xref: /freebsd/sys/netinet/if_ether.c (revision 349cc55c9796c4596a5b9904cd3281af295f878f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
32  */
33 
34 /*
35  * Ethernet address resolution protocol.
36  * TODO:
37  *	add "inuse/lock" bit (or ref. count) along with valid bit
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_inet.h"
44 
45 #include <sys/param.h>
46 #include <sys/eventhandler.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/socket.h>
56 #include <sys/syslog.h>
57 
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/netisr.h>
63 #include <net/ethernet.h>
64 #include <net/route.h>
65 #include <net/route/nhop.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_fib.h>
70 #include <netinet/in_var.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #ifdef INET
74 #include <netinet/ip_carp.h>
75 #endif
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define SIN(s) ((const struct sockaddr_in *)(s))
80 
81 static struct timeval arp_lastlog;
82 static int arp_curpps;
83 static int arp_maxpps = 1;
84 
85 /* Simple ARP state machine */
86 enum arp_llinfo_state {
87 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
88 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
89 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
90 	ARP_LLINFO_DELETED,	/* LLE is deleted */
91 };
92 
93 SYSCTL_DECL(_net_link_ether);
94 static SYSCTL_NODE(_net_link_ether, PF_INET, inet,
95     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
96     "");
97 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp,
98     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
99     "");
100 
101 /* timer values */
102 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60);	/* once resolved, good for 20
103 						 * minutes */
104 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
105 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
106 VNET_DEFINE_STATIC(int, arpt_down) = 20;	/* keep incomplete entries for
107 						 * 20 seconds */
108 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
109 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
110 VNET_PCPUSTAT_SYSINIT(arpstat);
111 
112 #ifdef VIMAGE
113 VNET_PCPUSTAT_SYSUNINIT(arpstat);
114 #endif /* VIMAGE */
115 
116 VNET_DEFINE_STATIC(int, arp_maxhold) = 16;
117 
118 #define	V_arpt_keep		VNET(arpt_keep)
119 #define	V_arpt_down		VNET(arpt_down)
120 #define	V_arpt_rexmit		VNET(arpt_rexmit)
121 #define	V_arp_maxtries		VNET(arp_maxtries)
122 #define	V_arp_proxyall		VNET(arp_proxyall)
123 #define	V_arp_maxhold		VNET(arp_maxhold)
124 
125 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
126 	&VNET_NAME(arpt_keep), 0,
127 	"ARP entry lifetime in seconds");
128 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
129 	&VNET_NAME(arp_maxtries), 0,
130 	"ARP resolution attempts before returning error");
131 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
132 	&VNET_NAME(arp_proxyall), 0,
133 	"Enable proxy ARP for all suitable requests");
134 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
135 	&VNET_NAME(arpt_down), 0,
136 	"Incomplete ARP entry lifetime in seconds");
137 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
138     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
139 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
140 	&VNET_NAME(arp_maxhold), 0,
141 	"Number of packets to hold per ARP entry");
142 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
143 	CTLFLAG_RW, &arp_maxpps, 0,
144 	"Maximum number of remotely triggered ARP messages that can be "
145 	"logged per second");
146 
147 /*
148  * Due to the exponential backoff algorithm used for the interval between GARP
149  * retransmissions, the maximum number of retransmissions is limited for
150  * sanity. This limit corresponds to a maximum interval between retransmissions
151  * of 2^16 seconds ~= 18 hours.
152  *
153  * Making this limit more dynamic is more complicated than worthwhile,
154  * especially since sending out GARPs spaced days apart would be of little
155  * use. A maximum dynamic limit would look something like:
156  *
157  * const int max = fls(INT_MAX / hz) - 1;
158  */
159 #define MAX_GARP_RETRANSMITS 16
160 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
161 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
162 
163 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
164     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
165     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
166     "Number of times to retransmit GARP packets;"
167     " 0 to disable, maximum of 16");
168 
169 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO;	/* Min. log(9) level. */
170 #define	V_arp_log_level		VNET(arp_log_level)
171 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
172 	&VNET_NAME(arp_log_level), 0,
173 	"Minimum log(9) level for recording rate limited arp log messages. "
174 	"The higher will be log more (emerg=0, info=6 (default), debug=7).");
175 #define	ARP_LOG(pri, ...)	do {					\
176 	if ((pri) <= V_arp_log_level &&					\
177 	    ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
178 		log((pri), "arp: " __VA_ARGS__);			\
179 } while (0)
180 
181 static void	arpintr(struct mbuf *);
182 static void	arptimer(void *);
183 #ifdef INET
184 static void	in_arpinput(struct mbuf *);
185 #endif
186 
187 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
188     struct ifnet *ifp, int bridged, struct llentry *la);
189 static void arp_mark_lle_reachable(struct llentry *la);
190 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
191 
192 static eventhandler_tag iflladdr_tag;
193 
194 static const struct netisr_handler arp_nh = {
195 	.nh_name = "arp",
196 	.nh_handler = arpintr,
197 	.nh_proto = NETISR_ARP,
198 	.nh_policy = NETISR_POLICY_SOURCE,
199 };
200 
201 /*
202  * Timeout routine.  Age arp_tab entries periodically.
