xref: /netbsd-src/sys/netinet/if_arp.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: if_arp.c,v 1.149 2009/11/20 02:14:57 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000, 2008 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  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1988, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)if_ether.c	8.2 (Berkeley) 9/26/94
62  */
63 
64 /*
65  * Ethernet address resolution protocol.
66  * TODO:
67  *	add "inuse/lock" bit (or ref. count) along with valid bit
68  */
69 
70 #include <sys/cdefs.h>
71 __KERNEL_RCSID(0, "$NetBSD: if_arp.c,v 1.149 2009/11/20 02:14:57 christos Exp $");
72 
73 #include "opt_ddb.h"
74 #include "opt_inet.h"
75 
76 #ifdef INET
77 
78 #include "bridge.h"
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/callout.h>
83 #include <sys/malloc.h>
84 #include <sys/mbuf.h>
85 #include <sys/socket.h>
86 #include <sys/time.h>
87 #include <sys/timetc.h>
88 #include <sys/kernel.h>
89 #include <sys/errno.h>
90 #include <sys/ioctl.h>
91 #include <sys/syslog.h>
92 #include <sys/proc.h>
93 #include <sys/protosw.h>
94 #include <sys/domain.h>
95 #include <sys/sysctl.h>
96 #include <sys/socketvar.h>
97 #include <sys/percpu.h>
98 
99 #include <net/ethertypes.h>
100 #include <net/if.h>
101 #include <net/if_dl.h>
102 #include <net/if_token.h>
103 #include <net/if_types.h>
104 #include <net/if_ether.h>
105 #include <net/route.h>
106 #include <net/net_stats.h>
107 
108 #include <netinet/in.h>
109 #include <netinet/in_systm.h>
110 #include <netinet/in_var.h>
111 #include <netinet/ip.h>
112 #include <netinet/if_inarp.h>
113 
114 #include "arcnet.h"
115 #if NARCNET > 0
116 #include <net/if_arc.h>
117 #endif
118 #include "fddi.h"
119 #if NFDDI > 0
120 #include <net/if_fddi.h>
121 #endif
122 #include "token.h"
123 #include "carp.h"
124 #if NCARP > 0
125 #include <netinet/ip_carp.h>
126 #endif
127 
128 #define SIN(s) ((struct sockaddr_in *)s)
129 #define SRP(s) ((struct sockaddr_inarp *)s)
130 
131 /*
132  * ARP trailer negotiation.  Trailer protocol is not IP specific,
133  * but ARP request/response use IP addresses.
134  */
135 #define ETHERTYPE_IPTRAILERS ETHERTYPE_TRAIL
136 
137 /* timer values */
138 int	arpt_prune = (5*60*1);	/* walk list every 5 minutes */
139 int	arpt_keep = (20*60);	/* once resolved, good for 20 more minutes */
140 int	arpt_down = 20;		/* once declared down, don't send for 20 secs */
141 int	arpt_refresh = (5*60);	/* time left before refreshing */
142 #define	rt_expire rt_rmx.rmx_expire
143 #define	rt_pksent rt_rmx.rmx_pksent
144 
145 static	struct sockaddr *arp_setgate(struct rtentry *, struct sockaddr *,
146 	    const struct sockaddr *);
147 static	void arptfree(struct llinfo_arp *);
148 static	void arptimer(void *);
149 static	struct llinfo_arp *arplookup1(struct mbuf *, const struct in_addr *,
150 				      int, int, struct rtentry *);
151 static	struct llinfo_arp *arplookup(struct mbuf *, const struct in_addr *,
152 					  int, int);
153 static	void in_arpinput(struct mbuf *);
154 
155 LIST_HEAD(, llinfo_arp) llinfo_arp;
156 struct	ifqueue arpintrq = {
157 	.ifq_head = NULL,
158 	.ifq_tail = NULL,
159 	.ifq_len = 0,
160 	.ifq_maxlen = 50,
161 	.ifq_drops = 0,
162 };
163 int	arp_inuse, arp_allocated, arp_intimer;
164 int	arp_maxtries = 5;
165 int	useloopback = 1;	/* use loopback interface for local traffic */
166 int	arpinit_done = 0;
167 
168 static percpu_t *arpstat_percpu;
169 
170 #define	ARP_STAT_GETREF()	_NET_STAT_GETREF(arpstat_percpu)
171 #define	ARP_STAT_PUTREF()	_NET_STAT_PUTREF(arpstat_percpu)
172 
173 #define	ARP_STATINC(x)		_NET_STATINC(arpstat_percpu, x)
174 #define	ARP_STATADD(x, v)	_NET_STATADD(arpstat_percpu, x, v)
175 
176 struct	callout arptimer_ch;
177 
178 /* revarp state */
179 struct	in_addr myip, srv_ip;
180 int	myip_initialized = 0;
181 int	revarp_in_progress = 0;
182 struct	ifnet *myip_ifp = NULL;
183 
184 #ifdef DDB
185 static void db_print_sa(const struct sockaddr *);
186 static void db_print_ifa(struct ifaddr *);
187 static void db_print_llinfo(void *);
188 static int db_show_rtentry(struct rtentry *, void *);
189 #endif
190 
191 /*
192  * this should be elsewhere.
193  */
194 
195 static char *
196 lla_snprintf(u_int8_t *, int);
197 
198 static char *
199 lla_snprintf(u_int8_t *adrp, int len)
200 {
201 #define NUMBUFS 3
202 	static char buf[NUMBUFS][16*3];
203 	static int bnum = 0;
204 
205 	int i;
206 	char *p;
207 
208 	p = buf[bnum];
209 
210 	*p++ = hexdigits[(*adrp)>>4];
211 	*p++ = hexdigits[(*adrp++)&0xf];
212 
213 	for (i=1; i<len && i<16; i++) {
214 		*p++ = ':';
215 		*p++ = hexdigits[(*adrp)>>4];
216 		*p++ = hexdigits[(*adrp++)&0xf];
217 	}
218 
219 	*p = 0;
220 	p = buf[bnum];
221 	bnum = (bnum + 1) % NUMBUFS;
222 	return p;
223 }
224 
225 DOMAIN_DEFINE(arpdomain);	/* forward declare and add to link set */
226 
227 const struct protosw arpsw[] = {
228 	{ .pr_type = 0,
229 	  .pr_domain = &arpdomain,
230 	  .pr_protocol = 0,
231 	  .pr_flags = 0,
232 	  .pr_input = 0,
233 	  .pr_output = 0,
234 	  .pr_ctlinput = 0,
235 	  .pr_ctloutput = 0,
236 	  .pr_usrreq =  0,
237 	  .pr_init = arp_init,
238 	  .pr_fasttimo = 0,
239 	  .pr_slowtimo = 0,
240 	  .pr_drain = arp_drain,
241 	}
242 };
243 
244 
245 struct domain arpdomain = {
246 	.dom_family = PF_ARP,
247 	.dom_name = "arp",
248 	.dom_protosw = arpsw,
249 	.dom_protoswNPROTOSW = &arpsw[__arraycount(arpsw)],
250 };
251 
252 /*
253  * ARP table locking.
254  *
255  * to prevent lossage vs. the arp_drain routine (which may be called at
256  * any time, including in a device driver context), we do two things:
257  *
258  * 1) manipulation of la->la_hold is done at splnet() (for all of
259  * about two instructions).
260  *
261  * 2) manipulation of the arp table's linked list is done under the
262  * protection of the ARP_LOCK; if arp_drain() or arptimer is called
263  * while the arp table is locked, we punt and try again later.
