xref: /netbsd-src/sys/netinet/if_arp.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: if_arp.c,v 1.130 2007/12/07 19:46:18 elad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Public Access Networks Corporation ("Panix").  It was developed under
9  * contract to Panix by Eric Haszlakiewicz and Thor Lancelot Simon.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1988, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. Neither the name of the University nor the names of its contributors
53  *    may be used to endorse or promote products derived from this software
54  *    without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66  * SUCH DAMAGE.
67  *
68  *	@(#)if_ether.c	8.2 (Berkeley) 9/26/94
69  */
70 
71 /*
72  * Ethernet address resolution protocol.
73  * TODO:
74  *	add "inuse/lock" bit (or ref. count) along with valid bit
75  */
76 
77 #include <sys/cdefs.h>
78 __KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1.130 2007/12/07 19:46:18 elad Exp $");
79 
80 #include "opt_ddb.h"
81 #include "opt_inet.h"
82 
83 #ifdef INET
84 
85 #include "bridge.h"
86 
87 #include <sys/param.h>
88 #include <sys/systm.h>
89 #include <sys/callout.h>
90 #include <sys/malloc.h>
91 #include <sys/mbuf.h>
92 #include <sys/socket.h>
93 #include <sys/time.h>
94 #include <sys/timetc.h>
95 #include <sys/kernel.h>
96 #include <sys/errno.h>
97 #include <sys/ioctl.h>
98 #include <sys/syslog.h>
99 #include <sys/proc.h>
100 #include <sys/protosw.h>
101 #include <sys/domain.h>
102 #include <sys/sysctl.h>
103 
104 #include <net/ethertypes.h>
105 #include <net/if.h>
106 #include <net/if_dl.h>
107 #include <net/if_token.h>
108 #include <net/if_types.h>
109 #include <net/if_ether.h>
110 #include <net/route.h>
111 
112 #include <netinet/in.h>
113 #include <netinet/in_systm.h>
114 #include <netinet/in_var.h>
115 #include <netinet/ip.h>
116 #include <netinet/if_inarp.h>
117 
118 #include "arcnet.h"
119 #if NARCNET > 0
120 #include <net/if_arc.h>
121 #endif
122 #include "fddi.h"
123 #if NFDDI > 0
124 #include <net/if_fddi.h>
125 #endif
126 #include "token.h"
127 #include "carp.h"
128 #if NCARP > 0
129 #include <netinet/ip_carp.h>
130 #endif
131 
132 #define SIN(s) ((struct sockaddr_in *)s)
133 #define SRP(s) ((struct sockaddr_inarp *)s)
134 
135 /*
136  * ARP trailer negotiation.  Trailer protocol is not IP specific,
137  * but ARP request/response use IP addresses.
138  */
139 #define ETHERTYPE_IPTRAILERS ETHERTYPE_TRAIL
140 
141 /* timer values */
142 int	arpt_prune = (5*60*1);	/* walk list every 5 minutes */
143 int	arpt_keep = (20*60);	/* once resolved, good for 20 more minutes */
144 int	arpt_down = 20;		/* once declared down, don't send for 20 secs */
145 int	arpt_refresh = (5*60);	/* time left before refreshing */
146 #define	rt_expire rt_rmx.rmx_expire
147 #define	rt_pksent rt_rmx.rmx_pksent
148 
149 static	struct sockaddr *arp_setgate(struct rtentry *, struct sockaddr *,
150 	    const struct sockaddr *);
151 static	void arptfree(struct llinfo_arp *);
152 static	void arptimer(void *);
153 static	struct llinfo_arp *arplookup1(struct mbuf *, const struct in_addr *,
154 				      int, int, struct rtentry *);
155 static	struct llinfo_arp *arplookup(struct mbuf *, const struct in_addr *,
156 					  int, int);
157 static	void in_arpinput(struct mbuf *);
158 
159 LIST_HEAD(, llinfo_arp) llinfo_arp;
160 struct	ifqueue arpintrq = {
161 	.ifq_head = NULL,
162 	.ifq_tail = NULL,
163 	.ifq_len = 0,
164 	.ifq_maxlen = 50,
165 	.ifq_drops = 0,
166 };
167 int	arp_inuse, arp_allocated, arp_intimer;
168 int	arp_maxtries = 5;
169 int	useloopback = 1;	/* use loopback interface for local traffic */
170 int	arpinit_done = 0;
171 
172 struct	arpstat arpstat;
173 struct	callout arptimer_ch;
174 
175 /* revarp state */
176 struct	in_addr myip, srv_ip;
177 int	myip_initialized = 0;
178 int	revarp_in_progress = 0;
179 struct	ifnet *myip_ifp = NULL;
180 
181 #ifdef DDB
182 static void db_print_sa(const struct sockaddr *);
183 static void db_print_ifa(struct ifaddr *);
184 static void db_print_llinfo(void *);
185 static int db_show_rtentry(struct rtentry *, void *);
186 #endif
187 
188 /*
189  * this should be elsewhere.
190  */
191 
192 static char *
193 lla_snprintf(u_int8_t *, int);
194 
195 static char *
196 lla_snprintf(u_int8_t *adrp, int len)
197 {
198 #define NUMBUFS 3
199 	static char buf[NUMBUFS][16*3];
200 	static int bnum = 0;
201 
202 	int i;
203 	char *p;
204 
205 	p = buf[bnum];
206 
207 	*p++ = hexdigits[(*adrp)>>4];
208 	*p++ = hexdigits[(*adrp++)&0xf];
209 
210 	for (i=1; i<len && i<16; i++) {
211 		*p++ = ':';
212 		*p++ = hexdigits[(*adrp)>>4];
213 		*p++ = hexdigits[(*adrp++)&0xf];
214 	}
215 
216 	*p = 0;
217 	p = buf[bnum];
218 	bnum = (bnum + 1) % NUMBUFS;
219 	return p;
220 }
221 
222 DOMAIN_DEFINE(arpdomain);	/* forward declare and add to link set */
223 
224 const struct protosw arpsw[] = {
225 	{ .pr_type = 0,
226 	  .pr_domain = &arpdomain,
227 	  .pr_protocol = 0,
228 	  .pr_flags = 0,
229 	  .pr_input = 0,
230 	  .pr_output = 0,
231 	  .pr_ctlinput = 0,
232 	  .pr_ctloutput = 0,
233 	  .pr_usrreq =  0,
234 	  .pr_init = 0,
235 	  .pr_fasttimo = 0,
236 	  .pr_slowtimo = 0,
237 	  .pr_drain = arp_drain,
238 	}
239 };
240 
241 
242 struct domain arpdomain = {
243 	.dom_family = PF_ARP,
244 	.dom_name = "arp",
245 	.dom_protosw = arpsw,
246 	.dom_protoswNPROTOSW = &arpsw[sizeof(arpsw)/sizeof(arpsw[0])],
247 };
248 
249 /*
250  * ARP table locking.
251  *
252  * to prevent lossage vs. the arp_drain routine (which may be called at
253  * any time, including in a device driver context), we do two things:
254  *
255  * 1) manipulation of la->la_hold is done at splnet() (for all of
256  * about two instructions).
