xref: /freebsd/sys/netinet/if_ether.c (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
32  */
33 
34 /*
35  * Ethernet address resolution protocol.
36  * TODO:
37  *	add "inuse/lock" bit (or ref. count) along with valid bit
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_inet.h"
44 
45 #include <sys/param.h>
46 #include <sys/eventhandler.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/rmlock.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/netisr.h>
64 #include <net/ethernet.h>
65 #include <net/route.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_fib.h>
70 #include <netinet/in_var.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #ifdef INET
74 #include <netinet/ip_carp.h>
75 #endif
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define SIN(s) ((const struct sockaddr_in *)(s))
80 
81 static struct timeval arp_lastlog;
82 static int arp_curpps;
83 static int arp_maxpps = 1;
84 
85 /* Simple ARP state machine */
86 enum arp_llinfo_state {
87 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
88 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
89 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
90 	ARP_LLINFO_DELETED,	/* LLE is deleted */
91 };
92 
93 SYSCTL_DECL(_net_link_ether);
94 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
95 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
96 
97 /* timer values */
98 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60);	/* once resolved, good for 20
99 						 * minutes */
100 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
101 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
102 VNET_DEFINE_STATIC(int, arpt_down) = 20;	/* keep incomplete entries for
103 						 * 20 seconds */
104 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
105 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
106 VNET_PCPUSTAT_SYSINIT(arpstat);
107 
108 #ifdef VIMAGE
109 VNET_PCPUSTAT_SYSUNINIT(arpstat);
110 #endif /* VIMAGE */
111 
112 VNET_DEFINE_STATIC(int, arp_maxhold) = 1;
113 
114 #define	V_arpt_keep		VNET(arpt_keep)
115 #define	V_arpt_down		VNET(arpt_down)
116 #define	V_arpt_rexmit		VNET(arpt_rexmit)
117 #define	V_arp_maxtries		VNET(arp_maxtries)
118 #define	V_arp_proxyall		VNET(arp_proxyall)
119 #define	V_arp_maxhold		VNET(arp_maxhold)
120 
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
122 	&VNET_NAME(arpt_keep), 0,
123 	"ARP entry lifetime in seconds");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
125 	&VNET_NAME(arp_maxtries), 0,
126 	"ARP resolution attempts before returning error");
127 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
128 	&VNET_NAME(arp_proxyall), 0,
129 	"Enable proxy ARP for all suitable requests");
130 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
131 	&VNET_NAME(arpt_down), 0,
132 	"Incomplete ARP entry lifetime in seconds");
133 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
134     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
136 	&VNET_NAME(arp_maxhold), 0,
137 	"Number of packets to hold per ARP entry");
138 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
139 	CTLFLAG_RW, &arp_maxpps, 0,
140 	"Maximum number of remotely triggered ARP messages that can be "
141 	"logged per second");
142 
143 /*
144  * Due to the exponential backoff algorithm used for the interval between GARP
145  * retransmissions, the maximum number of retransmissions is limited for
146  * sanity. This limit corresponds to a maximum interval between retransmissions
147  * of 2^16 seconds ~= 18 hours.
148  *
149  * Making this limit more dynamic is more complicated than worthwhile,
150  * especially since sending out GARPs spaced days apart would be of little
151  * use. A maximum dynamic limit would look something like:
152  *
153  * const int max = fls(INT_MAX / hz) - 1;
154  */
155 #define MAX_GARP_RETRANSMITS 16
156 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
157 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
158 
159 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
160     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
161     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
162     "Number of times to retransmit GARP packets;"
163     " 0 to disable, maximum of 16");
164 
165 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO;	/* Min. log(9) level. */
166 #define	V_arp_log_level		VNET(arp_log_level)
167 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
168 	&VNET_NAME(arp_log_level), 0,
169 	"Minimum log(9) level for recording rate limited arp log messages. "
170 	"The higher will be log more (emerg=0, info=6 (default), debug=7).");
171 #define	ARP_LOG(pri, ...)	do {					\
172 	if ((pri) <= V_arp_log_level &&					\
173 	    ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
174 		log((pri), "arp: " __VA_ARGS__);			\
175 } while (0)
176 
177 
178 static void	arpintr(struct mbuf *);
179 static void	arptimer(void *);
180 #ifdef INET
181 static void	in_arpinput(struct mbuf *);
182 #endif
183 
184 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
185     struct ifnet *ifp, int bridged, struct llentry *la);
186 static void arp_mark_lle_reachable(struct llentry *la);
187 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
188 
189 static eventhandler_tag iflladdr_tag;
190 
191 static const struct netisr_handler arp_nh = {
192 	.nh_name = "arp",
193 	.nh_handler = arpintr,
194 	.nh_proto = NETISR_ARP,
195 	.nh_policy = NETISR_POLICY_SOURCE,
196 };
197 
198 /*
199  * Timeout routine.  Age arp_tab entries periodically.
200  */
201 static void
202 arptimer(void *arg)
203 {
204 	struct llentry *lle = (struct llentry *)arg;
205 	struct ifnet *ifp;
206 	int r_skip_req;
207 
208 	if (lle->la_flags & LLE_STATIC) {
209 		return;
210 	}
211 	LLE_WLOCK(lle);
212 	if (callout_pending(&lle->lle_timer)) {
213 		/*
214 		 * Here we are a bit odd here in the treatment of
215 		 * active/pending. If the pending bit is set, it got
216 		 * rescheduled before I ran. The active
217 		 * bit we ignore, since if it was stopped
218 		 * in ll_tablefree() and was currently running
219 		 * it would have return 0 so the code would
220 		 * not have deleted it since the callout could
221 		 * not be stopped so we want to go through
222 		 * with the delete here now. If the callout
223 		 * was restarted, the pending bit will be back on and
224 		 * we just want to bail since the callout_reset would
225 		 * return 1 and our reference would have been removed
226 		 * by arpresolve() below.
