xref: /freebsd/sys/netinet/if_ether.c (revision 6683132d54bd6d589889e43dabdc53d35e38a028)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)if_ether.c	8.1 (Berkeley) 6/10/93
32  */
33 
34 /*
35  * Ethernet address resolution protocol.
36  * TODO:
37  *	add "inuse/lock" bit (or ref. count) along with valid bit
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_inet.h"
44 
45 #include <sys/param.h>
46 #include <sys/eventhandler.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/queue.h>
50 #include <sys/sysctl.h>
51 #include <sys/systm.h>
52 #include <sys/mbuf.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/rmlock.h>
56 #include <sys/socket.h>
57 #include <sys/syslog.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/netisr.h>
64 #include <net/ethernet.h>
65 #include <net/route.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_fib.h>
70 #include <netinet/in_var.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #ifdef INET
74 #include <netinet/ip_carp.h>
75 #endif
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define SIN(s) ((const struct sockaddr_in *)(s))
80 
81 static struct timeval arp_lastlog;
82 static int arp_curpps;
83 static int arp_maxpps = 1;
84 
85 /* Simple ARP state machine */
86 enum arp_llinfo_state {
87 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
88 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
89 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
90 	ARP_LLINFO_DELETED,	/* LLE is deleted */
91 };
92 
93 SYSCTL_DECL(_net_link_ether);
94 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
95 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
96 
97 /* timer values */
98 VNET_DEFINE_STATIC(int, arpt_keep) = (20*60);	/* once resolved, good for 20
99 						 * minutes */
100 VNET_DEFINE_STATIC(int, arp_maxtries) = 5;
101 VNET_DEFINE_STATIC(int, arp_proxyall) = 0;
102 VNET_DEFINE_STATIC(int, arpt_down) = 20;	/* keep incomplete entries for
103 						 * 20 seconds */
104 VNET_DEFINE_STATIC(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
105 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
106 VNET_PCPUSTAT_SYSINIT(arpstat);
107 
108 #ifdef VIMAGE
109 VNET_PCPUSTAT_SYSUNINIT(arpstat);
110 #endif /* VIMAGE */
111 
112 VNET_DEFINE_STATIC(int, arp_maxhold) = 1;
113 
114 #define	V_arpt_keep		VNET(arpt_keep)
115 #define	V_arpt_down		VNET(arpt_down)
116 #define	V_arpt_rexmit		VNET(arpt_rexmit)
117 #define	V_arp_maxtries		VNET(arp_maxtries)
118 #define	V_arp_proxyall		VNET(arp_proxyall)
119 #define	V_arp_maxhold		VNET(arp_maxhold)
120 
121 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
122 	&VNET_NAME(arpt_keep), 0,
123 	"ARP entry lifetime in seconds");
124 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
125 	&VNET_NAME(arp_maxtries), 0,
126 	"ARP resolution attempts before returning error");
127 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
128 	&VNET_NAME(arp_proxyall), 0,
129 	"Enable proxy ARP for all suitable requests");
130 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
131 	&VNET_NAME(arpt_down), 0,
132 	"Incomplete ARP entry lifetime in seconds");
133 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
134     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
135 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
136 	&VNET_NAME(arp_maxhold), 0,
137 	"Number of packets to hold per ARP entry");
138 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
139 	CTLFLAG_RW, &arp_maxpps, 0,
140 	"Maximum number of remotely triggered ARP messages that can be "
141 	"logged per second");
142 
143 /*
144  * Due to the exponential backoff algorithm used for the interval between GARP
145  * retransmissions, the maximum number of retransmissions is limited for
146  * sanity. This limit corresponds to a maximum interval between retransmissions
147  * of 2^16 seconds ~= 18 hours.
148  *
149  * Making this limit more dynamic is more complicated than worthwhile,
150  * especially since sending out GARPs spaced days apart would be of little
151  * use. A maximum dynamic limit would look something like:
152  *
153  * const int max = fls(INT_MAX / hz) - 1;
154  */
155 #define MAX_GARP_RETRANSMITS 16
156 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
157 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
158 
159 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
160     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
161     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
162     "Number of times to retransmit GARP packets;"
163     " 0 to disable, maximum of 16");
164 
165 VNET_DEFINE_STATIC(int, arp_log_level) = LOG_INFO;	/* Min. log(9) level. */
166 #define	V_arp_log_level		VNET(arp_log_level)
167 SYSCTL_INT(_net_link_ether_arp, OID_AUTO, log_level, CTLFLAG_VNET | CTLFLAG_RW,
168 	&VNET_NAME(arp_log_level), 0,
169 	"Minimum log(9) level for recording rate limited arp log messages. "
170 	"The higher will be log more (emerg=0, info=6 (default), debug=7).");
171 #define	ARP_LOG(pri, ...)	do {					\
172 	if ((pri) <= V_arp_log_level &&					\
173 	    ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
174 		log((pri), "arp: " __VA_ARGS__);			\
175 } while (0)
176 
177 
178 static void	arpintr(struct mbuf *);
179 static void	arptimer(void *);
180 #ifdef INET
181 static void	in_arpinput(struct mbuf *);
182 #endif
183 
184 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
185     struct ifnet *ifp, int bridged, struct llentry *la);
186 static void arp_mark_lle_reachable(struct llentry *la);
187 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
188 
189 static eventhandler_tag iflladdr_tag;
190 
191 static const struct netisr_handler arp_nh = {
192 	.nh_name = "arp",
193 	.nh_handler = arpintr,
194 	.nh_proto = NETISR_ARP,
195 	.nh_policy = NETISR_POLICY_SOURCE,
196 };
197 
198 /*
199  * Timeout routine.  Age arp_tab entries periodically.