203  */
204 static void
205 arptimer(void *arg)
206 {
207 	struct llentry *lle = (struct llentry *)arg;
208 	struct ifnet *ifp;
209 
210 	if (lle->la_flags & LLE_STATIC) {
211 		return;
212 	}
213 	LLE_WLOCK(lle);
214 	if (callout_pending(&lle->lle_timer)) {
215 		/*
216 		 * Here we are a bit odd here in the treatment of
217 		 * active/pending. If the pending bit is set, it got
218 		 * rescheduled before I ran. The active
219 		 * bit we ignore, since if it was stopped
220 		 * in ll_tablefree() and was currently running
221 		 * it would have return 0 so the code would
222 		 * not have deleted it since the callout could
223 		 * not be stopped so we want to go through
224 		 * with the delete here now. If the callout
225 		 * was restarted, the pending bit will be back on and
226 		 * we just want to bail since the callout_reset would
227 		 * return 1 and our reference would have been removed
228 		 * by arpresolve() below.
229 		 */
230 		LLE_WUNLOCK(lle);
231  		return;
232  	}
233 	ifp = lle->lle_tbl->llt_ifp;
234 	CURVNET_SET(ifp->if_vnet);
235 
236 	switch (lle->ln_state) {
237 	case ARP_LLINFO_REACHABLE:
238 
239 		/*
240 		 * Expiration time is approaching.
241 		 * Request usage feedback from the datapath.
242 		 * Change state and re-schedule ourselves.
243 		 */
244 		llentry_request_feedback(lle);
245 		lle->ln_state = ARP_LLINFO_VERIFY;
246 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
247 		LLE_WUNLOCK(lle);
248 		CURVNET_RESTORE();
249 		return;
250 	case ARP_LLINFO_VERIFY:
251 		if (llentry_get_hittime(lle) > 0 && lle->la_preempt > 0) {
252 			/* Entry was used, issue refresh request */
253 			struct epoch_tracker et;
254 			struct in_addr dst;
255 
256 			dst = lle->r_l3addr.addr4;
257 			lle->la_preempt--;
258 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
259 			LLE_WUNLOCK(lle);
260 			NET_EPOCH_ENTER(et);
261 			arprequest(ifp, NULL, &dst, NULL);
262 			NET_EPOCH_EXIT(et);
263 			CURVNET_RESTORE();
264 			return;
265 		}
266 		/* Nothing happened. Reschedule if not too late */
267 		if (lle->la_expire > time_uptime) {
268 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
269 			LLE_WUNLOCK(lle);
270 			CURVNET_RESTORE();
271 			return;
272 		}
273 		break;
274 	case ARP_LLINFO_INCOMPLETE:
275 	case ARP_LLINFO_DELETED:
276 		break;
277 	}
278 
279 	if ((lle->la_flags & LLE_DELETED) == 0) {
280 		int evt;
281 
282 		if (lle->la_flags & LLE_VALID)
283 			evt = LLENTRY_EXPIRED;
284 		else
285 			evt = LLENTRY_TIMEDOUT;
286 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
287 	}
288 
289 	callout_stop(&lle->lle_timer);
290 
291 	/* XXX: LOR avoidance. We still have ref on lle. */
292 	LLE_WUNLOCK(lle);
293 	IF_AFDATA_LOCK(ifp);
294 	LLE_WLOCK(lle);
295 
296 	/* Guard against race with other llentry_free(). */
297 	if (lle->la_flags & LLE_LINKED) {
298 		LLE_REMREF(lle);
299 		lltable_unlink_entry(lle->lle_tbl, lle);
300 	}
301 	IF_AFDATA_UNLOCK(ifp);
302 
303 	size_t pkts_dropped = llentry_free(lle);
304 
305 	ARPSTAT_ADD(dropped, pkts_dropped);
306 	ARPSTAT_INC(timeouts);
307 
308 	CURVNET_RESTORE();
309 }
310 
311 /*
312  * Stores link-layer header for @ifp in format suitable for if_output()
313  * into buffer @buf. Resulting header length is stored in @bufsize.
314  *
315  * Returns 0 on success.
