xref: /freebsd/sys/netinet/if_ether.c (revision 907b59d76938e654f0d040a888e8dfca3de1e222)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
30  */
31 
32 /*
33  * Ethernet address resolution protocol.
34  * TODO:
35  *	add "inuse/lock" bit (or ref. count) along with valid bit
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_inet.h"
42 
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/queue.h>
47 #include <sys/sysctl.h>
48 #include <sys/systm.h>
49 #include <sys/mbuf.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/rmlock.h>
53 #include <sys/socket.h>
54 #include <sys/syslog.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/netisr.h>
61 #include <net/ethernet.h>
62 #include <net/route.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_fib.h>
67 #include <netinet/in_var.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
70 #ifdef INET
71 #include <netinet/ip_carp.h>
72 #endif
73 
74 #include <security/mac/mac_framework.h>
75 
76 #define SIN(s) ((const struct sockaddr_in *)(s))
77 
78 static struct timeval arp_lastlog;
79 static int arp_curpps;
80 static int arp_maxpps = 1;
81 
82 /* Simple ARP state machine */
83 enum arp_llinfo_state {
84 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
85 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
86 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
87 	ARP_LLINFO_DELETED,	/* LLE is deleted */
88 };
89 
90 SYSCTL_DECL(_net_link_ether);
91 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
92 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
93 
94 /* timer values */
95 static VNET_DEFINE(int, arpt_keep) = (20*60);	/* once resolved, good for 20
96 						 * minutes */
97 static VNET_DEFINE(int, arp_maxtries) = 5;
98 static VNET_DEFINE(int, arp_proxyall) = 0;
99 static VNET_DEFINE(int, arpt_down) = 20;	/* keep incomplete entries for
100 						 * 20 seconds */
101 static VNET_DEFINE(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
102 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
103 VNET_PCPUSTAT_SYSINIT(arpstat);
104 
105 #ifdef VIMAGE
106 VNET_PCPUSTAT_SYSUNINIT(arpstat);
107 #endif /* VIMAGE */
108 
109 static VNET_DEFINE(int, arp_maxhold) = 1;
110 
111 #define	V_arpt_keep		VNET(arpt_keep)
112 #define	V_arpt_down		VNET(arpt_down)
113 #define	V_arpt_rexmit		VNET(arpt_rexmit)
114 #define	V_arp_maxtries		VNET(arp_maxtries)
115 #define	V_arp_proxyall		VNET(arp_proxyall)
116 #define	V_arp_maxhold		VNET(arp_maxhold)
117 
118 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
119 	&VNET_NAME(arpt_keep), 0,
120 	"ARP entry lifetime in seconds");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
122 	&VNET_NAME(arp_maxtries), 0,
123 	"ARP resolution attempts before returning error");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
125 	&VNET_NAME(arp_proxyall), 0,
126 	"Enable proxy ARP for all suitable requests");
127 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
128 	&VNET_NAME(arpt_down), 0,
129 	"Incomplete ARP entry lifetime in seconds");
130 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
131     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
133 	&VNET_NAME(arp_maxhold), 0,
134 	"Number of packets to hold per ARP entry");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
136 	CTLFLAG_RW, &arp_maxpps, 0,
137 	"Maximum number of remotely triggered ARP messages that can be "
138 	"logged per second");
139 
140 #define	ARP_LOG(pri, ...)	do {					\
141 	if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
142 		log((pri), "arp: " __VA_ARGS__);			\
143 } while (0)
144 
145 
146 static void	arpintr(struct mbuf *);
147 static void	arptimer(void *);
148 #ifdef INET
149 static void	in_arpinput(struct mbuf *);
150 #endif
151 
152 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
153     struct ifnet *ifp, int bridged, struct llentry *la);
154 static void arp_mark_lle_reachable(struct llentry *la);
155 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
156 
157 static eventhandler_tag iflladdr_tag;
158 
159 static const struct netisr_handler arp_nh = {
160 	.nh_name = "arp",
161 	.nh_handler = arpintr,
162 	.nh_proto = NETISR_ARP,
163 	.nh_policy = NETISR_POLICY_SOURCE,
164 };
165 
166 /*
167  * Timeout routine.  Age arp_tab entries periodically.
168  */
169 static void
170 arptimer(void *arg)
171 {
172 	struct llentry *lle = (struct llentry *)arg;
173 	struct ifnet *ifp;
174 	int r_skip_req;
175 
176 	if (lle->la_flags & LLE_STATIC) {
177 		return;
178 	}
179 	LLE_WLOCK(lle);
180 	if (callout_pending(&lle->lle_timer)) {
181 		/*
182 		 * Here we are a bit odd here in the treatment of
183 		 * active/pending. If the pending bit is set, it got
184 		 * rescheduled before I ran. The active
185 		 * bit we ignore, since if it was stopped
186 		 * in ll_tablefree() and was currently running
187 		 * it would have return 0 so the code would
188 		 * not have deleted it since the callout could
189 		 * not be stopped so we want to go through
190 		 * with the delete here now. If the callout
191 		 * was restarted, the pending bit will be back on and
192 		 * we just want to bail since the callout_reset would
193 		 * return 1 and our reference would have been removed
194 		 * by arpresolve() below.
