xref: /freebsd/sys/netinet/if_ether.c (revision e27abb6689c5733dd08ce240d5402a0de3a42254)
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	arp_init(void);
147 static void	arpintr(struct mbuf *);
148 static void	arptimer(void *);
149 #ifdef INET
150 static void	in_arpinput(struct mbuf *);
151 #endif
152 
153 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
154     struct ifnet *ifp, int bridged, struct llentry *la);
155 static void arp_mark_lle_reachable(struct llentry *la);
156 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
157 
158 static eventhandler_tag iflladdr_tag;
159 
160 static const struct netisr_handler arp_nh = {
161 	.nh_name = "arp",
162 	.nh_handler = arpintr,
163 	.nh_proto = NETISR_ARP,
164 	.nh_policy = NETISR_POLICY_SOURCE,
165 };
166 
167 /*
168  * Timeout routine.  Age arp_tab entries periodically.
169  */
170 static void
171 arptimer(void *arg)
172 {
173 	struct llentry *lle = (struct llentry *)arg;
174 	struct ifnet *ifp;
175 	int r_skip_req;
176 
177 	if (lle->la_flags & LLE_STATIC) {
178 		return;
179 	}
180 	LLE_WLOCK(lle);
181 	if (callout_pending(&lle->lle_timer)) {
182 		/*
183 		 * Here we are a bit odd here in the treatment of
184 		 * active/pending. If the pending bit is set, it got
185 		 * rescheduled before I ran. The active
186 		 * bit we ignore, since if it was stopped
187 		 * in ll_tablefree() and was currently running
188 		 * it would have return 0 so the code would
189 		 * not have deleted it since the callout could
190 		 * not be stopped so we want to go through
191 		 * with the delete here now. If the callout
192 		 * was restarted, the pending bit will be back on and
193 		 * we just want to bail since the callout_reset would
194 		 * return 1 and our reference would have been removed
195 		 * by arpresolve() below.
196 		 */
197 		LLE_WUNLOCK(lle);
198  		return;
199  	}
200 	ifp = lle->lle_tbl->llt_ifp;
201 	CURVNET_SET(ifp->if_vnet);
202 
203 	switch (lle->ln_state) {
204 	case ARP_LLINFO_REACHABLE:
205 
206 		/*
207 		 * Expiration time is approaching.
208 		 * Let's try to refresh entry if it is still
209 		 * in use.
210 		 *
211 		 * Set r_skip_req to get feedback from
212 		 * fast path. Change state and re-schedule
213 		 * ourselves.
214 		 */
215 		LLE_REQ_LOCK(lle);
216 		lle->r_skip_req = 1;
217 		LLE_REQ_UNLOCK(lle);
218 		lle->ln_state = ARP_LLINFO_VERIFY;
219 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
220 		LLE_WUNLOCK(lle);
221 		CURVNET_RESTORE();
222 		return;
223 	case ARP_LLINFO_VERIFY:
224 		LLE_REQ_LOCK(lle);
225 		r_skip_req = lle->r_skip_req;
226 		LLE_REQ_UNLOCK(lle);
227 
228 		if (r_skip_req == 0 && lle->la_preempt > 0) {
229 			/* Entry was used, issue refresh request */
230 			struct in_addr dst;
231 			dst = lle->r_l3addr.addr4;
232 			lle->la_preempt--;
233 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
234 			LLE_WUNLOCK(lle);
235 			arprequest(ifp, NULL, &dst, NULL);
236 			CURVNET_RESTORE();
237 			return;
238 		}
239 		/* Nothing happened. Reschedule if not too late */
240 		if (lle->la_expire > time_uptime) {
241 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
242 			LLE_WUNLOCK(lle);
243 			CURVNET_RESTORE();
244 			return;
245 		}
246 		break;
247 	case ARP_LLINFO_INCOMPLETE:
248 	case ARP_LLINFO_DELETED:
249 		break;
250 	}
251 
252 	if ((lle->la_flags & LLE_DELETED) == 0) {
253 		int evt;
254 
255 		if (lle->la_flags & LLE_VALID)
256 			evt = LLENTRY_EXPIRED;
257 		else
258 			evt = LLENTRY_TIMEDOUT;
259 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
260 	}
261 
262 	callout_stop(&lle->lle_timer);
263 
264 	/* XXX: LOR avoidance. We still have ref on lle. */
265 	LLE_WUNLOCK(lle);
266 	IF_AFDATA_LOCK(ifp);
267 	LLE_WLOCK(lle);
268 
269 	/* Guard against race with other llentry_free(). */
270 	if (lle->la_flags & LLE_LINKED) {
271 		LLE_REMREF(lle);
272 		lltable_unlink_entry(lle->lle_tbl, lle);
273 	}
274 	IF_AFDATA_UNLOCK(ifp);
275 
276 	size_t pkts_dropped = llentry_free(lle);
277 
278 	ARPSTAT_ADD(dropped, pkts_dropped);
279 	ARPSTAT_INC(timeouts);
280 
281 	CURVNET_RESTORE();
282 }
283 
284 /*
285  * Stores link-layer header for @ifp in format suitable for if_output()
286  * into buffer @buf. Resulting header length is stored in @bufsize.
