xref: /freebsd/sys/netinet/if_ether.c (revision 1f88aa09417f1cfb3929fd37531b1ab51213c2d6)
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/route/nhop.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_fib.h>
71 #include <netinet/in_var.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.h>
74 #ifdef INET
75 #include <netinet/ip_carp.h>
76 #endif
77 
78 #include <security/mac/mac_framework.h>
79 
80 #define SIN(s) ((const struct sockaddr_in *)(s))
81 
82 static struct timeval arp_lastlog;
83 static int arp_curpps;
84 static int arp_maxpps = 1;
85 
86 /* Simple ARP state machine */
87 enum arp_llinfo_state {
88 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
89 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
90 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
91 	ARP_LLINFO_DELETED,	/* LLE is deleted */
92 };
93 
94 SYSCTL_DECL(_net_link_ether);
95 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
96     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97     "");
98 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
99     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
100     "");
101 
102 /* timer values */
103 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60);	/* once resolved, good for 20
104 						 * minutes */
105 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
106 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
107 VNET_DEFINE_STATIC(int, arpt_down) = 20;	/* keep incomplete entries for
108 						 * 20 seconds */
109 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
110 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
111 VNET_PCPUSTAT_SYSINIT(arpstat);
112 
113 #ifdef VIMAGE
114 VNET_PCPUSTAT_SYSUNINIT(arpstat);
115 #endif /* VIMAGE */
116 
117 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
118 
119 #define	V_arpt_keep		VNET(arpt_keep)
120 #define	V_arpt_down		VNET(arpt_down)
121 #define	V_arpt_rexmit		VNET(arpt_rexmit)
122 #define	V_arp_maxtries		VNET(arp_maxtries)
123 #define	V_arp_proxyall		VNET(arp_proxyall)
124 #define	V_arp_maxhold		VNET(arp_maxhold)
125 
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
127 	&VNET_NAME(arpt_keep), 0,
128 	"ARP entry lifetime in seconds");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
130 	&VNET_NAME(arp_maxtries), 0,
131 	"ARP resolution attempts before returning error");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
133 	&VNET_NAME(arp_proxyall), 0,
134 	"Enable proxy ARP for all suitable requests");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
136 	&VNET_NAME(arpt_down), 0,
137 	"Incomplete ARP entry lifetime in seconds");
138 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
139     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
140 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
141 	&VNET_NAME(arp_maxhold), 0,
142 	"Number of packets to hold per ARP entry");
143 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
144 	CTLFLAG_RW, &arp_maxpps, 0,
145 	"Maximum number of remotely triggered ARP messages that can be "
146 	"logged per second");
147 
148 /*
149  * Due to the exponential backoff algorithm used for the interval between GARP
150  * retransmissions, the maximum number of retransmissions is limited for
151  * sanity. This limit corresponds to a maximum interval between retransmissions
152  * of 2^16 seconds ~= 18 hours.
153  *
154  * Making this limit more dynamic is more complicated than worthwhile,
155  * especially since sending out GARPs spaced days apart would be of little
156  * use. A maximum dynamic limit would look something like:
157  *
158  * const int max = fls(INT_MAX / hz) - 1;
159  */
160 #define MAX_GARP_RETRANSMITS 16
161 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
162 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
163 
164 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
165     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
166     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
167     "Number of times to retransmit GARP packets;"
168     " 0 to disable, maximum of 16");
169 
170 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO;	/* Min. log(9) level. */
171 #define	V_arp_log_level		VNET(arp_log_level)
172 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
173 	&VNET_NAME(arp_log_level), 0,
174 	"Minimum log(9) level for recording rate limited arp log messages. "
175 	"The higher will be log more (emerg=0, info=6 (default), debug=7).");
176 #define	ARP_LOG(pri, ...)	do {					\
177 	if ((pri) <= V_arp_log_level &&					\
178 	    ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
179 		log((pri), "arp: " __VA_ARGS__);			\
180 } while (0)
181 
182 static void	arpintr(struct mbuf *);
183 static void	arptimer(void *);
184 #ifdef INET
185 static void	in_arpinput(struct mbuf *);
186 #endif
187 
188 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
189     struct ifnet *ifp, int bridged, struct llentry *la);
190 static void arp_mark_lle_reachable(struct llentry *la);
191 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
192 
193 static eventhandler_tag iflladdr_tag;
194 
195 static const struct netisr_handler arp_nh = {
196 	.nh_name = "arp",
197 	.nh_handler = arpintr,
198 	.nh_proto = NETISR_ARP,
199 	.nh_policy = NETISR_POLICY_SOURCE,
200 };
201 
202 /*
203  * Timeout routine.  Age arp_tab entries periodically.
204  */
205 static void
206 arptimer(void *arg)
207 {
208 	struct llentry *lle = (struct llentry *)arg;
209 	struct ifnet *ifp;
210 
211 	if (lle->la_flags & LLE_STATIC) {
212 		return;
213 	}
214 	LLE_WLOCK(lle);
215 	if (callout_pending(&lle->lle_timer)) {
216 		/*
217 		 * Here we are a bit odd here in the treatment of
218 		 * active/pending. If the pending bit is set, it got
219 		 * rescheduled before I ran. The active
220 		 * bit we ignore, since if it was stopped
221 		 * in ll_tablefree() and was currently running
222 		 * it would have return 0 so the code would
223 		 * not have deleted it since the callout could
224 		 * not be stopped so we want to go through
225 		 * with the delete here now. If the callout
226 		 * was restarted, the pending bit will be back on and
227 		 * we just want to bail since the callout_reset would
228 		 * return 1 and our reference would have been removed
229 		 * by arpresolve() below.