195 		 */
196 		LLE_WUNLOCK(lle);
197  		return;
198  	}
199 	ifp = lle->lle_tbl->llt_ifp;
200 	CURVNET_SET(ifp->if_vnet);
201 
202 	switch (lle->ln_state) {
203 	case ARP_LLINFO_REACHABLE:
204 
205 		/*
206 		 * Expiration time is approaching.
207 		 * Let's try to refresh entry if it is still
208 		 * in use.
209 		 *
210 		 * Set r_skip_req to get feedback from
211 		 * fast path. Change state and re-schedule
212 		 * ourselves.
213 		 */
214 		LLE_REQ_LOCK(lle);
215 		lle->r_skip_req = 1;
216 		LLE_REQ_UNLOCK(lle);
217 		lle->ln_state = ARP_LLINFO_VERIFY;
218 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
219 		LLE_WUNLOCK(lle);
220 		CURVNET_RESTORE();
221 		return;
222 	case ARP_LLINFO_VERIFY:
223 		LLE_REQ_LOCK(lle);
224 		r_skip_req = lle->r_skip_req;
225 		LLE_REQ_UNLOCK(lle);
226 
227 		if (r_skip_req == 0 && lle->la_preempt > 0) {
228 			/* Entry was used, issue refresh request */
229 			struct in_addr dst;
230 			dst = lle->r_l3addr.addr4;
231 			lle->la_preempt--;
232 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
233 			LLE_WUNLOCK(lle);
234 			arprequest(ifp, NULL, &dst, NULL);
235 			CURVNET_RESTORE();
236 			return;
237 		}
238 		/* Nothing happened. Reschedule if not too late */
239 		if (lle->la_expire > time_uptime) {
240 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
241 			LLE_WUNLOCK(lle);
242 			CURVNET_RESTORE();
243 			return;
244 		}
245 		break;
246 	case ARP_LLINFO_INCOMPLETE:
247 	case ARP_LLINFO_DELETED:
248 		break;
249 	}
250 
251 	if ((lle->la_flags & LLE_DELETED) == 0) {
252 		int evt;
253 
254 		if (lle->la_flags & LLE_VALID)
255 			evt = LLENTRY_EXPIRED;
256 		else
257 			evt = LLENTRY_TIMEDOUT;
258 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
259 	}
260 
261 	callout_stop(&lle->lle_timer);
262 
263 	/* XXX: LOR avoidance. We still have ref on lle. */
264 	LLE_WUNLOCK(lle);
265 	IF_AFDATA_LOCK(ifp);
266 	LLE_WLOCK(lle);
267 
268 	/* Guard against race with other llentry_free(). */
269 	if (lle->la_flags & LLE_LINKED) {
270 		LLE_REMREF(lle);
271 		lltable_unlink_entry(lle->lle_tbl, lle);
272 	}
273 	IF_AFDATA_UNLOCK(ifp);
274 
275 	size_t pkts_dropped = llentry_free(lle);
276 
277 	ARPSTAT_ADD(dropped, pkts_dropped);
278 	ARPSTAT_INC(timeouts);
279 
280 	CURVNET_RESTORE();
281 }
282 
283 /*
284  * Stores link-layer header for @ifp in format suitable for if_output()
285  * into buffer @buf. Resulting header length is stored in @bufsize.
286  *
287  * Returns 0 on success.
288  */
289 static int
290 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
291     size_t *bufsize)
292 {
293 	struct if_encap_req ereq;
294 	int error;
295 
296 	bzero(buf, *bufsize);
297 	bzero(&ereq, sizeof(ereq));
298 	ereq.buf = buf;
299 	ereq.bufsize = *bufsize;
300 	ereq.rtype = IFENCAP_LL;
301 	ereq.family = AF_ARP;
302 	ereq.lladdr = ar_tha(ah);
303 	ereq.hdata = (u_char *)ah;
304 	if (bcast)
305 		ereq.flags = IFENCAP_FLAG_BROADCAST;
306 	error = ifp->if_requestencap(ifp, &ereq);
307 	if (error == 0)
308 		*bufsize = ereq.bufsize;
309 
310 	return (error);
311 }
312 
313 
314 /*
315  * Broadcast an ARP request. Caller specifies:
316  *	- arp header source ip address
317  *	- arp header target ip address
318  *	- arp header source ethernet address
319  */
320 void
321 arprequest(struct ifnet *ifp, const struct in_addr *sip,
322     const struct in_addr *tip, u_char *enaddr)
323 {
324 	struct mbuf *m;
325 	struct arphdr *ah;
326 	struct sockaddr sa;
327 	u_char *carpaddr = NULL;
328 	uint8_t linkhdr[LLE_MAX_LINKHDR];
329 	size_t linkhdrsize;
330 	struct route ro;
331 	int error;
332 
333 	if (sip == NULL) {
334 		/*
335 		 * The caller did not supply a source address, try to find
336 		 * a compatible one among those assigned to this interface.
