xref: /freebsd/sys/netinet/if_ether.c (revision f5147e312f43a9050468de539aeafa072caa1a60)
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/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 #include <sys/mbuf.h>
52 #include <sys/malloc.h>
53 #include <sys/proc.h>
54 #include <sys/rmlock.h>
55 #include <sys/socket.h>
56 #include <sys/syslog.h>
57 
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/netisr.h>
63 #include <net/ethernet.h>
64 #include <net/route.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_fib.h>
69 #include <netinet/in_var.h>
70 #include <net/if_llatbl.h>
71 #include <netinet/if_ether.h>
72 #ifdef INET
73 #include <netinet/ip_carp.h>
74 #endif
75 
76 #include <security/mac/mac_framework.h>
77 
78 #define SIN(s) ((const struct sockaddr_in *)(s))
79 
80 static struct timeval arp_lastlog;
81 static int arp_curpps;
82 static int arp_maxpps = 1;
83 
84 /* Simple ARP state machine */
85 enum arp_llinfo_state {
86 	ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */
87 	ARP_LLINFO_REACHABLE,	/* LLE is valid */
88 	ARP_LLINFO_VERIFY,	/* LLE is valid, need refresh */
89 	ARP_LLINFO_DELETED,	/* LLE is deleted */
90 };
91 
92 SYSCTL_DECL(_net_link_ether);
93 static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, "");
94 static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, "");
95 
96 /* timer values */
97 static VNET_DEFINE(int, arpt_keep) = (20*60);	/* once resolved, good for 20
98 						 * minutes */
99 static VNET_DEFINE(int, arp_maxtries) = 5;
100 static VNET_DEFINE(int, arp_proxyall) = 0;
101 static VNET_DEFINE(int, arpt_down) = 20;	/* keep incomplete entries for
102 						 * 20 seconds */
103 static VNET_DEFINE(int, arpt_rexmit) = 1;	/* retransmit arp entries, sec*/
104 VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
105 VNET_PCPUSTAT_SYSINIT(arpstat);
106 
107 #ifdef VIMAGE
108 VNET_PCPUSTAT_SYSUNINIT(arpstat);
109 #endif /* VIMAGE */
110 
111 static VNET_DEFINE(int, arp_maxhold) = 1;
112 
113 #define	V_arpt_keep		VNET(arpt_keep)
114 #define	V_arpt_down		VNET(arpt_down)
115 #define	V_arpt_rexmit		VNET(arpt_rexmit)
116 #define	V_arp_maxtries		VNET(arp_maxtries)
117 #define	V_arp_proxyall		VNET(arp_proxyall)
118 #define	V_arp_maxhold		VNET(arp_maxhold)
119 
120 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW,
121 	&VNET_NAME(arpt_keep), 0,
122 	"ARP entry lifetime in seconds");
123 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW,
124 	&VNET_NAME(arp_maxtries), 0,
125 	"ARP resolution attempts before returning error");
126 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW,
127 	&VNET_NAME(arp_proxyall), 0,
128 	"Enable proxy ARP for all suitable requests");
129 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW,
130 	&VNET_NAME(arpt_down), 0,
131 	"Incomplete ARP entry lifetime in seconds");
132 SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat,
133     arpstat, "ARP statistics (struct arpstat, net/if_arp.h)");
134 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW,
135 	&VNET_NAME(arp_maxhold), 0,
136 	"Number of packets to hold per ARP entry");
137 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second,
138 	CTLFLAG_RW, &arp_maxpps, 0,
139 	"Maximum number of remotely triggered ARP messages that can be "
140 	"logged per second");
141 
142 /*
143  * Due to the exponential backoff algorithm used for the interval between GARP
144  * retransmissions, the maximum number of retransmissions is limited for
145  * sanity. This limit corresponds to a maximum interval between retransmissions
146  * of 2^16 seconds ~= 18 hours.
147  *
148  * Making this limit more dynamic is more complicated than worthwhile,
149  * especially since sending out GARPs spaced days apart would be of little
150  * use. A maximum dynamic limit would look something like:
151  *
152  * const int max = fls(INT_MAX / hz) - 1;
153  */
154 #define MAX_GARP_RETRANSMITS 16
155 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS);
156 static int garp_rexmit_count = 0; /* GARP retransmission setting. */
157 
158 SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count,
159     CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE,
160     &garp_rexmit_count, 0, sysctl_garp_rexmit, "I",
161     "Number of times to retransmit GARP packets;"
162     " 0 to disable, maximum of 16");
163 
164 #define	ARP_LOG(pri, ...)	do {					\
165 	if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps))	\
166 		log((pri), "arp: " __VA_ARGS__);			\
167 } while (0)
168 
169 
170 static void	arpintr(struct mbuf *);
171 static void	arptimer(void *);
172 #ifdef INET
173 static void	in_arpinput(struct mbuf *);
174 #endif
175 
176 static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr,
177     struct ifnet *ifp, int bridged, struct llentry *la);
178 static void arp_mark_lle_reachable(struct llentry *la);
179 static void arp_iflladdr(void *arg __unused, struct ifnet *ifp);
180 
181 static eventhandler_tag iflladdr_tag;
182 
183 static const struct netisr_handler arp_nh = {
184 	.nh_name = "arp",
185 	.nh_handler = arpintr,
186 	.nh_proto = NETISR_ARP,
187 	.nh_policy = NETISR_POLICY_SOURCE,
188 };
189 
190 /*
191  * Timeout routine.  Age arp_tab entries periodically.
192  */
193 static void
194 arptimer(void *arg)
195 {
196 	struct llentry *lle = (struct llentry *)arg;
197 	struct ifnet *ifp;
198 	int r_skip_req;
199 
200 	if (lle->la_flags & LLE_STATIC) {
201 		return;
202 	}
203 	LLE_WLOCK(lle);
204 	if (callout_pending(&lle->lle_timer)) {
205 		/*
206 		 * Here we are a bit odd here in the treatment of
207 		 * active/pending. If the pending bit is set, it got
208 		 * rescheduled before I ran. The active
209 		 * bit we ignore, since if it was stopped
210 		 * in ll_tablefree() and was currently running
211 		 * it would have return 0 so the code would
212 		 * not have deleted it since the callout could
213 		 * not be stopped so we want to go through
214 		 * with the delete here now. If the callout
215 		 * was restarted, the pending bit will be back on and
216 		 * we just want to bail since the callout_reset would
217 		 * return 1 and our reference would have been removed
218 		 * by arpresolve() below.
