xref: /freebsd/sys/netinet/if_ether.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
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/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/route/nhop.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) = 16;
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 static void	arpintr(struct mbuf *);
182 static void	arptimer(void *);
183 #ifdef INET
184 static void	in_arpinput(struct mbuf *);
185 #endif
186 
187 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
188     struct ifnet *ifp, int bridged, struct llentry *la);
189 static void arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp);
190 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
191 
192 static eventhandler_tag iflladdr_tag;
193 
194 static const struct netisr_handler arp_nh = {
195 	.nh_name = "arp",
196 	.nh_handler = arpintr,
197 	.nh_proto = NETISR_ARP,
198 	.nh_policy = NETISR_POLICY_SOURCE,
199 };
200 
201 /*
202  * Timeout routine.  Age arp_tab entries periodically.
203  */
204 static void
205 arptimer(void *arg)
206 {
207 	struct llentry *lle = (struct llentry *)arg;
208 	struct ifnet *ifp;
209 
210 	if (lle->la_flags & LLE_STATIC) {
211 		return;
212 	}
213 	LLE_WLOCK(lle);
214 	if (callout_pending(&lle->lle_timer)) {
215 		/*
216 		 * Here we are a bit odd here in the treatment of
217 		 * active/pending. If the pending bit is set, it got
218 		 * rescheduled before I ran. The active
219 		 * bit we ignore, since if it was stopped
220 		 * in ll_tablefree() and was currently running
221 		 * it would have return 0 so the code would
222 		 * not have deleted it since the callout could
223 		 * not be stopped so we want to go through
224 		 * with the delete here now. If the callout
225 		 * was restarted, the pending bit will be back on and
226 		 * we just want to bail since the callout_reset would
227 		 * return 1 and our reference would have been removed
228 		 * by arpresolve() below.
229 		 */
230 		LLE_WUNLOCK(lle);
231  		return;
232  	}
233 	ifp = lle->lle_tbl->llt_ifp;
234 	CURVNET_SET(ifp->if_vnet);
235 
236 	switch (lle->ln_state) {
237 	case ARP_LLINFO_REACHABLE:
238 
239 		/*
240 		 * Expiration time is approaching.
241 		 * Request usage feedback from the datapath.
242 		 * Change state and re-schedule ourselves.
243 		 */
244 		llentry_request_feedback(lle);
245 		lle->ln_state = ARP_LLINFO_VERIFY;
246 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
247 		LLE_WUNLOCK(lle);
248 		CURVNET_RESTORE();
249 		return;
250 	case ARP_LLINFO_VERIFY:
251 		if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
252 			/* Entry was used, issue refresh request */
253 			struct epoch_tracker et;
254 			struct in_addr dst;
255 
256 			dst = lle->r_l3addr.addr4;
257 			lle->la_preempt--;
258 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
259 			LLE_WUNLOCK(lle);
260 			NET_EPOCH_ENTER(et);
261 			arprequest(ifp, NULL, &dst, NULL);
262 			NET_EPOCH_EXIT(et);
263 			CURVNET_RESTORE();
264 			return;
265 		}
266 		/* Nothing happened. Reschedule if not too late */
267 		if (lle->la_expire > time_uptime) {
268 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
269 			LLE_WUNLOCK(lle);
270 			CURVNET_RESTORE();
271 			return;
272 		}
273 		break;
274 	case ARP_LLINFO_INCOMPLETE:
275 	case ARP_LLINFO_DELETED:
276 		break;
277 	}
278 
279 	if ((lle->la_flags & LLE_DELETED) == 0) {
280 		int evt;
281 
282 		if (lle->la_flags & LLE_VALID)
283 			evt = LLENTRY_EXPIRED;
284 		else
285 			evt = LLENTRY_TIMEDOUT;
286 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
287 	}
288 
289 	callout_stop(&lle->lle_timer);
290 
291 	/* XXX: LOR avoidance. We still have ref on lle. */
292 	LLE_WUNLOCK(lle);
293 	IF_AFDATA_LOCK(ifp);
294 	LLE_WLOCK(lle);
295 
296 	/* Guard against race with other llentry_free(). */
297 	if (lle->la_flags & LLE_LINKED) {
298 		LLE_REMREF(lle);
299 		lltable_unlink_entry(lle->lle_tbl, lle);
300 	}
301 	IF_AFDATA_UNLOCK(ifp);
302 
303 	size_t pkts_dropped = llentry_free(lle);
304 
305 	ARPSTAT_ADD(dropped, pkts_dropped);
306 	ARPSTAT_INC(timeouts);
307 
308 	CURVNET_RESTORE();
309 }
310 
311 /*
312  * Stores link-layer header for @ifp in format suitable for if_output()
313  * into buffer @buf. Resulting header length is stored in @bufsize.
314  *
315  * Returns 0 on success.