337 		 */
338 		struct ifaddr *ifa;
339 
340 		IF_ADDR_RLOCK(ifp);
341 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
342 			if (ifa->ifa_addr->sa_family != AF_INET)
343 				continue;
344 
345 			if (ifa->ifa_carp) {
346 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
347 					continue;
348 				sip = &IA_SIN(ifa)->sin_addr;
349 			} else {
350 				carpaddr = NULL;
351 				sip = &IA_SIN(ifa)->sin_addr;
352 			}
353 
354 			if (0 == ((sip->s_addr ^ tip->s_addr) &
355 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
356 				break;  /* found it. */
357 		}
358 		IF_ADDR_RUNLOCK(ifp);
359 		if (sip == NULL) {
360 			printf("%s: cannot find matching address\n", __func__);
361 			return;
362 		}
363 	}
364 	if (enaddr == NULL)
365 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
366 
367 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
368 		return;
369 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
370 		2 * ifp->if_addrlen;
371 	m->m_pkthdr.len = m->m_len;
372 	M_ALIGN(m, m->m_len);
373 	ah = mtod(m, struct arphdr *);
374 	bzero((caddr_t)ah, m->m_len);
375 #ifdef MAC
376 	mac_netinet_arp_send(ifp, m);
377 #endif
378 	ah->ar_pro = htons(ETHERTYPE_IP);
379 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
380 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
381 	ah->ar_op = htons(ARPOP_REQUEST);
382 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
383 	bcopy(sip, ar_spa(ah), ah->ar_pln);
384 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
385 	sa.sa_family = AF_ARP;
386 	sa.sa_len = 2;
387 
388 	/* Calculate link header for sending frame */
389 	bzero(&ro, sizeof(ro));
390 	linkhdrsize = sizeof(linkhdr);
391 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
392 	if (error != 0 && error != EAFNOSUPPORT) {
393 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
394 		    if_name(ifp), error);
395 		return;
396 	}
397 
398 	ro.ro_prepend = linkhdr;
399 	ro.ro_plen = linkhdrsize;
400 	ro.ro_flags = 0;
401 
402 	m->m_flags |= M_BCAST;
403 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
404 	(*ifp->if_output)(ifp, m, &sa, &ro);
405 	ARPSTAT_INC(txrequests);
406 }
407 
408 
409 /*
410  * Resolve an IP address into an ethernet address - heavy version.
411  * Used internally by arpresolve().
412  * We have already checked than  we can't use existing lle without
413  * modification so we have to acquire LLE_EXCLUSIVE lle lock.
414  *
415  * On success, desten and flags are filled in and the function returns 0;
416  * If the packet must be held pending resolution, we return EWOULDBLOCK
417  * On other errors, we return the corresponding error code.
418  * Note that m_freem() handles NULL.
419  */
420 static int
421 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
422 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
423 	struct llentry **plle)
424 {
425 	struct llentry *la = NULL, *la_tmp;
426 	struct mbuf *curr = NULL;
427 	struct mbuf *next = NULL;
428 	int error, renew;
429 	char *lladdr;
430 	int ll_len;
431 
432 	if (pflags != NULL)
433 		*pflags = 0;
434 	if (plle != NULL)
435 		*plle = NULL;
436 
437 	if ((flags & LLE_CREATE) == 0) {
438 		IF_AFDATA_RLOCK(ifp);
439 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
440 		IF_AFDATA_RUNLOCK(ifp);
441 	}
442 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
443 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
444 		if (la == NULL) {
445 			log(LOG_DEBUG,
446 			    "arpresolve: can't allocate llinfo for %s on %s\n",
447 			    inet_ntoa(SIN(dst)->sin_addr), if_name(ifp));
448 			m_freem(m);
449 			return (EINVAL);
450 		}
451 
452 		IF_AFDATA_WLOCK(ifp);
453 		LLE_WLOCK(la);
454 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
455 		/* Prefer ANY existing lle over newly-created one */
456 		if (la_tmp == NULL)
457 			lltable_link_entry(LLTABLE(ifp), la);
458 		IF_AFDATA_WUNLOCK(ifp);
459 		if (la_tmp != NULL) {
460 			lltable_free_entry(LLTABLE(ifp), la);
461 			la = la_tmp;
462 		}
463 	}
464 	if (la == NULL) {
465 		m_freem(m);
466 		return (EINVAL);
467 	}
468 
469 	if ((la->la_flags & LLE_VALID) &&
470 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
471 		if (flags & LLE_ADDRONLY) {
472 			lladdr = la->ll_addr;
473 			ll_len = ifp->if_addrlen;
474 		} else {
475 			lladdr = la->r_linkdata;
476 			ll_len = la->r_hdrlen;
477 		}
478 		bcopy(lladdr, desten, ll_len);
479 
480 		/* Check if we have feedback request from arptimer() */
481 		if (la->r_skip_req != 0) {
482 			LLE_REQ_LOCK(la);
483 			la->r_skip_req = 0; /* Notify that entry was used */
484 			LLE_REQ_UNLOCK(la);
485 		}
486 		if (pflags != NULL)
487 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
488 		if (plle) {
489 			LLE_ADDREF(la);
490 			*plle = la;
491 		}
492 		LLE_WUNLOCK(la);
493 		return (0);
494 	}
495 
496 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
497 	/*
498 	 * There is an arptab entry, but no ethernet address
499 	 * response yet.  Add the mbuf to the list, dropping
500 	 * the oldest packet if we have exceeded the system
501 	 * setting.
502 	 */
503 	if (m != NULL) {
504 		if (la->la_numheld >= V_arp_maxhold) {
505 			if (la->la_hold != NULL) {
506 				next = la->la_hold->m_nextpkt;
507 				m_freem(la->la_hold);
508 				la->la_hold = next;
509 				la->la_numheld--;
510 				ARPSTAT_INC(dropped);
511 			}
512 		}
513 		if (la->la_hold != NULL) {
514 			curr = la->la_hold;
515 			while (curr->m_nextpkt != NULL)
516 				curr = curr->m_nextpkt;
517 			curr->m_nextpkt = m;
518 		} else
519 			la->la_hold = m;
520 		la->la_numheld++;
521 	}
522 	/*
523 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
524 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
525 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
526 	 * ARP request, but not faster than one request per second.