200  */
201 static void
202 arptimer(void *arg)
203 {
204 	struct llentry *lle = (struct llentry *)arg;
205 	struct ifnet *ifp;
206 	int r_skip_req;
207 
208 	if (lle->la_flags & LLE_STATIC) {
209 		return;
210 	}
211 	LLE_WLOCK(lle);
212 	if (callout_pending(&lle->lle_timer)) {
213 		/*
214 		 * Here we are a bit odd here in the treatment of
215 		 * active/pending. If the pending bit is set, it got
216 		 * rescheduled before I ran. The active
217 		 * bit we ignore, since if it was stopped
218 		 * in ll_tablefree() and was currently running
219 		 * it would have return 0 so the code would
220 		 * not have deleted it since the callout could
221 		 * not be stopped so we want to go through
222 		 * with the delete here now. If the callout
223 		 * was restarted, the pending bit will be back on and
224 		 * we just want to bail since the callout_reset would
225 		 * return 1 and our reference would have been removed
226 		 * by arpresolve() below.
227 		 */
228 		LLE_WUNLOCK(lle);
229  		return;
230  	}
231 	ifp = lle->lle_tbl->llt_ifp;
232 	CURVNET_SET(ifp->if_vnet);
233 
234 	switch (lle->ln_state) {
235 	case ARP_LLINFO_REACHABLE:
236 
237 		/*
238 		 * Expiration time is approaching.
239 		 * Let's try to refresh entry if it is still
240 		 * in use.
241 		 *
242 		 * Set r_skip_req to get feedback from
243 		 * fast path. Change state and re-schedule
244 		 * ourselves.
245 		 */
246 		LLE_REQ_LOCK(lle);
247 		lle->r_skip_req = 1;
248 		LLE_REQ_UNLOCK(lle);
249 		lle->ln_state = ARP_LLINFO_VERIFY;
250 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
251 		LLE_WUNLOCK(lle);
252 		CURVNET_RESTORE();
253 		return;
254 	case ARP_LLINFO_VERIFY:
255 		LLE_REQ_LOCK(lle);
256 		r_skip_req = lle->r_skip_req;
257 		LLE_REQ_UNLOCK(lle);
258 
259 		if (r_skip_req == 0 && lle->la_preempt > 0) {
260 			/* Entry was used, issue refresh request */
261 			struct in_addr dst;
262 			dst = lle->r_l3addr.addr4;
263 			lle->la_preempt--;
264 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
265 			LLE_WUNLOCK(lle);
266 			arprequest(ifp, NULL, &dst, NULL);
267 			CURVNET_RESTORE();
268 			return;
269 		}
270 		/* Nothing happened. Reschedule if not too late */
271 		if (lle->la_expire > time_uptime) {
272 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
273 			LLE_WUNLOCK(lle);
274 			CURVNET_RESTORE();
275 			return;
276 		}
277 		break;
278 	case ARP_LLINFO_INCOMPLETE:
279 	case ARP_LLINFO_DELETED:
280 		break;
281 	}
282 
283 	if ((lle->la_flags & LLE_DELETED) == 0) {
284 		int evt;
285 
286 		if (lle->la_flags & LLE_VALID)
287 			evt = LLENTRY_EXPIRED;
288 		else
289 			evt = LLENTRY_TIMEDOUT;
290 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
291 	}
292 
293 	callout_stop(&lle->lle_timer);
294 
295 	/* XXX: LOR avoidance. We still have ref on lle. */
296 	LLE_WUNLOCK(lle);
297 	IF_AFDATA_LOCK(ifp);
298 	LLE_WLOCK(lle);
299 
300 	/* Guard against race with other llentry_free(). */
301 	if (lle->la_flags & LLE_LINKED) {
302 		LLE_REMREF(lle);
303 		lltable_unlink_entry(lle->lle_tbl, lle);
304 	}
305 	IF_AFDATA_UNLOCK(ifp);
306 
307 	size_t pkts_dropped = llentry_free(lle);
308 
309 	ARPSTAT_ADD(dropped, pkts_dropped);
310 	ARPSTAT_INC(timeouts);
311 
312 	CURVNET_RESTORE();
313 }
314 
315 /*
316  * Stores link-layer header for @ifp in format suitable for if_output()
317  * into buffer @buf. Resulting header length is stored in @bufsize.
318  *
319  * Returns 0 on success.
320  */
321 static int
322 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
323     size_t *bufsize)
324 {
325 	struct if_encap_req ereq;
326 	int error;
327 
328 	bzero(buf, *bufsize);
329 	bzero(&ereq, sizeof(ereq));
330 	ereq.buf = buf;
331 	ereq.bufsize = *bufsize;
332 	ereq.rtype = IFENCAP_LL;
333 	ereq.family = AF_ARP;
334 	ereq.lladdr = ar_tha(ah);
335 	ereq.hdata = (u_char *)ah;
336 	if (bcast)
337 		ereq.flags = IFENCAP_FLAG_BROADCAST;
338 	error = ifp->if_requestencap(ifp, &ereq);
339 	if (error == 0)
340 		*bufsize = ereq.bufsize;
341 
342 	return (error);
343 }
344 
345 
346 /*
347  * Broadcast an ARP request. Caller specifies:
348  *	- arp header source ip address
349  *	- arp header target ip address
350  *	- arp header source ethernet address
351  */
352 static int
353 arprequest_internal(struct ifnet *ifp, const struct in_addr *sip,
354     const struct in_addr *tip, u_char *enaddr)
355 {
356 	struct mbuf *m;
357 	struct arphdr *ah;
358 	struct sockaddr sa;
359 	u_char *carpaddr = NULL;
360 	uint8_t linkhdr[LLE_MAX_LINKHDR];
361 	size_t linkhdrsize;
362 	struct route ro;
363 	int error;
364 
365 	if (sip == NULL) {
366 		/*
367 		 * The caller did not supply a source address, try to find
368 		 * a compatible one among those assigned to this interface.