287  *
288  * Returns 0 on success.
289  */
290 static int
291 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
292     size_t *bufsize)
293 {
294 	struct if_encap_req ereq;
295 	int error;
296 
297 	bzero(buf, *bufsize);
298 	bzero(&ereq, sizeof(ereq));
299 	ereq.buf = buf;
300 	ereq.bufsize = *bufsize;
301 	ereq.rtype = IFENCAP_LL;
302 	ereq.family = AF_ARP;
303 	ereq.lladdr = ar_tha(ah);
304 	ereq.hdata = (u_char *)ah;
305 	if (bcast)
306 		ereq.flags = IFENCAP_FLAG_BROADCAST;
307 	error = ifp->if_requestencap(ifp, &ereq);
308 	if (error == 0)
309 		*bufsize = ereq.bufsize;
310 
311 	return (error);
312 }
313 
314 
315 /*
316  * Broadcast an ARP request. Caller specifies:
317  *	- arp header source ip address
318  *	- arp header target ip address
319  *	- arp header source ethernet address
320  */
321 void
322 arprequest(struct ifnet *ifp, const struct in_addr *sip,
323     const struct in_addr *tip, u_char *enaddr)
324 {
325 	struct mbuf *m;
326 	struct arphdr *ah;
327 	struct sockaddr sa;
328 	u_char *carpaddr = NULL;
329 	uint8_t linkhdr[LLE_MAX_LINKHDR];
330 	size_t linkhdrsize;
331 	struct route ro;
332 	int error;
333 
334 	if (sip == NULL) {
335 		/*
336 		 * The caller did not supply a source address, try to find
337 		 * a compatible one among those assigned to this interface.
338 		 */
339 		struct ifaddr *ifa;
340 
341 		IF_ADDR_RLOCK(ifp);
342 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
343 			if (ifa->ifa_addr->sa_family != AF_INET)
344 				continue;
345 
346 			if (ifa->ifa_carp) {
347 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
348 					continue;
349 				sip = &IA_SIN(ifa)->sin_addr;
350 			} else {
351 				carpaddr = NULL;
352 				sip = &IA_SIN(ifa)->sin_addr;
353 			}
354 
355 			if (0 == ((sip->s_addr ^ tip->s_addr) &
356 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
357 				break;  /* found it. */
358 		}
359 		IF_ADDR_RUNLOCK(ifp);
360 		if (sip == NULL) {
361 			printf("%s: cannot find matching address\n", __func__);
362 			return;
363 		}
364 	}
365 	if (enaddr == NULL)
366 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
367 
368 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
369 		return;
370 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
371 		2 * ifp->if_addrlen;
372 	m->m_pkthdr.len = m->m_len;
373 	M_ALIGN(m, m->m_len);
374 	ah = mtod(m, struct arphdr *);
375 	bzero((caddr_t)ah, m->m_len);
376 #ifdef MAC
377 	mac_netinet_arp_send(ifp, m);
378 #endif
379 	ah->ar_pro = htons(ETHERTYPE_IP);
380 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
381 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
382 	ah->ar_op = htons(ARPOP_REQUEST);
383 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
384 	bcopy(sip, ar_spa(ah), ah->ar_pln);
385 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
386 	sa.sa_family = AF_ARP;
387 	sa.sa_len = 2;
388 
389 	/* Calculate link header for sending frame */
390 	bzero(&ro, sizeof(ro));
391 	linkhdrsize = sizeof(linkhdr);
392 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
393 	if (error != 0 && error != EAFNOSUPPORT) {
394 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
395 		    if_name(ifp), error);
396 		return;
397 	}
398 
399 	ro.ro_prepend = linkhdr;
400 	ro.ro_plen = linkhdrsize;
401 	ro.ro_flags = 0;
402 
403 	m->m_flags |= M_BCAST;
404 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
405 	(*ifp->if_output)(ifp, m, &sa, &ro);
406 	ARPSTAT_INC(txrequests);
407 }
408 
409 
410 /*
411  * Resolve an IP address into an ethernet address - heavy version.
412  * Used internally by arpresolve().
413  * We have already checked than  we can't use existing lle without
414  * modification so we have to acquire LLE_EXCLUSIVE lle lock.
415  *
416  * On success, desten and flags are filled in and the function returns 0;
417  * If the packet must be held pending resolution, we return EWOULDBLOCK
418  * On other errors, we return the corresponding error code.
419  * Note that m_freem() handles NULL.
