == 0) {
1224 		/* Try to perform LLE update */
1225 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1226 		    lladdr_off) == 0)
1227 			return;
1228 
1229 		/* Clear fast path feedback request if set */
1230 		la->r_skip_req = 0;
1231 	}
1232 
1233 	arp_mark_lle_reachable(la);
1234 
1235 	/*
1236 	 * The packets are all freed within the call to the output
1237 	 * routine.
1238 	 *
1239 	 * NB: The lock MUST be released before the call to the
1240 	 * output routine.
1241 	 */
1242 	if (la->la_hold != NULL) {
1243 		m_hold = la->la_hold;
1244 		la->la_hold = NULL;
1245 		la->la_numheld = 0;
1246 		lltable_fill_sa_entry(la, &sa);
1247 		LLE_WUNLOCK(la);
1248 		for (; m_hold != NULL; m_hold = m_hold_next) {
1249 			m_hold_next = m_hold->m_nextpkt;
1250 			m_hold->m_nextpkt = NULL;
1251 			/* Avoid confusing lower layers. */
1252 			m_clrprotoflags(m_hold);
1253 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1254 		}
1255 	} else
1256 		LLE_WUNLOCK(la);
1257 }
1258 
1259 static void
1260 arp_mark_lle_reachable(struct llentry *la)
1261 {
1262 	int canceled, wtime;
1263 
1264 	LLE_WLOCK_ASSERT(la);
1265 
1266 	la->ln_state = ARP_LLINFO_REACHABLE;
1267 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1268 
1269 	if (!(la->la_flags & LLE_STATIC)) {
1270 		LLE_ADDREF(la);
1271 		la->la_expire = time_uptime + V_arpt_keep;
1272 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1273 		if (wtime < 0)
1274 			wtime = V_arpt_keep;
1275 		canceled = callout_reset(&la->lle_timer,
1276 		    hz * wtime, arptimer, la);
1277 		if (canceled)
1278 			LLE_REMREF(la);
1279 	}
1280 	la->la_asked = 0;
1281 	la->la_preempt = V_arp_maxtries;
1282 }
1283 
1284 /*
1285  * Add permanent link-layer record for given interface address.
1286  */
1287 static __noinline void
1288 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1289 {
1290 	struct llentry *lle, *lle_tmp;
1291 
1292 	/*
1293 	 * Interface address LLE record is considered static
1294 	 * because kernel code relies on LLE_STATIC flag to check
1295 	 * if these entries can be rewriten by arp updates.
1296 	 */
1297 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1298 	if (lle == NULL) {
1299 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1300 		    "entry for interface address\n");
1301 		return;
1302 	}
1303 
1304 	IF_AFDATA_WLOCK(ifp);
1305 	LLE_WLOCK(lle);
1306 	/* Unlink any entry if exists */
1307 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1308 	if (lle_tmp != NULL)
1309 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1310 
1311 	lltable_link_entry(LLTABLE(ifp), lle);
1312 	IF_AFDATA_WUNLOCK(ifp);
1313 
1314 	if (lle_tmp != NULL)
1315 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1316 
1317 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1318 	LLE_WUNLOCK(lle);
1319 	if (lle_tmp != NULL)
1320 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1321 }
1322 
1323 /*
1324  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1325  * of valid values.
1326  */
1327 static int
1328 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1329 {
1330 	int error;
1331 	int rexmit_count = *(int *)arg1;
1332 
1333 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1334 
1335 	/* Enforce limits on any new value that may have been set. */
1336 	if (!error && req->newptr) {
1337 		/* A new value was set. */
1338 		if (rexmit_count < 0) {
1339 			rexmit_count = 0;
1340 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1341 			rexmit_count = MAX_GARP_RETRANSMITS;
1342 		}
1343 		*(int *)arg1 = rexmit_count;
1344 	}
1345 
1346 	return (error);
1347 }
1348 
1349 /*
1350  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1351  * retransmit it again. A pending callout owns a reference to the ifa.
1352  */
1353 static void
1354 garp_rexmit(void *arg)
1355 {
1356 	struct in_ifaddr *ia = arg;
1357 
1358 	if (callout_pending(&ia->ia_garp_timer) ||
1359 	    !callout_active(&ia->ia_garp_timer)) {
1360 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1361 		ifa_free(&ia->ia_ifa);
1362 		return;
1363 	}
1364 
1365 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1366 
1367 	/*
1368 	 * Drop lock while the ARP request is generated.