369 		 */
370 		struct epoch_tracker et;
371 		struct ifaddr *ifa;
372 
373 		NET_EPOCH_ENTER(et);
374 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
375 			if (ifa->ifa_addr->sa_family != AF_INET)
376 				continue;
377 
378 			if (ifa->ifa_carp) {
379 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
380 					continue;
381 				sip = &IA_SIN(ifa)->sin_addr;
382 			} else {
383 				carpaddr = NULL;
384 				sip = &IA_SIN(ifa)->sin_addr;
385 			}
386 
387 			if (0 == ((sip->s_addr ^ tip->s_addr) &
388 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
389 				break;  /* found it. */
390 		}
391 		NET_EPOCH_EXIT(et);
392 		if (sip == NULL) {
393 			printf("%s: cannot find matching address\n", __func__);
394 			return (EADDRNOTAVAIL);
395 		}
396 	}
397 	if (enaddr == NULL)
398 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
399 
400 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
401 		return (ENOMEM);
402 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
403 		2 * ifp->if_addrlen;
404 	m->m_pkthdr.len = m->m_len;
405 	M_ALIGN(m, m->m_len);
406 	ah = mtod(m, struct arphdr *);
407 	bzero((caddr_t)ah, m->m_len);
408 #ifdef MAC
409 	mac_netinet_arp_send(ifp, m);
410 #endif
411 	ah->ar_pro = htons(ETHERTYPE_IP);
412 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
413 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
414 	ah->ar_op = htons(ARPOP_REQUEST);
415 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
416 	bcopy(sip, ar_spa(ah), ah->ar_pln);
417 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
418 	sa.sa_family = AF_ARP;
419 	sa.sa_len = 2;
420 
421 	/* Calculate link header for sending frame */
422 	bzero(&ro, sizeof(ro));
423 	linkhdrsize = sizeof(linkhdr);
424 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
425 	if (error != 0 && error != EAFNOSUPPORT) {
426 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
427 		    if_name(ifp), error);
428 		return (error);
429 	}
430 
431 	ro.ro_prepend = linkhdr;
432 	ro.ro_plen = linkhdrsize;
433 	ro.ro_flags = 0;
434 
435 	m->m_flags |= M_BCAST;
436 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
437 	error = (*ifp->if_output)(ifp, m, &sa, &ro);
438 	ARPSTAT_INC(txrequests);
439 	if (error) {
440 		ARPSTAT_INC(txerrors);
441 		ARP_LOG(LOG_DEBUG, "Failed to send ARP packet on %s: %d\n",
442 		    if_name(ifp), error);
443 	}
444 	return (error);
445 }
446 
447 void
448 arprequest(struct ifnet *ifp, const struct in_addr *sip,
449     const struct in_addr *tip, u_char *enaddr)
450 {
451 
452 	(void) arprequest_internal(ifp, sip, tip, enaddr);
453 }
454 
455 /*
456  * Resolve an IP address into an ethernet address - heavy version.
457  * Used internally by arpresolve().
458  * We have already checked that we can't use an existing lle without
459  * modification so we have to acquire an LLE_EXCLUSIVE lle lock.
460  *
461  * On success, desten and pflags are filled in and the function returns 0;
462  * If the packet must be held pending resolution, we return EWOULDBLOCK
463  * On other errors, we return the corresponding error code.
464  * Note that m_freem() handles NULL.
465  */
466 static int
467 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
468 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
469 	struct llentry **plle)
470 {
471 	struct llentry *la = NULL, *la_tmp;
472 	struct mbuf *curr = NULL;
473 	struct mbuf *next = NULL;
474 	int error, renew;
475 	char *lladdr;
476 	int ll_len;
477 
478 	if (pflags != NULL)
479 		*pflags = 0;
480 	if (plle != NULL)
481 		*plle = NULL;
482 
483 	if ((flags & LLE_CREATE) == 0) {
484 		struct epoch_tracker et;
485 
486 		NET_EPOCH_ENTER(et);
487 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
488 		NET_EPOCH_EXIT(et);
489 	}
490 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
491 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
492 		if (la == NULL) {
493 			char addrbuf[INET_ADDRSTRLEN];
494 
495 			log(LOG_DEBUG,
496 			    "arpresolve: can't allocate llinfo for %s on %s\n",
497 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
498 			    if_name(ifp));
499 			m_freem(m);
500 			return (EINVAL);
501 		}
502 
503 		IF_AFDATA_WLOCK(ifp);
504 		LLE_WLOCK(la);
505 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
506 		/* Prefer ANY existing lle over newly-created one */
507 		if (la_tmp == NULL)
508 			lltable_link_entry(LLTABLE(ifp), la);
509 		IF_AFDATA_WUNLOCK(ifp);
510 		if (la_tmp != NULL) {
511 			lltable_free_entry(LLTABLE(ifp), la);
512 			la = la_tmp;
513 		}
514 	}
515 	if (la == NULL) {
516 		m_freem(m);
517 		return (EINVAL);
518 	}
519 
520 	if ((la->la_flags & LLE_VALID) &&
521 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
522 		if (flags & LLE_ADDRONLY) {
523 			lladdr = la->ll_addr;
524 			ll_len = ifp->if_addrlen;
525 		} else {
526 			lladdr = la->r_linkdata;
527 			ll_len = la->r_hdrlen;
528 		}
529 		bcopy(lladdr, desten, ll_len);
530 
531 		/* Notify LLE code that the entry was used by datapath */
532 		llentry_mark_used(la);
533 		if (pflags != NULL)
534 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
535 		if (plle) {
536 			LLE_ADDREF(la);
537 			*plle = la;
538 		}
539 		LLE_WUNLOCK(la);
540 		return (0);
541 	}
542 
543 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
544 	/*
545 	 * There is an arptab entry, but no ethernet address
546 	 * response yet.  Add the mbuf to the list, dropping
547 	 * the oldest packet if we have exceeded the system
548 	 * setting.
549 	 */
550 	if (m != NULL) {
551 		if (la->la_numheld >= V_arp_maxhold) {
552 			if (la->la_hold != NULL) {
553 				next = la->la_hold->m_nextpkt;
554 				m_freem(la->la_hold);
555 				la->la_hold = next;
556 				la->la_numheld--;
557 				ARPSTAT_INC(dropped);
558 			}
559 		}
560 		if (la->la_hold != NULL) {
561 			curr = la->la_hold;
562 			while (curr->m_nextpkt != NULL)
563 				curr = curr->m_nextpkt;
564 			curr->m_nextpkt = m;
565 		} else
566 			la->la_hold = m;
567 		la->la_numheld++;
568 	}
569 	/*
570 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
571 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
572 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
573 	 * ARP request, but not faster than one request per second.