316  */
317 static int
318 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
319     size_t *bufsize)
320 {
321 	struct if_encap_req ereq;
322 	int error;
323 
324 	bzero(buf, *bufsize);
325 	bzero(&ereq, sizeof(ereq));
326 	ereq.buf = buf;
327 	ereq.bufsize = *bufsize;
328 	ereq.rtype = IFENCAP_LL;
329 	ereq.family = AF_ARP;
330 	ereq.lladdr = ar_tha(ah);
331 	ereq.hdata = (u_char *)ah;
332 	if (bcast)
333 		ereq.flags = IFENCAP_FLAG_BROADCAST;
334 	error = ifp->if_requestencap(ifp, &ereq);
335 	if (error == 0)
336 		*bufsize = ereq.bufsize;
337 
338 	return (error);
339 }
340 
341 /*
342  * Broadcast an ARP request. Caller specifies:
343  *	- arp header source ip address
344  *	- arp header target ip address
345  *	- arp header source ethernet address
346  */
347 static int
348 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
349     const struct in_addr *tip, u_char *enaddr)
350 {
351 	struct mbuf *m;
352 	struct arphdr *ah;
353 	struct sockaddr sa;
354 	u_char *carpaddr = NULL;
355 	uint8_t linkhdr[LLE_MAX_LINKHDR];
356 	size_t linkhdrsize;
357 	struct route ro;
358 	int error;
359 
360 	NET_EPOCH_ASSERT();
361 
362 	if (sip == NULL) {
363 		/*
364 		 * The caller did not supply a source address, try to find
365 		 * a compatible one among those assigned to this interface.
366 		 */
367 		struct ifaddr *ifa;
368 
369 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
370 			if (ifa->ifa_addr->sa_family != AF_INET)
371 				continue;
372 
373 			if (ifa->ifa_carp) {
374 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
375 					continue;
376 				sip = &IA_SIN(ifa)->sin_addr;
377 			} else {
378 				carpaddr = NULL;
379 				sip = &IA_SIN(ifa)->sin_addr;
380 			}
381 
382 			if (0 == ((sip->s_addr ^ tip->s_addr) &
383 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
384 				break;  /* found it. */
385 		}
386 		if (sip == NULL) {
387 			printf("%s: cannot find matching address\n", __func__);
388 			return (EADDRNOTAVAIL);
389 		}
390 	}
391 	if (enaddr == NULL)
392 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
393 
394 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
395 		return (ENOMEM);
396 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
397 		2 * ifp->if_addrlen;
398 	m->m_pkthdr.len = m->m_len;
399 	M_ALIGN(m, m->m_len);
400 	ah = mtod(m, struct arphdr *);
401 	bzero((caddr_t)ah, m->m_len);
402 #ifdef MAC
403 	mac_netinet_arp_send(ifp, m);
404 #endif
405 	ah->ar_pro = htons(ETHERTYPE_IP);
406 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
407 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
408 	ah->ar_op = htons(ARPOP_REQUEST);
409 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
410 	bcopy(sip, ar_spa(ah), ah->ar_pln);
411 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
412 	sa.sa_family = AF_ARP;
413 	sa.sa_len = 2;
414 
415 	/* Calculate link header for sending frame */
416 	bzero(&ro, sizeof(ro));
417 	linkhdrsize = sizeof(linkhdr);
418 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
419 	if (error != 0 && error != EAFNOSUPPORT) {
420 		m_freem(m);
421 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
422 		    if_name(ifp), error);
423 		return (error);
424 	}
425 
426 	ro.ro_prepend = linkhdr;
427 	ro.ro_plen = linkhdrsize;
428 	ro.ro_flags = 0;
429 
430 	m->m_flags |= M_BCAST;
431 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
432 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
433 	ARPSTAT_INC(txrequests);
434 	if (error) {
435 		ARPSTAT_INC(txerrors);
436 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
437 		    if_name(ifp), error);
438 	}
439 	return (error);
440 }
441 
442 void
443 arprequest(struct ifnet *ifp, const struct in_addr *sip,
444     const struct in_addr *tip, u_char *enaddr)
445 {
446 
447 	(void) arprequest_internal(ifp, sip, tip, enaddr);
448 }
449 
450 /*
451  * Resolve an IP address into an ethernet address - heavy version.
452  * Used internally by arpresolve().
453  * We have already checked that we can't use an existing lle without
454  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
455  *
456  * On success, desten and pflags are filled in and the function returns 0;
457  * If the packet must be held pending resolution, we return EWOULDBLOCK
458  * On other errors, we return the corresponding error code.
459  * Note that m_freem() handles NULL.