230 		 */
231 		LLE_WUNLOCK(lle);
232  		return;
233  	}
234 	ifp = lle->lle_tbl->llt_ifp;
235 	CURVNET_SET(ifp->if_vnet);
236 
237 	switch (lle->ln_state) {
238 	case ARP_LLINFO_REACHABLE:
239 
240 		/*
241 		 * Expiration time is approaching.
242 		 * Request usage feedback from the datapath.
243 		 * Change state and re-schedule ourselves.
244 		 */
245 		llentry_request_feedback(lle);
246 		lle->ln_state = ARP_LLINFO_VERIFY;
247 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
248 		LLE_WUNLOCK(lle);
249 		CURVNET_RESTORE();
250 		return;
251 	case ARP_LLINFO_VERIFY:
252 		if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
253 			/* Entry was used, issue refresh request */
254 			struct epoch_tracker et;
255 			struct in_addr dst;
256 
257 			dst = lle->r_l3addr.addr4;
258 			lle->la_preempt--;
259 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
260 			LLE_WUNLOCK(lle);
261 			NET_EPOCH_ENTER(et);
262 			arprequest(ifp, NULL, &dst, NULL);
263 			NET_EPOCH_EXIT(et);
264 			CURVNET_RESTORE();
265 			return;
266 		}
267 		/* Nothing happened. Reschedule if not too late */
268 		if (lle->la_expire > time_uptime) {
269 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
270 			LLE_WUNLOCK(lle);
271 			CURVNET_RESTORE();
272 			return;
273 		}
274 		break;
275 	case ARP_LLINFO_INCOMPLETE:
276 	case ARP_LLINFO_DELETED:
277 		break;
278 	}
279 
280 	if ((lle->la_flags & LLE_DELETED) == 0) {
281 		int evt;
282 
283 		if (lle->la_flags & LLE_VALID)
284 			evt = LLENTRY_EXPIRED;
285 		else
286 			evt = LLENTRY_TIMEDOUT;
287 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
288 	}
289 
290 	callout_stop(&lle->lle_timer);
291 
292 	/* XXX: LOR avoidance. We still have ref on lle. */
293 	LLE_WUNLOCK(lle);
294 	IF_AFDATA_LOCK(ifp);
295 	LLE_WLOCK(lle);
296 
297 	/* Guard against race with other llentry_free(). */
298 	if (lle->la_flags & LLE_LINKED) {
299 		LLE_REMREF(lle);
300 		lltable_unlink_entry(lle->lle_tbl, lle);
301 	}
302 	IF_AFDATA_UNLOCK(ifp);
303 
304 	size_t pkts_dropped = llentry_free(lle);
305 
306 	ARPSTAT_ADD(dropped, pkts_dropped);
307 	ARPSTAT_INC(timeouts);
308 
309 	CURVNET_RESTORE();
310 }
311 
312 /*
313  * Stores link-layer header for @ifp in format suitable for if_output()
314  * into buffer @buf. Resulting header length is stored in @bufsize.
315  *
316  * Returns 0 on success.
317  */
318 static int
319 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
320     size_t *bufsize)
321 {
322 	struct if_encap_req ereq;
323 	int error;
324 
325 	bzero(buf, *bufsize);
326 	bzero(&ereq, sizeof(ereq));
327 	ereq.buf = buf;
328 	ereq.bufsize = *bufsize;
329 	ereq.rtype = IFENCAP_LL;
330 	ereq.family = AF_ARP;
331 	ereq.lladdr = ar_tha(ah);
332 	ereq.hdata = (u_char *)ah;
333 	if (bcast)
334 		ereq.flags = IFENCAP_FLAG_BROADCAST;
335 	error = ifp->if_requestencap(ifp, &ereq);
336 	if (error == 0)
337 		*bufsize = ereq.bufsize;
338 
339 	return (error);
340 }
341 
342 /*
343  * Broadcast an ARP request. Caller specifies:
344  *	- arp header source ip address
345  *	- arp header target ip address
346  *	- arp header source ethernet address
347  */
348 static int
349 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
350     const struct in_addr *tip, u_char *enaddr)
351 {
352 	struct mbuf *m;
353 	struct arphdr *ah;
354 	struct sockaddr sa;
355 	u_char *carpaddr = NULL;
356 	uint8_t linkhdr[LLE_MAX_LINKHDR];
357 	size_t linkhdrsize;
358 	struct route ro;
359 	int error;
360 
361 	NET_EPOCH_ASSERT();
362 
363 	if (sip == NULL) {
364 		/*
365 		 * The caller did not supply a source address, try to find
366 		 * a compatible one among those assigned to this interface.
367 		 */
368 		struct ifaddr *ifa;
369 
370 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
371 			if (ifa->ifa_addr->sa_family != AF_INET)
372 				continue;
373 
374 			if (ifa->ifa_carp) {
375 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
376 					continue;
377 				sip = &IA_SIN(ifa)->sin_addr;
378 			} else {
379 				carpaddr = NULL;
380 				sip = &IA_SIN(ifa)->sin_addr;
381 			}
382 
383 			if (0 == ((sip->s_addr ^ tip->s_addr) &
384 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
385 				break;  /* found it. */
386 		}
387 		if (sip == NULL) {
388 			printf("%s: cannot find matching address\n", __func__);
389 			return (EADDRNOTAVAIL);
390 		}
391 	}
392 	if (enaddr == NULL)
393 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
394 
395 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
396 		return (ENOMEM);
397 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
398 		2 * ifp->if_addrlen;
399 	m->m_pkthdr.len = m->m_len;
400 	M_ALIGN(m, m->m_len);
401 	ah = mtod(m, struct arphdr *);
402 	bzero((caddr_t)ah, m->m_len);
403 #ifdef MAC
404 	mac_netinet_arp_send(ifp, m);
405 #endif
406 	ah->ar_pro = htons(ETHERTYPE_IP);
407 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
408 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
409 	ah->ar_op = htons(ARPOP_REQUEST);
410 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
411 	bcopy(sip, ar_spa(ah), ah->ar_pln);
412 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
413 	sa.sa_family = AF_ARP;
414 	sa.sa_len = 2;
415 
416 	/* Calculate link header for sending frame */
417 	bzero(&ro, sizeof(ro));
418 	linkhdrsize = sizeof(linkhdr);
419 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
420 	if (error != 0 && error != EAFNOSUPPORT) {
421 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
422 		    if_name(ifp), error);
423 		return (error);
424 	}
425 
426 	ro.ro_prepend = linkhdr;
427 	ro.ro_plen = linkhdrsize;
428 	ro.ro_flags = 0;
429 
430 	m->m_flags |= M_BCAST;
431 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
432 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
433 	ARPSTAT_INC(txrequests);
434 	if (error) {
435 		ARPSTAT_INC(txerrors);
436 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
437 		    if_name(ifp), error);
438 	}
439 	return (error);
440 }
441 
442 void
443 arprequest(struct ifnet *ifp, const struct in_addr *sip,
444     const struct in_addr *tip, u_char *enaddr)
445 {
446 
447 	(void) arprequest_internal(ifp, sip, tip, enaddr);
448 }
449 
450 /*
451  * Resolve an IP address into an ethernet address - heavy version.