574 	 */
575 	if (la->la_asked < V_arp_maxtries)
576 		error = EWOULDBLOCK;	/* First request. */
577 	else
578 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
579 
580 	if (renew) {
581 		int canceled, e;
582 
583 		LLE_ADDREF(la);
584 		la->la_expire = time_uptime;
585 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
586 		    arptimer, la);
587 		if (canceled)
588 			LLE_REMREF(la);
589 		la->la_asked++;
590 		LLE_WUNLOCK(la);
591 		e = arprequest_internal(ifp, NULL, &SIN(dst)->sin_addr, NULL);
592 		/*
593 		 * Only overwrite 'error' in case of error; in case of success
594 		 * the proper return value was already set above.
595 		 */
596 		if (e != 0)
597 			return (e);
598 		return (error);
599 	}
600 
601 	LLE_WUNLOCK(la);
602 	return (error);
603 }
604 
605 /*
606  * Lookups link header based on an IP address.
607  * On input:
608  *    ifp is the interface we use
609  *    is_gw != 0 if @dst represents gateway to some destination
610  *    m is the mbuf. May be NULL if we don't have a packet.
611  *    dst is the next hop,
612  *    desten is the storage to put LL header.
613  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
614  *
615  * On success, full/partial link header and flags are filled in and
616  * the function returns 0.
617  * If the packet must be held pending resolution, we return EWOULDBLOCK
618  * On other errors, we return the corresponding error code.
619  * Note that m_freem() handles NULL.
620  */
621 int
622 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
623 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
624 	struct llentry **plle)
625 {
626 	struct epoch_tracker et;
627 	struct llentry *la = NULL;
628 
629 	if (pflags != NULL)
630 		*pflags = 0;
631 	if (plle != NULL)
632 		*plle = NULL;
633 
634 	if (m != NULL) {
635 		if (m->m_flags & M_BCAST) {
636 			/* broadcast */
637 			(void)memcpy(desten,
638 			    ifp->if_broadcastaddr, ifp->if_addrlen);
639 			return (0);
640 		}
641 		if (m->m_flags & M_MCAST) {
642 			/* multicast */
643 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
644 			return (0);
645 		}
646 	}
647 
648 	NET_EPOCH_ENTER(et);
649 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
650 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
651 		/* Entry found, let's copy lle info */
652 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
653 		if (pflags != NULL)
654 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
655 		/* Notify the LLE handling code that the entry was used. */
656 		llentry_mark_used(la);
657 		if (plle) {
658 			LLE_ADDREF(la);
659 			*plle = la;
660 			LLE_WUNLOCK(la);
661 		}
662 		NET_EPOCH_EXIT(et);
663 		return (0);
664 	}
665 	if (plle && la)
666 		LLE_WUNLOCK(la);
667 	NET_EPOCH_EXIT(et);
668 
669 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
670 	    desten, pflags, plle));
671 }
672 
673 /*
674  * Common length and type checks are done here,
675  * then the protocol-specific routine is called.
676  */
677 static void
678 arpintr(struct mbuf *m)
679 {
680 	struct arphdr *ar;
681 	struct ifnet *ifp;
682 	char *layer;
683 	int hlen;
684 
685 	ifp = m->m_pkthdr.rcvif;
686 
687 	if (m->m_len < sizeof(struct arphdr) &&
688 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
689 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
690 		    if_name(ifp));
691 		return;
692 	}
693 	ar = mtod(m, struct arphdr *);
694 
695 	/* Check if length is sufficient */
696 	if (m->m_len <  arphdr_len(ar)) {
697 		m = m_pullup(m, arphdr_len(ar));
698 		if (m == NULL) {
699 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
700 			    if_name(ifp));
701 			return;
702 		}
703 		ar = mtod(m, struct arphdr *);
704 	}
705 
706 	hlen = 0;
707 	layer = "";
708 	switch (ntohs(ar->ar_hrd)) {
709 	case ARPHRD_ETHER:
710 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
711 		layer = "ethernet";
712 		break;
713 	case ARPHRD_INFINIBAND:
714 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
715 		layer = "infiniband";
716 		break;
717 	case ARPHRD_IEEE1394:
718 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
719 		layer = "firewire";
720 
721 		/*
722 		 * Restrict too long hardware addresses.
723 		 * Currently we are capable of handling 20-byte
724 		 * addresses ( sizeof(lle->ll_addr) )
725 		 */
726 		if (ar->ar_hln >= 20)
727 			hlen = 16;
728 		break;
729 	default:
730 		ARP_LOG(LOG_NOTICE,
731 		    "packet with unknown hardware format 0x%02d received on "
732 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
733 		m_freem(m);
734 		return;
735 	}
736 
737 	if (hlen != 0 && hlen != ar->ar_hln) {
738 		ARP_LOG(LOG_NOTICE,
739 		    "packet with invalid %s address length %d received on %s\n",
740 		    layer, ar->ar_hln, if_name(ifp));
741 		m_freem(m);
742 		return;
743 	}
744 
745 	ARPSTAT_INC(received);
746 	switch (ntohs(ar->ar_pro)) {
747 #ifdef INET
748 	case ETHERTYPE_IP:
749 		in_arpinput(m);
750 		return;
751 #endif
752 	}
753 	m_freem(m);
754 }
755 
756 #ifdef INET
757 /*
758  * ARP for Internet protocols on 10 Mb/s Ethernet.
759  * Algorithm is that given in RFC 826.
760  * In addition, a sanity check is performed on the sender
761  * protocol address, to catch impersonators.
762  * We no longer handle negotiations for use of trailer protocol:
763  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
764  * along with IP replies if we wanted trailers sent to us,
765  * and also sent them in response to IP replies.
766  * This allowed either end to announce the desire to receive
767  * trailer packets.