219 		 */
220 		LLE_WUNLOCK(lle);
221  		return;
222  	}
223 	ifp = lle->lle_tbl->llt_ifp;
224 	CURVNET_SET(ifp->if_vnet);
225 
226 	switch (lle->ln_state) {
227 	case ARP_LLINFO_REACHABLE:
228 
229 		/*
230 		 * Expiration time is approaching.
231 		 * Let's try to refresh entry if it is still
232 		 * in use.
233 		 *
234 		 * Set r_skip_req to get feedback from
235 		 * fast path. Change state and re-schedule
236 		 * ourselves.
237 		 */
238 		LLE_REQ_LOCK(lle);
239 		lle->r_skip_req = 1;
240 		LLE_REQ_UNLOCK(lle);
241 		lle->ln_state = ARP_LLINFO_VERIFY;
242 		callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
243 		LLE_WUNLOCK(lle);
244 		CURVNET_RESTORE();
245 		return;
246 	case ARP_LLINFO_VERIFY:
247 		LLE_REQ_LOCK(lle);
248 		r_skip_req = lle->r_skip_req;
249 		LLE_REQ_UNLOCK(lle);
250 
251 		if (r_skip_req == 0 && lle->la_preempt > 0) {
252 			/* Entry was used, issue refresh request */
253 			struct in_addr dst;
254 			dst = lle->r_l3addr.addr4;
255 			lle->la_preempt--;
256 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
257 			LLE_WUNLOCK(lle);
258 			arprequest(ifp, NULL, &dst, NULL);
259 			CURVNET_RESTORE();
260 			return;
261 		}
262 		/* Nothing happened. Reschedule if not too late */
263 		if (lle->la_expire > time_uptime) {
264 			callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit);
265 			LLE_WUNLOCK(lle);
266 			CURVNET_RESTORE();
267 			return;
268 		}
269 		break;
270 	case ARP_LLINFO_INCOMPLETE:
271 	case ARP_LLINFO_DELETED:
272 		break;
273 	}
274 
275 	if ((lle->la_flags & LLE_DELETED) == 0) {
276 		int evt;
277 
278 		if (lle->la_flags & LLE_VALID)
279 			evt = LLENTRY_EXPIRED;
280 		else
281 			evt = LLENTRY_TIMEDOUT;
282 		EVENTHANDLER_INVOKE(lle_event, lle, evt);
283 	}
284 
285 	callout_stop(&lle->lle_timer);
286 
287 	/* XXX: LOR avoidance. We still have ref on lle. */
288 	LLE_WUNLOCK(lle);
289 	IF_AFDATA_LOCK(ifp);
290 	LLE_WLOCK(lle);
291 
292 	/* Guard against race with other llentry_free(). */
293 	if (lle->la_flags & LLE_LINKED) {
294 		LLE_REMREF(lle);
295 		lltable_unlink_entry(lle->lle_tbl, lle);
296 	}
297 	IF_AFDATA_UNLOCK(ifp);
298 
299 	size_t pkts_dropped = llentry_free(lle);
300 
301 	ARPSTAT_ADD(dropped, pkts_dropped);
302 	ARPSTAT_INC(timeouts);
303 
304 	CURVNET_RESTORE();
305 }
306 
307 /*
308  * Stores link-layer header for @ifp in format suitable for if_output()
309  * into buffer @buf. Resulting header length is stored in @bufsize.
310  *
311  * Returns 0 on success.
312  */
313 static int
314 arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf,
315     size_t *bufsize)
316 {
317 	struct if_encap_req ereq;
318 	int error;
319 
320 	bzero(buf, *bufsize);
321 	bzero(&ereq, sizeof(ereq));
322 	ereq.buf = buf;
323 	ereq.bufsize = *bufsize;
324 	ereq.rtype = IFENCAP_LL;
325 	ereq.family = AF_ARP;
326 	ereq.lladdr = ar_tha(ah);
327 	ereq.hdata = (u_char *)ah;
328 	if (bcast)
329 		ereq.flags = IFENCAP_FLAG_BROADCAST;
330 	error = ifp->if_requestencap(ifp, &ereq);
331 	if (error == 0)
332 		*bufsize = ereq.bufsize;
333 
334 	return (error);
335 }
336 
337 
338 /*
339  * Broadcast an ARP request. Caller specifies:
340  *	- arp header source ip address
341  *	- arp header target ip address
342  *	- arp header source ethernet address
343  */
344 void
345 arprequest(struct ifnet *ifp, const struct in_addr *sip,
346     const struct in_addr *tip, u_char *enaddr)
347 {
348 	struct mbuf *m;
349 	struct arphdr *ah;
350 	struct sockaddr sa;
351 	u_char *carpaddr = NULL;
352 	uint8_t linkhdr[LLE_MAX_LINKHDR];
353 	size_t linkhdrsize;
354 	struct route ro;
355 	int error;
356 
357 	if (sip == NULL) {
358 		/*
359 		 * The caller did not supply a source address, try to find
360 		 * a compatible one among those assigned to this interface.
