xref: /freebsd/sys/netinet/if_ether.c (revision 662c13053f4bf2d6245ba7e2b66c10d1cd5c1fb9)
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) = 1;
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 	int r_skip_req;
211 
212 	if (lle->la_flags & LLE_STATIC) {
213 		return;
214 	}
215 	LLE_WLOCK(lle);
216 	if (callout_pending(&lle->lle_timer)) {
217 		/*
218 		 * Here we are a bit odd here in the treatment of
219 		 * active/pending. If the pending bit is set, it got
220 		 * rescheduled before I ran. The active
221 		 * bit we ignore, since if it was stopped
222 		 * in ll_tablefree() and was currently running
223 		 * it would have return 0 so the code would
224 		 * not have deleted it since the callout could
225 		 * not be stopped so we want to go through
226 		 * with the delete here now. If the callout
227 		 * was restarted, the pending bit will be back on and
228 		 * we just want to bail since the callout_reset would
229 		 * return 1 and our reference would have been removed
230 		 * by arpresolve() below.
231 		 */
232 		LLE_WUNLOCK(lle);
233  		return;
234  	}
235 	ifp = lle->lle_tbl->llt_ifp;
236 	CURVNET_SET(ifp->if_vnet);
237 
238 	switch (lle->ln_state) {
239 	case ARP_LLINFO_REACHABLE:
240 
241 		/*
242 		 * Expiration time is approaching.
243 		 * Let's try to refresh entry if it is still
244 		 * in use.
245 		 *
246 		 * Set r_skip_req to get feedback from
247 		 * fast path. Change state and re-schedule
248 		 * ourselves.
249 		 */
250 		LLE_REQ_LOCK(lle);
251 		lle->r_skip_req = 1;
252 		LLE_REQ_UNLOCK(lle);
253 		lle->ln_state = ARP_LLINFO_VERIFY;
254 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
255 		LLE_WUNLOCK(lle);
256 		CURVNET_RESTORE();
257 		return;
258 	case ARP_LLINFO_VERIFY:
259 		LLE_REQ_LOCK(lle);
260 		r_skip_req = lle->r_skip_req;
261 		LLE_REQ_UNLOCK(lle);
262 
263 		if (r_skip_req == 0 && lle->la_preempt > 0) {
264 			/* Entry was used, issue refresh request */
265 			struct epoch_tracker et;
266 			struct in_addr dst;
267 
268 			dst = lle->r_l3addr.addr4;
269 			lle->la_preempt--;
270 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
271 			LLE_WUNLOCK(lle);
272 			NET_EPOCH_ENTER(et);
273 			arprequest(ifp, NULL, &dst, NULL);
274 			NET_EPOCH_EXIT(et);
275 			CURVNET_RESTORE();
276 			return;
277 		}
278 		/* Nothing happened. Reschedule if not too late */
279 		if (lle->la_expire > time_uptime) {
280 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
281 			LLE_WUNLOCK(lle);
282 			CURVNET_RESTORE();
283 			return;
284 		}
285 		break;
286 	case ARP_LLINFO_INCOMPLETE:
287 	case ARP_LLINFO_DELETED:
288 		break;
289 	}
290 
291 	if ((lle->la_flags & LLE_DELETED) == 0) {
292 		int evt;
293 
294 		if (lle->la_flags & LLE_VALID)
295 			evt = LLENTRY_EXPIRED;
296 		else
297 			evt = LLENTRY_TIMEDOUT;
298 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
299 	}
300 
301 	callout_stop(&lle->lle_timer);
302 
303 	/* XXX: LOR avoidance. We still have ref on lle. */
304 	LLE_WUNLOCK(lle);
305 	IF_AFDATA_LOCK(ifp);
306 	LLE_WLOCK(lle);
307 
308 	/* Guard against race with other llentry_free(). */
309 	if (lle->la_flags & LLE_LINKED) {
310 		LLE_REMREF(lle);
311 		lltable_unlink_entry(lle->lle_tbl, lle);
312 	}
313 	IF_AFDATA_UNLOCK(ifp);
314 
315 	size_t pkts_dropped = llentry_free(lle);
316 
317 	ARPSTAT_ADD(dropped, pkts_dropped);
318 	ARPSTAT_INC(timeouts);
319 
320 	CURVNET_RESTORE();
321 }
322 
323 /*
324  * Stores link-layer header for @ifp in format suitable for if_output()
325  * into buffer @buf. Resulting header length is stored in @bufsize.
326  *
327  * Returns 0 on success.