527 	 */
528 	if (la->la_asked < V_arp_maxtries)
529 		error = EWOULDBLOCK;	/* First request. */
530 	else
531 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
532 
533 	if (renew) {
534 		int canceled;
535 
536 		LLE_ADDREF(la);
537 		la->la_expire = time_uptime;
538 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
539 		    arptimer, la);
540 		if (canceled)
541 			LLE_REMREF(la);
542 		la->la_asked++;
543 		LLE_WUNLOCK(la);
544 		arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
545 		return (error);
546 	}
547 
548 	LLE_WUNLOCK(la);
549 	return (error);
550 }
551 
552 /*
553  * Resolve an IP address into an ethernet address.
554  */
555 int
556 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
557     char *desten, uint32_t *pflags, struct llentry **plle)
558 {
559 	int error;
560 
561 	flags |= LLE_ADDRONLY;
562 	error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
563 	return (error);
564 }
565 
566 
567 /*
568  * Lookups link header based on an IP address.
569  * On input:
570  *    ifp is the interface we use
571  *    is_gw != 0 if @dst represents gateway to some destination
572  *    m is the mbuf. May be NULL if we don't have a packet.
573  *    dst is the next hop,
574  *    desten is the storage to put LL header.
575  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
576  *
577  * On success, full/partial link header and flags are filled in and
578  * the function returns 0.
579  * If the packet must be held pending resolution, we return EWOULDBLOCK
580  * On other errors, we return the corresponding error code.
581  * Note that m_freem() handles NULL.
582  */
583 int
584 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
585 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
586 	struct llentry **plle)
587 {
588 	struct llentry *la = NULL;
589 
590 	if (pflags != NULL)
591 		*pflags = 0;
592 	if (plle != NULL)
593 		*plle = NULL;
594 
595 	if (m != NULL) {
596 		if (m->m_flags & M_BCAST) {
597 			/* broadcast */
598 			(void)memcpy(desten,
599 			    ifp->if_broadcastaddr, ifp->if_addrlen);
600 			return (0);
601 		}
602 		if (m->m_flags & M_MCAST) {
603 			/* multicast */
604 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
605 			return (0);
606 		}
607 	}
608 
609 	IF_AFDATA_RLOCK(ifp);
610 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
611 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
612 		/* Entry found, let's copy lle info */
613 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
614 		if (pflags != NULL)
615 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
616 		/* Check if we have feedback request from arptimer() */
617 		if (la->r_skip_req != 0) {
618 			LLE_REQ_LOCK(la);
619 			la->r_skip_req = 0; /* Notify that entry was used */
620 			LLE_REQ_UNLOCK(la);
621 		}
622 		if (plle) {
623 			LLE_ADDREF(la);
624 			*plle = la;
625 			LLE_WUNLOCK(la);
626 		}
627 		IF_AFDATA_RUNLOCK(ifp);
628 		return (0);
629 	}
630 	if (plle && la)
631 		LLE_WUNLOCK(la);
632 	IF_AFDATA_RUNLOCK(ifp);
633 
634 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
635 	    desten, pflags, plle));
636 }
637 
638 /*
639  * Common length and type checks are done here,
640  * then the protocol-specific routine is called.
641  */
642 static void
643 arpintr(struct mbuf *m)
644 {
645 	struct arphdr *ar;
646 	struct ifnet *ifp;
647 	char *layer;
648 	int hlen;
649 
650 	ifp = m->m_pkthdr.rcvif;
651 
652 	if (m->m_len < sizeof(struct arphdr) &&
653 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
654 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
655 		    if_name(ifp));
656 		return;
657 	}
658 	ar = mtod(m, struct arphdr *);
659 
660 	/* Check if length is sufficient */
661 	if (m->m_len <  arphdr_len(ar)) {
662 		m = m_pullup(m, arphdr_len(ar));
663 		if (m == NULL) {
664 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
665 			    if_name(ifp));
666 			return;
667 		}
668 		ar = mtod(m, struct arphdr *);
669 	}
670 
671 	hlen = 0;
672 	layer = "";
673 	switch (ntohs(ar->ar_hrd)) {
674 	case ARPHRD_ETHER:
675 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
676 		layer = "ethernet";
677 		break;
678 	case ARPHRD_IEEE802:
679 		hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
680 		layer = "fddi";
681 		break;
682 	case ARPHRD_ARCNET:
683 		hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
684 		layer = "arcnet";
685 		break;
686 	case ARPHRD_INFINIBAND:
687 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
688 		layer = "infiniband";
689 		break;
690 	case ARPHRD_IEEE1394:
691 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
692 		layer = "firewire";
693 
694 		/*
695 		 * Restrict too long hardware addresses.