264  */
265 
266 static int	arp_locked;
267 static inline int arp_lock_try(int);
268 static inline void arp_unlock(void);
269 
270 static inline int
271 arp_lock_try(int recurse)
272 {
273 	int s;
274 
275 	/*
276 	 * Use splvm() -- we're blocking things that would cause
277 	 * mbuf allocation.
278 	 */
279 	s = splvm();
280 	if (!recurse && arp_locked) {
281 		splx(s);
282 		return 0;
283 	}
284 	arp_locked++;
285 	splx(s);
286 	return 1;
287 }
288 
289 static inline void
290 arp_unlock(void)
291 {
292 	int s;
293 
294 	s = splvm();
295 	arp_locked--;
296 	splx(s);
297 }
298 
299 #ifdef DIAGNOSTIC
300 #define	ARP_LOCK(recurse)						\
301 do {									\
302 	if (arp_lock_try(recurse) == 0) {				\
303 		printf("%s:%d: arp already locked\n", __FILE__, __LINE__); \
304 		panic("arp_lock");					\
305 	}								\
306 } while (/*CONSTCOND*/ 0)
307 #define	ARP_LOCK_CHECK()						\
308 do {									\
309 	if (arp_locked == 0) {						\
310 		printf("%s:%d: arp lock not held\n", __FILE__, __LINE__); \
311 		panic("arp lock check");				\
312 	}								\
313 } while (/*CONSTCOND*/ 0)
314 #else
315 #define	ARP_LOCK(x)		(void) arp_lock_try(x)
316 #define	ARP_LOCK_CHECK()	/* nothing */
317 #endif
318 
319 #define	ARP_UNLOCK()		arp_unlock()
320 
321 static void sysctl_net_inet_arp_setup(struct sysctllog **);
322 
323 void
324 arp_init(void)
325 {
326 
327 	sysctl_net_inet_arp_setup(NULL);
328 	arpstat_percpu = percpu_alloc(sizeof(uint64_t) * ARP_NSTATS);
329 }
330 
331 /*
332  * ARP protocol drain routine.  Called when memory is in short supply.
333  * Called at splvm();  don't acquire softnet_lock as can be called from
334  * hardware interrupt handlers.
335  */
336 void
337 arp_drain(void)
338 {
339 	struct llinfo_arp *la, *nla;
340 	int count = 0;
341 	struct mbuf *mold;
342 
343 	KERNEL_LOCK(1, NULL);
344 
345 	if (arp_lock_try(0) == 0) {
346 		KERNEL_UNLOCK_ONE(NULL);
347 		return;
348 	}
349 
350 	for (la = LIST_FIRST(&llinfo_arp); la != NULL; la = nla) {
351 		nla = LIST_NEXT(la, la_list);
352 
353 		mold = la->la_hold;
354 		la->la_hold = 0;
355 
356 		if (mold) {
357 			m_freem(mold);
358 			count++;
359 		}
360 	}
361 	ARP_UNLOCK();
362 	ARP_STATADD(ARP_STAT_DFRDROPPED, count);
363 	KERNEL_UNLOCK_ONE(NULL);
364 }
365 
366 
367 /*
368  * Timeout routine.  Age arp_tab entries periodically.
369  */
370 /* ARGSUSED */
371 static void
372 arptimer(void *arg)
373 {
374 	struct llinfo_arp *la, *nla;
375 
376 	mutex_enter(softnet_lock);
377 	KERNEL_LOCK(1, NULL);
378 
379 	if (arp_lock_try(0) == 0) {
380 		/* get it later.. */
381 		KERNEL_UNLOCK_ONE(NULL);
382 		mutex_exit(softnet_lock);
383 		return;
384 	}
385 
386 	callout_reset(&arptimer_ch, arpt_prune * hz, arptimer, NULL);
387 	for (la = LIST_FIRST(&llinfo_arp); la != NULL; la = nla) {
388 		struct rtentry *rt = la->la_rt;
389 
390 		nla = LIST_NEXT(la, la_list);
391 		if (rt->rt_expire == 0)
392 			continue;
393 		if ((rt->rt_expire - time_second) < arpt_refresh &&
394 		    rt->rt_pksent > (time_second - arpt_keep)) {
395 			/*
396 			 * If the entry has been used during since last
397 			 * refresh, try to renew it before deleting.
398 			 */
399 			arprequest(rt->rt_ifp,
400 			    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
401 			    &satocsin(rt_getkey(rt))->sin_addr,
402 			    CLLADDR(rt->rt_ifp->if_sadl));
403 		} else if (rt->rt_expire <= time_second)
404 			arptfree(la); /* timer has expired; clear */
405 	}
406 
407 	ARP_UNLOCK();
408 
409 	KERNEL_UNLOCK_ONE(NULL);
410 	mutex_exit(softnet_lock);
411 }
412 
413 /*
414  * We set the gateway for RTF_CLONING routes to a "prototype"
415  * link-layer sockaddr whose interface type (if_type) and interface
416  * index (if_index) fields are prepared.
417  */
418 static struct sockaddr *
419 arp_setgate(struct rtentry *rt, struct sockaddr *gate,
420     const struct sockaddr *netmask)
421 {
422 	const struct ifnet *ifp = rt->rt_ifp;
423 	uint8_t namelen = strlen(ifp->if_xname);
424 	uint8_t addrlen = ifp->if_addrlen;
425 
426 	/*
427 	 * XXX: If this is a manually added route to interface
428 	 * such as older version of routed or gated might provide,
429 	 * restore cloning bit.
430 	 */
431 	if ((rt->rt_flags & RTF_HOST) == 0 && netmask != NULL &&
432 	    satocsin(netmask)->sin_addr.s_addr != 0xffffffff)
433 		rt->rt_flags |= RTF_CLONING;
434 	if (rt->rt_flags & RTF_CLONING) {
435 		union {
436 			struct sockaddr sa;
437 			struct sockaddr_storage ss;
438 			struct sockaddr_dl sdl;
439 		} u;
440 		/*
441 		 * Case 1: This route should come from a route to iface.
442 		 */
443 		sockaddr_dl_init(&u.sdl, sizeof(u.ss),
444 		    ifp->if_index, ifp->if_type, NULL, namelen, NULL, addrlen);
445 		rt_setgate(rt, &u.sa);
446 		gate = rt->rt_gateway;
447 	}
448 	return gate;
449 }
450 
451 /*
452  * Parallel to llc_rtrequest.
453  */
454 void
455 arp_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
456 {
457 	struct sockaddr *gate = rt->rt_gateway;
458 	struct llinfo_arp *la = (struct llinfo_arp *)rt->rt_llinfo;
459 	size_t allocsize;
460 	struct mbuf *mold;
461 	int s;
462 	struct in_ifaddr *ia;
463 	struct ifaddr *ifa;
464 	struct ifnet *ifp = rt->rt_ifp;
465 
466 	if (!arpinit_done) {
467 		arpinit_done = 1;
468 		/*
469 		 * We generate expiration times from time_second
470 		 * so avoid accidentally creating permanent routes.
471 		 */
472 		if (time_second == 0) {
473 			struct timespec ts;
474 			ts.tv_sec = 1;
475 			ts.tv_nsec = 0;
476 			tc_setclock(&ts);
477 		}
478 		callout_init(&arptimer_ch, CALLOUT_MPSAFE);
479 		callout_reset(&arptimer_ch, hz, arptimer, NULL);
480 	}
481 
482 	if (req == RTM_LLINFO_UPD) {
483 		struct in_addr *in;
484 
485 		if ((ifa = info->rti_ifa) == NULL)
486 			return;
487 
488 		in = &ifatoia(ifa)->ia_addr.sin_addr;
489 
490 		arprequest(ifa->ifa_ifp, in, in,
491 		    CLLADDR(ifa->ifa_ifp->if_sadl));
492 		return;
493 	}
494 
495 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
496 		if (req != RTM_ADD)
497 			return;
498 
499 		/*
500 		 * linklayers with particular link MTU limitation.