227 		 */
228 		LLE_WUNLOCK(lle);
229  		return;
230  	}
231 	ifp = lle->lle_tbl->llt_ifp;
232 	CURVNET_SET(ifp->if_vnet);
233 
234 	switch (lle->ln_state) {
235 	case ARP_LLINFO_REACHABLE:
236 
237 		/*
238 		 * Expiration time is approaching.
239 		 * Let's try to refresh entry if it is still
240 		 * in use.
241 		 *
242 		 * Set r_skip_req to get feedback from
243 		 * fast path. Change state and re-schedule
244 		 * ourselves.
245 		 */
246 		LLE_REQ_LOCK(lle);
247 		lle->r_skip_req = 1;
248 		LLE_REQ_UNLOCK(lle);
249 		lle->ln_state = ARP_LLINFO_VERIFY;
250 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
251 		LLE_WUNLOCK(lle);
252 		CURVNET_RESTORE();
253 		return;
254 	case ARP_LLINFO_VERIFY:
255 		LLE_REQ_LOCK(lle);
256 		r_skip_req = lle->r_skip_req;
257 		LLE_REQ_UNLOCK(lle);
258 
259 		if (r_skip_req == 0 && lle->la_preempt > 0) {
260 			/* Entry was used, issue refresh request */
261 			struct epoch_tracker et;
262 			struct in_addr dst;
263 
264 			dst = lle->r_l3addr.addr4;
265 			lle->la_preempt--;
266 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
267 			LLE_WUNLOCK(lle);
268 			NET_EPOCH_ENTER(et);
269 			arprequest(ifp, NULL, &dst, NULL);
270 			NET_EPOCH_EXIT(et);
271 			CURVNET_RESTORE();
272 			return;
273 		}
274 		/* Nothing happened. Reschedule if not too late */
275 		if (lle->la_expire > time_uptime) {
276 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
277 			LLE_WUNLOCK(lle);
278 			CURVNET_RESTORE();
279 			return;
280 		}
281 		break;
282 	case ARP_LLINFO_INCOMPLETE:
283 	case ARP_LLINFO_DELETED:
284 		break;
285 	}
286 
287 	if ((lle->la_flags & LLE_DELETED) == 0) {
288 		int evt;
289 
290 		if (lle->la_flags & LLE_VALID)
291 			evt = LLENTRY_EXPIRED;
292 		else
293 			evt = LLENTRY_TIMEDOUT;
294 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
295 	}
296 
297 	callout_stop(&lle->lle_timer);
298 
299 	/* XXX: LOR avoidance. We still have ref on lle. */
300 	LLE_WUNLOCK(lle);
301 	IF_AFDATA_LOCK(ifp);
302 	LLE_WLOCK(lle);
303 
304 	/* Guard against race with other llentry_free(). */
305 	if (lle->la_flags & LLE_LINKED) {
306 		LLE_REMREF(lle);
307 		lltable_unlink_entry(lle->lle_tbl, lle);
308 	}
309 	IF_AFDATA_UNLOCK(ifp);
310 
311 	size_t pkts_dropped = llentry_free(lle);
312 
313 	ARPSTAT_ADD(dropped, pkts_dropped);
314 	ARPSTAT_INC(timeouts);
315 
316 	CURVNET_RESTORE();
317 }
318 
319 /*
320  * Stores link-layer header for @ifp in format suitable for if_output()
321  * into buffer @buf. Resulting header length is stored in @bufsize.
322  *
323  * Returns 0 on success.
324  */
325 static int
326 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
327     size_t *bufsize)
328 {
329 	struct if_encap_req ereq;
330 	int error;
331 
332 	bzero(buf, *bufsize);
333 	bzero(&ereq, sizeof(ereq));
334 	ereq.buf = buf;
335 	ereq.bufsize = *bufsize;
336 	ereq.rtype = IFENCAP_LL;
337 	ereq.family = AF_ARP;
338 	ereq.lladdr = ar_tha(ah);
339 	ereq.hdata = (u_char *)ah;
340 	if (bcast)
341 		ereq.flags = IFENCAP_FLAG_BROADCAST;
342 	error = ifp->if_requestencap(ifp, &ereq);
343 	if (error == 0)
344 		*bufsize = ereq.bufsize;
345 
346 	return (error);
347 }
348 
349 
350 /*
351  * Broadcast an ARP request. Caller specifies:
352  *	- arp header source ip address
353  *	- arp header target ip address
354  *	- arp header source ethernet address
355  */
356 static int
357 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
358     const struct in_addr *tip, u_char *enaddr)
359 {
360 	struct mbuf *m;
361 	struct arphdr *ah;
362 	struct sockaddr sa;
363 	u_char *carpaddr = NULL;
364 	uint8_t linkhdr[LLE_MAX_LINKHDR];
365 	size_t linkhdrsize;
366 	struct route ro;
367 	int error;
368 
369 	NET_EPOCH_ASSERT();
370 
371 	if (sip == NULL) {
372 		/*
373 		 * The caller did not supply a source address, try to find
374 		 * a compatible one among those assigned to this interface.