316  */
317 static int
318 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
319     size_t *bufsize)
320 {
321 	struct if_encap_req ereq;
322 	int error;
323 
324 	bzero(buf, *bufsize);
325 	bzero(&ereq, sizeof(ereq));
326 	ereq.buf = buf;
327 	ereq.bufsize = *bufsize;
328 	ereq.rtype = IFENCAP_LL;
329 	ereq.family = AF_ARP;
330 	ereq.lladdr = ar_tha(ah);
331 	ereq.hdata = (u_char *)ah;
332 	if (bcast)
333 		ereq.flags = IFENCAP_FLAG_BROADCAST;
334 	error = ifp->if_requestencap(ifp, &ereq);
335 	if (error == 0)
336 		*bufsize = ereq.bufsize;
337 
338 	return (error);
339 }
340 
341 /*
342  * Broadcast an ARP request. Caller specifies:
343  *	- arp header source ip address
344  *	- arp header target ip address
345  *	- arp header source ethernet address
346  */
347 static int
348 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
349     const struct in_addr *tip, u_char *enaddr)
350 {
351 	struct mbuf *m;
352 	struct arphdr *ah;
353 	struct sockaddr sa;
354 	u_char *carpaddr = NULL;
355 	uint8_t linkhdr[LLE_MAX_LINKHDR];
356 	size_t linkhdrsize;
357 	struct route ro;
358 	int error;
359 
360 	NET_EPOCH_ASSERT();
361 
362 	if (sip == NULL) {
363 		/*
364 		 * The caller did not supply a source address, try to find
365 		 * a compatible one among those assigned to this interface.
366 		 */
367 		struct ifaddr *ifa;
368 
369 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
370 			if (ifa->ifa_addr->sa_family != AF_INET)
371 				continue;
372 
373 			if (ifa->ifa_carp) {
374 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
375 					continue;
376 				sip = &IA_SIN(ifa)->sin_addr;
377 			} else {
378 				carpaddr = NULL;
379 				sip = &IA_SIN(ifa)->sin_addr;
380 			}
381 
382 			if (0 == ((sip->s_addr ^ tip->s_addr) &
383 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
384 				break;  /* found it. */
385 		}
386 		if (sip == NULL) {
387 			printf("%s: cannot find matching address\n", __func__);
388 			return (EADDRNOTAVAIL);
389 		}
390 	}
391 	if (enaddr == NULL)
392 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
393 
394 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
395 		return (ENOMEM);
396 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
397 		2 * ifp->if_addrlen;
398 	m->m_pkthdr.len = m->m_len;
399 	M_ALIGN(m, m->m_len);
400 	ah = mtod(m, struct arphdr *);
401 	bzero((caddr_t)ah, m->m_len);
402 #ifdef MAC
403 	mac_netinet_arp_send(ifp, m);
404 #endif
405 	ah->ar_pro = htons(ETHERTYPE_IP);
406 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
407 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
408 	ah->ar_op = htons(ARPOP_REQUEST);
409 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
410 	bcopy(sip, ar_spa(ah), ah->ar_pln);
411 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
412 	sa.sa_family = AF_ARP;
413 	sa.sa_len = 2;
414 
415 	/* Calculate link header for sending frame */
416 	bzero(&ro, sizeof(ro));
417 	linkhdrsize = sizeof(linkhdr);
418 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
419 	if (error != 0 && error != EAFNOSUPPORT) {
420 		m_freem(m);
421 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
422 		    if_name(ifp), error);
423 		return (error);
424 	}
425 
426 	ro.ro_prepend = linkhdr;
427 	ro.ro_plen = linkhdrsize;
428 	ro.ro_flags = 0;
429 
430 	m->m_flags |= M_BCAST;
431 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
432 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
433 	ARPSTAT_INC(txrequests);
434 	if (error) {
435 		ARPSTAT_INC(txerrors);
436 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
437 		    if_name(ifp), error);
438 	}
439 	return (error);
440 }
441 
442 void
443 arprequest(struct ifnet *ifp, const struct in_addr *sip,
444     const struct in_addr *tip, u_char *enaddr)
445 {
446 
447 	(void) arprequest_internal(ifp, sip, tip, enaddr);
448 }
449 
450 /*
451  * Resolve an IP address into an ethernet address - heavy version.
452  * Used internally by arpresolve().
453  * We have already checked that we can't use an existing lle without
454  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
455  *
456  * On success, desten and pflags are filled in and the function returns 0;
457  * If the packet must be held pending resolution, we return EWOULDBLOCK
458  * On other errors, we return the corresponding error code.
459  * Note that m_freem() handles NULL.