257  *
258  * 2) manipulation of the arp table's linked list is done under the
259  * protection of the ARP_LOCK; if arp_drain() or arptimer is called
260  * while the arp table is locked, we punt and try again later.
261  */
262 
263 static int	arp_locked;
264 static inline int arp_lock_try(int);
265 static inline void arp_unlock(void);
266 
267 static inline int
268 arp_lock_try(int recurse)
269 {
270 	int s;
271 
272 	/*
273 	 * Use splvm() -- we're blocking things that would cause
274 	 * mbuf allocation.
275 	 */
276 	s = splvm();
277 	if (!recurse && arp_locked) {
278 		splx(s);
279 		return (0);
280 	}
281 	arp_locked++;
282 	splx(s);
283 	return (1);
284 }
285 
286 static inline void
287 arp_unlock(void)
288 {
289 	int s;
290 
291 	s = splvm();
292 	arp_locked--;
293 	splx(s);
294 }
295 
296 #ifdef DIAGNOSTIC
297 #define	ARP_LOCK(recurse)						\
298 do {									\
299 	if (arp_lock_try(recurse) == 0) {				\
300 		printf("%s:%d: arp already locked\n", __FILE__, __LINE__); \
301 		panic("arp_lock");					\
302 	}								\
303 } while (/*CONSTCOND*/ 0)
304 #define	ARP_LOCK_CHECK()						\
305 do {									\
306 	if (arp_locked == 0) {						\
307 		printf("%s:%d: arp lock not held\n", __FILE__, __LINE__); \
308 		panic("arp lock check");				\
309 	}								\
310 } while (/*CONSTCOND*/ 0)
311 #else
312 #define	ARP_LOCK(x)		(void) arp_lock_try(x)
313 #define	ARP_LOCK_CHECK()	/* nothing */
314 #endif
315 
316 #define	ARP_UNLOCK()		arp_unlock()
317 
318 /*
319  * ARP protocol drain routine.  Called when memory is in short supply.
320  * Called at splvm();
321  */
322 
323 void
324 arp_drain(void)
325 {
326 	struct llinfo_arp *la, *nla;
327 	int count = 0;
328 	struct mbuf *mold;
329 
330 	if (arp_lock_try(0) == 0) {
331 		printf("arp_drain: locked; punting\n");
332 		return;
333 	}
334 
335 	for (la = LIST_FIRST(&llinfo_arp); la != 0; la = nla) {
336 		nla = LIST_NEXT(la, la_list);
337 
338 		mold = la->la_hold;
339 		la->la_hold = 0;
340 
341 		if (mold) {
342 			m_freem(mold);
343 			count++;
344 		}
345 	}
346 	ARP_UNLOCK();
347 	arpstat.as_dfrdropped += count;
348 }
349 
350 
351 /*
352  * Timeout routine.  Age arp_tab entries periodically.
353  */
354 /* ARGSUSED */
355 static void
356 arptimer(void *arg)
357 {
358 	int s;
359 	struct llinfo_arp *la, *nla;
360 
361 	s = splsoftnet();
362 
363 	if (arp_lock_try(0) == 0) {
364 		/* get it later.. */
365 		splx(s);
366 		return;
367 	}
368 
369 	callout_reset(&arptimer_ch, arpt_prune * hz, arptimer, NULL);
370 	for (la = LIST_FIRST(&llinfo_arp); la != 0; la = nla) {
371 		struct rtentry *rt = la->la_rt;
372 
373 		nla = LIST_NEXT(la, la_list);
374 		if (rt->rt_expire == 0)
375 			continue;
376 		if ((rt->rt_expire - time_second) < arpt_refresh &&
377 		    rt->rt_pksent > (time_second - arpt_keep)) {
378 			/*
379 			 * If the entry has been used during since last
380 			 * refresh, try to renew it before deleting.
381 			 */
382 			arprequest(rt->rt_ifp,
383 			    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
384 			    &satocsin(rt_getkey(rt))->sin_addr,
385 			    CLLADDR(rt->rt_ifp->if_sadl));
386 		} else if (rt->rt_expire <= time_second)
387 			arptfree(la); /* timer has expired; clear */
388 	}
389 
390 	ARP_UNLOCK();
391 
392 	splx(s);
393 }
394 
395 /*
396  * We set the gateway for RTF_CLONING routes to a "prototype"
397  * link-layer sockaddr whose interface type (if_type) and interface
398  * index (if_index) fields are prepared.
399  */
400 static struct sockaddr *
401 arp_setgate(struct rtentry *rt, struct sockaddr *gate,
402     const struct sockaddr *netmask)
403 {
404 	const struct ifnet *ifp = rt->rt_ifp;
405 	uint8_t namelen = strlen(ifp->if_xname);
406 	uint8_t addrlen = ifp->if_addrlen;
407 
408 	/*
409 	 * XXX: If this is a manually added route to interface
410 	 * such as older version of routed or gated might provide,
411 	 * restore cloning bit.
412 	 */
413 	if ((rt->rt_flags & RTF_HOST) == 0 && netmask != NULL &&
414 	    satocsin(netmask)->sin_addr.s_addr != 0xffffffff)
415 		rt->rt_flags |= RTF_CLONING;
416 	if (rt->rt_flags & RTF_CLONING) {
417 		union {
418 			struct sockaddr sa;
419 			struct sockaddr_storage ss;
420 			struct sockaddr_dl sdl;
421 		} u;
422 		/*
423 		 * Case 1: This route should come from a route to iface.
424 		 */
425 		sockaddr_dl_init(&u.sdl, sizeof(u.ss),
426 		    ifp->if_index, ifp->if_type, NULL, namelen, NULL, addrlen);
427 		rt_setgate(rt, &u.sa);
428 		gate = rt->rt_gateway;
429 	}
430 	return gate;
431 }
432 
433 /*
434  * Parallel to llc_rtrequest.
435  */
436 void
437 arp_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
438 {
439 	struct sockaddr *gate = rt->rt_gateway;
440 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
441 	size_t allocsize;
442 	struct mbuf *mold;
443 	int s;
444 	struct in_ifaddr *ia;
445 	struct ifaddr *ifa;
446 	struct ifnet *ifp = rt->rt_ifp;
447 	uint8_t namelen = strlen(ifp->if_xname);
448 	uint8_t addrlen = ifp->if_addrlen;
449 
450 	if (!arpinit_done) {
451 		arpinit_done = 1;
452 		/*
453 		 * We generate expiration times from time_second
454 		 * so avoid accidentally creating permanent routes.
455 		 */
456 		if (time_second == 0) {
457 #ifdef __HAVE_TIMECOUNTER
458 			struct timespec ts;
459 			ts.tv_sec = 1;
460 			ts.tv_nsec = 0;
461 			tc_setclock(&ts);
462 #else /* !__HAVE_TIMECOUNTER */
463 			time.tv_sec++;
464 #endif /* !__HAVE_TIMECOUNTER */
465 		}
466 		callout_init(&arptimer_ch, 0);
467 		callout_reset(&arptimer_ch, hz, arptimer, NULL);
468 	}
469 
470 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
471 		if (req != RTM_ADD)
472 			return;
473 
474 		/*
475 		 * linklayers with particular link MTU limitation.