452  * Used internally by arpresolve().
453  * We have already checked that we can't use an existing lle without
454  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
455  *
456  * On success, desten and pflags are filled in and the function returns 0;
457  * If the packet must be held pending resolution, we return EWOULDBLOCK
458  * On other errors, we return the corresponding error code.
459  * Note that m_freem() handles NULL.
460  */
461 static int
462 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
463 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
464 	struct llentry **plle)
465 {
466 	struct llentry *la = NULL, *la_tmp;
467 	struct mbuf *curr = NULL;
468 	struct mbuf *next = NULL;
469 	int error, renew;
470 	char *lladdr;
471 	int ll_len;
472 
473 	NET_EPOCH_ASSERT();
474 
475 	if (pflags != NULL)
476 		*pflags = 0;
477 	if (plle != NULL)
478 		*plle = NULL;
479 
480 	if ((flags & LLE_CREATE) == 0)
481 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
482 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
483 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
484 		if (la == NULL) {
485 			char addrbuf[INET_ADDRSTRLEN];
486 
487 			log(LOG_DEBUG,
488 			    "arpresolve: can't allocate llinfo for %s on %s\n",
489 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
490 			    if_name(ifp));
491 			m_freem(m);
492 			return (EINVAL);
493 		}
494 
495 		IF_AFDATA_WLOCK(ifp);
496 		LLE_WLOCK(la);
497 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
498 		/* Prefer ANY existing lle over newly-created one */
499 		if (la_tmp == NULL)
500 			lltable_link_entry(LLTABLE(ifp), la);
501 		IF_AFDATA_WUNLOCK(ifp);
502 		if (la_tmp != NULL) {
503 			lltable_free_entry(LLTABLE(ifp), la);
504 			la = la_tmp;
505 		}
506 	}
507 	if (la == NULL) {
508 		m_freem(m);
509 		return (EINVAL);
510 	}
511 
512 	if ((la->la_flags & LLE_VALID) &&
513 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
514 		if (flags & LLE_ADDRONLY) {
515 			lladdr = la->ll_addr;
516 			ll_len = ifp->if_addrlen;
517 		} else {
518 			lladdr = la->r_linkdata;
519 			ll_len = la->r_hdrlen;
520 		}
521 		bcopy(lladdr, desten, ll_len);
522 
523 		/* Notify LLE code that the entry was used by datapath */
524 		llentry_provide_feedback(la);
525 		if (pflags != NULL)
526 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
527 		if (plle) {
528 			LLE_ADDREF(la);
529 			*plle = la;
530 		}
531 		LLE_WUNLOCK(la);
532 		return (0);
533 	}
534 
535 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
536 	/*
537 	 * There is an arptab entry, but no ethernet address
538 	 * response yet.  Add the mbuf to the list, dropping
539 	 * the oldest packet if we have exceeded the system
540 	 * setting.
541 	 */
542 	if (m != NULL) {
543 		if (la->la_numheld >= V_arp_maxhold) {
544 			if (la->la_hold != NULL) {
545 				next = la->la_hold->m_nextpkt;
546 				m_freem(la->la_hold);
547 				la->la_hold = next;
548 				la->la_numheld--;
549 				ARPSTAT_INC(dropped);
550 			}
551 		}
552 		if (la->la_hold != NULL) {
553 			curr = la->la_hold;
554 			while (curr->m_nextpkt != NULL)
555 				curr = curr->m_nextpkt;
556 			curr->m_nextpkt = m;
557 		} else
558 			la->la_hold = m;
559 		la->la_numheld++;
560 	}
561 	/*
562 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
563 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
564 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
565 	 * ARP request, but not faster than one request per second.
566 	 */
567 	if (la->la_asked < V_arp_maxtries)
568 		error = EWOULDBLOCK;	/* First request. */
569 	else
570 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
571 
572 	if (renew) {
573 		int canceled, e;
574 
575 		LLE_ADDREF(la);
576 		la->la_expire = time_uptime;
577 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
578 		    arptimer, la);
579 		if (canceled)
580 			LLE_REMREF(la);
581 		la->la_asked++;
582 		LLE_WUNLOCK(la);
583 		e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
584 		/*
585 		 * Only overwrite 'error' in case of error; in case of success
586 		 * the proper return value was already set above.
587 		 */
588 		if (e != 0)
589 			return (e);
590 		return (error);
591 	}
592 
593 	LLE_WUNLOCK(la);
594 	return (error);
595 }
596 
597 /*
598  * Lookups link header based on an IP address.