460  */
461 static int
462 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
463 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
464 	struct llentry **plle)
465 {
466 	struct llentry *la = NULL, *la_tmp;
467 	int error, renew;
468 	char *lladdr;
469 	int ll_len;
470 
471 	NET_EPOCH_ASSERT();
472 
473 	if (pflags != NULL)
474 		*pflags = 0;
475 	if (plle != NULL)
476 		*plle = NULL;
477 
478 	if ((flags & LLE_CREATE) == 0)
479 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
480 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
481 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
482 		if (la == NULL) {
483 			char addrbuf[INET_ADDRSTRLEN];
484 
485 			log(LOG_DEBUG,
486 			    "arpresolve: can't allocate llinfo for %s on %s\n",
487 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
488 			    if_name(ifp));
489 			m_freem(m);
490 			return (EINVAL);
491 		}
492 
493 		IF_AFDATA_WLOCK(ifp);
494 		LLE_WLOCK(la);
495 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
496 		/* Prefer ANY existing lle over newly-created one */
497 		if (la_tmp == NULL)
498 			lltable_link_entry(LLTABLE(ifp), la);
499 		IF_AFDATA_WUNLOCK(ifp);
500 		if (la_tmp != NULL) {
501 			lltable_free_entry(LLTABLE(ifp), la);
502 			la = la_tmp;
503 		}
504 	}
505 	if (la == NULL) {
506 		m_freem(m);
507 		return (EINVAL);
508 	}
509 
510 	if ((la->la_flags & LLE_VALID) &&
511 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
512 		if (flags & LLE_ADDRONLY) {
513 			lladdr = la->ll_addr;
514 			ll_len = ifp->if_addrlen;
515 		} else {
516 			lladdr = la->r_linkdata;
517 			ll_len = la->r_hdrlen;
518 		}
519 		bcopy(lladdr, desten, ll_len);
520 
521 		/* Notify LLE code that the entry was used by datapath */
522 		llentry_provide_feedback(la);
523 		if (pflags != NULL)
524 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
525 		if (plle) {
526 			LLE_ADDREF(la);
527 			*plle = la;
528 		}
529 		LLE_WUNLOCK(la);
530 		return (0);
531 	}
532 
533 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
534 
535 	/*
536 	 * There is an arptab entry, but no ethernet address
537 	 * response yet.  Add the mbuf to the list, dropping
538 	 * the oldest packet if we have exceeded the system
539 	 * setting.
540 	 */
541 	if (m != NULL) {
542 		size_t dropped = lltable_append_entry_queue(la, m, V_arp_maxhold);
543 		ARPSTAT_ADD(dropped, dropped);
544 	}
545 
546 	/*
547 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
548 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
549 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
550 	 * ARP request, but not faster than one request per second.
551 	 */
552 	if (la->la_asked < V_arp_maxtries)
553 		error = EWOULDBLOCK;	/* First request. */
554 	else
555 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
556 
557 	if (renew) {
558 		int canceled, e;
559 
560 		LLE_ADDREF(la);
561 		la->la_expire = time_uptime;
562 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
563 		    arptimer, la);
564 		if (canceled)
565 			LLE_REMREF(la);
566 		la->la_asked++;
567 		LLE_WUNLOCK(la);
568 		e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
569 		/*
570 		 * Only overwrite 'error' in case of error; in case of success
571 		 * the proper return value was already set above.
572 		 */
573 		if (e != 0)
574 			return (e);
575 		return (error);
576 	}
577 
578 	LLE_WUNLOCK(la);
579 	return (error);
580 }
581 
582 /*
583  * Lookups link header based on an IP address.
584  * On input:
585  *    ifp is the interface we use
586  *    is_gw != 0 if @dst represents gateway to some destination
587  *    m is the mbuf. May be NULL if we don't have a packet.
588  *    dst is the next hop,
589  *    desten is the storage to put LL header.
590  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
591  *
592  * On success, full/partial link header and flags are filled in and
593  * the function returns 0.
594  * If the packet must be held pending resolution, we return EWOULDBLOCK
595  * On other errors, we return the corresponding error code.
596  * Note that m_freem() handles NULL.
597  */
598 int
599 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
600 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
601 	struct llentry **plle)
602 {
603 	struct llentry *la = NULL;
604 
605 	NET_EPOCH_ASSERT();
606 
607 	if (pflags != NULL)
608 		*pflags = 0;
609 	if (plle != NULL)
610 		*plle = NULL;
611 
612 	if (m != NULL) {
613 		if (m->m_flags & M_BCAST) {
614 			/* broadcast */
615 			(void)memcpy(desten,
616 			    ifp->if_broadcastaddr, ifp->if_addrlen);
617 			return (0);
618 		}
619 		if (m->m_flags & M_MCAST) {
620 			/* multicast */
621 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
622 			return (0);
623 		}
624 	}
625 
626 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
627 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
628 		/* Entry found, let's copy lle info */
629 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
630 		if (pflags != NULL)
631 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
632 		/* Notify the LLE handling code that the entry was used. */
633 		llentry_provide_feedback(la);
634 		if (plle) {
635 			LLE_ADDREF(la);
636 			*plle = la;
637 			LLE_WUNLOCK(la);
638 		}
639 		return (0);
640 	}
641 	if (plle && la)
642 		LLE_WUNLOCK(la);
643 
644 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
645 	    desten, pflags, plle));
646 }
647 
648 /*
649  * Common length and type checks are done here,
650  * then the protocol-specific routine is called.
651  */
652 static void
653 arpintr(struct mbuf *m)
654 {
655 	struct arphdr *ar;
656 	struct ifnet *ifp;
657 	char *layer;
658 	int hlen;
659 
660 	ifp = m->m_pkthdr.rcvif;
661 
662 	if (m->m_len < sizeof(struct arphdr) &&
663 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
664 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
665 		    if_name(ifp));
666 		return;
667 	}
668 	ar = mtod(m, struct arphdr *);
669 
670 	/* Check if length is sufficient */
671 	if (m->m_len <  arphdr_len(ar)) {
672 		m = m_pullup(m, arphdr_len(ar));
673 		if (m == NULL) {
674 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
675 			    if_name(ifp));
676 			return;
677 		}
678 		ar = mtod(m, struct arphdr *);
679 	}
680 
681 	hlen = 0;
682 	layer = "";
683 	switch (ntohs(ar->ar_hrd)) {
684 	case ARPHRD_ETHER:
685 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
686 		layer = "ethernet";
687 		break;
688 	case ARPHRD_INFINIBAND:
689 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
690 		layer = "infiniband";
691 		break;
692 	case ARPHRD_IEEE1394:
693 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
694 		layer = "firewire";
695 
696 		/*
697 		 * Restrict too long hardware addresses.