501 		 */
502 		switch(ifp->if_type) {
503 #if NFDDI > 0
504 		case IFT_FDDI:
505 			if (ifp->if_mtu > FDDIIPMTU)
506 				rt->rt_rmx.rmx_mtu = FDDIIPMTU;
507 			break;
508 #endif
509 #if NARC > 0
510 		case IFT_ARCNET:
511 		    {
512 			int arcipifmtu;
513 
514 			if (ifp->if_flags & IFF_LINK0)
515 				arcipifmtu = arc_ipmtu;
516 			else
517 				arcipifmtu = ARCMTU;
518 			if (ifp->if_mtu > arcipifmtu)
519 				rt->rt_rmx.rmx_mtu = arcipifmtu;
520 			break;
521 		    }
522 #endif
523 		}
524 		return;
525 	}
526 
527 	ARP_LOCK(1);		/* we may already be locked here. */
528 
529 	switch (req) {
530 	case RTM_SETGATE:
531 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
532 		break;
533 	case RTM_ADD:
534 		gate = arp_setgate(rt, gate, info->rti_info[RTAX_NETMASK]);
535 		if (rt->rt_flags & RTF_CLONING) {
536 			/*
537 			 * Give this route an expiration time, even though
538 			 * it's a "permanent" route, so that routes cloned
539 			 * from it do not need their expiration time set.
540 			 */
541 			rt->rt_expire = time_second;
542 			/*
543 			 * linklayers with particular link MTU limitation.
544 			 */
545 			switch (ifp->if_type) {
546 #if NFDDI > 0
547 			case IFT_FDDI:
548 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
549 				    (rt->rt_rmx.rmx_mtu > FDDIIPMTU ||
550 				     (rt->rt_rmx.rmx_mtu == 0 &&
551 				      ifp->if_mtu > FDDIIPMTU)))
552 					rt->rt_rmx.rmx_mtu = FDDIIPMTU;
553 				break;
554 #endif
555 #if NARC > 0
556 			case IFT_ARCNET:
557 			    {
558 				int arcipifmtu;
559 				if (ifp->if_flags & IFF_LINK0)
560 					arcipifmtu = arc_ipmtu;
561 				else
562 					arcipifmtu = ARCMTU;
563 
564 				if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0 &&
565 				    (rt->rt_rmx.rmx_mtu > arcipifmtu ||
566 				     (rt->rt_rmx.rmx_mtu == 0 &&
567 				      ifp->if_mtu > arcipifmtu)))
568 					rt->rt_rmx.rmx_mtu = arcipifmtu;
569 				break;
570 			    }
571 #endif
572 			}
573 			break;
574 		}
575 		/* Announce a new entry if requested. */
576 		if (rt->rt_flags & RTF_ANNOUNCE) {
577 			arprequest(ifp,
578 			    &satocsin(rt_getkey(rt))->sin_addr,
579 			    &satocsin(rt_getkey(rt))->sin_addr,
580 			    CLLADDR(satocsdl(gate)));
581 		}
582 		/*FALLTHROUGH*/
583 	case RTM_RESOLVE:
584 		if (gate->sa_family != AF_LINK ||
585 		    gate->sa_len < sockaddr_dl_measure(0, ifp->if_addrlen)) {
586 			log(LOG_DEBUG, "arp_rtrequest: bad gateway value\n");
587 			break;
588 		}
589 		satosdl(gate)->sdl_type = ifp->if_type;
590 		satosdl(gate)->sdl_index = ifp->if_index;
591 		if (la != NULL)
592 			break; /* This happens on a route change */
593 		/*
594 		 * Case 2:  This route may come from cloning, or a manual route
595 		 * add with a LL address.
596 		 */
597 		switch (ifp->if_type) {
598 #if NTOKEN > 0
599 		case IFT_ISO88025:
600 			allocsize = sizeof(*la) + sizeof(struct token_rif);
601 			break;
602 #endif /* NTOKEN > 0 */
603 		default:
604 			allocsize = sizeof(*la);
605 		}
606 		R_Malloc(la, struct llinfo_arp *, allocsize);
607 		rt->rt_llinfo = (void *)la;
608 		if (la == NULL) {
609 			log(LOG_DEBUG, "arp_rtrequest: malloc failed\n");
610 			break;
611 		}
612 		arp_inuse++, arp_allocated++;
613 		memset(la, 0, allocsize);
614 		la->la_rt = rt;
615 		rt->rt_flags |= RTF_LLINFO;
616 		LIST_INSERT_HEAD(&llinfo_arp, la, la_list);
617 
618 		INADDR_TO_IA(satocsin(rt_getkey(rt))->sin_addr, ia);
619 		while (ia && ia->ia_ifp != ifp)
620 			NEXT_IA_WITH_SAME_ADDR(ia);
621 		if (ia) {
622 			/*
623 			 * This test used to be
624 			 *	if (lo0ifp->if_flags & IFF_UP)
625 			 * It allowed local traffic to be forced through
626 			 * the hardware by configuring the loopback down.
627 			 * However, it causes problems during network
628 			 * configuration for boards that can't receive
629 			 * packets they send.  It is now necessary to clear
630 			 * "useloopback" and remove the route to force
631 			 * traffic out to the hardware.
632 			 *
633 			 * In 4.4BSD, the above "if" statement checked
634 			 * rt->rt_ifa against rt_getkey(rt).  It was changed
635 			 * to the current form so that we can provide a
636 			 * better support for multiple IPv4 addresses on a
637 			 * interface.
638 			 */
639 			rt->rt_expire = 0;
640 			if (sockaddr_dl_init(satosdl(gate), gate->sa_len,
641 			    ifp->if_index, ifp->if_type, NULL, 0,
642 			    CLLADDR(ifp->if_sadl), ifp->if_addrlen) == NULL) {
643 				panic("%s(%s): sockaddr_dl_init cannot fail",
644 				    __func__, ifp->if_xname);
645 			}
646 			if (useloopback)
647 				ifp = rt->rt_ifp = lo0ifp;
648 			/*
649 			 * make sure to set rt->rt_ifa to the interface
650 			 * address we are using, otherwise we will have trouble
651 			 * with source address selection.