696 		 * Currently we are capable of handling 20-byte
697 		 * addresses ( sizeof(lle->ll_addr) )
698 		 */
699 		if (ar->ar_hln >= 20)
700 			hlen = 16;
701 		break;
702 	default:
703 		ARP_LOG(LOG_NOTICE,
704 		    "packet with unknown hardware format 0x%02d received on "
705 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
706 		m_freem(m);
707 		return;
708 	}
709 
710 	if (hlen != 0 && hlen != ar->ar_hln) {
711 		ARP_LOG(LOG_NOTICE,
712 		    "packet with invalid %s address length %d received on %s\n",
713 		    layer, ar->ar_hln, if_name(ifp));
714 		m_freem(m);
715 		return;
716 	}
717 
718 	ARPSTAT_INC(received);
719 	switch (ntohs(ar->ar_pro)) {
720 #ifdef INET
721 	case ETHERTYPE_IP:
722 		in_arpinput(m);
723 		return;
724 #endif
725 	}
726 	m_freem(m);
727 }
728 
729 #ifdef INET
730 /*
731  * ARP for Internet protocols on 10 Mb/s Ethernet.
732  * Algorithm is that given in RFC 826.
733  * In addition, a sanity check is performed on the sender
734  * protocol address, to catch impersonators.
735  * We no longer handle negotiations for use of trailer protocol:
736  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
737  * along with IP replies if we wanted trailers sent to us,
738  * and also sent them in response to IP replies.
739  * This allowed either end to announce the desire to receive
740  * trailer packets.
741  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
742  * but formerly didn't normally send requests.
743  */
744 static int log_arp_wrong_iface = 1;
745 static int log_arp_movements = 1;
746 static int log_arp_permanent_modify = 1;
747 static int allow_multicast = 0;
748 
749 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
750 	&log_arp_wrong_iface, 0,
751 	"log arp packets arriving on the wrong interface");
752 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
753 	&log_arp_movements, 0,
754 	"log arp replies from MACs different than the one in the cache");
755 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
756 	&log_arp_permanent_modify, 0,
757 	"log arp replies from MACs different than the one in the permanent arp entry");
758 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
759 	&allow_multicast, 0, "accept multicast addresses");
760 
761 static void
762 in_arpinput(struct mbuf *m)
763 {
764 	struct rm_priotracker in_ifa_tracker;
765 	struct arphdr *ah;
766 	struct ifnet *ifp = m->m_pkthdr.rcvif;
767 	struct llentry *la = NULL, *la_tmp;
768 	struct ifaddr *ifa;
769 	struct in_ifaddr *ia;
770 	struct sockaddr sa;
771 	struct in_addr isaddr, itaddr, myaddr;
772 	u_int8_t *enaddr = NULL;
773 	int op;
774 	int bridged = 0, is_bridge = 0;
775 	int carped;
776 	struct sockaddr_in sin;
777 	struct sockaddr *dst;
778 	struct nhop4_basic nh4;
779 	uint8_t linkhdr[LLE_MAX_LINKHDR];
780 	struct route ro;
781 	size_t linkhdrsize;
782 	int lladdr_off;
783 	int error;
784 
785 	sin.sin_len = sizeof(struct sockaddr_in);
786 	sin.sin_family = AF_INET;
787 	sin.sin_addr.s_addr = 0;
788 
789 	if (ifp->if_bridge)
790 		bridged = 1;
791 	if (ifp->if_type == IFT_BRIDGE)
792 		is_bridge = 1;
793 
794 	/*
795 	 * We already have checked that mbuf contains enough contiguous data
796 	 * to hold entire arp message according to the arp header.
797 	 */
798 	ah = mtod(m, struct arphdr *);
799 
800 	/*
801 	 * ARP is only for IPv4 so we can reject packets with
802 	 * a protocol length not equal to an IPv4 address.
803 	 */
804 	if (ah->ar_pln != sizeof(struct in_addr)) {
805 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
806 		    sizeof(struct in_addr));
807 		goto drop;
808 	}
809 
810 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
811 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
812 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
813 		goto drop;
814 	}
815 
816 	op = ntohs(ah->ar_op);
817 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
818 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
819 
820 	if (op == ARPOP_REPLY)
821 		ARPSTAT_INC(rxreplies);
822 
823 	/*
824 	 * For a bridge, we want to check the address irrespective
825 	 * of the receive interface. (This will change slightly
826 	 * when we have clusters of interfaces).
827 	 */
828 	IN_IFADDR_RLOCK(&in_ifa_tracker);
829 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
830 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
831 		    ia->ia_ifp == ifp) &&
832 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
833 		    (ia->ia_ifa.ifa_carp == NULL ||
834 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
835 			ifa_ref(&ia->ia_ifa);
836 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
837 			goto match;
838 		}
839 	}
840 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
841 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
842 		    ia->ia_ifp == ifp) &&
843 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
844 			ifa_ref(&ia->ia_ifa);
845 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
846 			goto match;
847 		}
848 
849 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
850   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
851   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
852   addr == ia->ia_addr.sin_addr.s_addr)
853 	/*
854 	 * Check the case when bridge shares its MAC address with
855 	 * some of its children, so packets are claimed by bridge
856 	 * itself (bridge_input() does it first), but they are really
857 	 * meant to be destined to the bridge member.
858 	 */
859 	if (is_bridge) {
860 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
861 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
862 				ifa_ref(&ia->ia_ifa);
863 				ifp = ia->ia_ifp;
864 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
865 				goto match;
866 			}
867 		}
868 	}
869 #undef BDG_MEMBER_MATCHES_ARP
870 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
871 
872 	/*
873 	 * No match, use the first inet address on the receive interface
874 	 * as a dummy address for the rest of the function.