460  */
461 static int
462 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
463 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
464 	struct llentry **plle)
465 {
466 	struct llentry *la = NULL, *la_tmp;
467 	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 arphdr *ah;
782 	struct ifnet *ifp = m->m_pkthdr.rcvif;
783 	struct llentry *la = NULL, *la_tmp;
784 	struct ifaddr *ifa;
785 	struct in_ifaddr *ia;
786 	struct sockaddr sa;
787 	struct in_addr isaddr, itaddr, myaddr;
788 	u_int8_t *enaddr = NULL;
789 	int op;
790 	int bridged = 0, is_bridge = 0;
791 	int carped;
792 	struct sockaddr_in sin;
793 	struct sockaddr *dst;
794 	struct nhop_object *nh;
795 	uint8_t linkhdr[LLE_MAX_LINKHDR];
796 	struct route ro;
797 	size_t linkhdrsize;
798 	int lladdr_off;
799 	int error;
800 	char addrbuf[INET_ADDRSTRLEN];
801 
802 	NET_EPOCH_ASSERT();
803 
804 	sin.sin_len = sizeof(struct sockaddr_in);
805 	sin.sin_family = AF_INET;
806 	sin.sin_addr.s_addr = 0;
807 
808 	if (ifp->if_bridge)
809 		bridged = 1;
810 	if (ifp->if_type == IFT_BRIDGE)
811 		is_bridge = 1;
812 
813 	/*
814 	 * We already have checked that mbuf contains enough contiguous data
815 	 * to hold entire arp message according to the arp header.
816 	 */
817 	ah = mtod(m, struct arphdr *);
818 
819 	/*
820 	 * ARP is only for IPv4 so we can reject packets with
821 	 * a protocol length not equal to an IPv4 address.
822 	 */
823 	if (ah->ar_pln != sizeof(struct in_addr)) {
824 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
825 		    sizeof(struct in_addr));
826 		goto drop;
827 	}
828 
829 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
830 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
831 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
832 		goto drop;
833 	}
834 
835 	op = ntohs(ah->ar_op);
836 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
837 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
838 
839 	if (op == ARPOP_REPLY)
840 		ARPSTAT_INC(rxreplies);
841 
842 	/*
843 	 * For a bridge, we want to check the address irrespective
844 	 * of the receive interface. (This will change slightly
845 	 * when we have clusters of interfaces).
846 	 */
847 	CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
848 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
849 		    ia->ia_ifp == ifp) &&
850 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
851 		    (ia->ia_ifa.ifa_carp == NULL ||
852 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
853 			ifa_ref(&ia->ia_ifa);
854 			goto match;
855 		}
856 	}
857 	CK_LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
858 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
859 		    ia->ia_ifp == ifp) &&
860 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
861 			ifa_ref(&ia->ia_ifa);
862 			goto match;
863 		}
864 
865 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
866   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
867   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
868   addr == ia->ia_addr.sin_addr.s_addr)
869 	/*
870 	 * Check the case when bridge shares its MAC address with
871 	 * some of its children, so packets are claimed by bridge
872 	 * itself (bridge_input() does it first), but they are really
873 	 * meant to be destined to the bridge member.
874 	 */
875 	if (is_bridge) {
876 		CK_LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
877 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
878 				ifa_ref(&ia->ia_ifa);
879 				ifp = ia->ia_ifp;
880 				goto match;
881 			}
882 		}
883 	}
884 #undef BDG_MEMBER_MATCHES_ARP
885 
886 	/*
887 	 * No match, use the first inet address on the receive interface
888 	 * as a dummy address for the rest of the function.
889 	 */
890 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
891 		if (ifa->ifa_addr->sa_family == AF_INET &&
892 		    (ifa->ifa_carp == NULL ||
893 		    (*carp_iamatch_p)(ifa, &enaddr))) {
894 			ia = ifatoia(ifa);
895 			ifa_ref(ifa);
896 			goto match;
897 		}
898 
899 	/*
900 	 * If bridging, fall back to using any inet address.
901 	 */
902 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL)
903 		goto drop;
904 	ifa_ref(&ia->ia_ifa);
905 match:
906 	if (!enaddr)
907 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
908 	carped = (ia->ia_ifa.ifa_carp != NULL);
909 	myaddr = ia->ia_addr.sin_addr;
910 	ifa_free(&ia->ia_ifa);
911 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
912 		goto drop;	/* it's from me, ignore it. */
913 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
914 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
915 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
916 		goto drop;
917 	}
918 
919 	if (ifp->if_addrlen != ah->ar_hln) {
920 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
921 		    "i/f %d (ignored)\n", ifp->if_addrlen,
922 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
923 		    ifp->if_addrlen);
924 		goto drop;
925 	}
926 
927 	/*
928 	 * Warn if another host is using the same IP address, but only if the
929 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
930 	 * case we suppress the warning to avoid false positive complaints of
931 	 * potential misconfiguration.