328  */
329 static int
330 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
331     size_t *bufsize)
332 {
333 	struct if_encap_req ereq;
334 	int error;
335 
336 	bzero(buf, *bufsize);
337 	bzero(&ereq, sizeof(ereq));
338 	ereq.buf = buf;
339 	ereq.bufsize = *bufsize;
340 	ereq.rtype = IFENCAP_LL;
341 	ereq.family = AF_ARP;
342 	ereq.lladdr = ar_tha(ah);
343 	ereq.hdata = (u_char *)ah;
344 	if (bcast)
345 		ereq.flags = IFENCAP_FLAG_BROADCAST;
346 	error = ifp->if_requestencap(ifp, &ereq);
347 	if (error == 0)
348 		*bufsize = ereq.bufsize;
349 
350 	return (error);
351 }
352 
353 /*
354  * Broadcast an ARP request. Caller specifies:
355  *	- arp header source ip address
356  *	- arp header target ip address
357  *	- arp header source ethernet address
358  */
359 static int
360 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
361     const struct in_addr *tip, u_char *enaddr)
362 {
363 	struct mbuf *m;
364 	struct arphdr *ah;
365 	struct sockaddr sa;
366 	u_char *carpaddr = NULL;
367 	uint8_t linkhdr[LLE_MAX_LINKHDR];
368 	size_t linkhdrsize;
369 	struct route ro;
370 	int error;
371 
372 	NET_EPOCH_ASSERT();
373 
374 	if (sip == NULL) {
375 		/*
376 		 * The caller did not supply a source address, try to find
377 		 * a compatible one among those assigned to this interface.
378 		 */
379 		struct ifaddr *ifa;
380 
381 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
382 			if (ifa->ifa_addr->sa_family != AF_INET)
383 				continue;
384 
385 			if (ifa->ifa_carp) {
386 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
387 					continue;
388 				sip = &IA_SIN(ifa)->sin_addr;
389 			} else {
390 				carpaddr = NULL;
391 				sip = &IA_SIN(ifa)->sin_addr;
392 			}
393 
394 			if (0 == ((sip->s_addr ^ tip->s_addr) &
395 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
396 				break;  /* found it. */
397 		}
398 		if (sip == NULL) {
399 			printf("%s: cannot find matching address\n", __func__);
400 			return (EADDRNOTAVAIL);
401 		}
402 	}
403 	if (enaddr == NULL)
404 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
405 
406 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
407 		return (ENOMEM);
408 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
409 		2 * ifp->if_addrlen;
410 	m->m_pkthdr.len = m->m_len;
411 	M_ALIGN(m, m->m_len);
412 	ah = mtod(m, struct arphdr *);
413 	bzero((caddr_t)ah, m->m_len);
414 #ifdef MAC
415 	mac_netinet_arp_send(ifp, m);
416 #endif
417 	ah->ar_pro = htons(ETHERTYPE_IP);
418 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
419 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
420 	ah->ar_op = htons(ARPOP_REQUEST);
421 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
422 	bcopy(sip, ar_spa(ah), ah->ar_pln);
423 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
424 	sa.sa_family = AF_ARP;
425 	sa.sa_len = 2;
426 
427 	/* Calculate link header for sending frame */
428 	bzero(&ro, sizeof(ro));
429 	linkhdrsize = sizeof(linkhdr);
430 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
431 	if (error != 0 && error != EAFNOSUPPORT) {
432 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
433 		    if_name(ifp), error);
434 		return (error);
435 	}
436 
437 	ro.ro_prepend = linkhdr;
438 	ro.ro_plen = linkhdrsize;
439 	ro.ro_flags = 0;
440 
441 	m->m_flags |= M_BCAST;
442 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
443 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
444 	ARPSTAT_INC(txrequests);
445 	if (error) {
446 		ARPSTAT_INC(txerrors);
447 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
448 		    if_name(ifp), error);
449 	}
450 	return (error);
451 }
452 
453 void
454 arprequest(struct ifnet *ifp, const struct in_addr *sip,
455     const struct in_addr *tip, u_char *enaddr)
456 {
457 
458 	(void) arprequest_internal(ifp, sip, tip, enaddr);
459 }
460 
461 /*
462  * Resolve an IP address into an ethernet address - heavy version.
463  * Used internally by arpresolve().
464  * We have already checked that we can't use an existing lle without
465  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
466  *
467  * On success, desten and pflags are filled in and the function returns 0;
468  * If the packet must be held pending resolution, we return EWOULDBLOCK
469  * On other errors, we return the corresponding error code.
470  * Note that m_freem() handles NULL.
471  */
472 static int
473 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
474 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
475 	struct llentry **plle)
476 {
477 	struct llentry *la = NULL, *la_tmp;
478 	struct mbuf *curr = NULL;
479 	struct mbuf *next = NULL;
480 	int error, renew;
481 	char *lladdr;
482 	int ll_len;
483 
484 	NET_EPOCH_ASSERT();
485 
486 	if (pflags != NULL)
487 		*pflags = 0;
488 	if (plle != NULL)
489 		*plle = NULL;
490 
491 	if ((flags & LLE_CREATE) == 0)
492 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
493 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
494 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
495 		if (la == NULL) {
496 			char addrbuf[INET_ADDRSTRLEN];
497 
498 			log(LOG_DEBUG,
499 			    "arpresolve: can't allocate llinfo for %s on %s\n",
500 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
501 			    if_name(ifp));
502 			m_freem(m);
503 			return (EINVAL);
504 		}
505 
506 		IF_AFDATA_WLOCK(ifp);
507 		LLE_WLOCK(la);
508 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
509 		/* Prefer ANY existing lle over newly-created one */
510 		if (la_tmp == NULL)
511 			lltable_link_entry(LLTABLE(ifp), la);
512 		IF_AFDATA_WUNLOCK(ifp);
513 		if (la_tmp != NULL) {
514 			lltable_free_entry(LLTABLE(ifp), la);
515 			la = la_tmp;
516 		}
517 	}
518 	if (la == NULL) {
519 		m_freem(m);
520 		return (EINVAL);
521 	}
522 
523 	if ((la->la_flags & LLE_VALID) &&
524 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
525 		if (flags & LLE_ADDRONLY) {
526 			lladdr = la->ll_addr;
527 			ll_len = ifp->if_addrlen;
528 		} else {
529 			lladdr = la->r_linkdata;
530 			ll_len = la->r_hdrlen;
531 		}
532 		bcopy(lladdr, desten, ll_len);
533 
534 		/* Notify LLE code that the entry was used by datapath */
535 		llentry_mark_used(la);
536 		if (pflags != NULL)
537 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
538 		if (plle) {
539 			LLE_ADDREF(la);
540 			*plle = la;
541 		}
542 		LLE_WUNLOCK(la);
543 		return (0);
544 	}
545 
546 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
547 	/*
548 	 * There is an arptab entry, but no ethernet address
549 	 * response yet.  Add the mbuf to the list, dropping
550 	 * the oldest packet if we have exceeded the system
551 	 * setting.