768  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
769  * but formerly didn't normally send requests.
770  */
771 static int log_arp_wrong_iface = 1;
772 static int log_arp_movements = 1;
773 static int log_arp_permanent_modify = 1;
774 static int allow_multicast = 0;
775 
776 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
777 	&log_arp_wrong_iface, 0,
778 	"log arp packets arriving on the wrong interface");
779 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
780 	&log_arp_movements, 0,
781 	"log arp replies from MACs different than the one in the cache");
782 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
783 	&log_arp_permanent_modify, 0,
784 	"log arp replies from MACs different than the one in the permanent arp entry");
785 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
786 	&allow_multicast, 0, "accept multicast addresses");
787 
788 static void
789 in_arpinput(struct mbuf *m)
790 {
791 	struct rm_priotracker in_ifa_tracker;
792 	struct arphdr *ah;
793 	struct ifnet *ifp = m->m_pkthdr.rcvif;
794 	struct llentry *la = NULL, *la_tmp;
795 	struct ifaddr *ifa;
796 	struct in_ifaddr *ia;
797 	struct sockaddr sa;
798 	struct in_addr isaddr, itaddr, myaddr;
799 	u_int8_t *enaddr = NULL;
800 	int op;
801 	int bridged = 0, is_bridge = 0;
802 	int carped;
803 	struct sockaddr_in sin;
804 	struct sockaddr *dst;
805 	struct nhop4_basic nh4;
806 	uint8_t linkhdr[LLE_MAX_LINKHDR];
807 	struct route ro;
808 	size_t linkhdrsize;
809 	int lladdr_off;
810 	int error;
811 	char addrbuf[INET_ADDRSTRLEN];
812 	struct epoch_tracker et;
813 
814 	sin.sin_len = sizeof(struct sockaddr_in);
815 	sin.sin_family = AF_INET;
816 	sin.sin_addr.s_addr = 0;
817 
818 	if (ifp->if_bridge)
819 		bridged = 1;
820 	if (ifp->if_type == IFT_BRIDGE)
821 		is_bridge = 1;
822 
823 	/*
824 	 * We already have checked that mbuf contains enough contiguous data
825 	 * to hold entire arp message according to the arp header.
826 	 */
827 	ah = mtod(m, struct arphdr *);
828 
829 	/*
830 	 * ARP is only for IPv4 so we can reject packets with
831 	 * a protocol length not equal to an IPv4 address.
832 	 */
833 	if (ah->ar_pln != sizeof(struct in_addr)) {
834 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
835 		    sizeof(struct in_addr));
836 		goto drop;
837 	}
838 
839 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
840 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
841 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
842 		goto drop;
843 	}
844 
845 	op = ntohs(ah->ar_op);
846 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
847 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
848 
849 	if (op == ARPOP_REPLY)
850 		ARPSTAT_INC(rxreplies);
851 
852 	/*
853 	 * For a bridge, we want to check the address irrespective
854 	 * of the receive interface. (This will change slightly
855 	 * when we have clusters of interfaces).
856 	 */
857 	IN_IFADDR_RLOCK(&in_ifa_tracker);
858 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
859 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
860 		    ia->ia_ifp == ifp) &&
861 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
862 		    (ia->ia_ifa.ifa_carp == NULL ||
863 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
864 			ifa_ref(&ia->ia_ifa);
865 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
866 			goto match;
867 		}
868 	}
869 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
870 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
871 		    ia->ia_ifp == ifp) &&
872 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
873 			ifa_ref(&ia->ia_ifa);
874 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
875 			goto match;
876 		}
877 
878 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
879   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
880   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
881   addr == ia->ia_addr.sin_addr.s_addr)
882 	/*
883 	 * Check the case when bridge shares its MAC address with
884 	 * some of its children, so packets are claimed by bridge
885 	 * itself (bridge_input() does it first), but they are really
886 	 * meant to be destined to the bridge member.
887 	 */
888 	if (is_bridge) {
889 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
890 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
891 				ifa_ref(&ia->ia_ifa);
892 				ifp = ia->ia_ifp;
893 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
894 				goto match;
895 			}
896 		}
897 	}
898 #undef BDG_MEMBER_MATCHES_ARP
899 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
900 
901 	/*
902 	 * No match, use the first inet address on the receive interface
903 	 * as a dummy address for the rest of the function.
904 	 */
905 	NET_EPOCH_ENTER(et);
906 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
907 		if (ifa->ifa_addr->sa_family == AF_INET &&
908 		    (ifa->ifa_carp == NULL ||
909 		    (*carp_iamatch_p)(ifa, &enaddr))) {
910 			ia = ifatoia(ifa);
911 			ifa_ref(ifa);
912 			NET_EPOCH_EXIT(et);
913 			goto match;
914 		}
915 	NET_EPOCH_EXIT(et);
916 
917 	/*
918 	 * If bridging, fall back to using any inet address.
919 	 */
920 	IN_IFADDR_RLOCK(&in_ifa_tracker);
921 	if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
922 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
923 		goto drop;
924 	}
925 	ifa_ref(&ia->ia_ifa);
926 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
927 match:
928 	if (!enaddr)
929 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
930 	carped = (ia->ia_ifa.ifa_carp != NULL);
931 	myaddr = ia->ia_addr.sin_addr;
932 	ifa_free(&ia->ia_ifa);
933 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
934 		goto drop;	/* it's from me, ignore it. */
935 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
936 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
937 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
938 		goto drop;
939 	}
940 
941 	if (ifp->if_addrlen != ah->ar_hln) {
942 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
943 		    "i/f %d (ignored)\n", ifp->if_addrlen,
944 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
945 		    ifp->if_addrlen);
946 		goto drop;
947 	}
948 
949 	/*
950 	 * Warn if another host is using the same IP address, but only if the
951 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
952 	 * case we suppress the warning to avoid false positive complaints of
953 	 * potential misconfiguration.