476 		 */
477 		switch(ifp->if_type) {
478 #if NFDDI > 0
479 		case IFT_FDDI:
480 			if (ifp->if_mtu > FDDIIPMTU)
481 				rt->rt_rmx.rmx_mtu = FDDIIPMTU;
482 			break;
483 #endif
484 #if NARC > 0
485 		case IFT_ARCNET:
486 		    {
487 			int arcipifmtu;
488 
489 			if (ifp->if_flags & IFF_LINK0)
490 				arcipifmtu = arc_ipmtu;
491 			else
492 				arcipifmtu = ARCMTU;
493 			if (ifp->if_mtu > arcipifmtu)
494 				rt->rt_rmx.rmx_mtu = arcipifmtu;
495 			break;
496 		    }
497 #endif
498 		}
499 		return;
500 	}
501 
502 	ARP_LOCK(1);		/* we may already be locked here. */
503 
504 	switch (req) {
505 
506 	case RTM_SETGATE:
507 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
508 		break;
509 	case RTM_ADD:
510 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
511 		if (rt->rt_flags & RTF_CLONING) {
512 			/*
513 			 * Give this route an expiration time, even though
514 			 * it's a "permanent" route, so that routes cloned
515 			 * from it do not need their expiration time set.
516 			 */
517 			rt->rt_expire = time_second;
518 			/*
519 			 * linklayers with particular link MTU limitation.
520 			 */
521 			switch (ifp->if_type) {
522 #if NFDDI > 0
523 			case IFT_FDDI:
524 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
525 				    (rt->rt_rmx.rmx_mtu > FDDIIPMTU ||
526 				     (rt->rt_rmx.rmx_mtu == 0 &&
527 				      ifp->if_mtu > FDDIIPMTU)))
528 					rt->rt_rmx.rmx_mtu = FDDIIPMTU;
529 				break;
530 #endif
531 #if NARC > 0
532 			case IFT_ARCNET:
533 			    {
534 				int arcipifmtu;
535 				if (ifp->if_flags & IFF_LINK0)
536 					arcipifmtu = arc_ipmtu;
537 				else
538 					arcipifmtu = ARCMTU;
539 
540 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
541 				    (rt->rt_rmx.rmx_mtu > arcipifmtu ||
542 				     (rt->rt_rmx.rmx_mtu == 0 &&
543 				      ifp->if_mtu > arcipifmtu)))
544 					rt->rt_rmx.rmx_mtu = arcipifmtu;
545 				break;
546 			    }
547 #endif
548 			}
549 			break;
550 		}
551 		/* Announce a new entry if requested. */
552 		if (rt->rt_flags & RTF_ANNOUNCE)
553 			arprequest(ifp,
554 			    &satocsin(rt_getkey(rt))->sin_addr,
555 			    &satocsin(rt_getkey(rt))->sin_addr,
556 			    CLLADDR(satocsdl(gate)));
557 		/*FALLTHROUGH*/
558 	case RTM_RESOLVE:
559 		if (gate->sa_family != AF_LINK ||
560 		    gate->sa_len < sockaddr_dl_measure(namelen, addrlen)) {
561 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
562 			break;
563 		}
564 		satosdl(gate)->sdl_type = ifp->if_type;
565 		satosdl(gate)->sdl_index = ifp->if_index;
566 		if (la != 0)
567 			break; /* This happens on a route change */
568 		/*
569 		 * Case 2:  This route may come from cloning, or a manual route
570 		 * add with a LL address.
571 		 */
572 		switch (satocsdl(gate)->sdl_type) {
573 #if NTOKEN > 0
574 		case IFT_ISO88025:
575 			allocsize = sizeof(*la) + sizeof(struct token_rif);
576 			break;
577 #endif /* NTOKEN > 0 */
578 		default:
579 			allocsize = sizeof(*la);
580 		}
581 		R_Malloc(la, struct llinfo_arp *, allocsize);
582 		rt->rt_llinfo = (void *)la;
583 		if (la == 0) {
584 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
585 			break;
586 		}
587 		arp_inuse++, arp_allocated++;
588 		Bzero(la, allocsize);
589 		la->la_rt = rt;
590 		rt->rt_flags |= RTF_LLINFO;
591 		LIST_INSERT_HEAD(&llinfo_arp, la, la_list);
592 
593 		INADDR_TO_IA(satocsin(rt_getkey(rt))->sin_addr, ia);
594 		while (ia && ia->ia_ifp != ifp)
595 			NEXT_IA_WITH_SAME_ADDR(ia);
596 		if (ia) {
597 			/*
598 			 * This test used to be
599 			 *	if (lo0ifp->if_flags & IFF_UP)
600 			 * It allowed local traffic to be forced through
601 			 * the hardware by configuring the loopback down.
602 			 * However, it causes problems during network
603 			 * configuration for boards that can't receive
604 			 * packets they send.  It is now necessary to clear
605 			 * "useloopback" and remove the route to force
606 			 * traffic out to the hardware.
607 			 *
608 			 * In 4.4BSD, the above "if" statement checked
609 			 * rt->rt_ifa against rt_getkey(rt).  It was changed
610 			 * to the current form so that we can provide a
611 			 * better support for multiple IPv4 addresses on a
612 			 * interface.
613 			 */
614 			rt->rt_expire = 0;
615 			(void)sockaddr_dl_setaddr(satosdl(gate), gate->sa_len,
616 			    CLLADDR(ifp->if_sadl), ifp->if_addrlen);
617 			if (useloopback)
618 				ifp = rt->rt_ifp = lo0ifp;
619 			/*
620 			 * make sure to set rt->rt_ifa to the interface
621 			 * address we are using, otherwise we will have trouble
622 			 * with source address selection.