698 		 * Currently we are capable of handling 20-byte
699 		 * addresses ( sizeof(lle->ll_addr) )
700 		 */
701 		if (ar->ar_hln >= 20)
702 			hlen = 16;
703 		break;
704 	default:
705 		ARP_LOG(LOG_NOTICE,
706 		    "packet with unknown hardware format 0x%02d received on "
707 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
708 		m_freem(m);
709 		return;
710 	}
711 
712 	if (hlen != 0 && hlen != ar->ar_hln) {
713 		ARP_LOG(LOG_NOTICE,
714 		    "packet with invalid %s address length %d received on %s\n",
715 		    layer, ar->ar_hln, if_name(ifp));
716 		m_freem(m);
717 		return;
718 	}
719 
720 	ARPSTAT_INC(received);
721 	switch (ntohs(ar->ar_pro)) {
722 #ifdef INET
723 	case ETHERTYPE_IP:
724 		in_arpinput(m);
725 		return;
726 #endif
727 	}
728 	m_freem(m);
729 }
730 
731 #ifdef INET
732 /*
733  * ARP for Internet protocols on 10 Mb/s Ethernet.
734  * Algorithm is that given in RFC 826.
735  * In addition, a sanity check is performed on the sender
736  * protocol address, to catch impersonators.
737  * We no longer handle negotiations for use of trailer protocol:
738  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
739  * along with IP replies if we wanted trailers sent to us,
740  * and also sent them in response to IP replies.
741  * This allowed either end to announce the desire to receive
742  * trailer packets.
743  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
744  * but formerly didn't normally send requests.
745  */
746 static int log_arp_wrong_iface = 1;
747 static int log_arp_movements = 1;
748 static int log_arp_permanent_modify = 1;
749 static int allow_multicast = 0;
750 
751 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
752 	&log_arp_wrong_iface, 0,
753 	"log arp packets arriving on the wrong interface");
754 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
755 	&log_arp_movements, 0,
756 	"log arp replies from MACs different than the one in the cache");
757 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
758 	&log_arp_permanent_modify, 0,
759 	"log arp replies from MACs different than the one in the permanent arp entry");
760 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
761 	&allow_multicast, 0, "accept multicast addresses");
762 
763 static void
764 in_arpinput(struct mbuf *m)
765 {
766 	struct arphdr *ah;
767 	struct ifnet *ifp = m->m_pkthdr.rcvif;
768 	struct llentry *la = NULL, *la_tmp;
769 	struct ifaddr *ifa;
770 	struct in_ifaddr *ia;
771 	struct sockaddr sa;
772 	struct in_addr isaddr, itaddr, myaddr;
773 	u_int8_t *enaddr = NULL;
774 	int op;
775 	int bridged = 0, is_bridge = 0;
776 	int carped;
777 	struct sockaddr_in sin;
778 	struct sockaddr *dst;
779 	struct nhop_object *nh;
780 	uint8_t linkhdr[LLE_MAX_LINKHDR];
781 	struct route ro;
782 	size_t linkhdrsize;
783 	int lladdr_off;
784 	int error;
785 	char addrbuf[INET_ADDRSTRLEN];
786 
787 	NET_EPOCH_ASSERT();
788 
789 	sin.sin_len = sizeof(struct sockaddr_in);
790 	sin.sin_family = AF_INET;
791 	sin.sin_addr.s_addr = 0;
792 
793 	if (ifp->if_bridge)
794 		bridged = 1;
795 	if (ifp->if_type == IFT_BRIDGE)
796 		is_bridge = 1;
797 
798 	/*
799 	 * We already have checked that mbuf contains enough contiguous data
800 	 * to hold entire arp message according to the arp header.
801 	 */
802 	ah = mtod(m, struct arphdr *);
803 
804 	/*
805 	 * ARP is only for IPv4 so we can reject packets with
806 	 * a protocol length not equal to an IPv4 address.
807 	 */
808 	if (ah->ar_pln != sizeof(struct in_addr)) {
809 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
810 		    sizeof(struct in_addr));
811 		goto drop;
812 	}
813 
814 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
815 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
816 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
817 		goto drop;
818 	}
819 
820 	op = ntohs(ah->ar_op);
821 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
822 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
823 
824 	if (op == ARPOP_REPLY)
825 		ARPSTAT_INC(rxreplies);
826 
827 	/*
828 	 * For a bridge, we want to check the address irrespective
829 	 * of the receive interface. (This will change slightly
830 	 * when we have clusters of interfaces).