652 			 */
653 			ifa = &ia->ia_ifa;
654 			if (ifa != rt->rt_ifa)
655 				rt_replace_ifa(rt, ifa);
656 		}
657 		break;
658 
659 	case RTM_DELETE:
660 		if (la == NULL)
661 			break;
662 		arp_inuse--;
663 		LIST_REMOVE(la, la_list);
664 		rt->rt_llinfo = NULL;
665 		rt->rt_flags &= ~RTF_LLINFO;
666 
667 		s = splnet();
668 		mold = la->la_hold;
669 		la->la_hold = 0;
670 		splx(s);
671 
672 		if (mold)
673 			m_freem(mold);
674 
675 		Free((void *)la);
676 	}
677 	ARP_UNLOCK();
678 }
679 
680 /*
681  * Broadcast an ARP request. Caller specifies:
682  *	- arp header source ip address
683  *	- arp header target ip address
684  *	- arp header source ethernet address
685  */
686 void
687 arprequest(struct ifnet *ifp,
688     const struct in_addr *sip, const struct in_addr *tip,
689     const u_int8_t *enaddr)
690 {
691 	struct mbuf *m;
692 	struct arphdr *ah;
693 	struct sockaddr sa;
694 	uint64_t *arps;
695 
696 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
697 		return;
698 	MCLAIM(m, &arpdomain.dom_mowner);
699 	switch (ifp->if_type) {
700 	case IFT_IEEE1394:
701 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
702 		    ifp->if_addrlen;
703 		break;
704 	default:
705 		m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) +
706 		    2 * ifp->if_addrlen;
707 		break;
708 	}
709 	m->m_pkthdr.len = m->m_len;
710 	MH_ALIGN(m, m->m_len);
711 	ah = mtod(m, struct arphdr *);
712 	memset(ah, 0, m->m_len);
713 	switch (ifp->if_type) {
714 	case IFT_IEEE1394:	/* RFC2734 */
715 		/* fill it now for ar_tpa computation */
716 		ah->ar_hrd = htons(ARPHRD_IEEE1394);
717 		break;
718 	default:
719 		/* ifp->if_output will fill ar_hrd */
720 		break;
721 	}
722 	ah->ar_pro = htons(ETHERTYPE_IP);
723 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
724 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
725 	ah->ar_op = htons(ARPOP_REQUEST);
726 	memcpy(ar_sha(ah), enaddr, ah->ar_hln);
727 	memcpy(ar_spa(ah), sip, ah->ar_pln);
728 	memcpy(ar_tpa(ah), tip, ah->ar_pln);
729 	sa.sa_family = AF_ARP;
730 	sa.sa_len = 2;
731 	m->m_flags |= M_BCAST;
732 	arps = ARP_STAT_GETREF();
733 	arps[ARP_STAT_SNDTOTAL]++;
734 	arps[ARP_STAT_SENDREQUEST]++;
735 	ARP_STAT_PUTREF();
736 	(*ifp->if_output)(ifp, m, &sa, NULL);
737 }
738 
739 /*
740  * Resolve an IP address into an ethernet address.  If success,
741  * desten is filled in.  If there is no entry in arptab,
742  * set one up and broadcast a request for the IP address.
743  * Hold onto this mbuf and resend it once the address
744  * is finally resolved.  A return value of 1 indicates
745  * that desten has been filled in and the packet should be sent
746  * normally; a 0 return indicates that the packet has been
747  * taken over here, either now or for later transmission.
748  */
749 int
750 arpresolve(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
751     const struct sockaddr *dst, u_char *desten)
752 {
753 	struct llinfo_arp *la;
754 	const struct sockaddr_dl *sdl;
755 	struct mbuf *mold;
756 	int s;
757 
758 	if ((la = arplookup1(m, &satocsin(dst)->sin_addr, 1, 0, rt)) != NULL)
759 		rt = la->la_rt;
760 
761 	if (la == NULL || rt == NULL) {
762 		ARP_STATINC(ARP_STAT_ALLOCFAIL);
763 		log(LOG_DEBUG,
764 		    "arpresolve: can't allocate llinfo on %s for %s\n",
765 		    ifp->if_xname, in_fmtaddr(satocsin(dst)->sin_addr));
766 		m_freem(m);
767 		return 0;
768 	}
769 	sdl = satocsdl(rt->rt_gateway);
770 	/*
771 	 * Check the address family and length is valid, the address
772 	 * is resolved; otherwise, try to resolve.
773 	 */
774 	if ((rt->rt_expire == 0 || rt->rt_expire > time_second) &&
775 	    sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0) {
776 		memcpy(desten, CLLADDR(sdl),
777 		    min(sdl->sdl_alen, ifp->if_addrlen));
778 		rt->rt_pksent = time_second; /* Time for last pkt sent */
779 		return 1;
780 	}
781 	/*
782 	 * There is an arptab entry, but no ethernet address
783 	 * response yet.  Replace the held mbuf with this
784 	 * latest one.
785 	 */
786 
787 	ARP_STATINC(ARP_STAT_DFRTOTAL);
788 	s = splnet();
789 	mold = la->la_hold;
790 	la->la_hold = m;
791 	splx(s);
792 
793 	if (mold) {
794 		ARP_STATINC(ARP_STAT_DFRDROPPED);
795 		m_freem(mold);
796 	}
797 
798 	/*
799 	 * Re-send the ARP request when appropriate.
800 	 */
801 #ifdef	DIAGNOSTIC
802 	if (rt->rt_expire == 0) {
803 		/* This should never happen. (Should it? -gwr) */
804 		printf("arpresolve: unresolved and rt_expire == 0\n");
805 		/* Set expiration time to now (expired). */
806 		rt->rt_expire = time_second;
807 	}
808 #endif
809 	if (rt->rt_expire) {
810 		rt->rt_flags &= ~RTF_REJECT;
811 		if (la->la_asked == 0 || rt->rt_expire != time_second) {
812 			rt->rt_expire = time_second;
813 			if (la->la_asked++ < arp_maxtries) {
814 				arprequest(ifp,
815 				    &satocsin(rt->rt_ifa->ifa_addr)->sin_addr,
816 				    &satocsin(dst)->sin_addr,
817 #if NCARP > 0
818 				    (rt->rt_ifp->if_type == IFT_CARP) ?
819 				    CLLADDR(rt->rt_ifp->if_sadl):
820 #endif
821 				    CLLADDR(ifp->if_sadl));
822 			} else {
823 				rt->rt_flags |= RTF_REJECT;
824 				rt->rt_expire += arpt_down;
825 				la->la_asked = 0;
826 			}
827 		}
828 	}
829 	return 0;
830 }
831 
832 /*
833  * Common length and type checks are done here,
834  * then the protocol-specific routine is called.
835  */
836 void
837 arpintr(void)
838 {
839 	struct mbuf *m;
840 	struct arphdr *ar;
841 	int s;
842 	int arplen;
843 
844 	mutex_enter(softnet_lock);
845 	KERNEL_LOCK(1, NULL);
846 	while (arpintrq.ifq_head) {
847 		s = splnet();
848 		IF_DEQUEUE(&arpintrq, m);
849 		splx(s);
850 		if (m == 0 || (m->m_flags & M_PKTHDR) == 0)
851 			panic("arpintr");
852 
853 		MCLAIM(m, &arpdomain.dom_mowner);
854 		ARP_STATINC(ARP_STAT_RCVTOTAL);
855 
856 		/*
857 		 * First, make sure we have at least struct arphdr.
858 		 */
859 		if (m->m_len < sizeof(struct arphdr) ||
860 		    (ar = mtod(m, struct arphdr *)) == NULL)
861 			goto badlen;
862 
863 		switch (m->m_pkthdr.rcvif->if_type) {
864 		case IFT_IEEE1394:
865 			arplen = sizeof(struct arphdr) +
866 			    ar->ar_hln + 2 * ar->ar_pln;
867 			break;
868 		default:
869 			arplen = sizeof(struct arphdr) +
870 			    2 * ar->ar_hln + 2 * ar->ar_pln;
871 			break;
872 		}
873 
874 		if (/* XXX ntohs(ar->ar_hrd) == ARPHRD_ETHER && */
875 		    m->m_len >= arplen)
876 			switch (ntohs(ar->ar_pro)) {
877 			case ETHERTYPE_IP:
878 			case ETHERTYPE_IPTRAILERS:
879 				in_arpinput(m);
880 				continue;
881 			default:
882 				ARP_STATINC(ARP_STAT_RCVBADPROTO);
883 			}
884 		else {
885 badlen:
886 			ARP_STATINC(ARP_STAT_RCVBADLEN);
887 		}
888 		m_freem(m);
889 	}
890 	KERNEL_UNLOCK_ONE(NULL);
891 	mutex_exit(softnet_lock);
892 }
893 
894 /*
895  * ARP for Internet protocols on 10 Mb/s Ethernet.
896  * Algorithm is that given in RFC 826.
897  * In addition, a sanity check is performed on the sender
898  * protocol address, to catch impersonators.