623 			 */
624 			ifa = &ia->ia_ifa;
625 			if (ifa != rt->rt_ifa)
626 				rt_replace_ifa(rt, ifa);
627 		}
628 		break;
629 
630 	case RTM_DELETE:
631 		if (la == 0)
632 			break;
633 		arp_inuse--;
634 		LIST_REMOVE(la, la_list);
635 		rt->rt_llinfo = 0;
636 		rt->rt_flags &= ~RTF_LLINFO;
637 
638 		s = splnet();
639 		mold = la->la_hold;
640 		la->la_hold = 0;
641 		splx(s);
642 
643 		if (mold)
644 			m_freem(mold);
645 
646 		Free((void *)la);
647 	}
648 	ARP_UNLOCK();
649 }
650 
651 /*
652  * Broadcast an ARP request. Caller specifies:
653  *	- arp header source ip address
654  *	- arp header target ip address
655  *	- arp header source ethernet address
656  */
657 void
658 arprequest(struct ifnet *ifp,
659     const struct in_addr *sip, const struct in_addr *tip,
660     const u_int8_t *enaddr)
661 {
662 	struct mbuf *m;
663 	struct arphdr *ah;
664 	struct sockaddr sa;
665 
666 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
667 		return;
668 	MCLAIM(m, &arpdomain.dom_mowner);
669 	switch (ifp->if_type) {
670 	case IFT_IEEE1394:
671 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
672 		    ifp->if_addrlen;
673 		break;
674 	default:
675 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
676 		    2 * ifp->if_addrlen;
677 		break;
678 	}
679 	m->m_pkthdr.len = m->m_len;
680 	MH_ALIGN(m, m->m_len);
681 	ah = mtod(m, struct arphdr *);
682 	bzero((void *)ah, m->m_len);
683 	switch (ifp->if_type) {
684 	case IFT_IEEE1394:	/* RFC2734 */
685 		/* fill it now for ar_tpa computation */
686 		ah->ar_hrd = htons(ARPHRD_IEEE1394);
687 		break;
688 	default:
689 		/* ifp->if_output will fill ar_hrd */
690 		break;
691 	}
692 	ah->ar_pro = htons(ETHERTYPE_IP);
693 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
694 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
695 	ah->ar_op = htons(ARPOP_REQUEST);
696 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
697 	memcpy(ar_spa(ah), sip, ah->ar_pln);
698 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
699 	sa.sa_family = AF_ARP;
700 	sa.sa_len = 2;
701 	m->m_flags |= M_BCAST;
702 	arpstat.as_sndtotal++;
703 	arpstat.as_sndrequest++;
704 	(*ifp->if_output)(ifp, m, &sa, NULL);
705 }
706 
707 /*
708  * Resolve an IP address into an ethernet address.  If success,
709  * desten is filled in.  If there is no entry in arptab,
710  * set one up and broadcast a request for the IP address.
711  * Hold onto this mbuf and resend it once the address
712  * is finally resolved.  A return value of 1 indicates
713  * that desten has been filled in and the packet should be sent
714  * normally; a 0 return indicates that the packet has been
715  * taken over here, either now or for later transmission.
716  */
717 int
718 arpresolve(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
719     const struct sockaddr *dst, u_char *desten)
720 {
721 	struct llinfo_arp *la;
722 	const struct sockaddr_dl *sdl;
723 	struct mbuf *mold;
724 	int s;
725 
726 	if ((la = arplookup1(m, &satocsin(dst)->sin_addr, 1, 0, rt)) != NULL)
727 		rt = la->la_rt;
728 
729 	if (la == 0 || rt == 0) {
730 		arpstat.as_allocfail++;
731 		log(LOG_DEBUG,
732 		    "arpresolve: can't allocate llinfo on %s for %s\n",
733 		    ifp->if_xname, in_fmtaddr(satocsin(dst)->sin_addr));
734 		m_freem(m);
735 		return (0);
736 	}
737 	sdl = satocsdl(rt->rt_gateway);
738 	/*
739 	 * Check the address family and length is valid, the address
740 	 * is resolved; otherwise, try to resolve.
741 	 */
742 	if ((rt->rt_expire == 0 || rt->rt_expire > time_second) &&
743 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
744 		bcopy(CLLADDR(sdl), desten,
745 		    min(sdl->sdl_alen, ifp->if_addrlen));
746 		rt->rt_pksent = time_second; /* Time for last pkt sent */
747 		return 1;
748 	}
749 	/*
750 	 * There is an arptab entry, but no ethernet address
751 	 * response yet.  Replace the held mbuf with this
752 	 * latest one.
753 	 */
754 
755 	arpstat.as_dfrtotal++;
756 	s = splnet();
757 	mold = la->la_hold;
758 	la->la_hold = m;
759 	splx(s);
760 
761 	if (mold) {
762 		arpstat.as_dfrdropped++;
763 		m_freem(mold);
764 	}
765 
766 	/*
767 	 * Re-send the ARP request when appropriate.
768 	 */
769 #ifdef	DIAGNOSTIC
770 	if (rt->rt_expire == 0) {
771 		/* This should never happen. (Should it? -gwr) */
772 		printf("arpresolve: unresolved and rt_expire == 0\n");
773 		/* Set expiration time to now (expired). */
774 		rt->rt_expire = time_second;
775 	}
776 #endif
777 	if (rt->rt_expire) {
778 		rt->rt_flags &= ~RTF_REJECT;
779 		if (la->la_asked == 0 || rt->rt_expire != time_second) {
780 			rt->rt_expire = time_second;
781 			if (la->la_asked++ < arp_maxtries)
782 				arprequest(ifp,
783 				    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
784 				    &satocsin(dst)->sin_addr,
785 #if NCARP > 0
786 				    (rt->rt_ifp->if_type == IFT_CARP) ?
787 				    CLLADDR(rt->rt_ifp->if_sadl):
788 #endif
789 				    CLLADDR(ifp->if_sadl));
790 			else {
791 				rt->rt_flags |= RTF_REJECT;
792 				rt->rt_expire += arpt_down;
793 				la->la_asked = 0;
794 			}
795 		}
796 	}
797 	return (0);
798 }
799 
800 /*
801  * Common length and type checks are done here,
802  * then the protocol-specific routine is called.
803  */
804 void
805 arpintr(void)
806 {
807 	struct mbuf *m;
808 	struct arphdr *ar;
809 	int s;
810 	int arplen;
811 
812 	while (arpintrq.ifq_head) {
813 		s = splnet();
814 		IF_DEQUEUE(&arpintrq, m);
815 		splx(s);
816 		if (m == 0 || (m->m_flags & M_PKTHDR) == 0)
817 			panic("arpintr");
818 
819 		MCLAIM(m, &arpdomain.dom_mowner);
820 		arpstat.as_rcvtotal++;
821 
822 		/*
823 		 * First, make sure we have at least struct arphdr.
824 		 */
825 		if (m->m_len < sizeof(struct arphdr) ||
826 		    (ar = mtod(m, struct arphdr *)) == NULL)
827 			goto badlen;
828 
829 		switch (m->m_pkthdr.rcvif->if_type) {
830 		case IFT_IEEE1394:
831 			arplen = sizeof(struct arphdr) +
832 			    ar->ar_hln + 2 * ar->ar_pln;
833 			break;
834 		default:
835 			arplen = sizeof(struct arphdr) +
836 			    2 * ar->ar_hln + 2 * ar->ar_pln;
837 			break;
838 		}
839 
840 		if (/* XXX ntohs(ar->ar_hrd) == ARPHRD_ETHER && */
841 		    m->m_len >= arplen)
842 			switch (ntohs(ar->ar_pro)) {
843 			case ETHERTYPE_IP:
844 			case ETHERTYPE_IPTRAILERS:
845 				in_arpinput(m);
846 				continue;
847 			default:
848 				arpstat.as_rcvbadproto++;
849 			}
850 		else {
851 badlen:
852 			arpstat.as_rcvbadlen++;
853 		}
854 		m_freem(m);
855 	}
856 }
857 
858 /*
859  * ARP for Internet protocols on 10 Mb/s Ethernet.
860  * Algorithm is that given in RFC 826.
861  * In addition, a sanity check is performed on the sender
862  * protocol address, to catch impersonators.
863  * We no longer handle negotiations for use of trailer protocol:
864  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
865  * along with IP replies if we wanted trailers sent to us,
866  * and also sent them in response to IP replies.