375 		 */
376 		struct ifaddr *ifa;
377 
378 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
379 			if (ifa->ifa_addr->sa_family != AF_INET)
380 				continue;
381 
382 			if (ifa->ifa_carp) {
383 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
384 					continue;
385 				sip = &IA_SIN(ifa)->sin_addr;
386 			} else {
387 				carpaddr = NULL;
388 				sip = &IA_SIN(ifa)->sin_addr;
389 			}
390 
391 			if (0 == ((sip->s_addr ^ tip->s_addr) &
392 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
393 				break;  /* found it. */
394 		}
395 		if (sip == NULL) {
396 			printf("%s: cannot find matching address\n", __func__);
397 			return (EADDRNOTAVAIL);
398 		}
399 	}
400 	if (enaddr == NULL)
401 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
402 
403 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
404 		return (ENOMEM);
405 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
406 		2 * ifp->if_addrlen;
407 	m->m_pkthdr.len = m->m_len;
408 	M_ALIGN(m, m->m_len);
409 	ah = mtod(m, struct arphdr *);
410 	bzero((caddr_t)ah, m->m_len);
411 #ifdef MAC
412 	mac_netinet_arp_send(ifp, m);
413 #endif
414 	ah->ar_pro = htons(ETHERTYPE_IP);
415 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
416 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
417 	ah->ar_op = htons(ARPOP_REQUEST);
418 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
419 	bcopy(sip, ar_spa(ah), ah->ar_pln);
420 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
421 	sa.sa_family = AF_ARP;
422 	sa.sa_len = 2;
423 
424 	/* Calculate link header for sending frame */
425 	bzero(&ro, sizeof(ro));
426 	linkhdrsize = sizeof(linkhdr);
427 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
428 	if (error != 0 && error != EAFNOSUPPORT) {
429 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
430 		    if_name(ifp), error);
431 		return (error);
432 	}
433 
434 	ro.ro_prepend = linkhdr;
435 	ro.ro_plen = linkhdrsize;
436 	ro.ro_flags = 0;
437 
438 	m->m_flags |= M_BCAST;
439 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
440 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
441 	ARPSTAT_INC(txrequests);
442 	if (error) {
443 		ARPSTAT_INC(txerrors);
444 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
445 		    if_name(ifp), error);
446 	}
447 	return (error);
448 }
449 
450 void
451 arprequest(struct ifnet *ifp, const struct in_addr *sip,
452     const struct in_addr *tip, u_char *enaddr)
453 {
454 
455 	(void) arprequest_internal(ifp, sip, tip, enaddr);
456 }
457 
458 /*
459  * Resolve an IP address into an ethernet address - heavy version.
460  * Used internally by arpresolve().
461  * We have already checked that we can't use an existing lle without
462  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
463  *
464  * On success, desten and pflags are filled in and the function returns 0;
465  * If the packet must be held pending resolution, we return EWOULDBLOCK
466  * On other errors, we return the corresponding error code.
467  * Note that m_freem() handles NULL.
468  */
469 static int
470 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
471 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
472 	struct llentry **plle)
473 {
474 	struct llentry *la = NULL, *la_tmp;
475 	struct mbuf *curr = NULL;
476 	struct mbuf *next = NULL;
477 	int error, renew;
478 	char *lladdr;
479 	int ll_len;
480 
481 	NET_EPOCH_ASSERT();
482 
483 	if (pflags != NULL)
484 		*pflags = 0;
485 	if (plle != NULL)
486 		*plle = NULL;
487 
488 	if ((flags & LLE_CREATE) == 0)
489 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
490 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
491 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
492 		if (la == NULL) {
493 			char addrbuf[INET_ADDRSTRLEN];
494 
495 			log(LOG_DEBUG,
496 			    "arpresolve: can't allocate llinfo for %s on %s\n",
497 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
498 			    if_name(ifp));
499 			m_freem(m);
500 			return (EINVAL);
501 		}
502 
503 		IF_AFDATA_WLOCK(ifp);
504 		LLE_WLOCK(la);
505 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
506 		/* Prefer ANY existing lle over newly-created one */
507 		if (la_tmp == NULL)
508 			lltable_link_entry(LLTABLE(ifp), la);
509 		IF_AFDATA_WUNLOCK(ifp);
510 		if (la_tmp != NULL) {
511 			lltable_free_entry(LLTABLE(ifp), la);
512 			la = la_tmp;
513 		}
514 	}
515 	if (la == NULL) {
516 		m_freem(m);
517 		return (EINVAL);
518 	}
519 
520 	if ((la->la_flags & LLE_VALID) &&
521 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
522 		if (flags & LLE_ADDRONLY) {
523 			lladdr = la->ll_addr;
524 			ll_len = ifp->if_addrlen;
525 		} else {
526 			lladdr = la->r_linkdata;
527 			ll_len = la->r_hdrlen;
528 		}
529 		bcopy(lladdr, desten, ll_len);
530 
531 		/* Notify LLE code that the entry was used by datapath */
532 		llentry_mark_used(la);
533 		if (pflags != NULL)
534 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
535 		if (plle) {
536 			LLE_ADDREF(la);
537 			*plle = la;
538 		}
539 		LLE_WUNLOCK(la);
540 		return (0);
541 	}
542 
543 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
544 	/*
545 	 * There is an arptab entry, but no ethernet address
546 	 * response yet.  Add the mbuf to the list, dropping
547 	 * the oldest packet if we have exceeded the system
548 	 * setting.
549 	 */
550 	if (m != NULL) {
551 		if (la->la_numheld >= V_arp_maxhold) {
552 			if (la->la_hold != NULL) {
553 				next = la->la_hold->m_nextpkt;
554 				m_freem(la->la_hold);
555 				la->la_hold = next;
556 				la->la_numheld--;
557 				ARPSTAT_INC(dropped);
558 			}
559 		}
560 		if (la->la_hold != NULL) {
561 			curr = la->la_hold;
562 			while (curr->m_nextpkt != NULL)
563 				curr = curr->m_nextpkt;
564 			curr->m_nextpkt = m;
565 		} else
566 			la->la_hold = m;
567 		la->la_numheld++;
568 	}
569 	/*
570 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
571 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
572 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
573 	 * ARP request, but not faster than one request per second.