420  */
421 static int
422 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
423 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
424 	struct llentry **plle)
425 {
426 	struct llentry *la = NULL, *la_tmp;
427 	struct mbuf *curr = NULL;
428 	struct mbuf *next = NULL;
429 	int error, renew;
430 	char *lladdr;
431 	int ll_len;
432 
433 	if (pflags != NULL)
434 		*pflags = 0;
435 	if (plle != NULL)
436 		*plle = NULL;
437 
438 	if ((flags & LLE_CREATE) == 0) {
439 		IF_AFDATA_RLOCK(ifp);
440 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
441 		IF_AFDATA_RUNLOCK(ifp);
442 	}
443 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
444 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
445 		if (la == NULL) {
446 			log(LOG_DEBUG,
447 			    "arpresolve: can't allocate llinfo for %s on %s\n",
448 			    inet_ntoa(SIN(dst)->sin_addr), if_name(ifp));
449 			m_freem(m);
450 			return (EINVAL);
451 		}
452 
453 		IF_AFDATA_WLOCK(ifp);
454 		LLE_WLOCK(la);
455 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
456 		/* Prefer ANY existing lle over newly-created one */
457 		if (la_tmp == NULL)
458 			lltable_link_entry(LLTABLE(ifp), la);
459 		IF_AFDATA_WUNLOCK(ifp);
460 		if (la_tmp != NULL) {
461 			lltable_free_entry(LLTABLE(ifp), la);
462 			la = la_tmp;
463 		}
464 	}
465 	if (la == NULL) {
466 		m_freem(m);
467 		return (EINVAL);
468 	}
469 
470 	if ((la->la_flags & LLE_VALID) &&
471 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
472 		if (flags & LLE_ADDRONLY) {
473 			lladdr = la->ll_addr;
474 			ll_len = ifp->if_addrlen;
475 		} else {
476 			lladdr = la->r_linkdata;
477 			ll_len = la->r_hdrlen;
478 		}
479 		bcopy(lladdr, desten, ll_len);
480 
481 		/* Check if we have feedback request from arptimer() */
482 		if (la->r_skip_req != 0) {
483 			LLE_REQ_LOCK(la);
484 			la->r_skip_req = 0; /* Notify that entry was used */
485 			LLE_REQ_UNLOCK(la);
486 		}
487 		if (pflags != NULL)
488 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
489 		if (plle) {
490 			LLE_ADDREF(la);
491 			*plle = la;
492 		}
493 		LLE_WUNLOCK(la);
494 		return (0);
495 	}
496 
497 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
498 	/*
499 	 * There is an arptab entry, but no ethernet address
500 	 * response yet.  Add the mbuf to the list, dropping
501 	 * the oldest packet if we have exceeded the system
502 	 * setting.
503 	 */
504 	if (m != NULL) {
505 		if (la->la_numheld >= V_arp_maxhold) {
506 			if (la->la_hold != NULL) {
507 				next = la->la_hold->m_nextpkt;
508 				m_freem(la->la_hold);
509 				la->la_hold = next;
510 				la->la_numheld--;
511 				ARPSTAT_INC(dropped);
512 			}
513 		}
514 		if (la->la_hold != NULL) {
515 			curr = la->la_hold;
516 			while (curr->m_nextpkt != NULL)
517 				curr = curr->m_nextpkt;
518 			curr->m_nextpkt = m;
519 		} else
520 			la->la_hold = m;
521 		la->la_numheld++;
522 	}
523 	/*
524 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
525 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
526 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
527 	 * ARP request, but not faster than one request per second.
528 	 */
529 	if (la->la_asked < V_arp_maxtries)
530 		error = EWOULDBLOCK;	/* First request. */
531 	else
532 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
533 
534 	if (renew) {
535 		int canceled;
536 
537 		LLE_ADDREF(la);
538 		la->la_expire = time_uptime;
539 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
540 		    arptimer, la);
541 		if (canceled)
542 			LLE_REMREF(la);
543 		la->la_asked++;
544 		LLE_WUNLOCK(la);
545 		arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
546 		return (error);
547 	}
548 
549 	LLE_WUNLOCK(la);
550 	return (error);
551 }
552 
553 /*
554  * Resolve an IP address into an ethernet address.
555  */
556 int
557 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
558     char *desten, uint32_t *pflags, struct llentry **plle)
559 {
560 	int error;
561 
562 	flags |= LLE_ADDRONLY;
563 	error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
564 	return (error);
565 }
566 
567 
568 /*
569  * Lookups link header based on an IP address.
570  * On input:
571  *    ifp is the interface we use
572  *    is_gw != 0 if @dst represents gateway to some destination
573  *    m is the mbuf. May be NULL if we don't have a packet.
574  *    dst is the next hop,
575  *    desten is the storage to put LL header.
576  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
577  *
578  * On success, full/partial link header and flags are filled in and
579  * the function returns 0.
580  * If the packet must be held pending resolution, we return EWOULDBLOCK
581  * On other errors, we return the corresponding error code.
