xref: /freebsd/sys/netinet/if_ether.c (revision 63d1fd5970ec814904aa0f4580b10a0d302d08b2)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
30  */
31 
32 /*
33  * Ethernet address resolution protocol.
34  * TODO:
35  *	add "inuse/lock" bit (or ref. count) along with valid bit
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_inet.h"
42 
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/queue.h>
47 #include <sys/sysctl.h>
48 #include <sys/systm.h>
49 #include <sys/mbuf.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/rmlock.h>
53 #include <sys/socket.h>
54 #include <sys/syslog.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/if_types.h>
60 #include <net/netisr.h>
61 #include <net/ethernet.h>
62 #include <net/route.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_fib.h>
67 #include <netinet/in_var.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
70 #ifdef INET
71 #include <netinet/ip_carp.h>
72 #endif
73 
74 #include <security/mac/mac_framework.h>
75 
76 #define SIN(s) ((const struct sockaddr_in *)(s))
77 
78 static struct timeval arp_lastlog;
79 static int arp_curpps;
80 static int arp_maxpps = 1;
81 
82 /* Simple ARP state machine */
83 enum arp_llinfo_state {
84 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
85 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
86 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
87 	ARP_LLINFO_DELETED,	/* LLE is deleted */
88 };
89 
90 SYSCTL_DECL(_net_link_ether);
91 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
92 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
93 
94 /* timer values */
95 static VNET_DEFINE(int, arpt_keep) = (20*60);	/* once resolved, good for 20
96 						 * minutes */
97 static VNET_DEFINE(int, arp_maxtries) = 5;
98 static VNET_DEFINE(int, arp_proxyall) = 0;
99 static VNET_DEFINE(int, arpt_down) = 20;	/* keep incomplete entries for
100 						 * 20 seconds */
101 static VNET_DEFINE(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
102 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
103 VNET_PCPUSTAT_SYSINIT(arpstat);
104 
105 #ifdef VIMAGE
106 VNET_PCPUSTAT_SYSUNINIT(arpstat);
107 #endif /* VIMAGE */
108 
109 static VNET_DEFINE(int, arp_maxhold) = 1;
110 
111 #define	V_arpt_keep		VNET(arpt_keep)
112 #define	V_arpt_down		VNET(arpt_down)
113 #define	V_arpt_rexmit		VNET(arpt_rexmit)
114 #define	V_arp_maxtries		VNET(arp_maxtries)
115 #define	V_arp_proxyall		VNET(arp_proxyall)
116 #define	V_arp_maxhold		VNET(arp_maxhold)
117 
118 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
119 	&VNET_NAME(arpt_keep), 0,
120 	"ARP entry lifetime in seconds");
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
122 	&VNET_NAME(arp_maxtries), 0,
123 	"ARP resolution attempts before returning error");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
125 	&VNET_NAME(arp_proxyall), 0,
126 	"Enable proxy ARP for all suitable requests");
127 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
128 	&VNET_NAME(arpt_down), 0,
129 	"Incomplete ARP entry lifetime in seconds");
130 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
131     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
132 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
133 	&VNET_NAME(arp_maxhold), 0,
134 	"Number of packets to hold per ARP entry");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
136 	CTLFLAG_RW, &arp_maxpps, 0,
137 	"Maximum number of remotely triggered ARP messages that can be "
138 	"logged per second");
139 
140 /*
141  * Due to the exponential backoff algorithm used for the interval between GARP
142  * retransmissions, the maximum number of retransmissions is limited for
143  * sanity. This limit corresponds to a maximum interval between retransmissions
144  * of 2^16 seconds ~= 18 hours.
145  *
146  * Making this limit more dynamic is more complicated than worthwhile,
147  * especially since sending out GARPs spaced days apart would be of little
148  * use. A maximum dynamic limit would look something like:
149  *
150  * const int max = fls(INT_MAX / hz) - 1;
151  */
152 #define MAX_GARP_RETRANSMITS 16
153 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
154 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
155 
156 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
157     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
158     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
159     "Number of times to retransmit GARP packets;"
160     " 0 to disable, maximum of 16");
161 
162 #define	ARP_LOG(pri, ...)	do {					\
163 	if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
164 		log((pri), "arp: " __VA_ARGS__);			\
165 } while (0)
166 
167 
168 static void	arpintr(struct mbuf *);
169 static void	arptimer(void *);
170 #ifdef INET
171 static void	in_arpinput(struct mbuf *);
172 #endif
173 
174 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
175     struct ifnet *ifp, int bridged, struct llentry *la);
176 static void arp_mark_lle_reachable(struct llentry *la);
177 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
178 
179 static eventhandler_tag iflladdr_tag;
180 
181 static const struct netisr_handler arp_nh = {
182 	.nh_name = "arp",
183 	.nh_handler = arpintr,
184 	.nh_proto = NETISR_ARP,
185 	.nh_policy = NETISR_POLICY_SOURCE,
186 };
187 
188 /*
189  * Timeout routine.  Age arp_tab entries periodically.
190  */
191 static void
192 arptimer(void *arg)
193 {
194 	struct llentry *lle = (struct llentry *)arg;
195 	struct ifnet *ifp;
196 	int r_skip_req;
197 
198 	if (lle->la_flags & LLE_STATIC) {
199 		return;
200 	}
201 	LLE_WLOCK(lle);
202 	if (callout_pending(&lle->lle_timer)) {
203 		/*
204 		 * Here we are a bit odd here in the treatment of
205 		 * active/pending. If the pending bit is set, it got
206 		 * rescheduled before I ran. The active
207 		 * bit we ignore, since if it was stopped
208 		 * in ll_tablefree() and was currently running
209 		 * it would have return 0 so the code would
210 		 * not have deleted it since the callout could
211 		 * not be stopped so we want to go through
212 		 * with the delete here now. If the callout
213 		 * was restarted, the pending bit will be back on and
214 		 * we just want to bail since the callout_reset would
215 		 * return 1 and our reference would have been removed
216 		 * by arpresolve() below.