361 		 */
362 		struct ifaddr *ifa;
363 
364 		IF_ADDR_RLOCK(ifp);
365 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
366 			if (ifa->ifa_addr->sa_family != AF_INET)
367 				continue;
368 
369 			if (ifa->ifa_carp) {
370 				if ((*carp_iamatch_p)(ifa, &carpaddr) == 0)
371 					continue;
372 				sip = &IA_SIN(ifa)->sin_addr;
373 			} else {
374 				carpaddr = NULL;
375 				sip = &IA_SIN(ifa)->sin_addr;
376 			}
377 
378 			if (0 == ((sip->s_addr ^ tip->s_addr) &
379 			    IA_MASKSIN(ifa)->sin_addr.s_addr))
380 				break;  /* found it. */
381 		}
382 		IF_ADDR_RUNLOCK(ifp);
383 		if (sip == NULL) {
384 			printf("%s: cannot find matching address\n", __func__);
385 			return;
386 		}
387 	}
388 	if (enaddr == NULL)
389 		enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp);
390 
391 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
392 		return;
393 	m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
394 		2 * ifp->if_addrlen;
395 	m->m_pkthdr.len = m->m_len;
396 	M_ALIGN(m, m->m_len);
397 	ah = mtod(m, struct arphdr *);
398 	bzero((caddr_t)ah, m->m_len);
399 #ifdef MAC
400 	mac_netinet_arp_send(ifp, m);
401 #endif
402 	ah->ar_pro = htons(ETHERTYPE_IP);
403 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
404 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
405 	ah->ar_op = htons(ARPOP_REQUEST);
406 	bcopy(enaddr, ar_sha(ah), ah->ar_hln);
407 	bcopy(sip, ar_spa(ah), ah->ar_pln);
408 	bcopy(tip, ar_tpa(ah), ah->ar_pln);
409 	sa.sa_family = AF_ARP;
410 	sa.sa_len = 2;
411 
412 	/* Calculate link header for sending frame */
413 	bzero(&ro, sizeof(ro));
414 	linkhdrsize = sizeof(linkhdr);
415 	error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize);
416 	if (error != 0 && error != EAFNOSUPPORT) {
417 		ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n",
418 		    if_name(ifp), error);
419 		return;
420 	}
421 
422 	ro.ro_prepend = linkhdr;
423 	ro.ro_plen = linkhdrsize;
424 	ro.ro_flags = 0;
425 
426 	m->m_flags |= M_BCAST;
427 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
428 	(*ifp->if_output)(ifp, m, &sa, &ro);
429 	ARPSTAT_INC(txrequests);
430 }
431 
432 
433 /*
434  * Resolve an IP address into an ethernet address - heavy version.
435  * Used internally by arpresolve().
436  * We have already checked than  we can't use existing lle without
437  * modification so we have to acquire LLE_EXCLUSIVE lle lock.
438  *
439  * On success, desten and flags are filled in and the function returns 0;
440  * If the packet must be held pending resolution, we return EWOULDBLOCK
441  * On other errors, we return the corresponding error code.
442  * Note that m_freem() handles NULL.
443  */
444 static int
445 arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m,
446 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
447 	struct llentry **plle)
448 {
449 	struct llentry *la = NULL, *la_tmp;
450 	struct mbuf *curr = NULL;
451 	struct mbuf *next = NULL;
452 	int error, renew;
453 	char *lladdr;
454 	int ll_len;
455 
456 	if (pflags != NULL)
457 		*pflags = 0;
458 	if (plle != NULL)
459 		*plle = NULL;
460 
461 	if ((flags & LLE_CREATE) == 0) {
462 		IF_AFDATA_RLOCK(ifp);
463 		la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
464 		IF_AFDATA_RUNLOCK(ifp);
465 	}
466 	if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) {
467 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
468 		if (la == NULL) {
469 			char addrbuf[INET_ADDRSTRLEN];
470 
471 			log(LOG_DEBUG,
472 			    "arpresolve: can't allocate llinfo for %s on %s\n",
473 			    inet_ntoa_r(SIN(dst)->sin_addr, addrbuf),
474 			    if_name(ifp));
475 			m_freem(m);
476 			return (EINVAL);
477 		}
478 
479 		IF_AFDATA_WLOCK(ifp);
480 		LLE_WLOCK(la);
481 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
482 		/* Prefer ANY existing lle over newly-created one */
483 		if (la_tmp == NULL)
484 			lltable_link_entry(LLTABLE(ifp), la);
485 		IF_AFDATA_WUNLOCK(ifp);
486 		if (la_tmp != NULL) {
487 			lltable_free_entry(LLTABLE(ifp), la);
488 			la = la_tmp;
489 		}
490 	}
491 	if (la == NULL) {
492 		m_freem(m);
493 		return (EINVAL);
494 	}
495 
496 	if ((la->la_flags & LLE_VALID) &&
497 	    ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) {
498 		if (flags & LLE_ADDRONLY) {
499 			lladdr = la->ll_addr;
500 			ll_len = ifp->if_addrlen;
501 		} else {
502 			lladdr = la->r_linkdata;
503 			ll_len = la->r_hdrlen;
504 		}
505 		bcopy(lladdr, desten, ll_len);
506 
507 		/* Notify LLE code that the entry was used by datapath */
508 		llentry_mark_used(la);
509 		if (pflags != NULL)
510 			*pflags = la->la_flags & (LLE_VALID|LLE_IFADDR);
511 		if (plle) {
512 			LLE_ADDREF(la);
513 			*plle = la;
514 		}
515 		LLE_WUNLOCK(la);
516 		return (0);
517 	}
518 
519 	renew = (la->la_asked == 0 || la->la_expire != time_uptime);
520 	/*
521 	 * There is an arptab entry, but no ethernet address
522 	 * response yet.  Add the mbuf to the list, dropping
523 	 * the oldest packet if we have exceeded the system
524 	 * setting.
525 	 */
526 	if (m != NULL) {
527 		if (la->la_numheld >= V_arp_maxhold) {
528 			if (la->la_hold != NULL) {
529 				next = la->la_hold->m_nextpkt;
530 				m_freem(la->la_hold);
531 				la->la_hold = next;
532 				la->la_numheld--;
533 				ARPSTAT_INC(dropped);
534 			}
535 		}
536 		if (la->la_hold != NULL) {
537 			curr = la->la_hold;
538 			while (curr->m_nextpkt != NULL)
539 				curr = curr->m_nextpkt;
540 			curr->m_nextpkt = m;
541 		} else
542 			la->la_hold = m;
543 		la->la_numheld++;
544 	}
545 	/*
546 	 * Return EWOULDBLOCK if we have tried less than arp_maxtries. It
547 	 * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH
548 	 * if we have already sent arp_maxtries ARP requests. Retransmit the
549 	 * ARP request, but not faster than one request per second.