867  * This allowed either end to announce the desire to receive
868  * trailer packets.
869  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
870  * but formerly didn't normally send requests.
871  */
872 static void
873 in_arpinput(struct mbuf *m)
874 {
875 	struct arphdr *ah;
876 	struct ifnet *ifp = m->m_pkthdr.rcvif;
877 	struct llinfo_arp *la = 0;
878 	struct rtentry  *rt;
879 	struct in_ifaddr *ia;
880 #if NBRIDGE > 0
881 	struct in_ifaddr *bridge_ia = NULL;
882 #endif
883 #if NCARP > 0
884 	u_int32_t count = 0, index = 0;
885 #endif
886 	struct sockaddr_dl *sdl;
887 	struct sockaddr sa;
888 	struct in_addr isaddr, itaddr, myaddr;
889 	int op;
890 	struct mbuf *mold;
891 	void *tha;
892 	int s;
893 
894 	if (__predict_false(m_makewritable(&m, 0, m->m_pkthdr.len, M_DONTWAIT)))
895 		goto out;
896 	ah = mtod(m, struct arphdr *);
897 	op = ntohs(ah->ar_op);
898 
899 	/*
900 	 * Fix up ah->ar_hrd if necessary, before using ar_tha() or
901 	 * ar_tpa().
902 	 */
903 	switch (ifp->if_type) {
904 	case IFT_IEEE1394:
905 		if (ntohs(ah->ar_hrd) == ARPHRD_IEEE1394)
906 			;
907 		else {
908 			/* XXX this is to make sure we compute ar_tha right */
909 			/* XXX check ar_hrd more strictly? */
910 			ah->ar_hrd = htons(ARPHRD_IEEE1394);
911 		}
912 		break;
913 	default:
914 		/* XXX check ar_hrd? */
915 		break;
916 	}
917 
918 	memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
919 	memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
920 
921 	if (m->m_flags & (M_BCAST|M_MCAST))
922 		arpstat.as_rcvmcast++;
923 
924 	/*
925 	 * If the target IP address is zero, ignore the packet.
926 	 * This prevents the code below from tring to answer
927 	 * when we are using IP address zero (booting).
928 	 */
929 	if (in_nullhost(itaddr)) {
930 		arpstat.as_rcvzerotpa++;
931 		goto out;
932 	}
933 
934 	/*
935 	 * If the source IP address is zero, this is most likely a
936 	 * confused host trying to use IP address zero. (Windoze?)
937 	 * XXX: Should we bother trying to reply to these?
938 	 */
939 	if (in_nullhost(isaddr)) {
940 		arpstat.as_rcvzerospa++;
941 		goto out;
942 	}
943 
944 	/*
945 	 * Search for a matching interface address
946 	 * or any address on the interface to use
947 	 * as a dummy address in the rest of this function
948 	 */
949 
950 	INADDR_TO_IA(itaddr, ia);
951 	while (ia != NULL) {
952 #if NCARP > 0
953 		if (ia->ia_ifp->if_type == IFT_CARP &&
954 		    ((ia->ia_ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
955 		    (IFF_UP|IFF_RUNNING))) {
956 			index++;
957 			if (ia->ia_ifp == m->m_pkthdr.rcvif &&
958 			    carp_iamatch(ia, ar_sha(ah),
959 			    &count, index)) {
960 				break;
961 				}
962 		} else
963 #endif
964 			    if (ia->ia_ifp == m->m_pkthdr.rcvif)
965 				break;
966 #if NBRIDGE > 0
967 		/*
968 		 * If the interface we received the packet on
969 		 * is part of a bridge, check to see if we need
970 		 * to "bridge" the packet to ourselves at this
971 		 * layer.  Note we still prefer a perfect match,
972 		 * but allow this weaker match if necessary.
973 		 */
974 		if (m->m_pkthdr.rcvif->if_bridge != NULL &&
975 		    m->m_pkthdr.rcvif->if_bridge == ia->ia_ifp->if_bridge)
976 			bridge_ia = ia;
977 #endif /* NBRIDGE > 0 */
978 
979 		NEXT_IA_WITH_SAME_ADDR(ia);
980 	}
981 
982 #if NBRIDGE > 0
983 	if (ia == NULL && bridge_ia != NULL) {
984 		ia = bridge_ia;
985 		ifp = bridge_ia->ia_ifp;
986 	}
987 #endif
988 
989 	if (ia == NULL) {
990 		INADDR_TO_IA(isaddr, ia);
991 		while ((ia != NULL) && ia->ia_ifp != m->m_pkthdr.rcvif)
992 			NEXT_IA_WITH_SAME_ADDR(ia);
993 
994 		if (ia == NULL) {
995 			IFP_TO_IA(ifp, ia);
996 			if (ia == NULL) {
997 				arpstat.as_rcvnoint++;
998 				goto out;
999 			}
1000 		}
1001 	}
1002 
1003 	myaddr = ia->ia_addr.sin_addr;
1004 
1005 	/* XXX checks for bridge case? */
1006 	if (!memcmp(ar_sha(ah), CLLADDR(ifp->if_sadl), ifp->if_addrlen)) {
1007 		arpstat.as_rcvlocalsha++;
1008 		goto out;	/* it's from me, ignore it. */
1009 	}
1010 
1011 	/* XXX checks for bridge case? */
1012 	if (!memcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1013 		arpstat.as_rcvbcastsha++;
1014 		log(LOG_ERR,
1015 		    "%s: arp: link address is broadcast for IP address %s!\n",
1016 		    ifp->if_xname, in_fmtaddr(isaddr));
1017 		goto out;
1018 	}
1019 
1020 	if (in_hosteq(isaddr, myaddr)) {
1021 		arpstat.as_rcvlocalspa++;
1022 		log(LOG_ERR,
1023 		   "duplicate IP address %s sent from link address %s\n",
1024 		   in_fmtaddr(isaddr), lla_snprintf(ar_sha(ah), ah->ar_hln));
1025 		itaddr = myaddr;
1026 		goto reply;
1027 	}
1028 	la = arplookup(m, &isaddr, in_hosteq(itaddr, myaddr), 0);
1029 	if (la && (rt = la->la_rt) && (sdl = satosdl(rt->rt_gateway))) {
1030 		if (sdl->sdl_alen &&
1031 		    memcmp(ar_sha(ah), CLLADDR(sdl), sdl->sdl_alen)) {
1032 			if (rt->rt_flags & RTF_STATIC) {
1033 				arpstat.as_rcvoverperm++;
1034 				log(LOG_INFO,
1035 				    "%s tried to overwrite permanent arp info"
1036 				    " for %s\n",
1037 				    lla_snprintf(ar_sha(ah), ah->ar_hln),
1038 				    in_fmtaddr(isaddr));
1039 				goto out;
1040 			} else if (rt->rt_ifp != ifp) {
1041 				arpstat.as_rcvoverint++;
1042 				log(LOG_INFO,
1043 				    "%s on %s tried to overwrite "
1044 				    "arp info for %s on %s\n",
1045 				    lla_snprintf(ar_sha(ah), ah->ar_hln),
1046 				    ifp->if_xname, in_fmtaddr(isaddr),
1047 				    rt->rt_ifp->if_xname);
1048 				    goto out;
1049 			} else {
1050 				arpstat.as_rcvover++;
1051 				log(LOG_INFO,
1052 				    "arp info overwritten for %s by %s\n",
1053 				    in_fmtaddr(isaddr),
1054 				    lla_snprintf(ar_sha(ah), ah->ar_hln));
1055 			}
1056 		}
1057 		/*
1058 		 * sanity check for the address length.