552 	 */
553 	if (m != NULL) {
554 		if (la->la_numheld >= V_arp_maxhold) {
555 			if (la->la_hold != NULL) {
556 				next = la->la_hold->m_nextpkt;
557 				m_freem(la->la_hold);
558 				la->la_hold = next;
559 				la->la_numheld--;
560 				ARPSTAT_INC(dropped);
561 			}
562 		}
563 		if (la->la_hold != NULL) {
564 			curr = la->la_hold;
565 			while (curr->m_nextpkt != NULL)
566 				curr = curr->m_nextpkt;
567 			curr->m_nextpkt = m;
568 		} else
569 			la->la_hold = m;
570 		la->la_numheld++;
571 	}
572 	/*
573 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
574 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
575 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
576 	 * ARP request, but not faster than one request per second.
577 	 */
578 	if (la->la_asked < V_arp_maxtries)
579 		error = EWOULDBLOCK;	/* First request. */
580 	else
581 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
582 
583 	if (renew) {
584 		int canceled, e;
585 
586 		LLE_ADDREF(la);
587 		la->la_expire = time_uptime;
588 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
589 		    arptimer, la);
590 		if (canceled)
591 			LLE_REMREF(la);
592 		la->la_asked++;
593 		LLE_WUNLOCK(la);
594 		e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
595 		/*
596 		 * Only overwrite 'error' in case of error; in case of success
597 		 * the proper return value was already set above.
598 		 */
599 		if (e != 0)
600 			return (e);
601 		return (error);
602 	}
603 
604 	LLE_WUNLOCK(la);
605 	return (error);
606 }
607 
608 /*
609  * Lookups link header based on an IP address.
610  * On input:
611  *    ifp is the interface we use
612  *    is_gw != 0 if @dst represents gateway to some destination
613  *    m is the mbuf. May be NULL if we don't have a packet.
614  *    dst is the next hop,
615  *    desten is the storage to put LL header.
616  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
617  *
618  * On success, full/partial link header and flags are filled in and
619  * the function returns 0.
620  * If the packet must be held pending resolution, we return EWOULDBLOCK
621  * On other errors, we return the corresponding error code.
622  * Note that m_freem() handles NULL.
623  */
624 int
625 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
626 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
627 	struct llentry **plle)
628 {
629 	struct llentry *la = NULL;
630 
631 	NET_EPOCH_ASSERT();
632 
633 	if (pflags != NULL)
634 		*pflags = 0;
635 	if (plle != NULL)
636 		*plle = NULL;
637 
638 	if (m != NULL) {
639 		if (m->m_flags & M_BCAST) {
640 			/* broadcast */
641 			(void)memcpy(desten,
642 			    ifp->if_broadcastaddr, ifp->if_addrlen);
643 			return (0);
644 		}
645 		if (m->m_flags & M_MCAST) {
646 			/* multicast */
647 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
648 			return (0);
649 		}
650 	}
651 
652 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
653 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
654 		/* Entry found, let's copy lle info */
655 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
656 		if (pflags != NULL)
657 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
658 		/* Notify the LLE handling code that the entry was used. */
659 		llentry_mark_used(la);
660 		if (plle) {
661 			LLE_ADDREF(la);
662 			*plle = la;
663 			LLE_WUNLOCK(la);
664 		}
665 		return (0);
666 	}
667 	if (plle && la)
668 		LLE_WUNLOCK(la);
669 
670 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
671 	    desten, pflags, plle));
672 }
673 
674 /*
675  * Common length and type checks are done here,
676  * then the protocol-specific routine is called.