932 	 */
933 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
934 	    myaddr.s_addr != 0) {
935 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
936 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
937 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
938 		itaddr = myaddr;
939 		ARPSTAT_INC(dupips);
940 		goto reply;
941 	}
942 	if (ifp->if_flags & IFF_STATICARP)
943 		goto reply;
944 
945 	bzero(&sin, sizeof(sin));
946 	sin.sin_len = sizeof(struct sockaddr_in);
947 	sin.sin_family = AF_INET;
948 	sin.sin_addr = isaddr;
949 	dst = (struct sockaddr *)&sin;
950 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
951 	if (la != NULL)
952 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
953 	else if (itaddr.s_addr == myaddr.s_addr) {
954 		/*
955 		 * Request/reply to our address, but no lle exists yet.
956 		 * Calculate full link prepend to use in lle.
957 		 */
958 		linkhdrsize = sizeof(linkhdr);
959 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
960 		    &linkhdrsize, &lladdr_off) != 0)
961 			goto reply;
962 
963 		/* Allocate new entry */
964 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
965 		if (la == NULL) {
966 			/*
967 			 * lle creation may fail if source address belongs
968 			 * to non-directly connected subnet. However, we
969 			 * will try to answer the request instead of dropping
970 			 * frame.
971 			 */
972 			goto reply;
973 		}
974 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
975 		    lladdr_off);
976 
977 		IF_AFDATA_WLOCK(ifp);
978 		LLE_WLOCK(la);
979 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
980 
981 		/*
982 		 * Check if lle still does not exists.
983 		 * If it does, that means that we either
984 		 * 1) have configured it explicitly, via
985 		 * 1a) 'arp -s' static entry or
986 		 * 1b) interface address static record
987 		 * or
988 		 * 2) it was the result of sending first packet to-host
989 		 * or
990 		 * 3) it was another arp reply packet we handled in
991 		 * different thread.
992 		 *
993 		 * In all cases except 3) we definitely need to prefer
994 		 * existing lle. For the sake of simplicity, prefer any
995 		 * existing lle over newly-create one.
996 		 */
997 		if (la_tmp == NULL)
998 			lltable_link_entry(LLTABLE(ifp), la);
999 		IF_AFDATA_WUNLOCK(ifp);
1000 
1001 		if (la_tmp == NULL) {
1002 			arp_mark_lle_reachable(la);
1003 			LLE_WUNLOCK(la);
1004 		} else {
1005 			/* Free newly-create entry and handle packet */
1006 			lltable_free_entry(LLTABLE(ifp), la);
1007 			la = la_tmp;
1008 			la_tmp = NULL;
1009 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1010 			/* arp_check_update_lle() returns @la unlocked */
1011 		}
1012 		la = NULL;
1013 	}
1014 reply:
1015 	if (op != ARPOP_REQUEST)
1016 		goto drop;
1017 	ARPSTAT_INC(rxrequests);
1018 
1019 	if (itaddr.s_addr == myaddr.s_addr) {
1020 		/* Shortcut.. the receiving interface is the target. */
1021 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1022 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1023 	} else {
1024 		/*
1025 		 * Destination address is not ours. Check if
1026 		 * proxyarp entry exists or proxyarp is turned on globally.
1027 		 */
1028 		struct llentry *lle;
1029 
1030 		sin.sin_addr = itaddr;
1031 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1032 
1033 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1034 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1035 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1036 			LLE_RUNLOCK(lle);
1037 		} else {
1038 			if (lle != NULL)
1039 				LLE_RUNLOCK(lle);
1040 
1041 			if (!V_arp_proxyall)
1042 				goto drop;
1043 
1044 			NET_EPOCH_ASSERT();
1045 			nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1046 			if (nh == NULL)
1047 				goto drop;
1048 
1049 			/*
1050 			 * Don't send proxies for nodes on the same interface
1051 			 * as this one came out of, or we'll get into a fight
1052 			 * over who claims what Ether address.
1053 			 */
1054 			if (nh->nh_ifp == ifp)
1055 				goto drop;
1056 
1057 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1058 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1059 
1060 			/*
1061 			 * Also check that the node which sent the ARP packet
1062 			 * is on the interface we expect it to be on. This
1063 			 * avoids ARP chaos if an interface is connected to the
1064 			 * wrong network.