831 	 */
832 	CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
833 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
834 		    ia->ia_ifp == ifp) &&
835 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
836 		    (ia->ia_ifa.ifa_carp == NULL ||
837 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
838 			ifa_ref(&ia->ia_ifa);
839 			goto match;
840 		}
841 	}
842 	CK_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 			goto match;
848 		}
849 
850 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
851   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
852   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
853   addr == ia->ia_addr.sin_addr.s_addr)
854 	/*
855 	 * Check the case when bridge shares its MAC address with
856 	 * some of its children, so packets are claimed by bridge
857 	 * itself (bridge_input() does it first), but they are really
858 	 * meant to be destined to the bridge member.
859 	 */
860 	if (is_bridge) {
861 		CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
862 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
863 				ifa_ref(&ia->ia_ifa);
864 				ifp = ia->ia_ifp;
865 				goto match;
866 			}
867 		}
868 	}
869 #undef BDG_MEMBER_MATCHES_ARP
870 
871 	/*
872 	 * No match, use the first inet address on the receive interface
873 	 * as a dummy address for the rest of the function.
874 	 */
875 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
876 		if (ifa->ifa_addr->sa_family == AF_INET &&
877 		    (ifa->ifa_carp == NULL ||
878 		    (*carp_iamatch_p)(ifa, &enaddr))) {
879 			ia = ifatoia(ifa);
880 			ifa_ref(ifa);
881 			goto match;
882 		}
883 
884 	/*
885 	 * If bridging, fall back to using any inet address.
886 	 */
887 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL)
888 		goto drop;
889 	ifa_ref(&ia->ia_ifa);
890 match:
891 	if (!enaddr)
892 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
893 	carped = (ia->ia_ifa.ifa_carp != NULL);
894 	myaddr = ia->ia_addr.sin_addr;
895 	ifa_free(&ia->ia_ifa);
896 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
897 		goto drop;	/* it's from me, ignore it. */
898 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
899 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
900 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
901 		goto drop;
902 	}
903 
904 	if (ifp->if_addrlen != ah->ar_hln) {
905 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
906 		    "i/f %d (ignored)\n", ifp->if_addrlen,
907 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
908 		    ifp->if_addrlen);
909 		goto drop;
910 	}
911 
912 	/*
913 	 * Warn if another host is using the same IP address, but only if the
914 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
915 	 * case we suppress the warning to avoid false positive complaints of
916 	 * potential misconfiguration.
917 	 */
918 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
919 	    myaddr.s_addr != 0) {
920 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
921 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
922 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
923 		itaddr = myaddr;
924 		ARPSTAT_INC(dupips);
925 		goto reply;
926 	}
927 	if (ifp->if_flags & IFF_STATICARP)
928 		goto reply;
929 
930 	bzero(&sin, sizeof(sin));
931 	sin.sin_len = sizeof(struct sockaddr_in);
932 	sin.sin_family = AF_INET;
933 	sin.sin_addr = isaddr;
934 	dst = (struct sockaddr *)&sin;
935 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
936 	if (la != NULL)
937 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
938 	else if (itaddr.s_addr == myaddr.s_addr) {
939 		/*
940 		 * Request/reply to our address, but no lle exists yet.
941 		 * Calculate full link prepend to use in lle.
942 		 */
943 		linkhdrsize = sizeof(linkhdr);
944 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
945 		    &linkhdrsize, &lladdr_off) != 0)
946 			goto reply;
947 
948 		/* Allocate new entry */
949 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
950 		if (la == NULL) {
951 			/*
952 			 * lle creation may fail if source address belongs
953 			 * to non-directly connected subnet. However, we
954 			 * will try to answer the request instead of dropping
955 			 * frame.
956 			 */
957 			goto reply;
958 		}
959 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
960 		    lladdr_off);
961 
962 		IF_AFDATA_WLOCK(ifp);
963 		LLE_WLOCK(la);
964 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
965 
966 		/*
967 		 * Check if lle still does not exists.
968 		 * If it does, that means that we either
969 		 * 1) have configured it explicitly, via
970 		 * 1a) 'arp -s' static entry or
971 		 * 1b) interface address static record
972 		 * or
973 		 * 2) it was the result of sending first packet to-host
974 		 * or
975 		 * 3) it was another arp reply packet we handled in
976 		 * different thread.
977 		 *
978 		 * In all cases except 3) we definitely need to prefer
979 		 * existing lle. For the sake of simplicity, prefer any
980 		 * existing lle over newly-create one.
981 		 */
982 		if (la_tmp == NULL)
983 			lltable_link_entry(LLTABLE(ifp), la);
984 		IF_AFDATA_WUNLOCK(ifp);
985 
986 		if (la_tmp == NULL) {
987 			arp_mark_lle_reachable(la, ifp);
988 			LLE_WUNLOCK(la);
989 		} else {
990 			/* Free newly-create entry and handle packet */
991 			lltable_free_entry(LLTABLE(ifp), la);
992 			la = la_tmp;
993 			la_tmp = NULL;
994 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
995 			/* arp_check_update_lle() returns @la unlocked */
996 		}
997 		la = NULL;
998 	}
999 reply:
1000 	if (op != ARPOP_REQUEST)
1001 		goto drop;
1002 	ARPSTAT_INC(rxrequests);
1003 
1004 	if (itaddr.s_addr == myaddr.s_addr) {
1005 		/* Shortcut.. the receiving interface is the target. */
1006 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1007 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1008 	} else {
1009 		/*
1010 		 * Destination address is not ours. Check if
1011 		 * proxyarp entry exists or proxyarp is turned on globally.