217 		 */
218 		LLE_WUNLOCK(lle);
219  		return;
220  	}
221 	ifp = lle->lle_tbl->llt_ifp;
222 	CURVNET_SET(ifp->if_vnet);
223 
224 	switch (lle->ln_state) {
225 	case ARP_LLINFO_REACHABLE:
226 
227 		/*
228 		 * Expiration time is approaching.
229 		 * Let's try to refresh entry if it is still
230 		 * in use.
231 		 *
232 		 * Set r_skip_req to get feedback from
233 		 * fast path. Change state and re-schedule
234 		 * ourselves.
235 		 */
236 		LLE_REQ_LOCK(lle);
237 		lle->r_skip_req = 1;
238 		LLE_REQ_UNLOCK(lle);
239 		lle->ln_state = ARP_LLINFO_VERIFY;
240 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
241 		LLE_WUNLOCK(lle);
242 		CURVNET_RESTORE();
243 		return;
244 	case ARP_LLINFO_VERIFY:
245 		LLE_REQ_LOCK(lle);
246 		r_skip_req = lle->r_skip_req;
247 		LLE_REQ_UNLOCK(lle);
248 
249 		if (r_skip_req == 0 && lle->la_preempt > 0) {
250 			/* Entry was used, issue refresh request */
251 			struct in_addr dst;
252 			dst = lle->r_l3addr.addr4;
253 			lle->la_preempt--;
254 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
255 			LLE_WUNLOCK(lle);
256 			arprequest(ifp, NULL, &dst, NULL);
257 			CURVNET_RESTORE();
258 			return;
259 		}
260 		/* Nothing happened. Reschedule if not too late */
261 		if (lle->la_expire > time_uptime) {
262 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
263 			LLE_WUNLOCK(lle);
264 			CURVNET_RESTORE();
265 			return;
266 		}
267 		break;
268 	case ARP_LLINFO_INCOMPLETE:
269 	case ARP_LLINFO_DELETED:
270 		break;
271 	}
272 
273 	if ((lle->la_flags & LLE_DELETED) == 0) {
274 		int evt;
275 
276 		if (lle->la_flags & LLE_VALID)
277 			evt = LLENTRY_EXPIRED;
278 		else
279 			evt = LLENTRY_TIMEDOUT;
280 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
281 	}
282 
283 	callout_stop(&lle->lle_timer);
284 
285 	/* XXX: LOR avoidance. We still have ref on lle. */
286 	LLE_WUNLOCK(lle);
287 	IF_AFDATA_LOCK(ifp);
288 	LLE_WLOCK(lle);
289 
290 	/* Guard against race with other llentry_free(). */
291 	if (lle->la_flags & LLE_LINKED) {
292 		LLE_REMREF(lle);
293 		lltable_unlink_entry(lle->lle_tbl, lle);
294 	}
295 	IF_AFDATA_UNLOCK(ifp);
296 
297 	size_t pkts_dropped = llentry_free(lle);
298 
299 	ARPSTAT_ADD(dropped, pkts_dropped);
300 	ARPSTAT_INC(timeouts);
301 
302 	CURVNET_RESTORE();
303 }
304 
305 /*
306  * Stores link-layer header for @ifp in format suitable for if_output()
307  * into buffer @buf. Resulting header length is stored in @bufsize.
308  *
309  * Returns 0 on success.
310  */
311 static int
312 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
313     size_t *bufsize)
314 {
315 	struct if_encap_req ereq;
316 	int error;
317 
318 	bzero(buf, *bufsize);
319 	bzero(&ereq, sizeof(ereq));
320 	ereq.buf = buf;
321 	ereq.bufsize = *bufsize;
322 	ereq.rtype = IFENCAP_LL;
323 	ereq.family = AF_ARP;
324 	ereq.lladdr = ar_tha(ah);
325 	ereq.hdata = (u_char *)ah;
326 	if (bcast)
327 		ereq.flags = IFENCAP_FLAG_BROADCAST;
328 	error = ifp->if_requestencap(ifp, &ereq);
329 	if (error == 0)
330 		*bufsize = ereq.bufsize;
331 
332 	return (error);
333 }
334 
335 
336 /*
337  * Broadcast an ARP request. Caller specifies:
338  *	- arp header source ip address
339  *	- arp header target ip address
340  *	- arp header source ethernet address
341  */
342 void
343 arprequest(struct ifnet *ifp, const struct in_addr *sip,
344     const struct in_addr *tip, u_char *enaddr)
345 {
346 	struct mbuf *m;
347 	struct arphdr *ah;
348 	struct sockaddr sa;
349 	u_char *carpaddr = NULL;
350 	uint8_t linkhdr[LLE_MAX_LINKHDR];
351 	size_t linkhdrsize;
352 	struct route ro;
353 	int error;
354 
355 	if (sip == NULL) {
356 		/*
357 		 * The caller did not supply a source address, try to find
358 		 * a compatible one among those assigned to this interface.