599  * On input:
600  *    ifp is the interface we use
601  *    is_gw != 0 if @dst represents gateway to some destination
602  *    m is the mbuf. May be NULL if we don't have a packet.
603  *    dst is the next hop,
604  *    desten is the storage to put LL header.
605  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
606  *
607  * On success, full/partial link header and flags are filled in and
608  * the function returns 0.
609  * If the packet must be held pending resolution, we return EWOULDBLOCK
610  * On other errors, we return the corresponding error code.
611  * Note that m_freem() handles NULL.
612  */
613 int
614 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
615 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
616 	struct llentry **plle)
617 {
618 	struct llentry *la = NULL;
619 
620 	NET_EPOCH_ASSERT();
621 
622 	if (pflags != NULL)
623 		*pflags = 0;
624 	if (plle != NULL)
625 		*plle = NULL;
626 
627 	if (m != NULL) {
628 		if (m->m_flags & M_BCAST) {
629 			/* broadcast */
630 			(void)memcpy(desten,
631 			    ifp->if_broadcastaddr, ifp->if_addrlen);
632 			return (0);
633 		}
634 		if (m->m_flags & M_MCAST) {
635 			/* multicast */
636 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
637 			return (0);
638 		}
639 	}
640 
641 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
642 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
643 		/* Entry found, let's copy lle info */
644 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
645 		if (pflags != NULL)
646 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
647 		/* Notify the LLE handling code that the entry was used. */
648 		llentry_provide_feedback(la);
649 		if (plle) {
650 			LLE_ADDREF(la);
651 			*plle = la;
652 			LLE_WUNLOCK(la);
653 		}
654 		return (0);
655 	}
656 	if (plle && la)
657 		LLE_WUNLOCK(la);
658 
659 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
660 	    desten, pflags, plle));
661 }
662 
663 /*
664  * Common length and type checks are done here,
665  * then the protocol-specific routine is called.
666  */
667 static void
668 arpintr(struct mbuf *m)
669 {
670 	struct arphdr *ar;
671 	struct ifnet *ifp;
672 	char *layer;
673 	int hlen;
674 
675 	ifp = m->m_pkthdr.rcvif;
676 
677 	if (m->m_len < sizeof(struct arphdr) &&
678 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
679 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
680 		    if_name(ifp));
681 		return;
682 	}
683 	ar = mtod(m, struct arphdr *);
684 
685 	/* Check if length is sufficient */
686 	if (m->m_len <  arphdr_len(ar)) {
687 		m = m_pullup(m, arphdr_len(ar));
688 		if (m == NULL) {
689 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
690 			    if_name(ifp));
691 			return;
692 		}
693 		ar = mtod(m, struct arphdr *);
694 	}
695 
696 	hlen = 0;
697 	layer = "";
698 	switch (ntohs(ar->ar_hrd)) {
699 	case ARPHRD_ETHER:
700 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
701 		layer = "ethernet";
702 		break;
703 	case ARPHRD_INFINIBAND:
704 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
705 		layer = "infiniband";
706 		break;
707 	case ARPHRD_IEEE1394:
708 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
709 		layer = "firewire";
710 
711 		/*
712 		 * Restrict too long hardware addresses.
713 		 * Currently we are capable of handling 20-byte
714 		 * addresses ( sizeof(lle->ll_addr) )
715 		 */
716 		if (ar->ar_hln >= 20)
717 			hlen = 16;
718 		break;
719 	default:
720 		ARP_LOG(LOG_NOTICE,
721 		    "packet with unknown hardware format 0x%02d received on "
722 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
723 		m_freem(m);
724 		return;
725 	}
726 
727 	if (hlen != 0 && hlen != ar->ar_hln) {
728 		ARP_LOG(LOG_NOTICE,
729 		    "packet with invalid %s address length %d received on %s\n",
730 		    layer, ar->ar_hln, if_name(ifp));
731 		m_freem(m);
732 		return;
733 	}
734 
735 	ARPSTAT_INC(received);
736 	switch (ntohs(ar->ar_pro)) {
737 #ifdef INET
738 	case ETHERTYPE_IP:
739 		in_arpinput(m);
740 		return;
741 #endif
742 	}
743 	m_freem(m);
744 }
745 
746 #ifdef INET
747 /*
748  * ARP for Internet protocols on 10 Mb/s Ethernet.
749  * Algorithm is that given in RFC 826.
750  * In addition, a sanity check is performed on the sender
751  * protocol address, to catch impersonators.
752  * We no longer handle negotiations for use of trailer protocol:
753  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
754  * along with IP replies if we wanted trailers sent to us,
755  * and also sent them in response to IP replies.
756  * This allowed either end to announce the desire to receive
757  * trailer packets.
758  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
759  * but formerly didn't normally send requests.
760  */
761 static int log_arp_wrong_iface = 1;
762 static int log_arp_movements = 1;
763 static int log_arp_permanent_modify = 1;
764 static int allow_multicast = 0;
765 
766 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
767 	&log_arp_wrong_iface, 0,
768 	"log arp packets arriving on the wrong interface");
769 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
770 	&log_arp_movements, 0,
771 	"log arp replies from MACs different than the one in the cache");
772 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
773 	&log_arp_permanent_modify, 0,
774 	"log arp replies from MACs different than the one in the permanent arp entry");
775 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
776 	&allow_multicast, 0, "accept multicast addresses");
777 
778 static void
779 in_arpinput(struct mbuf *m)
780 {
781 	struct rm_priotracker in_ifa_tracker;
782 	struct arphdr *ah;
783 	struct ifnet *ifp = m->m_pkthdr.rcvif;
784 	struct llentry *la = NULL, *la_tmp;
785 	struct ifaddr *ifa;
786 	struct in_ifaddr *ia;
787 	struct sockaddr sa;
788 	struct in_addr isaddr, itaddr, myaddr;
789 	u_int8_t *enaddr = NULL;
790 	int op;
791 	int bridged = 0, is_bridge = 0;
792 	int carped;
793 	struct sockaddr_in sin;
794 	struct sockaddr *dst;
795 	struct nhop_object *nh;
796 	uint8_t linkhdr[LLE_MAX_LINKHDR];
797 	struct route ro;
798 	size_t linkhdrsize;
799 	int lladdr_off;
800 	int error;
801 	char addrbuf[INET_ADDRSTRLEN];
802 
803 	NET_EPOCH_ASSERT();
804 
805 	sin.sin_len = sizeof(struct sockaddr_in);
806 	sin.sin_family = AF_INET;
807 	sin.sin_addr.s_addr = 0;
808 
809 	if (ifp->if_bridge)
810 		bridged = 1;
811 	if (ifp->if_type == IFT_BRIDGE)
812 		is_bridge = 1;
813 
814 	/*
815 	 * We already have checked that mbuf contains enough contiguous data
816 	 * to hold entire arp message according to the arp header.