899  * We no longer handle negotiations for use of trailer protocol:
900  * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent
901  * along with IP replies if we wanted trailers sent to us,
902  * and also sent them in response to IP replies.
903  * This allowed either end to announce the desire to receive
904  * trailer packets.
905  * We no longer reply to requests for ETHERTYPE_TRAIL protocol either,
906  * but formerly didn't normally send requests.
907  */
908 static void
909 in_arpinput(struct mbuf *m)
910 {
911 	struct arphdr *ah;
912 	struct ifnet *ifp = m->m_pkthdr.rcvif;
913 	struct llinfo_arp *la = NULL;
914 	struct rtentry  *rt;
915 	struct in_ifaddr *ia;
916 #if NBRIDGE > 0
917 	struct in_ifaddr *bridge_ia = NULL;
918 #endif
919 #if NCARP > 0
920 	u_int32_t count = 0, index = 0;
921 #endif
922 	struct sockaddr_dl *sdl;
923 	struct sockaddr sa;
924 	struct in_addr isaddr, itaddr, myaddr;
925 	int op;
926 	struct mbuf *mold;
927 	void *tha;
928 	int s;
929 	uint64_t *arps;
930 
931 	if (__predict_false(m_makewritable(&m, 0, m->m_pkthdr.len, M_DONTWAIT)))
932 		goto out;
933 	ah = mtod(m, struct arphdr *);
934 	op = ntohs(ah->ar_op);
935 
936 	/*
937 	 * Fix up ah->ar_hrd if necessary, before using ar_tha() or
938 	 * ar_tpa().
939 	 */
940 	switch (ifp->if_type) {
941 	case IFT_IEEE1394:
942 		if (ntohs(ah->ar_hrd) == ARPHRD_IEEE1394)
943 			;
944 		else {
945 			/* XXX this is to make sure we compute ar_tha right */
946 			/* XXX check ar_hrd more strictly? */
947 			ah->ar_hrd = htons(ARPHRD_IEEE1394);
948 		}
949 		break;
950 	default:
951 		/* XXX check ar_hrd? */
952 		break;
953 	}
954 
955 	memcpy(&isaddr, ar_spa(ah), sizeof (isaddr));
956 	memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr));
957 
958 	if (m->m_flags & (M_BCAST|M_MCAST))
959 		ARP_STATINC(ARP_STAT_RCVMCAST);
960 
961 	/*
962 	 * If the target IP address is zero, ignore the packet.
963 	 * This prevents the code below from tring to answer
964 	 * when we are using IP address zero (booting).
965 	 */
966 	if (in_nullhost(itaddr)) {
967 		ARP_STATINC(ARP_STAT_RCVZEROTPA);
968 		goto out;
969 	}
970 
971 
972 	/*
973 	 * Search for a matching interface address
974 	 * or any address on the interface to use
975 	 * as a dummy address in the rest of this function
976 	 */
977 
978 	INADDR_TO_IA(itaddr, ia);
979 	while (ia != NULL) {
980 #if NCARP > 0
981 		if (ia->ia_ifp->if_type == IFT_CARP &&
982 		    ((ia->ia_ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
983 		    (IFF_UP|IFF_RUNNING))) {
984 			index++;
985 			if (ia->ia_ifp == m->m_pkthdr.rcvif &&
986 			    carp_iamatch(ia, ar_sha(ah),
987 			    &count, index)) {
988 				break;
989 				}
990 		} else
991 #endif
992 			    if (ia->ia_ifp == m->m_pkthdr.rcvif)
993 				break;
994 #if NBRIDGE > 0
995 		/*
996 		 * If the interface we received the packet on
997 		 * is part of a bridge, check to see if we need
998 		 * to "bridge" the packet to ourselves at this
999 		 * layer.  Note we still prefer a perfect match,
1000 		 * but allow this weaker match if necessary.
1001 		 */
1002 		if (m->m_pkthdr.rcvif->if_bridge != NULL &&
1003 		    m->m_pkthdr.rcvif->if_bridge == ia->ia_ifp->if_bridge)
1004 			bridge_ia = ia;
1005 #endif /* NBRIDGE > 0 */
1006 
1007 		NEXT_IA_WITH_SAME_ADDR(ia);
1008 	}
1009 
1010 #if NBRIDGE > 0
1011 	if (ia == NULL && bridge_ia != NULL) {
1012 		ia = bridge_ia;
1013 		ifp = bridge_ia->ia_ifp;
1014 	}
1015 #endif
1016 
1017 	if (ia == NULL) {
1018 		INADDR_TO_IA(isaddr, ia);
1019 		while ((ia != NULL) && ia->ia_ifp != m->m_pkthdr.rcvif)
1020 			NEXT_IA_WITH_SAME_ADDR(ia);
1021 
1022 		if (ia == NULL) {
1023 			IFP_TO_IA(ifp, ia);
1024 			if (ia == NULL) {
1025 				ARP_STATINC(ARP_STAT_RCVNOINT);
1026 				goto out;
1027 			}
1028 		}
1029 	}
1030 
1031 	myaddr = ia->ia_addr.sin_addr;
1032 
1033 	/* XXX checks for bridge case? */
1034 	if (!memcmp(ar_sha(ah), CLLADDR(ifp->if_sadl), ifp->if_addrlen)) {
1035 		ARP_STATINC(ARP_STAT_RCVLOCALSHA);
1036 		goto out;	/* it's from me, ignore it. */
1037 	}
1038 
1039 	/* XXX checks for bridge case? */
1040 	if (!memcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) {
1041 		ARP_STATINC(ARP_STAT_RCVBCASTSHA);
1042 		log(LOG_ERR,
1043 		    "%s: arp: link address is broadcast for IP address %s!\n",
1044 		    ifp->if_xname, in_fmtaddr(isaddr));
1045 		goto out;
1046 	}
1047 
1048 	/*
1049 	 * If the source IP address is zero, this is an RFC 5227 ARP probe
1050 	 */
1051 	if (in_nullhost(isaddr)) {
1052 		ARP_STATINC(ARP_STAT_RCVZEROSPA);
1053 		goto reply;
1054 	}
1055 
1056 	if (in_hosteq(isaddr, myaddr)) {
1057 		ARP_STATINC(ARP_STAT_RCVLOCALSPA);
1058 		log(LOG_ERR,
1059 		   "duplicate IP address %s sent from link address %s\n",
1060 		   in_fmtaddr(isaddr), lla_snprintf(ar_sha(ah), ah->ar_hln));
1061 		itaddr = myaddr;
1062 		goto reply;
1063 	}
1064 	la = arplookup(m, &isaddr, in_hosteq(itaddr, myaddr), 0);
1065 	if (la != NULL && (rt = la->la_rt) && (sdl = satosdl(rt->rt_gateway))) {
1066 		if (sdl->sdl_alen &&
1067 		    memcmp(ar_sha(ah), CLLADDR(sdl), sdl->sdl_alen)) {
1068 			if (rt->rt_flags & RTF_STATIC) {
1069 				ARP_STATINC(ARP_STAT_RCVOVERPERM);
1070 				log(LOG_INFO,
1071 				    "%s tried to overwrite permanent arp info"
1072 				    " for %s\n",
1073 				    lla_snprintf(ar_sha(ah), ah->ar_hln),
1074 				    in_fmtaddr(isaddr));
1075 				goto out;
1076 			} else if (rt->rt_ifp != ifp) {
1077 				ARP_STATINC(ARP_STAT_RCVOVERINT);
1078 				log(LOG_INFO,
1079 				    "%s on %s tried to overwrite "
1080 				    "arp info for %s on %s\n",
1081 				    lla_snprintf(ar_sha(ah), ah->ar_hln),
1082 				    ifp->if_xname, in_fmtaddr(isaddr),
1083 				    rt->rt_ifp->if_xname);
1084 				    goto out;
1085 			} else {
1086 				ARP_STATINC(ARP_STAT_RCVOVER);
1087 				log(LOG_INFO,
1088 				    "arp info overwritten for %s by %s\n",
1089 				    in_fmtaddr(isaddr),
1090 				    lla_snprintf(ar_sha(ah), ah->ar_hln));
1091 			}
1092 		}
1093 		/*
1094 		 * sanity check for the address length.