574 	 */
575 	if (la->la_asked < V_arp_maxtries)
576 		error = EWOULDBLOCK;	/* First request. */
577 	else
578 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
579 
580 	if (renew) {
581 		int canceled, e;
582 
583 		LLE_ADDREF(la);
584 		la->la_expire = time_uptime;
585 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
586 		    arptimer, la);
587 		if (canceled)
588 			LLE_REMREF(la);
589 		la->la_asked++;
590 		LLE_WUNLOCK(la);
591 		e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
592 		/*
593 		 * Only overwrite 'error' in case of error; in case of success
594 		 * the proper return value was already set above.
595 		 */
596 		if (e != 0)
597 			return (e);
598 		return (error);
599 	}
600 
601 	LLE_WUNLOCK(la);
602 	return (error);
603 }
604 
605 /*
606  * Lookups link header based on an IP address.
607  * On input:
608  *    ifp is the interface we use
609  *    is_gw != 0 if @dst represents gateway to some destination
610  *    m is the mbuf. May be NULL if we don't have a packet.
611  *    dst is the next hop,
612  *    desten is the storage to put LL header.
613  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
614  *
615  * On success, full/partial link header and flags are filled in and
616  * the function returns 0.
617  * If the packet must be held pending resolution, we return EWOULDBLOCK
618  * On other errors, we return the corresponding error code.
619  * Note that m_freem() handles NULL.
620  */
621 int
622 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
623 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
624 	struct llentry **plle)
625 {
626 	struct llentry *la = NULL;
627 
628 	NET_EPOCH_ASSERT();
629 
630 	if (pflags != NULL)
631 		*pflags = 0;
632 	if (plle != NULL)
633 		*plle = NULL;
634 
635 	if (m != NULL) {
636 		if (m->m_flags & M_BCAST) {
637 			/* broadcast */
638 			(void)memcpy(desten,
639 			    ifp->if_broadcastaddr, ifp->if_addrlen);
640 			return (0);
641 		}
642 		if (m->m_flags & M_MCAST) {
643 			/* multicast */
644 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
645 			return (0);
646 		}
647 	}
648 
649 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
650 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
651 		/* Entry found, let's copy lle info */
652 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
653 		if (pflags != NULL)
654 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
655 		/* Notify the LLE handling code that the entry was used. */
656 		llentry_mark_used(la);
657 		if (plle) {
658 			LLE_ADDREF(la);
659 			*plle = la;
660 			LLE_WUNLOCK(la);
661 		}
662 		return (0);
663 	}
664 	if (plle && la)
665 		LLE_WUNLOCK(la);
666 
667 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
668 	    desten, pflags, plle));
669 }
670 
671 /*
672  * Common length and type checks are done here,
673  * then the protocol-specific routine is called.
674  */
675 static void
676 arpintr(struct mbuf *m)
677 {
678 	struct arphdr *ar;
679 	struct ifnet *ifp;
680 	char *layer;
681 	int hlen;
682 
683 	ifp = m->m_pkthdr.rcvif;
684 
685 	if (m->m_len < sizeof(struct arphdr) &&
686 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
687 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
688 		    if_name(ifp));
689 		return;
690 	}
691 	ar = mtod(m, struct arphdr *);
692 
693 	/* Check if length is sufficient */
694 	if (m->m_len <  arphdr_len(ar)) {
695 		m = m_pullup(m, arphdr_len(ar));
696 		if (m == NULL) {
697 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
698 			    if_name(ifp));
699 			return;
700 		}
701 		ar = mtod(m, struct arphdr *);
702 	}
703 
704 	hlen = 0;
705 	layer = "";
706 	switch (ntohs(ar->ar_hrd)) {
707 	case ARPHRD_ETHER:
708 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
709 		layer = "ethernet";
710 		break;
711 	case ARPHRD_INFINIBAND:
712 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
713 		layer = "infiniband";
714 		break;
715 	case ARPHRD_IEEE1394:
716 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
717 		layer = "firewire";
718 
719 		/*
720 		 * Restrict too long hardware addresses.
721 		 * Currently we are capable of handling 20-byte
722 		 * addresses ( sizeof(lle->ll_addr) )
723 		 */
724 		if (ar->ar_hln >= 20)
725 			hlen = 16;
726 		break;
727 	default:
728 		ARP_LOG(LOG_NOTICE,
729 		    "packet with unknown hardware format 0x%02d received on "
730 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
731 		m_freem(m);
732 		return;
733 	}
734 
735 	if (hlen != 0 && hlen != ar->ar_hln) {
736 		ARP_LOG(LOG_NOTICE,
737 		    "packet with invalid %s address length %d received on %s\n",
738 		    layer, ar->ar_hln, if_name(ifp));
739 		m_freem(m);
740 		return;
741 	}
742 
743 	ARPSTAT_INC(received);
744 	switch (ntohs(ar->ar_pro)) {
745 #ifdef INET
746 	case ETHERTYPE_IP:
747 		in_arpinput(m);
748 		return;
749 #endif
750 	}
751 	m_freem(m);
752 }
753 
754 #ifdef INET
755 /*
756  * ARP for Internet protocols on 10 Mb/s Ethernet.
757  * Algorithm is that given in RFC 826.
758  * In addition, a sanity check is performed on the sender
759  * protocol address, to catch impersonators.
760  * We no longer handle negotiations for use of trailer protocol:
761  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
762  * along with IP replies if we wanted trailers sent to us,
763  * and also sent them in response to IP replies.
764  * This allowed either end to announce the desire to receive
765  * trailer packets.
766  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
767  * but formerly didn't normally send requests.