359 		 */
360 		struct ifaddr *ifa;
361 
362 		IF_ADDR_RLOCK(ifp);
363 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
364 			if (ifa->ifa_addr->sa_family != AF_INET)
365 				continue;
366 
367 			if (ifa->ifa_carp) {
368 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
369 					continue;
370 				sip = &IA_SIN(ifa)->sin_addr;
371 			} else {
372 				carpaddr = NULL;
373 				sip = &IA_SIN(ifa)->sin_addr;
374 			}
375 
376 			if (0 == ((sip->s_addr ^ tip->s_addr) &
377 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
378 				break;  /* found it. */
379 		}
380 		IF_ADDR_RUNLOCK(ifp);
381 		if (sip == NULL) {
382 			printf("%s: cannot find matching address\n", __func__);
383 			return;
384 		}
385 	}
386 	if (enaddr == NULL)
387 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
388 
389 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
390 		return;
391 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
392 		2 * ifp->if_addrlen;
393 	m->m_pkthdr.len = m->m_len;
394 	M_ALIGN(m, m->m_len);
395 	ah = mtod(m, struct arphdr *);
396 	bzero((caddr_t)ah, m->m_len);
397 #ifdef MAC
398 	mac_netinet_arp_send(ifp, m);
399 #endif
400 	ah->ar_pro = htons(ETHERTYPE_IP);
401 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
402 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
403 	ah->ar_op = htons(ARPOP_REQUEST);
404 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
405 	bcopy(sip, ar_spa(ah), ah->ar_pln);
406 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
407 	sa.sa_family = AF_ARP;
408 	sa.sa_len = 2;
409 
410 	/* Calculate link header for sending frame */
411 	bzero(&ro, sizeof(ro));
412 	linkhdrsize = sizeof(linkhdr);
413 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
414 	if (error != 0 && error != EAFNOSUPPORT) {
415 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
416 		    if_name(ifp), error);
417 		return;
418 	}
419 
420 	ro.ro_prepend = linkhdr;
421 	ro.ro_plen = linkhdrsize;
422 	ro.ro_flags = 0;
423 
424 	m->m_flags |= M_BCAST;
425 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
426 	(*ifp->if_output)(ifp, m, &sa, &ro);
427 	ARPSTAT_INC(txrequests);
428 }
429 
430 
431 /*
432  * Resolve an IP address into an ethernet address - heavy version.
433  * Used internally by arpresolve().
434  * We have already checked than  we can't use existing lle without
435  * modification so we have to acquire LLE_EXCLUSIVE lle lock.
436  *
437  * On success, desten and flags are filled in and the function returns 0;
438  * If the packet must be held pending resolution, we return EWOULDBLOCK
439  * On other errors, we return the corresponding error code.
440  * Note that m_freem() handles NULL.
441  */
442 static int
443 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
444 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
445 	struct llentry **plle)
446 {
447 	struct llentry *la = NULL, *la_tmp;
448 	struct mbuf *curr = NULL;
449 	struct mbuf *next = NULL;
450 	int error, renew;
451 	char *lladdr;
452 	int ll_len;
453 
454 	if (pflags != NULL)
455 		*pflags = 0;
456 	if (plle != NULL)
457 		*plle = NULL;
458 
459 	if ((flags & LLE_CREATE) == 0) {
460 		IF_AFDATA_RLOCK(ifp);
461 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
462 		IF_AFDATA_RUNLOCK(ifp);
463 	}
464 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
465 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
466 		if (la == NULL) {
467 			log(LOG_DEBUG,
468 			    "arpresolve: can't allocate llinfo for %s on %s\n",
469 			    inet_ntoa(SIN(dst)->sin_addr), if_name(ifp));
470 			m_freem(m);
471 			return (EINVAL);
472 		}
473 
474 		IF_AFDATA_WLOCK(ifp);
475 		LLE_WLOCK(la);
476 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
477 		/* Prefer ANY existing lle over newly-created one */
478 		if (la_tmp == NULL)
479 			lltable_link_entry(LLTABLE(ifp), la);
480 		IF_AFDATA_WUNLOCK(ifp);
481 		if (la_tmp != NULL) {
482 			lltable_free_entry(LLTABLE(ifp), la);
483 			la = la_tmp;
484 		}
485 	}
486 	if (la == NULL) {
487 		m_freem(m);
488 		return (EINVAL);
489 	}
490 
491 	if ((la->la_flags & LLE_VALID) &&
492 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
493 		if (flags & LLE_ADDRONLY) {
494 			lladdr = la->ll_addr;
495 			ll_len = ifp->if_addrlen;
496 		} else {
497 			lladdr = la->r_linkdata;
498 			ll_len = la->r_hdrlen;
499 		}
500 		bcopy(lladdr, desten, ll_len);
501 
502 		/* Check if we have feedback request from arptimer() */
503 		if (la->r_skip_req != 0) {
504 			LLE_REQ_LOCK(la);
505 			la->r_skip_req = 0; /* Notify that entry was used */
506 			LLE_REQ_UNLOCK(la);
507 		}
508 		if (pflags != NULL)
509 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
510 		if (plle) {
511 			LLE_ADDREF(la);
512 			*plle = la;
513 		}
514 		LLE_WUNLOCK(la);
515 		return (0);
516 	}
517 
518 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
519 	/*
520 	 * There is an arptab entry, but no ethernet address
521 	 * response yet.  Add the mbuf to the list, dropping
522 	 * the oldest packet if we have exceeded the system
523 	 * setting.
524 	 */
525 	if (m != NULL) {
526 		if (la->la_numheld >= V_arp_maxhold) {
527 			if (la->la_hold != NULL) {
528 				next = la->la_hold->m_nextpkt;
529 				m_freem(la->la_hold);
530 				la->la_hold = next;
531 				la->la_numheld--;
532 				ARPSTAT_INC(dropped);
533 			}
534 		}
535 		if (la->la_hold != NULL) {
536 			curr = la->la_hold;
537 			while (curr->m_nextpkt != NULL)
538 				curr = curr->m_nextpkt;
539 			curr->m_nextpkt = m;
540 		} else
541 			la->la_hold = m;
542 		la->la_numheld++;
543 	}
544 	/*
545 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
546 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
547 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
548 	 * ARP request, but not faster than one request per second.
549 	 */
550 	if (la->la_asked < V_arp_maxtries)
551 		error = EWOULDBLOCK;	/* First request. */
552 	else
553 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
554 
555 	if (renew) {
556 		int canceled;
557 
558 		LLE_ADDREF(la);
559 		la->la_expire = time_uptime;
560 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
561 		    arptimer, la);
562 		if (canceled)
563 			LLE_REMREF(la);
564 		la->la_asked++;
565 		LLE_WUNLOCK(la);
566 		arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
567 		return (error);
568 	}
569 
570 	LLE_WUNLOCK(la);
571 	return (error);
572 }
573 
574 /*
575  * Resolve an IP address into an ethernet address.