1059 		 * XXX this does not work for protocols with variable address
1060 		 * length. -is
1061 		 */
1062 		if (sdl->sdl_alen &&
1063 		    sdl->sdl_alen != ah->ar_hln) {
1064 			arpstat.as_rcvlenchg++;
1065 			log(LOG_WARNING,
1066 			    "arp from %s: new addr len %d, was %d",
1067 			    in_fmtaddr(isaddr), ah->ar_hln, sdl->sdl_alen);
1068 		}
1069 		if (ifp->if_addrlen != ah->ar_hln) {
1070 			arpstat.as_rcvbadlen++;
1071 			log(LOG_WARNING,
1072 			    "arp from %s: addr len: new %d, i/f %d (ignored)",
1073 			    in_fmtaddr(isaddr), ah->ar_hln,
1074 			    ifp->if_addrlen);
1075 			goto reply;
1076 		}
1077 #if NTOKEN > 0
1078 		/*
1079 		 * XXX uses m_data and assumes the complete answer including
1080 		 * XXX token-ring headers is in the same buf
1081 		 */
1082 		if (ifp->if_type == IFT_ISO88025) {
1083 			struct token_header *trh;
1084 
1085 			trh = (struct token_header *)M_TRHSTART(m);
1086 			if (trh->token_shost[0] & TOKEN_RI_PRESENT) {
1087 				struct token_rif	*rif;
1088 				size_t	riflen;
1089 
1090 				rif = TOKEN_RIF(trh);
1091 				riflen = (ntohs(rif->tr_rcf) &
1092 				    TOKEN_RCF_LEN_MASK) >> 8;
1093 
1094 				if (riflen > 2 &&
1095 				    riflen < sizeof(struct token_rif) &&
1096 				    (riflen & 1) == 0) {
1097 					rif->tr_rcf ^= htons(TOKEN_RCF_DIRECTION);
1098 					rif->tr_rcf &= htons(~TOKEN_RCF_BROADCAST_MASK);
1099 					bcopy(rif, TOKEN_RIF(la), riflen);
1100 				}
1101 			}
1102 		}
1103 #endif /* NTOKEN > 0 */
1104 		(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, ar_sha(ah),
1105 		    ah->ar_hln);
1106 		if (rt->rt_expire)
1107 			rt->rt_expire = time_second + arpt_keep;
1108 		rt->rt_flags &= ~RTF_REJECT;
1109 		la->la_asked = 0;
1110 
1111 		s = splnet();
1112 		mold = la->la_hold;
1113 		la->la_hold = 0;
1114 		splx(s);
1115 
1116 		if (mold) {
1117 			arpstat.as_dfrsent++;
1118 			(*ifp->if_output)(ifp, mold, rt_getkey(rt), rt);
1119 		}
1120 	}
1121 reply:
1122 	if (op != ARPOP_REQUEST) {
1123 		if (op == ARPOP_REPLY)
1124 			arpstat.as_rcvreply++;
1125 	out:
1126 		m_freem(m);
1127 		return;
1128 	}
1129 	arpstat.as_rcvrequest++;
1130 	if (in_hosteq(itaddr, myaddr)) {
1131 		/* I am the target */
1132 		tha = ar_tha(ah);
1133 		if (tha)
1134 			memcpy(tha, ar_sha(ah), ah->ar_hln);
1135 		memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
1136 	} else {
1137 		la = arplookup(m, &itaddr, 0, SIN_PROXY);
1138 		if (la == 0)
1139 			goto out;
1140 		rt = la->la_rt;
1141 		if (rt->rt_ifp->if_type == IFT_CARP &&
1142 		    m->m_pkthdr.rcvif->if_type != IFT_CARP)
1143 			goto out;
1144 		tha = ar_tha(ah);
1145 		if (tha)
1146 			memcpy(tha, ar_sha(ah), ah->ar_hln);
1147 		sdl = satosdl(rt->rt_gateway);
1148 		memcpy(ar_sha(ah), CLLADDR(sdl), ah->ar_hln);
1149 	}
1150 
1151 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1152 	memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1153 	ah->ar_op = htons(ARPOP_REPLY);
1154 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1155 	switch (ifp->if_type) {
1156 	case IFT_IEEE1394:
1157 		/*
1158 		 * ieee1394 arp reply is broadcast
1159 		 */
1160 		m->m_flags &= ~M_MCAST;
1161 		m->m_flags |= M_BCAST;
1162 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + ah->ar_hln;
1163 		break;
1164 
1165 	default:
1166 		m->m_flags &= ~(M_BCAST|M_MCAST); /* never reply by broadcast */
1167 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1168 		break;
1169 	}
1170 	m->m_pkthdr.len = m->m_len;
1171 	sa.sa_family = AF_ARP;
1172 	sa.sa_len = 2;
1173 	arpstat.as_sndtotal++;
1174 	arpstat.as_sndreply++;
1175 	(*ifp->if_output)(ifp, m, &sa, (struct rtentry *)0);
1176 	return;
1177 }
1178 
1179 /*
1180  * Free an arp entry.
1181  */
1182 static void arptfree(struct llinfo_arp *la)
1183 {
1184 	struct rtentry *rt = la->la_rt;
1185 	struct sockaddr_dl *sdl;
1186 
1187 	ARP_LOCK_CHECK();
1188 
1189 	if (rt == 0)
1190 		panic("arptfree");
1191 	if (rt->rt_refcnt > 0 && (sdl = satosdl(rt->rt_gateway)) &&
1192 	    sdl->sdl_family == AF_LINK) {
1193 		sdl->sdl_alen = 0;
1194 		la->la_asked = 0;
1195 		rt->rt_flags &= ~RTF_REJECT;
1196 		return;
1197 	}
1198 	rtrequest(RTM_DELETE, rt_getkey(rt), NULL, rt_mask(rt), 0, NULL);
1199 }
1200 
1201 static struct llinfo_arp *
1202 arplookup(struct mbuf *m, const struct in_addr *addr, int create, int proxy)
1203 {
1204 	return arplookup1(m, addr, create, proxy, NULL);
1205 }
1206 
1207 /*
1208  * Lookup or enter a new address in arptab.