677  */
678 static void
679 arpintr(struct mbuf *m)
680 {
681 	struct arphdr *ar;
682 	struct ifnet *ifp;
683 	char *layer;
684 	int hlen;
685 
686 	ifp = m->m_pkthdr.rcvif;
687 
688 	if (m->m_len < sizeof(struct arphdr) &&
689 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
690 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
691 		    if_name(ifp));
692 		return;
693 	}
694 	ar = mtod(m, struct arphdr *);
695 
696 	/* Check if length is sufficient */
697 	if (m->m_len <  arphdr_len(ar)) {
698 		m = m_pullup(m, arphdr_len(ar));
699 		if (m == NULL) {
700 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
701 			    if_name(ifp));
702 			return;
703 		}
704 		ar = mtod(m, struct arphdr *);
705 	}
706 
707 	hlen = 0;
708 	layer = "";
709 	switch (ntohs(ar->ar_hrd)) {
710 	case ARPHRD_ETHER:
711 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
712 		layer = "ethernet";
713 		break;
714 	case ARPHRD_INFINIBAND:
715 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
716 		layer = "infiniband";
717 		break;
718 	case ARPHRD_IEEE1394:
719 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
720 		layer = "firewire";
721 
722 		/*
723 		 * Restrict too long hardware addresses.
724 		 * Currently we are capable of handling 20-byte
725 		 * addresses ( sizeof(lle->ll_addr) )
726 		 */
727 		if (ar->ar_hln >= 20)
728 			hlen = 16;
729 		break;
730 	default:
731 		ARP_LOG(LOG_NOTICE,
732 		    "packet with unknown hardware format 0x%02d received on "
733 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
734 		m_freem(m);
735 		return;
736 	}
737 
738 	if (hlen != 0 && hlen != ar->ar_hln) {
739 		ARP_LOG(LOG_NOTICE,
740 		    "packet with invalid %s address length %d received on %s\n",
741 		    layer, ar->ar_hln, if_name(ifp));
742 		m_freem(m);
743 		return;
744 	}
745 
746 	ARPSTAT_INC(received);
747 	switch (ntohs(ar->ar_pro)) {
748 #ifdef INET
749 	case ETHERTYPE_IP:
750 		in_arpinput(m);
751 		return;
752 #endif
753 	}
754 	m_freem(m);
755 }
756 
757 #ifdef INET
758 /*
759  * ARP for Internet protocols on 10 Mb/s Ethernet.
760  * Algorithm is that given in RFC 826.
761  * In addition, a sanity check is performed on the sender
762  * protocol address, to catch impersonators.
763  * We no longer handle negotiations for use of trailer protocol:
764  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
765  * along with IP replies if we wanted trailers sent to us,
766  * and also sent them in response to IP replies.
767  * This allowed either end to announce the desire to receive
768  * trailer packets.
769  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
770  * but formerly didn't normally send requests.
771  */
772 static int log_arp_wrong_iface = 1;
773 static int log_arp_movements = 1;
774 static int log_arp_permanent_modify = 1;
775 static int allow_multicast = 0;
776 
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
778 	&log_arp_wrong_iface, 0,
779 	"log arp packets arriving on the wrong interface");
780 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
781 	&log_arp_movements, 0,
782 	"log arp replies from MACs different than the one in the cache");
783 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
784 	&log_arp_permanent_modify, 0,
785 	"log arp replies from MACs different than the one in the permanent arp entry");
786 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
787 	&allow_multicast, 0, "accept multicast addresses");
788 
789 static void
790 in_arpinput(struct mbuf *m)
791 {
792 	struct rm_priotracker in_ifa_tracker;
793 	struct arphdr *ah;
794 	struct ifnet *ifp = m->m_pkthdr.rcvif;
795 	struct llentry *la = NULL, *la_tmp;
796 	struct ifaddr *ifa;
797 	struct in_ifaddr *ia;
798 	struct sockaddr sa;
799 	struct in_addr isaddr, itaddr, myaddr;
800 	u_int8_t *enaddr = NULL;
801 	int op;
802 	int bridged = 0, is_bridge = 0;
803 	int carped;
804 	struct sockaddr_in sin;
805 	struct sockaddr *dst;
806 	struct nhop_object *nh;
807 	uint8_t linkhdr[LLE_MAX_LINKHDR];
808 	struct route ro;
809 	size_t linkhdrsize;
810 	int lladdr_off;
811 	int error;
812 	char addrbuf[INET_ADDRSTRLEN];
813 
814 	NET_EPOCH_ASSERT();
815 
816 	sin.sin_len = sizeof(struct sockaddr_in);
817 	sin.sin_family = AF_INET;
818 	sin.sin_addr.s_addr = 0;
819 
820 	if (ifp->if_bridge)
821 		bridged = 1;
822 	if (ifp->if_type == IFT_BRIDGE)
823 		is_bridge = 1;
824 
825 	/*
826 	 * We already have checked that mbuf contains enough contiguous data
827 	 * to hold entire arp message according to the arp header.
828 	 */
829 	ah = mtod(m, struct arphdr *);
830 
831 	/*
832 	 * ARP is only for IPv4 so we can reject packets with
833 	 * a protocol length not equal to an IPv4 address.
834 	 */
835 	if (ah->ar_pln != sizeof(struct in_addr)) {
836 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
837 		    sizeof(struct in_addr));
838 		goto drop;
839 	}
840 
841 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
842 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
843 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
844 		goto drop;
845 	}
846 
847 	op = ntohs(ah->ar_op);
848 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
849 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
850 
851 	if (op == ARPOP_REPLY)
852 		ARPSTAT_INC(rxreplies);
853 
854 	/*
855 	 * For a bridge, we want to check the address irrespective
856 	 * of the receive interface. (This will change slightly
857 	 * when we have clusters of interfaces).