576  */
577 int
578 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
579     char *desten, uint32_t *pflags, struct llentry **plle)
580 {
581 	int error;
582 
583 	flags |= LLE_ADDRONLY;
584 	error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
585 	return (error);
586 }
587 
588 
589 /*
590  * Lookups link header based on an IP address.
591  * On input:
592  *    ifp is the interface we use
593  *    is_gw != 0 if @dst represents gateway to some destination
594  *    m is the mbuf. May be NULL if we don't have a packet.
595  *    dst is the next hop,
596  *    desten is the storage to put LL header.
597  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
598  *
599  * On success, full/partial link header and flags are filled in and
600  * the function returns 0.
601  * If the packet must be held pending resolution, we return EWOULDBLOCK
602  * On other errors, we return the corresponding error code.
603  * Note that m_freem() handles NULL.
604  */
605 int
606 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
607 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
608 	struct llentry **plle)
609 {
610 	struct llentry *la = NULL;
611 
612 	if (pflags != NULL)
613 		*pflags = 0;
614 	if (plle != NULL)
615 		*plle = NULL;
616 
617 	if (m != NULL) {
618 		if (m->m_flags & M_BCAST) {
619 			/* broadcast */
620 			(void)memcpy(desten,
621 			    ifp->if_broadcastaddr, ifp->if_addrlen);
622 			return (0);
623 		}
624 		if (m->m_flags & M_MCAST) {
625 			/* multicast */
626 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
627 			return (0);
628 		}
629 	}
630 
631 	IF_AFDATA_RLOCK(ifp);
632 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
633 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
634 		/* Entry found, let's copy lle info */
635 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
636 		if (pflags != NULL)
637 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
638 		/* Check if we have feedback request from arptimer() */
639 		if (la->r_skip_req != 0) {
640 			LLE_REQ_LOCK(la);
641 			la->r_skip_req = 0; /* Notify that entry was used */
642 			LLE_REQ_UNLOCK(la);
643 		}
644 		if (plle) {
645 			LLE_ADDREF(la);
646 			*plle = la;
647 			LLE_WUNLOCK(la);
648 		}
649 		IF_AFDATA_RUNLOCK(ifp);
650 		return (0);
651 	}
652 	if (plle && la)
653 		LLE_WUNLOCK(la);
654 	IF_AFDATA_RUNLOCK(ifp);
655 
656 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
657 	    desten, pflags, plle));
658 }
659 
660 /*
661  * Common length and type checks are done here,
662  * then the protocol-specific routine is called.
663  */
664 static void
665 arpintr(struct mbuf *m)
666 {
667 	struct arphdr *ar;
668 	struct ifnet *ifp;
669 	char *layer;
670 	int hlen;
671 
672 	ifp = m->m_pkthdr.rcvif;
673 
674 	if (m->m_len < sizeof(struct arphdr) &&
675 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
676 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
677 		    if_name(ifp));
678 		return;
679 	}
680 	ar = mtod(m, struct arphdr *);
681 
682 	/* Check if length is sufficient */
683 	if (m->m_len <  arphdr_len(ar)) {
684 		m = m_pullup(m, arphdr_len(ar));
685 		if (m == NULL) {
686 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
687 			    if_name(ifp));
688 			return;
689 		}
690 		ar = mtod(m, struct arphdr *);
691 	}
692 
693 	hlen = 0;
694 	layer = "";
695 	switch (ntohs(ar->ar_hrd)) {
696 	case ARPHRD_ETHER:
697 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
698 		layer = "ethernet";
699 		break;
700 	case ARPHRD_IEEE802:
701 		hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
702 		layer = "fddi";
703 		break;
704 	case ARPHRD_ARCNET:
705 		hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
706 		layer = "arcnet";
707 		break;
708 	case ARPHRD_INFINIBAND:
709 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
710 		layer = "infiniband";
711 		break;
712 	case ARPHRD_IEEE1394:
713 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
714 		layer = "firewire";
715 
716 		/*
717 		 * Restrict too long hardware addresses.
718 		 * Currently we are capable of handling 20-byte
719 		 * addresses ( sizeof(lle->ll_addr) )
720 		 */
721 		if (ar->ar_hln >= 20)
722 			hlen = 16;
723 		break;
724 	default:
725 		ARP_LOG(LOG_NOTICE,
726 		    "packet with unknown hardware format 0x%02d received on "
727 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
728 		m_freem(m);
729 		return;
730 	}
731 
732 	if (hlen != 0 && hlen != ar->ar_hln) {
733 		ARP_LOG(LOG_NOTICE,
734 		    "packet with invalid %s address length %d received on %s\n",
735 		    layer, ar->ar_hln, if_name(ifp));
736 		m_freem(m);
737 		return;
738 	}
739 
740 	ARPSTAT_INC(received);
741 	switch (ntohs(ar->ar_pro)) {
742 #ifdef INET
743 	case ETHERTYPE_IP:
744 		in_arpinput(m);
745 		return;
746 #endif
747 	}
748 	m_freem(m);
749 }
750 
751 #ifdef INET
752 /*
753  * ARP for Internet protocols on 10 Mb/s Ethernet.
754  * Algorithm is that given in RFC 826.
755  * In addition, a sanity check is performed on the sender
756  * protocol address, to catch impersonators.
757  * We no longer handle negotiations for use of trailer protocol:
758  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
759  * along with IP replies if we wanted trailers sent to us,
760  * and also sent them in response to IP replies.
761  * This allowed either end to announce the desire to receive
762  * trailer packets.
763  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
764  * but formerly didn't normally send requests.