875 	 */
876 	IF_ADDR_RLOCK(ifp);
877 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
878 		if (ifa->ifa_addr->sa_family == AF_INET &&
879 		    (ifa->ifa_carp == NULL ||
880 		    (*carp_iamatch_p)(ifa, &enaddr))) {
881 			ia = ifatoia(ifa);
882 			ifa_ref(ifa);
883 			IF_ADDR_RUNLOCK(ifp);
884 			goto match;
885 		}
886 	IF_ADDR_RUNLOCK(ifp);
887 
888 	/*
889 	 * If bridging, fall back to using any inet address.
890 	 */
891 	IN_IFADDR_RLOCK(&in_ifa_tracker);
892 	if (!bridged || (ia = TAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
893 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
894 		goto drop;
895 	}
896 	ifa_ref(&ia->ia_ifa);
897 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
898 match:
899 	if (!enaddr)
900 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
901 	carped = (ia->ia_ifa.ifa_carp != NULL);
902 	myaddr = ia->ia_addr.sin_addr;
903 	ifa_free(&ia->ia_ifa);
904 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
905 		goto drop;	/* it's from me, ignore it. */
906 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
907 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
908 		    "%s!\n", inet_ntoa(isaddr));
909 		goto drop;
910 	}
911 
912 	if (ifp->if_addrlen != ah->ar_hln) {
913 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
914 		    "i/f %d (ignored)\n", ifp->if_addrlen,
915 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
916 		    ifp->if_addrlen);
917 		goto drop;
918 	}
919 
920 	/*
921 	 * Warn if another host is using the same IP address, but only if the
922 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
923 	 * case we suppress the warning to avoid false positive complaints of
924 	 * potential misconfiguration.
925 	 */
926 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
927 	    myaddr.s_addr != 0) {
928 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
929 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
930 		   inet_ntoa(isaddr), ifp->if_xname);
931 		itaddr = myaddr;
932 		ARPSTAT_INC(dupips);
933 		goto reply;
934 	}
935 	if (ifp->if_flags & IFF_STATICARP)
936 		goto reply;
937 
938 	bzero(&sin, sizeof(sin));
939 	sin.sin_len = sizeof(struct sockaddr_in);
940 	sin.sin_family = AF_INET;
941 	sin.sin_addr = isaddr;
942 	dst = (struct sockaddr *)&sin;
943 	IF_AFDATA_RLOCK(ifp);
944 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
945 	IF_AFDATA_RUNLOCK(ifp);
946 	if (la != NULL)
947 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
948 	else if (itaddr.s_addr == myaddr.s_addr) {
949 		/*
950 		 * Request/reply to our address, but no lle exists yet.
951 		 * Calculate full link prepend to use in lle.
952 		 */
953 		linkhdrsize = sizeof(linkhdr);
954 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
955 		    &linkhdrsize, &lladdr_off) != 0)
956 			goto reply;
957 
958 		/* Allocate new entry */
959 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
960 		if (la == NULL) {
961 
962 			/*
963 			 * lle creation may fail if source address belongs
964 			 * to non-directly connected subnet. However, we
965 			 * will try to answer the request instead of dropping
966 			 * frame.
967 			 */
968 			goto reply;
969 		}
970 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
971 		    lladdr_off);
972 
973 		IF_AFDATA_WLOCK(ifp);
974 		LLE_WLOCK(la);
975 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
976 
977 		/*
978 		 * Check if lle still does not exists.
979 		 * If it does, that means that we either
980 		 * 1) have configured it explicitly, via
981 		 * 1a) 'arp -s' static entry or
982 		 * 1b) interface address static record
983 		 * or
984 		 * 2) it was the result of sending first packet to-host
985 		 * or
986 		 * 3) it was another arp reply packet we handled in
987 		 * different thread.
988 		 *
989 		 * In all cases except 3) we definitely need to prefer
990 		 * existing lle. For the sake of simplicity, prefer any
991 		 * existing lle over newly-create one.
992 		 */
993 		if (la_tmp == NULL)
994 			lltable_link_entry(LLTABLE(ifp), la);
995 		IF_AFDATA_WUNLOCK(ifp);
996 
997 		if (la_tmp == NULL) {
998 			arp_mark_lle_reachable(la);
999 			LLE_WUNLOCK(la);
1000 		} else {
1001 			/* Free newly-create entry and handle packet */
1002 			lltable_free_entry(LLTABLE(ifp), la);
1003 			la = la_tmp;
1004 			la_tmp = NULL;
1005 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1006 			/* arp_check_update_lle() returns @la unlocked */
1007 		}
1008 		la = NULL;
1009 	}
1010 reply:
1011 	if (op != ARPOP_REQUEST)
1012 		goto drop;
1013 	ARPSTAT_INC(rxrequests);
1014 
1015 	if (itaddr.s_addr == myaddr.s_addr) {
1016 		/* Shortcut.. the receiving interface is the target. */
1017 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1018 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1019 	} else {
1020 		struct llentry *lle = NULL;
1021 
1022 		sin.sin_addr = itaddr;
1023 		IF_AFDATA_RLOCK(ifp);
1024 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1025 		IF_AFDATA_RUNLOCK(ifp);
1026 
1027 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1028 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1029 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1030 			LLE_RUNLOCK(lle);
1031 		} else {
1032 
1033 			if (lle != NULL)
1034 				LLE_RUNLOCK(lle);
1035 
1036 			if (!V_arp_proxyall)
1037 				goto drop;
1038 
1039 			/* XXX MRT use table 0 for arp reply  */
1040 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1041 				goto drop;
1042 
1043 			/*
1044 			 * Don't send proxies for nodes on the same interface
1045 			 * as this one came out of, or we'll get into a fight
1046 			 * over who claims what Ether address.