954 	 */
955 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
956 	    myaddr.s_addr != 0) {
957 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
958 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
959 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
960 		itaddr = myaddr;
961 		ARPSTAT_INC(dupips);
962 		goto reply;
963 	}
964 	if (ifp->if_flags & IFF_STATICARP)
965 		goto reply;
966 
967 	bzero(&sin, sizeof(sin));
968 	sin.sin_len = sizeof(struct sockaddr_in);
969 	sin.sin_family = AF_INET;
970 	sin.sin_addr = isaddr;
971 	dst = (struct sockaddr *)&sin;
972 	NET_EPOCH_ENTER(et);
973 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
974 	NET_EPOCH_EXIT(et);
975 	if (la != NULL)
976 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
977 	else if (itaddr.s_addr == myaddr.s_addr) {
978 		/*
979 		 * Request/reply to our address, but no lle exists yet.
980 		 * Calculate full link prepend to use in lle.
981 		 */
982 		linkhdrsize = sizeof(linkhdr);
983 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
984 		    &linkhdrsize, &lladdr_off) != 0)
985 			goto reply;
986 
987 		/* Allocate new entry */
988 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
989 		if (la == NULL) {
990 
991 			/*
992 			 * lle creation may fail if source address belongs
993 			 * to non-directly connected subnet. However, we
994 			 * will try to answer the request instead of dropping
995 			 * frame.
996 			 */
997 			goto reply;
998 		}
999 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1000 		    lladdr_off);
1001 
1002 		IF_AFDATA_WLOCK(ifp);
1003 		LLE_WLOCK(la);
1004 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1005 
1006 		/*
1007 		 * Check if lle still does not exists.
1008 		 * If it does, that means that we either
1009 		 * 1) have configured it explicitly, via
1010 		 * 1a) 'arp -s' static entry or
1011 		 * 1b) interface address static record
1012 		 * or
1013 		 * 2) it was the result of sending first packet to-host
1014 		 * or
1015 		 * 3) it was another arp reply packet we handled in
1016 		 * different thread.
1017 		 *
1018 		 * In all cases except 3) we definitely need to prefer
1019 		 * existing lle. For the sake of simplicity, prefer any
1020 		 * existing lle over newly-create one.
1021 		 */
1022 		if (la_tmp == NULL)
1023 			lltable_link_entry(LLTABLE(ifp), la);
1024 		IF_AFDATA_WUNLOCK(ifp);
1025 
1026 		if (la_tmp == NULL) {
1027 			arp_mark_lle_reachable(la);
1028 			LLE_WUNLOCK(la);
1029 		} else {
1030 			/* Free newly-create entry and handle packet */
1031 			lltable_free_entry(LLTABLE(ifp), la);
1032 			la = la_tmp;
1033 			la_tmp = NULL;
1034 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1035 			/* arp_check_update_lle() returns @la unlocked */
1036 		}
1037 		la = NULL;
1038 	}
1039 reply:
1040 	if (op != ARPOP_REQUEST)
1041 		goto drop;
1042 	ARPSTAT_INC(rxrequests);
1043 
1044 	if (itaddr.s_addr == myaddr.s_addr) {
1045 		/* Shortcut.. the receiving interface is the target. */
1046 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1047 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1048 	} else {
1049 		struct llentry *lle = NULL;
1050 
1051 		sin.sin_addr = itaddr;
1052 		NET_EPOCH_ENTER(et);
1053 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1054 		NET_EPOCH_EXIT(et);
1055 
1056 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1057 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1058 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1059 			LLE_RUNLOCK(lle);
1060 		} else {
1061 
1062 			if (lle != NULL)
1063 				LLE_RUNLOCK(lle);
1064 
1065 			if (!V_arp_proxyall)
1066 				goto drop;
1067 
1068 			/* XXX MRT use table 0 for arp reply  */
1069 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1070 				goto drop;
1071 
1072 			/*
1073 			 * Don't send proxies for nodes on the same interface
1074 			 * as this one came out of, or we'll get into a fight
1075 			 * over who claims what Ether address.
1076 			 */
1077 			if (nh4.nh_ifp == ifp)
1078 				goto drop;
1079 
1080 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1081 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1082 
1083 			/*
1084 			 * Also check that the node which sent the ARP packet
1085 			 * is on the interface we expect it to be on. This
1086 			 * avoids ARP chaos if an interface is connected to the
1087 			 * wrong network.
1088 			 */
1089 
1090 			/* XXX MRT use table 0 for arp checks */
1091 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1092 				goto drop;
1093 			if (nh4.nh_ifp != ifp) {
1094 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1095 				    " from %s via %s\n",
1096 				    inet_ntoa_r(isaddr, addrbuf),
1097 				    ifp->if_xname);
1098 				goto drop;
1099 			}
1100 
1101 #ifdef DEBUG_PROXY
1102 			printf("arp: proxying for %s\n",
1103 			    inet_ntoa_r(itaddr, addrbuf));
1104 #endif
1105 		}
1106 	}
1107 
1108 	if (itaddr.s_addr == myaddr.s_addr &&
1109 	    IN_LINKLOCAL(ntohl(itaddr.s_addr))) {
1110 		/* RFC 3927 link-local IPv4; always reply by broadcast. */
1111 #ifdef DEBUG_LINKLOCAL
1112 		printf("arp: sending reply for link-local addr %s\n",
1113 		    inet_ntoa_r(itaddr, addrbuf));
1114 #endif
1115 		m->m_flags |= M_BCAST;
1116 		m->m_flags &= ~M_MCAST;
1117 	} else {
1118 		/* default behaviour; never reply by broadcast. */
1119 		m->m_flags &= ~(M_BCAST|M_MCAST);
1120 	}
1121 	(void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1122 	(void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1123 	ah->ar_op = htons(ARPOP_REPLY);
1124 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1125 	m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1126 	m->m_pkthdr.len = m->m_len;
1127 	m->m_pkthdr.rcvif = NULL;
1128 	sa.sa_family = AF_ARP;
1129 	sa.sa_len = 2;
1130 
1131 	/* Calculate link header for sending frame */
1132 	bzero(&ro, sizeof(ro));
1133 	linkhdrsize = sizeof(linkhdr);
1134 	error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize);
1135 
1136 	/*
1137 	 * arp_fillheader() may fail due to lack of support inside encap request
1138 	 * routing. This is not necessary an error, AF_ARP can/should be handled
1139 	 * by if_output().