550 	 */
551 	if (la->la_asked < V_arp_maxtries)
552 		error = EWOULDBLOCK;	/* First request. */
553 	else
554 		error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN;
555 
556 	if (renew) {
557 		int canceled;
558 
559 		LLE_ADDREF(la);
560 		la->la_expire = time_uptime;
561 		canceled = callout_reset(&la->lle_timer, hz * V_arpt_down,
562 		    arptimer, la);
563 		if (canceled)
564 			LLE_REMREF(la);
565 		la->la_asked++;
566 		LLE_WUNLOCK(la);
567 		arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL);
568 		return (error);
569 	}
570 
571 	LLE_WUNLOCK(la);
572 	return (error);
573 }
574 
575 /*
576  * Resolve an IP address into an ethernet address.
577  */
578 int
579 arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
580     char *desten, uint32_t *pflags, struct llentry **plle)
581 {
582 	int error;
583 
584 	flags |= LLE_ADDRONLY;
585 	error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle);
586 	return (error);
587 }
588 
589 
590 /*
591  * Lookups link header based on an IP address.
592  * On input:
593  *    ifp is the interface we use
594  *    is_gw != 0 if @dst represents gateway to some destination
595  *    m is the mbuf. May be NULL if we don't have a packet.
596  *    dst is the next hop,
597  *    desten is the storage to put LL header.
598  *    flags returns subset of lle flags: LLE_VALID | LLE_IFADDR
599  *
600  * On success, full/partial link header and flags are filled in and
601  * the function returns 0.
602  * If the packet must be held pending resolution, we return EWOULDBLOCK
603  * On other errors, we return the corresponding error code.
604  * Note that m_freem() handles NULL.
605  */
606 int
607 arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
608 	const struct sockaddr *dst, u_char *desten, uint32_t *pflags,
609 	struct llentry **plle)
610 {
611 	struct llentry *la = NULL;
612 
613 	if (pflags != NULL)
614 		*pflags = 0;
615 	if (plle != NULL)
616 		*plle = NULL;
617 
618 	if (m != NULL) {
619 		if (m->m_flags & M_BCAST) {
620 			/* broadcast */
621 			(void)memcpy(desten,
622 			    ifp->if_broadcastaddr, ifp->if_addrlen);
623 			return (0);
624 		}
625 		if (m->m_flags & M_MCAST) {
626 			/* multicast */
627 			ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten);
628 			return (0);
629 		}
630 	}
631 
632 	IF_AFDATA_RLOCK(ifp);
633 	la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst);
634 	if (la != NULL && (la->r_flags & RLLE_VALID) != 0) {
635 		/* Entry found, let's copy lle info */
636 		bcopy(la->r_linkdata, desten, la->r_hdrlen);
637 		if (pflags != NULL)
638 			*pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR);
639 		/* Notify the LLE handling code that the entry was used. */
640 		llentry_mark_used(la);
641 		if (plle) {
642 			LLE_ADDREF(la);
643 			*plle = la;
644 			LLE_WUNLOCK(la);
645 		}
646 		IF_AFDATA_RUNLOCK(ifp);
647 		return (0);
648 	}
649 	if (plle && la)
650 		LLE_WUNLOCK(la);
651 	IF_AFDATA_RUNLOCK(ifp);
652 
653 	return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst,
654 	    desten, pflags, plle));
655 }
656 
657 /*
658  * Common length and type checks are done here,
659  * then the protocol-specific routine is called.
660  */
661 static void
662 arpintr(struct mbuf *m)
663 {
664 	struct arphdr *ar;
665 	struct ifnet *ifp;
666 	char *layer;
667 	int hlen;
668 
669 	ifp = m->m_pkthdr.rcvif;
670 
671 	if (m->m_len < sizeof(struct arphdr) &&
672 	    ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) {
673 		ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n",
674 		    if_name(ifp));
675 		return;
676 	}
677 	ar = mtod(m, struct arphdr *);
678 
679 	/* Check if length is sufficient */
680 	if (m->m_len <  arphdr_len(ar)) {
681 		m = m_pullup(m, arphdr_len(ar));
682 		if (m == NULL) {
683 			ARP_LOG(LOG_NOTICE, "short packet received on %s\n",
684 			    if_name(ifp));
685 			return;
686 		}
687 		ar = mtod(m, struct arphdr *);
688 	}
689 
690 	hlen = 0;
691 	layer = "";
692 	switch (ntohs(ar->ar_hrd)) {
693 	case ARPHRD_ETHER:
694 		hlen = ETHER_ADDR_LEN; /* RFC 826 */
695 		layer = "ethernet";
696 		break;
697 	case ARPHRD_IEEE802:
698 		hlen = 6; /* RFC 1390, FDDI_ADDR_LEN */
699 		layer = "fddi";
700 		break;
701 	case ARPHRD_ARCNET:
702 		hlen = 1; /* RFC 1201, ARC_ADDR_LEN */
703 		layer = "arcnet";
704 		break;
705 	case ARPHRD_INFINIBAND:
706 		hlen = 20;	/* RFC 4391, INFINIBAND_ALEN */
707 		layer = "infiniband";
708 		break;
709 	case ARPHRD_IEEE1394:
710 		hlen = 0; /* SHALL be 16 */ /* RFC 2734 */
711 		layer = "firewire";
712 
713 		/*
714 		 * Restrict too long hardware addresses.
715 		 * Currently we are capable of handling 20-byte
716 		 * addresses ( sizeof(lle->ll_addr) )
717 		 */
718 		if (ar->ar_hln >= 20)
719 			hlen = 16;
720 		break;
721 	default:
722 		ARP_LOG(LOG_NOTICE,
723 		    "packet with unknown hardware format 0x%02d received on "
724 		    "%s\n", ntohs(ar->ar_hrd), if_name(ifp));
725 		m_freem(m);
726 		return;
727 	}
728 
729 	if (hlen != 0 && hlen != ar->ar_hln) {
730 		ARP_LOG(LOG_NOTICE,
731 		    "packet with invalid %s address length %d received on %s\n",
732 		    layer, ar->ar_hln, if_name(ifp));
733 		m_freem(m);
734 		return;
735 	}
736 
737 	ARPSTAT_INC(received);
738 	switch (ntohs(ar->ar_pro)) {
739 #ifdef INET
740 	case ETHERTYPE_IP:
741 		in_arpinput(m);
742 		return;
743 #endif
744 	}
745 	m_freem(m);
746 }
747 
748 #ifdef INET
749 /*
750  * ARP for Internet protocols on 10 Mb/s Ethernet.