1047 			 */
1048 			if (nh4.nh_ifp == ifp)
1049 				goto drop;
1050 
1051 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1052 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1053 
1054 			/*
1055 			 * Also check that the node which sent the ARP packet
1056 			 * is on the interface we expect it to be on. This
1057 			 * avoids ARP chaos if an interface is connected to the
1058 			 * wrong network.
1059 			 */
1060 
1061 			/* XXX MRT use table 0 for arp checks */
1062 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1063 				goto drop;
1064 			if (nh4.nh_ifp != ifp) {
1065 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1066 				    " from %s via %s\n",
1067 				    inet_ntoa(isaddr), ifp->if_xname);
1068 				goto drop;
1069 			}
1070 
1071 #ifdef DEBUG_PROXY
1072 			printf("arp: proxying for %s\n", inet_ntoa(itaddr));
1073 #endif
1074 		}
1075 	}
1076 
1077 	if (itaddr.s_addr == myaddr.s_addr &&
1078 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1079 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1080 #ifdef DEBUG_LINKLOCAL
1081 		printf("arp: sending reply for link-local addr %s\n",
1082 		    inet_ntoa(itaddr));
1083 #endif
1084 		m->m_flags |= M_BCAST;
1085 		m->m_flags &= ~M_MCAST;
1086 	} else {
1087 		/* default behaviour; never reply by broadcast. */
1088 		m->m_flags &= ~(M_BCAST|M_MCAST);
1089 	}
1090 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1091 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1092 	ah->ar_op = htons(ARPOP_REPLY);
1093 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1094 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1095 	m->m_pkthdr.len = m->m_len;
1096 	m->m_pkthdr.rcvif = NULL;
1097 	sa.sa_family = AF_ARP;
1098 	sa.sa_len = 2;
1099 
1100 	/* Calculate link header for sending frame */
1101 	bzero(&ro, sizeof(ro));
1102 	linkhdrsize = sizeof(linkhdr);
1103 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1104 
1105 	/*
1106 	 * arp_fillheader() may fail due to lack of support inside encap request
1107 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1108 	 * by if_output().
1109 	 */
1110 	if (error != 0 && error != EAFNOSUPPORT) {
1111 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1112 		    if_name(ifp), error);
1113 		return;
1114 	}
1115 
1116 	ro.ro_prepend = linkhdr;
1117 	ro.ro_plen = linkhdrsize;
1118 	ro.ro_flags = 0;
1119 
1120 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1121 	(*ifp->if_output)(ifp, m, &sa, &ro);
1122 	ARPSTAT_INC(txreplies);
1123 	return;
1124 
1125 drop:
1126 	m_freem(m);
1127 }
1128 #endif
1129 
1130 /*
1131  * Checks received arp data against existing @la.
1132  * Updates lle state/performs notification if necessary.
1133  */
1134 static void
1135 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1136     int bridged, struct llentry *la)
1137 {
1138 	struct sockaddr sa;
1139 	struct mbuf *m_hold, *m_hold_next;
1140 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1141 	size_t linkhdrsize;
1142 	int lladdr_off;
1143 
1144 	LLE_WLOCK_ASSERT(la);
1145 
1146 	/* the following is not an error when doing bridging */
1147 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1148 		if (log_arp_wrong_iface)
1149 			ARP_LOG(LOG_WARNING, "%s is on %s "
1150 			    "but got reply from %*D on %s\n",
1151 			    inet_ntoa(isaddr),
1152 			    la->lle_tbl->llt_ifp->if_xname,
1153 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1154 			    ifp->if_xname);
1155 		LLE_WUNLOCK(la);
1156 		return;
1157 	}
1158 	if ((la->la_flags & LLE_VALID) &&
1159 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1160 		if (la->la_flags & LLE_STATIC) {
1161 			LLE_WUNLOCK(la);
1162 			if (log_arp_permanent_modify)
1163 				ARP_LOG(LOG_ERR,
1164 				    "%*D attempts to modify "
1165 				    "permanent entry for %s on %s\n",
1166 				    ifp->if_addrlen,
1167 				    (u_char *)ar_sha(ah), ":",
1168 				    inet_ntoa(isaddr), ifp->if_xname);
1169 			return;
1170 		}
1171 		if (log_arp_movements) {
1172 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1173 			    "to %*D on %s\n",
1174 			    inet_ntoa(isaddr),
1175 			    ifp->if_addrlen,
1176 			    (u_char *)&la->ll_addr, ":",
1177 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1178 			    ifp->if_xname);
1179 		}
1180 	}
1181 
1182 	/* Calculate full link prepend to use in lle */
1183 	linkhdrsize = sizeof(linkhdr);
1184 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1185 	    &linkhdrsize, &lladdr_off) != 0)
1186 		return;
1187 
1188 	/* Check if something has changed */
1189 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1190 	    (la->la_flags & LLE_VALID) == 0) {
1191 		/* Try to perform LLE update */
1192 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1193 		    lladdr_off) == 0)
1194 			return;
1195 
1196 		/* Clear fast path feedback request if set */
1197 		la->r_skip_req = 0;
1198 	}
1199 
1200 	arp_mark_lle_reachable(la);
1201 
1202 	/*
1203 	 * The packets are all freed within the call to the output
1204 	 * routine.