817 	 */
818 	ah = mtod(m, struct arphdr *);
819 
820 	/*
821 	 * ARP is only for IPv4 so we can reject packets with
822 	 * a protocol length not equal to an IPv4 address.
823 	 */
824 	if (ah->ar_pln != sizeof(struct in_addr)) {
825 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
826 		    sizeof(struct in_addr));
827 		goto drop;
828 	}
829 
830 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
831 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
832 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
833 		goto drop;
834 	}
835 
836 	op = ntohs(ah->ar_op);
837 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
838 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
839 
840 	if (op == ARPOP_REPLY)
841 		ARPSTAT_INC(rxreplies);
842 
843 	/*
844 	 * For a bridge, we want to check the address irrespective
845 	 * of the receive interface. (This will change slightly
846 	 * when we have clusters of interfaces).
847 	 */
848 	IN_IFADDR_RLOCK(&in_ifa_tracker);
849 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
850 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
851 		    ia->ia_ifp == ifp) &&
852 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
853 		    (ia->ia_ifa.ifa_carp == NULL ||
854 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
855 			ifa_ref(&ia->ia_ifa);
856 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
857 			goto match;
858 		}
859 	}
860 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
861 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
862 		    ia->ia_ifp == ifp) &&
863 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
864 			ifa_ref(&ia->ia_ifa);
865 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
866 			goto match;
867 		}
868 
869 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
870   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
871   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
872   addr == ia->ia_addr.sin_addr.s_addr)
873 	/*
874 	 * Check the case when bridge shares its MAC address with
875 	 * some of its children, so packets are claimed by bridge
876 	 * itself (bridge_input() does it first), but they are really
877 	 * meant to be destined to the bridge member.
878 	 */
879 	if (is_bridge) {
880 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
881 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
882 				ifa_ref(&ia->ia_ifa);
883 				ifp = ia->ia_ifp;
884 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
885 				goto match;
886 			}
887 		}
888 	}
889 #undef BDG_MEMBER_MATCHES_ARP
890 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
891 
892 	/*
893 	 * No match, use the first inet address on the receive interface
894 	 * as a dummy address for the rest of the function.
895 	 */
896 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
897 		if (ifa->ifa_addr->sa_family == AF_INET &&
898 		    (ifa->ifa_carp == NULL ||
899 		    (*carp_iamatch_p)(ifa, &enaddr))) {
900 			ia = ifatoia(ifa);
901 			ifa_ref(ifa);
902 			goto match;
903 		}
904 
905 	/*
906 	 * If bridging, fall back to using any inet address.
907 	 */
908 	IN_IFADDR_RLOCK(&in_ifa_tracker);
909 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
910 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
911 		goto drop;
912 	}
913 	ifa_ref(&ia->ia_ifa);
914 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
915 match:
916 	if (!enaddr)
917 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
918 	carped = (ia->ia_ifa.ifa_carp != NULL);
919 	myaddr = ia->ia_addr.sin_addr;
920 	ifa_free(&ia->ia_ifa);
921 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
922 		goto drop;	/* it's from me, ignore it. */
923 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
924 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
925 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
926 		goto drop;
927 	}
928 
929 	if (ifp->if_addrlen != ah->ar_hln) {
930 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
931 		    "i/f %d (ignored)\n", ifp->if_addrlen,
932 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
933 		    ifp->if_addrlen);
934 		goto drop;
935 	}
936 
937 	/*
938 	 * Warn if another host is using the same IP address, but only if the
939 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
940 	 * case we suppress the warning to avoid false positive complaints of
941 	 * potential misconfiguration.
942 	 */
943 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
944 	    myaddr.s_addr != 0) {
945 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
946 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
947 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
948 		itaddr = myaddr;
949 		ARPSTAT_INC(dupips);
950 		goto reply;
951 	}
952 	if (ifp->if_flags & IFF_STATICARP)
953 		goto reply;
954 
955 	bzero(&sin, sizeof(sin));
956 	sin.sin_len = sizeof(struct sockaddr_in);
957 	sin.sin_family = AF_INET;
958 	sin.sin_addr = isaddr;
959 	dst = (struct sockaddr *)&sin;
960 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
961 	if (la != NULL)
962 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
963 	else if (itaddr.s_addr == myaddr.s_addr) {
964 		/*
965 		 * Request/reply to our address, but no lle exists yet.
966 		 * Calculate full link prepend to use in lle.
967 		 */
968 		linkhdrsize = sizeof(linkhdr);
969 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
970 		    &linkhdrsize, &lladdr_off) != 0)
971 			goto reply;
972 
973 		/* Allocate new entry */
974 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
975 		if (la == NULL) {
976 			/*
977 			 * lle creation may fail if source address belongs
978 			 * to non-directly connected subnet. However, we
979 			 * will try to answer the request instead of dropping
980 			 * frame.