1209  */
1210 static struct llinfo_arp *
1211 arplookup1(struct mbuf *m, const struct in_addr *addr, int create, int proxy,
1212     struct rtentry *rt0)
1213 {
1214 	struct arphdr *ah;
1215 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1216 	struct rtentry *rt;
1217 	static struct sockaddr_inarp sin;
1218 	const char *why = 0;
1219 
1220 	ah = mtod(m, struct arphdr *);
1221 	if (rt0 == NULL) {
1222 		sin.sin_len = sizeof(sin);
1223 		sin.sin_family = AF_INET;
1224 		sin.sin_addr = *addr;
1225 		sin.sin_other = proxy ? SIN_PROXY : 0;
1226 		rt = rtalloc1(sintosa(&sin), create);
1227 		if (rt == NULL)
1228 			return (NULL);
1229 		rt->rt_refcnt--;
1230 	} else
1231 		rt = rt0;
1232 
1233 #define	IS_LLINFO(__rt)							  \
1234 	(((__rt)->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) == RTF_LLINFO && \
1235 	 (__rt)->rt_gateway->sa_family == AF_LINK)
1236 
1237 
1238 	if (IS_LLINFO(rt))
1239 		return ((struct llinfo_arp *)rt->rt_llinfo);
1240 
1241 	if (create) {
1242 		if (rt->rt_flags & RTF_GATEWAY)
1243 			why = "host is not on local network";
1244 		else if ((rt->rt_flags & RTF_LLINFO) == 0) {
1245 			arpstat.as_allocfail++;
1246 			why = "could not allocate llinfo";
1247 		} else
1248 			why = "gateway route is not ours";
1249 		log(LOG_DEBUG, "arplookup: unable to enter address"
1250 		    " for %s@%s on %s (%s)\n",
1251 		    in_fmtaddr(*addr), lla_snprintf(ar_sha(ah), ah->ar_hln),
1252 		    (ifp) ? ifp->if_xname : 0, why);
1253 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_CLONED) != 0) {
1254 			rtrequest(RTM_DELETE, rt_getkey(rt),
1255 		    	    rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
1256 		}
1257 	}
1258 	return (0);
1259 }
1260 
1261 int
1262 arpioctl(u_long cmd, void *data)
1263 {
1264 
1265 	return (EOPNOTSUPP);
1266 }
1267 
1268 void
1269 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1270 {
1271 	struct in_addr *ip;
1272 
1273 	/*
1274 	 * Warn the user if another station has this IP address,
1275 	 * but only if the interface IP address is not zero.
1276 	 */
1277 	ip = &IA_SIN(ifa)->sin_addr;
1278 	if (!in_nullhost(*ip))
1279 		arprequest(ifp, ip, ip, CLLADDR(ifp->if_sadl));
1280 
1281 	ifa->ifa_rtrequest = arp_rtrequest;
1282 	ifa->ifa_flags |= RTF_CLONING;
1283 }
1284 
1285 /*
1286  * Called from 10 Mb/s Ethernet interrupt handlers
1287  * when ether packet type ETHERTYPE_REVARP
1288  * is received.  Common length and type checks are done here,
1289  * then the protocol-specific routine is called.
1290  */
1291 void
1292 revarpinput(struct mbuf *m)
1293 {
1294 	struct arphdr *ar;
1295 
1296 	if (m->m_len < sizeof(struct arphdr))
1297 		goto out;
1298 	ar = mtod(m, struct arphdr *);
1299 #if 0 /* XXX I don't think we need this... and it will prevent other LL */
1300 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER)
1301 		goto out;
1302 #endif
1303 	if (m->m_len < sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln))
1304 		goto out;
1305 	switch (ntohs(ar->ar_pro)) {
1306 	case ETHERTYPE_IP:
1307 	case ETHERTYPE_IPTRAILERS:
1308 		in_revarpinput(m);
1309 		return;
1310 
1311 	default:
1312 		break;
1313 	}
1314 out:
1315 	m_freem(m);
1316 }
1317 
1318 /*
1319  * RARP for Internet protocols on 10 Mb/s Ethernet.
1320  * Algorithm is that given in RFC 903.
1321  * We are only using for bootstrap purposes to get an ip address for one of
1322  * our interfaces.  Thus we support no user-interface.
1323  *
1324  * Since the contents of the RARP reply are specific to the interface that
1325  * sent the request, this code must ensure that they are properly associated.
1326  *
1327  * Note: also supports ARP via RARP packets, per the RFC.
1328  */
1329 void
1330 in_revarpinput(struct mbuf *m)
1331 {
1332 	struct ifnet *ifp;
1333 	struct arphdr *ah;
1334 	void *tha;
1335 	int op;
1336 
1337 	ah = mtod(m, struct arphdr *);
1338 	op = ntohs(ah->ar_op);
1339 
1340 	switch (m->m_pkthdr.rcvif->if_type) {
1341 	case IFT_IEEE1394:
1342 		/* ARP without target hardware address is not supported */
1343 		goto out;
1344 	default:
1345 		break;
1346 	}
1347 
1348 	switch (op) {
1349 	case ARPOP_REQUEST:
1350 	case ARPOP_REPLY:	/* per RFC */
1351 		in_arpinput(m);
1352 		return;
1353 	case ARPOP_REVREPLY:
1354 		break;
1355 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
1356 	default:
1357 		goto out;
1358 	}
1359 	if (!revarp_in_progress)
1360 		goto out;
1361 	ifp = m->m_pkthdr.rcvif;
1362 	if (ifp != myip_ifp) /* !same interface */
1363 		goto out;
1364 	if (myip_initialized)
1365 		goto wake;
1366 	tha = ar_tha(ah);
1367 	KASSERT(tha);
1368 	if (bcmp(tha, CLLADDR(ifp->if_sadl), ifp->if_sadl->sdl_alen))
1369 		goto out;
1370 	memcpy(&srv_ip, ar_spa(ah), sizeof(srv_ip));
1371 	memcpy(&myip, ar_tpa(ah), sizeof(myip));
1372 	myip_initialized = 1;
1373 wake:	/* Do wakeup every time in case it was missed. */
1374 	wakeup((void *)&myip);
1375 
1376 out:
1377 	m_freem(m);
1378 }
1379 
1380 /*
1381  * Send a RARP request for the ip address of the specified interface.
1382  * The request should be RFC 903-compliant.
1383  */
1384 void
1385 revarprequest(struct ifnet *ifp)
1386 {
1387 	struct sockaddr sa;
1388 	struct mbuf *m;
1389 	struct arphdr *ah;
1390 	void *tha;
1391 
1392 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
1393 		return;
1394 	MCLAIM(m, &arpdomain.dom_mowner);
1395 	m->m_len = sizeof(*ah) + 2*sizeof(struct in_addr) +
1396 	    2*ifp->if_addrlen;
1397 	m->m_pkthdr.len = m->m_len;
1398 	MH_ALIGN(m, m->m_len);
1399 	ah = mtod(m, struct arphdr *);
1400 	bzero((void *)ah, m->m_len);
1401 	ah->ar_pro = htons(ETHERTYPE_IP);
1402 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
1403 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
1404 	ah->ar_op = htons(ARPOP_REVREQUEST);
1405 
1406 	memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
1407 	tha = ar_tha(ah);
1408 	KASSERT(tha);
1409 	bcopy(CLLADDR(ifp->if_sadl), tha, ah->ar_hln);
1410 
1411 	sa.sa_family = AF_ARP;
1412 	sa.sa_len = 2;
1413 	m->m_flags |= M_BCAST;
1414 	(*ifp->if_output)(ifp, m, &sa, (struct rtentry *)0);
1415 
1416 }
1417 
1418 /*
1419  * RARP for the ip address of the specified interface, but also
1420  * save the ip address of the server that sent the answer.