1205 	 *
1206 	 * NB: The lock MUST be released before the call to the
1207 	 * output routine.
1208 	 */
1209 	if (la->la_hold != NULL) {
1210 		m_hold = la->la_hold;
1211 		la->la_hold = NULL;
1212 		la->la_numheld = 0;
1213 		lltable_fill_sa_entry(la, &sa);
1214 		LLE_WUNLOCK(la);
1215 		for (; m_hold != NULL; m_hold = m_hold_next) {
1216 			m_hold_next = m_hold->m_nextpkt;
1217 			m_hold->m_nextpkt = NULL;
1218 			/* Avoid confusing lower layers. */
1219 			m_clrprotoflags(m_hold);
1220 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1221 		}
1222 	} else
1223 		LLE_WUNLOCK(la);
1224 }
1225 
1226 static void
1227 arp_mark_lle_reachable(struct llentry *la)
1228 {
1229 	int canceled, wtime;
1230 
1231 	LLE_WLOCK_ASSERT(la);
1232 
1233 	la->ln_state = ARP_LLINFO_REACHABLE;
1234 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1235 
1236 	if (!(la->la_flags & LLE_STATIC)) {
1237 		LLE_ADDREF(la);
1238 		la->la_expire = time_uptime + V_arpt_keep;
1239 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1240 		if (wtime < 0)
1241 			wtime = V_arpt_keep;
1242 		canceled = callout_reset(&la->lle_timer,
1243 		    hz * wtime, arptimer, la);
1244 		if (canceled)
1245 			LLE_REMREF(la);
1246 	}
1247 	la->la_asked = 0;
1248 	la->la_preempt = V_arp_maxtries;
1249 }
1250 
1251 /*
1252  * Add pernament link-layer record for given interface address.
1253  */
1254 static __noinline void
1255 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1256 {
1257 	struct llentry *lle, *lle_tmp;
1258 
1259 	/*
1260 	 * Interface address LLE record is considered static
1261 	 * because kernel code relies on LLE_STATIC flag to check
1262 	 * if these entries can be rewriten by arp updates.
1263 	 */
1264 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1265 	if (lle == NULL) {
1266 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1267 		    "entry for interface address\n");
1268 		return;
1269 	}
1270 
1271 	IF_AFDATA_WLOCK(ifp);
1272 	LLE_WLOCK(lle);
1273 	/* Unlink any entry if exists */
1274 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1275 	if (lle_tmp != NULL)
1276 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1277 
1278 	lltable_link_entry(LLTABLE(ifp), lle);
1279 	IF_AFDATA_WUNLOCK(ifp);
1280 
1281 	if (lle_tmp != NULL)
1282 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1283 
1284 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1285 	LLE_WUNLOCK(lle);
1286 	if (lle_tmp != NULL)
1287 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1288 }
1289 
1290 void
1291 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1292 {
1293 	const struct sockaddr_in *dst_in;
1294 	const struct sockaddr *dst;
1295 
1296 	if (ifa->ifa_carp != NULL)
1297 		return;
1298 
1299 	dst = ifa->ifa_addr;
1300 	dst_in = (const struct sockaddr_in *)dst;
1301 
1302 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1303 		return;
1304 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1305 
1306 	arp_add_ifa_lle(ifp, dst);
1307 }
1308 
1309 void
1310 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1311 {
1312 
1313 	if (ntohl(addr.s_addr) != INADDR_ANY)
1314 		arprequest(ifp, &addr, &addr, enaddr);
1315 }
1316 
1317 /*
1318  * Sends gratuitous ARPs for each ifaddr to notify other
1319  * nodes about the address change.
1320  */
1321 static __noinline void
1322 arp_handle_ifllchange(struct ifnet *ifp)
1323 {
1324 	struct ifaddr *ifa;
1325 
1326 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1327 		if (ifa->ifa_addr->sa_family == AF_INET)
1328 			arp_ifinit(ifp, ifa);
1329 	}
1330 }
1331 
1332 /*
1333  * A handler for interface link layer address change event.
1334  */
1335 static void
1336 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1337 {
1338 
1339 	lltable_update_ifaddr(LLTABLE(ifp));
1340 
1341 	if ((ifp->if_flags & IFF_UP) != 0)
1342 		arp_handle_ifllchange(ifp);
1343 }
1344 
1345 static void
1346 vnet_arp_init(void)
1347 {
1348 
1349 	if (IS_DEFAULT_VNET(curvnet)) {
1350 		netisr_register(&arp_nh);
1351 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1352 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1353 	}
1354 #ifdef VIMAGE
1355 	else
1356 		netisr_register_vnet(&arp_nh);
1357 #endif
1358 }
1359 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1360     vnet_arp_init, 0);
1361 
1362 #ifdef VIMAGE
1363 /*
1364  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1365  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1366  */
1367 static void
1368 vnet_arp_destroy(__unused void *arg)
1369 {
1370 
1371 	netisr_unregister_vnet(&arp_nh);
1372 }
1373 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1374     vnet_arp_destroy, NULL);
1375 #endif
1376