981 			 */
982 			goto reply;
983 		}
984 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
985 		    lladdr_off);
986 
987 		IF_AFDATA_WLOCK(ifp);
988 		LLE_WLOCK(la);
989 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
990 
991 		/*
992 		 * Check if lle still does not exists.
993 		 * If it does, that means that we either
994 		 * 1) have configured it explicitly, via
995 		 * 1a) 'arp -s' static entry or
996 		 * 1b) interface address static record
997 		 * or
998 		 * 2) it was the result of sending first packet to-host
999 		 * or
1000 		 * 3) it was another arp reply packet we handled in
1001 		 * different thread.
1002 		 *
1003 		 * In all cases except 3) we definitely need to prefer
1004 		 * existing lle. For the sake of simplicity, prefer any
1005 		 * existing lle over newly-create one.
1006 		 */
1007 		if (la_tmp == NULL)
1008 			lltable_link_entry(LLTABLE(ifp), la);
1009 		IF_AFDATA_WUNLOCK(ifp);
1010 
1011 		if (la_tmp == NULL) {
1012 			arp_mark_lle_reachable(la);
1013 			LLE_WUNLOCK(la);
1014 		} else {
1015 			/* Free newly-create entry and handle packet */
1016 			lltable_free_entry(LLTABLE(ifp), la);
1017 			la = la_tmp;
1018 			la_tmp = NULL;
1019 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1020 			/* arp_check_update_lle() returns @la unlocked */
1021 		}
1022 		la = NULL;
1023 	}
1024 reply:
1025 	if (op != ARPOP_REQUEST)
1026 		goto drop;
1027 	ARPSTAT_INC(rxrequests);
1028 
1029 	if (itaddr.s_addr == myaddr.s_addr) {
1030 		/* Shortcut.. the receiving interface is the target. */
1031 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1032 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1033 	} else {
1034 		/*
1035 		 * Destination address is not ours. Check if
1036 		 * proxyarp entry exists or proxyarp is turned on globally.
1037 		 */
1038 		struct llentry *lle;
1039 
1040 		sin.sin_addr = itaddr;
1041 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1042 
1043 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1044 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1045 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1046 			LLE_RUNLOCK(lle);
1047 		} else {
1048 			if (lle != NULL)
1049 				LLE_RUNLOCK(lle);
1050 
1051 			if (!V_arp_proxyall)
1052 				goto drop;
1053 
1054 			NET_EPOCH_ASSERT();
1055 			nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1056 			if (nh == NULL)
1057 				goto drop;
1058 
1059 			/*
1060 			 * Don't send proxies for nodes on the same interface
1061 			 * as this one came out of, or we'll get into a fight
1062 			 * over who claims what Ether address.
1063 			 */
1064 			if (nh->nh_ifp == ifp)
1065 				goto drop;
1066 
1067 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1068 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1069 
1070 			/*
1071 			 * Also check that the node which sent the ARP packet
1072 			 * is on the interface we expect it to be on. This
1073 			 * avoids ARP chaos if an interface is connected to the
1074 			 * wrong network.
1075 			 */
1076 
1077 			nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1078 			if (nh == NULL)
1079 				goto drop;
1080 			if (nh->nh_ifp != ifp) {
1081 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1082 				    " from %s via %s\n",
1083 				    inet_ntoa_r(isaddr, addrbuf),
1084 				    ifp->if_xname);
1085 				goto drop;
1086 			}
1087 
1088 #ifdef DEBUG_PROXY
1089 			printf("arp: proxying for %s\n",
1090 			    inet_ntoa_r(itaddr, addrbuf));
1091 #endif
1092 		}
1093 	}
1094 
1095 	if (itaddr.s_addr == myaddr.s_addr &&
1096 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1097 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1098 #ifdef DEBUG_LINKLOCAL
1099 		printf("arp: sending reply for link-local addr %s\n",
1100 		    inet_ntoa_r(itaddr, addrbuf));
1101 #endif
1102 		m->m_flags |= M_BCAST;
1103 		m->m_flags &= ~M_MCAST;
1104 	} else {
1105 		/* default behaviour; never reply by broadcast. */
1106 		m->m_flags &= ~(M_BCAST|M_MCAST);
1107 	}
1108 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1109 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1110 	ah->ar_op = htons(ARPOP_REPLY);
1111 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1112 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1113 	m->m_pkthdr.len = m->m_len;
1114 	m->m_pkthdr.rcvif = NULL;
1115 	sa.sa_family = AF_ARP;
1116 	sa.sa_len = 2;
1117 
1118 	/* Calculate link header for sending frame */
1119 	bzero(&ro, sizeof(ro));
1120 	linkhdrsize = sizeof(linkhdr);
1121 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1122 
1123 	/*
1124 	 * arp_fillheader() may fail due to lack of support inside encap request
1125 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1126 	 * by if_output().
1127 	 */
1128 	if (error != 0 && error != EAFNOSUPPORT) {
1129 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1130 		    if_name(ifp), error);
1131 		return;
1132 	}
1133 
1134 	ro.ro_prepend = linkhdr;
1135 	ro.ro_plen = linkhdrsize;
1136 	ro.ro_flags = 0;
1137 
1138 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1139 	(*ifp->if_output)(ifp, m, &sa, &ro);
1140 	ARPSTAT_INC(txreplies);
1141 	return;
1142 
1143 drop:
1144 	m_freem(m);
1145 }
1146 #endif
1147 
1148 static struct mbuf *
1149 arp_grab_holdchain(struct llentry *la)
1150 {
1151 	struct mbuf *chain;
1152 
1153 	LLE_WLOCK_ASSERT(la);
1154 
1155 	chain = la->la_hold;
1156 	la->la_hold = NULL;
1157 	la->la_numheld = 0;
1158 
1159 	return (chain);
1160 }
1161 
1162 static void
1163 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1164 {
1165 	struct mbuf *m_hold, *m_hold_next;
1166 	struct sockaddr_in sin;
1167 
1168 	NET_EPOCH_ASSERT();
1169 
1170 	struct route ro = {
1171 		.ro_prepend = la->r_linkdata,
1172 		.ro_plen = la->r_hdrlen,
1173 	};
1174 
1175 	lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1176 
1177 	for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1178 		m_hold_next = m_hold->m_nextpkt;
1179 		m_hold->m_nextpkt = NULL;
1180 		/* Avoid confusing lower layers. */
1181 		m_clrprotoflags(m_hold);
1182 		(*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1183 	}
1184 }
1185 
1186 /*
1187  * Checks received arp data against existing @la.