582  * Note that m_freem() handles NULL.
583  */
584 int
585 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
586 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
587 	struct llentry **plle)
588 {
589 	struct llentry *la = NULL;
590 
591 	if (pflags != NULL)
592 		*pflags = 0;
593 	if (plle != NULL)
594 		*plle = NULL;
595 
596 	if (m != NULL) {
597 		if (m->m_flags & M_BCAST) {
598 			/* broadcast */
599 			(void)memcpy(desten,
600 			    ifp->if_broadcastaddr, ifp->if_addrlen);
601 			return (0);
602 		}
603 		if (m->m_flags & M_MCAST) {
604 			/* multicast */
605 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
606 			return (0);
607 		}
608 	}
609 
610 	IF_AFDATA_RLOCK(ifp);
611 	la = lla_lookup(LLTABLE(ifp), LLE_UNLOCKED, dst);
612 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
613 		/* Entry found, let's copy lle info */
614 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
615 		if (pflags != NULL)
616 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
617 		/* Check if we have feedback request from arptimer() */
618 		if (la->r_skip_req != 0) {
619 			LLE_REQ_LOCK(la);
620 			la->r_skip_req = 0; /* Notify that entry was used */
621 			LLE_REQ_UNLOCK(la);
622 		}
623 		IF_AFDATA_RUNLOCK(ifp);
624 		return (0);
625 	}
626 	IF_AFDATA_RUNLOCK(ifp);
627 
628 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
629 	    desten, pflags, plle));
630 }
631 
632 /*
633  * Common length and type checks are done here,
634  * then the protocol-specific routine is called.
635  */
636 static void
637 arpintr(struct mbuf *m)
638 {
639 	struct arphdr *ar;
640 	struct ifnet *ifp;
641 	char *layer;
642 	int hlen;
643 
644 	ifp = m->m_pkthdr.rcvif;
645 
646 	if (m->m_len < sizeof(struct arphdr) &&
647 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
648 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
649 		    if_name(ifp));
650 		return;
651 	}
652 	ar = mtod(m, struct arphdr *);
653 
654 	/* Check if length is sufficient */
655 	if (m->m_len <  arphdr_len(ar)) {
656 		m = m_pullup(m, arphdr_len(ar));
657 		if (m == NULL) {
658 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
659 			    if_name(ifp));
660 			return;
661 		}
662 		ar = mtod(m, struct arphdr *);
663 	}
664 
665 	hlen = 0;
666 	layer = "";
667 	switch (ntohs(ar->ar_hrd)) {
668 	case ARPHRD_ETHER:
669 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
670 		layer = "ethernet";
671 		break;
672 	case ARPHRD_IEEE802:
673 		hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
674 		layer = "fddi";
675 		break;
676 	case ARPHRD_ARCNET:
677 		hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
678 		layer = "arcnet";
679 		break;
680 	case ARPHRD_INFINIBAND:
681 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
682 		layer = "infiniband";
683 		break;
684 	case ARPHRD_IEEE1394:
685 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
686 		layer = "firewire";
687 
688 		/*
689 		 * Restrict too long hardware addresses.
690 		 * Currently we are capable of handling 20-byte
691 		 * addresses ( sizeof(lle->ll_addr) )
692 		 */
693 		if (ar->ar_hln >= 20)
694 			hlen = 16;
695 		break;
696 	default:
697 		ARP_LOG(LOG_NOTICE,
698 		    "packet with unknown hardware format 0x%02d received on "
699 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
700 		m_freem(m);
701 		return;
702 	}
703 
704 	if (hlen != 0 && hlen != ar->ar_hln) {
705 		ARP_LOG(LOG_NOTICE,
706 		    "packet with invalid %s address length %d received on %s\n",
707 		    layer, ar->ar_hln, if_name(ifp));
708 		m_freem(m);
709 		return;
710 	}
711 
712 	ARPSTAT_INC(received);
713 	switch (ntohs(ar->ar_pro)) {
714 #ifdef INET
715 	case ETHERTYPE_IP:
716 		in_arpinput(m);
717 		return;
718 #endif
719 	}
720 	m_freem(m);
721 }
722 
723 #ifdef INET
724 /*
725  * ARP for Internet protocols on 10 Mb/s Ethernet.
726  * Algorithm is that given in RFC 826.
727  * In addition, a sanity check is performed on the sender
728  * protocol address, to catch impersonators.
729  * We no longer handle negotiations for use of trailer protocol:
730  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
731  * along with IP replies if we wanted trailers sent to us,
732  * and also sent them in response to IP replies.
733  * This allowed either end to announce the desire to receive
734  * trailer packets.
735  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
736  * but formerly didn't normally send requests.