1012 		 */
1013 		struct llentry *lle;
1014 
1015 		sin.sin_addr = itaddr;
1016 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1017 
1018 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1019 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1020 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1021 			LLE_RUNLOCK(lle);
1022 		} else {
1023 			if (lle != NULL)
1024 				LLE_RUNLOCK(lle);
1025 
1026 			if (!V_arp_proxyall)
1027 				goto drop;
1028 
1029 			NET_EPOCH_ASSERT();
1030 			nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1031 			if (nh == NULL)
1032 				goto drop;
1033 
1034 			/*
1035 			 * Don't send proxies for nodes on the same interface
1036 			 * as this one came out of, or we'll get into a fight
1037 			 * over who claims what Ether address.
1038 			 */
1039 			if (nh->nh_ifp == ifp)
1040 				goto drop;
1041 
1042 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1043 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1044 
1045 			/*
1046 			 * Also check that the node which sent the ARP packet
1047 			 * is on the interface we expect it to be on. This
1048 			 * avoids ARP chaos if an interface is connected to the
1049 			 * wrong network.
1050 			 */
1051 
1052 			nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1053 			if (nh == NULL)
1054 				goto drop;
1055 			if (nh->nh_ifp != ifp) {
1056 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1057 				    " from %s via %s\n",
1058 				    inet_ntoa_r(isaddr, addrbuf),
1059 				    ifp->if_xname);
1060 				goto drop;
1061 			}
1062 
1063 #ifdef DEBUG_PROXY
1064 			printf("arp: proxying for %s\n",
1065 			    inet_ntoa_r(itaddr, addrbuf));
1066 #endif
1067 		}
1068 	}
1069 
1070 	if (itaddr.s_addr == myaddr.s_addr &&
1071 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1072 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1073 #ifdef DEBUG_LINKLOCAL
1074 		printf("arp: sending reply for link-local addr %s\n",
1075 		    inet_ntoa_r(itaddr, addrbuf));
1076 #endif
1077 		m->m_flags |= M_BCAST;
1078 		m->m_flags &= ~M_MCAST;
1079 	} else {
1080 		/* default behaviour; never reply by broadcast. */
1081 		m->m_flags &= ~(M_BCAST|M_MCAST);
1082 	}
1083 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1084 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1085 	ah->ar_op = htons(ARPOP_REPLY);
1086 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1087 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1088 	m->m_pkthdr.len = m->m_len;
1089 	m->m_pkthdr.rcvif = NULL;
1090 	sa.sa_family = AF_ARP;
1091 	sa.sa_len = 2;
1092 
1093 	/* Calculate link header for sending frame */
1094 	bzero(&ro, sizeof(ro));
1095 	linkhdrsize = sizeof(linkhdr);
1096 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1097 
1098 	/*
1099 	 * arp_fillheader() may fail due to lack of support inside encap request
1100 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1101 	 * by if_output().
1102 	 */
1103 	if (error != 0 && error != EAFNOSUPPORT) {
1104 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1105 		    if_name(ifp), error);
1106 		goto drop;
1107 	}
1108 
1109 	ro.ro_prepend = linkhdr;
1110 	ro.ro_plen = linkhdrsize;
1111 	ro.ro_flags = 0;
1112 
1113 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1114 	(*ifp->if_output)(ifp, m, &sa, &ro);
1115 	ARPSTAT_INC(txreplies);
1116 	return;
1117 
1118 drop:
1119 	m_freem(m);
1120 }
1121 #endif
1122 
1123 static struct mbuf *
1124 arp_grab_holdchain(struct llentry *la)
1125 {
1126 	struct mbuf *chain;
1127 
1128 	LLE_WLOCK_ASSERT(la);
1129 
1130 	chain = la->la_hold;
1131 	la->la_hold = NULL;
1132 	la->la_numheld = 0;
1133 
1134 	return (chain);
1135 }
1136 
1137 static void
1138 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1139 {
1140 	struct mbuf *m_hold, *m_hold_next;
1141 	struct sockaddr_in sin;
1142 
1143 	NET_EPOCH_ASSERT();
1144 
1145 	struct route ro = {
1146 		.ro_prepend = la->r_linkdata,
1147 		.ro_plen = la->r_hdrlen,
1148 	};
1149 
1150 	lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1151 
1152 	for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1153 		m_hold_next = m_hold->m_nextpkt;
1154 		m_hold->m_nextpkt = NULL;
1155 		/* Avoid confusing lower layers. */
1156 		m_clrprotoflags(m_hold);
1157 		(*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1158 	}
1159 }
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 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1170 	size_t linkhdrsize;
1171 	int lladdr_off;
1172 	char addrbuf[INET_ADDRSTRLEN];
1173 
1174 	LLE_WLOCK_ASSERT(la);
1175 
1176 	/* the following is not an error when doing bridging */
1177 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1178 		if (log_arp_wrong_iface)
1179 			ARP_LOG(LOG_WARNING, "%s is on %s "
1180 			    "but got reply from %*D on %s\n",