1188  * Updates lle state/performs notification if necessary.
1189  */
1190 static void
1191 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1192     int bridged, struct llentry *la)
1193 {
1194 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1195 	size_t linkhdrsize;
1196 	int lladdr_off;
1197 	char addrbuf[INET_ADDRSTRLEN];
1198 
1199 	LLE_WLOCK_ASSERT(la);
1200 
1201 	/* the following is not an error when doing bridging */
1202 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1203 		if (log_arp_wrong_iface)
1204 			ARP_LOG(LOG_WARNING, "%s is on %s "
1205 			    "but got reply from %*D on %s\n",
1206 			    inet_ntoa_r(isaddr, addrbuf),
1207 			    la->lle_tbl->llt_ifp->if_xname,
1208 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1209 			    ifp->if_xname);
1210 		LLE_WUNLOCK(la);
1211 		return;
1212 	}
1213 	if ((la->la_flags & LLE_VALID) &&
1214 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1215 		if (la->la_flags & LLE_STATIC) {
1216 			LLE_WUNLOCK(la);
1217 			if (log_arp_permanent_modify)
1218 				ARP_LOG(LOG_ERR,
1219 				    "%*D attempts to modify "
1220 				    "permanent entry for %s on %s\n",
1221 				    ifp->if_addrlen,
1222 				    (u_char *)ar_sha(ah), ":",
1223 				    inet_ntoa_r(isaddr, addrbuf),
1224 				    ifp->if_xname);
1225 			return;
1226 		}
1227 		if (log_arp_movements) {
1228 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1229 			    "to %*D on %s\n",
1230 			    inet_ntoa_r(isaddr, addrbuf),
1231 			    ifp->if_addrlen,
1232 			    (u_char *)la->ll_addr, ":",
1233 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1234 			    ifp->if_xname);
1235 		}
1236 	}
1237 
1238 	/* Calculate full link prepend to use in lle */
1239 	linkhdrsize = sizeof(linkhdr);
1240 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1241 	    &linkhdrsize, &lladdr_off) != 0)
1242 		return;
1243 
1244 	/* Check if something has changed */
1245 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1246 	    (la->la_flags & LLE_VALID) == 0) {
1247 		/* Try to perform LLE update */
1248 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1249 		    lladdr_off) == 0)
1250 			return;
1251 
1252 		/* Clear fast path feedback request if set */
1253 		llentry_mark_used(la);
1254 	}
1255 
1256 	arp_mark_lle_reachable(la);
1257 
1258 	/*
1259 	 * The packets are all freed within the call to the output
1260 	 * routine.
1261 	 *
1262 	 * NB: The lock MUST be released before the call to the
1263 	 * output routine.
1264 	 */
1265 	if (la->la_hold != NULL) {
1266 		struct mbuf *chain;
1267 
1268 		chain = arp_grab_holdchain(la);
1269 		LLE_WUNLOCK(la);
1270 		arp_flush_holdchain(ifp, la, chain);
1271 	} else
1272 		LLE_WUNLOCK(la);
1273 }
1274 
1275 static void
1276 arp_mark_lle_reachable(struct llentry *la)
1277 {
1278 	int canceled, wtime;
1279 
1280 	LLE_WLOCK_ASSERT(la);
1281 
1282 	la->ln_state = ARP_LLINFO_REACHABLE;
1283 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1284 
1285 	if (!(la->la_flags & LLE_STATIC)) {
1286 		LLE_ADDREF(la);
1287 		la->la_expire = time_uptime + V_arpt_keep;
1288 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1289 		if (wtime < 0)
1290 			wtime = V_arpt_keep;
1291 		canceled = callout_reset(&la->lle_timer,
1292 		    hz * wtime, arptimer, la);
1293 		if (canceled)
1294 			LLE_REMREF(la);
1295 	}
1296 	la->la_asked = 0;
1297 	la->la_preempt = V_arp_maxtries;
1298 }
1299 
1300 /*
1301  * Add permanent link-layer record for given interface address.
1302  */
1303 static __noinline void
1304 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1305 {
1306 	struct llentry *lle, *lle_tmp;
1307 
1308 	/*
1309 	 * Interface address LLE record is considered static
1310 	 * because kernel code relies on LLE_STATIC flag to check
1311 	 * if these entries can be rewriten by arp updates.
1312 	 */
1313 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1314 	if (lle == NULL) {
1315 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1316 		    "entry for interface address\n");
1317 		return;
1318 	}
1319 
1320 	IF_AFDATA_WLOCK(ifp);
1321 	LLE_WLOCK(lle);
1322 	/* Unlink any entry if exists */
1323 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1324 	if (lle_tmp != NULL)
1325 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1326 
1327 	lltable_link_entry(LLTABLE(ifp), lle);
1328 	IF_AFDATA_WUNLOCK(ifp);
1329 
1330 	if (lle_tmp != NULL)
1331 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1332 
1333 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1334 	LLE_WUNLOCK(lle);
1335 	if (lle_tmp != NULL)
1336 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1337 }
1338 
1339 /*
1340  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1341  * of valid values.