858 	 */
859 	IN_IFADDR_RLOCK(&in_ifa_tracker);
860 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
861 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
862 		    ia->ia_ifp == ifp) &&
863 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
864 		    (ia->ia_ifa.ifa_carp == NULL ||
865 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
866 			ifa_ref(&ia->ia_ifa);
867 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
868 			goto match;
869 		}
870 	}
871 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
872 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
873 		    ia->ia_ifp == ifp) &&
874 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
875 			ifa_ref(&ia->ia_ifa);
876 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
877 			goto match;
878 		}
879 
880 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
881   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
882   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
883   addr == ia->ia_addr.sin_addr.s_addr)
884 	/*
885 	 * Check the case when bridge shares its MAC address with
886 	 * some of its children, so packets are claimed by bridge
887 	 * itself (bridge_input() does it first), but they are really
888 	 * meant to be destined to the bridge member.
889 	 */
890 	if (is_bridge) {
891 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
892 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
893 				ifa_ref(&ia->ia_ifa);
894 				ifp = ia->ia_ifp;
895 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
896 				goto match;
897 			}
898 		}
899 	}
900 #undef BDG_MEMBER_MATCHES_ARP
901 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
902 
903 	/*
904 	 * No match, use the first inet address on the receive interface
905 	 * as a dummy address for the rest of the function.
906 	 */
907 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
908 		if (ifa->ifa_addr->sa_family == AF_INET &&
909 		    (ifa->ifa_carp == NULL ||
910 		    (*carp_iamatch_p)(ifa, &enaddr))) {
911 			ia = ifatoia(ifa);
912 			ifa_ref(ifa);
913 			goto match;
914 		}
915 
916 	/*
917 	 * If bridging, fall back to using any inet address.
918 	 */
919 	IN_IFADDR_RLOCK(&in_ifa_tracker);
920 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
921 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
922 		goto drop;
923 	}
924 	ifa_ref(&ia->ia_ifa);
925 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
926 match:
927 	if (!enaddr)
928 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
929 	carped = (ia->ia_ifa.ifa_carp != NULL);
930 	myaddr = ia->ia_addr.sin_addr;
931 	ifa_free(&ia->ia_ifa);
932 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
933 		goto drop;	/* it's from me, ignore it. */
934 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
935 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
936 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
937 		goto drop;
938 	}
939 
940 	if (ifp->if_addrlen != ah->ar_hln) {
941 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
942 		    "i/f %d (ignored)\n", ifp->if_addrlen,
943 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
944 		    ifp->if_addrlen);
945 		goto drop;
946 	}
947 
948 	/*
949 	 * Warn if another host is using the same IP address, but only if the
950 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
951 	 * case we suppress the warning to avoid false positive complaints of
952 	 * potential misconfiguration.
953 	 */
954 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
955 	    myaddr.s_addr != 0) {
956 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
957 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
958 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
959 		itaddr = myaddr;
960 		ARPSTAT_INC(dupips);
961 		goto reply;
962 	}
963 	if (ifp->if_flags & IFF_STATICARP)
964 		goto reply;
965 
966 	bzero(&sin, sizeof(sin));
967 	sin.sin_len = sizeof(struct sockaddr_in);
968 	sin.sin_family = AF_INET;
969 	sin.sin_addr = isaddr;
970 	dst = (struct sockaddr *)&sin;
971 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
972 	if (la != NULL)
973 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
974 	else if (itaddr.s_addr == myaddr.s_addr) {
975 		/*
976 		 * Request/reply to our address, but no lle exists yet.
977 		 * Calculate full link prepend to use in lle.
978 		 */
979 		linkhdrsize = sizeof(linkhdr);
980 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
981 		    &linkhdrsize, &lladdr_off) != 0)
982 			goto reply;
983 
984 		/* Allocate new entry */
985 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
986 		if (la == NULL) {
987 			/*
988 			 * lle creation may fail if source address belongs
989 			 * to non-directly connected subnet. However, we
990 			 * will try to answer the request instead of dropping
991 			 * frame.
992 			 */
993 			goto reply;
994 		}
995 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
996 		    lladdr_off);
997 
998 		IF_AFDATA_WLOCK(ifp);
999 		LLE_WLOCK(la);
1000 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1001 
1002 		/*
1003 		 * Check if lle still does not exists.
1004 		 * If it does, that means that we either
1005 		 * 1) have configured it explicitly, via
1006 		 * 1a) 'arp -s' static entry or
1007 		 * 1b) interface address static record
1008 		 * or
1009 		 * 2) it was the result of sending first packet to-host
1010 		 * or
1011 		 * 3) it was another arp reply packet we handled in
1012 		 * different thread.
1013 		 *
1014 		 * In all cases except 3) we definitely need to prefer
1015 		 * existing lle. For the sake of simplicity, prefer any
1016 		 * existing lle over newly-create one.