1181 			    inet_ntoa_r(isaddr, addrbuf),
1182 			    la->lle_tbl->llt_ifp->if_xname,
1183 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1184 			    ifp->if_xname);
1185 		LLE_WUNLOCK(la);
1186 		return;
1187 	}
1188 	if ((la->la_flags & LLE_VALID) &&
1189 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1190 		if (la->la_flags & LLE_STATIC) {
1191 			LLE_WUNLOCK(la);
1192 			if (log_arp_permanent_modify)
1193 				ARP_LOG(LOG_ERR,
1194 				    "%*D attempts to modify "
1195 				    "permanent entry for %s on %s\n",
1196 				    ifp->if_addrlen,
1197 				    (u_char *)ar_sha(ah), ":",
1198 				    inet_ntoa_r(isaddr, addrbuf),
1199 				    ifp->if_xname);
1200 			return;
1201 		}
1202 		if (log_arp_movements) {
1203 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1204 			    "to %*D on %s\n",
1205 			    inet_ntoa_r(isaddr, addrbuf),
1206 			    ifp->if_addrlen,
1207 			    (u_char *)la->ll_addr, ":",
1208 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1209 			    ifp->if_xname);
1210 		}
1211 	}
1212 
1213 	/* Calculate full link prepend to use in lle */
1214 	linkhdrsize = sizeof(linkhdr);
1215 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1216 	    &linkhdrsize, &lladdr_off) != 0) {
1217 		LLE_WUNLOCK(la);
1218 		return;
1219 	}
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 			LLE_WUNLOCK(la);
1228 			return;
1229 		}
1230 
1231 		/* Clear fast path feedback request if set */
1232 		llentry_mark_used(la);
1233 	}
1234 
1235 	arp_mark_lle_reachable(la, ifp);
1236 
1237 	/*
1238 	 * The packets are all freed within the call to the output
1239 	 * routine.
1240 	 *
1241 	 * NB: The lock MUST be released before the call to the
1242 	 * output routine.
1243 	 */
1244 	if (la->la_hold != NULL) {
1245 		struct mbuf *chain;
1246 
1247 		chain = arp_grab_holdchain(la);
1248 		LLE_WUNLOCK(la);
1249 		arp_flush_holdchain(ifp, la, chain);
1250 	} else
1251 		LLE_WUNLOCK(la);
1252 }
1253 
1254 static void
1255 arp_mark_lle_reachable(struct llentry *la, struct ifnet *ifp)
1256 {
1257 	int canceled, wtime;
1258 
1259 	LLE_WLOCK_ASSERT(la);
1260 
1261 	la->ln_state = ARP_LLINFO_REACHABLE;
1262 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1263 
1264 	if ((ifp->if_flags & IFF_STICKYARP) != 0)
1265 		la->la_flags |= LLE_STATIC;
1266 
1267 	if (!(la->la_flags & LLE_STATIC)) {
1268 		LLE_ADDREF(la);
1269 		la->la_expire = time_uptime + V_arpt_keep;
1270 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1271 		if (wtime < 0)
1272 			wtime = V_arpt_keep;
1273 		canceled = callout_reset(&la->lle_timer,
1274 		    hz * wtime, arptimer, la);
1275 		if (canceled)
1276 			LLE_REMREF(la);
1277 	}
1278 	la->la_asked = 0;
1279 	la->la_preempt = V_arp_maxtries;
1280 }
1281 
1282 /*
1283  * Add permanent link-layer record for given interface address.
1284  */
1285 static __noinline void
1286 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1287 {
1288 	struct llentry *lle, *lle_tmp;
1289 
1290 	/*
1291 	 * Interface address LLE record is considered static
1292 	 * because kernel code relies on LLE_STATIC flag to check
1293 	 * if these entries can be rewriten by arp updates.
1294 	 */
1295 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1296 	if (lle == NULL) {
1297 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1298 		    "entry for interface address\n");
1299 		return;
1300 	}
1301 
1302 	IF_AFDATA_WLOCK(ifp);
1303 	LLE_WLOCK(lle);
1304 	/* Unlink any entry if exists */
1305 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1306 	if (lle_tmp != NULL)
1307 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1308 
1309 	lltable_link_entry(LLTABLE(ifp), lle);
1310 	IF_AFDATA_WUNLOCK(ifp);
1311 
1312 	if (lle_tmp != NULL)
1313 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1314 
1315 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1316 	LLE_WUNLOCK(lle);
1317 	if (lle_tmp != NULL)
1318 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1319 }
1320 
1321 /*
1322  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1323  * of valid values.
1324  */
1325 static int
1326 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1327 {
1328 	int error;
1329 	int rexmit_count = *(int *)arg1;
1330 
1331 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1332 
1333 	/* Enforce limits on any new value that may have been set. */
1334 	if (!error && req->newptr) {
1335 		/* A new value was set. */
1336 		if (rexmit_count < 0) {
1337 			rexmit_count = 0;
1338 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1339 			rexmit_count = MAX_GARP_RETRANSMITS;
1340 		}
1341 		*(int *)arg1 = rexmit_count;
1342 	}
1343 
1344 	return (error);
1345 }
1346 
1347 /*
1348  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1349  * retransmit it again. A pending callout owns a reference to the ifa.