1140 	 */
1141 	if (error != 0 && error != EAFNOSUPPORT) {
1142 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
1143 		    if_name(ifp), error);
1144 		return;
1145 	}
1146 
1147 	ro.ro_prepend = linkhdr;
1148 	ro.ro_plen = linkhdrsize;
1149 	ro.ro_flags = 0;
1150 
1151 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
1152 	(*ifp->if_output)(ifp, m, &sa, &ro);
1153 	ARPSTAT_INC(txreplies);
1154 	return;
1155 
1156 drop:
1157 	m_freem(m);
1158 }
1159 #endif
1160 
1161 /*
1162  * Checks received arp data against existing @la.
1163  * Updates lle state/performs notification if necessary.
1164  */
1165 static void
1166 arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp,
1167     int bridged, struct llentry *la)
1168 {
1169 	struct sockaddr sa;
1170 	struct mbuf *m_hold, *m_hold_next;
1171 	uint8_t linkhdr[LLE_MAX_LINKHDR];
1172 	size_t linkhdrsize;
1173 	int lladdr_off;
1174 	char addrbuf[INET_ADDRSTRLEN];
1175 
1176 	LLE_WLOCK_ASSERT(la);
1177 
1178 	/* the following is not an error when doing bridging */
1179 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1180 		if (log_arp_wrong_iface)
1181 			ARP_LOG(LOG_WARNING, "%s is on %s "
1182 			    "but got reply from %*D on %s\n",
1183 			    inet_ntoa_r(isaddr, addrbuf),
1184 			    la->lle_tbl->llt_ifp->if_xname,
1185 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1186 			    ifp->if_xname);
1187 		LLE_WUNLOCK(la);
1188 		return;
1189 	}
1190 	if ((la->la_flags & LLE_VALID) &&
1191 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1192 		if (la->la_flags & LLE_STATIC) {
1193 			LLE_WUNLOCK(la);
1194 			if (log_arp_permanent_modify)
1195 				ARP_LOG(LOG_ERR,
1196 				    "%*D attempts to modify "
1197 				    "permanent entry for %s on %s\n",
1198 				    ifp->if_addrlen,
1199 				    (u_char *)ar_sha(ah), ":",
1200 				    inet_ntoa_r(isaddr, addrbuf),
1201 				    ifp->if_xname);
1202 			return;
1203 		}
1204 		if (log_arp_movements) {
1205 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1206 			    "to %*D on %s\n",
1207 			    inet_ntoa_r(isaddr, addrbuf),
1208 			    ifp->if_addrlen,
1209 			    (u_char *)la->ll_addr, ":",
1210 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1211 			    ifp->if_xname);
1212 		}
1213 	}
1214 
1215 	/* Calculate full link prepend to use in lle */
1216 	linkhdrsize = sizeof(linkhdr);
1217 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1218 	    &linkhdrsize, &lladdr_off) != 0)
1219 		return;
1220 
1221 	/* Check if something has changed */
1222 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1223 	    (la->la_flags & LLE_VALID) == 0) {
1224 		/* Try to perform LLE update */
1225 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1226 		    lladdr_off) == 0)
1227 			return;
1228 
1229 		/* Clear fast path feedback request if set */
1230 		la->r_skip_req = 0;
1231 	}
1232 
1233 	arp_mark_lle_reachable(la);
1234 
1235 	/*
1236 	 * The packets are all freed within the call to the output
1237 	 * routine.
1238 	 *
1239 	 * NB: The lock MUST be released before the call to the
1240 	 * output routine.
1241 	 */
1242 	if (la->la_hold != NULL) {
1243 		m_hold = la->la_hold;
1244 		la->la_hold = NULL;
1245 		la->la_numheld = 0;
1246 		lltable_fill_sa_entry(la, &sa);
1247 		LLE_WUNLOCK(la);
1248 		for (; m_hold != NULL; m_hold = m_hold_next) {
1249 			m_hold_next = m_hold->m_nextpkt;
1250 			m_hold->m_nextpkt = NULL;
1251 			/* Avoid confusing lower layers. */
1252 			m_clrprotoflags(m_hold);
1253 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1254 		}
1255 	} else
1256 		LLE_WUNLOCK(la);
1257 }
1258 
1259 static void
1260 arp_mark_lle_reachable(struct llentry *la)
1261 {
1262 	int canceled, wtime;
1263 
1264 	LLE_WLOCK_ASSERT(la);
1265 
1266 	la->ln_state = ARP_LLINFO_REACHABLE;
1267 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1268 
1269 	if (!(la->la_flags & LLE_STATIC)) {
1270 		LLE_ADDREF(la);
1271 		la->la_expire = time_uptime + V_arpt_keep;
1272 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1273 		if (wtime < 0)
1274 			wtime = V_arpt_keep;
1275 		canceled = callout_reset(&la->lle_timer,
1276 		    hz * wtime, arptimer, la);
1277 		if (canceled)
1278 			LLE_REMREF(la);
1279 	}
1280 	la->la_asked = 0;
1281 	la->la_preempt = V_arp_maxtries;
1282 }
1283 
1284 /*
1285  * Add permanent link-layer record for given interface address.
1286  */
1287 static __noinline void
1288 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1289 {
1290 	struct llentry *lle, *lle_tmp;
1291 
1292 	/*
1293 	 * Interface address LLE record is considered static
1294 	 * because kernel code relies on LLE_STATIC flag to check
1295 	 * if these entries can be rewriten by arp updates.