1017 		 */
1018 		if (la_tmp == NULL)
1019 			lltable_link_entry(LLTABLE(ifp), la);
1020 		IF_AFDATA_WUNLOCK(ifp);
1021 
1022 		if (la_tmp == NULL) {
1023 			arp_mark_lle_reachable(la);
1024 			LLE_WUNLOCK(la);
1025 		} else {
1026 			/* Free newly-create entry and handle packet */
1027 			lltable_free_entry(LLTABLE(ifp), la);
1028 			la = la_tmp;
1029 			la_tmp = NULL;
1030 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1031 			/* arp_check_update_lle() returns @la unlocked */
1032 		}
1033 		la = NULL;
1034 	}
1035 reply:
1036 	if (op != ARPOP_REQUEST)
1037 		goto drop;
1038 	ARPSTAT_INC(rxrequests);
1039 
1040 	if (itaddr.s_addr == myaddr.s_addr) {
1041 		/* Shortcut.. the receiving interface is the target. */
1042 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1043 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1044 	} else {
1045 		/*
1046 		 * Destination address is not ours. Check if
1047 		 * proxyarp entry exists or proxyarp is turned on globally.
1048 		 */
1049 		struct llentry *lle;
1050 
1051 		sin.sin_addr = itaddr;
1052 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1053 
1054 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1055 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1056 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1057 			LLE_RUNLOCK(lle);
1058 		} else {
1059 			if (lle != NULL)
1060 				LLE_RUNLOCK(lle);
1061 
1062 			if (!V_arp_proxyall)
1063 				goto drop;
1064 
1065 			NET_EPOCH_ASSERT();
1066 			nh = fib4_lookup(ifp->if_fib, itaddr, 0, 0, 0);
1067 			if (nh == NULL)
1068 				goto drop;
1069 
1070 			/*
1071 			 * Don't send proxies for nodes on the same interface
1072 			 * as this one came out of, or we'll get into a fight
1073 			 * over who claims what Ether address.
1074 			 */
1075 			if (nh->nh_ifp == ifp)
1076 				goto drop;
1077 
1078 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1079 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1080 
1081 			/*
1082 			 * Also check that the node which sent the ARP packet
1083 			 * is on the interface we expect it to be on. This
1084 			 * avoids ARP chaos if an interface is connected to the
1085 			 * wrong network.
1086 			 */
1087 
1088 			nh = fib4_lookup(ifp->if_fib, isaddr, 0, 0, 0);
1089 			if (nh == NULL)
1090 				goto drop;
1091 			if (nh->nh_ifp != ifp) {
1092 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1093 				    " from %s via %s\n",
1094 				    inet_ntoa_r(isaddr, addrbuf),
1095 				    ifp->if_xname);
1096 				goto drop;
1097 			}
1098 
1099 #ifdef DEBUG_PROXY
1100 			printf("arp: proxying for %s\n",
1101 			    inet_ntoa_r(itaddr, addrbuf));
1102 #endif
1103 		}
1104 	}
1105 
1106 	if (itaddr.s_addr == myaddr.s_addr &&
1107 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1108 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1109 #ifdef DEBUG_LINKLOCAL
1110 		printf("arp: sending reply for link-local addr %s\n",
1111 		    inet_ntoa_r(itaddr, addrbuf));
1112 #endif
1113 		m->m_flags |= M_BCAST;
1114 		m->m_flags &= ~M_MCAST;
1115 	} else {
1116 		/* default behaviour; never reply by broadcast. */
1117 		m->m_flags &= ~(M_BCAST|M_MCAST);
1118 	}
1119 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1120 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1121 	ah->ar_op = htons(ARPOP_REPLY);
1122 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1123 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1124 	m->m_pkthdr.len = m->m_len;
1125 	m->m_pkthdr.rcvif = NULL;
1126 	sa.sa_family = AF_ARP;
1127 	sa.sa_len = 2;
1128 
1129 	/* Calculate link header for sending frame */
1130 	bzero(&ro, sizeof(ro));
1131 	linkhdrsize = sizeof(linkhdr);
1132 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1133 
1134 	/*
1135 	 * arp_fillheader() may fail due to lack of support inside encap request
1136 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1137 	 * by if_output().
1138 	 */
1139 	if (error != 0 && error != EAFNOSUPPORT) {
1140 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1141 		    if_name(ifp), error);
1142 		return;
1143 	}
1144 
1145 	ro.ro_prepend = linkhdr;
1146 	ro.ro_plen = linkhdrsize;
1147 	ro.ro_flags = 0;
1148 
1149 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1150 	(*ifp->if_output)(ifp, m, &sa, &ro);
1151 	ARPSTAT_INC(txreplies);
1152 	return;
1153 
1154 drop:
1155 	m_freem(m);
1156 }
1157 #endif
1158 
1159 /*
1160  * Checks received arp data against existing @la.
1161  * Updates lle state/performs notification if necessary.