737  */
738 static int log_arp_wrong_iface = 1;
739 static int log_arp_movements = 1;
740 static int log_arp_permanent_modify = 1;
741 static int allow_multicast = 0;
742 
743 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
744 	&log_arp_wrong_iface, 0,
745 	"log arp packets arriving on the wrong interface");
746 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
747 	&log_arp_movements, 0,
748 	"log arp replies from MACs different than the one in the cache");
749 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
750 	&log_arp_permanent_modify, 0,
751 	"log arp replies from MACs different than the one in the permanent arp entry");
752 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
753 	&allow_multicast, 0, "accept multicast addresses");
754 
755 static void
756 in_arpinput(struct mbuf *m)
757 {
758 	struct rm_priotracker in_ifa_tracker;
759 	struct arphdr *ah;
760 	struct ifnet *ifp = m->m_pkthdr.rcvif;
761 	struct llentry *la = NULL, *la_tmp;
762 	struct ifaddr *ifa;
763 	struct in_ifaddr *ia;
764 	struct sockaddr sa;
765 	struct in_addr isaddr, itaddr, myaddr;
766 	u_int8_t *enaddr = NULL;
767 	int op;
768 	int bridged = 0, is_bridge = 0;
769 	int carped;
770 	struct sockaddr_in sin;
771 	struct sockaddr *dst;
772 	struct nhop4_basic nh4;
773 	uint8_t linkhdr[LLE_MAX_LINKHDR];
774 	struct route ro;
775 	size_t linkhdrsize;
776 	int lladdr_off;
777 	int error;
778 
779 	sin.sin_len = sizeof(struct sockaddr_in);
780 	sin.sin_family = AF_INET;
781 	sin.sin_addr.s_addr = 0;
782 
783 	if (ifp->if_bridge)
784 		bridged = 1;
785 	if (ifp->if_type == IFT_BRIDGE)
786 		is_bridge = 1;
787 
788 	/*
789 	 * We already have checked that mbuf contains enough contiguous data
790 	 * to hold entire arp message according to the arp header.
791 	 */
792 	ah = mtod(m, struct arphdr *);
793 
794 	/*
795 	 * ARP is only for IPv4 so we can reject packets with
796 	 * a protocol length not equal to an IPv4 address.
797 	 */
798 	if (ah->ar_pln != sizeof(struct in_addr)) {
799 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
800 		    sizeof(struct in_addr));
801 		goto drop;
802 	}
803 
804 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
805 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
806 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
807 		goto drop;
808 	}
809 
810 	op = ntohs(ah->ar_op);
811 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
812 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
813 
814 	if (op == ARPOP_REPLY)
815 		ARPSTAT_INC(rxreplies);
816 
817 	/*
818 	 * For a bridge, we want to check the address irrespective
819 	 * of the receive interface. (This will change slightly
820 	 * when we have clusters of interfaces).
821 	 */
822 	IN_IFADDR_RLOCK(&in_ifa_tracker);
823 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
824 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
825 		    ia->ia_ifp == ifp) &&
826 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
827 		    (ia->ia_ifa.ifa_carp == NULL ||
828 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
829 			ifa_ref(&ia->ia_ifa);
830 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
831 			goto match;
832 		}
833 	}
834 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
835 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
836 		    ia->ia_ifp == ifp) &&
837 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
838 			ifa_ref(&ia->ia_ifa);
839 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
840 			goto match;
841 		}
842 
843 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
844   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
845   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
846   addr == ia->ia_addr.sin_addr.s_addr)
847 	/*
848 	 * Check the case when bridge shares its MAC address with
849 	 * some of its children, so packets are claimed by bridge
850 	 * itself (bridge_input() does it first), but they are really
851 	 * meant to be destined to the bridge member.
852 	 */
853 	if (is_bridge) {
854 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
855 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
856 				ifa_ref(&ia->ia_ifa);
857 				ifp = ia->ia_ifp;
858 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
859 				goto match;
860 			}
861 		}
862 	}
863 #undef BDG_MEMBER_MATCHES_ARP
864 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
865 
866 	/*
867 	 * No match, use the first inet address on the receive interface
868 	 * as a dummy address for the rest of the function.
869 	 */
870 	IF_ADDR_RLOCK(ifp);
871 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
872 		if (ifa->ifa_addr->sa_family == AF_INET &&
873 		    (ifa->ifa_carp == NULL ||
874 		    (*carp_iamatch_p)(ifa, &enaddr))) {
875 			ia = ifatoia(ifa);
876 			ifa_ref(ifa);
877 			IF_ADDR_RUNLOCK(ifp);
878 			goto match;
879 		}
880 	IF_ADDR_RUNLOCK(ifp);
881 
882 	/*
883 	 * If bridging, fall back to using any inet address.