765  */
766 static int log_arp_wrong_iface = 1;
767 static int log_arp_movements = 1;
768 static int log_arp_permanent_modify = 1;
769 static int allow_multicast = 0;
770 
771 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
772 	&log_arp_wrong_iface, 0,
773 	"log arp packets arriving on the wrong interface");
774 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
775 	&log_arp_movements, 0,
776 	"log arp replies from MACs different than the one in the cache");
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
778 	&log_arp_permanent_modify, 0,
779 	"log arp replies from MACs different than the one in the permanent arp entry");
780 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
781 	&allow_multicast, 0, "accept multicast addresses");
782 
783 static void
784 in_arpinput(struct mbuf *m)
785 {
786 	struct rm_priotracker in_ifa_tracker;
787 	struct arphdr *ah;
788 	struct ifnet *ifp = m->m_pkthdr.rcvif;
789 	struct llentry *la = NULL, *la_tmp;
790 	struct ifaddr *ifa;
791 	struct in_ifaddr *ia;
792 	struct sockaddr sa;
793 	struct in_addr isaddr, itaddr, myaddr;
794 	u_int8_t *enaddr = NULL;
795 	int op;
796 	int bridged = 0, is_bridge = 0;
797 	int carped;
798 	struct sockaddr_in sin;
799 	struct sockaddr *dst;
800 	struct nhop4_basic nh4;
801 	uint8_t linkhdr[LLE_MAX_LINKHDR];
802 	struct route ro;
803 	size_t linkhdrsize;
804 	int lladdr_off;
805 	int error;
806 
807 	sin.sin_len = sizeof(struct sockaddr_in);
808 	sin.sin_family = AF_INET;
809 	sin.sin_addr.s_addr = 0;
810 
811 	if (ifp->if_bridge)
812 		bridged = 1;
813 	if (ifp->if_type == IFT_BRIDGE)
814 		is_bridge = 1;
815 
816 	/*
817 	 * We already have checked that mbuf contains enough contiguous data
818 	 * to hold entire arp message according to the arp header.
819 	 */
820 	ah = mtod(m, struct arphdr *);
821 
822 	/*
823 	 * ARP is only for IPv4 so we can reject packets with
824 	 * a protocol length not equal to an IPv4 address.
825 	 */
826 	if (ah->ar_pln != sizeof(struct in_addr)) {
827 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
828 		    sizeof(struct in_addr));
829 		goto drop;
830 	}
831 
832 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
833 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
834 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
835 		goto drop;
836 	}
837 
838 	op = ntohs(ah->ar_op);
839 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
840 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
841 
842 	if (op == ARPOP_REPLY)
843 		ARPSTAT_INC(rxreplies);
844 
845 	/*
846 	 * For a bridge, we want to check the address irrespective
847 	 * of the receive interface. (This will change slightly
848 	 * when we have clusters of interfaces).
849 	 */
850 	IN_IFADDR_RLOCK(&in_ifa_tracker);
851 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
852 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
853 		    ia->ia_ifp == ifp) &&
854 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
855 		    (ia->ia_ifa.ifa_carp == NULL ||
856 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
857 			ifa_ref(&ia->ia_ifa);
858 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
859 			goto match;
860 		}
861 	}
862 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
863 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
864 		    ia->ia_ifp == ifp) &&
865 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
866 			ifa_ref(&ia->ia_ifa);
867 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
868 			goto match;
869 		}
870 
871 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
872   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
873   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
874   addr == ia->ia_addr.sin_addr.s_addr)
875 	/*
876 	 * Check the case when bridge shares its MAC address with
877 	 * some of its children, so packets are claimed by bridge
878 	 * itself (bridge_input() does it first), but they are really
879 	 * meant to be destined to the bridge member.
880 	 */
881 	if (is_bridge) {
882 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
883 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
884 				ifa_ref(&ia->ia_ifa);
885 				ifp = ia->ia_ifp;
886 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
887 				goto match;
888 			}
889 		}
890 	}
891 #undef BDG_MEMBER_MATCHES_ARP
892 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
893 
894 	/*
895 	 * No match, use the first inet address on the receive interface
896 	 * as a dummy address for the rest of the function.
897 	 */
898 	IF_ADDR_RLOCK(ifp);
899 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
900 		if (ifa->ifa_addr->sa_family == AF_INET &&
901 		    (ifa->ifa_carp == NULL ||
902 		    (*carp_iamatch_p)(ifa, &enaddr))) {
903 			ia = ifatoia(ifa);
904 			ifa_ref(ifa);
905 			IF_ADDR_RUNLOCK(ifp);
906 			goto match;
907 		}
908 	IF_ADDR_RUNLOCK(ifp);
909 
910 	/*
911 	 * If bridging, fall back to using any inet address.
912 	 */
913 	IN_IFADDR_RLOCK(&in_ifa_tracker);
914 	if (!bridged || (ia = TAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
915 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
916 		goto drop;
917 	}
918 	ifa_ref(&ia->ia_ifa);
919 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
920 match:
921 	if (!enaddr)
922 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
923 	carped = (ia->ia_ifa.ifa_carp != NULL);
924 	myaddr = ia->ia_addr.sin_addr;
925 	ifa_free(&ia->ia_ifa);
926 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
927 		goto drop;	/* it's from me, ignore it. */
928 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
929 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
930 		    "%s!\n", inet_ntoa(isaddr));
931 		goto drop;
932 	}
933 
934 	if (ifp->if_addrlen != ah->ar_hln) {
935 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
936 		    "i/f %d (ignored)\n", ifp->if_addrlen,
937 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
938 		    ifp->if_addrlen);
939 		goto drop;
940 	}
941 
942 	/*
943 	 * Warn if another host is using the same IP address, but only if the
944 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
945 	 * case we suppress the warning to avoid false positive complaints of
946 	 * potential misconfiguration.