751  * Algorithm is that given in RFC 826.
752  * In addition, a sanity check is performed on the sender
753  * protocol address, to catch impersonators.
754  * We no longer handle negotiations for use of trailer protocol:
755  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
756  * along with IP replies if we wanted trailers sent to us,
757  * and also sent them in response to IP replies.
758  * This allowed either end to announce the desire to receive
759  * trailer packets.
760  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
761  * but formerly didn't normally send requests.
762  */
763 static int log_arp_wrong_iface = 1;
764 static int log_arp_movements = 1;
765 static int log_arp_permanent_modify = 1;
766 static int allow_multicast = 0;
767 
768 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW,
769 	&log_arp_wrong_iface, 0,
770 	"log arp packets arriving on the wrong interface");
771 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW,
772 	&log_arp_movements, 0,
773 	"log arp replies from MACs different than the one in the cache");
774 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW,
775 	&log_arp_permanent_modify, 0,
776 	"log arp replies from MACs different than the one in the permanent arp entry");
777 SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW,
778 	&allow_multicast, 0, "accept multicast addresses");
779 
780 static void
781 in_arpinput(struct mbuf *m)
782 {
783 	struct rm_priotracker in_ifa_tracker;
784 	struct arphdr *ah;
785 	struct ifnet *ifp = m->m_pkthdr.rcvif;
786 	struct llentry *la = NULL, *la_tmp;
787 	struct ifaddr *ifa;
788 	struct in_ifaddr *ia;
789 	struct sockaddr sa;
790 	struct in_addr isaddr, itaddr, myaddr;
791 	u_int8_t *enaddr = NULL;
792 	int op;
793 	int bridged = 0, is_bridge = 0;
794 	int carped;
795 	struct sockaddr_in sin;
796 	struct sockaddr *dst;
797 	struct nhop4_basic nh4;
798 	uint8_t linkhdr[LLE_MAX_LINKHDR];
799 	struct route ro;
800 	size_t linkhdrsize;
801 	int lladdr_off;
802 	int error;
803 	char addrbuf[INET_ADDRSTRLEN];
804 
805 	sin.sin_len = sizeof(struct sockaddr_in);
806 	sin.sin_family = AF_INET;
807 	sin.sin_addr.s_addr = 0;
808 
809 	if (ifp->if_bridge)
810 		bridged = 1;
811 	if (ifp->if_type == IFT_BRIDGE)
812 		is_bridge = 1;
813 
814 	/*
815 	 * We already have checked that mbuf contains enough contiguous data
816 	 * to hold entire arp message according to the arp header.
817 	 */
818 	ah = mtod(m, struct arphdr *);
819 
820 	/*
821 	 * ARP is only for IPv4 so we can reject packets with
822 	 * a protocol length not equal to an IPv4 address.
823 	 */
824 	if (ah->ar_pln != sizeof(struct in_addr)) {
825 		ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n",
826 		    sizeof(struct in_addr));
827 		goto drop;
828 	}
829 
830 	if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) {
831 		ARP_LOG(LOG_NOTICE, "%*D is multicast\n",
832 		    ifp->if_addrlen, (u_char *)ar_sha(ah), ":");
833 		goto drop;
834 	}
835 
836 	op = ntohs(ah->ar_op);
837 	(void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
838 	(void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
839 
840 	if (op == ARPOP_REPLY)
841 		ARPSTAT_INC(rxreplies);
842 
843 	/*
844 	 * For a bridge, we want to check the address irrespective
845 	 * of the receive interface. (This will change slightly
846 	 * when we have clusters of interfaces).
847 	 */
848 	IN_IFADDR_RLOCK(&in_ifa_tracker);
849 	LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
850 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
851 		    ia->ia_ifp == ifp) &&
852 		    itaddr.s_addr == ia->ia_addr.sin_addr.s_addr &&
853 		    (ia->ia_ifa.ifa_carp == NULL ||
854 		    (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) {
855 			ifa_ref(&ia->ia_ifa);
856 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
857 			goto match;
858 		}
859 	}
860 	LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash)
861 		if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) ||
862 		    ia->ia_ifp == ifp) &&
863 		    isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) {
864 			ifa_ref(&ia->ia_ifa);
865 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
866 			goto match;
867 		}
868 
869 #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia)				\
870   (ia->ia_ifp->if_bridge == ifp->if_softc &&				\
871   !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) &&	\
872   addr == ia->ia_addr.sin_addr.s_addr)
873 	/*
874 	 * Check the case when bridge shares its MAC address with
875 	 * some of its children, so packets are claimed by bridge
876 	 * itself (bridge_input() does it first), but they are really
877 	 * meant to be destined to the bridge member.
878 	 */
879 	if (is_bridge) {
880 		LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) {
881 			if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) {
882 				ifa_ref(&ia->ia_ifa);
883 				ifp = ia->ia_ifp;
884 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
885 				goto match;
886 			}
887 		}
888 	}
889 #undef BDG_MEMBER_MATCHES_ARP
890 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
891 
892 	/*
893 	 * No match, use the first inet address on the receive interface
894 	 * as a dummy address for the rest of the function.
895 	 */
896 	IF_ADDR_RLOCK(ifp);
897 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
898 		if (ifa->ifa_addr->sa_family == AF_INET &&
899 		    (ifa->ifa_carp == NULL ||
900 		    (*carp_iamatch_p)(ifa, &enaddr))) {
901 			ia = ifatoia(ifa);
902 			ifa_ref(ifa);
903 			IF_ADDR_RUNLOCK(ifp);
904 			goto match;
905 		}
906 	IF_ADDR_RUNLOCK(ifp);
907 
908 	/*
909 	 * If bridging, fall back to using any inet address.