884 	 */
885 	IN_IFADDR_RLOCK(&in_ifa_tracker);
886 	if (!bridged || (ia = TAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
887 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
888 		goto drop;
889 	}
890 	ifa_ref(&ia->ia_ifa);
891 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
892 match:
893 	if (!enaddr)
894 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
895 	carped = (ia->ia_ifa.ifa_carp != NULL);
896 	myaddr = ia->ia_addr.sin_addr;
897 	ifa_free(&ia->ia_ifa);
898 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
899 		goto drop;	/* it's from me, ignore it. */
900 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
901 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
902 		    "%s!\n", inet_ntoa(isaddr));
903 		goto drop;
904 	}
905 
906 	if (ifp->if_addrlen != ah->ar_hln) {
907 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
908 		    "i/f %d (ignored)\n", ifp->if_addrlen,
909 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
910 		    ifp->if_addrlen);
911 		goto drop;
912 	}
913 
914 	/*
915 	 * Warn if another host is using the same IP address, but only if the
916 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
917 	 * case we suppress the warning to avoid false positive complaints of
918 	 * potential misconfiguration.
919 	 */
920 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
921 	    myaddr.s_addr != 0) {
922 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
923 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
924 		   inet_ntoa(isaddr), ifp->if_xname);
925 		itaddr = myaddr;
926 		ARPSTAT_INC(dupips);
927 		goto reply;
928 	}
929 	if (ifp->if_flags & IFF_STATICARP)
930 		goto reply;
931 
932 	bzero(&sin, sizeof(sin));
933 	sin.sin_len = sizeof(struct sockaddr_in);
934 	sin.sin_family = AF_INET;
935 	sin.sin_addr = isaddr;
936 	dst = (struct sockaddr *)&sin;
937 	IF_AFDATA_RLOCK(ifp);
938 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
939 	IF_AFDATA_RUNLOCK(ifp);
940 	if (la != NULL)
941 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
942 	else if (itaddr.s_addr == myaddr.s_addr) {
943 		/*
944 		 * Request/reply to our address, but no lle exists yet.
945 		 * Calculate full link prepend to use in lle.
946 		 */
947 		linkhdrsize = sizeof(linkhdr);
948 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
949 		    &linkhdrsize, &lladdr_off) != 0)
950 			goto reply;
951 
952 		/* Allocate new entry */
953 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
954 		if (la == NULL) {
955 
956 			/*
957 			 * lle creation may fail if source address belongs
958 			 * to non-directly connected subnet. However, we
959 			 * will try to answer the request instead of dropping
960 			 * frame.
961 			 */
962 			goto reply;
963 		}
964 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
965 		    lladdr_off);
966 
967 		IF_AFDATA_WLOCK(ifp);
968 		LLE_WLOCK(la);
969 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
970 
971 		/*
972 		 * Check if lle still does not exists.
973 		 * If it does, that means that we either
974 		 * 1) have configured it explicitly, via
975 		 * 1a) 'arp -s' static entry or
976 		 * 1b) interface address static record
977 		 * or
978 		 * 2) it was the result of sending first packet to-host
979 		 * or
980 		 * 3) it was another arp reply packet we handled in
981 		 * different thread.
982 		 *
983 		 * In all cases except 3) we definitely need to prefer
984 		 * existing lle. For the sake of simplicity, prefer any
985 		 * existing lle over newly-create one.
986 		 */
987 		if (la_tmp == NULL)
988 			lltable_link_entry(LLTABLE(ifp), la);
989 		IF_AFDATA_WUNLOCK(ifp);
990 
991 		if (la_tmp == NULL) {
992 			arp_mark_lle_reachable(la);
993 			LLE_WUNLOCK(la);
994 		} else {
995 			/* Free newly-create entry and handle packet */
996 			lltable_free_entry(LLTABLE(ifp), la);
997 			la = la_tmp;
998 			la_tmp = NULL;
999 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1000 			/* arp_check_update_lle() returns @la unlocked */
1001 		}
1002 		la = NULL;
1003 	}
1004 reply:
1005 	if (op != ARPOP_REQUEST)
1006 		goto drop;
1007 	ARPSTAT_INC(rxrequests);
1008 
1009 	if (itaddr.s_addr == myaddr.s_addr) {
1010 		/* Shortcut.. the receiving interface is the target. */
1011 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1012 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1013 	} else {
1014 		struct llentry *lle = NULL;
1015 
1016 		sin.sin_addr = itaddr;
1017 		IF_AFDATA_RLOCK(ifp);
1018 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1019 		IF_AFDATA_RUNLOCK(ifp);
1020 
1021 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1022 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1023 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1024 			LLE_RUNLOCK(lle);
1025 		} else {
1026 
1027 			if (lle != NULL)
1028 				LLE_RUNLOCK(lle);
1029 
1030 			if (!V_arp_proxyall)
1031 				goto drop;
1032 
1033 			/* XXX MRT use table 0 for arp reply  */
1034 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1035 				goto drop;
1036 
1037 			/*
1038 			 * Don't send proxies for nodes on the same interface
1039 			 * as this one came out of, or we'll get into a fight
1040 			 * over who claims what Ether address.