1421  * Timeout if no response is received.
1422  */
1423 int
1424 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
1425     struct in_addr *clnt_in)
1426 {
1427 	int result, count = 20;
1428 
1429 	myip_initialized = 0;
1430 	myip_ifp = ifp;
1431 
1432 	revarp_in_progress = 1;
1433 	while (count--) {
1434 		revarprequest(ifp);
1435 		result = tsleep((void *)&myip, PSOCK, "revarp", hz/2);
1436 		if (result != EWOULDBLOCK)
1437 			break;
1438 	}
1439 	revarp_in_progress = 0;
1440 
1441 	if (!myip_initialized)
1442 		return ENETUNREACH;
1443 
1444 	bcopy((void *)&srv_ip, serv_in, sizeof(*serv_in));
1445 	bcopy((void *)&myip, clnt_in, sizeof(*clnt_in));
1446 	return 0;
1447 }
1448 
1449 
1450 
1451 #ifdef DDB
1452 
1453 #include <machine/db_machdep.h>
1454 #include <ddb/db_interface.h>
1455 #include <ddb/db_output.h>
1456 
1457 static void
1458 db_print_sa(const struct sockaddr *sa)
1459 {
1460 	int len;
1461 	const u_char *p;
1462 
1463 	if (sa == 0) {
1464 		db_printf("[NULL]");
1465 		return;
1466 	}
1467 
1468 	p = (const u_char *)sa;
1469 	len = sa->sa_len;
1470 	db_printf("[");
1471 	while (len > 0) {
1472 		db_printf("%d", *p);
1473 		p++; len--;
1474 		if (len) db_printf(",");
1475 	}
1476 	db_printf("]\n");
1477 }
1478 
1479 static void
1480 db_print_ifa(struct ifaddr *ifa)
1481 {
1482 	if (ifa == 0)
1483 		return;
1484 	db_printf("  ifa_addr=");
1485 	db_print_sa(ifa->ifa_addr);
1486 	db_printf("  ifa_dsta=");
1487 	db_print_sa(ifa->ifa_dstaddr);
1488 	db_printf("  ifa_mask=");
1489 	db_print_sa(ifa->ifa_netmask);
1490 	db_printf("  flags=0x%x,refcnt=%d,metric=%d\n",
1491 			  ifa->ifa_flags,
1492 			  ifa->ifa_refcnt,
1493 			  ifa->ifa_metric);
1494 }
1495 
1496 static void
1497 db_print_llinfo(void *li)
1498 {
1499 	struct llinfo_arp *la;
1500 
1501 	if (li == 0)
1502 		return;
1503 	la = (struct llinfo_arp *)li;
1504 	db_printf("  la_rt=%p la_hold=%p, la_asked=0x%lx\n",
1505 			  la->la_rt, la->la_hold, la->la_asked);
1506 }
1507 
1508 /*
1509  * Function to pass to rt_walktree().
1510  * Return non-zero error to abort walk.
1511  */
1512 static int
1513 db_show_rtentry(struct rtentry *rt, void *w)
1514 {
1515 	db_printf("rtentry=%p", rt);
1516 
1517 	db_printf(" flags=0x%x refcnt=%d use=%ld expire=%ld\n",
1518 			  rt->rt_flags, rt->rt_refcnt,
1519 			  rt->rt_use, rt->rt_expire);
1520 
1521 	db_printf(" key="); db_print_sa(rt_getkey(rt));
1522 	db_printf(" mask="); db_print_sa(rt_mask(rt));
1523 	db_printf(" gw="); db_print_sa(rt->rt_gateway);
1524 
1525 	db_printf(" ifp=%p ", rt->rt_ifp);
1526 	if (rt->rt_ifp)
1527 		db_printf("(%s)", rt->rt_ifp->if_xname);
1528 	else
1529 		db_printf("(NULL)");
1530 
1531 	db_printf(" ifa=%p\n", rt->rt_ifa);
1532 	db_print_ifa(rt->rt_ifa);
1533 
1534 	db_printf(" gwroute=%p llinfo=%p\n",
1535 			  rt->rt_gwroute, rt->rt_llinfo);
1536 	db_print_llinfo(rt->rt_llinfo);
1537 
1538 	return (0);
1539 }
1540 
1541 /*
1542  * Function to print all the route trees.
1543  * Use this from ddb:  "show arptab"
1544  */
1545 void
1546 db_show_arptab(db_expr_t addr, bool have_addr,
1547     db_expr_t count, const char *modif)
1548 {
1549 	rt_walktree(AF_INET, db_show_rtentry, NULL);
1550 }
1551 #endif
1552 
1553 SYSCTL_SETUP(sysctl_net_inet_arp_setup, "sysctl net.inet.arp subtree setup")
1554 {
1555 	const struct sysctlnode *node;
1556 
1557 	sysctl_createv(clog, 0, NULL, NULL,
1558 			CTLFLAG_PERMANENT,
1559 			CTLTYPE_NODE, "net", NULL,
1560 			NULL, 0, NULL, 0,
1561 			CTL_NET, CTL_EOL);
1562 	sysctl_createv(clog, 0, NULL, NULL,
1563 			CTLFLAG_PERMANENT,
1564 			CTLTYPE_NODE, "inet", NULL,
1565 			NULL, 0, NULL, 0,
1566 			CTL_NET, PF_INET, CTL_EOL);
1567 	sysctl_createv(clog, 0, NULL, &node,
1568 			CTLFLAG_PERMANENT,
1569 			CTLTYPE_NODE, "arp",
1570 			SYSCTL_DESCR("Address Resolution Protocol"),
1571 			NULL, 0, NULL, 0,
1572 			CTL_NET, PF_INET, CTL_CREATE, CTL_EOL);
1573 
1574 	sysctl_createv(clog, 0, NULL, NULL,
1575 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1576 			CTLTYPE_INT, "prune",
1577 			SYSCTL_DESCR("ARP cache pruning interval"),
1578 			NULL, 0, &arpt_prune, 0,
1579 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1580 
1581 	sysctl_createv(clog, 0, NULL, NULL,
1582 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1583 			CTLTYPE_INT, "keep",
1584 			SYSCTL_DESCR("Valid ARP entry lifetime"),
1585 			NULL, 0, &arpt_keep, 0,
1586 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1587 
1588 	sysctl_createv(clog, 0, NULL, NULL,
1589 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1590 			CTLTYPE_INT, "down",
1591 			SYSCTL_DESCR("Failed ARP entry lifetime"),
1592 			NULL, 0, &arpt_down, 0,
1593 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1594 
1595 	sysctl_createv(clog, 0, NULL, NULL,
1596 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1597 			CTLTYPE_INT, "refresh",
1598 			SYSCTL_DESCR("ARP entry refresh interval"),
1599 			NULL, 0, &arpt_refresh, 0,
1600 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1601 }
1602 
1603 #endif /* INET */
1604