1350  */
1351 static void
1352 garp_rexmit(void *arg)
1353 {
1354 	struct in_ifaddr *ia = arg;
1355 
1356 	if (callout_pending(&ia->ia_garp_timer) ||
1357 	    !callout_active(&ia->ia_garp_timer)) {
1358 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1359 		ifa_free(&ia->ia_ifa);
1360 		return;
1361 	}
1362 
1363 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1364 
1365 	/*
1366 	 * Drop lock while the ARP request is generated.
1367 	 */
1368 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1369 
1370 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1371 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1372 
1373 	/*
1374 	 * Increment the count of retransmissions. If the count has reached the
1375 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1376 	 * the callout to retransmit another GARP packet.
1377 	 */
1378 	++ia->ia_garp_count;
1379 	if (ia->ia_garp_count >= garp_rexmit_count) {
1380 		ifa_free(&ia->ia_ifa);
1381 	} else {
1382 		int rescheduled;
1383 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1384 		rescheduled = callout_reset(&ia->ia_garp_timer,
1385 		    (1 << ia->ia_garp_count) * hz,
1386 		    garp_rexmit, ia);
1387 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1388 		if (rescheduled) {
1389 			ifa_free(&ia->ia_ifa);
1390 		}
1391 	}
1392 
1393 	CURVNET_RESTORE();
1394 }
1395 
1396 /*
1397  * Start the GARP retransmit timer.
1398  *
1399  * A single GARP is always transmitted when an IPv4 address is added
1400  * to an interface and that is usually sufficient. However, in some
1401  * circumstances, such as when a shared address is passed between
1402  * cluster nodes, this single GARP may occasionally be dropped or
1403  * lost. This can lead to neighbors on the network link working with a
1404  * stale ARP cache and sending packets destined for that address to
1405  * the node that previously owned the address, which may not respond.
1406  *
1407  * To avoid this situation, GARP retransmits can be enabled by setting
1408  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1409  * than zero. The setting represents the maximum number of
1410  * retransmissions. The interval between retransmissions is calculated
1411  * using an exponential backoff algorithm, doubling each time, so the
1412  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1413  */
1414 static void
1415 garp_timer_start(struct ifaddr *ifa)
1416 {
1417 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1418 
1419 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1420 	ia->ia_garp_count = 0;
1421 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1422 	    garp_rexmit, ia) == 0) {
1423 		ifa_ref(ifa);
1424 	}
1425 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1426 }
1427 
1428 void
1429 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1430 {
1431 	struct epoch_tracker et;
1432 	const struct sockaddr_in *dst_in;
1433 	const struct sockaddr *dst;
1434 
1435 	if (ifa->ifa_carp != NULL)
1436 		return;
1437 
1438 	dst = ifa->ifa_addr;
1439 	dst_in = (const struct sockaddr_in *)dst;
1440 
1441 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1442 		return;
1443 	NET_EPOCH_ENTER(et);
1444 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1445 	NET_EPOCH_EXIT(et);
1446 	if (garp_rexmit_count > 0) {
1447 		garp_timer_start(ifa);
1448 	}
1449 
1450 	arp_add_ifa_lle(ifp, dst);
1451 }
1452 
1453 void
1454 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1455 {
1456 
1457 	if (ntohl(addr.s_addr) != INADDR_ANY)
1458 		arprequest(ifp, &addr, &addr, enaddr);
1459 }
1460 
1461 /*
1462  * Sends gratuitous ARPs for each ifaddr to notify other
1463  * nodes about the address change.
1464  */
1465 static __noinline void
1466 arp_handle_ifllchange(struct ifnet *ifp)
1467 {
1468 	struct ifaddr *ifa;
1469 
1470 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1471 		if (ifa->ifa_addr->sa_family == AF_INET)
1472 			arp_ifinit(ifp, ifa);
1473 	}
1474 }
1475 
1476 /*
1477  * A handler for interface link layer address change event.
1478  */
1479 static void
1480 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1481 {
1482 	/* if_bridge can update its lladdr during if_vmove(), after we've done
1483 	 * if_detach_internal()/dom_ifdetach(). */
1484 	if (ifp->if_afdata[AF_INET] == NULL)
1485 		return;
1486 
1487 	lltable_update_ifaddr(LLTABLE(ifp));
1488 
1489 	if ((ifp->if_flags & IFF_UP) != 0)
1490 		arp_handle_ifllchange(ifp);
1491 }
1492 
1493 static void
1494 vnet_arp_init(void)
1495 {
1496 
1497 	if (IS_DEFAULT_VNET(curvnet)) {
1498 		netisr_register(&arp_nh);
1499 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1500 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1501 	}
1502 #ifdef VIMAGE
1503 	else
1504 		netisr_register_vnet(&arp_nh);
1505 #endif
1506 }
1507 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1508     vnet_arp_init, 0);
1509 
1510 #ifdef VIMAGE
1511 /*
1512  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1513  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1514  */
1515 static void
1516 vnet_arp_destroy(__unused void *arg)
1517 {
1518 
1519 	netisr_unregister_vnet(&arp_nh);
1520 }
1521 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1522     vnet_arp_destroy, NULL);
1523 #endif
1524