1065 			 */
1066 
1067 			nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1068 			if (nh == NULL)
1069 				goto drop;
1070 			if (nh->nh_ifp != ifp) {
1071 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1072 				    " from %s via %s\n",
1073 				    inet_ntoa_r(isaddr, addrbuf),
1074 				    ifp->if_xname);
1075 				goto drop;
1076 			}
1077 
1078 #ifdef DEBUG_PROXY
1079 			printf("arp: proxying for %s\n",
1080 			    inet_ntoa_r(itaddr, addrbuf));
1081 #endif
1082 		}
1083 	}
1084 
1085 	if (itaddr.s_addr == myaddr.s_addr &&
1086 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1087 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1088 #ifdef DEBUG_LINKLOCAL
1089 		printf("arp: sending reply for link-local addr %s\n",
1090 		    inet_ntoa_r(itaddr, addrbuf));
1091 #endif
1092 		m->m_flags |= M_BCAST;
1093 		m->m_flags &= ~M_MCAST;
1094 	} else {
1095 		/* default behaviour; never reply by broadcast. */
1096 		m->m_flags &= ~(M_BCAST|M_MCAST);
1097 	}
1098 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1099 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1100 	ah->ar_op = htons(ARPOP_REPLY);
1101 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1102 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1103 	m->m_pkthdr.len = m->m_len;
1104 	m->m_pkthdr.rcvif = NULL;
1105 	sa.sa_family = AF_ARP;
1106 	sa.sa_len = 2;
1107 
1108 	/* Calculate link header for sending frame */
1109 	bzero(&ro, sizeof(ro));
1110 	linkhdrsize = sizeof(linkhdr);
1111 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1112 
1113 	/*
1114 	 * arp_fillheader() may fail due to lack of support inside encap request
1115 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1116 	 * by if_output().
1117 	 */
1118 	if (error != 0 && error != EAFNOSUPPORT) {
1119 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1120 		    if_name(ifp), error);
1121 		goto drop;
1122 	}
1123 
1124 	ro.ro_prepend = linkhdr;
1125 	ro.ro_plen = linkhdrsize;
1126 	ro.ro_flags = 0;
1127 
1128 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1129 	(*ifp->if_output)(ifp, m, &sa, &ro);
1130 	ARPSTAT_INC(txreplies);
1131 	return;
1132 
1133 drop:
1134 	m_freem(m);
1135 }
1136 #endif
1137 
1138 static struct mbuf *
1139 arp_grab_holdchain(struct llentry *la)
1140 {
1141 	struct mbuf *chain;
1142 
1143 	LLE_WLOCK_ASSERT(la);
1144 
1145 	chain = la->la_hold;
1146 	la->la_hold = NULL;
1147 	la->la_numheld = 0;
1148 
1149 	return (chain);
1150 }
1151 
1152 static void
1153 arp_flush_holdchain(struct ifnet *ifp, struct llentry *la, struct mbuf *chain)
1154 {
1155 	struct mbuf *m_hold, *m_hold_next;
1156 	struct sockaddr_in sin;
1157 
1158 	NET_EPOCH_ASSERT();
1159 
1160 	struct route ro = {
1161 		.ro_prepend = la->r_linkdata,
1162 		.ro_plen = la->r_hdrlen,
1163 	};
1164 
1165 	lltable_fill_sa_entry(la, (struct sockaddr *)&sin);
1166 
1167 	for (m_hold = chain; m_hold != NULL; m_hold = m_hold_next) {
1168 		m_hold_next = m_hold->m_nextpkt;
1169 		m_hold->m_nextpkt = NULL;
1170 		/* Avoid confusing lower layers. */
1171 		m_clrprotoflags(m_hold);
1172 		(*ifp->if_output)(ifp, m_hold, (struct sockaddr *)&sin, &ro);
1173 	}
1174 }
1175 
1176 /*
1177  * Checks received arp data against existing @la.
1178  * Updates lle state/performs notification if necessary.
1179  */
1180 static void
1181 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1182     int bridged, struct llentry *la)
1183 {
1184 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1185 	size_t linkhdrsize;
1186 	int lladdr_off;
1187 	char addrbuf[INET_ADDRSTRLEN];
1188 
1189 	LLE_WLOCK_ASSERT(la);
1190 
1191 	/* the following is not an error when doing bridging */
1192 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1193 		if (log_arp_wrong_iface)
1194 			ARP_LOG(LOG_WARNING, "%s is on %s "
1195 			    "but got reply from %*D on %s\n",
1196 			    inet_ntoa_r(isaddr, addrbuf),
1197 			    la->lle_tbl->llt_ifp->if_xname,
1198 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1199 			    ifp->if_xname);
1200 		LLE_WUNLOCK(la);
1201 		return;
1202 	}
1203 	if ((la->la_flags & LLE_VALID) &&
1204 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1205 		if (la->la_flags & LLE_STATIC) {
1206 			LLE_WUNLOCK(la);
1207 			if (log_arp_permanent_modify)
1208 				ARP_LOG(LOG_ERR,
1209 				    "%*D attempts to modify "
1210 				    "permanent entry for %s on %s\n",
1211 				    ifp->if_addrlen,
1212 				    (u_char *)ar_sha(ah), ":",
1213 				    inet_ntoa_r(isaddr, addrbuf),
1214 				    ifp->if_xname);
1215 			return;
1216 		}
1217 		if (log_arp_movements) {
1218 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1219 			    "to %*D on %s\n",
1220 			    inet_ntoa_r(isaddr, addrbuf),
1221 			    ifp->if_addrlen,
1222 			    (u_char *)la->ll_addr, ":",
1223 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1224 			    ifp->if_xname);
1225 		}
1226 	}
1227 
1228 	/* Calculate full link prepend to use in lle */
1229 	linkhdrsize = sizeof(linkhdr);
1230 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1231 	    &linkhdrsize, &lladdr_off) != 0) {
1232 		LLE_WUNLOCK(la);
1233 		return;
1234 	}
1235 
1236 	/* Check if something has changed */
1237 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1238 	    (la->la_flags & LLE_VALID) == 0) {
1239 		/* Try to perform LLE update */
1240 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1241 		    lladdr_off) == 0) {
1242 			LLE_WUNLOCK(la);
1243 			return;
1244 		}
1245 
1246 		/* Clear fast path feedback request if set */
1247 		llentry_mark_used(la);
1248 	}
1249 
1250 	arp_mark_lle_reachable(la);
1251 
1252 	/*
1253 	 * The packets are all freed within the call to the output
1254 	 * routine.