1296 	 */
1297 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1298 	if (lle == NULL) {
1299 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1300 		    "entry for interface address\n");
1301 		return;
1302 	}
1303 
1304 	IF_AFDATA_WLOCK(ifp);
1305 	LLE_WLOCK(lle);
1306 	/* Unlink any entry if exists */
1307 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1308 	if (lle_tmp != NULL)
1309 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1310 
1311 	lltable_link_entry(LLTABLE(ifp), lle);
1312 	IF_AFDATA_WUNLOCK(ifp);
1313 
1314 	if (lle_tmp != NULL)
1315 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1316 
1317 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1318 	LLE_WUNLOCK(lle);
1319 	if (lle_tmp != NULL)
1320 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1321 }
1322 
1323 /*
1324  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1325  * of valid values.
1326  */
1327 static int
1328 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1329 {
1330 	int error;
1331 	int rexmit_count = *(int *)arg1;
1332 
1333 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1334 
1335 	/* Enforce limits on any new value that may have been set. */
1336 	if (!error && req->newptr) {
1337 		/* A new value was set. */
1338 		if (rexmit_count < 0) {
1339 			rexmit_count = 0;
1340 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1341 			rexmit_count = MAX_GARP_RETRANSMITS;
1342 		}
1343 		*(int *)arg1 = rexmit_count;
1344 	}
1345 
1346 	return (error);
1347 }
1348 
1349 /*
1350  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1351  * retransmit it again. A pending callout owns a reference to the ifa.
1352  */
1353 static void
1354 garp_rexmit(void *arg)
1355 {
1356 	struct in_ifaddr *ia = arg;
1357 
1358 	if (callout_pending(&ia->ia_garp_timer) ||
1359 	    !callout_active(&ia->ia_garp_timer)) {
1360 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1361 		ifa_free(&ia->ia_ifa);
1362 		return;
1363 	}
1364 
1365 	CURVNET_SET(ia->ia_ifa.ifa_ifp->if_vnet);
1366 
1367 	/*
1368 	 * Drop lock while the ARP request is generated.
1369 	 */
1370 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1371 
1372 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1373 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1374 
1375 	/*
1376 	 * Increment the count of retransmissions. If the count has reached the
1377 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1378 	 * the callout to retransmit another GARP packet.
1379 	 */
1380 	++ia->ia_garp_count;
1381 	if (ia->ia_garp_count >= garp_rexmit_count) {
1382 		ifa_free(&ia->ia_ifa);
1383 	} else {
1384 		int rescheduled;
1385 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1386 		rescheduled = callout_reset(&ia->ia_garp_timer,
1387 		    (1 << ia->ia_garp_count) * hz,
1388 		    garp_rexmit, ia);
1389 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1390 		if (rescheduled) {
1391 			ifa_free(&ia->ia_ifa);
1392 		}
1393 	}
1394 
1395 	CURVNET_RESTORE();
1396 }
1397 
1398 /*
1399  * Start the GARP retransmit timer.
1400  *
1401  * A single GARP is always transmitted when an IPv4 address is added
1402  * to an interface and that is usually sufficient. However, in some
1403  * circumstances, such as when a shared address is passed between
1404  * cluster nodes, this single GARP may occasionally be dropped or
1405  * lost. This can lead to neighbors on the network link working with a
1406  * stale ARP cache and sending packets destined for that address to
1407  * the node that previously owned the address, which may not respond.
1408  *
1409  * To avoid this situation, GARP retransmits can be enabled by setting
1410  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1411  * than zero. The setting represents the maximum number of
1412  * retransmissions. The interval between retransmissions is calculated
1413  * using an exponential backoff algorithm, doubling each time, so the
1414  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1415  */
1416 static void
1417 garp_timer_start(struct ifaddr *ifa)
1418 {
1419 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1420 
1421 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1422 	ia->ia_garp_count = 0;
1423 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1424 	    garp_rexmit, ia) == 0) {
1425 		ifa_ref(ifa);
1426 	}
1427 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1428 }
1429 
1430 void
1431 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1432 {
1433 	const struct sockaddr_in *dst_in;
1434 	const struct sockaddr *dst;
1435 
1436 	if (ifa->ifa_carp != NULL)
1437 		return;
1438 
1439 	dst = ifa->ifa_addr;
1440 	dst_in = (const struct sockaddr_in *)dst;
1441 
1442 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1443 		return;
1444 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1445 	if (garp_rexmit_count > 0) {
1446 		garp_timer_start(ifa);
1447 	}
1448 
1449 	arp_add_ifa_lle(ifp, dst);
1450 }
1451 
1452 void
1453 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1454 {
1455 
1456 	if (ntohl(addr.s_addr) != INADDR_ANY)
1457 		arprequest(ifp, &addr, &addr, enaddr);
1458 }
1459 
1460 /*
1461  * Sends gratuitous ARPs for each ifaddr to notify other
1462  * nodes about the address change.
1463  */
1464 static __noinline void
1465 arp_handle_ifllchange(struct ifnet *ifp)
1466 {
1467 	struct ifaddr *ifa;
1468 
1469 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1470 		if (ifa->ifa_addr->sa_family == AF_INET)
1471 			arp_ifinit(ifp, ifa);
1472 	}
1473 }
1474 
1475 /*
1476  * A handler for interface link layer address change event.
1477  */
1478 static void
1479 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1480 {
1481 
1482 	lltable_update_ifaddr(LLTABLE(ifp));
1483 
1484 	if ((ifp->if_flags & IFF_UP) != 0)
1485 		arp_handle_ifllchange(ifp);
1486 }
1487 
1488 static void
1489 vnet_arp_init(void)
1490 {
1491 
1492 	if (IS_DEFAULT_VNET(curvnet)) {
1493 		netisr_register(&arp_nh);
1494 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1495 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1496 	}
1497 #ifdef VIMAGE
1498 	else
1499 		netisr_register_vnet(&arp_nh);
1500 #endif
1501 }
1502 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1503     vnet_arp_init, 0);
1504 
1505 #ifdef VIMAGE
1506 /*
1507  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1508  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1509  */
1510 static void
1511 vnet_arp_destroy(__unused void *arg)
1512 {
1513 
1514 	netisr_unregister_vnet(&arp_nh);
1515 }
1516 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1517     vnet_arp_destroy, NULL);
1518 #endif
1519