1162  */
1163 static void
1164 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1165     int bridged, struct llentry *la)
1166 {
1167 	struct sockaddr sa;
1168 	struct mbuf *m_hold, *m_hold_next;
1169 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1170 	size_t linkhdrsize;
1171 	int lladdr_off;
1172 	char addrbuf[INET_ADDRSTRLEN];
1173 
1174 	LLE_WLOCK_ASSERT(la);
1175 
1176 	/* the following is not an error when doing bridging */
1177 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1178 		if (log_arp_wrong_iface)
1179 			ARP_LOG(LOG_WARNING, "%s is on %s "
1180 			    "but got reply from %*D on %s\n",
1181 			    inet_ntoa_r(isaddr, addrbuf),
1182 			    la->lle_tbl->llt_ifp->if_xname,
1183 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1184 			    ifp->if_xname);
1185 		LLE_WUNLOCK(la);
1186 		return;
1187 	}
1188 	if ((la->la_flags & LLE_VALID) &&
1189 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1190 		if (la->la_flags & LLE_STATIC) {
1191 			LLE_WUNLOCK(la);
1192 			if (log_arp_permanent_modify)
1193 				ARP_LOG(LOG_ERR,
1194 				    "%*D attempts to modify "
1195 				    "permanent entry for %s on %s\n",
1196 				    ifp->if_addrlen,
1197 				    (u_char *)ar_sha(ah), ":",
1198 				    inet_ntoa_r(isaddr, addrbuf),
1199 				    ifp->if_xname);
1200 			return;
1201 		}
1202 		if (log_arp_movements) {
1203 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1204 			    "to %*D on %s\n",
1205 			    inet_ntoa_r(isaddr, addrbuf),
1206 			    ifp->if_addrlen,
1207 			    (u_char *)la->ll_addr, ":",
1208 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1209 			    ifp->if_xname);
1210 		}
1211 	}
1212 
1213 	/* Calculate full link prepend to use in lle */
1214 	linkhdrsize = sizeof(linkhdr);
1215 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1216 	    &linkhdrsize, &lladdr_off) != 0)
1217 		return;
1218 
1219 	/* Check if something has changed */
1220 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1221 	    (la->la_flags & LLE_VALID) == 0) {
1222 		/* Try to perform LLE update */
1223 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1224 		    lladdr_off) == 0)
1225 			return;
1226 
1227 		/* Clear fast path feedback request if set */
1228 		la->r_skip_req = 0;
1229 	}
1230 
1231 	arp_mark_lle_reachable(la);
1232 
1233 	/*
1234 	 * The packets are all freed within the call to the output
1235 	 * routine.
1236 	 *
1237 	 * NB: The lock MUST be released before the call to the
1238 	 * output routine.
1239 	 */
1240 	if (la->la_hold != NULL) {
1241 		m_hold = la->la_hold;
1242 		la->la_hold = NULL;
1243 		la->la_numheld = 0;
1244 		lltable_fill_sa_entry(la, &sa);
1245 		LLE_WUNLOCK(la);
1246 		for (; m_hold != NULL; m_hold = m_hold_next) {
1247 			m_hold_next = m_hold->m_nextpkt;
1248 			m_hold->m_nextpkt = NULL;
1249 			/* Avoid confusing lower layers. */
1250 			m_clrprotoflags(m_hold);
1251 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1252 		}
1253 	} else
1254 		LLE_WUNLOCK(la);
1255 }
1256 
1257 static void
1258 arp_mark_lle_reachable(struct llentry *la)
1259 {
1260 	int canceled, wtime;
1261 
1262 	LLE_WLOCK_ASSERT(la);
1263 
1264 	la->ln_state = ARP_LLINFO_REACHABLE;
1265 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1266 
1267 	if (!(la->la_flags & LLE_STATIC)) {
1268 		LLE_ADDREF(la);
1269 		la->la_expire = time_uptime + V_arpt_keep;
1270 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1271 		if (wtime < 0)
1272 			wtime = V_arpt_keep;
1273 		canceled = callout_reset(&la->lle_timer,
1274 		    hz * wtime, arptimer, la);
1275 		if (canceled)
1276 			LLE_REMREF(la);
1277 	}
1278 	la->la_asked = 0;
1279 	la->la_preempt = V_arp_maxtries;
1280 }
1281 
1282 /*
1283  * Add permanent link-layer record for given interface address.
1284  */
1285 static __noinline void
1286 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1287 {
1288 	struct llentry *lle, *lle_tmp;
1289 
1290 	/*
1291 	 * Interface address LLE record is considered static
1292 	 * because kernel code relies on LLE_STATIC flag to check
1293 	 * if these entries can be rewriten by arp updates.
1294 	 */
1295 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1296 	if (lle == NULL) {
1297 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1298 		    "entry for interface address\n");
1299 		return;
1300 	}
1301 
1302 	IF_AFDATA_WLOCK(ifp);
1303 	LLE_WLOCK(lle);
1304 	/* Unlink any entry if exists */
1305 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1306 	if (lle_tmp != NULL)
1307 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1308 
1309 	lltable_link_entry(LLTABLE(ifp), lle);
1310 	IF_AFDATA_WUNLOCK(ifp);
1311 
1312 	if (lle_tmp != NULL)
1313 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1314 
1315 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1316 	LLE_WUNLOCK(lle);
1317 	if (lle_tmp != NULL)
1318 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1319 }
1320 
1321 /*
1322  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1323  * of valid values.
1324  */
1325 static int
1326 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1327 {
1328 	int error;
1329 	int rexmit_count = *(int *)arg1;
1330 
1331 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1332 
1333 	/* Enforce limits on any new value that may have been set. */
1334 	if (!error && req->newptr) {
1335 		/* A new value was set. */
1336 		if (rexmit_count < 0) {
1337 			rexmit_count = 0;
1338 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1339 			rexmit_count = MAX_GARP_RETRANSMITS;
1340 		}
1341 		*(int *)arg1 = rexmit_count;
1342 	}
1343 
1344 	return (error);
1345 }
1346 
1347 /*
1348  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1349  * retransmit it again. A pending callout owns a reference to the ifa.