1255 	 *
1256 	 * NB: The lock MUST be released before the call to the
1257 	 * output routine.
1258 	 */
1259 	if (la->la_hold != NULL) {
1260 		struct mbuf *chain;
1261 
1262 		chain = arp_grab_holdchain(la);
1263 		LLE_WUNLOCK(la);
1264 		arp_flush_holdchain(ifp, la, chain);
1265 	} else
1266 		LLE_WUNLOCK(la);
1267 }
1268 
1269 static void
1270 arp_mark_lle_reachable(struct llentry *la)
1271 {
1272 	int canceled, wtime;
1273 
1274 	LLE_WLOCK_ASSERT(la);
1275 
1276 	la->ln_state = ARP_LLINFO_REACHABLE;
1277 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1278 
1279 	if (!(la->la_flags & LLE_STATIC)) {
1280 		LLE_ADDREF(la);
1281 		la->la_expire = time_uptime + V_arpt_keep;
1282 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1283 		if (wtime < 0)
1284 			wtime = V_arpt_keep;
1285 		canceled = callout_reset(&la->lle_timer,
1286 		    hz * wtime, arptimer, la);
1287 		if (canceled)
1288 			LLE_REMREF(la);
1289 	}
1290 	la->la_asked = 0;
1291 	la->la_preempt = V_arp_maxtries;
1292 }
1293 
1294 /*
1295  * Add permanent link-layer record for given interface address.
1296  */
1297 static __noinline void
1298 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1299 {
1300 	struct llentry *lle, *lle_tmp;
1301 
1302 	/*
1303 	 * Interface address LLE record is considered static
1304 	 * because kernel code relies on LLE_STATIC flag to check
1305 	 * if these entries can be rewriten by arp updates.
1306 	 */
1307 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1308 	if (lle == NULL) {
1309 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1310 		    "entry for interface address\n");
1311 		return;
1312 	}
1313 
1314 	IF_AFDATA_WLOCK(ifp);
1315 	LLE_WLOCK(lle);
1316 	/* Unlink any entry if exists */
1317 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1318 	if (lle_tmp != NULL)
1319 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1320 
1321 	lltable_link_entry(LLTABLE(ifp), lle);
1322 	IF_AFDATA_WUNLOCK(ifp);
1323 
1324 	if (lle_tmp != NULL)
1325 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1326 
1327 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1328 	LLE_WUNLOCK(lle);
1329 	if (lle_tmp != NULL)
1330 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1331 }
1332 
1333 /*
1334  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1335  * of valid values.
1336  */
1337 static int
1338 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1339 {
1340 	int error;
1341 	int rexmit_count = *(int *)arg1;
1342 
1343 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1344 
1345 	/* Enforce limits on any new value that may have been set. */
1346 	if (!error && req->newptr) {
1347 		/* A new value was set. */
1348 		if (rexmit_count < 0) {
1349 			rexmit_count = 0;
1350 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1351 			rexmit_count = MAX_GARP_RETRANSMITS;
1352 		}
1353 		*(int *)arg1 = rexmit_count;
1354 	}
1355 
1356 	return (error);
1357 }
1358 
1359 /*
1360  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1361  * retransmit it again. A pending callout owns a reference to the ifa.
1362  */
1363 static void
1364 garp_rexmit(void *arg)
1365 {
1366 	struct in_ifaddr *ia = arg;
1367 
1368 	if (callout_pending(&ia->ia_garp_timer) ||
1369 	    !callout_active(&ia->ia_garp_timer)) {
1370 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1371 		ifa_free(&ia->ia_ifa);
1372 		return;
1373 	}
1374 
1375 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1376 
1377 	/*
1378 	 * Drop lock while the ARP request is generated.