768  */
769 static int log_arp_wrong_iface = 1;
770 static int log_arp_movements = 1;
771 static int log_arp_permanent_modify = 1;
772 static int allow_multicast = 0;
773 
774 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
775 	&log_arp_wrong_iface, 0,
776 	"log arp packets arriving on the wrong interface");
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
778 	&log_arp_movements, 0,
779 	"log arp replies from MACs different than the one in the cache");
780 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
781 	&log_arp_permanent_modify, 0,
782 	"log arp replies from MACs different than the one in the permanent arp entry");
783 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
784 	&allow_multicast, 0, "accept multicast addresses");
785 
786 static void
787 in_arpinput(struct mbuf *m)
788 {
789 	struct rm_priotracker in_ifa_tracker;
790 	struct arphdr *ah;
791 	struct ifnet *ifp = m->m_pkthdr.rcvif;
792 	struct llentry *la = NULL, *la_tmp;
793 	struct ifaddr *ifa;
794 	struct in_ifaddr *ia;
795 	struct sockaddr sa;
796 	struct in_addr isaddr, itaddr, myaddr;
797 	u_int8_t *enaddr = NULL;
798 	int op;
799 	int bridged = 0, is_bridge = 0;
800 	int carped;
801 	struct sockaddr_in sin;
802 	struct sockaddr *dst;
803 	struct nhop4_basic nh4;
804 	uint8_t linkhdr[LLE_MAX_LINKHDR];
805 	struct route ro;
806 	size_t linkhdrsize;
807 	int lladdr_off;
808 	int error;
809 	char addrbuf[INET_ADDRSTRLEN];
810 
811 	NET_EPOCH_ASSERT();
812 
813 	sin.sin_len = sizeof(struct sockaddr_in);
814 	sin.sin_family = AF_INET;
815 	sin.sin_addr.s_addr = 0;
816 
817 	if (ifp->if_bridge)
818 		bridged = 1;
819 	if (ifp->if_type == IFT_BRIDGE)
820 		is_bridge = 1;
821 
822 	/*
823 	 * We already have checked that mbuf contains enough contiguous data
824 	 * to hold entire arp message according to the arp header.
825 	 */
826 	ah = mtod(m, struct arphdr *);
827 
828 	/*
829 	 * ARP is only for IPv4 so we can reject packets with
830 	 * a protocol length not equal to an IPv4 address.
831 	 */
832 	if (ah->ar_pln != sizeof(struct in_addr)) {
833 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
834 		    sizeof(struct in_addr));
835 		goto drop;
836 	}
837 
838 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
839 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
840 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
841 		goto drop;
842 	}
843 
844 	op = ntohs(ah->ar_op);
845 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
846 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
847 
848 	if (op == ARPOP_REPLY)
849 		ARPSTAT_INC(rxreplies);
850 
851 	/*
852 	 * For a bridge, we want to check the address irrespective
853 	 * of the receive interface. (This will change slightly
854 	 * when we have clusters of interfaces).
855 	 */
856 	IN_IFADDR_RLOCK(&in_ifa_tracker);
857 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
858 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
859 		    ia->ia_ifp == ifp) &&
860 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
861 		    (ia->ia_ifa.ifa_carp == NULL ||
862 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
863 			ifa_ref(&ia->ia_ifa);
864 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
865 			goto match;
866 		}
867 	}
868 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
869 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
870 		    ia->ia_ifp == ifp) &&
871 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
872 			ifa_ref(&ia->ia_ifa);
873 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
874 			goto match;
875 		}
876 
877 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
878   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
879   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
880   addr == ia->ia_addr.sin_addr.s_addr)
881 	/*
882 	 * Check the case when bridge shares its MAC address with
883 	 * some of its children, so packets are claimed by bridge
884 	 * itself (bridge_input() does it first), but they are really
885 	 * meant to be destined to the bridge member.
886 	 */
887 	if (is_bridge) {
888 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
889 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
890 				ifa_ref(&ia->ia_ifa);
891 				ifp = ia->ia_ifp;
892 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
893 				goto match;
894 			}
895 		}
896 	}
897 #undef BDG_MEMBER_MATCHES_ARP
898 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
899 
900 	/*
901 	 * No match, use the first inet address on the receive interface
902 	 * as a dummy address for the rest of the function.
903 	 */
904 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
905 		if (ifa->ifa_addr->sa_family == AF_INET &&
906 		    (ifa->ifa_carp == NULL ||
907 		    (*carp_iamatch_p)(ifa, &enaddr))) {
908 			ia = ifatoia(ifa);
909 			ifa_ref(ifa);
910 			goto match;
911 		}
912 
913 	/*
914 	 * If bridging, fall back to using any inet address.
915 	 */
916 	IN_IFADDR_RLOCK(&in_ifa_tracker);
917 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
918 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
919 		goto drop;
920 	}
921 	ifa_ref(&ia->ia_ifa);
922 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
923 match:
924 	if (!enaddr)
925 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
926 	carped = (ia->ia_ifa.ifa_carp != NULL);
927 	myaddr = ia->ia_addr.sin_addr;
928 	ifa_free(&ia->ia_ifa);
929 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
930 		goto drop;	/* it's from me, ignore it. */
931 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
932 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
933 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
934 		goto drop;
935 	}
936 
937 	if (ifp->if_addrlen != ah->ar_hln) {
938 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
939 		    "i/f %d (ignored)\n", ifp->if_addrlen,
940 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
941 		    ifp->if_addrlen);
942 		goto drop;
943 	}
944 
945 	/*
946 	 * Warn if another host is using the same IP address, but only if the
947 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
948 	 * case we suppress the warning to avoid false positive complaints of
949 	 * potential misconfiguration.
950 	 */
951 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
952 	    myaddr.s_addr != 0) {
953 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
954 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
955 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
956 		itaddr = myaddr;
957 		ARPSTAT_INC(dupips);
958 		goto reply;
959 	}
960 	if (ifp->if_flags & IFF_STATICARP)
961 		goto reply;
962 
963 	bzero(&sin, sizeof(sin));
964 	sin.sin_len = sizeof(struct sockaddr_in);
965 	sin.sin_family = AF_INET;
966 	sin.sin_addr = isaddr;
967 	dst = (struct sockaddr *)&sin;
968 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
969 	if (la != NULL)
970 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
971 	else if (itaddr.s_addr == myaddr.s_addr) {
972 		/*
973 		 * Request/reply to our address, but no lle exists yet.