910 	 */
911 	IN_IFADDR_RLOCK(&in_ifa_tracker);
912 	if (!bridged || (ia = TAILQ_FIRST(&V_in_ifaddrhead)) == NULL) {
913 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
914 		goto drop;
915 	}
916 	ifa_ref(&ia->ia_ifa);
917 	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
918 match:
919 	if (!enaddr)
920 		enaddr = (u_int8_t *)IF_LLADDR(ifp);
921 	carped = (ia->ia_ifa.ifa_carp != NULL);
922 	myaddr = ia->ia_addr.sin_addr;
923 	ifa_free(&ia->ia_ifa);
924 	if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen))
925 		goto drop;	/* it's from me, ignore it. */
926 	if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
927 		ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address "
928 		    "%s!\n", inet_ntoa_r(isaddr, addrbuf));
929 		goto drop;
930 	}
931 
932 	if (ifp->if_addrlen != ah->ar_hln) {
933 		ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, "
934 		    "i/f %d (ignored)\n", ifp->if_addrlen,
935 		    (u_char *) ar_sha(ah), ":", ah->ar_hln,
936 		    ifp->if_addrlen);
937 		goto drop;
938 	}
939 
940 	/*
941 	 * Warn if another host is using the same IP address, but only if the
942 	 * IP address isn't 0.0.0.0, which is used for DHCP only, in which
943 	 * case we suppress the warning to avoid false positive complaints of
944 	 * potential misconfiguration.
945 	 */
946 	if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr &&
947 	    myaddr.s_addr != 0) {
948 		ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n",
949 		   ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
950 		   inet_ntoa_r(isaddr, addrbuf), ifp->if_xname);
951 		itaddr = myaddr;
952 		ARPSTAT_INC(dupips);
953 		goto reply;
954 	}
955 	if (ifp->if_flags & IFF_STATICARP)
956 		goto reply;
957 
958 	bzero(&sin, sizeof(sin));
959 	sin.sin_len = sizeof(struct sockaddr_in);
960 	sin.sin_family = AF_INET;
961 	sin.sin_addr = isaddr;
962 	dst = (struct sockaddr *)&sin;
963 	IF_AFDATA_RLOCK(ifp);
964 	la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
965 	IF_AFDATA_RUNLOCK(ifp);
966 	if (la != NULL)
967 		arp_check_update_lle(ah, isaddr, ifp, bridged, la);
968 	else if (itaddr.s_addr == myaddr.s_addr) {
969 		/*
970 		 * Request/reply to our address, but no lle exists yet.
971 		 * Calculate full link prepend to use in lle.
972 		 */
973 		linkhdrsize = sizeof(linkhdr);
974 		if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
975 		    &linkhdrsize, &lladdr_off) != 0)
976 			goto reply;
977 
978 		/* Allocate new entry */
979 		la = lltable_alloc_entry(LLTABLE(ifp), 0, dst);
980 		if (la == NULL) {
981 
982 			/*
983 			 * lle creation may fail if source address belongs
984 			 * to non-directly connected subnet. However, we
985 			 * will try to answer the request instead of dropping
986 			 * frame.
987 			 */
988 			goto reply;
989 		}
990 		lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
991 		    lladdr_off);
992 
993 		IF_AFDATA_WLOCK(ifp);
994 		LLE_WLOCK(la);
995 		la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
996 
997 		/*
998 		 * Check if lle still does not exists.
999 		 * If it does, that means that we either
1000 		 * 1) have configured it explicitly, via
1001 		 * 1a) 'arp -s' static entry or
1002 		 * 1b) interface address static record
1003 		 * or
1004 		 * 2) it was the result of sending first packet to-host
1005 		 * or
1006 		 * 3) it was another arp reply packet we handled in
1007 		 * different thread.
1008 		 *
1009 		 * In all cases except 3) we definitely need to prefer
1010 		 * existing lle. For the sake of simplicity, prefer any
1011 		 * existing lle over newly-create one.
1012 		 */
1013 		if (la_tmp == NULL)
1014 			lltable_link_entry(LLTABLE(ifp), la);
1015 		IF_AFDATA_WUNLOCK(ifp);
1016 
1017 		if (la_tmp == NULL) {
1018 			arp_mark_lle_reachable(la);
1019 			LLE_WUNLOCK(la);
1020 		} else {
1021 			/* Free newly-create entry and handle packet */
1022 			lltable_free_entry(LLTABLE(ifp), la);
1023 			la = la_tmp;
1024 			la_tmp = NULL;
1025 			arp_check_update_lle(ah, isaddr, ifp, bridged, la);
1026 			/* arp_check_update_lle() returns @la unlocked */
1027 		}
1028 		la = NULL;
1029 	}
1030 reply:
1031 	if (op != ARPOP_REQUEST)
1032 		goto drop;
1033 	ARPSTAT_INC(rxrequests);
1034 
1035 	if (itaddr.s_addr == myaddr.s_addr) {
1036 		/* Shortcut.. the receiving interface is the target. */
1037 		(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1038 		(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1039 	} else {
1040 		struct llentry *lle = NULL;
1041 
1042 		sin.sin_addr = itaddr;
1043 		IF_AFDATA_RLOCK(ifp);
1044 		lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin);
1045 		IF_AFDATA_RUNLOCK(ifp);
1046 
1047 		if ((lle != NULL) && (lle->la_flags & LLE_PUB)) {
1048 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1049 			(void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln);
1050 			LLE_RUNLOCK(lle);
1051 		} else {
1052 
1053 			if (lle != NULL)
1054 				LLE_RUNLOCK(lle);
1055 
1056 			if (!V_arp_proxyall)
1057 				goto drop;
1058 
1059 			/* XXX MRT use table 0 for arp reply  */
1060 			if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0)
1061 				goto drop;
1062 
1063 			/*
1064 			 * Don't send proxies for nodes on the same interface
1065 			 * as this one came out of, or we'll get into a fight
1066 			 * over who claims what Ether address.
1067 			 */
1068 			if (nh4.nh_ifp == ifp)
1069 				goto drop;
1070 
1071 			(void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln);
1072 			(void)memcpy(ar_sha(ah), enaddr, ah->ar_hln);
1073 
1074 			/*
1075 			 * Also check that the node which sent the ARP packet
1076 			 * is on the interface we expect it to be on. This
1077 			 * avoids ARP chaos if an interface is connected to the
1078 			 * wrong network.