1369 	 */
1370 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1371 
1372 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1373 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1374 
1375 	/*
1376 	 * Increment the count of retransmissions. If the count has reached the
1377 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1378 	 * the callout to retransmit another GARP packet.
1379 	 */
1380 	++ia->ia_garp_count;
1381 	if (ia->ia_garp_count >= garp_rexmit_count) {
1382 		ifa_free(&ia->ia_ifa);
1383 	} else {
1384 		int rescheduled;
1385 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1386 		rescheduled = callout_reset(&ia->ia_garp_timer,
1387 		    (1 << ia->ia_garp_count) * hz,
1388 		    garp_rexmit, ia);
1389 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1390 		if (rescheduled) {
1391 			ifa_free(&ia->ia_ifa);
1392 		}
1393 	}
1394 
1395 	CURVNET_RESTORE();
1396 }
1397 
1398 /*
1399  * Start the GARP retransmit timer.
1400  *
1401  * A single GARP is always transmitted when an IPv4 address is added
1402  * to an interface and that is usually sufficient. However, in some
1403  * circumstances, such as when a shared address is passed between
1404  * cluster nodes, this single GARP may occasionally be dropped or
1405  * lost. This can lead to neighbors on the network link working with a
1406  * stale ARP cache and sending packets destined for that address to
1407  * the node that previously owned the address, which may not respond.
1408  *
1409  * To avoid this situation, GARP retransmits can be enabled by setting
1410  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1411  * than zero. The setting represents the maximum number of
1412  * retransmissions. The interval between retransmissions is calculated
1413  * using an exponential backoff algorithm, doubling each time, so the
1414  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1415  */
1416 static void
1417 garp_timer_start(struct ifaddr *ifa)
1418 {
1419 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1420 
1421 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1422 	ia->ia_garp_count = 0;
1423 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1424 	    garp_rexmit, ia) == 0) {
1425 		ifa_ref(ifa);
1426 	}
1427 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1428 }
1429 
1430 void
1431 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1432 {
1433 	struct epoch_tracker et;
1434 	const struct sockaddr_in *dst_in;
1435 	const struct sockaddr *dst;
1436 
1437 	if (ifa->ifa_carp != NULL)
1438 		return;
1439 
1440 	dst = ifa->ifa_addr;
1441 	dst_in = (const struct sockaddr_in *)dst;
1442 
1443 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1444 		return;
1445 	NET_EPOCH_ENTER(et);
1446 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1447 	NET_EPOCH_EXIT(et);
1448 	if (garp_rexmit_count > 0) {
1449 		garp_timer_start(ifa);
1450 	}
1451 
1452 	arp_add_ifa_lle(ifp, dst);
1453 }
1454 
1455 void
1456 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1457 {
1458 
1459 	if (ntohl(addr.s_addr) != INADDR_ANY)
1460 		arprequest(ifp, &addr, &addr, enaddr);
1461 }
1462 
1463 /*
1464  * Sends gratuitous ARPs for each ifaddr to notify other
1465  * nodes about the address change.
1466  */
1467 static __noinline void
1468 arp_handle_ifllchange(struct ifnet *ifp)
1469 {
1470 	struct ifaddr *ifa;
1471 
1472 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1473 		if (ifa->ifa_addr->sa_family == AF_INET)
1474 			arp_ifinit(ifp, ifa);
1475 	}
1476 }
1477 
1478 /*
1479  * A handler for interface link layer address change event.
1480  */
1481 static void
1482 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1483 {
1484 
1485 	lltable_update_ifaddr(LLTABLE(ifp));
1486 
1487 	if ((ifp->if_flags & IFF_UP) != 0)
1488 		arp_handle_ifllchange(ifp);
1489 }
1490 
1491 static void
1492 vnet_arp_init(void)
1493 {
1494 
1495 	if (IS_DEFAULT_VNET(curvnet)) {
1496 		netisr_register(&arp_nh);
1497 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1498 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1499 	}
1500 #ifdef VIMAGE
1501 	else
1502 		netisr_register_vnet(&arp_nh);
1503 #endif
1504 }
1505 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1506     vnet_arp_init, 0);
1507 
1508 #ifdef VIMAGE
1509 /*
1510  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1511  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1512  */
1513 static void
1514 vnet_arp_destroy(__unused void *arg)
1515 {
1516 
1517 	netisr_unregister_vnet(&arp_nh);
1518 }
1519 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1520     vnet_arp_destroy, NULL);
1521 #endif
1522