1379 	 */
1380 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1381 
1382 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1383 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1384 
1385 	/*
1386 	 * Increment the count of retransmissions. If the count has reached the
1387 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1388 	 * the callout to retransmit another GARP packet.
1389 	 */
1390 	++ia->ia_garp_count;
1391 	if (ia->ia_garp_count >= garp_rexmit_count) {
1392 		ifa_free(&ia->ia_ifa);
1393 	} else {
1394 		int rescheduled;
1395 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1396 		rescheduled = callout_reset(&ia->ia_garp_timer,
1397 		    (1 << ia->ia_garp_count) * hz,
1398 		    garp_rexmit, ia);
1399 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1400 		if (rescheduled) {
1401 			ifa_free(&ia->ia_ifa);
1402 		}
1403 	}
1404 
1405 	CURVNET_RESTORE();
1406 }
1407 
1408 /*
1409  * Start the GARP retransmit timer.
1410  *
1411  * A single GARP is always transmitted when an IPv4 address is added
1412  * to an interface and that is usually sufficient. However, in some
1413  * circumstances, such as when a shared address is passed between
1414  * cluster nodes, this single GARP may occasionally be dropped or
1415  * lost. This can lead to neighbors on the network link working with a
1416  * stale ARP cache and sending packets destined for that address to
1417  * the node that previously owned the address, which may not respond.
1418  *
1419  * To avoid this situation, GARP retransmits can be enabled by setting
1420  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1421  * than zero. The setting represents the maximum number of
1422  * retransmissions. The interval between retransmissions is calculated
1423  * using an exponential backoff algorithm, doubling each time, so the
1424  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1425  */
1426 static void
1427 garp_timer_start(struct ifaddr *ifa)
1428 {
1429 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1430 
1431 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1432 	ia->ia_garp_count = 0;
1433 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1434 	    garp_rexmit, ia) == 0) {
1435 		ifa_ref(ifa);
1436 	}
1437 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1438 }
1439 
1440 void
1441 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1442 {
1443 	struct epoch_tracker et;
1444 	const struct sockaddr_in *dst_in;
1445 	const struct sockaddr *dst;
1446 
1447 	if (ifa->ifa_carp != NULL)
1448 		return;
1449 
1450 	dst = ifa->ifa_addr;
1451 	dst_in = (const struct sockaddr_in *)dst;
1452 
1453 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1454 		return;
1455 	NET_EPOCH_ENTER(et);
1456 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1457 	NET_EPOCH_EXIT(et);
1458 	if (garp_rexmit_count > 0) {
1459 		garp_timer_start(ifa);
1460 	}
1461 
1462 	arp_add_ifa_lle(ifp, dst);
1463 }
1464 
1465 void
1466 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1467 {
1468 
1469 	if (ntohl(addr.s_addr) != INADDR_ANY)
1470 		arprequest(ifp, &addr, &addr, enaddr);
1471 }
1472 
1473 /*
1474  * Sends gratuitous ARPs for each ifaddr to notify other
1475  * nodes about the address change.
1476  */
1477 static __noinline void
1478 arp_handle_ifllchange(struct ifnet *ifp)
1479 {
1480 	struct ifaddr *ifa;
1481 
1482 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1483 		if (ifa->ifa_addr->sa_family == AF_INET)
1484 			arp_ifinit(ifp, ifa);
1485 	}
1486 }
1487 
1488 /*
1489  * A handler for interface link layer address change event.
1490  */
1491 static void
1492 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1493 {
1494 	/* if_bridge can update its lladdr during if_vmove(), after we've done
1495 	 * if_detach_internal()/dom_ifdetach(). */
1496 	if (ifp->if_afdata[AF_INET] == NULL)
1497 		return;
1498 
1499 	lltable_update_ifaddr(LLTABLE(ifp));
1500 
1501 	if ((ifp->if_flags & IFF_UP) != 0)
1502 		arp_handle_ifllchange(ifp);
1503 }
1504 
1505 static void
1506 vnet_arp_init(void)
1507 {
1508 
1509 	if (IS_DEFAULT_VNET(curvnet)) {
1510 		netisr_register(&arp_nh);
1511 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1512 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1513 	}
1514 #ifdef VIMAGE
1515 	else
1516 		netisr_register_vnet(&arp_nh);
1517 #endif
1518 }
1519 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1520     vnet_arp_init, 0);
1521 
1522 #ifdef VIMAGE
1523 /*
1524  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1525  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1526  */
1527 static void
1528 vnet_arp_destroy(__unused void *arg)
1529 {
1530 
1531 	netisr_unregister_vnet(&arp_nh);
1532 }
1533 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1534     vnet_arp_destroy, NULL);
1535 #endif
1536