974 		 * Calculate full link prepend to use in lle.
975 		 */
976 		linkhdrsize = sizeof(linkhdr);
977 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
978 		    &linkhdrsize, &lladdr_off) != 0)
979 			goto reply;
980 
981 		/* Allocate new entry */
982 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
983 		if (la == NULL) {
984 
985 			/*
986 			 * lle creation may fail if source address belongs
987 			 * to non-directly connected subnet. However, we
988 			 * will try to answer the request instead of dropping
989 			 * frame.
990 			 */
991 			goto reply;
992 		}
993 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
994 		    lladdr_off);
995 
996 		IF_AFDATA_WLOCK(ifp);
997 		LLE_WLOCK(la);
998 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
999 
1000 		/*
1001 		 * Check if lle still does not exists.
1002 		 * If it does, that means that we either
1003 		 * 1) have configured it explicitly, via
1004 		 * 1a) 'arp -s' static entry or
1005 		 * 1b) interface address static record
1006 		 * or
1007 		 * 2) it was the result of sending first packet to-host
1008 		 * or
1009 		 * 3) it was another arp reply packet we handled in
1010 		 * different thread.
1011 		 *
1012 		 * In all cases except 3) we definitely need to prefer
1013 		 * existing lle. For the sake of simplicity, prefer any
1014 		 * existing lle over newly-create one.
1015 		 */
1016 		if (la_tmp == NULL)
1017 			lltable_link_entry(LLTABLE(ifp), la);
1018 		IF_AFDATA_WUNLOCK(ifp);
1019 
1020 		if (la_tmp == NULL) {
1021 			arp_mark_lle_reachable(la);
1022 			LLE_WUNLOCK(la);
1023 		} else {
1024 			/* Free newly-create entry and handle packet */
1025 			lltable_free_entry(LLTABLE(ifp), la);
1026 			la = la_tmp;
1027 			la_tmp = NULL;
1028 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1029 			/* arp_check_update_lle() returns @la unlocked */
1030 		}
1031 		la = NULL;
1032 	}
1033 reply:
1034 	if (op != ARPOP_REQUEST)
1035 		goto drop;
1036 	ARPSTAT_INC(rxrequests);
1037 
1038 	if (itaddr.s_addr == myaddr.s_addr) {
1039 		/* Shortcut.. the receiving interface is the target. */
1040 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1041 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1042 	} else {
1043 		struct llentry *lle = NULL;
1044 
1045 		sin.sin_addr = itaddr;
1046 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1047 
1048 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1049 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1050 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1051 			LLE_RUNLOCK(lle);
1052 		} else {
1053 
1054 			if (lle != NULL)
1055 				LLE_RUNLOCK(lle);
1056 
1057 			if (!V_arp_proxyall)
1058 				goto drop;
1059 
1060 			/* XXX MRT use table 0 for arp reply  */
1061 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1062 				goto drop;
1063 
1064 			/*
1065 			 * Don't send proxies for nodes on the same interface
1066 			 * as this one came out of, or we'll get into a fight
1067 			 * over who claims what Ether address.
1068 			 */
1069 			if (nh4.nh_ifp == ifp)
1070 				goto drop;
1071 
1072 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1073 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1074 
1075 			/*
1076 			 * Also check that the node which sent the ARP packet
1077 			 * is on the interface we expect it to be on. This
1078 			 * avoids ARP chaos if an interface is connected to the
1079 			 * wrong network.
1080 			 */
1081 
1082 			/* XXX MRT use table 0 for arp checks */
1083 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1084 				goto drop;
1085 			if (nh4.nh_ifp != ifp) {
1086 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1087 				    " from %s via %s\n",
1088 				    inet_ntoa_r(isaddr, addrbuf),
1089 				    ifp->if_xname);
1090 				goto drop;
1091 			}
1092 
1093 #ifdef DEBUG_PROXY
1094 			printf("arp: proxying for %s\n",
1095 			    inet_ntoa_r(itaddr, addrbuf));
1096 #endif
1097 		}
1098 	}
1099 
1100 	if (itaddr.s_addr == myaddr.s_addr &&
1101 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1102 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1103 #ifdef DEBUG_LINKLOCAL
1104 		printf("arp: sending reply for link-local addr %s\n",
1105 		    inet_ntoa_r(itaddr, addrbuf));
1106 #endif
1107 		m->m_flags |= M_BCAST;
1108 		m->m_flags &= ~M_MCAST;
1109 	} else {
1110 		/* default behaviour; never reply by broadcast. */
1111 		m->m_flags &= ~(M_BCAST|M_MCAST);
1112 	}
1113 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1114 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1115 	ah->ar_op = htons(ARPOP_REPLY);
1116 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1117 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1118 	m->m_pkthdr.len = m->m_len;
1119 	m->m_pkthdr.rcvif = NULL;
1120 	sa.sa_family = AF_ARP;
1121 	sa.sa_len = 2;
1122 
1123 	/* Calculate link header for sending frame */
1124 	bzero(&ro, sizeof(ro));
1125 	linkhdrsize = sizeof(linkhdr);
1126 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1127 
1128 	/*
1129 	 * arp_fillheader() may fail due to lack of support inside encap request
1130 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1131 	 * by if_output().