1079 			 */
1080 
1081 			/* XXX MRT use table 0 for arp checks */
1082 			if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0)
1083 				goto drop;
1084 			if (nh4.nh_ifp != ifp) {
1085 				ARP_LOG(LOG_INFO, "proxy: ignoring request"
1086 				    " from %s via %s\n",
1087 				    inet_ntoa_r(isaddr, addrbuf),
1088 				    ifp->if_xname);
1089 				goto drop;
1090 			}
1091 
1092 #ifdef DEBUG_PROXY
1093 			printf("arp: proxying for %s\n",
1094 			    inet_ntoa_r(itaddr, addrbuf));
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_r(itaddr, addrbuf));
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 	char addrbuf[INET_ADDRSTRLEN];
1166 
1167 	LLE_WLOCK_ASSERT(la);
1168 
1169 	/* the following is not an error when doing bridging */
1170 	if (!bridged && la->lle_tbl->llt_ifp != ifp) {
1171 		if (log_arp_wrong_iface)
1172 			ARP_LOG(LOG_WARNING, "%s is on %s "
1173 			    "but got reply from %*D on %s\n",
1174 			    inet_ntoa_r(isaddr, addrbuf),
1175 			    la->lle_tbl->llt_ifp->if_xname,
1176 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1177 			    ifp->if_xname);
1178 		LLE_WUNLOCK(la);
1179 		return;
1180 	}
1181 	if ((la->la_flags & LLE_VALID) &&
1182 	    bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) {
1183 		if (la->la_flags & LLE_STATIC) {
1184 			LLE_WUNLOCK(la);
1185 			if (log_arp_permanent_modify)
1186 				ARP_LOG(LOG_ERR,
1187 				    "%*D attempts to modify "
1188 				    "permanent entry for %s on %s\n",
1189 				    ifp->if_addrlen,
1190 				    (u_char *)ar_sha(ah), ":",
1191 				    inet_ntoa_r(isaddr, addrbuf),
1192 				    ifp->if_xname);
1193 			return;
1194 		}
1195 		if (log_arp_movements) {
1196 			ARP_LOG(LOG_INFO, "%s moved from %*D "
1197 			    "to %*D on %s\n",
1198 			    inet_ntoa_r(isaddr, addrbuf),
1199 			    ifp->if_addrlen,
1200 			    (u_char *)la->ll_addr, ":",
1201 			    ifp->if_addrlen, (u_char *)ar_sha(ah), ":",
1202 			    ifp->if_xname);
1203 		}
1204 	}
1205 
1206 	/* Calculate full link prepend to use in lle */
1207 	linkhdrsize = sizeof(linkhdr);
1208 	if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr,
1209 	    &linkhdrsize, &lladdr_off) != 0)
1210 		return;
1211 
1212 	/* Check if something has changed */
1213 	if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 ||
1214 	    (la->la_flags & LLE_VALID) == 0) {
1215 		/* Try to perform LLE update */
1216 		if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize,
1217 		    lladdr_off) == 0)
1218 			return;
1219 
1220 		/* Clear fast path feedback request if set */
1221 		la->r_skip_req = 0;
1222 	}
1223 
1224 	arp_mark_lle_reachable(la);
1225 
1226 	/*
1227 	 * The packets are all freed within the call to the output
1228 	 * routine.
1229 	 *
1230 	 * NB: The lock MUST be released before the call to the
1231 	 * output routine.
1232 	 */
1233 	if (la->la_hold != NULL) {
1234 		m_hold = la->la_hold;
1235 		la->la_hold = NULL;
1236 		la->la_numheld = 0;
1237 		lltable_fill_sa_entry(la, &sa);
1238 		LLE_WUNLOCK(la);
1239 		for (; m_hold != NULL; m_hold = m_hold_next) {
1240 			m_hold_next = m_hold->m_nextpkt;
1241 			m_hold->m_nextpkt = NULL;
1242 			/* Avoid confusing lower layers. */
1243 			m_clrprotoflags(m_hold);
1244 			(*ifp->if_output)(ifp, m_hold, &sa, NULL);
1245 		}
1246 	} else
1247 		LLE_WUNLOCK(la);
1248 }
1249 
1250 static void
1251 arp_mark_lle_reachable(struct llentry *la)
1252 {
1253 	int canceled, wtime;
1254 
1255 	LLE_WLOCK_ASSERT(la);
1256 
1257 	la->ln_state = ARP_LLINFO_REACHABLE;
1258 	EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED);
1259 
1260 	if (!(la->la_flags & LLE_STATIC)) {
1261 		LLE_ADDREF(la);
1262 		la->la_expire = time_uptime + V_arpt_keep;
1263 		wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit;
1264 		if (wtime < 0)
1265 			wtime = V_arpt_keep;
1266 		canceled = callout_reset(&la->lle_timer,
1267 		    hz * wtime, arptimer, la);
1268 		if (canceled)
1269 			LLE_REMREF(la);
1270 	}
1271 	la->la_asked = 0;
1272 	la->la_preempt = V_arp_maxtries;
1273 }
1274 
1275 /*
1276  * Add permanent link-layer record for given interface address.
1277  */
1278 static __noinline void
1279 arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst)
1280 {
1281 	struct llentry *lle, *lle_tmp;
1282 
1283 	/*
1284 	 * Interface address LLE record is considered static
1285 	 * because kernel code relies on LLE_STATIC flag to check
1286 	 * if these entries can be rewriten by arp updates.
1287 	 */
1288 	lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst);
1289 	if (lle == NULL) {
1290 		log(LOG_INFO, "arp_ifinit: cannot create arp "
1291 		    "entry for interface address\n");
1292 		return;
1293 	}
1294 
1295 	IF_AFDATA_WLOCK(ifp);
1296 	LLE_WLOCK(lle);
1297 	/* Unlink any entry if exists */
1298 	lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst);
1299 	if (lle_tmp != NULL)
1300 		lltable_unlink_entry(LLTABLE(ifp), lle_tmp);
1301 
1302 	lltable_link_entry(LLTABLE(ifp), lle);
1303 	IF_AFDATA_WUNLOCK(ifp);
1304 
1305 	if (lle_tmp != NULL)
1306 		EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED);
1307 
1308 	EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED);
1309 	LLE_WUNLOCK(lle);
1310 	if (lle_tmp != NULL)
1311 		lltable_free_entry(LLTABLE(ifp), lle_tmp);
1312 }
1313 
1314 /*
1315  * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range
1316  * of valid values.