1350  */
1351 static void
1352 garp_rexmit(void *arg)
1353 {
1354 	struct in_ifaddr *ia = arg;
1355 
1356 	if (callout_pending(&ia->ia_garp_timer) ||
1357 	    !callout_active(&ia->ia_garp_timer)) {
1358 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1359 		ifa_free(&ia->ia_ifa);
1360 		return;
1361 	}
1362 
1363 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1364 
1365 	/*
1366 	 * Drop lock while the ARP request is generated.
1367 	 */
1368 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1369 
1370 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1371 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1372 
1373 	/*
1374 	 * Increment the count of retransmissions. If the count has reached the
1375 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1376 	 * the callout to retransmit another GARP packet.
1377 	 */
1378 	++ia->ia_garp_count;
1379 	if (ia->ia_garp_count >= garp_rexmit_count) {
1380 		ifa_free(&ia->ia_ifa);
1381 	} else {
1382 		int rescheduled;
1383 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1384 		rescheduled = callout_reset(&ia->ia_garp_timer,
1385 		    (1 << ia->ia_garp_count) * hz,
1386 		    garp_rexmit, ia);
1387 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1388 		if (rescheduled) {
1389 			ifa_free(&ia->ia_ifa);
1390 		}
1391 	}
1392 
1393 	CURVNET_RESTORE();
1394 }
1395 
1396 /*
1397  * Start the GARP retransmit timer.
1398  *
1399  * A single GARP is always transmitted when an IPv4 address is added
1400  * to an interface and that is usually sufficient. However, in some
1401  * circumstances, such as when a shared address is passed between
1402  * cluster nodes, this single GARP may occasionally be dropped or
1403  * lost. This can lead to neighbors on the network link working with a
1404  * stale ARP cache and sending packets destined for that address to
1405  * the node that previously owned the address, which may not respond.
1406  *
1407  * To avoid this situation, GARP retransmits can be enabled by setting
1408  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1409  * than zero. The setting represents the maximum number of
1410  * retransmissions. The interval between retransmissions is calculated
1411  * using an exponential backoff algorithm, doubling each time, so the
1412  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1413  */
1414 static void
1415 garp_timer_start(struct ifaddr *ifa)
1416 {
1417 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1418 
1419 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1420 	ia->ia_garp_count = 0;
1421 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1422 	    garp_rexmit, ia) == 0) {
1423 		ifa_ref(ifa);
1424 	}
1425 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1426 }
1427 
1428 void
1429 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1430 {
1431 	struct epoch_tracker et;
1432 	const struct sockaddr_in *dst_in;
1433 	const struct sockaddr *dst;
1434 
1435 	if (ifa->ifa_carp != NULL)
1436 		return;
1437 
1438 	dst = ifa->ifa_addr;
1439 	dst_in = (const struct sockaddr_in *)dst;
1440 
1441 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1442 		return;
1443 	NET_EPOCH_ENTER(et);
1444 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1445 	NET_EPOCH_EXIT(et);
1446 	if (garp_rexmit_count > 0) {
1447 		garp_timer_start(ifa);
1448 	}
1449 
1450 	arp_add_ifa_lle(ifp, dst);
1451 }
1452 
1453 void
1454 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1455 {
1456 
1457 	if (ntohl(addr.s_addr) != INADDR_ANY)
1458 		arprequest(ifp, &addr, &addr, enaddr);
1459 }
1460 
1461 /*
1462  * Sends gratuitous ARPs for each ifaddr to notify other
1463  * nodes about the address change.
1464  */
1465 static __noinline void
1466 arp_handle_ifllchange(struct ifnet *ifp)
1467 {
1468 	struct ifaddr *ifa;
1469 
1470 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1471 		if (ifa->ifa_addr->sa_family == AF_INET)
1472 			arp_ifinit(ifp, ifa);
1473 	}
1474 }
1475 
1476 /*
1477  * A handler for interface link layer address change event.
1478  */
1479 static void
1480 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1481 {
1482 
1483 	lltable_update_ifaddr(LLTABLE(ifp));
1484 
1485 	if ((ifp->if_flags & IFF_UP) != 0)
1486 		arp_handle_ifllchange(ifp);
1487 }
1488 
1489 static void
1490 vnet_arp_init(void)
1491 {
1492 
1493 	if (IS_DEFAULT_VNET(curvnet)) {
1494 		netisr_register(&arp_nh);
1495 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1496 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1497 	}
1498 #ifdef VIMAGE
1499 	else
1500 		netisr_register_vnet(&arp_nh);
1501 #endif
1502 }
1503 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1504     vnet_arp_init, 0);
1505 
1506 #ifdef VIMAGE
1507 /*
1508  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1509  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1510  */
1511 static void
1512 vnet_arp_destroy(__unused void *arg)
1513 {
1514 
1515 	netisr_unregister_vnet(&arp_nh);
1516 }
1517 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1518     vnet_arp_destroy, NULL);
1519 #endif
1520