1342  */
1343 static int
1344 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1345 {
1346 	int error;
1347 	int rexmit_count = *(int *)arg1;
1348 
1349 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1350 
1351 	/* Enforce limits on any new value that may have been set. */
1352 	if (!error && req->newptr) {
1353 		/* A new value was set. */
1354 		if (rexmit_count < 0) {
1355 			rexmit_count = 0;
1356 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1357 			rexmit_count = MAX_GARP_RETRANSMITS;
1358 		}
1359 		*(int *)arg1 = rexmit_count;
1360 	}
1361 
1362 	return (error);
1363 }
1364 
1365 /*
1366  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1367  * retransmit it again. A pending callout owns a reference to the ifa.
1368  */
1369 static void
1370 garp_rexmit(void *arg)
1371 {
1372 	struct in_ifaddr *ia = arg;
1373 
1374 	if (callout_pending(&ia->ia_garp_timer) ||
1375 	    !callout_active(&ia->ia_garp_timer)) {
1376 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1377 		ifa_free(&ia->ia_ifa);
1378 		return;
1379 	}
1380 
1381 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1382 
1383 	/*
1384 	 * Drop lock while the ARP request is generated.
1385 	 */
1386 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1387 
1388 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1389 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1390 
1391 	/*
1392 	 * Increment the count of retransmissions. If the count has reached the
1393 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1394 	 * the callout to retransmit another GARP packet.
1395 	 */
1396 	++ia->ia_garp_count;
1397 	if (ia->ia_garp_count >= garp_rexmit_count) {
1398 		ifa_free(&ia->ia_ifa);
1399 	} else {
1400 		int rescheduled;
1401 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1402 		rescheduled = callout_reset(&ia->ia_garp_timer,
1403 		    (1 << ia->ia_garp_count) * hz,
1404 		    garp_rexmit, ia);
1405 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1406 		if (rescheduled) {
1407 			ifa_free(&ia->ia_ifa);
1408 		}
1409 	}
1410 
1411 	CURVNET_RESTORE();
1412 }
1413 
1414 /*
1415  * Start the GARP retransmit timer.
1416  *
1417  * A single GARP is always transmitted when an IPv4 address is added
1418  * to an interface and that is usually sufficient. However, in some
1419  * circumstances, such as when a shared address is passed between
1420  * cluster nodes, this single GARP may occasionally be dropped or
1421  * lost. This can lead to neighbors on the network link working with a
1422  * stale ARP cache and sending packets destined for that address to
1423  * the node that previously owned the address, which may not respond.
1424  *
1425  * To avoid this situation, GARP retransmits can be enabled by setting
1426  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1427  * than zero. The setting represents the maximum number of
1428  * retransmissions. The interval between retransmissions is calculated
1429  * using an exponential backoff algorithm, doubling each time, so the
1430  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1431  */
1432 static void
1433 garp_timer_start(struct ifaddr *ifa)
1434 {
1435 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1436 
1437 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1438 	ia->ia_garp_count = 0;
1439 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1440 	    garp_rexmit, ia) == 0) {
1441 		ifa_ref(ifa);
1442 	}
1443 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1444 }
1445 
1446 void
1447 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1448 {
1449 	struct epoch_tracker et;
1450 	const struct sockaddr_in *dst_in;
1451 	const struct sockaddr *dst;
1452 
1453 	if (ifa->ifa_carp != NULL)
1454 		return;
1455 
1456 	dst = ifa->ifa_addr;
1457 	dst_in = (const struct sockaddr_in *)dst;
1458 
1459 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1460 		return;
1461 	NET_EPOCH_ENTER(et);
1462 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1463 	NET_EPOCH_EXIT(et);
1464 	if (garp_rexmit_count > 0) {
1465 		garp_timer_start(ifa);
1466 	}
1467 
1468 	arp_add_ifa_lle(ifp, dst);
1469 }
1470 
1471 void
1472 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1473 {
1474 
1475 	if (ntohl(addr.s_addr) != INADDR_ANY)
1476 		arprequest(ifp, &addr, &addr, enaddr);
1477 }
1478 
1479 /*
1480  * Sends gratuitous ARPs for each ifaddr to notify other
1481  * nodes about the address change.
1482  */
1483 static __noinline void
1484 arp_handle_ifllchange(struct ifnet *ifp)
1485 {
1486 	struct ifaddr *ifa;
1487 
1488 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1489 		if (ifa->ifa_addr->sa_family == AF_INET)
1490 			arp_ifinit(ifp, ifa);
1491 	}
1492 }
1493 
1494 /*
1495  * A handler for interface link layer address change event.
1496  */
1497 static void
1498 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1499 {
1500 	/* if_bridge can update its lladdr during if_vmove(), after we've done
1501 	 * if_detach_internal()/dom_ifdetach(). */
1502 	if (ifp->if_afdata[AF_INET] == NULL)
1503 		return;
1504 
1505 	lltable_update_ifaddr(LLTABLE(ifp));
1506 
1507 	if ((ifp->if_flags & IFF_UP) != 0)
1508 		arp_handle_ifllchange(ifp);
1509 }
1510 
1511 static void
1512 vnet_arp_init(void)
1513 {
1514 
1515 	if (IS_DEFAULT_VNET(curvnet)) {
1516 		netisr_register(&arp_nh);
1517 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1518 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1519 	}
1520 #ifdef VIMAGE
1521 	else
1522 		netisr_register_vnet(&arp_nh);
1523 #endif
1524 }
1525 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1526     vnet_arp_init, 0);
1527 
1528 #ifdef VIMAGE
1529 /*
1530  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1531  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1532  */
1533 static void
1534 vnet_arp_destroy(__unused void *arg)
1535 {
1536 
1537 	netisr_unregister_vnet(&arp_nh);
1538 }
1539 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1540     vnet_arp_destroy, NULL);
1541 #endif
1542