1095 		 * XXX this does not work for protocols with variable address
1096 		 * length. -is
1097 		 */
1098 		if (sdl->sdl_alen &&
1099 		    sdl->sdl_alen != ah->ar_hln) {
1100 			ARP_STATINC(ARP_STAT_RCVLENCHG);
1101 			log(LOG_WARNING,
1102 			    "arp from %s: new addr len %d, was %d\n",
1103 			    in_fmtaddr(isaddr), ah->ar_hln, sdl->sdl_alen);
1104 		}
1105 		if (ifp->if_addrlen != ah->ar_hln) {
1106 			ARP_STATINC(ARP_STAT_RCVBADLEN);
1107 			log(LOG_WARNING,
1108 			    "arp from %s: addr len: new %d, i/f %d (ignored)\n",
1109 			    in_fmtaddr(isaddr), ah->ar_hln,
1110 			    ifp->if_addrlen);
1111 			goto reply;
1112 		}
1113 #if NTOKEN > 0
1114 		/*
1115 		 * XXX uses m_data and assumes the complete answer including
1116 		 * XXX token-ring headers is in the same buf
1117 		 */
1118 		if (ifp->if_type == IFT_ISO88025) {
1119 			struct token_header *trh;
1120 
1121 			trh = (struct token_header *)M_TRHSTART(m);
1122 			if (trh->token_shost[0] & TOKEN_RI_PRESENT) {
1123 				struct token_rif	*rif;
1124 				size_t	riflen;
1125 
1126 				rif = TOKEN_RIF(trh);
1127 				riflen = (ntohs(rif->tr_rcf) &
1128 				    TOKEN_RCF_LEN_MASK) >> 8;
1129 
1130 				if (riflen > 2 &&
1131 				    riflen < sizeof(struct token_rif) &&
1132 				    (riflen & 1) == 0) {
1133 					rif->tr_rcf ^= htons(TOKEN_RCF_DIRECTION);
1134 					rif->tr_rcf &= htons(~TOKEN_RCF_BROADCAST_MASK);
1135 					memcpy(TOKEN_RIF(la), rif, riflen);
1136 				}
1137 			}
1138 		}
1139 #endif /* NTOKEN > 0 */
1140 		(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, ar_sha(ah),
1141 		    ah->ar_hln);
1142 		if (rt->rt_expire)
1143 			rt->rt_expire = time_second + arpt_keep;
1144 		rt->rt_flags &= ~RTF_REJECT;
1145 		la->la_asked = 0;
1146 
1147 		s = splnet();
1148 		mold = la->la_hold;
1149 		la->la_hold = 0;
1150 		splx(s);
1151 
1152 		if (mold) {
1153 			ARP_STATINC(ARP_STAT_DFRSENT);
1154 			(*ifp->if_output)(ifp, mold, rt_getkey(rt), rt);
1155 		}
1156 	}
1157 reply:
1158 	if (op != ARPOP_REQUEST) {
1159 		if (op == ARPOP_REPLY)
1160 			ARP_STATINC(ARP_STAT_RCVREPLY);
1161 	out:
1162 		m_freem(m);
1163 		return;
1164 	}
1165 	ARP_STATINC(ARP_STAT_RCVREQUEST);
1166 	if (in_hosteq(itaddr, myaddr)) {
1167 		/* I am the target */
1168 		tha = ar_tha(ah);
1169 		if (tha)
1170 			memcpy(tha, ar_sha(ah), ah->ar_hln);
1171 		memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
1172 	} else {
1173 		la = arplookup(m, &itaddr, 0, SIN_PROXY);
1174 		if (la == NULL)
1175 			goto out;
1176 		rt = la->la_rt;
1177 		if (rt->rt_ifp->if_type == IFT_CARP &&
1178 		    m->m_pkthdr.rcvif->if_type != IFT_CARP)
1179 			goto out;
1180 		tha = ar_tha(ah);
1181 		if (tha)
1182 			memcpy(tha, ar_sha(ah), ah->ar_hln);
1183 		sdl = satosdl(rt->rt_gateway);
1184 		memcpy(ar_sha(ah), CLLADDR(sdl), ah->ar_hln);
1185 	}
1186 
1187 	memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln);
1188 	memcpy(ar_spa(ah), &itaddr, ah->ar_pln);
1189 	ah->ar_op = htons(ARPOP_REPLY);
1190 	ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */
1191 	switch (ifp->if_type) {
1192 	case IFT_IEEE1394:
1193 		/*
1194 		 * ieee1394 arp reply is broadcast
1195 		 */
1196 		m->m_flags &= ~M_MCAST;
1197 		m->m_flags |= M_BCAST;
1198 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + ah->ar_hln;
1199 		break;
1200 
1201 	default:
1202 		m->m_flags &= ~(M_BCAST|M_MCAST); /* never reply by broadcast */
1203 		m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln);
1204 		break;
1205 	}
1206 	m->m_pkthdr.len = m->m_len;
1207 	sa.sa_family = AF_ARP;
1208 	sa.sa_len = 2;
1209 	arps = ARP_STAT_GETREF();
1210 	arps[ARP_STAT_SNDTOTAL]++;
1211 	arps[ARP_STAT_SNDREPLY]++;
1212 	ARP_STAT_PUTREF();
1213 	(*ifp->if_output)(ifp, m, &sa, (struct rtentry *)0);
1214 	return;
1215 }
1216 
1217 /*
1218  * Free an arp entry.
1219  */
1220 static void arptfree(struct llinfo_arp *la)
1221 {
1222 	struct rtentry *rt = la->la_rt;
1223 	struct sockaddr_dl *sdl;
1224 
1225 	ARP_LOCK_CHECK();
1226 
1227 	if (rt == NULL)
1228 		panic("arptfree");
1229 	if (rt->rt_refcnt > 0 && (sdl = satosdl(rt->rt_gateway)) &&
1230 	    sdl->sdl_family == AF_LINK) {
1231 		sdl->sdl_alen = 0;
1232 		la->la_asked = 0;
1233 		rt->rt_flags &= ~RTF_REJECT;
1234 		return;
1235 	}
1236 	rtrequest(RTM_DELETE, rt_getkey(rt), NULL, rt_mask(rt), 0, NULL);
1237 }
1238 
1239 static struct llinfo_arp *
1240 arplookup(struct mbuf *m, const struct in_addr *addr, int create, int proxy)
1241 {
1242 	return arplookup1(m, addr, create, proxy, NULL);
1243 }
1244 
1245 /*
1246  * Lookup or enter a new address in arptab.