1132 	 */
1133 	if (error != 0 && error != EAFNOSUPPORT) {
1134 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1135 		    if_name(ifp), error);
1136 		return;
1137 	}
1138 
1139 	ro.ro_prepend = linkhdr;
1140 	ro.ro_plen = linkhdrsize;
1141 	ro.ro_flags = 0;
1142 
1143 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1144 	(*ifp->if_output)(ifp, m, &sa, &ro);
1145 	ARPSTAT_INC(txreplies);
1146 	return;
1147 
1148 drop:
1149 	m_freem(m);
1150 }
1151 #endif
1152 
1153 /*
1154  * Checks received arp data against existing @la.
1155  * Updates lle state/performs notification if necessary.
1156  */
1157 static void
1158 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1159     int bridged, struct llentry *la)
1160 {
1161 	struct sockaddr sa;
1162 	struct mbuf *m_hold, *m_hold_next;
1163 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1164 	size_t linkhdrsize;
1165 	int lladdr_off;
1166 	char addrbuf[INET_ADDRSTRLEN];
1167 
1168 	LLE_WLOCK_ASSERT(la);
1169 
1170 	/* the following is not an error when doing bridging */
1171 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1172 		if (log_arp_wrong_iface)
1173 			ARP_LOG(LOG_WARNING, "%s is on %s "
1174 			    "but got reply from %*D on %s\n",
1175 			    inet_ntoa_r(isaddr, addrbuf),
1176 			    la->lle_tbl->llt_ifp->if_xname,
1177 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1178 			    ifp->if_xname);
1179 		LLE_WUNLOCK(la);
1180 		return;
1181 	}
1182 	if ((la->la_flags & LLE_VALID) &&
1183 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1184 		if (la->la_flags & LLE_STATIC) {
1185 			LLE_WUNLOCK(la);
1186 			if (log_arp_permanent_modify)
1187 				ARP_LOG(LOG_ERR,
1188 				    "%*D attempts to modify "
1189 				    "permanent entry for %s on %s\n",
1190 				    ifp->if_addrlen,
1191 				    (u_char *)ar_sha(ah), ":",
1192 				    inet_ntoa_r(isaddr, addrbuf),
1193 				    ifp->if_xname);
1194 			return;
1195 		}
1196 		if (log_arp_movements) {
1197 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1198 			    "to %*D on %s\n",
1199 			    inet_ntoa_r(isaddr, addrbuf),
1200 			    ifp->if_addrlen,
1201 			    (u_char *)la->ll_addr, ":",
1202 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1203 			    ifp->if_xname);
1204 		}
1205 	}
1206 
1207 	/* Calculate full link prepend to use in lle */
1208 	linkhdrsize = sizeof(linkhdr);
1209 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1210 	    &linkhdrsize, &lladdr_off) != 0)
1211 		return;
1212 
1213 	/* Check if something has changed */
1214 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1215 	    (la->la_flags & LLE_VALID) == 0) {
1216 		/* Try to perform LLE update */
1217 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1218 		    lladdr_off) == 0)
1219 			return;
1220 
1221 		/* Clear fast path feedback request if set */
1222 		la->r_skip_req = 0;
1223 	}
1224 
1225 	arp_mark_lle_reachable(la);
1226 
1227 	/*
1228 	 * The packets are all freed within the call to the output
1229 	 * routine.
1230 	 *
1231 	 * NB: The lock MUST be released before the call to the
1232 	 * output routine.
1233 	 */
1234 	if (la->la_hold != NULL) {
1235 		m_hold = la->la_hold;
1236 		la->la_hold = NULL;
1237 		la->la_numheld = 0;
1238 		lltable_fill_sa_entry(la, &sa);
1239 		LLE_WUNLOCK(la);
1240 		for (; m_hold != NULL; m_hold = m_hold_next) {
1241 			m_hold_next = m_hold->m_nextpkt;
1242 			m_hold->m_nextpkt = NULL;
1243 			/* Avoid confusing lower layers. */
1244 			m_clrprotoflags(m_hold);
1245 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1246 		}
1247 	} else
1248 		LLE_WUNLOCK(la);
1249 }
1250 
1251 static void
1252 arp_mark_lle_reachable(struct llentry *la)
1253 {
1254 	int canceled, wtime;
1255 
1256 	LLE_WLOCK_ASSERT(la);
1257 
1258 	la->ln_state = ARP_LLINFO_REACHABLE;
1259 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1260 
1261 	if (!(la->la_flags & LLE_STATIC)) {
1262 		LLE_ADDREF(la);
1263 		la->la_expire = time_uptime + V_arpt_keep;
1264 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1265 		if (wtime < 0)
1266 			wtime = V_arpt_keep;
1267 		canceled = callout_reset(&la->lle_timer,
1268 		    hz * wtime, arptimer, la);
1269 		if (canceled)
1270 			LLE_REMREF(la);
1271 	}
1272 	la->la_asked = 0;
1273 	la->la_preempt = V_arp_maxtries;
1274 }
1275 
1276 /*
1277  * Add permanent link-layer record for given interface address.
1278  */
1279 static __noinline void
1280 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1281 {
1282 	struct llentry *lle, *lle_tmp;
1283 
1284 	/*
1285 	 * Interface address LLE record is considered static
1286 	 * because kernel code relies on LLE_STATIC flag to check
1287 	 * if these entries can be rewriten by arp updates.
1288 	 */
1289 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1290 	if (lle == NULL) {
1291 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1292 		    "entry for interface address\n");
1293 		return;
1294 	}
1295 
1296 	IF_AFDATA_WLOCK(ifp);
1297 	LLE_WLOCK(lle);
1298 	/* Unlink any entry if exists */
1299 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1300 	if (lle_tmp != NULL)
1301 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1302 
1303 	lltable_link_entry(LLTABLE(ifp), lle);
1304 	IF_AFDATA_WUNLOCK(ifp);
1305 
1306 	if (lle_tmp != NULL)
1307 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1308 
1309 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1310 	LLE_WUNLOCK(lle);
1311 	if (lle_tmp != NULL)
1312 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1313 }
1314 
1315 /*
1316  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1317  * of valid values.