947 	 */
948 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
949 	    myaddr.s_addr != 0) {
950 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
951 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
952 		   inet_ntoa(isaddr), ifp->if_xname);
953 		itaddr = myaddr;
954 		ARPSTAT_INC(dupips);
955 		goto reply;
956 	}
957 	if (ifp->if_flags & IFF_STATICARP)
958 		goto reply;
959 
960 	bzero(&sin, sizeof(sin));
961 	sin.sin_len = sizeof(struct sockaddr_in);
962 	sin.sin_family = AF_INET;
963 	sin.sin_addr = isaddr;
964 	dst = (struct sockaddr *)&sin;
965 	IF_AFDATA_RLOCK(ifp);
966 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
967 	IF_AFDATA_RUNLOCK(ifp);
968 	if (la != NULL)
969 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
970 	else if (itaddr.s_addr == myaddr.s_addr) {
971 		/*
972 		 * Request/reply to our address, but no lle exists yet.
973 		 * Calculate full link prepend to use in lle.
974 		 */
975 		linkhdrsize = sizeof(linkhdr);
976 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
977 		    &linkhdrsize, &lladdr_off) != 0)
978 			goto reply;
979 
980 		/* Allocate new entry */
981 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
982 		if (la == NULL) {
983 
984 			/*
985 			 * lle creation may fail if source address belongs
986 			 * to non-directly connected subnet. However, we
987 			 * will try to answer the request instead of dropping
988 			 * frame.
989 			 */
990 			goto reply;
991 		}
992 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
993 		    lladdr_off);
994 
995 		IF_AFDATA_WLOCK(ifp);
996 		LLE_WLOCK(la);
997 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
998 
999 		/*
1000 		 * Check if lle still does not exists.
1001 		 * If it does, that means that we either
1002 		 * 1) have configured it explicitly, via
1003 		 * 1a) 'arp -s' static entry or
1004 		 * 1b) interface address static record
1005 		 * or
1006 		 * 2) it was the result of sending first packet to-host
1007 		 * or
1008 		 * 3) it was another arp reply packet we handled in
1009 		 * different thread.
1010 		 *
1011 		 * In all cases except 3) we definitely need to prefer
1012 		 * existing lle. For the sake of simplicity, prefer any
1013 		 * existing lle over newly-create one.
1014 		 */
1015 		if (la_tmp == NULL)
1016 			lltable_link_entry(LLTABLE(ifp), la);
1017 		IF_AFDATA_WUNLOCK(ifp);
1018 
1019 		if (la_tmp == NULL) {
1020 			arp_mark_lle_reachable(la);
1021 			LLE_WUNLOCK(la);
1022 		} else {
1023 			/* Free newly-create entry and handle packet */
1024 			lltable_free_entry(LLTABLE(ifp), la);
1025 			la = la_tmp;
1026 			la_tmp = NULL;
1027 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1028 			/* arp_check_update_lle() returns @la unlocked */
1029 		}
1030 		la = NULL;
1031 	}
1032 reply:
1033 	if (op != ARPOP_REQUEST)
1034 		goto drop;
1035 	ARPSTAT_INC(rxrequests);
1036 
1037 	if (itaddr.s_addr == myaddr.s_addr) {
1038 		/* Shortcut.. the receiving interface is the target. */
1039 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1040 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1041 	} else {
1042 		struct llentry *lle = NULL;
1043 
1044 		sin.sin_addr = itaddr;
1045 		IF_AFDATA_RLOCK(ifp);
1046 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1047 		IF_AFDATA_RUNLOCK(ifp);
1048 
1049 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1050 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1051 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1052 			LLE_RUNLOCK(lle);
1053 		} else {
1054 
1055 			if (lle != NULL)
1056 				LLE_RUNLOCK(lle);
1057 
1058 			if (!V_arp_proxyall)
1059 				goto drop;
1060 
1061 			/* XXX MRT use table 0 for arp reply  */
1062 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1063 				goto drop;
1064 
1065 			/*
1066 			 * Don't send proxies for nodes on the same interface
1067 			 * as this one came out of, or we'll get into a fight
1068 			 * over who claims what Ether address.
1069 			 */
1070 			if (nh4.nh_ifp == ifp)
1071 				goto drop;
1072 
1073 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1074 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1075 
1076 			/*
1077 			 * Also check that the node which sent the ARP packet
1078 			 * is on the interface we expect it to be on. This
1079 			 * avoids ARP chaos if an interface is connected to the
1080 			 * wrong network.
1081 			 */
1082 
1083 			/* XXX MRT use table 0 for arp checks */
1084 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1085 				goto drop;
1086 			if (nh4.nh_ifp != ifp) {
1087 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1088 				    " from %s via %s\n",
1089 				    inet_ntoa(isaddr), ifp->if_xname);
1090 				goto drop;
1091 			}
1092 
1093 #ifdef DEBUG_PROXY
1094 			printf("arp: proxying for %s\n", inet_ntoa(itaddr));
1095 #endif
1096 		}
1097 	}
1098 
1099 	if (itaddr.s_addr == myaddr.s_addr &&
1100 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1101 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1102 #ifdef DEBUG_LINKLOCAL
1103 		printf("arp: sending reply for link-local addr %s\n",
1104 		    inet_ntoa(itaddr));
1105 #endif
1106 		m->m_flags |= M_BCAST;
1107 		m->m_flags &= ~M_MCAST;
1108 	} else {
1109 		/* default behaviour; never reply by broadcast. */
1110 		m->m_flags &= ~(M_BCAST|M_MCAST);
1111 	}
1112 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1113 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1114 	ah->ar_op = htons(ARPOP_REPLY);
1115 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1116 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1117 	m->m_pkthdr.len = m->m_len;
1118 	m->m_pkthdr.rcvif = NULL;
1119 	sa.sa_family = AF_ARP;
1120 	sa.sa_len = 2;
1121 
1122 	/* Calculate link header for sending frame */
1123 	bzero(&ro, sizeof(ro));
1124 	linkhdrsize = sizeof(linkhdr);
1125 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1126 
1127 	/*
1128 	 * arp_fillheader() may fail due to lack of support inside encap request
1129 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1130 	 * by if_output().