1247  */
1248 static struct llinfo_arp *
1249 arplookup1(struct mbuf *m, const struct in_addr *addr, int create, int proxy,
1250     struct rtentry *rt0)
1251 {
1252 	struct arphdr *ah;
1253 	struct ifnet *ifp = m->m_pkthdr.rcvif;
1254 	struct rtentry *rt;
1255 	struct sockaddr_inarp sin;
1256 	const char *why = NULL;
1257 
1258 	ah = mtod(m, struct arphdr *);
1259 	if (rt0 == NULL) {
1260 		memset(&sin, 0, sizeof(sin));
1261 		sin.sin_len = sizeof(sin);
1262 		sin.sin_family = AF_INET;
1263 		sin.sin_addr = *addr;
1264 		sin.sin_other = proxy ? SIN_PROXY : 0;
1265 		rt = rtalloc1(sintosa(&sin), create);
1266 		if (rt == NULL)
1267 			return NULL;
1268 		rt->rt_refcnt--;
1269 	} else
1270 		rt = rt0;
1271 
1272 #define	IS_LLINFO(__rt)							  \
1273 	(((__rt)->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) == RTF_LLINFO && \
1274 	 (__rt)->rt_gateway->sa_family == AF_LINK)
1275 
1276 
1277 	if (IS_LLINFO(rt))
1278 		return (struct llinfo_arp *)rt->rt_llinfo;
1279 
1280 	if (create) {
1281 		if (rt->rt_flags & RTF_GATEWAY)
1282 			why = "host is not on local network";
1283 		else if ((rt->rt_flags & RTF_LLINFO) == 0) {
1284 			ARP_STATINC(ARP_STAT_ALLOCFAIL);
1285 			why = "could not allocate llinfo";
1286 		} else
1287 			why = "gateway route is not ours";
1288 		log(LOG_DEBUG, "arplookup: unable to enter address"
1289 		    " for %s@%s on %s (%s)\n",
1290 		    in_fmtaddr(*addr), lla_snprintf(ar_sha(ah), ah->ar_hln),
1291 		    (ifp) ? ifp->if_xname : "null", why);
1292 		if (rt->rt_refcnt <= 0 && (rt->rt_flags & RTF_CLONED) != 0) {
1293 			rtrequest(RTM_DELETE, rt_getkey(rt),
1294 		    	    rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL);
1295 		}
1296 	}
1297 	return NULL;
1298 }
1299 
1300 int
1301 arpioctl(u_long cmd, void *data)
1302 {
1303 
1304 	return EOPNOTSUPP;
1305 }
1306 
1307 void
1308 arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa)
1309 {
1310 	struct in_addr *ip;
1311 
1312 	/*
1313 	 * Warn the user if another station has this IP address,
1314 	 * but only if the interface IP address is not zero.
1315 	 */
1316 	ip = &IA_SIN(ifa)->sin_addr;
1317 	if (!in_nullhost(*ip))
1318 		arprequest(ifp, ip, ip, CLLADDR(ifp->if_sadl));
1319 
1320 	ifa->ifa_rtrequest = arp_rtrequest;
1321 	ifa->ifa_flags |= RTF_CLONING;
1322 }
1323 
1324 /*
1325  * Called from 10 Mb/s Ethernet interrupt handlers
1326  * when ether packet type ETHERTYPE_REVARP
1327  * is received.  Common length and type checks are done here,
1328  * then the protocol-specific routine is called.
1329  */
1330 void
1331 revarpinput(struct mbuf *m)
1332 {
1333 	struct arphdr *ar;
1334 
1335 	if (m->m_len < sizeof(struct arphdr))
1336 		goto out;
1337 	ar = mtod(m, struct arphdr *);
1338 #if 0 /* XXX I don't think we need this... and it will prevent other LL */
1339 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER)
1340 		goto out;
1341 #endif
1342 	if (m->m_len < sizeof(struct arphdr) + 2 * (ar->ar_hln + ar->ar_pln))
1343 		goto out;
1344 	switch (ntohs(ar->ar_pro)) {
1345 	case ETHERTYPE_IP:
1346 	case ETHERTYPE_IPTRAILERS:
1347 		in_revarpinput(m);
1348 		return;
1349 
1350 	default:
1351 		break;
1352 	}
1353 out:
1354 	m_freem(m);
1355 }
1356 
1357 /*
1358  * RARP for Internet protocols on 10 Mb/s Ethernet.
1359  * Algorithm is that given in RFC 903.
1360  * We are only using for bootstrap purposes to get an ip address for one of
1361  * our interfaces.  Thus we support no user-interface.
1362  *
1363  * Since the contents of the RARP reply are specific to the interface that
1364  * sent the request, this code must ensure that they are properly associated.
1365  *
1366  * Note: also supports ARP via RARP packets, per the RFC.
1367  */
1368 void
1369 in_revarpinput(struct mbuf *m)
1370 {
1371 	struct ifnet *ifp;
1372 	struct arphdr *ah;
1373 	void *tha;
1374 	int op;
1375 
1376 	ah = mtod(m, struct arphdr *);
1377 	op = ntohs(ah->ar_op);
1378 
1379 	switch (m->m_pkthdr.rcvif->if_type) {
1380 	case IFT_IEEE1394:
1381 		/* ARP without target hardware address is not supported */
1382 		goto out;
1383 	default:
1384 		break;
1385 	}
1386 
1387 	switch (op) {
1388 	case ARPOP_REQUEST:
1389 	case ARPOP_REPLY:	/* per RFC */
1390 		in_arpinput(m);
1391 		return;
1392 	case ARPOP_REVREPLY:
1393 		break;
1394 	case ARPOP_REVREQUEST:	/* handled by rarpd(8) */
1395 	default:
1396 		goto out;
1397 	}
1398 	if (!revarp_in_progress)
1399 		goto out;
1400 	ifp = m->m_pkthdr.rcvif;
1401 	if (ifp != myip_ifp) /* !same interface */
1402 		goto out;
1403 	if (myip_initialized)
1404 		goto wake;
1405 	tha = ar_tha(ah);
1406 	if (tha == NULL)
1407 		goto out;
1408 	if (memcmp(tha, CLLADDR(ifp->if_sadl), ifp->if_sadl->sdl_alen))
1409 		goto out;
1410 	memcpy(&srv_ip, ar_spa(ah), sizeof(srv_ip));
1411 	memcpy(&myip, ar_tpa(ah), sizeof(myip));
1412 	myip_initialized = 1;
1413 wake:	/* Do wakeup every time in case it was missed. */
1414 	wakeup((void *)&myip);
1415 
1416 out:
1417 	m_freem(m);
1418 }
1419 
1420 /*
1421  * Send a RARP request for the ip address of the specified interface.
1422  * The request should be RFC 903-compliant.
1423  */
1424 void
1425 revarprequest(struct ifnet *ifp)
1426 {
1427 	struct sockaddr sa;
1428 	struct mbuf *m;
1429 	struct arphdr *ah;
1430 	void *tha;
1431 
1432 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
1433 		return;
1434 	MCLAIM(m, &arpdomain.dom_mowner);
1435 	m->m_len = sizeof(*ah) + 2*sizeof(struct in_addr) +
1436 	    2*ifp->if_addrlen;
1437 	m->m_pkthdr.len = m->m_len;
1438 	MH_ALIGN(m, m->m_len);
1439 	ah = mtod(m, struct arphdr *);
1440 	memset(ah, 0, m->m_len);
1441 	ah->ar_pro = htons(ETHERTYPE_IP);
1442 	ah->ar_hln = ifp->if_addrlen;		/* hardware address length */
1443 	ah->ar_pln = sizeof(struct in_addr);	/* protocol address length */
1444 	ah->ar_op = htons(ARPOP_REVREQUEST);
1445 
1446 	memcpy(ar_sha(ah), CLLADDR(ifp->if_sadl), ah->ar_hln);
1447 	tha = ar_tha(ah);
1448 	if (tha == NULL)
1449 		return;
1450 	memcpy(tha, CLLADDR(ifp->if_sadl), ah->ar_hln);
1451 
1452 	sa.sa_family = AF_ARP;
1453 	sa.sa_len = 2;
1454 	m->m_flags |= M_BCAST;
1455 	(*ifp->if_output)(ifp, m, &sa, NULL);
1456 
1457 }
1458 
1459 /*
1460  * RARP for the ip address of the specified interface, but also
1461  * save the ip address of the server that sent the answer.