1318  */
1319 static int
1320 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1321 {
1322 	int error;
1323 	int rexmit_count = *(int *)arg1;
1324 
1325 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1326 
1327 	/* Enforce limits on any new value that may have been set. */
1328 	if (!error && req->newptr) {
1329 		/* A new value was set. */
1330 		if (rexmit_count < 0) {
1331 			rexmit_count = 0;
1332 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1333 			rexmit_count = MAX_GARP_RETRANSMITS;
1334 		}
1335 		*(int *)arg1 = rexmit_count;
1336 	}
1337 
1338 	return (error);
1339 }
1340 
1341 /*
1342  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1343  * retransmit it again. A pending callout owns a reference to the ifa.
1344  */
1345 static void
1346 garp_rexmit(void *arg)
1347 {
1348 	struct in_ifaddr *ia = arg;
1349 
1350 	if (callout_pending(&ia->ia_garp_timer) ||
1351 	    !callout_active(&ia->ia_garp_timer)) {
1352 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1353 		ifa_free(&ia->ia_ifa);
1354 		return;
1355 	}
1356 
1357 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1358 
1359 	/*
1360 	 * Drop lock while the ARP request is generated.
1361 	 */
1362 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1363 
1364 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1365 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1366 
1367 	/*
1368 	 * Increment the count of retransmissions. If the count has reached the
1369 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1370 	 * the callout to retransmit another GARP packet.
1371 	 */
1372 	++ia->ia_garp_count;
1373 	if (ia->ia_garp_count >= garp_rexmit_count) {
1374 		ifa_free(&ia->ia_ifa);
1375 	} else {
1376 		int rescheduled;
1377 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1378 		rescheduled = callout_reset(&ia->ia_garp_timer,
1379 		    (1 << ia->ia_garp_count) * hz,
1380 		    garp_rexmit, ia);
1381 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1382 		if (rescheduled) {
1383 			ifa_free(&ia->ia_ifa);
1384 		}
1385 	}
1386 
1387 	CURVNET_RESTORE();
1388 }
1389 
1390 /*
1391  * Start the GARP retransmit timer.
1392  *
1393  * A single GARP is always transmitted when an IPv4 address is added
1394  * to an interface and that is usually sufficient. However, in some
1395  * circumstances, such as when a shared address is passed between
1396  * cluster nodes, this single GARP may occasionally be dropped or
1397  * lost. This can lead to neighbors on the network link working with a
1398  * stale ARP cache and sending packets destined for that address to
1399  * the node that previously owned the address, which may not respond.
1400  *
1401  * To avoid this situation, GARP retransmits can be enabled by setting
1402  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1403  * than zero. The setting represents the maximum number of
1404  * retransmissions. The interval between retransmissions is calculated
1405  * using an exponential backoff algorithm, doubling each time, so the
1406  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1407  */
1408 static void
1409 garp_timer_start(struct ifaddr *ifa)
1410 {
1411 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1412 
1413 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1414 	ia->ia_garp_count = 0;
1415 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1416 	    garp_rexmit, ia) == 0) {
1417 		ifa_ref(ifa);
1418 	}
1419 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1420 }
1421 
1422 void
1423 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1424 {
1425 	struct epoch_tracker et;
1426 	const struct sockaddr_in *dst_in;
1427 	const struct sockaddr *dst;
1428 
1429 	if (ifa->ifa_carp != NULL)
1430 		return;
1431 
1432 	dst = ifa->ifa_addr;
1433 	dst_in = (const struct sockaddr_in *)dst;
1434 
1435 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1436 		return;
1437 	NET_EPOCH_ENTER(et);
1438 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1439 	NET_EPOCH_EXIT(et);
1440 	if (garp_rexmit_count > 0) {
1441 		garp_timer_start(ifa);
1442 	}
1443 
1444 	arp_add_ifa_lle(ifp, dst);
1445 }
1446 
1447 void
1448 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1449 {
1450 
1451 	if (ntohl(addr.s_addr) != INADDR_ANY)
1452 		arprequest(ifp, &addr, &addr, enaddr);
1453 }
1454 
1455 /*
1456  * Sends gratuitous ARPs for each ifaddr to notify other
1457  * nodes about the address change.
1458  */
1459 static __noinline void
1460 arp_handle_ifllchange(struct ifnet *ifp)
1461 {
1462 	struct ifaddr *ifa;
1463 
1464 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1465 		if (ifa->ifa_addr->sa_family == AF_INET)
1466 			arp_ifinit(ifp, ifa);
1467 	}
1468 }
1469 
1470 /*
1471  * A handler for interface link layer address change event.
1472  */
1473 static void
1474 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1475 {
1476 
1477 	lltable_update_ifaddr(LLTABLE(ifp));
1478 
1479 	if ((ifp->if_flags & IFF_UP) != 0)
1480 		arp_handle_ifllchange(ifp);
1481 }
1482 
1483 static void
1484 vnet_arp_init(void)
1485 {
1486 
1487 	if (IS_DEFAULT_VNET(curvnet)) {
1488 		netisr_register(&arp_nh);
1489 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1490 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1491 	}
1492 #ifdef VIMAGE
1493 	else
1494 		netisr_register_vnet(&arp_nh);
1495 #endif
1496 }
1497 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1498     vnet_arp_init, 0);
1499 
1500 #ifdef VIMAGE
1501 /*
1502  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1503  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1504  */
1505 static void
1506 vnet_arp_destroy(__unused void *arg)
1507 {
1508 
1509 	netisr_unregister_vnet(&arp_nh);
1510 }
1511 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1512     vnet_arp_destroy, NULL);
1513 #endif
1514