1131 	 */
1132 	if (error != 0 && error != EAFNOSUPPORT) {
1133 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1134 		    if_name(ifp), error);
1135 		return;
1136 	}
1137 
1138 	ro.ro_prepend = linkhdr;
1139 	ro.ro_plen = linkhdrsize;
1140 	ro.ro_flags = 0;
1141 
1142 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1143 	(*ifp->if_output)(ifp, m, &sa, &ro);
1144 	ARPSTAT_INC(txreplies);
1145 	return;
1146 
1147 drop:
1148 	m_freem(m);
1149 }
1150 #endif
1151 
1152 /*
1153  * Checks received arp data against existing @la.
1154  * Updates lle state/performs notification if necessary.
1155  */
1156 static void
1157 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1158     int bridged, struct llentry *la)
1159 {
1160 	struct sockaddr sa;
1161 	struct mbuf *m_hold, *m_hold_next;
1162 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1163 	size_t linkhdrsize;
1164 	int lladdr_off;
1165 
1166 	LLE_WLOCK_ASSERT(la);
1167 
1168 	/* the following is not an error when doing bridging */
1169 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1170 		if (log_arp_wrong_iface)
1171 			ARP_LOG(LOG_WARNING, "%s is on %s "
1172 			    "but got reply from %*D on %s\n",
1173 			    inet_ntoa(isaddr),
1174 			    la->lle_tbl->llt_ifp->if_xname,
1175 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1176 			    ifp->if_xname);
1177 		LLE_WUNLOCK(la);
1178 		return;
1179 	}
1180 	if ((la->la_flags & LLE_VALID) &&
1181 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1182 		if (la->la_flags & LLE_STATIC) {
1183 			LLE_WUNLOCK(la);
1184 			if (log_arp_permanent_modify)
1185 				ARP_LOG(LOG_ERR,
1186 				    "%*D attempts to modify "
1187 				    "permanent entry for %s on %s\n",
1188 				    ifp->if_addrlen,
1189 				    (u_char *)ar_sha(ah), ":",
1190 				    inet_ntoa(isaddr), ifp->if_xname);
1191 			return;
1192 		}
1193 		if (log_arp_movements) {
1194 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1195 			    "to %*D on %s\n",
1196 			    inet_ntoa(isaddr),
1197 			    ifp->if_addrlen,
1198 			    (u_char *)&la->ll_addr, ":",
1199 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1200 			    ifp->if_xname);
1201 		}
1202 	}
1203 
1204 	/* Calculate full link prepend to use in lle */
1205 	linkhdrsize = sizeof(linkhdr);
1206 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1207 	    &linkhdrsize, &lladdr_off) != 0)
1208 		return;
1209 
1210 	/* Check if something has changed */
1211 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1212 	    (la->la_flags & LLE_VALID) == 0) {
1213 		/* Try to perform LLE update */
1214 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1215 		    lladdr_off) == 0)
1216 			return;
1217 
1218 		/* Clear fast path feedback request if set */
1219 		la->r_skip_req = 0;
1220 	}
1221 
1222 	arp_mark_lle_reachable(la);
1223 
1224 	/*
1225 	 * The packets are all freed within the call to the output
1226 	 * routine.
1227 	 *
1228 	 * NB: The lock MUST be released before the call to the
1229 	 * output routine.
1230 	 */
1231 	if (la->la_hold != NULL) {
1232 		m_hold = la->la_hold;
1233 		la->la_hold = NULL;
1234 		la->la_numheld = 0;
1235 		lltable_fill_sa_entry(la, &sa);
1236 		LLE_WUNLOCK(la);
1237 		for (; m_hold != NULL; m_hold = m_hold_next) {
1238 			m_hold_next = m_hold->m_nextpkt;
1239 			m_hold->m_nextpkt = NULL;
1240 			/* Avoid confusing lower layers. */
1241 			m_clrprotoflags(m_hold);
1242 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1243 		}
1244 	} else
1245 		LLE_WUNLOCK(la);
1246 }
1247 
1248 static void
1249 arp_mark_lle_reachable(struct llentry *la)
1250 {
1251 	int canceled, wtime;
1252 
1253 	LLE_WLOCK_ASSERT(la);
1254 
1255 	la->ln_state = ARP_LLINFO_REACHABLE;
1256 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1257 
1258 	if (!(la->la_flags & LLE_STATIC)) {
1259 		LLE_ADDREF(la);
1260 		la->la_expire = time_uptime + V_arpt_keep;
1261 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1262 		if (wtime < 0)
1263 			wtime = V_arpt_keep;
1264 		canceled = callout_reset(&la->lle_timer,
1265 		    hz * wtime, arptimer, la);
1266 		if (canceled)
1267 			LLE_REMREF(la);
1268 	}
1269 	la->la_asked = 0;
1270 	la->la_preempt = V_arp_maxtries;
1271 }
1272 
1273 /*
1274  * Add pernament link-layer record for given interface address.
1275  */
1276 static __noinline void
1277 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1278 {
1279 	struct llentry *lle, *lle_tmp;
1280 
1281 	/*
1282 	 * Interface address LLE record is considered static
1283 	 * because kernel code relies on LLE_STATIC flag to check
1284 	 * if these entries can be rewriten by arp updates.