1041 			 */
1042 			if (nh4.nh_ifp == ifp)
1043 				goto drop;
1044 
1045 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1046 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1047 
1048 			/*
1049 			 * Also check that the node which sent the ARP packet
1050 			 * is on the interface we expect it to be on. This
1051 			 * avoids ARP chaos if an interface is connected to the
1052 			 * wrong network.
1053 			 */
1054 
1055 			/* XXX MRT use table 0 for arp checks */
1056 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1057 				goto drop;
1058 			if (nh4.nh_ifp != ifp) {
1059 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1060 				    " from %s via %s\n",
1061 				    inet_ntoa(isaddr), ifp->if_xname);
1062 				goto drop;
1063 			}
1064 
1065 #ifdef DEBUG_PROXY
1066 			printf("arp: proxying for %s\n", inet_ntoa(itaddr));
1067 #endif
1068 		}
1069 	}
1070 
1071 	if (itaddr.s_addr == myaddr.s_addr &&
1072 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1073 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1074 #ifdef DEBUG_LINKLOCAL
1075 		printf("arp: sending reply for link-local addr %s\n",
1076 		    inet_ntoa(itaddr));
1077 #endif
1078 		m->m_flags |= M_BCAST;
1079 		m->m_flags &= ~M_MCAST;
1080 	} else {
1081 		/* default behaviour; never reply by broadcast. */
1082 		m->m_flags &= ~(M_BCAST|M_MCAST);
1083 	}
1084 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1085 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1086 	ah->ar_op = htons(ARPOP_REPLY);
1087 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1088 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1089 	m->m_pkthdr.len = m->m_len;
1090 	m->m_pkthdr.rcvif = NULL;
1091 	sa.sa_family = AF_ARP;
1092 	sa.sa_len = 2;
1093 
1094 	/* Calculate link header for sending frame */
1095 	bzero(&ro, sizeof(ro));
1096 	linkhdrsize = sizeof(linkhdr);
1097 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1098 
1099 	/*
1100 	 * arp_fillheader() may fail due to lack of support inside encap request
1101 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1102 	 * by if_output().
1103 	 */
1104 	if (error != 0 && error != EAFNOSUPPORT) {
1105 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1106 		    if_name(ifp), error);
1107 		return;
1108 	}
1109 
1110 	ro.ro_prepend = linkhdr;
1111 	ro.ro_plen = linkhdrsize;
1112 	ro.ro_flags = 0;
1113 
1114 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1115 	(*ifp->if_output)(ifp, m, &sa, &ro);
1116 	ARPSTAT_INC(txreplies);
1117 	return;
1118 
1119 drop:
1120 	m_freem(m);
1121 }
1122 #endif
1123 
1124 /*
1125  * Checks received arp data against existing @la.
1126  * Updates lle state/performs notification if necessary.
1127  */
1128 static void
1129 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1130     int bridged, struct llentry *la)
1131 {
1132 	struct sockaddr sa;
1133 	struct mbuf *m_hold, *m_hold_next;
1134 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1135 	size_t linkhdrsize;
1136 	int lladdr_off;
1137 
1138 	LLE_WLOCK_ASSERT(la);
1139 
1140 	/* the following is not an error when doing bridging */
1141 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1142 		if (log_arp_wrong_iface)
1143 			ARP_LOG(LOG_WARNING, "%s is on %s "
1144 			    "but got reply from %*D on %s\n",
1145 			    inet_ntoa(isaddr),
1146 			    la->lle_tbl->llt_ifp->if_xname,
1147 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1148 			    ifp->if_xname);
1149 		LLE_WUNLOCK(la);
1150 		return;
1151 	}
1152 	if ((la->la_flags & LLE_VALID) &&
1153 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1154 		if (la->la_flags & LLE_STATIC) {
1155 			LLE_WUNLOCK(la);
1156 			if (log_arp_permanent_modify)
1157 				ARP_LOG(LOG_ERR,
1158 				    "%*D attempts to modify "
1159 				    "permanent entry for %s on %s\n",
1160 				    ifp->if_addrlen,
1161 				    (u_char *)ar_sha(ah), ":",
1162 				    inet_ntoa(isaddr), ifp->if_xname);
1163 			return;
1164 		}
1165 		if (log_arp_movements) {
1166 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1167 			    "to %*D on %s\n",
1168 			    inet_ntoa(isaddr),
1169 			    ifp->if_addrlen,
1170 			    (u_char *)&la->ll_addr, ":",
1171 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1172 			    ifp->if_xname);
1173 		}
1174 	}
1175 
1176 	/* Calculate full link prepend to use in lle */
1177 	linkhdrsize = sizeof(linkhdr);
1178 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1179 	    &linkhdrsize, &lladdr_off) != 0)
1180 		return;
1181 
1182 	/* Check if something has changed */
1183 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1184 	    (la->la_flags & LLE_VALID) == 0) {
1185 		/* Try to perform LLE update */
1186 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1187 		    lladdr_off) == 0)
1188 			return;
1189 
1190 		/* Clear fast path feedback request if set */
1191 		la->r_skip_req = 0;
1192 	}
1193 
1194 	arp_mark_lle_reachable(la);
1195 
1196 	/*
1197 	 * The packets are all freed within the call to the output
1198 	 * routine.