1462  * Timeout if no response is received.
1463  */
1464 int
1465 revarpwhoarewe(struct ifnet *ifp, struct in_addr *serv_in,
1466     struct in_addr *clnt_in)
1467 {
1468 	int result, count = 20;
1469 
1470 	myip_initialized = 0;
1471 	myip_ifp = ifp;
1472 
1473 	revarp_in_progress = 1;
1474 	while (count--) {
1475 		revarprequest(ifp);
1476 		result = tsleep((void *)&myip, PSOCK, "revarp", hz/2);
1477 		if (result != EWOULDBLOCK)
1478 			break;
1479 	}
1480 	revarp_in_progress = 0;
1481 
1482 	if (!myip_initialized)
1483 		return ENETUNREACH;
1484 
1485 	memcpy(serv_in, &srv_ip, sizeof(*serv_in));
1486 	memcpy(clnt_in, &myip, sizeof(*clnt_in));
1487 	return 0;
1488 }
1489 
1490 
1491 
1492 #ifdef DDB
1493 
1494 #include <machine/db_machdep.h>
1495 #include <ddb/db_interface.h>
1496 #include <ddb/db_output.h>
1497 
1498 static void
1499 db_print_sa(const struct sockaddr *sa)
1500 {
1501 	int len;
1502 	const u_char *p;
1503 
1504 	if (sa == NULL) {
1505 		db_printf("[NULL]");
1506 		return;
1507 	}
1508 
1509 	p = (const u_char *)sa;
1510 	len = sa->sa_len;
1511 	db_printf("[");
1512 	while (len > 0) {
1513 		db_printf("%d", *p);
1514 		p++; len--;
1515 		if (len) db_printf(",");
1516 	}
1517 	db_printf("]\n");
1518 }
1519 
1520 static void
1521 db_print_ifa(struct ifaddr *ifa)
1522 {
1523 	if (ifa == NULL)
1524 		return;
1525 	db_printf("  ifa_addr=");
1526 	db_print_sa(ifa->ifa_addr);
1527 	db_printf("  ifa_dsta=");
1528 	db_print_sa(ifa->ifa_dstaddr);
1529 	db_printf("  ifa_mask=");
1530 	db_print_sa(ifa->ifa_netmask);
1531 	db_printf("  flags=0x%x,refcnt=%d,metric=%d\n",
1532 			  ifa->ifa_flags,
1533 			  ifa->ifa_refcnt,
1534 			  ifa->ifa_metric);
1535 }
1536 
1537 static void
1538 db_print_llinfo(void *li)
1539 {
1540 	struct llinfo_arp *la;
1541 
1542 	if (li == NULL)
1543 		return;
1544 	la = (struct llinfo_arp *)li;
1545 	db_printf("  la_rt=%p la_hold=%p, la_asked=0x%lx\n",
1546 			  la->la_rt, la->la_hold, la->la_asked);
1547 }
1548 
1549 /*
1550  * Function to pass to rt_walktree().
1551  * Return non-zero error to abort walk.
1552  */
1553 static int
1554 db_show_rtentry(struct rtentry *rt, void *w)
1555 {
1556 	db_printf("rtentry=%p", rt);
1557 
1558 	db_printf(" flags=0x%x refcnt=%d use=%ld expire=%lld\n",
1559 			  rt->rt_flags, rt->rt_refcnt,
1560 			  rt->rt_use, (long long)rt->rt_expire);
1561 
1562 	db_printf(" key="); db_print_sa(rt_getkey(rt));
1563 	db_printf(" mask="); db_print_sa(rt_mask(rt));
1564 	db_printf(" gw="); db_print_sa(rt->rt_gateway);
1565 
1566 	db_printf(" ifp=%p ", rt->rt_ifp);
1567 	if (rt->rt_ifp)
1568 		db_printf("(%s)", rt->rt_ifp->if_xname);
1569 	else
1570 		db_printf("(NULL)");
1571 
1572 	db_printf(" ifa=%p\n", rt->rt_ifa);
1573 	db_print_ifa(rt->rt_ifa);
1574 
1575 	db_printf(" gwroute=%p llinfo=%p\n",
1576 			  rt->rt_gwroute, rt->rt_llinfo);
1577 	db_print_llinfo(rt->rt_llinfo);
1578 
1579 	return 0;
1580 }
1581 
1582 /*
1583  * Function to print all the route trees.
1584  * Use this from ddb:  "show arptab"
1585  */
1586 void
1587 db_show_arptab(db_expr_t addr, bool have_addr,
1588     db_expr_t count, const char *modif)
1589 {
1590 	rt_walktree(AF_INET, db_show_rtentry, NULL);
1591 }
1592 #endif
1593 
1594 static int
1595 sysctl_net_inet_arp_stats(SYSCTLFN_ARGS)
1596 {
1597 
1598 	return NETSTAT_SYSCTL(arpstat_percpu, ARP_NSTATS);
1599 }
1600 
1601 static void
1602 sysctl_net_inet_arp_setup(struct sysctllog **clog)
1603 {
1604 	const struct sysctlnode *node;
1605 
1606 	sysctl_createv(clog, 0, NULL, NULL,
1607 			CTLFLAG_PERMANENT,
1608 			CTLTYPE_NODE, "net", NULL,
1609 			NULL, 0, NULL, 0,
1610 			CTL_NET, CTL_EOL);
1611 	sysctl_createv(clog, 0, NULL, NULL,
1612 			CTLFLAG_PERMANENT,
1613 			CTLTYPE_NODE, "inet", NULL,
1614 			NULL, 0, NULL, 0,
1615 			CTL_NET, PF_INET, CTL_EOL);
1616 	sysctl_createv(clog, 0, NULL, &node,
1617 			CTLFLAG_PERMANENT,
1618 			CTLTYPE_NODE, "arp",
1619 			SYSCTL_DESCR("Address Resolution Protocol"),
1620 			NULL, 0, NULL, 0,
1621 			CTL_NET, PF_INET, CTL_CREATE, CTL_EOL);
1622 
1623 	sysctl_createv(clog, 0, NULL, NULL,
1624 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1625 			CTLTYPE_INT, "prune",
1626 			SYSCTL_DESCR("ARP cache pruning interval"),
1627 			NULL, 0, &arpt_prune, 0,
1628 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1629 
1630 	sysctl_createv(clog, 0, NULL, NULL,
1631 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1632 			CTLTYPE_INT, "keep",
1633 			SYSCTL_DESCR("Valid ARP entry lifetime"),
1634 			NULL, 0, &arpt_keep, 0,
1635 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1636 
1637 	sysctl_createv(clog, 0, NULL, NULL,
1638 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1639 			CTLTYPE_INT, "down",
1640 			SYSCTL_DESCR("Failed ARP entry lifetime"),
1641 			NULL, 0, &arpt_down, 0,
1642 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1643 
1644 	sysctl_createv(clog, 0, NULL, NULL,
1645 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1646 			CTLTYPE_INT, "refresh",
1647 			SYSCTL_DESCR("ARP entry refresh interval"),
1648 			NULL, 0, &arpt_refresh, 0,
1649 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1650 
1651 	sysctl_createv(clog, 0, NULL, NULL,
1652 			CTLFLAG_PERMANENT,
1653 			CTLTYPE_STRUCT, "stats",
1654 			SYSCTL_DESCR("ARP statistics"),
1655 			sysctl_net_inet_arp_stats, 0, NULL, 0,
1656 			CTL_NET,PF_INET, node->sysctl_num, CTL_CREATE, CTL_EOL);
1657 }
1658 
1659 #endif /* INET */
1660