1317  */
1318 static int
1319 sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS)
1320 {
1321 	int error;
1322 	int rexmit_count = *(int *)arg1;
1323 
1324 	error = sysctl_handle_int(oidp, &rexmit_count, 0, req);
1325 
1326 	/* Enforce limits on any new value that may have been set. */
1327 	if (!error && req->newptr) {
1328 		/* A new value was set. */
1329 		if (rexmit_count < 0) {
1330 			rexmit_count = 0;
1331 		} else if (rexmit_count > MAX_GARP_RETRANSMITS) {
1332 			rexmit_count = MAX_GARP_RETRANSMITS;
1333 		}
1334 		*(int *)arg1 = rexmit_count;
1335 	}
1336 
1337 	return (error);
1338 }
1339 
1340 /*
1341  * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to
1342  * retransmit it again. A pending callout owns a reference to the ifa.
1343  */
1344 static void
1345 garp_rexmit(void *arg)
1346 {
1347 	struct in_ifaddr *ia = arg;
1348 
1349 	if (callout_pending(&ia->ia_garp_timer) ||
1350 	    !callout_active(&ia->ia_garp_timer)) {
1351 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1352 		ifa_free(&ia->ia_ifa);
1353 		return;
1354 	}
1355 
1356 	/*
1357 	 * Drop lock while the ARP request is generated.
1358 	 */
1359 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1360 
1361 	arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr,
1362 	    &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp));
1363 
1364 	/*
1365 	 * Increment the count of retransmissions. If the count has reached the
1366 	 * maximum value, stop sending the GARP packets. Otherwise, schedule
1367 	 * the callout to retransmit another GARP packet.
1368 	 */
1369 	++ia->ia_garp_count;
1370 	if (ia->ia_garp_count >= garp_rexmit_count) {
1371 		ifa_free(&ia->ia_ifa);
1372 	} else {
1373 		int rescheduled;
1374 		IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1375 		rescheduled = callout_reset(&ia->ia_garp_timer,
1376 		    (1 << ia->ia_garp_count) * hz,
1377 		    garp_rexmit, ia);
1378 		IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1379 		if (rescheduled) {
1380 			ifa_free(&ia->ia_ifa);
1381 		}
1382 	}
1383 }
1384 
1385 /*
1386  * Start the GARP retransmit timer.
1387  *
1388  * A single GARP is always transmitted when an IPv4 address is added
1389  * to an interface and that is usually sufficient. However, in some
1390  * circumstances, such as when a shared address is passed between
1391  * cluster nodes, this single GARP may occasionally be dropped or
1392  * lost. This can lead to neighbors on the network link working with a
1393  * stale ARP cache and sending packets destined for that address to
1394  * the node that previously owned the address, which may not respond.
1395  *
1396  * To avoid this situation, GARP retransmits can be enabled by setting
1397  * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater
1398  * than zero. The setting represents the maximum number of
1399  * retransmissions. The interval between retransmissions is calculated
1400  * using an exponential backoff algorithm, doubling each time, so the
1401  * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds).
1402  */
1403 static void
1404 garp_timer_start(struct ifaddr *ifa)
1405 {
1406 	struct in_ifaddr *ia = (struct in_ifaddr *) ifa;
1407 
1408 	IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp);
1409 	ia->ia_garp_count = 0;
1410 	if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz,
1411 	    garp_rexmit, ia) == 0) {
1412 		ifa_ref(ifa);
1413 	}
1414 	IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp);
1415 }
1416 
1417 void
1418 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1419 {
1420 	const struct sockaddr_in *dst_in;
1421 	const struct sockaddr *dst;
1422 
1423 	if (ifa->ifa_carp != NULL)
1424 		return;
1425 
1426 	dst = ifa->ifa_addr;
1427 	dst_in = (const struct sockaddr_in *)dst;
1428 
1429 	if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY)
1430 		return;
1431 	arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp));
1432 	if (garp_rexmit_count > 0) {
1433 		garp_timer_start(ifa);
1434 	}
1435 
1436 	arp_add_ifa_lle(ifp, dst);
1437 }
1438 
1439 void
1440 arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr)
1441 {
1442 
1443 	if (ntohl(addr.s_addr) != INADDR_ANY)
1444 		arprequest(ifp, &addr, &addr, enaddr);
1445 }
1446 
1447 /*
1448  * Sends gratuitous ARPs for each ifaddr to notify other
1449  * nodes about the address change.
1450  */
1451 static __noinline void
1452 arp_handle_ifllchange(struct ifnet *ifp)
1453 {
1454 	struct ifaddr *ifa;
1455 
1456 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1457 		if (ifa->ifa_addr->sa_family == AF_INET)
1458 			arp_ifinit(ifp, ifa);
1459 	}
1460 }
1461 
1462 /*
1463  * A handler for interface link layer address change event.
1464  */
1465 static void
1466 arp_iflladdr(void *arg __unused, struct ifnet *ifp)
1467 {
1468 
1469 	lltable_update_ifaddr(LLTABLE(ifp));
1470 
1471 	if ((ifp->if_flags & IFF_UP) != 0)
1472 		arp_handle_ifllchange(ifp);
1473 }
1474 
1475 static void
1476 vnet_arp_init(void)
1477 {
1478 
1479 	if (IS_DEFAULT_VNET(curvnet)) {
1480 		netisr_register(&arp_nh);
1481 		iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event,
1482 		    arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
1483 	}
1484 #ifdef VIMAGE
1485 	else
1486 		netisr_register_vnet(&arp_nh);
1487 #endif
1488 }
1489 VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND,
1490     vnet_arp_init, 0);
1491 
1492 #ifdef VIMAGE
1493 /*
1494  * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH
1495  * lookups after destroying the hash.  Ideally this would go on SI_ORDER_3.5.
1496  */
1497 static void
1498 vnet_arp_destroy(__unused void *arg)
1499 {
1500 
1501 	netisr_unregister_vnet(&arp_nh);
1502 }
1503 VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD,
1504     vnet_arp_destroy, NULL);
1505 #endif
1506