1199 	 *
1200 	 * NB: The lock MUST be released before the call to the
1201 	 * output routine.
1202 	 */
1203 	if (la->la_hold != NULL) {
1204 		m_hold = la->la_hold;
1205 		la->la_hold = NULL;
1206 		la->la_numheld = 0;
1207 		lltable_fill_sa_entry(la, &sa);
1208 		LLE_WUNLOCK(la);
1209 		for (; m_hold != NULL; m_hold = m_hold_next) {
1210 			m_hold_next = m_hold->m_nextpkt;
1211 			m_hold->m_nextpkt = NULL;
1212 			/* Avoid confusing lower layers. */
1213 			m_clrprotoflags(m_hold);
1214 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1215 		}
1216 	} else
1217 		LLE_WUNLOCK(la);
1218 }
1219 
1220 static void
1221 arp_mark_lle_reachable(struct llentry *la)
1222 {
1223 	int canceled, wtime;
1224 
1225 	LLE_WLOCK_ASSERT(la);
1226 
1227 	la->ln_state = ARP_LLINFO_REACHABLE;
1228 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1229 
1230 	if (!(la->la_flags & LLE_STATIC)) {
1231 		LLE_ADDREF(la);
1232 		la->la_expire = time_uptime + V_arpt_keep;
1233 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1234 		if (wtime < 0)
1235 			wtime = V_arpt_keep;
1236 		canceled = callout_reset(&la->lle_timer,
1237 		    hz * wtime, arptimer, la);
1238 		if (canceled)
1239 			LLE_REMREF(la);
1240 	}
1241 	la->la_asked = 0;
1242 	la->la_preempt = V_arp_maxtries;
1243 }
1244 
1245 /*
1246  * Add pernament link-layer record for given interface address.
1247  */
1248 static __noinline void
1249 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1250 {
1251 	struct llentry *lle, *lle_tmp;
1252 
1253 	/*
1254 	 * Interface address LLE record is considered static
1255 	 * because kernel code relies on LLE_STATIC flag to check
1256 	 * if these entries can be rewriten by arp updates.
1257 	 */
1258 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1259 	if (lle == NULL) {
1260 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1261 		    "entry for interface address\n");
1262 		return;
1263 	}
1264 
1265 	IF_AFDATA_WLOCK(ifp);
1266 	LLE_WLOCK(lle);
1267 	/* Unlink any entry if exists */
1268 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1269 	if (lle_tmp != NULL)
1270 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1271 
1272 	lltable_link_entry(LLTABLE(ifp), lle);
1273 	IF_AFDATA_WUNLOCK(ifp);
1274 
1275 	if (lle_tmp != NULL)
1276 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1277 
1278 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1279 	LLE_WUNLOCK(lle);
1280 	if (lle_tmp != NULL)
1281 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1282 }
1283 
1284 void
1285 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1286 {
1287 	const struct sockaddr_in *dst_in;
1288 	const struct sockaddr *dst;
1289 
1290 	if (ifa->ifa_carp != NULL)
1291 		return;
1292 
1293 	dst = ifa->ifa_addr;
1294 	dst_in = (const struct sockaddr_in *)dst;
1295 
1296 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1297 		return;
1298 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1299 
1300 	arp_add_ifa_lle(ifp, dst);
1301 }
1302 
1303 void
1304 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1305 {
1306 
1307 	if (ntohl(addr.s_addr) != INADDR_ANY)
1308 		arprequest(ifp, &addr, &addr, enaddr);
1309 }
1310 
1311 /*
1312  * Sends gratuitous ARPs for each ifaddr to notify other
1313  * nodes about the address change.
1314  */
1315 static __noinline void
1316 arp_handle_ifllchange(struct ifnet *ifp)
1317 {
1318 	struct ifaddr *ifa;
1319 
1320 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1321 		if (ifa->ifa_addr->sa_family == AF_INET)
1322 			arp_ifinit(ifp, ifa);
1323 	}
1324 }
1325 
1326 /*
1327  * A handler for interface link layer address change event.
1328  */
1329 static void
1330 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1331 {
1332 
1333 	lltable_update_ifaddr(LLTABLE(ifp));
1334 
1335 	if ((ifp->if_flags & IFF_UP) != 0)
1336 		arp_handle_ifllchange(ifp);
1337 }
1338 
1339 static void
1340 arp_init(void)
1341 {
1342 
1343 	netisr_register(&arp_nh);
1344 	if (IS_DEFAULT_VNET(curvnet))
1345 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1346 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1347 }
1348 SYSINIT(arp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, arp_init, 0);
1349