1285 	 */
1286 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1287 	if (lle == NULL) {
1288 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1289 		    "entry for interface address\n");
1290 		return;
1291 	}
1292 
1293 	IF_AFDATA_WLOCK(ifp);
1294 	LLE_WLOCK(lle);
1295 	/* Unlink any entry if exists */
1296 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1297 	if (lle_tmp != NULL)
1298 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1299 
1300 	lltable_link_entry(LLTABLE(ifp), lle);
1301 	IF_AFDATA_WUNLOCK(ifp);
1302 
1303 	if (lle_tmp != NULL)
1304 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1305 
1306 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1307 	LLE_WUNLOCK(lle);
1308 	if (lle_tmp != NULL)
1309 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1310 }
1311 
1312 /*
1313  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1314  * of valid values.
1315  */
1316 static int
1317 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1318 {
1319 	int error;
1320 	int rexmit_count = *(int *)arg1;
1321 
1322 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1323 
1324 	/* Enforce limits on any new value that may have been set. */
1325 	if (!error && req->newptr) {
1326 		/* A new value was set. */
1327 		if (rexmit_count < 0) {
1328 			rexmit_count = 0;
1329 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1330 			rexmit_count = MAX_GARP_RETRANSMITS;
1331 		}
1332 		*(int *)arg1 = rexmit_count;
1333 	}
1334 
1335 	return (error);
1336 }
1337 
1338 /*
1339  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1340  * retransmit it again. A pending callout owns a reference to the ifa.
1341  */
1342 static void
1343 garp_rexmit(void *arg)
1344 {
1345 	struct in_ifaddr *ia = arg;
1346 
1347 	if (callout_pending(&ia->ia_garp_timer) ||
1348 	    !callout_active(&ia->ia_garp_timer)) {
1349 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1350 		ifa_free(&ia->ia_ifa);
1351 		return;
1352 	}
1353 
1354 	/*
1355 	 * Drop lock while the ARP request is generated.
1356 	 */
1357 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1358 
1359 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1360 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1361 
1362 	/*
1363 	 * Increment the count of retransmissions. If the count has reached the
1364 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1365 	 * the callout to retransmit another GARP packet.
1366 	 */
1367 	++ia->ia_garp_count;
1368 	if (ia->ia_garp_count >= garp_rexmit_count) {
1369 		ifa_free(&ia->ia_ifa);
1370 	} else {
1371 		int rescheduled;
1372 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1373 		rescheduled = callout_reset(&ia->ia_garp_timer,
1374 		    (1 << ia->ia_garp_count) * hz,
1375 		    garp_rexmit, ia);
1376 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1377 		if (rescheduled) {
1378 			ifa_free(&ia->ia_ifa);
1379 		}
1380 	}
1381 }
1382 
1383 /*
1384  * Start the GARP retransmit timer.
1385  *
1386  * A single GARP is always transmitted when an IPv4 address is added
1387  * to an interface and that is usually sufficient. However, in some
1388  * circumstances, such as when a shared address is passed between
1389  * cluster nodes, this single GARP may occasionally be dropped or
1390  * lost. This can lead to neighbors on the network link working with a
1391  * stale ARP cache and sending packets destined for that address to
1392  * the node that previously owned the address, which may not respond.
1393  *
1394  * To avoid this situation, GARP retransmits can be enabled by setting
1395  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1396  * than zero. The setting represents the maximum number of
1397  * retransmissions. The interval between retransmissions is calculated
1398  * using an exponential backoff algorithm, doubling each time, so the
1399  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1400  */
1401 static void
1402 garp_timer_start(struct ifaddr *ifa)
1403 {
1404 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1405 
1406 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1407 	ia->ia_garp_count = 0;
1408 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1409 	    garp_rexmit, ia) == 0) {
1410 		ifa_ref(ifa);
1411 	}
1412 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1413 }
1414 
1415 void
1416 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1417 {
1418 	const struct sockaddr_in *dst_in;
1419 	const struct sockaddr *dst;
1420 
1421 	if (ifa->ifa_carp != NULL)
1422 		return;
1423 
1424 	dst = ifa->ifa_addr;
1425 	dst_in = (const struct sockaddr_in *)dst;
1426 
1427 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1428 		return;
1429 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1430 	if (garp_rexmit_count > 0) {
1431 		garp_timer_start(ifa);
1432 	}
1433 
1434 	arp_add_ifa_lle(ifp, dst);
1435 }
1436 
1437 void
1438 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1439 {
1440 
1441 	if (ntohl(addr.s_addr) != INADDR_ANY)
1442 		arprequest(ifp, &addr, &addr, enaddr);
1443 }
1444 
1445 /*
1446  * Sends gratuitous ARPs for each ifaddr to notify other
1447  * nodes about the address change.
1448  */
1449 static __noinline void
1450 arp_handle_ifllchange(struct ifnet *ifp)
1451 {
1452 	struct ifaddr *ifa;
1453 
1454 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1455 		if (ifa->ifa_addr->sa_family == AF_INET)
1456 			arp_ifinit(ifp, ifa);
1457 	}
1458 }
1459 
1460 /*
1461  * A handler for interface link layer address change event.
1462  */
1463 static void
1464 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1465 {
1466 
1467 	lltable_update_ifaddr(LLTABLE(ifp));
1468 
1469 	if ((ifp->if_flags & IFF_UP) != 0)
1470 		arp_handle_ifllchange(ifp);
1471 }
1472 
1473 static void
1474 vnet_arp_init(void)
1475 {
1476 
1477 	if (IS_DEFAULT_VNET(curvnet)) {
1478 		netisr_register(&arp_nh);
1479 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1480 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1481 	}
1482 #ifdef VIMAGE
1483 	else
1484 		netisr_register_vnet(&arp_nh);
1485 #endif
1486 }
1487 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1488     vnet_arp_init, 0);
1489 
1490 #ifdef VIMAGE
1491 /*
1492  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1493  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1494  */
1495 static void
1496 vnet_arp_destroy(__unused void *arg)
1497 {
1498 
1499 	netisr_unregister_vnet(&arp_nh);
1500 }
1501 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1502     vnet_arp_destroy, NULL);
1503 #endif
1504